1457c8996SThomas Gleixner // SPDX-License-Identifier: GPL-2.0-only 21da177e4SLinus Torvalds /* 31da177e4SLinus Torvalds * linux/mm/nommu.c 41da177e4SLinus Torvalds * 51da177e4SLinus Torvalds * Replacement code for mm functions to support CPU's that don't 61da177e4SLinus Torvalds * have any form of memory management unit (thus no virtual memory). 71da177e4SLinus Torvalds * 81da177e4SLinus Torvalds * See Documentation/nommu-mmap.txt 91da177e4SLinus Torvalds * 108feae131SDavid Howells * Copyright (c) 2004-2008 David Howells <dhowells@redhat.com> 111da177e4SLinus Torvalds * Copyright (c) 2000-2003 David McCullough <davidm@snapgear.com> 121da177e4SLinus Torvalds * Copyright (c) 2000-2001 D Jeff Dionne <jeff@uClinux.org> 131da177e4SLinus Torvalds * Copyright (c) 2002 Greg Ungerer <gerg@snapgear.com> 1429c185e5SPaul Mundt * Copyright (c) 2007-2010 Paul Mundt <lethal@linux-sh.org> 151da177e4SLinus Torvalds */ 161da177e4SLinus Torvalds 17b1de0d13SMitchel Humpherys #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt 18b1de0d13SMitchel Humpherys 19b95f1b31SPaul Gortmaker #include <linux/export.h> 201da177e4SLinus Torvalds #include <linux/mm.h> 216e84f315SIngo Molnar #include <linux/sched/mm.h> 22615d6e87SDavidlohr Bueso #include <linux/vmacache.h> 231da177e4SLinus Torvalds #include <linux/mman.h> 241da177e4SLinus Torvalds #include <linux/swap.h> 251da177e4SLinus Torvalds #include <linux/file.h> 261da177e4SLinus Torvalds #include <linux/highmem.h> 271da177e4SLinus Torvalds #include <linux/pagemap.h> 281da177e4SLinus Torvalds #include <linux/slab.h> 291da177e4SLinus Torvalds #include <linux/vmalloc.h> 301da177e4SLinus Torvalds #include <linux/blkdev.h> 311da177e4SLinus Torvalds #include <linux/backing-dev.h> 323b32123dSGideon Israel Dsouza #include <linux/compiler.h> 331da177e4SLinus Torvalds #include <linux/mount.h> 341da177e4SLinus Torvalds #include <linux/personality.h> 351da177e4SLinus Torvalds #include <linux/security.h> 361da177e4SLinus Torvalds #include <linux/syscalls.h> 37120a795dSAl Viro #include <linux/audit.h> 38b1de0d13SMitchel Humpherys #include <linux/printk.h> 391da177e4SLinus Torvalds 407c0f6ba6SLinus Torvalds #include <linux/uaccess.h> 411da177e4SLinus Torvalds #include <asm/tlb.h> 421da177e4SLinus Torvalds #include <asm/tlbflush.h> 43eb8cdec4SBernd Schmidt #include <asm/mmu_context.h> 448feae131SDavid Howells #include "internal.h" 458feae131SDavid Howells 461da177e4SLinus Torvalds void *high_memory; 47944b6874SArnd Bergmann EXPORT_SYMBOL(high_memory); 481da177e4SLinus Torvalds struct page *mem_map; 491da177e4SLinus Torvalds unsigned long max_mapnr; 505b8bf307Sgchen gchen EXPORT_SYMBOL(max_mapnr); 514266c97aSHugh Dickins unsigned long highest_memmap_pfn; 52fc4d5c29SDavid Howells int sysctl_nr_trim_pages = CONFIG_NOMMU_INITIAL_TRIM_EXCESS; 531da177e4SLinus Torvalds int heap_stack_gap = 0; 541da177e4SLinus Torvalds 5533e5d769SDavid Howells atomic_long_t mmap_pages_allocated; 568feae131SDavid Howells 571da177e4SLinus Torvalds EXPORT_SYMBOL(mem_map); 581da177e4SLinus Torvalds 598feae131SDavid Howells /* list of mapped, potentially shareable regions */ 608feae131SDavid Howells static struct kmem_cache *vm_region_jar; 618feae131SDavid Howells struct rb_root nommu_region_tree = RB_ROOT; 628feae131SDavid Howells DECLARE_RWSEM(nommu_region_sem); 631da177e4SLinus Torvalds 64f0f37e2fSAlexey Dobriyan const struct vm_operations_struct generic_file_vm_ops = { 651da177e4SLinus Torvalds }; 661da177e4SLinus Torvalds 671da177e4SLinus Torvalds /* 681da177e4SLinus Torvalds * Return the total memory allocated for this pointer, not 691da177e4SLinus Torvalds * just what the caller asked for. 701da177e4SLinus Torvalds * 711da177e4SLinus Torvalds * Doesn't have to be accurate, i.e. may have races. 721da177e4SLinus Torvalds */ 731da177e4SLinus Torvalds unsigned int kobjsize(const void *objp) 741da177e4SLinus Torvalds { 751da177e4SLinus Torvalds struct page *page; 761da177e4SLinus Torvalds 774016a139SMichael Hennerich /* 784016a139SMichael Hennerich * If the object we have should not have ksize performed on it, 794016a139SMichael Hennerich * return size of 0 804016a139SMichael Hennerich */ 815a1603beSPaul Mundt if (!objp || !virt_addr_valid(objp)) 826cfd53fcSPaul Mundt return 0; 836cfd53fcSPaul Mundt 846cfd53fcSPaul Mundt page = virt_to_head_page(objp); 856cfd53fcSPaul Mundt 866cfd53fcSPaul Mundt /* 876cfd53fcSPaul Mundt * If the allocator sets PageSlab, we know the pointer came from 886cfd53fcSPaul Mundt * kmalloc(). 896cfd53fcSPaul Mundt */ 901da177e4SLinus Torvalds if (PageSlab(page)) 911da177e4SLinus Torvalds return ksize(objp); 921da177e4SLinus Torvalds 936cfd53fcSPaul Mundt /* 94ab2e83eaSPaul Mundt * If it's not a compound page, see if we have a matching VMA 95ab2e83eaSPaul Mundt * region. This test is intentionally done in reverse order, 96ab2e83eaSPaul Mundt * so if there's no VMA, we still fall through and hand back 97ab2e83eaSPaul Mundt * PAGE_SIZE for 0-order pages. 98ab2e83eaSPaul Mundt */ 99ab2e83eaSPaul Mundt if (!PageCompound(page)) { 100ab2e83eaSPaul Mundt struct vm_area_struct *vma; 101ab2e83eaSPaul Mundt 102ab2e83eaSPaul Mundt vma = find_vma(current->mm, (unsigned long)objp); 103ab2e83eaSPaul Mundt if (vma) 104ab2e83eaSPaul Mundt return vma->vm_end - vma->vm_start; 105ab2e83eaSPaul Mundt } 106ab2e83eaSPaul Mundt 107ab2e83eaSPaul Mundt /* 1086cfd53fcSPaul Mundt * The ksize() function is only guaranteed to work for pointers 1095a1603beSPaul Mundt * returned by kmalloc(). So handle arbitrary pointers here. 1106cfd53fcSPaul Mundt */ 1115a1603beSPaul Mundt return PAGE_SIZE << compound_order(page); 1121da177e4SLinus Torvalds } 1131da177e4SLinus Torvalds 114dfc2f91aSPaul Mundt /** 115dfc2f91aSPaul Mundt * follow_pfn - look up PFN at a user virtual address 116dfc2f91aSPaul Mundt * @vma: memory mapping 117dfc2f91aSPaul Mundt * @address: user virtual address 118dfc2f91aSPaul Mundt * @pfn: location to store found PFN 119dfc2f91aSPaul Mundt * 120dfc2f91aSPaul Mundt * Only IO mappings and raw PFN mappings are allowed. 121dfc2f91aSPaul Mundt * 122dfc2f91aSPaul Mundt * Returns zero and the pfn at @pfn on success, -ve otherwise. 123dfc2f91aSPaul Mundt */ 124dfc2f91aSPaul Mundt int follow_pfn(struct vm_area_struct *vma, unsigned long address, 125dfc2f91aSPaul Mundt unsigned long *pfn) 126dfc2f91aSPaul Mundt { 127dfc2f91aSPaul Mundt if (!(vma->vm_flags & (VM_IO | VM_PFNMAP))) 128dfc2f91aSPaul Mundt return -EINVAL; 129dfc2f91aSPaul Mundt 130dfc2f91aSPaul Mundt *pfn = address >> PAGE_SHIFT; 131dfc2f91aSPaul Mundt return 0; 132dfc2f91aSPaul Mundt } 133dfc2f91aSPaul Mundt EXPORT_SYMBOL(follow_pfn); 134dfc2f91aSPaul Mundt 135f1c4069eSJoonsoo Kim LIST_HEAD(vmap_area_list); 1361da177e4SLinus Torvalds 137b3bdda02SChristoph Lameter void vfree(const void *addr) 1381da177e4SLinus Torvalds { 1391da177e4SLinus Torvalds kfree(addr); 1401da177e4SLinus Torvalds } 141b5073173SPaul Mundt EXPORT_SYMBOL(vfree); 1421da177e4SLinus Torvalds 143dd0fc66fSAl Viro void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot) 1441da177e4SLinus Torvalds { 1451da177e4SLinus Torvalds /* 1468518609dSRobert P. J. Day * You can't specify __GFP_HIGHMEM with kmalloc() since kmalloc() 1478518609dSRobert P. J. Day * returns only a logical address. 1481da177e4SLinus Torvalds */ 14984097518SNick Piggin return kmalloc(size, (gfp_mask | __GFP_COMP) & ~__GFP_HIGHMEM); 1501da177e4SLinus Torvalds } 151b5073173SPaul Mundt EXPORT_SYMBOL(__vmalloc); 1521da177e4SLinus Torvalds 153a7c3e901SMichal Hocko void *__vmalloc_node_flags(unsigned long size, int node, gfp_t flags) 154a7c3e901SMichal Hocko { 155a7c3e901SMichal Hocko return __vmalloc(size, flags, PAGE_KERNEL); 156a7c3e901SMichal Hocko } 157a7c3e901SMichal Hocko 158f905bc44SPaul Mundt void *vmalloc_user(unsigned long size) 159f905bc44SPaul Mundt { 160f905bc44SPaul Mundt void *ret; 161f905bc44SPaul Mundt 16219809c2dSMichal Hocko ret = __vmalloc(size, GFP_KERNEL | __GFP_ZERO, PAGE_KERNEL); 163f905bc44SPaul Mundt if (ret) { 164f905bc44SPaul Mundt struct vm_area_struct *vma; 165f905bc44SPaul Mundt 166f905bc44SPaul Mundt down_write(¤t->mm->mmap_sem); 167f905bc44SPaul Mundt vma = find_vma(current->mm, (unsigned long)ret); 168f905bc44SPaul Mundt if (vma) 169f905bc44SPaul Mundt vma->vm_flags |= VM_USERMAP; 170f905bc44SPaul Mundt up_write(¤t->mm->mmap_sem); 171f905bc44SPaul Mundt } 172f905bc44SPaul Mundt 173f905bc44SPaul Mundt return ret; 174f905bc44SPaul Mundt } 175f905bc44SPaul Mundt EXPORT_SYMBOL(vmalloc_user); 176f905bc44SPaul Mundt 177b3bdda02SChristoph Lameter struct page *vmalloc_to_page(const void *addr) 1781da177e4SLinus Torvalds { 1791da177e4SLinus Torvalds return virt_to_page(addr); 1801da177e4SLinus Torvalds } 181b5073173SPaul Mundt EXPORT_SYMBOL(vmalloc_to_page); 1821da177e4SLinus Torvalds 183b3bdda02SChristoph Lameter unsigned long vmalloc_to_pfn(const void *addr) 1841da177e4SLinus Torvalds { 1851da177e4SLinus Torvalds return page_to_pfn(virt_to_page(addr)); 1861da177e4SLinus Torvalds } 187b5073173SPaul Mundt EXPORT_SYMBOL(vmalloc_to_pfn); 1881da177e4SLinus Torvalds 1891da177e4SLinus Torvalds long vread(char *buf, char *addr, unsigned long count) 1901da177e4SLinus Torvalds { 1919bde916bSChen Gang /* Don't allow overflow */ 1929bde916bSChen Gang if ((unsigned long) buf + count < count) 1939bde916bSChen Gang count = -(unsigned long) buf; 1949bde916bSChen Gang 1951da177e4SLinus Torvalds memcpy(buf, addr, count); 1961da177e4SLinus Torvalds return count; 1971da177e4SLinus Torvalds } 1981da177e4SLinus Torvalds 1991da177e4SLinus Torvalds long vwrite(char *buf, char *addr, unsigned long count) 2001da177e4SLinus Torvalds { 2011da177e4SLinus Torvalds /* Don't allow overflow */ 2021da177e4SLinus Torvalds if ((unsigned long) addr + count < count) 2031da177e4SLinus Torvalds count = -(unsigned long) addr; 2041da177e4SLinus Torvalds 2051da177e4SLinus Torvalds memcpy(addr, buf, count); 206ac714904SChoi Gi-yong return count; 2071da177e4SLinus Torvalds } 2081da177e4SLinus Torvalds 2091da177e4SLinus Torvalds /* 210e1c05067SMasahiro Yamada * vmalloc - allocate virtually contiguous memory 2111da177e4SLinus Torvalds * 2121da177e4SLinus Torvalds * @size: allocation size 2131da177e4SLinus Torvalds * 2141da177e4SLinus Torvalds * Allocate enough pages to cover @size from the page level 215e1c05067SMasahiro Yamada * allocator and map them into contiguous kernel virtual space. 2161da177e4SLinus Torvalds * 217c1c8897fSMichael Opdenacker * For tight control over page level allocator and protection flags 2181da177e4SLinus Torvalds * use __vmalloc() instead. 2191da177e4SLinus Torvalds */ 2201da177e4SLinus Torvalds void *vmalloc(unsigned long size) 2211da177e4SLinus Torvalds { 2221da177e4SLinus Torvalds return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL); 2231da177e4SLinus Torvalds } 224f6138882SAndrew Morton EXPORT_SYMBOL(vmalloc); 225f6138882SAndrew Morton 226e1ca7788SDave Young /* 227e1c05067SMasahiro Yamada * vzalloc - allocate virtually contiguous memory with zero fill 228e1ca7788SDave Young * 229e1ca7788SDave Young * @size: allocation size 230e1ca7788SDave Young * 231e1ca7788SDave Young * Allocate enough pages to cover @size from the page level 232e1c05067SMasahiro Yamada * allocator and map them into contiguous kernel virtual space. 233e1ca7788SDave Young * The memory allocated is set to zero. 234e1ca7788SDave Young * 235e1ca7788SDave Young * For tight control over page level allocator and protection flags 236e1ca7788SDave Young * use __vmalloc() instead. 237e1ca7788SDave Young */ 238e1ca7788SDave Young void *vzalloc(unsigned long size) 239e1ca7788SDave Young { 240e1ca7788SDave Young return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO, 241e1ca7788SDave Young PAGE_KERNEL); 242e1ca7788SDave Young } 243e1ca7788SDave Young EXPORT_SYMBOL(vzalloc); 244e1ca7788SDave Young 245e1ca7788SDave Young /** 246e1ca7788SDave Young * vmalloc_node - allocate memory on a specific node 247e1ca7788SDave Young * @size: allocation size 248e1ca7788SDave Young * @node: numa node 249e1ca7788SDave Young * 250e1ca7788SDave Young * Allocate enough pages to cover @size from the page level 251e1ca7788SDave Young * allocator and map them into contiguous kernel virtual space. 252e1ca7788SDave Young * 253e1ca7788SDave Young * For tight control over page level allocator and protection flags 254e1ca7788SDave Young * use __vmalloc() instead. 255e1ca7788SDave Young */ 256f6138882SAndrew Morton void *vmalloc_node(unsigned long size, int node) 257f6138882SAndrew Morton { 258f6138882SAndrew Morton return vmalloc(size); 259f6138882SAndrew Morton } 2609a14f653SPaul Mundt EXPORT_SYMBOL(vmalloc_node); 261e1ca7788SDave Young 262e1ca7788SDave Young /** 263e1ca7788SDave Young * vzalloc_node - allocate memory on a specific node with zero fill 264e1ca7788SDave Young * @size: allocation size 265e1ca7788SDave Young * @node: numa node 266e1ca7788SDave Young * 267e1ca7788SDave Young * Allocate enough pages to cover @size from the page level 268e1ca7788SDave Young * allocator and map them into contiguous kernel virtual space. 269e1ca7788SDave Young * The memory allocated is set to zero. 270e1ca7788SDave Young * 271e1ca7788SDave Young * For tight control over page level allocator and protection flags 272e1ca7788SDave Young * use __vmalloc() instead. 273e1ca7788SDave Young */ 274e1ca7788SDave Young void *vzalloc_node(unsigned long size, int node) 275e1ca7788SDave Young { 276e1ca7788SDave Young return vzalloc(size); 277e1ca7788SDave Young } 278e1ca7788SDave Young EXPORT_SYMBOL(vzalloc_node); 2791da177e4SLinus Torvalds 2801af446edSPaul Mundt /** 2811af446edSPaul Mundt * vmalloc_exec - allocate virtually contiguous, executable memory 2821af446edSPaul Mundt * @size: allocation size 2831af446edSPaul Mundt * 2841af446edSPaul Mundt * Kernel-internal function to allocate enough pages to cover @size 2851af446edSPaul Mundt * the page level allocator and map them into contiguous and 2861af446edSPaul Mundt * executable kernel virtual space. 2871af446edSPaul Mundt * 2881af446edSPaul Mundt * For tight control over page level allocator and protection flags 2891af446edSPaul Mundt * use __vmalloc() instead. 2901af446edSPaul Mundt */ 2911af446edSPaul Mundt 2921af446edSPaul Mundt void *vmalloc_exec(unsigned long size) 2931af446edSPaul Mundt { 2941af446edSPaul Mundt return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC); 2951af446edSPaul Mundt } 2961af446edSPaul Mundt 297b5073173SPaul Mundt /** 298b5073173SPaul Mundt * vmalloc_32 - allocate virtually contiguous memory (32bit addressable) 2991da177e4SLinus Torvalds * @size: allocation size 3001da177e4SLinus Torvalds * 3011da177e4SLinus Torvalds * Allocate enough 32bit PA addressable pages to cover @size from the 302e1c05067SMasahiro Yamada * page level allocator and map them into contiguous kernel virtual space. 3031da177e4SLinus Torvalds */ 3041da177e4SLinus Torvalds void *vmalloc_32(unsigned long size) 3051da177e4SLinus Torvalds { 3061da177e4SLinus Torvalds return __vmalloc(size, GFP_KERNEL, PAGE_KERNEL); 3071da177e4SLinus Torvalds } 308b5073173SPaul Mundt EXPORT_SYMBOL(vmalloc_32); 309b5073173SPaul Mundt 310b5073173SPaul Mundt /** 311b5073173SPaul Mundt * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory 312b5073173SPaul Mundt * @size: allocation size 313b5073173SPaul Mundt * 314b5073173SPaul Mundt * The resulting memory area is 32bit addressable and zeroed so it can be 315b5073173SPaul Mundt * mapped to userspace without leaking data. 316f905bc44SPaul Mundt * 317f905bc44SPaul Mundt * VM_USERMAP is set on the corresponding VMA so that subsequent calls to 318f905bc44SPaul Mundt * remap_vmalloc_range() are permissible. 319b5073173SPaul Mundt */ 320b5073173SPaul Mundt void *vmalloc_32_user(unsigned long size) 321b5073173SPaul Mundt { 322f905bc44SPaul Mundt /* 323f905bc44SPaul Mundt * We'll have to sort out the ZONE_DMA bits for 64-bit, 324f905bc44SPaul Mundt * but for now this can simply use vmalloc_user() directly. 325f905bc44SPaul Mundt */ 326f905bc44SPaul Mundt return vmalloc_user(size); 327b5073173SPaul Mundt } 328b5073173SPaul Mundt EXPORT_SYMBOL(vmalloc_32_user); 3291da177e4SLinus Torvalds 3301da177e4SLinus Torvalds void *vmap(struct page **pages, unsigned int count, unsigned long flags, pgprot_t prot) 3311da177e4SLinus Torvalds { 3321da177e4SLinus Torvalds BUG(); 3331da177e4SLinus Torvalds return NULL; 3341da177e4SLinus Torvalds } 335b5073173SPaul Mundt EXPORT_SYMBOL(vmap); 3361da177e4SLinus Torvalds 337b3bdda02SChristoph Lameter void vunmap(const void *addr) 3381da177e4SLinus Torvalds { 3391da177e4SLinus Torvalds BUG(); 3401da177e4SLinus Torvalds } 341b5073173SPaul Mundt EXPORT_SYMBOL(vunmap); 3421da177e4SLinus Torvalds 343eb6434d9SPaul Mundt void *vm_map_ram(struct page **pages, unsigned int count, int node, pgprot_t prot) 344eb6434d9SPaul Mundt { 345eb6434d9SPaul Mundt BUG(); 346eb6434d9SPaul Mundt return NULL; 347eb6434d9SPaul Mundt } 348eb6434d9SPaul Mundt EXPORT_SYMBOL(vm_map_ram); 349eb6434d9SPaul Mundt 350eb6434d9SPaul Mundt void vm_unmap_ram(const void *mem, unsigned int count) 351eb6434d9SPaul Mundt { 352eb6434d9SPaul Mundt BUG(); 353eb6434d9SPaul Mundt } 354eb6434d9SPaul Mundt EXPORT_SYMBOL(vm_unmap_ram); 355eb6434d9SPaul Mundt 356eb6434d9SPaul Mundt void vm_unmap_aliases(void) 357eb6434d9SPaul Mundt { 358eb6434d9SPaul Mundt } 359eb6434d9SPaul Mundt EXPORT_SYMBOL_GPL(vm_unmap_aliases); 360eb6434d9SPaul Mundt 3611da177e4SLinus Torvalds /* 3621eeb66a1SChristoph Hellwig * Implement a stub for vmalloc_sync_all() if the architecture chose not to 3631eeb66a1SChristoph Hellwig * have one. 3641eeb66a1SChristoph Hellwig */ 3653b32123dSGideon Israel Dsouza void __weak vmalloc_sync_all(void) 3661eeb66a1SChristoph Hellwig { 3671eeb66a1SChristoph Hellwig } 3681eeb66a1SChristoph Hellwig 369cd12909cSDavid Vrabel struct vm_struct *alloc_vm_area(size_t size, pte_t **ptes) 37029c185e5SPaul Mundt { 37129c185e5SPaul Mundt BUG(); 37229c185e5SPaul Mundt return NULL; 37329c185e5SPaul Mundt } 37429c185e5SPaul Mundt EXPORT_SYMBOL_GPL(alloc_vm_area); 37529c185e5SPaul Mundt 37629c185e5SPaul Mundt void free_vm_area(struct vm_struct *area) 37729c185e5SPaul Mundt { 37829c185e5SPaul Mundt BUG(); 37929c185e5SPaul Mundt } 38029c185e5SPaul Mundt EXPORT_SYMBOL_GPL(free_vm_area); 38129c185e5SPaul Mundt 382b5073173SPaul Mundt int vm_insert_page(struct vm_area_struct *vma, unsigned long addr, 383b5073173SPaul Mundt struct page *page) 384b5073173SPaul Mundt { 385b5073173SPaul Mundt return -EINVAL; 386b5073173SPaul Mundt } 387b5073173SPaul Mundt EXPORT_SYMBOL(vm_insert_page); 388b5073173SPaul Mundt 389a667d745SSouptick Joarder int vm_map_pages(struct vm_area_struct *vma, struct page **pages, 390a667d745SSouptick Joarder unsigned long num) 391a667d745SSouptick Joarder { 392a667d745SSouptick Joarder return -EINVAL; 393a667d745SSouptick Joarder } 394a667d745SSouptick Joarder EXPORT_SYMBOL(vm_map_pages); 395a667d745SSouptick Joarder 396a667d745SSouptick Joarder int vm_map_pages_zero(struct vm_area_struct *vma, struct page **pages, 397a667d745SSouptick Joarder unsigned long num) 398a667d745SSouptick Joarder { 399a667d745SSouptick Joarder return -EINVAL; 400a667d745SSouptick Joarder } 401a667d745SSouptick Joarder EXPORT_SYMBOL(vm_map_pages_zero); 402a667d745SSouptick Joarder 4031eeb66a1SChristoph Hellwig /* 4041da177e4SLinus Torvalds * sys_brk() for the most part doesn't need the global kernel 4051da177e4SLinus Torvalds * lock, except when an application is doing something nasty 4061da177e4SLinus Torvalds * like trying to un-brk an area that has already been mapped 4071da177e4SLinus Torvalds * to a regular file. in this case, the unmapping will need 4081da177e4SLinus Torvalds * to invoke file system routines that need the global lock. 4091da177e4SLinus Torvalds */ 4106a6160a7SHeiko Carstens SYSCALL_DEFINE1(brk, unsigned long, brk) 4111da177e4SLinus Torvalds { 4121da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 4131da177e4SLinus Torvalds 4141da177e4SLinus Torvalds if (brk < mm->start_brk || brk > mm->context.end_brk) 4151da177e4SLinus Torvalds return mm->brk; 4161da177e4SLinus Torvalds 4171da177e4SLinus Torvalds if (mm->brk == brk) 4181da177e4SLinus Torvalds return mm->brk; 4191da177e4SLinus Torvalds 4201da177e4SLinus Torvalds /* 4211da177e4SLinus Torvalds * Always allow shrinking brk 4221da177e4SLinus Torvalds */ 4231da177e4SLinus Torvalds if (brk <= mm->brk) { 4241da177e4SLinus Torvalds mm->brk = brk; 4251da177e4SLinus Torvalds return brk; 4261da177e4SLinus Torvalds } 4271da177e4SLinus Torvalds 4281da177e4SLinus Torvalds /* 4291da177e4SLinus Torvalds * Ok, looks good - let it rip. 4301da177e4SLinus Torvalds */ 431cfe79c00SMike Frysinger flush_icache_range(mm->brk, brk); 4321da177e4SLinus Torvalds return mm->brk = brk; 4331da177e4SLinus Torvalds } 4341da177e4SLinus Torvalds 4358feae131SDavid Howells /* 4363edf41d8Sseokhoon.yoon * initialise the percpu counter for VM and region record slabs 4378feae131SDavid Howells */ 4388feae131SDavid Howells void __init mmap_init(void) 4391da177e4SLinus Torvalds { 44000a62ce9SKOSAKI Motohiro int ret; 44100a62ce9SKOSAKI Motohiro 442908c7f19STejun Heo ret = percpu_counter_init(&vm_committed_as, 0, GFP_KERNEL); 44300a62ce9SKOSAKI Motohiro VM_BUG_ON(ret); 4445d097056SVladimir Davydov vm_region_jar = KMEM_CACHE(vm_region, SLAB_PANIC|SLAB_ACCOUNT); 4458feae131SDavid Howells } 4461da177e4SLinus Torvalds 4478feae131SDavid Howells /* 4488feae131SDavid Howells * validate the region tree 4498feae131SDavid Howells * - the caller must hold the region lock 4508feae131SDavid Howells */ 4518feae131SDavid Howells #ifdef CONFIG_DEBUG_NOMMU_REGIONS 4528feae131SDavid Howells static noinline void validate_nommu_regions(void) 4538feae131SDavid Howells { 4548feae131SDavid Howells struct vm_region *region, *last; 4558feae131SDavid Howells struct rb_node *p, *lastp; 4561da177e4SLinus Torvalds 4578feae131SDavid Howells lastp = rb_first(&nommu_region_tree); 4588feae131SDavid Howells if (!lastp) 4598feae131SDavid Howells return; 4608feae131SDavid Howells 4618feae131SDavid Howells last = rb_entry(lastp, struct vm_region, vm_rb); 462c9427bc0SGeliang Tang BUG_ON(last->vm_end <= last->vm_start); 463c9427bc0SGeliang Tang BUG_ON(last->vm_top < last->vm_end); 4648feae131SDavid Howells 4658feae131SDavid Howells while ((p = rb_next(lastp))) { 4668feae131SDavid Howells region = rb_entry(p, struct vm_region, vm_rb); 4678feae131SDavid Howells last = rb_entry(lastp, struct vm_region, vm_rb); 4688feae131SDavid Howells 469c9427bc0SGeliang Tang BUG_ON(region->vm_end <= region->vm_start); 470c9427bc0SGeliang Tang BUG_ON(region->vm_top < region->vm_end); 471c9427bc0SGeliang Tang BUG_ON(region->vm_start < last->vm_top); 4728feae131SDavid Howells 4738feae131SDavid Howells lastp = p; 4741da177e4SLinus Torvalds } 4751da177e4SLinus Torvalds } 4768feae131SDavid Howells #else 47733e5d769SDavid Howells static void validate_nommu_regions(void) 47833e5d769SDavid Howells { 47933e5d769SDavid Howells } 4808feae131SDavid Howells #endif 4818feae131SDavid Howells 4828feae131SDavid Howells /* 4838feae131SDavid Howells * add a region into the global tree 4848feae131SDavid Howells */ 4858feae131SDavid Howells static void add_nommu_region(struct vm_region *region) 4868feae131SDavid Howells { 4878feae131SDavid Howells struct vm_region *pregion; 4888feae131SDavid Howells struct rb_node **p, *parent; 4898feae131SDavid Howells 4908feae131SDavid Howells validate_nommu_regions(); 4918feae131SDavid Howells 4928feae131SDavid Howells parent = NULL; 4938feae131SDavid Howells p = &nommu_region_tree.rb_node; 4948feae131SDavid Howells while (*p) { 4958feae131SDavid Howells parent = *p; 4968feae131SDavid Howells pregion = rb_entry(parent, struct vm_region, vm_rb); 4978feae131SDavid Howells if (region->vm_start < pregion->vm_start) 4988feae131SDavid Howells p = &(*p)->rb_left; 4998feae131SDavid Howells else if (region->vm_start > pregion->vm_start) 5008feae131SDavid Howells p = &(*p)->rb_right; 5018feae131SDavid Howells else if (pregion == region) 5028feae131SDavid Howells return; 5038feae131SDavid Howells else 5048feae131SDavid Howells BUG(); 5058feae131SDavid Howells } 5068feae131SDavid Howells 5078feae131SDavid Howells rb_link_node(®ion->vm_rb, parent, p); 5088feae131SDavid Howells rb_insert_color(®ion->vm_rb, &nommu_region_tree); 5098feae131SDavid Howells 5108feae131SDavid Howells validate_nommu_regions(); 5118feae131SDavid Howells } 5128feae131SDavid Howells 5138feae131SDavid Howells /* 5148feae131SDavid Howells * delete a region from the global tree 5158feae131SDavid Howells */ 5168feae131SDavid Howells static void delete_nommu_region(struct vm_region *region) 5178feae131SDavid Howells { 5188feae131SDavid Howells BUG_ON(!nommu_region_tree.rb_node); 5198feae131SDavid Howells 5208feae131SDavid Howells validate_nommu_regions(); 5218feae131SDavid Howells rb_erase(®ion->vm_rb, &nommu_region_tree); 5228feae131SDavid Howells validate_nommu_regions(); 5238feae131SDavid Howells } 5248feae131SDavid Howells 5258feae131SDavid Howells /* 5268feae131SDavid Howells * free a contiguous series of pages 5278feae131SDavid Howells */ 5288feae131SDavid Howells static void free_page_series(unsigned long from, unsigned long to) 5298feae131SDavid Howells { 5308feae131SDavid Howells for (; from < to; from += PAGE_SIZE) { 5318feae131SDavid Howells struct page *page = virt_to_page(from); 5328feae131SDavid Howells 53333e5d769SDavid Howells atomic_long_dec(&mmap_pages_allocated); 5348feae131SDavid Howells put_page(page); 5358feae131SDavid Howells } 5368feae131SDavid Howells } 5378feae131SDavid Howells 5388feae131SDavid Howells /* 5398feae131SDavid Howells * release a reference to a region 54033e5d769SDavid Howells * - the caller must hold the region semaphore for writing, which this releases 541dd8632a1SPaul Mundt * - the region may not have been added to the tree yet, in which case vm_top 5428feae131SDavid Howells * will equal vm_start 5438feae131SDavid Howells */ 5448feae131SDavid Howells static void __put_nommu_region(struct vm_region *region) 5458feae131SDavid Howells __releases(nommu_region_sem) 5468feae131SDavid Howells { 5478feae131SDavid Howells BUG_ON(!nommu_region_tree.rb_node); 5488feae131SDavid Howells 5491e2ae599SDavid Howells if (--region->vm_usage == 0) { 550dd8632a1SPaul Mundt if (region->vm_top > region->vm_start) 5518feae131SDavid Howells delete_nommu_region(region); 5528feae131SDavid Howells up_write(&nommu_region_sem); 5538feae131SDavid Howells 5548feae131SDavid Howells if (region->vm_file) 5558feae131SDavid Howells fput(region->vm_file); 5568feae131SDavid Howells 5578feae131SDavid Howells /* IO memory and memory shared directly out of the pagecache 5588feae131SDavid Howells * from ramfs/tmpfs mustn't be released here */ 55922cc877bSLeon Romanovsky if (region->vm_flags & VM_MAPPED_COPY) 560dd8632a1SPaul Mundt free_page_series(region->vm_start, region->vm_top); 5618feae131SDavid Howells kmem_cache_free(vm_region_jar, region); 5628feae131SDavid Howells } else { 5638feae131SDavid Howells up_write(&nommu_region_sem); 5648feae131SDavid Howells } 5658feae131SDavid Howells } 5668feae131SDavid Howells 5678feae131SDavid Howells /* 5688feae131SDavid Howells * release a reference to a region 5698feae131SDavid Howells */ 5708feae131SDavid Howells static void put_nommu_region(struct vm_region *region) 5718feae131SDavid Howells { 5728feae131SDavid Howells down_write(&nommu_region_sem); 5738feae131SDavid Howells __put_nommu_region(region); 5748feae131SDavid Howells } 5751da177e4SLinus Torvalds 5763034097aSDavid Howells /* 5773034097aSDavid Howells * add a VMA into a process's mm_struct in the appropriate place in the list 5788feae131SDavid Howells * and tree and add to the address space's page tree also if not an anonymous 5798feae131SDavid Howells * page 5803034097aSDavid Howells * - should be called with mm->mmap_sem held writelocked 5813034097aSDavid Howells */ 5828feae131SDavid Howells static void add_vma_to_mm(struct mm_struct *mm, struct vm_area_struct *vma) 5833034097aSDavid Howells { 5846038def0SNamhyung Kim struct vm_area_struct *pvma, *prev; 5858feae131SDavid Howells struct address_space *mapping; 5866038def0SNamhyung Kim struct rb_node **p, *parent, *rb_prev; 5873034097aSDavid Howells 5888feae131SDavid Howells BUG_ON(!vma->vm_region); 5898feae131SDavid Howells 5908feae131SDavid Howells mm->map_count++; 5918feae131SDavid Howells vma->vm_mm = mm; 5928feae131SDavid Howells 5938feae131SDavid Howells /* add the VMA to the mapping */ 5948feae131SDavid Howells if (vma->vm_file) { 5958feae131SDavid Howells mapping = vma->vm_file->f_mapping; 5968feae131SDavid Howells 59783cde9e8SDavidlohr Bueso i_mmap_lock_write(mapping); 5988feae131SDavid Howells flush_dcache_mmap_lock(mapping); 5996b2dbba8SMichel Lespinasse vma_interval_tree_insert(vma, &mapping->i_mmap); 6008feae131SDavid Howells flush_dcache_mmap_unlock(mapping); 60183cde9e8SDavidlohr Bueso i_mmap_unlock_write(mapping); 6028feae131SDavid Howells } 6038feae131SDavid Howells 6048feae131SDavid Howells /* add the VMA to the tree */ 6056038def0SNamhyung Kim parent = rb_prev = NULL; 6068feae131SDavid Howells p = &mm->mm_rb.rb_node; 6078feae131SDavid Howells while (*p) { 6088feae131SDavid Howells parent = *p; 6098feae131SDavid Howells pvma = rb_entry(parent, struct vm_area_struct, vm_rb); 6108feae131SDavid Howells 6118feae131SDavid Howells /* sort by: start addr, end addr, VMA struct addr in that order 6128feae131SDavid Howells * (the latter is necessary as we may get identical VMAs) */ 6138feae131SDavid Howells if (vma->vm_start < pvma->vm_start) 6148feae131SDavid Howells p = &(*p)->rb_left; 6156038def0SNamhyung Kim else if (vma->vm_start > pvma->vm_start) { 6166038def0SNamhyung Kim rb_prev = parent; 6178feae131SDavid Howells p = &(*p)->rb_right; 6186038def0SNamhyung Kim } else if (vma->vm_end < pvma->vm_end) 6198feae131SDavid Howells p = &(*p)->rb_left; 6206038def0SNamhyung Kim else if (vma->vm_end > pvma->vm_end) { 6216038def0SNamhyung Kim rb_prev = parent; 6228feae131SDavid Howells p = &(*p)->rb_right; 6236038def0SNamhyung Kim } else if (vma < pvma) 6248feae131SDavid Howells p = &(*p)->rb_left; 6256038def0SNamhyung Kim else if (vma > pvma) { 6266038def0SNamhyung Kim rb_prev = parent; 6278feae131SDavid Howells p = &(*p)->rb_right; 6286038def0SNamhyung Kim } else 6298feae131SDavid Howells BUG(); 6308feae131SDavid Howells } 6318feae131SDavid Howells 6328feae131SDavid Howells rb_link_node(&vma->vm_rb, parent, p); 6338feae131SDavid Howells rb_insert_color(&vma->vm_rb, &mm->mm_rb); 6348feae131SDavid Howells 6358feae131SDavid Howells /* add VMA to the VMA list also */ 6366038def0SNamhyung Kim prev = NULL; 6376038def0SNamhyung Kim if (rb_prev) 6386038def0SNamhyung Kim prev = rb_entry(rb_prev, struct vm_area_struct, vm_rb); 6393034097aSDavid Howells 6406038def0SNamhyung Kim __vma_link_list(mm, vma, prev, parent); 6418feae131SDavid Howells } 6428feae131SDavid Howells 6438feae131SDavid Howells /* 6448feae131SDavid Howells * delete a VMA from its owning mm_struct and address space 6458feae131SDavid Howells */ 6468feae131SDavid Howells static void delete_vma_from_mm(struct vm_area_struct *vma) 6478feae131SDavid Howells { 648615d6e87SDavidlohr Bueso int i; 6498feae131SDavid Howells struct address_space *mapping; 6508feae131SDavid Howells struct mm_struct *mm = vma->vm_mm; 651615d6e87SDavidlohr Bueso struct task_struct *curr = current; 6528feae131SDavid Howells 6538feae131SDavid Howells mm->map_count--; 654615d6e87SDavidlohr Bueso for (i = 0; i < VMACACHE_SIZE; i++) { 655615d6e87SDavidlohr Bueso /* if the vma is cached, invalidate the entire cache */ 656314ff785SIngo Molnar if (curr->vmacache.vmas[i] == vma) { 657e020d5bdSSteven Miao vmacache_invalidate(mm); 658615d6e87SDavidlohr Bueso break; 659615d6e87SDavidlohr Bueso } 660615d6e87SDavidlohr Bueso } 6618feae131SDavid Howells 6628feae131SDavid Howells /* remove the VMA from the mapping */ 6638feae131SDavid Howells if (vma->vm_file) { 6648feae131SDavid Howells mapping = vma->vm_file->f_mapping; 6658feae131SDavid Howells 66683cde9e8SDavidlohr Bueso i_mmap_lock_write(mapping); 6678feae131SDavid Howells flush_dcache_mmap_lock(mapping); 6686b2dbba8SMichel Lespinasse vma_interval_tree_remove(vma, &mapping->i_mmap); 6698feae131SDavid Howells flush_dcache_mmap_unlock(mapping); 67083cde9e8SDavidlohr Bueso i_mmap_unlock_write(mapping); 6718feae131SDavid Howells } 6728feae131SDavid Howells 6738feae131SDavid Howells /* remove from the MM's tree and list */ 6748feae131SDavid Howells rb_erase(&vma->vm_rb, &mm->mm_rb); 675b951bf2cSNamhyung Kim 676b951bf2cSNamhyung Kim if (vma->vm_prev) 677b951bf2cSNamhyung Kim vma->vm_prev->vm_next = vma->vm_next; 678b951bf2cSNamhyung Kim else 679b951bf2cSNamhyung Kim mm->mmap = vma->vm_next; 680b951bf2cSNamhyung Kim 681b951bf2cSNamhyung Kim if (vma->vm_next) 682b951bf2cSNamhyung Kim vma->vm_next->vm_prev = vma->vm_prev; 6838feae131SDavid Howells } 6848feae131SDavid Howells 6858feae131SDavid Howells /* 6868feae131SDavid Howells * destroy a VMA record 6878feae131SDavid Howells */ 6888feae131SDavid Howells static void delete_vma(struct mm_struct *mm, struct vm_area_struct *vma) 6898feae131SDavid Howells { 6908feae131SDavid Howells if (vma->vm_ops && vma->vm_ops->close) 6918feae131SDavid Howells vma->vm_ops->close(vma); 692e9714acfSKonstantin Khlebnikov if (vma->vm_file) 6938feae131SDavid Howells fput(vma->vm_file); 6948feae131SDavid Howells put_nommu_region(vma->vm_region); 6953928d4f5SLinus Torvalds vm_area_free(vma); 6963034097aSDavid Howells } 6973034097aSDavid Howells 6983034097aSDavid Howells /* 6993034097aSDavid Howells * look up the first VMA in which addr resides, NULL if none 7003034097aSDavid Howells * - should be called with mm->mmap_sem at least held readlocked 7013034097aSDavid Howells */ 7023034097aSDavid Howells struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr) 7033034097aSDavid Howells { 7048feae131SDavid Howells struct vm_area_struct *vma; 7053034097aSDavid Howells 7068feae131SDavid Howells /* check the cache first */ 707615d6e87SDavidlohr Bueso vma = vmacache_find(mm, addr); 708615d6e87SDavidlohr Bueso if (likely(vma)) 7098feae131SDavid Howells return vma; 7108feae131SDavid Howells 711e922c4c5SNamhyung Kim /* trawl the list (there may be multiple mappings in which addr 7128feae131SDavid Howells * resides) */ 713e922c4c5SNamhyung Kim for (vma = mm->mmap; vma; vma = vma->vm_next) { 7148feae131SDavid Howells if (vma->vm_start > addr) 7158feae131SDavid Howells return NULL; 7168feae131SDavid Howells if (vma->vm_end > addr) { 717615d6e87SDavidlohr Bueso vmacache_update(addr, vma); 7188feae131SDavid Howells return vma; 7193034097aSDavid Howells } 7208feae131SDavid Howells } 7213034097aSDavid Howells 7223034097aSDavid Howells return NULL; 7233034097aSDavid Howells } 7243034097aSDavid Howells EXPORT_SYMBOL(find_vma); 7253034097aSDavid Howells 7263034097aSDavid Howells /* 727930e652aSDavid Howells * find a VMA 728930e652aSDavid Howells * - we don't extend stack VMAs under NOMMU conditions 729930e652aSDavid Howells */ 730930e652aSDavid Howells struct vm_area_struct *find_extend_vma(struct mm_struct *mm, unsigned long addr) 731930e652aSDavid Howells { 7327561e8caSDavid Howells return find_vma(mm, addr); 733930e652aSDavid Howells } 734930e652aSDavid Howells 7358feae131SDavid Howells /* 7368feae131SDavid Howells * expand a stack to a given address 7378feae131SDavid Howells * - not supported under NOMMU conditions 7388feae131SDavid Howells */ 73957c8f63eSGreg Ungerer int expand_stack(struct vm_area_struct *vma, unsigned long address) 74057c8f63eSGreg Ungerer { 74157c8f63eSGreg Ungerer return -ENOMEM; 74257c8f63eSGreg Ungerer } 74357c8f63eSGreg Ungerer 744930e652aSDavid Howells /* 7456fa5f80bSDavid Howells * look up the first VMA exactly that exactly matches addr 7466fa5f80bSDavid Howells * - should be called with mm->mmap_sem at least held readlocked 7476fa5f80bSDavid Howells */ 7488feae131SDavid Howells static struct vm_area_struct *find_vma_exact(struct mm_struct *mm, 7498feae131SDavid Howells unsigned long addr, 7508feae131SDavid Howells unsigned long len) 7511da177e4SLinus Torvalds { 7521da177e4SLinus Torvalds struct vm_area_struct *vma; 7538feae131SDavid Howells unsigned long end = addr + len; 7541da177e4SLinus Torvalds 7558feae131SDavid Howells /* check the cache first */ 756615d6e87SDavidlohr Bueso vma = vmacache_find_exact(mm, addr, end); 757615d6e87SDavidlohr Bueso if (vma) 7581da177e4SLinus Torvalds return vma; 7598feae131SDavid Howells 760e922c4c5SNamhyung Kim /* trawl the list (there may be multiple mappings in which addr 7618feae131SDavid Howells * resides) */ 762e922c4c5SNamhyung Kim for (vma = mm->mmap; vma; vma = vma->vm_next) { 7638feae131SDavid Howells if (vma->vm_start < addr) 7648feae131SDavid Howells continue; 7658feae131SDavid Howells if (vma->vm_start > addr) 7668feae131SDavid Howells return NULL; 7678feae131SDavid Howells if (vma->vm_end == end) { 768615d6e87SDavidlohr Bueso vmacache_update(addr, vma); 7698feae131SDavid Howells return vma; 7708feae131SDavid Howells } 7711da177e4SLinus Torvalds } 7721da177e4SLinus Torvalds 7731da177e4SLinus Torvalds return NULL; 7741da177e4SLinus Torvalds } 7751da177e4SLinus Torvalds 7763034097aSDavid Howells /* 7771da177e4SLinus Torvalds * determine whether a mapping should be permitted and, if so, what sort of 7781da177e4SLinus Torvalds * mapping we're capable of supporting 7791da177e4SLinus Torvalds */ 7801da177e4SLinus Torvalds static int validate_mmap_request(struct file *file, 7811da177e4SLinus Torvalds unsigned long addr, 7821da177e4SLinus Torvalds unsigned long len, 7831da177e4SLinus Torvalds unsigned long prot, 7841da177e4SLinus Torvalds unsigned long flags, 7851da177e4SLinus Torvalds unsigned long pgoff, 7861da177e4SLinus Torvalds unsigned long *_capabilities) 7871da177e4SLinus Torvalds { 7888feae131SDavid Howells unsigned long capabilities, rlen; 7891da177e4SLinus Torvalds int ret; 7901da177e4SLinus Torvalds 7911da177e4SLinus Torvalds /* do the simple checks first */ 79222cc877bSLeon Romanovsky if (flags & MAP_FIXED) 7931da177e4SLinus Torvalds return -EINVAL; 7941da177e4SLinus Torvalds 7951da177e4SLinus Torvalds if ((flags & MAP_TYPE) != MAP_PRIVATE && 7961da177e4SLinus Torvalds (flags & MAP_TYPE) != MAP_SHARED) 7971da177e4SLinus Torvalds return -EINVAL; 7981da177e4SLinus Torvalds 799f81cff0dSMike Frysinger if (!len) 8001da177e4SLinus Torvalds return -EINVAL; 8011da177e4SLinus Torvalds 802f81cff0dSMike Frysinger /* Careful about overflows.. */ 8038feae131SDavid Howells rlen = PAGE_ALIGN(len); 8048feae131SDavid Howells if (!rlen || rlen > TASK_SIZE) 805f81cff0dSMike Frysinger return -ENOMEM; 806f81cff0dSMike Frysinger 8071da177e4SLinus Torvalds /* offset overflow? */ 8088feae131SDavid Howells if ((pgoff + (rlen >> PAGE_SHIFT)) < pgoff) 809f81cff0dSMike Frysinger return -EOVERFLOW; 8101da177e4SLinus Torvalds 8111da177e4SLinus Torvalds if (file) { 8121da177e4SLinus Torvalds /* files must support mmap */ 81372c2d531SAl Viro if (!file->f_op->mmap) 8141da177e4SLinus Torvalds return -ENODEV; 8151da177e4SLinus Torvalds 8161da177e4SLinus Torvalds /* work out if what we've got could possibly be shared 8171da177e4SLinus Torvalds * - we support chardevs that provide their own "memory" 8181da177e4SLinus Torvalds * - we support files/blockdevs that are memory backed 8191da177e4SLinus Torvalds */ 820b4caecd4SChristoph Hellwig if (file->f_op->mmap_capabilities) { 821b4caecd4SChristoph Hellwig capabilities = file->f_op->mmap_capabilities(file); 822b4caecd4SChristoph Hellwig } else { 8231da177e4SLinus Torvalds /* no explicit capabilities set, so assume some 8241da177e4SLinus Torvalds * defaults */ 825496ad9aaSAl Viro switch (file_inode(file)->i_mode & S_IFMT) { 8261da177e4SLinus Torvalds case S_IFREG: 8271da177e4SLinus Torvalds case S_IFBLK: 828b4caecd4SChristoph Hellwig capabilities = NOMMU_MAP_COPY; 8291da177e4SLinus Torvalds break; 8301da177e4SLinus Torvalds 8311da177e4SLinus Torvalds case S_IFCHR: 8321da177e4SLinus Torvalds capabilities = 833b4caecd4SChristoph Hellwig NOMMU_MAP_DIRECT | 834b4caecd4SChristoph Hellwig NOMMU_MAP_READ | 835b4caecd4SChristoph Hellwig NOMMU_MAP_WRITE; 8361da177e4SLinus Torvalds break; 8371da177e4SLinus Torvalds 8381da177e4SLinus Torvalds default: 8391da177e4SLinus Torvalds return -EINVAL; 8401da177e4SLinus Torvalds } 8411da177e4SLinus Torvalds } 8421da177e4SLinus Torvalds 8431da177e4SLinus Torvalds /* eliminate any capabilities that we can't support on this 8441da177e4SLinus Torvalds * device */ 8451da177e4SLinus Torvalds if (!file->f_op->get_unmapped_area) 846b4caecd4SChristoph Hellwig capabilities &= ~NOMMU_MAP_DIRECT; 8476e242a1cSAl Viro if (!(file->f_mode & FMODE_CAN_READ)) 848b4caecd4SChristoph Hellwig capabilities &= ~NOMMU_MAP_COPY; 8491da177e4SLinus Torvalds 85028d7a6aeSGraff Yang /* The file shall have been opened with read permission. */ 85128d7a6aeSGraff Yang if (!(file->f_mode & FMODE_READ)) 85228d7a6aeSGraff Yang return -EACCES; 85328d7a6aeSGraff Yang 8541da177e4SLinus Torvalds if (flags & MAP_SHARED) { 8551da177e4SLinus Torvalds /* do checks for writing, appending and locking */ 8561da177e4SLinus Torvalds if ((prot & PROT_WRITE) && 8571da177e4SLinus Torvalds !(file->f_mode & FMODE_WRITE)) 8581da177e4SLinus Torvalds return -EACCES; 8591da177e4SLinus Torvalds 860496ad9aaSAl Viro if (IS_APPEND(file_inode(file)) && 8611da177e4SLinus Torvalds (file->f_mode & FMODE_WRITE)) 8621da177e4SLinus Torvalds return -EACCES; 8631da177e4SLinus Torvalds 864d7a06983SJeff Layton if (locks_verify_locked(file)) 8651da177e4SLinus Torvalds return -EAGAIN; 8661da177e4SLinus Torvalds 867b4caecd4SChristoph Hellwig if (!(capabilities & NOMMU_MAP_DIRECT)) 8681da177e4SLinus Torvalds return -ENODEV; 8691da177e4SLinus Torvalds 8701da177e4SLinus Torvalds /* we mustn't privatise shared mappings */ 871b4caecd4SChristoph Hellwig capabilities &= ~NOMMU_MAP_COPY; 872ac714904SChoi Gi-yong } else { 8731da177e4SLinus Torvalds /* we're going to read the file into private memory we 8741da177e4SLinus Torvalds * allocate */ 875b4caecd4SChristoph Hellwig if (!(capabilities & NOMMU_MAP_COPY)) 8761da177e4SLinus Torvalds return -ENODEV; 8771da177e4SLinus Torvalds 8781da177e4SLinus Torvalds /* we don't permit a private writable mapping to be 8791da177e4SLinus Torvalds * shared with the backing device */ 8801da177e4SLinus Torvalds if (prot & PROT_WRITE) 881b4caecd4SChristoph Hellwig capabilities &= ~NOMMU_MAP_DIRECT; 8821da177e4SLinus Torvalds } 8831da177e4SLinus Torvalds 884b4caecd4SChristoph Hellwig if (capabilities & NOMMU_MAP_DIRECT) { 885b4caecd4SChristoph Hellwig if (((prot & PROT_READ) && !(capabilities & NOMMU_MAP_READ)) || 886b4caecd4SChristoph Hellwig ((prot & PROT_WRITE) && !(capabilities & NOMMU_MAP_WRITE)) || 887b4caecd4SChristoph Hellwig ((prot & PROT_EXEC) && !(capabilities & NOMMU_MAP_EXEC)) 8883c7b2045SBernd Schmidt ) { 889b4caecd4SChristoph Hellwig capabilities &= ~NOMMU_MAP_DIRECT; 8903c7b2045SBernd Schmidt if (flags & MAP_SHARED) { 89122cc877bSLeon Romanovsky pr_warn("MAP_SHARED not completely supported on !MMU\n"); 8923c7b2045SBernd Schmidt return -EINVAL; 8933c7b2045SBernd Schmidt } 8943c7b2045SBernd Schmidt } 8953c7b2045SBernd Schmidt } 8963c7b2045SBernd Schmidt 8971da177e4SLinus Torvalds /* handle executable mappings and implied executable 8981da177e4SLinus Torvalds * mappings */ 89990f8572bSEric W. Biederman if (path_noexec(&file->f_path)) { 9001da177e4SLinus Torvalds if (prot & PROT_EXEC) 9011da177e4SLinus Torvalds return -EPERM; 902ac714904SChoi Gi-yong } else if ((prot & PROT_READ) && !(prot & PROT_EXEC)) { 9031da177e4SLinus Torvalds /* handle implication of PROT_EXEC by PROT_READ */ 9041da177e4SLinus Torvalds if (current->personality & READ_IMPLIES_EXEC) { 905b4caecd4SChristoph Hellwig if (capabilities & NOMMU_MAP_EXEC) 9061da177e4SLinus Torvalds prot |= PROT_EXEC; 9071da177e4SLinus Torvalds } 908ac714904SChoi Gi-yong } else if ((prot & PROT_READ) && 9091da177e4SLinus Torvalds (prot & PROT_EXEC) && 910b4caecd4SChristoph Hellwig !(capabilities & NOMMU_MAP_EXEC) 9111da177e4SLinus Torvalds ) { 9121da177e4SLinus Torvalds /* backing file is not executable, try to copy */ 913b4caecd4SChristoph Hellwig capabilities &= ~NOMMU_MAP_DIRECT; 9141da177e4SLinus Torvalds } 915ac714904SChoi Gi-yong } else { 9161da177e4SLinus Torvalds /* anonymous mappings are always memory backed and can be 9171da177e4SLinus Torvalds * privately mapped 9181da177e4SLinus Torvalds */ 919b4caecd4SChristoph Hellwig capabilities = NOMMU_MAP_COPY; 9201da177e4SLinus Torvalds 9211da177e4SLinus Torvalds /* handle PROT_EXEC implication by PROT_READ */ 9221da177e4SLinus Torvalds if ((prot & PROT_READ) && 9231da177e4SLinus Torvalds (current->personality & READ_IMPLIES_EXEC)) 9241da177e4SLinus Torvalds prot |= PROT_EXEC; 9251da177e4SLinus Torvalds } 9261da177e4SLinus Torvalds 9271da177e4SLinus Torvalds /* allow the security API to have its say */ 928e5467859SAl Viro ret = security_mmap_addr(addr); 929e5467859SAl Viro if (ret < 0) 930e5467859SAl Viro return ret; 9311da177e4SLinus Torvalds 9321da177e4SLinus Torvalds /* looks okay */ 9331da177e4SLinus Torvalds *_capabilities = capabilities; 9341da177e4SLinus Torvalds return 0; 9351da177e4SLinus Torvalds } 9361da177e4SLinus Torvalds 9371da177e4SLinus Torvalds /* 9381da177e4SLinus Torvalds * we've determined that we can make the mapping, now translate what we 9391da177e4SLinus Torvalds * now know into VMA flags 9401da177e4SLinus Torvalds */ 9411da177e4SLinus Torvalds static unsigned long determine_vm_flags(struct file *file, 9421da177e4SLinus Torvalds unsigned long prot, 9431da177e4SLinus Torvalds unsigned long flags, 9441da177e4SLinus Torvalds unsigned long capabilities) 9451da177e4SLinus Torvalds { 9461da177e4SLinus Torvalds unsigned long vm_flags; 9471da177e4SLinus Torvalds 948e6bfb709SDave Hansen vm_flags = calc_vm_prot_bits(prot, 0) | calc_vm_flag_bits(flags); 9491da177e4SLinus Torvalds /* vm_flags |= mm->def_flags; */ 9501da177e4SLinus Torvalds 951b4caecd4SChristoph Hellwig if (!(capabilities & NOMMU_MAP_DIRECT)) { 9521da177e4SLinus Torvalds /* attempt to share read-only copies of mapped file chunks */ 9533c7b2045SBernd Schmidt vm_flags |= VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC; 9541da177e4SLinus Torvalds if (file && !(prot & PROT_WRITE)) 9551da177e4SLinus Torvalds vm_flags |= VM_MAYSHARE; 9563c7b2045SBernd Schmidt } else { 9571da177e4SLinus Torvalds /* overlay a shareable mapping on the backing device or inode 9581da177e4SLinus Torvalds * if possible - used for chardevs, ramfs/tmpfs/shmfs and 9591da177e4SLinus Torvalds * romfs/cramfs */ 960b4caecd4SChristoph Hellwig vm_flags |= VM_MAYSHARE | (capabilities & NOMMU_VMFLAGS); 9611da177e4SLinus Torvalds if (flags & MAP_SHARED) 9623c7b2045SBernd Schmidt vm_flags |= VM_SHARED; 9631da177e4SLinus Torvalds } 9641da177e4SLinus Torvalds 9651da177e4SLinus Torvalds /* refuse to let anyone share private mappings with this process if 9661da177e4SLinus Torvalds * it's being traced - otherwise breakpoints set in it may interfere 9671da177e4SLinus Torvalds * with another untraced process 9681da177e4SLinus Torvalds */ 969a288eeccSTejun Heo if ((flags & MAP_PRIVATE) && current->ptrace) 9701da177e4SLinus Torvalds vm_flags &= ~VM_MAYSHARE; 9711da177e4SLinus Torvalds 9721da177e4SLinus Torvalds return vm_flags; 9731da177e4SLinus Torvalds } 9741da177e4SLinus Torvalds 9751da177e4SLinus Torvalds /* 9768feae131SDavid Howells * set up a shared mapping on a file (the driver or filesystem provides and 9778feae131SDavid Howells * pins the storage) 9781da177e4SLinus Torvalds */ 9798feae131SDavid Howells static int do_mmap_shared_file(struct vm_area_struct *vma) 9801da177e4SLinus Torvalds { 9811da177e4SLinus Torvalds int ret; 9821da177e4SLinus Torvalds 983f74ac015SMiklos Szeredi ret = call_mmap(vma->vm_file, vma); 984dd8632a1SPaul Mundt if (ret == 0) { 985dd8632a1SPaul Mundt vma->vm_region->vm_top = vma->vm_region->vm_end; 986645d83c5SDavid Howells return 0; 987dd8632a1SPaul Mundt } 9881da177e4SLinus Torvalds if (ret != -ENOSYS) 9891da177e4SLinus Torvalds return ret; 9901da177e4SLinus Torvalds 9913fa30460SDavid Howells /* getting -ENOSYS indicates that direct mmap isn't possible (as 9923fa30460SDavid Howells * opposed to tried but failed) so we can only give a suitable error as 9933fa30460SDavid Howells * it's not possible to make a private copy if MAP_SHARED was given */ 9941da177e4SLinus Torvalds return -ENODEV; 9951da177e4SLinus Torvalds } 9961da177e4SLinus Torvalds 9971da177e4SLinus Torvalds /* 9981da177e4SLinus Torvalds * set up a private mapping or an anonymous shared mapping 9991da177e4SLinus Torvalds */ 10008feae131SDavid Howells static int do_mmap_private(struct vm_area_struct *vma, 10018feae131SDavid Howells struct vm_region *region, 1002645d83c5SDavid Howells unsigned long len, 1003645d83c5SDavid Howells unsigned long capabilities) 10041da177e4SLinus Torvalds { 1005dbc8358cSJoonsoo Kim unsigned long total, point; 10061da177e4SLinus Torvalds void *base; 10078feae131SDavid Howells int ret, order; 10081da177e4SLinus Torvalds 10091da177e4SLinus Torvalds /* invoke the file's mapping function so that it can keep track of 10101da177e4SLinus Torvalds * shared mappings on devices or memory 10111da177e4SLinus Torvalds * - VM_MAYSHARE will be set if it may attempt to share 10121da177e4SLinus Torvalds */ 1013b4caecd4SChristoph Hellwig if (capabilities & NOMMU_MAP_DIRECT) { 1014f74ac015SMiklos Szeredi ret = call_mmap(vma->vm_file, vma); 1015dd8632a1SPaul Mundt if (ret == 0) { 10161da177e4SLinus Torvalds /* shouldn't return success if we're not sharing */ 1017dd8632a1SPaul Mundt BUG_ON(!(vma->vm_flags & VM_MAYSHARE)); 1018dd8632a1SPaul Mundt vma->vm_region->vm_top = vma->vm_region->vm_end; 1019645d83c5SDavid Howells return 0; 10201da177e4SLinus Torvalds } 1021dd8632a1SPaul Mundt if (ret != -ENOSYS) 1022dd8632a1SPaul Mundt return ret; 10231da177e4SLinus Torvalds 10241da177e4SLinus Torvalds /* getting an ENOSYS error indicates that direct mmap isn't 10251da177e4SLinus Torvalds * possible (as opposed to tried but failed) so we'll try to 10261da177e4SLinus Torvalds * make a private copy of the data and map that instead */ 10271da177e4SLinus Torvalds } 10281da177e4SLinus Torvalds 10298feae131SDavid Howells 10301da177e4SLinus Torvalds /* allocate some memory to hold the mapping 10311da177e4SLinus Torvalds * - note that this may not return a page-aligned address if the object 10321da177e4SLinus Torvalds * we're allocating is smaller than a page 10331da177e4SLinus Torvalds */ 1034f67d9b15SBob Liu order = get_order(len); 10358feae131SDavid Howells total = 1 << order; 1036f67d9b15SBob Liu point = len >> PAGE_SHIFT; 1037dd8632a1SPaul Mundt 1038dbc8358cSJoonsoo Kim /* we don't want to allocate a power-of-2 sized page set */ 103922cc877bSLeon Romanovsky if (sysctl_nr_trim_pages && total - point >= sysctl_nr_trim_pages) 1040dbc8358cSJoonsoo Kim total = point; 10418feae131SDavid Howells 1042da616534SJoonsoo Kim base = alloc_pages_exact(total << PAGE_SHIFT, GFP_KERNEL); 1043dbc8358cSJoonsoo Kim if (!base) 1044dbc8358cSJoonsoo Kim goto enomem; 10458feae131SDavid Howells 1046dbc8358cSJoonsoo Kim atomic_long_add(total, &mmap_pages_allocated); 1047dbc8358cSJoonsoo Kim 10488feae131SDavid Howells region->vm_flags = vma->vm_flags |= VM_MAPPED_COPY; 10498feae131SDavid Howells region->vm_start = (unsigned long) base; 1050f67d9b15SBob Liu region->vm_end = region->vm_start + len; 1051dd8632a1SPaul Mundt region->vm_top = region->vm_start + (total << PAGE_SHIFT); 10528feae131SDavid Howells 10538feae131SDavid Howells vma->vm_start = region->vm_start; 10548feae131SDavid Howells vma->vm_end = region->vm_start + len; 10551da177e4SLinus Torvalds 10561da177e4SLinus Torvalds if (vma->vm_file) { 10571da177e4SLinus Torvalds /* read the contents of a file into the copy */ 10581da177e4SLinus Torvalds loff_t fpos; 10591da177e4SLinus Torvalds 10601da177e4SLinus Torvalds fpos = vma->vm_pgoff; 10611da177e4SLinus Torvalds fpos <<= PAGE_SHIFT; 10621da177e4SLinus Torvalds 1063b4bf802aSChristoph Hellwig ret = kernel_read(vma->vm_file, base, len, &fpos); 10641da177e4SLinus Torvalds if (ret < 0) 10651da177e4SLinus Torvalds goto error_free; 10661da177e4SLinus Torvalds 10671da177e4SLinus Torvalds /* clear the last little bit */ 1068f67d9b15SBob Liu if (ret < len) 1069f67d9b15SBob Liu memset(base + ret, 0, len - ret); 10701da177e4SLinus Torvalds 1071bfd40eafSKirill A. Shutemov } else { 1072bfd40eafSKirill A. Shutemov vma_set_anonymous(vma); 10731da177e4SLinus Torvalds } 10741da177e4SLinus Torvalds 10751da177e4SLinus Torvalds return 0; 10761da177e4SLinus Torvalds 10771da177e4SLinus Torvalds error_free: 10787223bb4aSNamhyung Kim free_page_series(region->vm_start, region->vm_top); 10798feae131SDavid Howells region->vm_start = vma->vm_start = 0; 10808feae131SDavid Howells region->vm_end = vma->vm_end = 0; 1081dd8632a1SPaul Mundt region->vm_top = 0; 10821da177e4SLinus Torvalds return ret; 10831da177e4SLinus Torvalds 10841da177e4SLinus Torvalds enomem: 1085b1de0d13SMitchel Humpherys pr_err("Allocation of length %lu from process %d (%s) failed\n", 108605ae6fa3SGreg Ungerer len, current->pid, current->comm); 10879af744d7SMichal Hocko show_free_areas(0, NULL); 10881da177e4SLinus Torvalds return -ENOMEM; 10891da177e4SLinus Torvalds } 10901da177e4SLinus Torvalds 10911da177e4SLinus Torvalds /* 10921da177e4SLinus Torvalds * handle mapping creation for uClinux 10931da177e4SLinus Torvalds */ 10941fcfd8dbSOleg Nesterov unsigned long do_mmap(struct file *file, 10951da177e4SLinus Torvalds unsigned long addr, 10961da177e4SLinus Torvalds unsigned long len, 10971da177e4SLinus Torvalds unsigned long prot, 10981da177e4SLinus Torvalds unsigned long flags, 10991fcfd8dbSOleg Nesterov vm_flags_t vm_flags, 1100bebeb3d6SMichel Lespinasse unsigned long pgoff, 1101897ab3e0SMike Rapoport unsigned long *populate, 1102897ab3e0SMike Rapoport struct list_head *uf) 11031da177e4SLinus Torvalds { 11048feae131SDavid Howells struct vm_area_struct *vma; 11058feae131SDavid Howells struct vm_region *region; 11061da177e4SLinus Torvalds struct rb_node *rb; 11071fcfd8dbSOleg Nesterov unsigned long capabilities, result; 11081da177e4SLinus Torvalds int ret; 11091da177e4SLinus Torvalds 111041badc15SMichel Lespinasse *populate = 0; 1111bebeb3d6SMichel Lespinasse 11121da177e4SLinus Torvalds /* decide whether we should attempt the mapping, and if so what sort of 11131da177e4SLinus Torvalds * mapping */ 11141da177e4SLinus Torvalds ret = validate_mmap_request(file, addr, len, prot, flags, pgoff, 11151da177e4SLinus Torvalds &capabilities); 111622cc877bSLeon Romanovsky if (ret < 0) 11171da177e4SLinus Torvalds return ret; 11181da177e4SLinus Torvalds 111906aab5a3SDavid Howells /* we ignore the address hint */ 112006aab5a3SDavid Howells addr = 0; 1121f67d9b15SBob Liu len = PAGE_ALIGN(len); 112206aab5a3SDavid Howells 11231da177e4SLinus Torvalds /* we've determined that we can make the mapping, now translate what we 11241da177e4SLinus Torvalds * now know into VMA flags */ 11251fcfd8dbSOleg Nesterov vm_flags |= determine_vm_flags(file, prot, flags, capabilities); 11261da177e4SLinus Torvalds 11278feae131SDavid Howells /* we're going to need to record the mapping */ 11288feae131SDavid Howells region = kmem_cache_zalloc(vm_region_jar, GFP_KERNEL); 11298feae131SDavid Howells if (!region) 11308feae131SDavid Howells goto error_getting_region; 11311da177e4SLinus Torvalds 1132490fc053SLinus Torvalds vma = vm_area_alloc(current->mm); 11338feae131SDavid Howells if (!vma) 11348feae131SDavid Howells goto error_getting_vma; 11351da177e4SLinus Torvalds 11361e2ae599SDavid Howells region->vm_usage = 1; 11378feae131SDavid Howells region->vm_flags = vm_flags; 11388feae131SDavid Howells region->vm_pgoff = pgoff; 11398feae131SDavid Howells 11408feae131SDavid Howells vma->vm_flags = vm_flags; 11418feae131SDavid Howells vma->vm_pgoff = pgoff; 11428feae131SDavid Howells 11438feae131SDavid Howells if (file) { 1144cb0942b8SAl Viro region->vm_file = get_file(file); 1145cb0942b8SAl Viro vma->vm_file = get_file(file); 11468feae131SDavid Howells } 11478feae131SDavid Howells 11488feae131SDavid Howells down_write(&nommu_region_sem); 11498feae131SDavid Howells 11508feae131SDavid Howells /* if we want to share, we need to check for regions created by other 11511da177e4SLinus Torvalds * mmap() calls that overlap with our proposed mapping 11528feae131SDavid Howells * - we can only share with a superset match on most regular files 11531da177e4SLinus Torvalds * - shared mappings on character devices and memory backed files are 11541da177e4SLinus Torvalds * permitted to overlap inexactly as far as we are concerned for in 11551da177e4SLinus Torvalds * these cases, sharing is handled in the driver or filesystem rather 11561da177e4SLinus Torvalds * than here 11571da177e4SLinus Torvalds */ 11581da177e4SLinus Torvalds if (vm_flags & VM_MAYSHARE) { 11598feae131SDavid Howells struct vm_region *pregion; 11608feae131SDavid Howells unsigned long pglen, rpglen, pgend, rpgend, start; 11611da177e4SLinus Torvalds 11628feae131SDavid Howells pglen = (len + PAGE_SIZE - 1) >> PAGE_SHIFT; 11638feae131SDavid Howells pgend = pgoff + pglen; 1164165b2392SDavid Howells 11658feae131SDavid Howells for (rb = rb_first(&nommu_region_tree); rb; rb = rb_next(rb)) { 11668feae131SDavid Howells pregion = rb_entry(rb, struct vm_region, vm_rb); 11671da177e4SLinus Torvalds 11688feae131SDavid Howells if (!(pregion->vm_flags & VM_MAYSHARE)) 11691da177e4SLinus Torvalds continue; 11701da177e4SLinus Torvalds 11711da177e4SLinus Torvalds /* search for overlapping mappings on the same file */ 1172496ad9aaSAl Viro if (file_inode(pregion->vm_file) != 1173496ad9aaSAl Viro file_inode(file)) 11741da177e4SLinus Torvalds continue; 11751da177e4SLinus Torvalds 11768feae131SDavid Howells if (pregion->vm_pgoff >= pgend) 11771da177e4SLinus Torvalds continue; 11781da177e4SLinus Torvalds 11798feae131SDavid Howells rpglen = pregion->vm_end - pregion->vm_start; 11808feae131SDavid Howells rpglen = (rpglen + PAGE_SIZE - 1) >> PAGE_SHIFT; 11818feae131SDavid Howells rpgend = pregion->vm_pgoff + rpglen; 11828feae131SDavid Howells if (pgoff >= rpgend) 11831da177e4SLinus Torvalds continue; 11841da177e4SLinus Torvalds 11858feae131SDavid Howells /* handle inexactly overlapping matches between 11868feae131SDavid Howells * mappings */ 11878feae131SDavid Howells if ((pregion->vm_pgoff != pgoff || rpglen != pglen) && 11888feae131SDavid Howells !(pgoff >= pregion->vm_pgoff && pgend <= rpgend)) { 11898feae131SDavid Howells /* new mapping is not a subset of the region */ 1190b4caecd4SChristoph Hellwig if (!(capabilities & NOMMU_MAP_DIRECT)) 11911da177e4SLinus Torvalds goto sharing_violation; 11921da177e4SLinus Torvalds continue; 11931da177e4SLinus Torvalds } 11941da177e4SLinus Torvalds 11958feae131SDavid Howells /* we've found a region we can share */ 11961e2ae599SDavid Howells pregion->vm_usage++; 11978feae131SDavid Howells vma->vm_region = pregion; 11988feae131SDavid Howells start = pregion->vm_start; 11998feae131SDavid Howells start += (pgoff - pregion->vm_pgoff) << PAGE_SHIFT; 12008feae131SDavid Howells vma->vm_start = start; 12018feae131SDavid Howells vma->vm_end = start + len; 12021da177e4SLinus Torvalds 120322cc877bSLeon Romanovsky if (pregion->vm_flags & VM_MAPPED_COPY) 12048feae131SDavid Howells vma->vm_flags |= VM_MAPPED_COPY; 120522cc877bSLeon Romanovsky else { 12068feae131SDavid Howells ret = do_mmap_shared_file(vma); 12078feae131SDavid Howells if (ret < 0) { 12088feae131SDavid Howells vma->vm_region = NULL; 12098feae131SDavid Howells vma->vm_start = 0; 12108feae131SDavid Howells vma->vm_end = 0; 12111e2ae599SDavid Howells pregion->vm_usage--; 12128feae131SDavid Howells pregion = NULL; 12138feae131SDavid Howells goto error_just_free; 12141da177e4SLinus Torvalds } 12158feae131SDavid Howells } 12168feae131SDavid Howells fput(region->vm_file); 12178feae131SDavid Howells kmem_cache_free(vm_region_jar, region); 12188feae131SDavid Howells region = pregion; 12198feae131SDavid Howells result = start; 12208feae131SDavid Howells goto share; 12218feae131SDavid Howells } 12221da177e4SLinus Torvalds 12231da177e4SLinus Torvalds /* obtain the address at which to make a shared mapping 12241da177e4SLinus Torvalds * - this is the hook for quasi-memory character devices to 12251da177e4SLinus Torvalds * tell us the location of a shared mapping 12261da177e4SLinus Torvalds */ 1227b4caecd4SChristoph Hellwig if (capabilities & NOMMU_MAP_DIRECT) { 12281da177e4SLinus Torvalds addr = file->f_op->get_unmapped_area(file, addr, len, 12291da177e4SLinus Torvalds pgoff, flags); 1230bb005a59SNamhyung Kim if (IS_ERR_VALUE(addr)) { 12311da177e4SLinus Torvalds ret = addr; 1232bb005a59SNamhyung Kim if (ret != -ENOSYS) 12338feae131SDavid Howells goto error_just_free; 12341da177e4SLinus Torvalds 12351da177e4SLinus Torvalds /* the driver refused to tell us where to site 12361da177e4SLinus Torvalds * the mapping so we'll have to attempt to copy 12371da177e4SLinus Torvalds * it */ 1238bb005a59SNamhyung Kim ret = -ENODEV; 1239b4caecd4SChristoph Hellwig if (!(capabilities & NOMMU_MAP_COPY)) 12408feae131SDavid Howells goto error_just_free; 12411da177e4SLinus Torvalds 1242b4caecd4SChristoph Hellwig capabilities &= ~NOMMU_MAP_DIRECT; 12438feae131SDavid Howells } else { 12448feae131SDavid Howells vma->vm_start = region->vm_start = addr; 12458feae131SDavid Howells vma->vm_end = region->vm_end = addr + len; 12461da177e4SLinus Torvalds } 12471da177e4SLinus Torvalds } 12481da177e4SLinus Torvalds } 12491da177e4SLinus Torvalds 12508feae131SDavid Howells vma->vm_region = region; 12511da177e4SLinus Torvalds 1252645d83c5SDavid Howells /* set up the mapping 1253b4caecd4SChristoph Hellwig * - the region is filled in if NOMMU_MAP_DIRECT is still set 1254645d83c5SDavid Howells */ 12551da177e4SLinus Torvalds if (file && vma->vm_flags & VM_SHARED) 12568feae131SDavid Howells ret = do_mmap_shared_file(vma); 12571da177e4SLinus Torvalds else 1258645d83c5SDavid Howells ret = do_mmap_private(vma, region, len, capabilities); 12591da177e4SLinus Torvalds if (ret < 0) 1260645d83c5SDavid Howells goto error_just_free; 1261645d83c5SDavid Howells add_nommu_region(region); 12628feae131SDavid Howells 1263ea637639SJie Zhang /* clear anonymous mappings that don't ask for uninitialized data */ 1264*0bf5f949SChristoph Hellwig if (!vma->vm_file && 1265*0bf5f949SChristoph Hellwig (!IS_ENABLED(CONFIG_MMAP_ALLOW_UNINITIALIZED) || 1266*0bf5f949SChristoph Hellwig !(flags & MAP_UNINITIALIZED))) 1267ea637639SJie Zhang memset((void *)region->vm_start, 0, 1268ea637639SJie Zhang region->vm_end - region->vm_start); 1269ea637639SJie Zhang 12701da177e4SLinus Torvalds /* okay... we have a mapping; now we have to register it */ 12718feae131SDavid Howells result = vma->vm_start; 12721da177e4SLinus Torvalds 12731da177e4SLinus Torvalds current->mm->total_vm += len >> PAGE_SHIFT; 12741da177e4SLinus Torvalds 12758feae131SDavid Howells share: 12768feae131SDavid Howells add_vma_to_mm(current->mm, vma); 12771da177e4SLinus Torvalds 1278cfe79c00SMike Frysinger /* we flush the region from the icache only when the first executable 1279cfe79c00SMike Frysinger * mapping of it is made */ 1280cfe79c00SMike Frysinger if (vma->vm_flags & VM_EXEC && !region->vm_icache_flushed) { 1281cfe79c00SMike Frysinger flush_icache_range(region->vm_start, region->vm_end); 1282cfe79c00SMike Frysinger region->vm_icache_flushed = true; 1283cfe79c00SMike Frysinger } 12841da177e4SLinus Torvalds 1285cfe79c00SMike Frysinger up_write(&nommu_region_sem); 12861da177e4SLinus Torvalds 12878feae131SDavid Howells return result; 12881da177e4SLinus Torvalds 12898feae131SDavid Howells error_just_free: 12908feae131SDavid Howells up_write(&nommu_region_sem); 12918feae131SDavid Howells error: 129289a86402SDavid Howells if (region->vm_file) 12938feae131SDavid Howells fput(region->vm_file); 12948feae131SDavid Howells kmem_cache_free(vm_region_jar, region); 129589a86402SDavid Howells if (vma->vm_file) 12968feae131SDavid Howells fput(vma->vm_file); 12973928d4f5SLinus Torvalds vm_area_free(vma); 12981da177e4SLinus Torvalds return ret; 12991da177e4SLinus Torvalds 13001da177e4SLinus Torvalds sharing_violation: 13018feae131SDavid Howells up_write(&nommu_region_sem); 130222cc877bSLeon Romanovsky pr_warn("Attempt to share mismatched mappings\n"); 13038feae131SDavid Howells ret = -EINVAL; 13048feae131SDavid Howells goto error; 13051da177e4SLinus Torvalds 13061da177e4SLinus Torvalds error_getting_vma: 13078feae131SDavid Howells kmem_cache_free(vm_region_jar, region); 130822cc877bSLeon Romanovsky pr_warn("Allocation of vma for %lu byte allocation from process %d failed\n", 13091da177e4SLinus Torvalds len, current->pid); 13109af744d7SMichal Hocko show_free_areas(0, NULL); 13111da177e4SLinus Torvalds return -ENOMEM; 13121da177e4SLinus Torvalds 13138feae131SDavid Howells error_getting_region: 131422cc877bSLeon Romanovsky pr_warn("Allocation of vm region for %lu byte allocation from process %d failed\n", 13151da177e4SLinus Torvalds len, current->pid); 13169af744d7SMichal Hocko show_free_areas(0, NULL); 13171da177e4SLinus Torvalds return -ENOMEM; 13181da177e4SLinus Torvalds } 13196be5ceb0SLinus Torvalds 1320a90f590aSDominik Brodowski unsigned long ksys_mmap_pgoff(unsigned long addr, unsigned long len, 1321a90f590aSDominik Brodowski unsigned long prot, unsigned long flags, 1322a90f590aSDominik Brodowski unsigned long fd, unsigned long pgoff) 132366f0dc48SHugh Dickins { 132466f0dc48SHugh Dickins struct file *file = NULL; 132566f0dc48SHugh Dickins unsigned long retval = -EBADF; 132666f0dc48SHugh Dickins 1327120a795dSAl Viro audit_mmap_fd(fd, flags); 132866f0dc48SHugh Dickins if (!(flags & MAP_ANONYMOUS)) { 132966f0dc48SHugh Dickins file = fget(fd); 133066f0dc48SHugh Dickins if (!file) 133166f0dc48SHugh Dickins goto out; 133266f0dc48SHugh Dickins } 133366f0dc48SHugh Dickins 133466f0dc48SHugh Dickins flags &= ~(MAP_EXECUTABLE | MAP_DENYWRITE); 133566f0dc48SHugh Dickins 1336ad1ed293SGreg Ungerer retval = vm_mmap_pgoff(file, addr, len, prot, flags, pgoff); 133766f0dc48SHugh Dickins 133866f0dc48SHugh Dickins if (file) 133966f0dc48SHugh Dickins fput(file); 134066f0dc48SHugh Dickins out: 134166f0dc48SHugh Dickins return retval; 134266f0dc48SHugh Dickins } 134366f0dc48SHugh Dickins 1344a90f590aSDominik Brodowski SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len, 1345a90f590aSDominik Brodowski unsigned long, prot, unsigned long, flags, 1346a90f590aSDominik Brodowski unsigned long, fd, unsigned long, pgoff) 1347a90f590aSDominik Brodowski { 1348a90f590aSDominik Brodowski return ksys_mmap_pgoff(addr, len, prot, flags, fd, pgoff); 1349a90f590aSDominik Brodowski } 1350a90f590aSDominik Brodowski 1351a4679373SChristoph Hellwig #ifdef __ARCH_WANT_SYS_OLD_MMAP 1352a4679373SChristoph Hellwig struct mmap_arg_struct { 1353a4679373SChristoph Hellwig unsigned long addr; 1354a4679373SChristoph Hellwig unsigned long len; 1355a4679373SChristoph Hellwig unsigned long prot; 1356a4679373SChristoph Hellwig unsigned long flags; 1357a4679373SChristoph Hellwig unsigned long fd; 1358a4679373SChristoph Hellwig unsigned long offset; 1359a4679373SChristoph Hellwig }; 1360a4679373SChristoph Hellwig 1361a4679373SChristoph Hellwig SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg) 1362a4679373SChristoph Hellwig { 1363a4679373SChristoph Hellwig struct mmap_arg_struct a; 1364a4679373SChristoph Hellwig 1365a4679373SChristoph Hellwig if (copy_from_user(&a, arg, sizeof(a))) 1366a4679373SChristoph Hellwig return -EFAULT; 13671824cb75SAlexander Kuleshov if (offset_in_page(a.offset)) 1368a4679373SChristoph Hellwig return -EINVAL; 1369a4679373SChristoph Hellwig 1370a90f590aSDominik Brodowski return ksys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd, 1371a4679373SChristoph Hellwig a.offset >> PAGE_SHIFT); 1372a4679373SChristoph Hellwig } 1373a4679373SChristoph Hellwig #endif /* __ARCH_WANT_SYS_OLD_MMAP */ 1374a4679373SChristoph Hellwig 13751da177e4SLinus Torvalds /* 13768feae131SDavid Howells * split a vma into two pieces at address 'addr', a new vma is allocated either 13778feae131SDavid Howells * for the first part or the tail. 13781da177e4SLinus Torvalds */ 13798feae131SDavid Howells int split_vma(struct mm_struct *mm, struct vm_area_struct *vma, 13808feae131SDavid Howells unsigned long addr, int new_below) 13811da177e4SLinus Torvalds { 13828feae131SDavid Howells struct vm_area_struct *new; 13838feae131SDavid Howells struct vm_region *region; 13848feae131SDavid Howells unsigned long npages; 13851da177e4SLinus Torvalds 1386779c1023SDavid Howells /* we're only permitted to split anonymous regions (these should have 1387779c1023SDavid Howells * only a single usage on the region) */ 1388779c1023SDavid Howells if (vma->vm_file) 13898feae131SDavid Howells return -ENOMEM; 13901da177e4SLinus Torvalds 13918feae131SDavid Howells if (mm->map_count >= sysctl_max_map_count) 13928feae131SDavid Howells return -ENOMEM; 13931da177e4SLinus Torvalds 13948feae131SDavid Howells region = kmem_cache_alloc(vm_region_jar, GFP_KERNEL); 13958feae131SDavid Howells if (!region) 13968feae131SDavid Howells return -ENOMEM; 13978feae131SDavid Howells 13983928d4f5SLinus Torvalds new = vm_area_dup(vma); 13998feae131SDavid Howells if (!new) { 14008feae131SDavid Howells kmem_cache_free(vm_region_jar, region); 14018feae131SDavid Howells return -ENOMEM; 14021da177e4SLinus Torvalds } 14031da177e4SLinus Torvalds 14048feae131SDavid Howells /* most fields are the same, copy all, and then fixup */ 14058feae131SDavid Howells *region = *vma->vm_region; 14068feae131SDavid Howells new->vm_region = region; 14078feae131SDavid Howells 14088feae131SDavid Howells npages = (addr - vma->vm_start) >> PAGE_SHIFT; 14098feae131SDavid Howells 14108feae131SDavid Howells if (new_below) { 1411dd8632a1SPaul Mundt region->vm_top = region->vm_end = new->vm_end = addr; 14128feae131SDavid Howells } else { 14138feae131SDavid Howells region->vm_start = new->vm_start = addr; 14148feae131SDavid Howells region->vm_pgoff = new->vm_pgoff += npages; 14151da177e4SLinus Torvalds } 14168feae131SDavid Howells 14178feae131SDavid Howells if (new->vm_ops && new->vm_ops->open) 14188feae131SDavid Howells new->vm_ops->open(new); 14198feae131SDavid Howells 14208feae131SDavid Howells delete_vma_from_mm(vma); 14218feae131SDavid Howells down_write(&nommu_region_sem); 14228feae131SDavid Howells delete_nommu_region(vma->vm_region); 14238feae131SDavid Howells if (new_below) { 14248feae131SDavid Howells vma->vm_region->vm_start = vma->vm_start = addr; 14258feae131SDavid Howells vma->vm_region->vm_pgoff = vma->vm_pgoff += npages; 14268feae131SDavid Howells } else { 14278feae131SDavid Howells vma->vm_region->vm_end = vma->vm_end = addr; 1428dd8632a1SPaul Mundt vma->vm_region->vm_top = addr; 14298feae131SDavid Howells } 14308feae131SDavid Howells add_nommu_region(vma->vm_region); 14318feae131SDavid Howells add_nommu_region(new->vm_region); 14328feae131SDavid Howells up_write(&nommu_region_sem); 14338feae131SDavid Howells add_vma_to_mm(mm, vma); 14348feae131SDavid Howells add_vma_to_mm(mm, new); 14358feae131SDavid Howells return 0; 14368feae131SDavid Howells } 14378feae131SDavid Howells 14388feae131SDavid Howells /* 14398feae131SDavid Howells * shrink a VMA by removing the specified chunk from either the beginning or 14408feae131SDavid Howells * the end 14418feae131SDavid Howells */ 14428feae131SDavid Howells static int shrink_vma(struct mm_struct *mm, 14438feae131SDavid Howells struct vm_area_struct *vma, 14448feae131SDavid Howells unsigned long from, unsigned long to) 14458feae131SDavid Howells { 14468feae131SDavid Howells struct vm_region *region; 14478feae131SDavid Howells 14488feae131SDavid Howells /* adjust the VMA's pointers, which may reposition it in the MM's tree 14498feae131SDavid Howells * and list */ 14508feae131SDavid Howells delete_vma_from_mm(vma); 14518feae131SDavid Howells if (from > vma->vm_start) 14528feae131SDavid Howells vma->vm_end = from; 14538feae131SDavid Howells else 14548feae131SDavid Howells vma->vm_start = to; 14558feae131SDavid Howells add_vma_to_mm(mm, vma); 14568feae131SDavid Howells 14578feae131SDavid Howells /* cut the backing region down to size */ 14588feae131SDavid Howells region = vma->vm_region; 14591e2ae599SDavid Howells BUG_ON(region->vm_usage != 1); 14608feae131SDavid Howells 14618feae131SDavid Howells down_write(&nommu_region_sem); 14628feae131SDavid Howells delete_nommu_region(region); 1463dd8632a1SPaul Mundt if (from > region->vm_start) { 1464dd8632a1SPaul Mundt to = region->vm_top; 1465dd8632a1SPaul Mundt region->vm_top = region->vm_end = from; 1466dd8632a1SPaul Mundt } else { 14678feae131SDavid Howells region->vm_start = to; 1468dd8632a1SPaul Mundt } 14698feae131SDavid Howells add_nommu_region(region); 14708feae131SDavid Howells up_write(&nommu_region_sem); 14718feae131SDavid Howells 14728feae131SDavid Howells free_page_series(from, to); 14738feae131SDavid Howells return 0; 14741da177e4SLinus Torvalds } 14751da177e4SLinus Torvalds 14763034097aSDavid Howells /* 14773034097aSDavid Howells * release a mapping 14788feae131SDavid Howells * - under NOMMU conditions the chunk to be unmapped must be backed by a single 14798feae131SDavid Howells * VMA, though it need not cover the whole VMA 14803034097aSDavid Howells */ 1481897ab3e0SMike Rapoport int do_munmap(struct mm_struct *mm, unsigned long start, size_t len, struct list_head *uf) 14821da177e4SLinus Torvalds { 14838feae131SDavid Howells struct vm_area_struct *vma; 1484f67d9b15SBob Liu unsigned long end; 14858feae131SDavid Howells int ret; 14861da177e4SLinus Torvalds 1487f67d9b15SBob Liu len = PAGE_ALIGN(len); 14888feae131SDavid Howells if (len == 0) 14891da177e4SLinus Torvalds return -EINVAL; 14901da177e4SLinus Torvalds 1491f67d9b15SBob Liu end = start + len; 1492f67d9b15SBob Liu 14938feae131SDavid Howells /* find the first potentially overlapping VMA */ 14948feae131SDavid Howells vma = find_vma(mm, start); 14958feae131SDavid Howells if (!vma) { 1496ac714904SChoi Gi-yong static int limit; 149733e5d769SDavid Howells if (limit < 5) { 149822cc877bSLeon Romanovsky pr_warn("munmap of memory not mmapped by process %d (%s): 0x%lx-0x%lx\n", 149933e5d769SDavid Howells current->pid, current->comm, 150033e5d769SDavid Howells start, start + len - 1); 150133e5d769SDavid Howells limit++; 150233e5d769SDavid Howells } 15038feae131SDavid Howells return -EINVAL; 15048feae131SDavid Howells } 15051da177e4SLinus Torvalds 15068feae131SDavid Howells /* we're allowed to split an anonymous VMA but not a file-backed one */ 15078feae131SDavid Howells if (vma->vm_file) { 15088feae131SDavid Howells do { 150922cc877bSLeon Romanovsky if (start > vma->vm_start) 15108feae131SDavid Howells return -EINVAL; 15118feae131SDavid Howells if (end == vma->vm_end) 15128feae131SDavid Howells goto erase_whole_vma; 1513d75a310cSNamhyung Kim vma = vma->vm_next; 1514d75a310cSNamhyung Kim } while (vma); 15158feae131SDavid Howells return -EINVAL; 15168feae131SDavid Howells } else { 15178feae131SDavid Howells /* the chunk must be a subset of the VMA found */ 15188feae131SDavid Howells if (start == vma->vm_start && end == vma->vm_end) 15198feae131SDavid Howells goto erase_whole_vma; 152022cc877bSLeon Romanovsky if (start < vma->vm_start || end > vma->vm_end) 15218feae131SDavid Howells return -EINVAL; 15221824cb75SAlexander Kuleshov if (offset_in_page(start)) 15238feae131SDavid Howells return -EINVAL; 15241824cb75SAlexander Kuleshov if (end != vma->vm_end && offset_in_page(end)) 15258feae131SDavid Howells return -EINVAL; 15268feae131SDavid Howells if (start != vma->vm_start && end != vma->vm_end) { 15278feae131SDavid Howells ret = split_vma(mm, vma, start, 1); 152822cc877bSLeon Romanovsky if (ret < 0) 15298feae131SDavid Howells return ret; 15308feae131SDavid Howells } 15318feae131SDavid Howells return shrink_vma(mm, vma, start, end); 15328feae131SDavid Howells } 15331da177e4SLinus Torvalds 15348feae131SDavid Howells erase_whole_vma: 15358feae131SDavid Howells delete_vma_from_mm(vma); 15368feae131SDavid Howells delete_vma(mm, vma); 15371da177e4SLinus Torvalds return 0; 15381da177e4SLinus Torvalds } 1539b5073173SPaul Mundt EXPORT_SYMBOL(do_munmap); 15401da177e4SLinus Torvalds 1541bfce281cSAl Viro int vm_munmap(unsigned long addr, size_t len) 15423034097aSDavid Howells { 1543bfce281cSAl Viro struct mm_struct *mm = current->mm; 15443034097aSDavid Howells int ret; 15453034097aSDavid Howells 15463034097aSDavid Howells down_write(&mm->mmap_sem); 1547897ab3e0SMike Rapoport ret = do_munmap(mm, addr, len, NULL); 15483034097aSDavid Howells up_write(&mm->mmap_sem); 15493034097aSDavid Howells return ret; 15503034097aSDavid Howells } 1551a46ef99dSLinus Torvalds EXPORT_SYMBOL(vm_munmap); 1552a46ef99dSLinus Torvalds 1553a46ef99dSLinus Torvalds SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len) 1554a46ef99dSLinus Torvalds { 1555bfce281cSAl Viro return vm_munmap(addr, len); 1556a46ef99dSLinus Torvalds } 15573034097aSDavid Howells 15583034097aSDavid Howells /* 15598feae131SDavid Howells * release all the mappings made in a process's VM space 15603034097aSDavid Howells */ 15611da177e4SLinus Torvalds void exit_mmap(struct mm_struct *mm) 15621da177e4SLinus Torvalds { 15638feae131SDavid Howells struct vm_area_struct *vma; 15641da177e4SLinus Torvalds 15658feae131SDavid Howells if (!mm) 15668feae131SDavid Howells return; 15678feae131SDavid Howells 15681da177e4SLinus Torvalds mm->total_vm = 0; 15691da177e4SLinus Torvalds 15708feae131SDavid Howells while ((vma = mm->mmap)) { 15718feae131SDavid Howells mm->mmap = vma->vm_next; 15728feae131SDavid Howells delete_vma_from_mm(vma); 15738feae131SDavid Howells delete_vma(mm, vma); 157404c34961SSteven J. Magnani cond_resched(); 15751da177e4SLinus Torvalds } 15761da177e4SLinus Torvalds } 15771da177e4SLinus Torvalds 15785d22fc25SLinus Torvalds int vm_brk(unsigned long addr, unsigned long len) 15791da177e4SLinus Torvalds { 15801da177e4SLinus Torvalds return -ENOMEM; 15811da177e4SLinus Torvalds } 15821da177e4SLinus Torvalds 15831da177e4SLinus Torvalds /* 15846fa5f80bSDavid Howells * expand (or shrink) an existing mapping, potentially moving it at the same 15856fa5f80bSDavid Howells * time (controlled by the MREMAP_MAYMOVE flag and available VM space) 15861da177e4SLinus Torvalds * 15876fa5f80bSDavid Howells * under NOMMU conditions, we only permit changing a mapping's size, and only 15888feae131SDavid Howells * as long as it stays within the region allocated by do_mmap_private() and the 15898feae131SDavid Howells * block is not shareable 15901da177e4SLinus Torvalds * 15916fa5f80bSDavid Howells * MREMAP_FIXED is not supported under NOMMU conditions 15921da177e4SLinus Torvalds */ 15934b377babSAl Viro static unsigned long do_mremap(unsigned long addr, 15941da177e4SLinus Torvalds unsigned long old_len, unsigned long new_len, 15951da177e4SLinus Torvalds unsigned long flags, unsigned long new_addr) 15961da177e4SLinus Torvalds { 15976fa5f80bSDavid Howells struct vm_area_struct *vma; 15981da177e4SLinus Torvalds 15991da177e4SLinus Torvalds /* insanity checks first */ 1600f67d9b15SBob Liu old_len = PAGE_ALIGN(old_len); 1601f67d9b15SBob Liu new_len = PAGE_ALIGN(new_len); 16028feae131SDavid Howells if (old_len == 0 || new_len == 0) 16031da177e4SLinus Torvalds return (unsigned long) -EINVAL; 16041da177e4SLinus Torvalds 16051824cb75SAlexander Kuleshov if (offset_in_page(addr)) 16068feae131SDavid Howells return -EINVAL; 16078feae131SDavid Howells 16081da177e4SLinus Torvalds if (flags & MREMAP_FIXED && new_addr != addr) 16091da177e4SLinus Torvalds return (unsigned long) -EINVAL; 16101da177e4SLinus Torvalds 16118feae131SDavid Howells vma = find_vma_exact(current->mm, addr, old_len); 16126fa5f80bSDavid Howells if (!vma) 16131da177e4SLinus Torvalds return (unsigned long) -EINVAL; 16141da177e4SLinus Torvalds 16156fa5f80bSDavid Howells if (vma->vm_end != vma->vm_start + old_len) 16161da177e4SLinus Torvalds return (unsigned long) -EFAULT; 16171da177e4SLinus Torvalds 16186fa5f80bSDavid Howells if (vma->vm_flags & VM_MAYSHARE) 16191da177e4SLinus Torvalds return (unsigned long) -EPERM; 16201da177e4SLinus Torvalds 16218feae131SDavid Howells if (new_len > vma->vm_region->vm_end - vma->vm_region->vm_start) 16221da177e4SLinus Torvalds return (unsigned long) -ENOMEM; 16231da177e4SLinus Torvalds 16241da177e4SLinus Torvalds /* all checks complete - do it */ 16256fa5f80bSDavid Howells vma->vm_end = vma->vm_start + new_len; 16266fa5f80bSDavid Howells return vma->vm_start; 16276fa5f80bSDavid Howells } 16286fa5f80bSDavid Howells 16296a6160a7SHeiko Carstens SYSCALL_DEFINE5(mremap, unsigned long, addr, unsigned long, old_len, 16306a6160a7SHeiko Carstens unsigned long, new_len, unsigned long, flags, 16316a6160a7SHeiko Carstens unsigned long, new_addr) 16326fa5f80bSDavid Howells { 16336fa5f80bSDavid Howells unsigned long ret; 16346fa5f80bSDavid Howells 16356fa5f80bSDavid Howells down_write(¤t->mm->mmap_sem); 16366fa5f80bSDavid Howells ret = do_mremap(addr, old_len, new_len, flags, new_addr); 16376fa5f80bSDavid Howells up_write(¤t->mm->mmap_sem); 16386fa5f80bSDavid Howells return ret; 16391da177e4SLinus Torvalds } 16401da177e4SLinus Torvalds 1641df06b37fSKeith Busch struct page *follow_page(struct vm_area_struct *vma, unsigned long address, 1642df06b37fSKeith Busch unsigned int foll_flags) 16431da177e4SLinus Torvalds { 16441da177e4SLinus Torvalds return NULL; 16451da177e4SLinus Torvalds } 16461da177e4SLinus Torvalds 16478f3b1327SBob Liu int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr, 16488f3b1327SBob Liu unsigned long pfn, unsigned long size, pgprot_t prot) 16491da177e4SLinus Torvalds { 16508f3b1327SBob Liu if (addr != (pfn << PAGE_SHIFT)) 16518f3b1327SBob Liu return -EINVAL; 16528f3b1327SBob Liu 1653314e51b9SKonstantin Khlebnikov vma->vm_flags |= VM_IO | VM_PFNMAP | VM_DONTEXPAND | VM_DONTDUMP; 165466aa2b4bSGreg Ungerer return 0; 16551da177e4SLinus Torvalds } 165622c4af40SLuke Yang EXPORT_SYMBOL(remap_pfn_range); 16571da177e4SLinus Torvalds 16583c0b9de6SLinus Torvalds int vm_iomap_memory(struct vm_area_struct *vma, phys_addr_t start, unsigned long len) 16593c0b9de6SLinus Torvalds { 16603c0b9de6SLinus Torvalds unsigned long pfn = start >> PAGE_SHIFT; 16613c0b9de6SLinus Torvalds unsigned long vm_len = vma->vm_end - vma->vm_start; 16623c0b9de6SLinus Torvalds 16633c0b9de6SLinus Torvalds pfn += vma->vm_pgoff; 16643c0b9de6SLinus Torvalds return io_remap_pfn_range(vma, vma->vm_start, pfn, vm_len, vma->vm_page_prot); 16653c0b9de6SLinus Torvalds } 16663c0b9de6SLinus Torvalds EXPORT_SYMBOL(vm_iomap_memory); 16673c0b9de6SLinus Torvalds 1668f905bc44SPaul Mundt int remap_vmalloc_range(struct vm_area_struct *vma, void *addr, 1669f905bc44SPaul Mundt unsigned long pgoff) 1670f905bc44SPaul Mundt { 1671f905bc44SPaul Mundt unsigned int size = vma->vm_end - vma->vm_start; 1672f905bc44SPaul Mundt 1673f905bc44SPaul Mundt if (!(vma->vm_flags & VM_USERMAP)) 1674f905bc44SPaul Mundt return -EINVAL; 1675f905bc44SPaul Mundt 1676f905bc44SPaul Mundt vma->vm_start = (unsigned long)(addr + (pgoff << PAGE_SHIFT)); 1677f905bc44SPaul Mundt vma->vm_end = vma->vm_start + size; 1678f905bc44SPaul Mundt 1679f905bc44SPaul Mundt return 0; 1680f905bc44SPaul Mundt } 1681f905bc44SPaul Mundt EXPORT_SYMBOL(remap_vmalloc_range); 1682f905bc44SPaul Mundt 16831da177e4SLinus Torvalds unsigned long arch_get_unmapped_area(struct file *file, unsigned long addr, 16841da177e4SLinus Torvalds unsigned long len, unsigned long pgoff, unsigned long flags) 16851da177e4SLinus Torvalds { 16861da177e4SLinus Torvalds return -ENOMEM; 16871da177e4SLinus Torvalds } 16881da177e4SLinus Torvalds 16892bcd6454SSouptick Joarder vm_fault_t filemap_fault(struct vm_fault *vmf) 1690b0e15190SDavid Howells { 1691b0e15190SDavid Howells BUG(); 1692d0217ac0SNick Piggin return 0; 1693b0e15190SDavid Howells } 1694b5073173SPaul Mundt EXPORT_SYMBOL(filemap_fault); 16950ec76a11SDavid Howells 169682b0f8c3SJan Kara void filemap_map_pages(struct vm_fault *vmf, 1697bae473a4SKirill A. Shutemov pgoff_t start_pgoff, pgoff_t end_pgoff) 1698f1820361SKirill A. Shutemov { 1699f1820361SKirill A. Shutemov BUG(); 1700f1820361SKirill A. Shutemov } 1701f1820361SKirill A. Shutemov EXPORT_SYMBOL(filemap_map_pages); 1702f1820361SKirill A. Shutemov 170384d77d3fSEric W. Biederman int __access_remote_vm(struct task_struct *tsk, struct mm_struct *mm, 1704442486ecSLorenzo Stoakes unsigned long addr, void *buf, int len, unsigned int gup_flags) 17050ec76a11SDavid Howells { 17060ec76a11SDavid Howells struct vm_area_struct *vma; 1707442486ecSLorenzo Stoakes int write = gup_flags & FOLL_WRITE; 17080ec76a11SDavid Howells 17091e426fe2SKonstantin Khlebnikov if (down_read_killable(&mm->mmap_sem)) 17101e426fe2SKonstantin Khlebnikov return 0; 17110ec76a11SDavid Howells 17120ec76a11SDavid Howells /* the access must start within one of the target process's mappings */ 17130159b141SDavid Howells vma = find_vma(mm, addr); 17140159b141SDavid Howells if (vma) { 17150ec76a11SDavid Howells /* don't overrun this mapping */ 17160ec76a11SDavid Howells if (addr + len >= vma->vm_end) 17170ec76a11SDavid Howells len = vma->vm_end - addr; 17180ec76a11SDavid Howells 17190ec76a11SDavid Howells /* only read or write mappings where it is permitted */ 1720d00c7b99SDavid Howells if (write && vma->vm_flags & VM_MAYWRITE) 17217959722bSJie Zhang copy_to_user_page(vma, NULL, addr, 17227959722bSJie Zhang (void *) addr, buf, len); 1723d00c7b99SDavid Howells else if (!write && vma->vm_flags & VM_MAYREAD) 17247959722bSJie Zhang copy_from_user_page(vma, NULL, addr, 17257959722bSJie Zhang buf, (void *) addr, len); 17260ec76a11SDavid Howells else 17270ec76a11SDavid Howells len = 0; 17280ec76a11SDavid Howells } else { 17290ec76a11SDavid Howells len = 0; 17300ec76a11SDavid Howells } 17310ec76a11SDavid Howells 17320ec76a11SDavid Howells up_read(&mm->mmap_sem); 1733f55f199bSMike Frysinger 1734f55f199bSMike Frysinger return len; 1735f55f199bSMike Frysinger } 1736f55f199bSMike Frysinger 1737f55f199bSMike Frysinger /** 1738b7701a5fSMike Rapoport * access_remote_vm - access another process' address space 1739f55f199bSMike Frysinger * @mm: the mm_struct of the target address space 1740f55f199bSMike Frysinger * @addr: start address to access 1741f55f199bSMike Frysinger * @buf: source or destination buffer 1742f55f199bSMike Frysinger * @len: number of bytes to transfer 17436347e8d5SLorenzo Stoakes * @gup_flags: flags modifying lookup behaviour 1744f55f199bSMike Frysinger * 1745f55f199bSMike Frysinger * The caller must hold a reference on @mm. 1746f55f199bSMike Frysinger */ 1747f55f199bSMike Frysinger int access_remote_vm(struct mm_struct *mm, unsigned long addr, 17486347e8d5SLorenzo Stoakes void *buf, int len, unsigned int gup_flags) 1749f55f199bSMike Frysinger { 17506347e8d5SLorenzo Stoakes return __access_remote_vm(NULL, mm, addr, buf, len, gup_flags); 1751f55f199bSMike Frysinger } 1752f55f199bSMike Frysinger 1753f55f199bSMike Frysinger /* 1754f55f199bSMike Frysinger * Access another process' address space. 1755f55f199bSMike Frysinger * - source/target buffer must be kernel space 1756f55f199bSMike Frysinger */ 1757f307ab6dSLorenzo Stoakes int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len, 1758f307ab6dSLorenzo Stoakes unsigned int gup_flags) 1759f55f199bSMike Frysinger { 1760f55f199bSMike Frysinger struct mm_struct *mm; 1761f55f199bSMike Frysinger 1762f55f199bSMike Frysinger if (addr + len < addr) 1763f55f199bSMike Frysinger return 0; 1764f55f199bSMike Frysinger 1765f55f199bSMike Frysinger mm = get_task_mm(tsk); 1766f55f199bSMike Frysinger if (!mm) 1767f55f199bSMike Frysinger return 0; 1768f55f199bSMike Frysinger 1769f307ab6dSLorenzo Stoakes len = __access_remote_vm(tsk, mm, addr, buf, len, gup_flags); 1770f55f199bSMike Frysinger 17710ec76a11SDavid Howells mmput(mm); 17720ec76a11SDavid Howells return len; 17730ec76a11SDavid Howells } 1774fcd35857SCatalin Marinas EXPORT_SYMBOL_GPL(access_process_vm); 17757e660872SDavid Howells 17767e660872SDavid Howells /** 17777e660872SDavid Howells * nommu_shrink_inode_mappings - Shrink the shared mappings on an inode 17787e660872SDavid Howells * @inode: The inode to check 17797e660872SDavid Howells * @size: The current filesize of the inode 17807e660872SDavid Howells * @newsize: The proposed filesize of the inode 17817e660872SDavid Howells * 17827e660872SDavid Howells * Check the shared mappings on an inode on behalf of a shrinking truncate to 17837e660872SDavid Howells * make sure that that any outstanding VMAs aren't broken and then shrink the 17847e660872SDavid Howells * vm_regions that extend that beyond so that do_mmap_pgoff() doesn't 17857e660872SDavid Howells * automatically grant mappings that are too large. 17867e660872SDavid Howells */ 17877e660872SDavid Howells int nommu_shrink_inode_mappings(struct inode *inode, size_t size, 17887e660872SDavid Howells size_t newsize) 17897e660872SDavid Howells { 17907e660872SDavid Howells struct vm_area_struct *vma; 17917e660872SDavid Howells struct vm_region *region; 17927e660872SDavid Howells pgoff_t low, high; 17937e660872SDavid Howells size_t r_size, r_top; 17947e660872SDavid Howells 17957e660872SDavid Howells low = newsize >> PAGE_SHIFT; 17967e660872SDavid Howells high = (size + PAGE_SIZE - 1) >> PAGE_SHIFT; 17977e660872SDavid Howells 17987e660872SDavid Howells down_write(&nommu_region_sem); 17991acf2e04SDavidlohr Bueso i_mmap_lock_read(inode->i_mapping); 18007e660872SDavid Howells 18017e660872SDavid Howells /* search for VMAs that fall within the dead zone */ 18026b2dbba8SMichel Lespinasse vma_interval_tree_foreach(vma, &inode->i_mapping->i_mmap, low, high) { 18037e660872SDavid Howells /* found one - only interested if it's shared out of the page 18047e660872SDavid Howells * cache */ 18057e660872SDavid Howells if (vma->vm_flags & VM_SHARED) { 18061acf2e04SDavidlohr Bueso i_mmap_unlock_read(inode->i_mapping); 18077e660872SDavid Howells up_write(&nommu_region_sem); 18087e660872SDavid Howells return -ETXTBSY; /* not quite true, but near enough */ 18097e660872SDavid Howells } 18107e660872SDavid Howells } 18117e660872SDavid Howells 18127e660872SDavid Howells /* reduce any regions that overlap the dead zone - if in existence, 18137e660872SDavid Howells * these will be pointed to by VMAs that don't overlap the dead zone 18147e660872SDavid Howells * 18157e660872SDavid Howells * we don't check for any regions that start beyond the EOF as there 18167e660872SDavid Howells * shouldn't be any 18177e660872SDavid Howells */ 18181acf2e04SDavidlohr Bueso vma_interval_tree_foreach(vma, &inode->i_mapping->i_mmap, 0, ULONG_MAX) { 18197e660872SDavid Howells if (!(vma->vm_flags & VM_SHARED)) 18207e660872SDavid Howells continue; 18217e660872SDavid Howells 18227e660872SDavid Howells region = vma->vm_region; 18237e660872SDavid Howells r_size = region->vm_top - region->vm_start; 18247e660872SDavid Howells r_top = (region->vm_pgoff << PAGE_SHIFT) + r_size; 18257e660872SDavid Howells 18267e660872SDavid Howells if (r_top > newsize) { 18277e660872SDavid Howells region->vm_top -= r_top - newsize; 18287e660872SDavid Howells if (region->vm_end > region->vm_top) 18297e660872SDavid Howells region->vm_end = region->vm_top; 18307e660872SDavid Howells } 18317e660872SDavid Howells } 18327e660872SDavid Howells 18331acf2e04SDavidlohr Bueso i_mmap_unlock_read(inode->i_mapping); 18347e660872SDavid Howells up_write(&nommu_region_sem); 18357e660872SDavid Howells return 0; 18367e660872SDavid Howells } 1837c9b1d098SAndrew Shewmaker 1838c9b1d098SAndrew Shewmaker /* 1839c9b1d098SAndrew Shewmaker * Initialise sysctl_user_reserve_kbytes. 1840c9b1d098SAndrew Shewmaker * 1841c9b1d098SAndrew Shewmaker * This is intended to prevent a user from starting a single memory hogging 1842c9b1d098SAndrew Shewmaker * process, such that they cannot recover (kill the hog) in OVERCOMMIT_NEVER 1843c9b1d098SAndrew Shewmaker * mode. 1844c9b1d098SAndrew Shewmaker * 1845c9b1d098SAndrew Shewmaker * The default value is min(3% of free memory, 128MB) 1846c9b1d098SAndrew Shewmaker * 128MB is enough to recover with sshd/login, bash, and top/kill. 1847c9b1d098SAndrew Shewmaker */ 1848c9b1d098SAndrew Shewmaker static int __meminit init_user_reserve(void) 1849c9b1d098SAndrew Shewmaker { 1850c9b1d098SAndrew Shewmaker unsigned long free_kbytes; 1851c9b1d098SAndrew Shewmaker 1852c41f012aSMichal Hocko free_kbytes = global_zone_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10); 1853c9b1d098SAndrew Shewmaker 1854c9b1d098SAndrew Shewmaker sysctl_user_reserve_kbytes = min(free_kbytes / 32, 1UL << 17); 1855c9b1d098SAndrew Shewmaker return 0; 1856c9b1d098SAndrew Shewmaker } 1857a4bc6fc7SPaul Gortmaker subsys_initcall(init_user_reserve); 18584eeab4f5SAndrew Shewmaker 18594eeab4f5SAndrew Shewmaker /* 18604eeab4f5SAndrew Shewmaker * Initialise sysctl_admin_reserve_kbytes. 18614eeab4f5SAndrew Shewmaker * 18624eeab4f5SAndrew Shewmaker * The purpose of sysctl_admin_reserve_kbytes is to allow the sys admin 18634eeab4f5SAndrew Shewmaker * to log in and kill a memory hogging process. 18644eeab4f5SAndrew Shewmaker * 18654eeab4f5SAndrew Shewmaker * Systems with more than 256MB will reserve 8MB, enough to recover 18664eeab4f5SAndrew Shewmaker * with sshd, bash, and top in OVERCOMMIT_GUESS. Smaller systems will 18674eeab4f5SAndrew Shewmaker * only reserve 3% of free pages by default. 18684eeab4f5SAndrew Shewmaker */ 18694eeab4f5SAndrew Shewmaker static int __meminit init_admin_reserve(void) 18704eeab4f5SAndrew Shewmaker { 18714eeab4f5SAndrew Shewmaker unsigned long free_kbytes; 18724eeab4f5SAndrew Shewmaker 1873c41f012aSMichal Hocko free_kbytes = global_zone_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10); 18744eeab4f5SAndrew Shewmaker 18754eeab4f5SAndrew Shewmaker sysctl_admin_reserve_kbytes = min(free_kbytes / 32, 1UL << 13); 18764eeab4f5SAndrew Shewmaker return 0; 18774eeab4f5SAndrew Shewmaker } 1878a4bc6fc7SPaul Gortmaker subsys_initcall(init_admin_reserve); 1879