11da177e4SLinus Torvalds /* 21da177e4SLinus Torvalds * linux/mm/nommu.c 31da177e4SLinus Torvalds * 41da177e4SLinus Torvalds * Replacement code for mm functions to support CPU's that don't 51da177e4SLinus Torvalds * have any form of memory management unit (thus no virtual memory). 61da177e4SLinus Torvalds * 71da177e4SLinus Torvalds * See Documentation/nommu-mmap.txt 81da177e4SLinus Torvalds * 98feae131SDavid Howells * Copyright (c) 2004-2008 David Howells <dhowells@redhat.com> 101da177e4SLinus Torvalds * Copyright (c) 2000-2003 David McCullough <davidm@snapgear.com> 111da177e4SLinus Torvalds * Copyright (c) 2000-2001 D Jeff Dionne <jeff@uClinux.org> 121da177e4SLinus Torvalds * Copyright (c) 2002 Greg Ungerer <gerg@snapgear.com> 1329c185e5SPaul Mundt * Copyright (c) 2007-2010 Paul Mundt <lethal@linux-sh.org> 141da177e4SLinus Torvalds */ 151da177e4SLinus Torvalds 16*b1de0d13SMitchel Humpherys #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt 17*b1de0d13SMitchel Humpherys 18b95f1b31SPaul Gortmaker #include <linux/export.h> 191da177e4SLinus Torvalds #include <linux/mm.h> 20615d6e87SDavidlohr Bueso #include <linux/vmacache.h> 211da177e4SLinus Torvalds #include <linux/mman.h> 221da177e4SLinus Torvalds #include <linux/swap.h> 231da177e4SLinus Torvalds #include <linux/file.h> 241da177e4SLinus Torvalds #include <linux/highmem.h> 251da177e4SLinus Torvalds #include <linux/pagemap.h> 261da177e4SLinus Torvalds #include <linux/slab.h> 271da177e4SLinus Torvalds #include <linux/vmalloc.h> 281da177e4SLinus Torvalds #include <linux/blkdev.h> 291da177e4SLinus Torvalds #include <linux/backing-dev.h> 303b32123dSGideon Israel Dsouza #include <linux/compiler.h> 311da177e4SLinus Torvalds #include <linux/mount.h> 321da177e4SLinus Torvalds #include <linux/personality.h> 331da177e4SLinus Torvalds #include <linux/security.h> 341da177e4SLinus Torvalds #include <linux/syscalls.h> 35120a795dSAl Viro #include <linux/audit.h> 36cf4aebc2SClark Williams #include <linux/sched/sysctl.h> 37*b1de0d13SMitchel Humpherys #include <linux/printk.h> 381da177e4SLinus Torvalds 391da177e4SLinus Torvalds #include <asm/uaccess.h> 401da177e4SLinus Torvalds #include <asm/tlb.h> 411da177e4SLinus Torvalds #include <asm/tlbflush.h> 42eb8cdec4SBernd Schmidt #include <asm/mmu_context.h> 438feae131SDavid Howells #include "internal.h" 448feae131SDavid Howells 458feae131SDavid Howells #if 0 468feae131SDavid Howells #define kenter(FMT, ...) \ 478feae131SDavid Howells printk(KERN_DEBUG "==> %s("FMT")\n", __func__, ##__VA_ARGS__) 488feae131SDavid Howells #define kleave(FMT, ...) \ 498feae131SDavid Howells printk(KERN_DEBUG "<== %s()"FMT"\n", __func__, ##__VA_ARGS__) 508feae131SDavid Howells #define kdebug(FMT, ...) \ 518feae131SDavid Howells printk(KERN_DEBUG "xxx" FMT"yyy\n", ##__VA_ARGS__) 528feae131SDavid Howells #else 538feae131SDavid Howells #define kenter(FMT, ...) \ 548feae131SDavid Howells no_printk(KERN_DEBUG "==> %s("FMT")\n", __func__, ##__VA_ARGS__) 558feae131SDavid Howells #define kleave(FMT, ...) \ 568feae131SDavid Howells no_printk(KERN_DEBUG "<== %s()"FMT"\n", __func__, ##__VA_ARGS__) 578feae131SDavid Howells #define kdebug(FMT, ...) \ 588feae131SDavid Howells no_printk(KERN_DEBUG FMT"\n", ##__VA_ARGS__) 598feae131SDavid Howells #endif 601da177e4SLinus Torvalds 611da177e4SLinus Torvalds void *high_memory; 621da177e4SLinus Torvalds struct page *mem_map; 631da177e4SLinus Torvalds unsigned long max_mapnr; 644266c97aSHugh Dickins unsigned long highest_memmap_pfn; 6500a62ce9SKOSAKI Motohiro struct percpu_counter vm_committed_as; 661da177e4SLinus Torvalds int sysctl_overcommit_memory = OVERCOMMIT_GUESS; /* heuristic overcommit */ 671da177e4SLinus Torvalds int sysctl_overcommit_ratio = 50; /* default is 50% */ 6849f0ce5fSJerome Marchand unsigned long sysctl_overcommit_kbytes __read_mostly; 691da177e4SLinus Torvalds int sysctl_max_map_count = DEFAULT_MAX_MAP_COUNT; 70fc4d5c29SDavid Howells int sysctl_nr_trim_pages = CONFIG_NOMMU_INITIAL_TRIM_EXCESS; 71c9b1d098SAndrew Shewmaker unsigned long sysctl_user_reserve_kbytes __read_mostly = 1UL << 17; /* 128MB */ 724eeab4f5SAndrew Shewmaker unsigned long sysctl_admin_reserve_kbytes __read_mostly = 1UL << 13; /* 8MB */ 731da177e4SLinus Torvalds int heap_stack_gap = 0; 741da177e4SLinus Torvalds 7533e5d769SDavid Howells atomic_long_t mmap_pages_allocated; 768feae131SDavid Howells 77997071bcSK. Y. Srinivasan /* 78997071bcSK. Y. Srinivasan * The global memory commitment made in the system can be a metric 79997071bcSK. Y. Srinivasan * that can be used to drive ballooning decisions when Linux is hosted 80997071bcSK. Y. Srinivasan * as a guest. On Hyper-V, the host implements a policy engine for dynamically 81997071bcSK. Y. Srinivasan * balancing memory across competing virtual machines that are hosted. 82997071bcSK. Y. Srinivasan * Several metrics drive this policy engine including the guest reported 83997071bcSK. Y. Srinivasan * memory commitment. 84997071bcSK. Y. Srinivasan */ 85997071bcSK. Y. Srinivasan unsigned long vm_memory_committed(void) 86997071bcSK. Y. Srinivasan { 87997071bcSK. Y. Srinivasan return percpu_counter_read_positive(&vm_committed_as); 88997071bcSK. Y. Srinivasan } 89997071bcSK. Y. Srinivasan 90997071bcSK. Y. Srinivasan EXPORT_SYMBOL_GPL(vm_memory_committed); 91997071bcSK. Y. Srinivasan 921da177e4SLinus Torvalds EXPORT_SYMBOL(mem_map); 931da177e4SLinus Torvalds 948feae131SDavid Howells /* list of mapped, potentially shareable regions */ 958feae131SDavid Howells static struct kmem_cache *vm_region_jar; 968feae131SDavid Howells struct rb_root nommu_region_tree = RB_ROOT; 978feae131SDavid Howells DECLARE_RWSEM(nommu_region_sem); 981da177e4SLinus Torvalds 99f0f37e2fSAlexey Dobriyan const struct vm_operations_struct generic_file_vm_ops = { 1001da177e4SLinus Torvalds }; 1011da177e4SLinus Torvalds 1021da177e4SLinus Torvalds /* 1031da177e4SLinus Torvalds * Return the total memory allocated for this pointer, not 1041da177e4SLinus Torvalds * just what the caller asked for. 1051da177e4SLinus Torvalds * 1061da177e4SLinus Torvalds * Doesn't have to be accurate, i.e. may have races. 1071da177e4SLinus Torvalds */ 1081da177e4SLinus Torvalds unsigned int kobjsize(const void *objp) 1091da177e4SLinus Torvalds { 1101da177e4SLinus Torvalds struct page *page; 1111da177e4SLinus Torvalds 1124016a139SMichael Hennerich /* 1134016a139SMichael Hennerich * If the object we have should not have ksize performed on it, 1144016a139SMichael Hennerich * return size of 0 1154016a139SMichael Hennerich */ 1165a1603beSPaul Mundt if (!objp || !virt_addr_valid(objp)) 1176cfd53fcSPaul Mundt return 0; 1186cfd53fcSPaul Mundt 1196cfd53fcSPaul Mundt page = virt_to_head_page(objp); 1206cfd53fcSPaul Mundt 1216cfd53fcSPaul Mundt /* 1226cfd53fcSPaul Mundt * If the allocator sets PageSlab, we know the pointer came from 1236cfd53fcSPaul Mundt * kmalloc(). 1246cfd53fcSPaul Mundt */ 1251da177e4SLinus Torvalds if (PageSlab(page)) 1261da177e4SLinus Torvalds return ksize(objp); 1271da177e4SLinus Torvalds 1286cfd53fcSPaul Mundt /* 129ab2e83eaSPaul Mundt * If it's not a compound page, see if we have a matching VMA 130ab2e83eaSPaul Mundt * region. This test is intentionally done in reverse order, 131ab2e83eaSPaul Mundt * so if there's no VMA, we still fall through and hand back 132ab2e83eaSPaul Mundt * PAGE_SIZE for 0-order pages. 133ab2e83eaSPaul Mundt */ 134ab2e83eaSPaul Mundt if (!PageCompound(page)) { 135ab2e83eaSPaul Mundt struct vm_area_struct *vma; 136ab2e83eaSPaul Mundt 137ab2e83eaSPaul Mundt vma = find_vma(current->mm, (unsigned long)objp); 138ab2e83eaSPaul Mundt if (vma) 139ab2e83eaSPaul Mundt return vma->vm_end - vma->vm_start; 140ab2e83eaSPaul Mundt } 141ab2e83eaSPaul Mundt 142ab2e83eaSPaul Mundt /* 1436cfd53fcSPaul Mundt * The ksize() function is only guaranteed to work for pointers 1445a1603beSPaul Mundt * returned by kmalloc(). So handle arbitrary pointers here. 1456cfd53fcSPaul Mundt */ 1465a1603beSPaul Mundt return PAGE_SIZE << compound_order(page); 1471da177e4SLinus Torvalds } 1481da177e4SLinus Torvalds 14928a35716SMichel Lespinasse long __get_user_pages(struct task_struct *tsk, struct mm_struct *mm, 15028a35716SMichel Lespinasse unsigned long start, unsigned long nr_pages, 15128a35716SMichel Lespinasse unsigned int foll_flags, struct page **pages, 15228a35716SMichel Lespinasse struct vm_area_struct **vmas, int *nonblocking) 1531da177e4SLinus Torvalds { 154910e46daSSonic Zhang struct vm_area_struct *vma; 1557b4d5b8bSDavid Howells unsigned long vm_flags; 1567b4d5b8bSDavid Howells int i; 1577b4d5b8bSDavid Howells 1587b4d5b8bSDavid Howells /* calculate required read or write permissions. 15958fa879eSHugh Dickins * If FOLL_FORCE is set, we only require the "MAY" flags. 1607b4d5b8bSDavid Howells */ 16158fa879eSHugh Dickins vm_flags = (foll_flags & FOLL_WRITE) ? 16258fa879eSHugh Dickins (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD); 16358fa879eSHugh Dickins vm_flags &= (foll_flags & FOLL_FORCE) ? 16458fa879eSHugh Dickins (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE); 1651da177e4SLinus Torvalds 1669d73777eSPeter Zijlstra for (i = 0; i < nr_pages; i++) { 1677561e8caSDavid Howells vma = find_vma(mm, start); 168910e46daSSonic Zhang if (!vma) 1697b4d5b8bSDavid Howells goto finish_or_fault; 1707b4d5b8bSDavid Howells 1717b4d5b8bSDavid Howells /* protect what we can, including chardevs */ 1721c3aff1cSHugh Dickins if ((vma->vm_flags & (VM_IO | VM_PFNMAP)) || 1731c3aff1cSHugh Dickins !(vm_flags & vma->vm_flags)) 1747b4d5b8bSDavid Howells goto finish_or_fault; 175910e46daSSonic Zhang 1761da177e4SLinus Torvalds if (pages) { 1771da177e4SLinus Torvalds pages[i] = virt_to_page(start); 1781da177e4SLinus Torvalds if (pages[i]) 1791da177e4SLinus Torvalds page_cache_get(pages[i]); 1801da177e4SLinus Torvalds } 1811da177e4SLinus Torvalds if (vmas) 182910e46daSSonic Zhang vmas[i] = vma; 183e1ee65d8SDavid Howells start = (start + PAGE_SIZE) & PAGE_MASK; 1841da177e4SLinus Torvalds } 1857b4d5b8bSDavid Howells 1867b4d5b8bSDavid Howells return i; 1877b4d5b8bSDavid Howells 1887b4d5b8bSDavid Howells finish_or_fault: 1897b4d5b8bSDavid Howells return i ? : -EFAULT; 1901da177e4SLinus Torvalds } 191b291f000SNick Piggin 192b291f000SNick Piggin /* 193b291f000SNick Piggin * get a list of pages in an address range belonging to the specified process 194b291f000SNick Piggin * and indicate the VMA that covers each page 195b291f000SNick Piggin * - this is potentially dodgy as we may end incrementing the page count of a 196b291f000SNick Piggin * slab page or a secondary page from a compound page 197b291f000SNick Piggin * - don't permit access to VMAs that don't support it, such as I/O mappings 198b291f000SNick Piggin */ 19928a35716SMichel Lespinasse long get_user_pages(struct task_struct *tsk, struct mm_struct *mm, 20028a35716SMichel Lespinasse unsigned long start, unsigned long nr_pages, 20128a35716SMichel Lespinasse int write, int force, struct page **pages, 20228a35716SMichel Lespinasse struct vm_area_struct **vmas) 203b291f000SNick Piggin { 204b291f000SNick Piggin int flags = 0; 205b291f000SNick Piggin 206b291f000SNick Piggin if (write) 20758fa879eSHugh Dickins flags |= FOLL_WRITE; 208b291f000SNick Piggin if (force) 20958fa879eSHugh Dickins flags |= FOLL_FORCE; 210b291f000SNick Piggin 21153a7706dSMichel Lespinasse return __get_user_pages(tsk, mm, start, nr_pages, flags, pages, vmas, 21253a7706dSMichel Lespinasse NULL); 213b291f000SNick Piggin } 21466aa2b4bSGreg Ungerer EXPORT_SYMBOL(get_user_pages); 21566aa2b4bSGreg Ungerer 216dfc2f91aSPaul Mundt /** 217dfc2f91aSPaul Mundt * follow_pfn - look up PFN at a user virtual address 218dfc2f91aSPaul Mundt * @vma: memory mapping 219dfc2f91aSPaul Mundt * @address: user virtual address 220dfc2f91aSPaul Mundt * @pfn: location to store found PFN 221dfc2f91aSPaul Mundt * 222dfc2f91aSPaul Mundt * Only IO mappings and raw PFN mappings are allowed. 223dfc2f91aSPaul Mundt * 224dfc2f91aSPaul Mundt * Returns zero and the pfn at @pfn on success, -ve otherwise. 225dfc2f91aSPaul Mundt */ 226dfc2f91aSPaul Mundt int follow_pfn(struct vm_area_struct *vma, unsigned long address, 227dfc2f91aSPaul Mundt unsigned long *pfn) 228dfc2f91aSPaul Mundt { 229dfc2f91aSPaul Mundt if (!(vma->vm_flags & (VM_IO | VM_PFNMAP))) 230dfc2f91aSPaul Mundt return -EINVAL; 231dfc2f91aSPaul Mundt 232dfc2f91aSPaul Mundt *pfn = address >> PAGE_SHIFT; 233dfc2f91aSPaul Mundt return 0; 234dfc2f91aSPaul Mundt } 235dfc2f91aSPaul Mundt EXPORT_SYMBOL(follow_pfn); 236dfc2f91aSPaul Mundt 237f1c4069eSJoonsoo Kim LIST_HEAD(vmap_area_list); 2381da177e4SLinus Torvalds 239b3bdda02SChristoph Lameter void vfree(const void *addr) 2401da177e4SLinus Torvalds { 2411da177e4SLinus Torvalds kfree(addr); 2421da177e4SLinus Torvalds } 243b5073173SPaul Mundt EXPORT_SYMBOL(vfree); 2441da177e4SLinus Torvalds 245dd0fc66fSAl Viro void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot) 2461da177e4SLinus Torvalds { 2471da177e4SLinus Torvalds /* 2488518609dSRobert P. J. Day * You can't specify __GFP_HIGHMEM with kmalloc() since kmalloc() 2498518609dSRobert P. J. Day * returns only a logical address. 2501da177e4SLinus Torvalds */ 25184097518SNick Piggin return kmalloc(size, (gfp_mask | __GFP_COMP) & ~__GFP_HIGHMEM); 2521da177e4SLinus Torvalds } 253b5073173SPaul Mundt EXPORT_SYMBOL(__vmalloc); 2541da177e4SLinus Torvalds 255f905bc44SPaul Mundt void *vmalloc_user(unsigned long size) 256f905bc44SPaul Mundt { 257f905bc44SPaul Mundt void *ret; 258f905bc44SPaul Mundt 259f905bc44SPaul Mundt ret = __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO, 260f905bc44SPaul Mundt PAGE_KERNEL); 261f905bc44SPaul Mundt if (ret) { 262f905bc44SPaul Mundt struct vm_area_struct *vma; 263f905bc44SPaul Mundt 264f905bc44SPaul Mundt down_write(¤t->mm->mmap_sem); 265f905bc44SPaul Mundt vma = find_vma(current->mm, (unsigned long)ret); 266f905bc44SPaul Mundt if (vma) 267f905bc44SPaul Mundt vma->vm_flags |= VM_USERMAP; 268f905bc44SPaul Mundt up_write(¤t->mm->mmap_sem); 269f905bc44SPaul Mundt } 270f905bc44SPaul Mundt 271f905bc44SPaul Mundt return ret; 272f905bc44SPaul Mundt } 273f905bc44SPaul Mundt EXPORT_SYMBOL(vmalloc_user); 274f905bc44SPaul Mundt 275b3bdda02SChristoph Lameter struct page *vmalloc_to_page(const void *addr) 2761da177e4SLinus Torvalds { 2771da177e4SLinus Torvalds return virt_to_page(addr); 2781da177e4SLinus Torvalds } 279b5073173SPaul Mundt EXPORT_SYMBOL(vmalloc_to_page); 2801da177e4SLinus Torvalds 281b3bdda02SChristoph Lameter unsigned long vmalloc_to_pfn(const void *addr) 2821da177e4SLinus Torvalds { 2831da177e4SLinus Torvalds return page_to_pfn(virt_to_page(addr)); 2841da177e4SLinus Torvalds } 285b5073173SPaul Mundt EXPORT_SYMBOL(vmalloc_to_pfn); 2861da177e4SLinus Torvalds 2871da177e4SLinus Torvalds long vread(char *buf, char *addr, unsigned long count) 2881da177e4SLinus Torvalds { 2899bde916bSChen Gang /* Don't allow overflow */ 2909bde916bSChen Gang if ((unsigned long) buf + count < count) 2919bde916bSChen Gang count = -(unsigned long) buf; 2929bde916bSChen Gang 2931da177e4SLinus Torvalds memcpy(buf, addr, count); 2941da177e4SLinus Torvalds return count; 2951da177e4SLinus Torvalds } 2961da177e4SLinus Torvalds 2971da177e4SLinus Torvalds long vwrite(char *buf, char *addr, unsigned long count) 2981da177e4SLinus Torvalds { 2991da177e4SLinus Torvalds /* Don't allow overflow */ 3001da177e4SLinus Torvalds if ((unsigned long) addr + count < count) 3011da177e4SLinus Torvalds count = -(unsigned long) addr; 3021da177e4SLinus Torvalds 3031da177e4SLinus Torvalds memcpy(addr, buf, count); 304ac714904SChoi Gi-yong return count; 3051da177e4SLinus Torvalds } 3061da177e4SLinus Torvalds 3071da177e4SLinus Torvalds /* 3081da177e4SLinus Torvalds * vmalloc - allocate virtually continguos memory 3091da177e4SLinus Torvalds * 3101da177e4SLinus Torvalds * @size: allocation size 3111da177e4SLinus Torvalds * 3121da177e4SLinus Torvalds * Allocate enough pages to cover @size from the page level 3131da177e4SLinus Torvalds * allocator and map them into continguos kernel virtual space. 3141da177e4SLinus Torvalds * 315c1c8897fSMichael Opdenacker * For tight control over page level allocator and protection flags 3161da177e4SLinus Torvalds * use __vmalloc() instead. 3171da177e4SLinus Torvalds */ 3181da177e4SLinus Torvalds void *vmalloc(unsigned long size) 3191da177e4SLinus Torvalds { 3201da177e4SLinus Torvalds return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL); 3211da177e4SLinus Torvalds } 322f6138882SAndrew Morton EXPORT_SYMBOL(vmalloc); 323f6138882SAndrew Morton 324e1ca7788SDave Young /* 325e1ca7788SDave Young * vzalloc - allocate virtually continguos memory with zero fill 326e1ca7788SDave Young * 327e1ca7788SDave Young * @size: allocation size 328e1ca7788SDave Young * 329e1ca7788SDave Young * Allocate enough pages to cover @size from the page level 330e1ca7788SDave Young * allocator and map them into continguos kernel virtual space. 331e1ca7788SDave Young * The memory allocated is set to zero. 332e1ca7788SDave Young * 333e1ca7788SDave Young * For tight control over page level allocator and protection flags 334e1ca7788SDave Young * use __vmalloc() instead. 335e1ca7788SDave Young */ 336e1ca7788SDave Young void *vzalloc(unsigned long size) 337e1ca7788SDave Young { 338e1ca7788SDave Young return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO, 339e1ca7788SDave Young PAGE_KERNEL); 340e1ca7788SDave Young } 341e1ca7788SDave Young EXPORT_SYMBOL(vzalloc); 342e1ca7788SDave Young 343e1ca7788SDave Young /** 344e1ca7788SDave Young * vmalloc_node - allocate memory on a specific node 345e1ca7788SDave Young * @size: allocation size 346e1ca7788SDave Young * @node: numa node 347e1ca7788SDave Young * 348e1ca7788SDave Young * Allocate enough pages to cover @size from the page level 349e1ca7788SDave Young * allocator and map them into contiguous kernel virtual space. 350e1ca7788SDave Young * 351e1ca7788SDave Young * For tight control over page level allocator and protection flags 352e1ca7788SDave Young * use __vmalloc() instead. 353e1ca7788SDave Young */ 354f6138882SAndrew Morton void *vmalloc_node(unsigned long size, int node) 355f6138882SAndrew Morton { 356f6138882SAndrew Morton return vmalloc(size); 357f6138882SAndrew Morton } 3589a14f653SPaul Mundt EXPORT_SYMBOL(vmalloc_node); 359e1ca7788SDave Young 360e1ca7788SDave Young /** 361e1ca7788SDave Young * vzalloc_node - allocate memory on a specific node with zero fill 362e1ca7788SDave Young * @size: allocation size 363e1ca7788SDave Young * @node: numa node 364e1ca7788SDave Young * 365e1ca7788SDave Young * Allocate enough pages to cover @size from the page level 366e1ca7788SDave Young * allocator and map them into contiguous kernel virtual space. 367e1ca7788SDave Young * The memory allocated is set to zero. 368e1ca7788SDave Young * 369e1ca7788SDave Young * For tight control over page level allocator and protection flags 370e1ca7788SDave Young * use __vmalloc() instead. 371e1ca7788SDave Young */ 372e1ca7788SDave Young void *vzalloc_node(unsigned long size, int node) 373e1ca7788SDave Young { 374e1ca7788SDave Young return vzalloc(size); 375e1ca7788SDave Young } 376e1ca7788SDave Young EXPORT_SYMBOL(vzalloc_node); 3771da177e4SLinus Torvalds 3781af446edSPaul Mundt #ifndef PAGE_KERNEL_EXEC 3791af446edSPaul Mundt # define PAGE_KERNEL_EXEC PAGE_KERNEL 3801af446edSPaul Mundt #endif 3811af446edSPaul Mundt 3821af446edSPaul Mundt /** 3831af446edSPaul Mundt * vmalloc_exec - allocate virtually contiguous, executable memory 3841af446edSPaul Mundt * @size: allocation size 3851af446edSPaul Mundt * 3861af446edSPaul Mundt * Kernel-internal function to allocate enough pages to cover @size 3871af446edSPaul Mundt * the page level allocator and map them into contiguous and 3881af446edSPaul Mundt * executable kernel virtual space. 3891af446edSPaul Mundt * 3901af446edSPaul Mundt * For tight control over page level allocator and protection flags 3911af446edSPaul Mundt * use __vmalloc() instead. 3921af446edSPaul Mundt */ 3931af446edSPaul Mundt 3941af446edSPaul Mundt void *vmalloc_exec(unsigned long size) 3951af446edSPaul Mundt { 3961af446edSPaul Mundt return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC); 3971af446edSPaul Mundt } 3981af446edSPaul Mundt 399b5073173SPaul Mundt /** 400b5073173SPaul Mundt * vmalloc_32 - allocate virtually contiguous memory (32bit addressable) 4011da177e4SLinus Torvalds * @size: allocation size 4021da177e4SLinus Torvalds * 4031da177e4SLinus Torvalds * Allocate enough 32bit PA addressable pages to cover @size from the 4041da177e4SLinus Torvalds * page level allocator and map them into continguos kernel virtual space. 4051da177e4SLinus Torvalds */ 4061da177e4SLinus Torvalds void *vmalloc_32(unsigned long size) 4071da177e4SLinus Torvalds { 4081da177e4SLinus Torvalds return __vmalloc(size, GFP_KERNEL, PAGE_KERNEL); 4091da177e4SLinus Torvalds } 410b5073173SPaul Mundt EXPORT_SYMBOL(vmalloc_32); 411b5073173SPaul Mundt 412b5073173SPaul Mundt /** 413b5073173SPaul Mundt * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory 414b5073173SPaul Mundt * @size: allocation size 415b5073173SPaul Mundt * 416b5073173SPaul Mundt * The resulting memory area is 32bit addressable and zeroed so it can be 417b5073173SPaul Mundt * mapped to userspace without leaking data. 418f905bc44SPaul Mundt * 419f905bc44SPaul Mundt * VM_USERMAP is set on the corresponding VMA so that subsequent calls to 420f905bc44SPaul Mundt * remap_vmalloc_range() are permissible. 421b5073173SPaul Mundt */ 422b5073173SPaul Mundt void *vmalloc_32_user(unsigned long size) 423b5073173SPaul Mundt { 424f905bc44SPaul Mundt /* 425f905bc44SPaul Mundt * We'll have to sort out the ZONE_DMA bits for 64-bit, 426f905bc44SPaul Mundt * but for now this can simply use vmalloc_user() directly. 427f905bc44SPaul Mundt */ 428f905bc44SPaul Mundt return vmalloc_user(size); 429b5073173SPaul Mundt } 430b5073173SPaul Mundt EXPORT_SYMBOL(vmalloc_32_user); 4311da177e4SLinus Torvalds 4321da177e4SLinus Torvalds void *vmap(struct page **pages, unsigned int count, unsigned long flags, pgprot_t prot) 4331da177e4SLinus Torvalds { 4341da177e4SLinus Torvalds BUG(); 4351da177e4SLinus Torvalds return NULL; 4361da177e4SLinus Torvalds } 437b5073173SPaul Mundt EXPORT_SYMBOL(vmap); 4381da177e4SLinus Torvalds 439b3bdda02SChristoph Lameter void vunmap(const void *addr) 4401da177e4SLinus Torvalds { 4411da177e4SLinus Torvalds BUG(); 4421da177e4SLinus Torvalds } 443b5073173SPaul Mundt EXPORT_SYMBOL(vunmap); 4441da177e4SLinus Torvalds 445eb6434d9SPaul Mundt void *vm_map_ram(struct page **pages, unsigned int count, int node, pgprot_t prot) 446eb6434d9SPaul Mundt { 447eb6434d9SPaul Mundt BUG(); 448eb6434d9SPaul Mundt return NULL; 449eb6434d9SPaul Mundt } 450eb6434d9SPaul Mundt EXPORT_SYMBOL(vm_map_ram); 451eb6434d9SPaul Mundt 452eb6434d9SPaul Mundt void vm_unmap_ram(const void *mem, unsigned int count) 453eb6434d9SPaul Mundt { 454eb6434d9SPaul Mundt BUG(); 455eb6434d9SPaul Mundt } 456eb6434d9SPaul Mundt EXPORT_SYMBOL(vm_unmap_ram); 457eb6434d9SPaul Mundt 458eb6434d9SPaul Mundt void vm_unmap_aliases(void) 459eb6434d9SPaul Mundt { 460eb6434d9SPaul Mundt } 461eb6434d9SPaul Mundt EXPORT_SYMBOL_GPL(vm_unmap_aliases); 462eb6434d9SPaul Mundt 4631da177e4SLinus Torvalds /* 4641eeb66a1SChristoph Hellwig * Implement a stub for vmalloc_sync_all() if the architecture chose not to 4651eeb66a1SChristoph Hellwig * have one. 4661eeb66a1SChristoph Hellwig */ 4673b32123dSGideon Israel Dsouza void __weak vmalloc_sync_all(void) 4681eeb66a1SChristoph Hellwig { 4691eeb66a1SChristoph Hellwig } 4701eeb66a1SChristoph Hellwig 47129c185e5SPaul Mundt /** 47229c185e5SPaul Mundt * alloc_vm_area - allocate a range of kernel address space 47329c185e5SPaul Mundt * @size: size of the area 47429c185e5SPaul Mundt * 47529c185e5SPaul Mundt * Returns: NULL on failure, vm_struct on success 47629c185e5SPaul Mundt * 47729c185e5SPaul Mundt * This function reserves a range of kernel address space, and 47829c185e5SPaul Mundt * allocates pagetables to map that range. No actual mappings 47929c185e5SPaul Mundt * are created. If the kernel address space is not shared 48029c185e5SPaul Mundt * between processes, it syncs the pagetable across all 48129c185e5SPaul Mundt * processes. 48229c185e5SPaul Mundt */ 483cd12909cSDavid Vrabel struct vm_struct *alloc_vm_area(size_t size, pte_t **ptes) 48429c185e5SPaul Mundt { 48529c185e5SPaul Mundt BUG(); 48629c185e5SPaul Mundt return NULL; 48729c185e5SPaul Mundt } 48829c185e5SPaul Mundt EXPORT_SYMBOL_GPL(alloc_vm_area); 48929c185e5SPaul Mundt 49029c185e5SPaul Mundt void free_vm_area(struct vm_struct *area) 49129c185e5SPaul Mundt { 49229c185e5SPaul Mundt BUG(); 49329c185e5SPaul Mundt } 49429c185e5SPaul Mundt EXPORT_SYMBOL_GPL(free_vm_area); 49529c185e5SPaul Mundt 496b5073173SPaul Mundt int vm_insert_page(struct vm_area_struct *vma, unsigned long addr, 497b5073173SPaul Mundt struct page *page) 498b5073173SPaul Mundt { 499b5073173SPaul Mundt return -EINVAL; 500b5073173SPaul Mundt } 501b5073173SPaul Mundt EXPORT_SYMBOL(vm_insert_page); 502b5073173SPaul Mundt 5031eeb66a1SChristoph Hellwig /* 5041da177e4SLinus Torvalds * sys_brk() for the most part doesn't need the global kernel 5051da177e4SLinus Torvalds * lock, except when an application is doing something nasty 5061da177e4SLinus Torvalds * like trying to un-brk an area that has already been mapped 5071da177e4SLinus Torvalds * to a regular file. in this case, the unmapping will need 5081da177e4SLinus Torvalds * to invoke file system routines that need the global lock. 5091da177e4SLinus Torvalds */ 5106a6160a7SHeiko Carstens SYSCALL_DEFINE1(brk, unsigned long, brk) 5111da177e4SLinus Torvalds { 5121da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 5131da177e4SLinus Torvalds 5141da177e4SLinus Torvalds if (brk < mm->start_brk || brk > mm->context.end_brk) 5151da177e4SLinus Torvalds return mm->brk; 5161da177e4SLinus Torvalds 5171da177e4SLinus Torvalds if (mm->brk == brk) 5181da177e4SLinus Torvalds return mm->brk; 5191da177e4SLinus Torvalds 5201da177e4SLinus Torvalds /* 5211da177e4SLinus Torvalds * Always allow shrinking brk 5221da177e4SLinus Torvalds */ 5231da177e4SLinus Torvalds if (brk <= mm->brk) { 5241da177e4SLinus Torvalds mm->brk = brk; 5251da177e4SLinus Torvalds return brk; 5261da177e4SLinus Torvalds } 5271da177e4SLinus Torvalds 5281da177e4SLinus Torvalds /* 5291da177e4SLinus Torvalds * Ok, looks good - let it rip. 5301da177e4SLinus Torvalds */ 531cfe79c00SMike Frysinger flush_icache_range(mm->brk, brk); 5321da177e4SLinus Torvalds return mm->brk = brk; 5331da177e4SLinus Torvalds } 5341da177e4SLinus Torvalds 5358feae131SDavid Howells /* 5368feae131SDavid Howells * initialise the VMA and region record slabs 5378feae131SDavid Howells */ 5388feae131SDavid Howells void __init mmap_init(void) 5391da177e4SLinus Torvalds { 54000a62ce9SKOSAKI Motohiro int ret; 54100a62ce9SKOSAKI Motohiro 54200a62ce9SKOSAKI Motohiro ret = percpu_counter_init(&vm_committed_as, 0); 54300a62ce9SKOSAKI Motohiro VM_BUG_ON(ret); 54433e5d769SDavid Howells vm_region_jar = KMEM_CACHE(vm_region, SLAB_PANIC); 5458feae131SDavid Howells } 5461da177e4SLinus Torvalds 5478feae131SDavid Howells /* 5488feae131SDavid Howells * validate the region tree 5498feae131SDavid Howells * - the caller must hold the region lock 5508feae131SDavid Howells */ 5518feae131SDavid Howells #ifdef CONFIG_DEBUG_NOMMU_REGIONS 5528feae131SDavid Howells static noinline void validate_nommu_regions(void) 5538feae131SDavid Howells { 5548feae131SDavid Howells struct vm_region *region, *last; 5558feae131SDavid Howells struct rb_node *p, *lastp; 5561da177e4SLinus Torvalds 5578feae131SDavid Howells lastp = rb_first(&nommu_region_tree); 5588feae131SDavid Howells if (!lastp) 5598feae131SDavid Howells return; 5608feae131SDavid Howells 5618feae131SDavid Howells last = rb_entry(lastp, struct vm_region, vm_rb); 56233e5d769SDavid Howells BUG_ON(unlikely(last->vm_end <= last->vm_start)); 56333e5d769SDavid Howells BUG_ON(unlikely(last->vm_top < last->vm_end)); 5648feae131SDavid Howells 5658feae131SDavid Howells while ((p = rb_next(lastp))) { 5668feae131SDavid Howells region = rb_entry(p, struct vm_region, vm_rb); 5678feae131SDavid Howells last = rb_entry(lastp, struct vm_region, vm_rb); 5688feae131SDavid Howells 56933e5d769SDavid Howells BUG_ON(unlikely(region->vm_end <= region->vm_start)); 57033e5d769SDavid Howells BUG_ON(unlikely(region->vm_top < region->vm_end)); 57133e5d769SDavid Howells BUG_ON(unlikely(region->vm_start < last->vm_top)); 5728feae131SDavid Howells 5738feae131SDavid Howells lastp = p; 5741da177e4SLinus Torvalds } 5751da177e4SLinus Torvalds } 5768feae131SDavid Howells #else 57733e5d769SDavid Howells static void validate_nommu_regions(void) 57833e5d769SDavid Howells { 57933e5d769SDavid Howells } 5808feae131SDavid Howells #endif 5818feae131SDavid Howells 5828feae131SDavid Howells /* 5838feae131SDavid Howells * add a region into the global tree 5848feae131SDavid Howells */ 5858feae131SDavid Howells static void add_nommu_region(struct vm_region *region) 5868feae131SDavid Howells { 5878feae131SDavid Howells struct vm_region *pregion; 5888feae131SDavid Howells struct rb_node **p, *parent; 5898feae131SDavid Howells 5908feae131SDavid Howells validate_nommu_regions(); 5918feae131SDavid Howells 5928feae131SDavid Howells parent = NULL; 5938feae131SDavid Howells p = &nommu_region_tree.rb_node; 5948feae131SDavid Howells while (*p) { 5958feae131SDavid Howells parent = *p; 5968feae131SDavid Howells pregion = rb_entry(parent, struct vm_region, vm_rb); 5978feae131SDavid Howells if (region->vm_start < pregion->vm_start) 5988feae131SDavid Howells p = &(*p)->rb_left; 5998feae131SDavid Howells else if (region->vm_start > pregion->vm_start) 6008feae131SDavid Howells p = &(*p)->rb_right; 6018feae131SDavid Howells else if (pregion == region) 6028feae131SDavid Howells return; 6038feae131SDavid Howells else 6048feae131SDavid Howells BUG(); 6058feae131SDavid Howells } 6068feae131SDavid Howells 6078feae131SDavid Howells rb_link_node(®ion->vm_rb, parent, p); 6088feae131SDavid Howells rb_insert_color(®ion->vm_rb, &nommu_region_tree); 6098feae131SDavid Howells 6108feae131SDavid Howells validate_nommu_regions(); 6118feae131SDavid Howells } 6128feae131SDavid Howells 6138feae131SDavid Howells /* 6148feae131SDavid Howells * delete a region from the global tree 6158feae131SDavid Howells */ 6168feae131SDavid Howells static void delete_nommu_region(struct vm_region *region) 6178feae131SDavid Howells { 6188feae131SDavid Howells BUG_ON(!nommu_region_tree.rb_node); 6198feae131SDavid Howells 6208feae131SDavid Howells validate_nommu_regions(); 6218feae131SDavid Howells rb_erase(®ion->vm_rb, &nommu_region_tree); 6228feae131SDavid Howells validate_nommu_regions(); 6238feae131SDavid Howells } 6248feae131SDavid Howells 6258feae131SDavid Howells /* 6268feae131SDavid Howells * free a contiguous series of pages 6278feae131SDavid Howells */ 6288feae131SDavid Howells static void free_page_series(unsigned long from, unsigned long to) 6298feae131SDavid Howells { 6308feae131SDavid Howells for (; from < to; from += PAGE_SIZE) { 6318feae131SDavid Howells struct page *page = virt_to_page(from); 6328feae131SDavid Howells 6338feae131SDavid Howells kdebug("- free %lx", from); 63433e5d769SDavid Howells atomic_long_dec(&mmap_pages_allocated); 6358feae131SDavid Howells if (page_count(page) != 1) 63633e5d769SDavid Howells kdebug("free page %p: refcount not one: %d", 63733e5d769SDavid Howells page, page_count(page)); 6388feae131SDavid Howells put_page(page); 6398feae131SDavid Howells } 6408feae131SDavid Howells } 6418feae131SDavid Howells 6428feae131SDavid Howells /* 6438feae131SDavid Howells * release a reference to a region 64433e5d769SDavid Howells * - the caller must hold the region semaphore for writing, which this releases 645dd8632a1SPaul Mundt * - the region may not have been added to the tree yet, in which case vm_top 6468feae131SDavid Howells * will equal vm_start 6478feae131SDavid Howells */ 6488feae131SDavid Howells static void __put_nommu_region(struct vm_region *region) 6498feae131SDavid Howells __releases(nommu_region_sem) 6508feae131SDavid Howells { 6511e2ae599SDavid Howells kenter("%p{%d}", region, region->vm_usage); 6528feae131SDavid Howells 6538feae131SDavid Howells BUG_ON(!nommu_region_tree.rb_node); 6548feae131SDavid Howells 6551e2ae599SDavid Howells if (--region->vm_usage == 0) { 656dd8632a1SPaul Mundt if (region->vm_top > region->vm_start) 6578feae131SDavid Howells delete_nommu_region(region); 6588feae131SDavid Howells up_write(&nommu_region_sem); 6598feae131SDavid Howells 6608feae131SDavid Howells if (region->vm_file) 6618feae131SDavid Howells fput(region->vm_file); 6628feae131SDavid Howells 6638feae131SDavid Howells /* IO memory and memory shared directly out of the pagecache 6648feae131SDavid Howells * from ramfs/tmpfs mustn't be released here */ 6658feae131SDavid Howells if (region->vm_flags & VM_MAPPED_COPY) { 6668feae131SDavid Howells kdebug("free series"); 667dd8632a1SPaul Mundt free_page_series(region->vm_start, region->vm_top); 6688feae131SDavid Howells } 6698feae131SDavid Howells kmem_cache_free(vm_region_jar, region); 6708feae131SDavid Howells } else { 6718feae131SDavid Howells up_write(&nommu_region_sem); 6728feae131SDavid Howells } 6738feae131SDavid Howells } 6748feae131SDavid Howells 6758feae131SDavid Howells /* 6768feae131SDavid Howells * release a reference to a region 6778feae131SDavid Howells */ 6788feae131SDavid Howells static void put_nommu_region(struct vm_region *region) 6798feae131SDavid Howells { 6808feae131SDavid Howells down_write(&nommu_region_sem); 6818feae131SDavid Howells __put_nommu_region(region); 6828feae131SDavid Howells } 6831da177e4SLinus Torvalds 6843034097aSDavid Howells /* 685eb8cdec4SBernd Schmidt * update protection on a vma 686eb8cdec4SBernd Schmidt */ 687eb8cdec4SBernd Schmidt static void protect_vma(struct vm_area_struct *vma, unsigned long flags) 688eb8cdec4SBernd Schmidt { 689eb8cdec4SBernd Schmidt #ifdef CONFIG_MPU 690eb8cdec4SBernd Schmidt struct mm_struct *mm = vma->vm_mm; 691eb8cdec4SBernd Schmidt long start = vma->vm_start & PAGE_MASK; 692eb8cdec4SBernd Schmidt while (start < vma->vm_end) { 693eb8cdec4SBernd Schmidt protect_page(mm, start, flags); 694eb8cdec4SBernd Schmidt start += PAGE_SIZE; 695eb8cdec4SBernd Schmidt } 696eb8cdec4SBernd Schmidt update_protections(mm); 697eb8cdec4SBernd Schmidt #endif 698eb8cdec4SBernd Schmidt } 699eb8cdec4SBernd Schmidt 700eb8cdec4SBernd Schmidt /* 7013034097aSDavid Howells * add a VMA into a process's mm_struct in the appropriate place in the list 7028feae131SDavid Howells * and tree and add to the address space's page tree also if not an anonymous 7038feae131SDavid Howells * page 7043034097aSDavid Howells * - should be called with mm->mmap_sem held writelocked 7053034097aSDavid Howells */ 7068feae131SDavid Howells static void add_vma_to_mm(struct mm_struct *mm, struct vm_area_struct *vma) 7073034097aSDavid Howells { 7086038def0SNamhyung Kim struct vm_area_struct *pvma, *prev; 7098feae131SDavid Howells struct address_space *mapping; 7106038def0SNamhyung Kim struct rb_node **p, *parent, *rb_prev; 7113034097aSDavid Howells 7128feae131SDavid Howells kenter(",%p", vma); 7138feae131SDavid Howells 7148feae131SDavid Howells BUG_ON(!vma->vm_region); 7158feae131SDavid Howells 7168feae131SDavid Howells mm->map_count++; 7178feae131SDavid Howells vma->vm_mm = mm; 7188feae131SDavid Howells 719eb8cdec4SBernd Schmidt protect_vma(vma, vma->vm_flags); 720eb8cdec4SBernd Schmidt 7218feae131SDavid Howells /* add the VMA to the mapping */ 7228feae131SDavid Howells if (vma->vm_file) { 7238feae131SDavid Howells mapping = vma->vm_file->f_mapping; 7248feae131SDavid Howells 725918e556eSDavid Howells mutex_lock(&mapping->i_mmap_mutex); 7268feae131SDavid Howells flush_dcache_mmap_lock(mapping); 7276b2dbba8SMichel Lespinasse vma_interval_tree_insert(vma, &mapping->i_mmap); 7288feae131SDavid Howells flush_dcache_mmap_unlock(mapping); 729918e556eSDavid Howells mutex_unlock(&mapping->i_mmap_mutex); 7308feae131SDavid Howells } 7318feae131SDavid Howells 7328feae131SDavid Howells /* add the VMA to the tree */ 7336038def0SNamhyung Kim parent = rb_prev = NULL; 7348feae131SDavid Howells p = &mm->mm_rb.rb_node; 7358feae131SDavid Howells while (*p) { 7368feae131SDavid Howells parent = *p; 7378feae131SDavid Howells pvma = rb_entry(parent, struct vm_area_struct, vm_rb); 7388feae131SDavid Howells 7398feae131SDavid Howells /* sort by: start addr, end addr, VMA struct addr in that order 7408feae131SDavid Howells * (the latter is necessary as we may get identical VMAs) */ 7418feae131SDavid Howells if (vma->vm_start < pvma->vm_start) 7428feae131SDavid Howells p = &(*p)->rb_left; 7436038def0SNamhyung Kim else if (vma->vm_start > pvma->vm_start) { 7446038def0SNamhyung Kim rb_prev = parent; 7458feae131SDavid Howells p = &(*p)->rb_right; 7466038def0SNamhyung Kim } else if (vma->vm_end < pvma->vm_end) 7478feae131SDavid Howells p = &(*p)->rb_left; 7486038def0SNamhyung Kim else if (vma->vm_end > pvma->vm_end) { 7496038def0SNamhyung Kim rb_prev = parent; 7508feae131SDavid Howells p = &(*p)->rb_right; 7516038def0SNamhyung Kim } else if (vma < pvma) 7528feae131SDavid Howells p = &(*p)->rb_left; 7536038def0SNamhyung Kim else if (vma > pvma) { 7546038def0SNamhyung Kim rb_prev = parent; 7558feae131SDavid Howells p = &(*p)->rb_right; 7566038def0SNamhyung Kim } else 7578feae131SDavid Howells BUG(); 7588feae131SDavid Howells } 7598feae131SDavid Howells 7608feae131SDavid Howells rb_link_node(&vma->vm_rb, parent, p); 7618feae131SDavid Howells rb_insert_color(&vma->vm_rb, &mm->mm_rb); 7628feae131SDavid Howells 7638feae131SDavid Howells /* add VMA to the VMA list also */ 7646038def0SNamhyung Kim prev = NULL; 7656038def0SNamhyung Kim if (rb_prev) 7666038def0SNamhyung Kim prev = rb_entry(rb_prev, struct vm_area_struct, vm_rb); 7673034097aSDavid Howells 7686038def0SNamhyung Kim __vma_link_list(mm, vma, prev, parent); 7698feae131SDavid Howells } 7708feae131SDavid Howells 7718feae131SDavid Howells /* 7728feae131SDavid Howells * delete a VMA from its owning mm_struct and address space 7738feae131SDavid Howells */ 7748feae131SDavid Howells static void delete_vma_from_mm(struct vm_area_struct *vma) 7758feae131SDavid Howells { 776615d6e87SDavidlohr Bueso int i; 7778feae131SDavid Howells struct address_space *mapping; 7788feae131SDavid Howells struct mm_struct *mm = vma->vm_mm; 779615d6e87SDavidlohr Bueso struct task_struct *curr = current; 7808feae131SDavid Howells 7818feae131SDavid Howells kenter("%p", vma); 7828feae131SDavid Howells 783eb8cdec4SBernd Schmidt protect_vma(vma, 0); 784eb8cdec4SBernd Schmidt 7858feae131SDavid Howells mm->map_count--; 786615d6e87SDavidlohr Bueso for (i = 0; i < VMACACHE_SIZE; i++) { 787615d6e87SDavidlohr Bueso /* if the vma is cached, invalidate the entire cache */ 788615d6e87SDavidlohr Bueso if (curr->vmacache[i] == vma) { 789615d6e87SDavidlohr Bueso vmacache_invalidate(curr->mm); 790615d6e87SDavidlohr Bueso break; 791615d6e87SDavidlohr Bueso } 792615d6e87SDavidlohr Bueso } 7938feae131SDavid Howells 7948feae131SDavid Howells /* remove the VMA from the mapping */ 7958feae131SDavid Howells if (vma->vm_file) { 7968feae131SDavid Howells mapping = vma->vm_file->f_mapping; 7978feae131SDavid Howells 798918e556eSDavid Howells mutex_lock(&mapping->i_mmap_mutex); 7998feae131SDavid Howells flush_dcache_mmap_lock(mapping); 8006b2dbba8SMichel Lespinasse vma_interval_tree_remove(vma, &mapping->i_mmap); 8018feae131SDavid Howells flush_dcache_mmap_unlock(mapping); 802918e556eSDavid Howells mutex_unlock(&mapping->i_mmap_mutex); 8038feae131SDavid Howells } 8048feae131SDavid Howells 8058feae131SDavid Howells /* remove from the MM's tree and list */ 8068feae131SDavid Howells rb_erase(&vma->vm_rb, &mm->mm_rb); 807b951bf2cSNamhyung Kim 808b951bf2cSNamhyung Kim if (vma->vm_prev) 809b951bf2cSNamhyung Kim vma->vm_prev->vm_next = vma->vm_next; 810b951bf2cSNamhyung Kim else 811b951bf2cSNamhyung Kim mm->mmap = vma->vm_next; 812b951bf2cSNamhyung Kim 813b951bf2cSNamhyung Kim if (vma->vm_next) 814b951bf2cSNamhyung Kim vma->vm_next->vm_prev = vma->vm_prev; 8158feae131SDavid Howells } 8168feae131SDavid Howells 8178feae131SDavid Howells /* 8188feae131SDavid Howells * destroy a VMA record 8198feae131SDavid Howells */ 8208feae131SDavid Howells static void delete_vma(struct mm_struct *mm, struct vm_area_struct *vma) 8218feae131SDavid Howells { 8228feae131SDavid Howells kenter("%p", vma); 8238feae131SDavid Howells if (vma->vm_ops && vma->vm_ops->close) 8248feae131SDavid Howells vma->vm_ops->close(vma); 825e9714acfSKonstantin Khlebnikov if (vma->vm_file) 8268feae131SDavid Howells fput(vma->vm_file); 8278feae131SDavid Howells put_nommu_region(vma->vm_region); 8288feae131SDavid Howells kmem_cache_free(vm_area_cachep, vma); 8293034097aSDavid Howells } 8303034097aSDavid Howells 8313034097aSDavid Howells /* 8323034097aSDavid Howells * look up the first VMA in which addr resides, NULL if none 8333034097aSDavid Howells * - should be called with mm->mmap_sem at least held readlocked 8343034097aSDavid Howells */ 8353034097aSDavid Howells struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr) 8363034097aSDavid Howells { 8378feae131SDavid Howells struct vm_area_struct *vma; 8383034097aSDavid Howells 8398feae131SDavid Howells /* check the cache first */ 840615d6e87SDavidlohr Bueso vma = vmacache_find(mm, addr); 841615d6e87SDavidlohr Bueso if (likely(vma)) 8428feae131SDavid Howells return vma; 8438feae131SDavid Howells 844e922c4c5SNamhyung Kim /* trawl the list (there may be multiple mappings in which addr 8458feae131SDavid Howells * resides) */ 846e922c4c5SNamhyung Kim for (vma = mm->mmap; vma; vma = vma->vm_next) { 8478feae131SDavid Howells if (vma->vm_start > addr) 8488feae131SDavid Howells return NULL; 8498feae131SDavid Howells if (vma->vm_end > addr) { 850615d6e87SDavidlohr Bueso vmacache_update(addr, vma); 8518feae131SDavid Howells return vma; 8523034097aSDavid Howells } 8538feae131SDavid Howells } 8543034097aSDavid Howells 8553034097aSDavid Howells return NULL; 8563034097aSDavid Howells } 8573034097aSDavid Howells EXPORT_SYMBOL(find_vma); 8583034097aSDavid Howells 8593034097aSDavid Howells /* 860930e652aSDavid Howells * find a VMA 861930e652aSDavid Howells * - we don't extend stack VMAs under NOMMU conditions 862930e652aSDavid Howells */ 863930e652aSDavid Howells struct vm_area_struct *find_extend_vma(struct mm_struct *mm, unsigned long addr) 864930e652aSDavid Howells { 8657561e8caSDavid Howells return find_vma(mm, addr); 866930e652aSDavid Howells } 867930e652aSDavid Howells 8688feae131SDavid Howells /* 8698feae131SDavid Howells * expand a stack to a given address 8708feae131SDavid Howells * - not supported under NOMMU conditions 8718feae131SDavid Howells */ 87257c8f63eSGreg Ungerer int expand_stack(struct vm_area_struct *vma, unsigned long address) 87357c8f63eSGreg Ungerer { 87457c8f63eSGreg Ungerer return -ENOMEM; 87557c8f63eSGreg Ungerer } 87657c8f63eSGreg Ungerer 877930e652aSDavid Howells /* 8786fa5f80bSDavid Howells * look up the first VMA exactly that exactly matches addr 8796fa5f80bSDavid Howells * - should be called with mm->mmap_sem at least held readlocked 8806fa5f80bSDavid Howells */ 8818feae131SDavid Howells static struct vm_area_struct *find_vma_exact(struct mm_struct *mm, 8828feae131SDavid Howells unsigned long addr, 8838feae131SDavid Howells unsigned long len) 8841da177e4SLinus Torvalds { 8851da177e4SLinus Torvalds struct vm_area_struct *vma; 8868feae131SDavid Howells unsigned long end = addr + len; 8871da177e4SLinus Torvalds 8888feae131SDavid Howells /* check the cache first */ 889615d6e87SDavidlohr Bueso vma = vmacache_find_exact(mm, addr, end); 890615d6e87SDavidlohr Bueso if (vma) 8911da177e4SLinus Torvalds return vma; 8928feae131SDavid Howells 893e922c4c5SNamhyung Kim /* trawl the list (there may be multiple mappings in which addr 8948feae131SDavid Howells * resides) */ 895e922c4c5SNamhyung Kim for (vma = mm->mmap; vma; vma = vma->vm_next) { 8968feae131SDavid Howells if (vma->vm_start < addr) 8978feae131SDavid Howells continue; 8988feae131SDavid Howells if (vma->vm_start > addr) 8998feae131SDavid Howells return NULL; 9008feae131SDavid Howells if (vma->vm_end == end) { 901615d6e87SDavidlohr Bueso vmacache_update(addr, vma); 9028feae131SDavid Howells return vma; 9038feae131SDavid Howells } 9041da177e4SLinus Torvalds } 9051da177e4SLinus Torvalds 9061da177e4SLinus Torvalds return NULL; 9071da177e4SLinus Torvalds } 9081da177e4SLinus Torvalds 9093034097aSDavid Howells /* 9101da177e4SLinus Torvalds * determine whether a mapping should be permitted and, if so, what sort of 9111da177e4SLinus Torvalds * mapping we're capable of supporting 9121da177e4SLinus Torvalds */ 9131da177e4SLinus Torvalds static int validate_mmap_request(struct file *file, 9141da177e4SLinus Torvalds unsigned long addr, 9151da177e4SLinus Torvalds unsigned long len, 9161da177e4SLinus Torvalds unsigned long prot, 9171da177e4SLinus Torvalds unsigned long flags, 9181da177e4SLinus Torvalds unsigned long pgoff, 9191da177e4SLinus Torvalds unsigned long *_capabilities) 9201da177e4SLinus Torvalds { 9218feae131SDavid Howells unsigned long capabilities, rlen; 9221da177e4SLinus Torvalds int ret; 9231da177e4SLinus Torvalds 9241da177e4SLinus Torvalds /* do the simple checks first */ 92506aab5a3SDavid Howells if (flags & MAP_FIXED) { 9261da177e4SLinus Torvalds printk(KERN_DEBUG 9271da177e4SLinus Torvalds "%d: Can't do fixed-address/overlay mmap of RAM\n", 9281da177e4SLinus Torvalds current->pid); 9291da177e4SLinus Torvalds return -EINVAL; 9301da177e4SLinus Torvalds } 9311da177e4SLinus Torvalds 9321da177e4SLinus Torvalds if ((flags & MAP_TYPE) != MAP_PRIVATE && 9331da177e4SLinus Torvalds (flags & MAP_TYPE) != MAP_SHARED) 9341da177e4SLinus Torvalds return -EINVAL; 9351da177e4SLinus Torvalds 936f81cff0dSMike Frysinger if (!len) 9371da177e4SLinus Torvalds return -EINVAL; 9381da177e4SLinus Torvalds 939f81cff0dSMike Frysinger /* Careful about overflows.. */ 9408feae131SDavid Howells rlen = PAGE_ALIGN(len); 9418feae131SDavid Howells if (!rlen || rlen > TASK_SIZE) 942f81cff0dSMike Frysinger return -ENOMEM; 943f81cff0dSMike Frysinger 9441da177e4SLinus Torvalds /* offset overflow? */ 9458feae131SDavid Howells if ((pgoff + (rlen >> PAGE_SHIFT)) < pgoff) 946f81cff0dSMike Frysinger return -EOVERFLOW; 9471da177e4SLinus Torvalds 9481da177e4SLinus Torvalds if (file) { 9491da177e4SLinus Torvalds /* validate file mapping requests */ 9501da177e4SLinus Torvalds struct address_space *mapping; 9511da177e4SLinus Torvalds 9521da177e4SLinus Torvalds /* files must support mmap */ 95372c2d531SAl Viro if (!file->f_op->mmap) 9541da177e4SLinus Torvalds return -ENODEV; 9551da177e4SLinus Torvalds 9561da177e4SLinus Torvalds /* work out if what we've got could possibly be shared 9571da177e4SLinus Torvalds * - we support chardevs that provide their own "memory" 9581da177e4SLinus Torvalds * - we support files/blockdevs that are memory backed 9591da177e4SLinus Torvalds */ 9601da177e4SLinus Torvalds mapping = file->f_mapping; 9611da177e4SLinus Torvalds if (!mapping) 962496ad9aaSAl Viro mapping = file_inode(file)->i_mapping; 9631da177e4SLinus Torvalds 9641da177e4SLinus Torvalds capabilities = 0; 9651da177e4SLinus Torvalds if (mapping && mapping->backing_dev_info) 9661da177e4SLinus Torvalds capabilities = mapping->backing_dev_info->capabilities; 9671da177e4SLinus Torvalds 9681da177e4SLinus Torvalds if (!capabilities) { 9691da177e4SLinus Torvalds /* no explicit capabilities set, so assume some 9701da177e4SLinus Torvalds * defaults */ 971496ad9aaSAl Viro switch (file_inode(file)->i_mode & S_IFMT) { 9721da177e4SLinus Torvalds case S_IFREG: 9731da177e4SLinus Torvalds case S_IFBLK: 9741da177e4SLinus Torvalds capabilities = BDI_CAP_MAP_COPY; 9751da177e4SLinus Torvalds break; 9761da177e4SLinus Torvalds 9771da177e4SLinus Torvalds case S_IFCHR: 9781da177e4SLinus Torvalds capabilities = 9791da177e4SLinus Torvalds BDI_CAP_MAP_DIRECT | 9801da177e4SLinus Torvalds BDI_CAP_READ_MAP | 9811da177e4SLinus Torvalds BDI_CAP_WRITE_MAP; 9821da177e4SLinus Torvalds break; 9831da177e4SLinus Torvalds 9841da177e4SLinus Torvalds default: 9851da177e4SLinus Torvalds return -EINVAL; 9861da177e4SLinus Torvalds } 9871da177e4SLinus Torvalds } 9881da177e4SLinus Torvalds 9891da177e4SLinus Torvalds /* eliminate any capabilities that we can't support on this 9901da177e4SLinus Torvalds * device */ 9911da177e4SLinus Torvalds if (!file->f_op->get_unmapped_area) 9921da177e4SLinus Torvalds capabilities &= ~BDI_CAP_MAP_DIRECT; 9931da177e4SLinus Torvalds if (!file->f_op->read) 9941da177e4SLinus Torvalds capabilities &= ~BDI_CAP_MAP_COPY; 9951da177e4SLinus Torvalds 99628d7a6aeSGraff Yang /* The file shall have been opened with read permission. */ 99728d7a6aeSGraff Yang if (!(file->f_mode & FMODE_READ)) 99828d7a6aeSGraff Yang return -EACCES; 99928d7a6aeSGraff Yang 10001da177e4SLinus Torvalds if (flags & MAP_SHARED) { 10011da177e4SLinus Torvalds /* do checks for writing, appending and locking */ 10021da177e4SLinus Torvalds if ((prot & PROT_WRITE) && 10031da177e4SLinus Torvalds !(file->f_mode & FMODE_WRITE)) 10041da177e4SLinus Torvalds return -EACCES; 10051da177e4SLinus Torvalds 1006496ad9aaSAl Viro if (IS_APPEND(file_inode(file)) && 10071da177e4SLinus Torvalds (file->f_mode & FMODE_WRITE)) 10081da177e4SLinus Torvalds return -EACCES; 10091da177e4SLinus Torvalds 1010d7a06983SJeff Layton if (locks_verify_locked(file)) 10111da177e4SLinus Torvalds return -EAGAIN; 10121da177e4SLinus Torvalds 10131da177e4SLinus Torvalds if (!(capabilities & BDI_CAP_MAP_DIRECT)) 10141da177e4SLinus Torvalds return -ENODEV; 10151da177e4SLinus Torvalds 10161da177e4SLinus Torvalds /* we mustn't privatise shared mappings */ 10171da177e4SLinus Torvalds capabilities &= ~BDI_CAP_MAP_COPY; 1018ac714904SChoi Gi-yong } else { 10191da177e4SLinus Torvalds /* we're going to read the file into private memory we 10201da177e4SLinus Torvalds * allocate */ 10211da177e4SLinus Torvalds if (!(capabilities & BDI_CAP_MAP_COPY)) 10221da177e4SLinus Torvalds return -ENODEV; 10231da177e4SLinus Torvalds 10241da177e4SLinus Torvalds /* we don't permit a private writable mapping to be 10251da177e4SLinus Torvalds * shared with the backing device */ 10261da177e4SLinus Torvalds if (prot & PROT_WRITE) 10271da177e4SLinus Torvalds capabilities &= ~BDI_CAP_MAP_DIRECT; 10281da177e4SLinus Torvalds } 10291da177e4SLinus Torvalds 10303c7b2045SBernd Schmidt if (capabilities & BDI_CAP_MAP_DIRECT) { 10313c7b2045SBernd Schmidt if (((prot & PROT_READ) && !(capabilities & BDI_CAP_READ_MAP)) || 10323c7b2045SBernd Schmidt ((prot & PROT_WRITE) && !(capabilities & BDI_CAP_WRITE_MAP)) || 10333c7b2045SBernd Schmidt ((prot & PROT_EXEC) && !(capabilities & BDI_CAP_EXEC_MAP)) 10343c7b2045SBernd Schmidt ) { 10353c7b2045SBernd Schmidt capabilities &= ~BDI_CAP_MAP_DIRECT; 10363c7b2045SBernd Schmidt if (flags & MAP_SHARED) { 10373c7b2045SBernd Schmidt printk(KERN_WARNING 10383c7b2045SBernd Schmidt "MAP_SHARED not completely supported on !MMU\n"); 10393c7b2045SBernd Schmidt return -EINVAL; 10403c7b2045SBernd Schmidt } 10413c7b2045SBernd Schmidt } 10423c7b2045SBernd Schmidt } 10433c7b2045SBernd Schmidt 10441da177e4SLinus Torvalds /* handle executable mappings and implied executable 10451da177e4SLinus Torvalds * mappings */ 1046e9536ae7SJosef Sipek if (file->f_path.mnt->mnt_flags & MNT_NOEXEC) { 10471da177e4SLinus Torvalds if (prot & PROT_EXEC) 10481da177e4SLinus Torvalds return -EPERM; 1049ac714904SChoi Gi-yong } else if ((prot & PROT_READ) && !(prot & PROT_EXEC)) { 10501da177e4SLinus Torvalds /* handle implication of PROT_EXEC by PROT_READ */ 10511da177e4SLinus Torvalds if (current->personality & READ_IMPLIES_EXEC) { 10521da177e4SLinus Torvalds if (capabilities & BDI_CAP_EXEC_MAP) 10531da177e4SLinus Torvalds prot |= PROT_EXEC; 10541da177e4SLinus Torvalds } 1055ac714904SChoi Gi-yong } else if ((prot & PROT_READ) && 10561da177e4SLinus Torvalds (prot & PROT_EXEC) && 10571da177e4SLinus Torvalds !(capabilities & BDI_CAP_EXEC_MAP) 10581da177e4SLinus Torvalds ) { 10591da177e4SLinus Torvalds /* backing file is not executable, try to copy */ 10601da177e4SLinus Torvalds capabilities &= ~BDI_CAP_MAP_DIRECT; 10611da177e4SLinus Torvalds } 1062ac714904SChoi Gi-yong } else { 10631da177e4SLinus Torvalds /* anonymous mappings are always memory backed and can be 10641da177e4SLinus Torvalds * privately mapped 10651da177e4SLinus Torvalds */ 10661da177e4SLinus Torvalds capabilities = BDI_CAP_MAP_COPY; 10671da177e4SLinus Torvalds 10681da177e4SLinus Torvalds /* handle PROT_EXEC implication by PROT_READ */ 10691da177e4SLinus Torvalds if ((prot & PROT_READ) && 10701da177e4SLinus Torvalds (current->personality & READ_IMPLIES_EXEC)) 10711da177e4SLinus Torvalds prot |= PROT_EXEC; 10721da177e4SLinus Torvalds } 10731da177e4SLinus Torvalds 10741da177e4SLinus Torvalds /* allow the security API to have its say */ 1075e5467859SAl Viro ret = security_mmap_addr(addr); 1076e5467859SAl Viro if (ret < 0) 1077e5467859SAl Viro return ret; 10781da177e4SLinus Torvalds 10791da177e4SLinus Torvalds /* looks okay */ 10801da177e4SLinus Torvalds *_capabilities = capabilities; 10811da177e4SLinus Torvalds return 0; 10821da177e4SLinus Torvalds } 10831da177e4SLinus Torvalds 10841da177e4SLinus Torvalds /* 10851da177e4SLinus Torvalds * we've determined that we can make the mapping, now translate what we 10861da177e4SLinus Torvalds * now know into VMA flags 10871da177e4SLinus Torvalds */ 10881da177e4SLinus Torvalds static unsigned long determine_vm_flags(struct file *file, 10891da177e4SLinus Torvalds unsigned long prot, 10901da177e4SLinus Torvalds unsigned long flags, 10911da177e4SLinus Torvalds unsigned long capabilities) 10921da177e4SLinus Torvalds { 10931da177e4SLinus Torvalds unsigned long vm_flags; 10941da177e4SLinus Torvalds 10951da177e4SLinus Torvalds vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags); 10961da177e4SLinus Torvalds /* vm_flags |= mm->def_flags; */ 10971da177e4SLinus Torvalds 10981da177e4SLinus Torvalds if (!(capabilities & BDI_CAP_MAP_DIRECT)) { 10991da177e4SLinus Torvalds /* attempt to share read-only copies of mapped file chunks */ 11003c7b2045SBernd Schmidt vm_flags |= VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC; 11011da177e4SLinus Torvalds if (file && !(prot & PROT_WRITE)) 11021da177e4SLinus Torvalds vm_flags |= VM_MAYSHARE; 11033c7b2045SBernd Schmidt } else { 11041da177e4SLinus Torvalds /* overlay a shareable mapping on the backing device or inode 11051da177e4SLinus Torvalds * if possible - used for chardevs, ramfs/tmpfs/shmfs and 11061da177e4SLinus Torvalds * romfs/cramfs */ 11073c7b2045SBernd Schmidt vm_flags |= VM_MAYSHARE | (capabilities & BDI_CAP_VMFLAGS); 11081da177e4SLinus Torvalds if (flags & MAP_SHARED) 11093c7b2045SBernd Schmidt vm_flags |= VM_SHARED; 11101da177e4SLinus Torvalds } 11111da177e4SLinus Torvalds 11121da177e4SLinus Torvalds /* refuse to let anyone share private mappings with this process if 11131da177e4SLinus Torvalds * it's being traced - otherwise breakpoints set in it may interfere 11141da177e4SLinus Torvalds * with another untraced process 11151da177e4SLinus Torvalds */ 1116a288eeccSTejun Heo if ((flags & MAP_PRIVATE) && current->ptrace) 11171da177e4SLinus Torvalds vm_flags &= ~VM_MAYSHARE; 11181da177e4SLinus Torvalds 11191da177e4SLinus Torvalds return vm_flags; 11201da177e4SLinus Torvalds } 11211da177e4SLinus Torvalds 11221da177e4SLinus Torvalds /* 11238feae131SDavid Howells * set up a shared mapping on a file (the driver or filesystem provides and 11248feae131SDavid Howells * pins the storage) 11251da177e4SLinus Torvalds */ 11268feae131SDavid Howells static int do_mmap_shared_file(struct vm_area_struct *vma) 11271da177e4SLinus Torvalds { 11281da177e4SLinus Torvalds int ret; 11291da177e4SLinus Torvalds 11301da177e4SLinus Torvalds ret = vma->vm_file->f_op->mmap(vma->vm_file, vma); 1131dd8632a1SPaul Mundt if (ret == 0) { 1132dd8632a1SPaul Mundt vma->vm_region->vm_top = vma->vm_region->vm_end; 1133645d83c5SDavid Howells return 0; 1134dd8632a1SPaul Mundt } 11351da177e4SLinus Torvalds if (ret != -ENOSYS) 11361da177e4SLinus Torvalds return ret; 11371da177e4SLinus Torvalds 11383fa30460SDavid Howells /* getting -ENOSYS indicates that direct mmap isn't possible (as 11393fa30460SDavid Howells * opposed to tried but failed) so we can only give a suitable error as 11403fa30460SDavid Howells * it's not possible to make a private copy if MAP_SHARED was given */ 11411da177e4SLinus Torvalds return -ENODEV; 11421da177e4SLinus Torvalds } 11431da177e4SLinus Torvalds 11441da177e4SLinus Torvalds /* 11451da177e4SLinus Torvalds * set up a private mapping or an anonymous shared mapping 11461da177e4SLinus Torvalds */ 11478feae131SDavid Howells static int do_mmap_private(struct vm_area_struct *vma, 11488feae131SDavid Howells struct vm_region *region, 1149645d83c5SDavid Howells unsigned long len, 1150645d83c5SDavid Howells unsigned long capabilities) 11511da177e4SLinus Torvalds { 11528feae131SDavid Howells struct page *pages; 1153f67d9b15SBob Liu unsigned long total, point, n; 11541da177e4SLinus Torvalds void *base; 11558feae131SDavid Howells int ret, order; 11561da177e4SLinus Torvalds 11571da177e4SLinus Torvalds /* invoke the file's mapping function so that it can keep track of 11581da177e4SLinus Torvalds * shared mappings on devices or memory 11591da177e4SLinus Torvalds * - VM_MAYSHARE will be set if it may attempt to share 11601da177e4SLinus Torvalds */ 1161645d83c5SDavid Howells if (capabilities & BDI_CAP_MAP_DIRECT) { 11621da177e4SLinus Torvalds ret = vma->vm_file->f_op->mmap(vma->vm_file, vma); 1163dd8632a1SPaul Mundt if (ret == 0) { 11641da177e4SLinus Torvalds /* shouldn't return success if we're not sharing */ 1165dd8632a1SPaul Mundt BUG_ON(!(vma->vm_flags & VM_MAYSHARE)); 1166dd8632a1SPaul Mundt vma->vm_region->vm_top = vma->vm_region->vm_end; 1167645d83c5SDavid Howells return 0; 11681da177e4SLinus Torvalds } 1169dd8632a1SPaul Mundt if (ret != -ENOSYS) 1170dd8632a1SPaul Mundt return ret; 11711da177e4SLinus Torvalds 11721da177e4SLinus Torvalds /* getting an ENOSYS error indicates that direct mmap isn't 11731da177e4SLinus Torvalds * possible (as opposed to tried but failed) so we'll try to 11741da177e4SLinus Torvalds * make a private copy of the data and map that instead */ 11751da177e4SLinus Torvalds } 11761da177e4SLinus Torvalds 11778feae131SDavid Howells 11781da177e4SLinus Torvalds /* allocate some memory to hold the mapping 11791da177e4SLinus Torvalds * - note that this may not return a page-aligned address if the object 11801da177e4SLinus Torvalds * we're allocating is smaller than a page 11811da177e4SLinus Torvalds */ 1182f67d9b15SBob Liu order = get_order(len); 11838feae131SDavid Howells kdebug("alloc order %d for %lx", order, len); 11848feae131SDavid Howells 11858feae131SDavid Howells pages = alloc_pages(GFP_KERNEL, order); 11868feae131SDavid Howells if (!pages) 11871da177e4SLinus Torvalds goto enomem; 11881da177e4SLinus Torvalds 11898feae131SDavid Howells total = 1 << order; 119033e5d769SDavid Howells atomic_long_add(total, &mmap_pages_allocated); 11911da177e4SLinus Torvalds 1192f67d9b15SBob Liu point = len >> PAGE_SHIFT; 1193dd8632a1SPaul Mundt 1194dd8632a1SPaul Mundt /* we allocated a power-of-2 sized page set, so we may want to trim off 1195dd8632a1SPaul Mundt * the excess */ 1196dd8632a1SPaul Mundt if (sysctl_nr_trim_pages && total - point >= sysctl_nr_trim_pages) { 11978feae131SDavid Howells while (total > point) { 11988feae131SDavid Howells order = ilog2(total - point); 11998feae131SDavid Howells n = 1 << order; 12008feae131SDavid Howells kdebug("shave %lu/%lu @%lu", n, total - point, total); 120133e5d769SDavid Howells atomic_long_sub(n, &mmap_pages_allocated); 12028feae131SDavid Howells total -= n; 12038feae131SDavid Howells set_page_refcounted(pages + total); 12048feae131SDavid Howells __free_pages(pages + total, order); 12058feae131SDavid Howells } 1206dd8632a1SPaul Mundt } 12078feae131SDavid Howells 12088feae131SDavid Howells for (point = 1; point < total; point++) 12098feae131SDavid Howells set_page_refcounted(&pages[point]); 12108feae131SDavid Howells 12118feae131SDavid Howells base = page_address(pages); 12128feae131SDavid Howells region->vm_flags = vma->vm_flags |= VM_MAPPED_COPY; 12138feae131SDavid Howells region->vm_start = (unsigned long) base; 1214f67d9b15SBob Liu region->vm_end = region->vm_start + len; 1215dd8632a1SPaul Mundt region->vm_top = region->vm_start + (total << PAGE_SHIFT); 12168feae131SDavid Howells 12178feae131SDavid Howells vma->vm_start = region->vm_start; 12188feae131SDavid Howells vma->vm_end = region->vm_start + len; 12191da177e4SLinus Torvalds 12201da177e4SLinus Torvalds if (vma->vm_file) { 12211da177e4SLinus Torvalds /* read the contents of a file into the copy */ 12221da177e4SLinus Torvalds mm_segment_t old_fs; 12231da177e4SLinus Torvalds loff_t fpos; 12241da177e4SLinus Torvalds 12251da177e4SLinus Torvalds fpos = vma->vm_pgoff; 12261da177e4SLinus Torvalds fpos <<= PAGE_SHIFT; 12271da177e4SLinus Torvalds 12281da177e4SLinus Torvalds old_fs = get_fs(); 12291da177e4SLinus Torvalds set_fs(KERNEL_DS); 1230f67d9b15SBob Liu ret = vma->vm_file->f_op->read(vma->vm_file, base, len, &fpos); 12311da177e4SLinus Torvalds set_fs(old_fs); 12321da177e4SLinus Torvalds 12331da177e4SLinus Torvalds if (ret < 0) 12341da177e4SLinus Torvalds goto error_free; 12351da177e4SLinus Torvalds 12361da177e4SLinus Torvalds /* clear the last little bit */ 1237f67d9b15SBob Liu if (ret < len) 1238f67d9b15SBob Liu memset(base + ret, 0, len - ret); 12391da177e4SLinus Torvalds 12401da177e4SLinus Torvalds } 12411da177e4SLinus Torvalds 12421da177e4SLinus Torvalds return 0; 12431da177e4SLinus Torvalds 12441da177e4SLinus Torvalds error_free: 12457223bb4aSNamhyung Kim free_page_series(region->vm_start, region->vm_top); 12468feae131SDavid Howells region->vm_start = vma->vm_start = 0; 12478feae131SDavid Howells region->vm_end = vma->vm_end = 0; 1248dd8632a1SPaul Mundt region->vm_top = 0; 12491da177e4SLinus Torvalds return ret; 12501da177e4SLinus Torvalds 12511da177e4SLinus Torvalds enomem: 1252*b1de0d13SMitchel Humpherys pr_err("Allocation of length %lu from process %d (%s) failed\n", 125305ae6fa3SGreg Ungerer len, current->pid, current->comm); 12547bf02ea2SDavid Rientjes show_free_areas(0); 12551da177e4SLinus Torvalds return -ENOMEM; 12561da177e4SLinus Torvalds } 12571da177e4SLinus Torvalds 12581da177e4SLinus Torvalds /* 12591da177e4SLinus Torvalds * handle mapping creation for uClinux 12601da177e4SLinus Torvalds */ 1261e3fc629dSAl Viro unsigned long do_mmap_pgoff(struct file *file, 12621da177e4SLinus Torvalds unsigned long addr, 12631da177e4SLinus Torvalds unsigned long len, 12641da177e4SLinus Torvalds unsigned long prot, 12651da177e4SLinus Torvalds unsigned long flags, 1266bebeb3d6SMichel Lespinasse unsigned long pgoff, 126741badc15SMichel Lespinasse unsigned long *populate) 12681da177e4SLinus Torvalds { 12698feae131SDavid Howells struct vm_area_struct *vma; 12708feae131SDavid Howells struct vm_region *region; 12711da177e4SLinus Torvalds struct rb_node *rb; 12728feae131SDavid Howells unsigned long capabilities, vm_flags, result; 12731da177e4SLinus Torvalds int ret; 12741da177e4SLinus Torvalds 12758feae131SDavid Howells kenter(",%lx,%lx,%lx,%lx,%lx", addr, len, prot, flags, pgoff); 12768feae131SDavid Howells 127741badc15SMichel Lespinasse *populate = 0; 1278bebeb3d6SMichel Lespinasse 12791da177e4SLinus Torvalds /* decide whether we should attempt the mapping, and if so what sort of 12801da177e4SLinus Torvalds * mapping */ 12811da177e4SLinus Torvalds ret = validate_mmap_request(file, addr, len, prot, flags, pgoff, 12821da177e4SLinus Torvalds &capabilities); 12838feae131SDavid Howells if (ret < 0) { 12848feae131SDavid Howells kleave(" = %d [val]", ret); 12851da177e4SLinus Torvalds return ret; 12868feae131SDavid Howells } 12871da177e4SLinus Torvalds 128806aab5a3SDavid Howells /* we ignore the address hint */ 128906aab5a3SDavid Howells addr = 0; 1290f67d9b15SBob Liu len = PAGE_ALIGN(len); 129106aab5a3SDavid Howells 12921da177e4SLinus Torvalds /* we've determined that we can make the mapping, now translate what we 12931da177e4SLinus Torvalds * now know into VMA flags */ 12941da177e4SLinus Torvalds vm_flags = determine_vm_flags(file, prot, flags, capabilities); 12951da177e4SLinus Torvalds 12968feae131SDavid Howells /* we're going to need to record the mapping */ 12978feae131SDavid Howells region = kmem_cache_zalloc(vm_region_jar, GFP_KERNEL); 12988feae131SDavid Howells if (!region) 12998feae131SDavid Howells goto error_getting_region; 13001da177e4SLinus Torvalds 13018feae131SDavid Howells vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL); 13028feae131SDavid Howells if (!vma) 13038feae131SDavid Howells goto error_getting_vma; 13041da177e4SLinus Torvalds 13051e2ae599SDavid Howells region->vm_usage = 1; 13068feae131SDavid Howells region->vm_flags = vm_flags; 13078feae131SDavid Howells region->vm_pgoff = pgoff; 13088feae131SDavid Howells 13095beb4930SRik van Riel INIT_LIST_HEAD(&vma->anon_vma_chain); 13108feae131SDavid Howells vma->vm_flags = vm_flags; 13118feae131SDavid Howells vma->vm_pgoff = pgoff; 13128feae131SDavid Howells 13138feae131SDavid Howells if (file) { 1314cb0942b8SAl Viro region->vm_file = get_file(file); 1315cb0942b8SAl Viro vma->vm_file = get_file(file); 13168feae131SDavid Howells } 13178feae131SDavid Howells 13188feae131SDavid Howells down_write(&nommu_region_sem); 13198feae131SDavid Howells 13208feae131SDavid Howells /* if we want to share, we need to check for regions created by other 13211da177e4SLinus Torvalds * mmap() calls that overlap with our proposed mapping 13228feae131SDavid Howells * - we can only share with a superset match on most regular files 13231da177e4SLinus Torvalds * - shared mappings on character devices and memory backed files are 13241da177e4SLinus Torvalds * permitted to overlap inexactly as far as we are concerned for in 13251da177e4SLinus Torvalds * these cases, sharing is handled in the driver or filesystem rather 13261da177e4SLinus Torvalds * than here 13271da177e4SLinus Torvalds */ 13281da177e4SLinus Torvalds if (vm_flags & VM_MAYSHARE) { 13298feae131SDavid Howells struct vm_region *pregion; 13308feae131SDavid Howells unsigned long pglen, rpglen, pgend, rpgend, start; 13311da177e4SLinus Torvalds 13328feae131SDavid Howells pglen = (len + PAGE_SIZE - 1) >> PAGE_SHIFT; 13338feae131SDavid Howells pgend = pgoff + pglen; 1334165b2392SDavid Howells 13358feae131SDavid Howells for (rb = rb_first(&nommu_region_tree); rb; rb = rb_next(rb)) { 13368feae131SDavid Howells pregion = rb_entry(rb, struct vm_region, vm_rb); 13371da177e4SLinus Torvalds 13388feae131SDavid Howells if (!(pregion->vm_flags & VM_MAYSHARE)) 13391da177e4SLinus Torvalds continue; 13401da177e4SLinus Torvalds 13411da177e4SLinus Torvalds /* search for overlapping mappings on the same file */ 1342496ad9aaSAl Viro if (file_inode(pregion->vm_file) != 1343496ad9aaSAl Viro file_inode(file)) 13441da177e4SLinus Torvalds continue; 13451da177e4SLinus Torvalds 13468feae131SDavid Howells if (pregion->vm_pgoff >= pgend) 13471da177e4SLinus Torvalds continue; 13481da177e4SLinus Torvalds 13498feae131SDavid Howells rpglen = pregion->vm_end - pregion->vm_start; 13508feae131SDavid Howells rpglen = (rpglen + PAGE_SIZE - 1) >> PAGE_SHIFT; 13518feae131SDavid Howells rpgend = pregion->vm_pgoff + rpglen; 13528feae131SDavid Howells if (pgoff >= rpgend) 13531da177e4SLinus Torvalds continue; 13541da177e4SLinus Torvalds 13558feae131SDavid Howells /* handle inexactly overlapping matches between 13568feae131SDavid Howells * mappings */ 13578feae131SDavid Howells if ((pregion->vm_pgoff != pgoff || rpglen != pglen) && 13588feae131SDavid Howells !(pgoff >= pregion->vm_pgoff && pgend <= rpgend)) { 13598feae131SDavid Howells /* new mapping is not a subset of the region */ 13601da177e4SLinus Torvalds if (!(capabilities & BDI_CAP_MAP_DIRECT)) 13611da177e4SLinus Torvalds goto sharing_violation; 13621da177e4SLinus Torvalds continue; 13631da177e4SLinus Torvalds } 13641da177e4SLinus Torvalds 13658feae131SDavid Howells /* we've found a region we can share */ 13661e2ae599SDavid Howells pregion->vm_usage++; 13678feae131SDavid Howells vma->vm_region = pregion; 13688feae131SDavid Howells start = pregion->vm_start; 13698feae131SDavid Howells start += (pgoff - pregion->vm_pgoff) << PAGE_SHIFT; 13708feae131SDavid Howells vma->vm_start = start; 13718feae131SDavid Howells vma->vm_end = start + len; 13721da177e4SLinus Torvalds 13738feae131SDavid Howells if (pregion->vm_flags & VM_MAPPED_COPY) { 13748feae131SDavid Howells kdebug("share copy"); 13758feae131SDavid Howells vma->vm_flags |= VM_MAPPED_COPY; 13768feae131SDavid Howells } else { 13778feae131SDavid Howells kdebug("share mmap"); 13788feae131SDavid Howells ret = do_mmap_shared_file(vma); 13798feae131SDavid Howells if (ret < 0) { 13808feae131SDavid Howells vma->vm_region = NULL; 13818feae131SDavid Howells vma->vm_start = 0; 13828feae131SDavid Howells vma->vm_end = 0; 13831e2ae599SDavid Howells pregion->vm_usage--; 13848feae131SDavid Howells pregion = NULL; 13858feae131SDavid Howells goto error_just_free; 13861da177e4SLinus Torvalds } 13878feae131SDavid Howells } 13888feae131SDavid Howells fput(region->vm_file); 13898feae131SDavid Howells kmem_cache_free(vm_region_jar, region); 13908feae131SDavid Howells region = pregion; 13918feae131SDavid Howells result = start; 13928feae131SDavid Howells goto share; 13938feae131SDavid Howells } 13941da177e4SLinus Torvalds 13951da177e4SLinus Torvalds /* obtain the address at which to make a shared mapping 13961da177e4SLinus Torvalds * - this is the hook for quasi-memory character devices to 13971da177e4SLinus Torvalds * tell us the location of a shared mapping 13981da177e4SLinus Torvalds */ 1399645d83c5SDavid Howells if (capabilities & BDI_CAP_MAP_DIRECT) { 14001da177e4SLinus Torvalds addr = file->f_op->get_unmapped_area(file, addr, len, 14011da177e4SLinus Torvalds pgoff, flags); 1402bb005a59SNamhyung Kim if (IS_ERR_VALUE(addr)) { 14031da177e4SLinus Torvalds ret = addr; 1404bb005a59SNamhyung Kim if (ret != -ENOSYS) 14058feae131SDavid Howells goto error_just_free; 14061da177e4SLinus Torvalds 14071da177e4SLinus Torvalds /* the driver refused to tell us where to site 14081da177e4SLinus Torvalds * the mapping so we'll have to attempt to copy 14091da177e4SLinus Torvalds * it */ 1410bb005a59SNamhyung Kim ret = -ENODEV; 14111da177e4SLinus Torvalds if (!(capabilities & BDI_CAP_MAP_COPY)) 14128feae131SDavid Howells goto error_just_free; 14131da177e4SLinus Torvalds 14141da177e4SLinus Torvalds capabilities &= ~BDI_CAP_MAP_DIRECT; 14158feae131SDavid Howells } else { 14168feae131SDavid Howells vma->vm_start = region->vm_start = addr; 14178feae131SDavid Howells vma->vm_end = region->vm_end = addr + len; 14181da177e4SLinus Torvalds } 14191da177e4SLinus Torvalds } 14201da177e4SLinus Torvalds } 14211da177e4SLinus Torvalds 14228feae131SDavid Howells vma->vm_region = region; 14231da177e4SLinus Torvalds 1424645d83c5SDavid Howells /* set up the mapping 1425645d83c5SDavid Howells * - the region is filled in if BDI_CAP_MAP_DIRECT is still set 1426645d83c5SDavid Howells */ 14271da177e4SLinus Torvalds if (file && vma->vm_flags & VM_SHARED) 14288feae131SDavid Howells ret = do_mmap_shared_file(vma); 14291da177e4SLinus Torvalds else 1430645d83c5SDavid Howells ret = do_mmap_private(vma, region, len, capabilities); 14311da177e4SLinus Torvalds if (ret < 0) 1432645d83c5SDavid Howells goto error_just_free; 1433645d83c5SDavid Howells add_nommu_region(region); 14348feae131SDavid Howells 1435ea637639SJie Zhang /* clear anonymous mappings that don't ask for uninitialized data */ 1436ea637639SJie Zhang if (!vma->vm_file && !(flags & MAP_UNINITIALIZED)) 1437ea637639SJie Zhang memset((void *)region->vm_start, 0, 1438ea637639SJie Zhang region->vm_end - region->vm_start); 1439ea637639SJie Zhang 14401da177e4SLinus Torvalds /* okay... we have a mapping; now we have to register it */ 14418feae131SDavid Howells result = vma->vm_start; 14421da177e4SLinus Torvalds 14431da177e4SLinus Torvalds current->mm->total_vm += len >> PAGE_SHIFT; 14441da177e4SLinus Torvalds 14458feae131SDavid Howells share: 14468feae131SDavid Howells add_vma_to_mm(current->mm, vma); 14471da177e4SLinus Torvalds 1448cfe79c00SMike Frysinger /* we flush the region from the icache only when the first executable 1449cfe79c00SMike Frysinger * mapping of it is made */ 1450cfe79c00SMike Frysinger if (vma->vm_flags & VM_EXEC && !region->vm_icache_flushed) { 1451cfe79c00SMike Frysinger flush_icache_range(region->vm_start, region->vm_end); 1452cfe79c00SMike Frysinger region->vm_icache_flushed = true; 1453cfe79c00SMike Frysinger } 14541da177e4SLinus Torvalds 1455cfe79c00SMike Frysinger up_write(&nommu_region_sem); 14561da177e4SLinus Torvalds 14578feae131SDavid Howells kleave(" = %lx", result); 14588feae131SDavid Howells return result; 14591da177e4SLinus Torvalds 14608feae131SDavid Howells error_just_free: 14618feae131SDavid Howells up_write(&nommu_region_sem); 14628feae131SDavid Howells error: 146389a86402SDavid Howells if (region->vm_file) 14648feae131SDavid Howells fput(region->vm_file); 14658feae131SDavid Howells kmem_cache_free(vm_region_jar, region); 146689a86402SDavid Howells if (vma->vm_file) 14678feae131SDavid Howells fput(vma->vm_file); 14688feae131SDavid Howells kmem_cache_free(vm_area_cachep, vma); 14698feae131SDavid Howells kleave(" = %d", ret); 14701da177e4SLinus Torvalds return ret; 14711da177e4SLinus Torvalds 14721da177e4SLinus Torvalds sharing_violation: 14738feae131SDavid Howells up_write(&nommu_region_sem); 14748feae131SDavid Howells printk(KERN_WARNING "Attempt to share mismatched mappings\n"); 14758feae131SDavid Howells ret = -EINVAL; 14768feae131SDavid Howells goto error; 14771da177e4SLinus Torvalds 14781da177e4SLinus Torvalds error_getting_vma: 14798feae131SDavid Howells kmem_cache_free(vm_region_jar, region); 14808feae131SDavid Howells printk(KERN_WARNING "Allocation of vma for %lu byte allocation" 14818feae131SDavid Howells " from process %d failed\n", 14821da177e4SLinus Torvalds len, current->pid); 14837bf02ea2SDavid Rientjes show_free_areas(0); 14841da177e4SLinus Torvalds return -ENOMEM; 14851da177e4SLinus Torvalds 14868feae131SDavid Howells error_getting_region: 14878feae131SDavid Howells printk(KERN_WARNING "Allocation of vm region for %lu byte allocation" 14888feae131SDavid Howells " from process %d failed\n", 14891da177e4SLinus Torvalds len, current->pid); 14907bf02ea2SDavid Rientjes show_free_areas(0); 14911da177e4SLinus Torvalds return -ENOMEM; 14921da177e4SLinus Torvalds } 14936be5ceb0SLinus Torvalds 149466f0dc48SHugh Dickins SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len, 149566f0dc48SHugh Dickins unsigned long, prot, unsigned long, flags, 149666f0dc48SHugh Dickins unsigned long, fd, unsigned long, pgoff) 149766f0dc48SHugh Dickins { 149866f0dc48SHugh Dickins struct file *file = NULL; 149966f0dc48SHugh Dickins unsigned long retval = -EBADF; 150066f0dc48SHugh Dickins 1501120a795dSAl Viro audit_mmap_fd(fd, flags); 150266f0dc48SHugh Dickins if (!(flags & MAP_ANONYMOUS)) { 150366f0dc48SHugh Dickins file = fget(fd); 150466f0dc48SHugh Dickins if (!file) 150566f0dc48SHugh Dickins goto out; 150666f0dc48SHugh Dickins } 150766f0dc48SHugh Dickins 150866f0dc48SHugh Dickins flags &= ~(MAP_EXECUTABLE | MAP_DENYWRITE); 150966f0dc48SHugh Dickins 1510ad1ed293SGreg Ungerer retval = vm_mmap_pgoff(file, addr, len, prot, flags, pgoff); 151166f0dc48SHugh Dickins 151266f0dc48SHugh Dickins if (file) 151366f0dc48SHugh Dickins fput(file); 151466f0dc48SHugh Dickins out: 151566f0dc48SHugh Dickins return retval; 151666f0dc48SHugh Dickins } 151766f0dc48SHugh Dickins 1518a4679373SChristoph Hellwig #ifdef __ARCH_WANT_SYS_OLD_MMAP 1519a4679373SChristoph Hellwig struct mmap_arg_struct { 1520a4679373SChristoph Hellwig unsigned long addr; 1521a4679373SChristoph Hellwig unsigned long len; 1522a4679373SChristoph Hellwig unsigned long prot; 1523a4679373SChristoph Hellwig unsigned long flags; 1524a4679373SChristoph Hellwig unsigned long fd; 1525a4679373SChristoph Hellwig unsigned long offset; 1526a4679373SChristoph Hellwig }; 1527a4679373SChristoph Hellwig 1528a4679373SChristoph Hellwig SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg) 1529a4679373SChristoph Hellwig { 1530a4679373SChristoph Hellwig struct mmap_arg_struct a; 1531a4679373SChristoph Hellwig 1532a4679373SChristoph Hellwig if (copy_from_user(&a, arg, sizeof(a))) 1533a4679373SChristoph Hellwig return -EFAULT; 1534a4679373SChristoph Hellwig if (a.offset & ~PAGE_MASK) 1535a4679373SChristoph Hellwig return -EINVAL; 1536a4679373SChristoph Hellwig 1537a4679373SChristoph Hellwig return sys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd, 1538a4679373SChristoph Hellwig a.offset >> PAGE_SHIFT); 1539a4679373SChristoph Hellwig } 1540a4679373SChristoph Hellwig #endif /* __ARCH_WANT_SYS_OLD_MMAP */ 1541a4679373SChristoph Hellwig 15421da177e4SLinus Torvalds /* 15438feae131SDavid Howells * split a vma into two pieces at address 'addr', a new vma is allocated either 15448feae131SDavid Howells * for the first part or the tail. 15451da177e4SLinus Torvalds */ 15468feae131SDavid Howells int split_vma(struct mm_struct *mm, struct vm_area_struct *vma, 15478feae131SDavid Howells unsigned long addr, int new_below) 15481da177e4SLinus Torvalds { 15498feae131SDavid Howells struct vm_area_struct *new; 15508feae131SDavid Howells struct vm_region *region; 15518feae131SDavid Howells unsigned long npages; 15521da177e4SLinus Torvalds 15538feae131SDavid Howells kenter(""); 15541da177e4SLinus Torvalds 1555779c1023SDavid Howells /* we're only permitted to split anonymous regions (these should have 1556779c1023SDavid Howells * only a single usage on the region) */ 1557779c1023SDavid Howells if (vma->vm_file) 15588feae131SDavid Howells return -ENOMEM; 15591da177e4SLinus Torvalds 15608feae131SDavid Howells if (mm->map_count >= sysctl_max_map_count) 15618feae131SDavid Howells return -ENOMEM; 15621da177e4SLinus Torvalds 15638feae131SDavid Howells region = kmem_cache_alloc(vm_region_jar, GFP_KERNEL); 15648feae131SDavid Howells if (!region) 15658feae131SDavid Howells return -ENOMEM; 15668feae131SDavid Howells 15678feae131SDavid Howells new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL); 15688feae131SDavid Howells if (!new) { 15698feae131SDavid Howells kmem_cache_free(vm_region_jar, region); 15708feae131SDavid Howells return -ENOMEM; 15711da177e4SLinus Torvalds } 15721da177e4SLinus Torvalds 15738feae131SDavid Howells /* most fields are the same, copy all, and then fixup */ 15748feae131SDavid Howells *new = *vma; 15758feae131SDavid Howells *region = *vma->vm_region; 15768feae131SDavid Howells new->vm_region = region; 15778feae131SDavid Howells 15788feae131SDavid Howells npages = (addr - vma->vm_start) >> PAGE_SHIFT; 15798feae131SDavid Howells 15808feae131SDavid Howells if (new_below) { 1581dd8632a1SPaul Mundt region->vm_top = region->vm_end = new->vm_end = addr; 15828feae131SDavid Howells } else { 15838feae131SDavid Howells region->vm_start = new->vm_start = addr; 15848feae131SDavid Howells region->vm_pgoff = new->vm_pgoff += npages; 15851da177e4SLinus Torvalds } 15868feae131SDavid Howells 15878feae131SDavid Howells if (new->vm_ops && new->vm_ops->open) 15888feae131SDavid Howells new->vm_ops->open(new); 15898feae131SDavid Howells 15908feae131SDavid Howells delete_vma_from_mm(vma); 15918feae131SDavid Howells down_write(&nommu_region_sem); 15928feae131SDavid Howells delete_nommu_region(vma->vm_region); 15938feae131SDavid Howells if (new_below) { 15948feae131SDavid Howells vma->vm_region->vm_start = vma->vm_start = addr; 15958feae131SDavid Howells vma->vm_region->vm_pgoff = vma->vm_pgoff += npages; 15968feae131SDavid Howells } else { 15978feae131SDavid Howells vma->vm_region->vm_end = vma->vm_end = addr; 1598dd8632a1SPaul Mundt vma->vm_region->vm_top = addr; 15998feae131SDavid Howells } 16008feae131SDavid Howells add_nommu_region(vma->vm_region); 16018feae131SDavid Howells add_nommu_region(new->vm_region); 16028feae131SDavid Howells up_write(&nommu_region_sem); 16038feae131SDavid Howells add_vma_to_mm(mm, vma); 16048feae131SDavid Howells add_vma_to_mm(mm, new); 16058feae131SDavid Howells return 0; 16068feae131SDavid Howells } 16078feae131SDavid Howells 16088feae131SDavid Howells /* 16098feae131SDavid Howells * shrink a VMA by removing the specified chunk from either the beginning or 16108feae131SDavid Howells * the end 16118feae131SDavid Howells */ 16128feae131SDavid Howells static int shrink_vma(struct mm_struct *mm, 16138feae131SDavid Howells struct vm_area_struct *vma, 16148feae131SDavid Howells unsigned long from, unsigned long to) 16158feae131SDavid Howells { 16168feae131SDavid Howells struct vm_region *region; 16178feae131SDavid Howells 16188feae131SDavid Howells kenter(""); 16198feae131SDavid Howells 16208feae131SDavid Howells /* adjust the VMA's pointers, which may reposition it in the MM's tree 16218feae131SDavid Howells * and list */ 16228feae131SDavid Howells delete_vma_from_mm(vma); 16238feae131SDavid Howells if (from > vma->vm_start) 16248feae131SDavid Howells vma->vm_end = from; 16258feae131SDavid Howells else 16268feae131SDavid Howells vma->vm_start = to; 16278feae131SDavid Howells add_vma_to_mm(mm, vma); 16288feae131SDavid Howells 16298feae131SDavid Howells /* cut the backing region down to size */ 16308feae131SDavid Howells region = vma->vm_region; 16311e2ae599SDavid Howells BUG_ON(region->vm_usage != 1); 16328feae131SDavid Howells 16338feae131SDavid Howells down_write(&nommu_region_sem); 16348feae131SDavid Howells delete_nommu_region(region); 1635dd8632a1SPaul Mundt if (from > region->vm_start) { 1636dd8632a1SPaul Mundt to = region->vm_top; 1637dd8632a1SPaul Mundt region->vm_top = region->vm_end = from; 1638dd8632a1SPaul Mundt } else { 16398feae131SDavid Howells region->vm_start = to; 1640dd8632a1SPaul Mundt } 16418feae131SDavid Howells add_nommu_region(region); 16428feae131SDavid Howells up_write(&nommu_region_sem); 16438feae131SDavid Howells 16448feae131SDavid Howells free_page_series(from, to); 16458feae131SDavid Howells return 0; 16461da177e4SLinus Torvalds } 16471da177e4SLinus Torvalds 16483034097aSDavid Howells /* 16493034097aSDavid Howells * release a mapping 16508feae131SDavid Howells * - under NOMMU conditions the chunk to be unmapped must be backed by a single 16518feae131SDavid Howells * VMA, though it need not cover the whole VMA 16523034097aSDavid Howells */ 16538feae131SDavid Howells int do_munmap(struct mm_struct *mm, unsigned long start, size_t len) 16541da177e4SLinus Torvalds { 16558feae131SDavid Howells struct vm_area_struct *vma; 1656f67d9b15SBob Liu unsigned long end; 16578feae131SDavid Howells int ret; 16581da177e4SLinus Torvalds 16598feae131SDavid Howells kenter(",%lx,%zx", start, len); 16601da177e4SLinus Torvalds 1661f67d9b15SBob Liu len = PAGE_ALIGN(len); 16628feae131SDavid Howells if (len == 0) 16631da177e4SLinus Torvalds return -EINVAL; 16641da177e4SLinus Torvalds 1665f67d9b15SBob Liu end = start + len; 1666f67d9b15SBob Liu 16678feae131SDavid Howells /* find the first potentially overlapping VMA */ 16688feae131SDavid Howells vma = find_vma(mm, start); 16698feae131SDavid Howells if (!vma) { 1670ac714904SChoi Gi-yong static int limit; 167133e5d769SDavid Howells if (limit < 5) { 16728feae131SDavid Howells printk(KERN_WARNING 167333e5d769SDavid Howells "munmap of memory not mmapped by process %d" 167433e5d769SDavid Howells " (%s): 0x%lx-0x%lx\n", 167533e5d769SDavid Howells current->pid, current->comm, 167633e5d769SDavid Howells start, start + len - 1); 167733e5d769SDavid Howells limit++; 167833e5d769SDavid Howells } 16798feae131SDavid Howells return -EINVAL; 16808feae131SDavid Howells } 16811da177e4SLinus Torvalds 16828feae131SDavid Howells /* we're allowed to split an anonymous VMA but not a file-backed one */ 16838feae131SDavid Howells if (vma->vm_file) { 16848feae131SDavid Howells do { 16858feae131SDavid Howells if (start > vma->vm_start) { 16868feae131SDavid Howells kleave(" = -EINVAL [miss]"); 16878feae131SDavid Howells return -EINVAL; 16888feae131SDavid Howells } 16898feae131SDavid Howells if (end == vma->vm_end) 16908feae131SDavid Howells goto erase_whole_vma; 1691d75a310cSNamhyung Kim vma = vma->vm_next; 1692d75a310cSNamhyung Kim } while (vma); 16938feae131SDavid Howells kleave(" = -EINVAL [split file]"); 16948feae131SDavid Howells return -EINVAL; 16958feae131SDavid Howells } else { 16968feae131SDavid Howells /* the chunk must be a subset of the VMA found */ 16978feae131SDavid Howells if (start == vma->vm_start && end == vma->vm_end) 16988feae131SDavid Howells goto erase_whole_vma; 16998feae131SDavid Howells if (start < vma->vm_start || end > vma->vm_end) { 17008feae131SDavid Howells kleave(" = -EINVAL [superset]"); 17018feae131SDavid Howells return -EINVAL; 17028feae131SDavid Howells } 17038feae131SDavid Howells if (start & ~PAGE_MASK) { 17048feae131SDavid Howells kleave(" = -EINVAL [unaligned start]"); 17058feae131SDavid Howells return -EINVAL; 17068feae131SDavid Howells } 17078feae131SDavid Howells if (end != vma->vm_end && end & ~PAGE_MASK) { 17088feae131SDavid Howells kleave(" = -EINVAL [unaligned split]"); 17098feae131SDavid Howells return -EINVAL; 17108feae131SDavid Howells } 17118feae131SDavid Howells if (start != vma->vm_start && end != vma->vm_end) { 17128feae131SDavid Howells ret = split_vma(mm, vma, start, 1); 17138feae131SDavid Howells if (ret < 0) { 17148feae131SDavid Howells kleave(" = %d [split]", ret); 17158feae131SDavid Howells return ret; 17168feae131SDavid Howells } 17178feae131SDavid Howells } 17188feae131SDavid Howells return shrink_vma(mm, vma, start, end); 17198feae131SDavid Howells } 17201da177e4SLinus Torvalds 17218feae131SDavid Howells erase_whole_vma: 17228feae131SDavid Howells delete_vma_from_mm(vma); 17238feae131SDavid Howells delete_vma(mm, vma); 17248feae131SDavid Howells kleave(" = 0"); 17251da177e4SLinus Torvalds return 0; 17261da177e4SLinus Torvalds } 1727b5073173SPaul Mundt EXPORT_SYMBOL(do_munmap); 17281da177e4SLinus Torvalds 1729bfce281cSAl Viro int vm_munmap(unsigned long addr, size_t len) 17303034097aSDavid Howells { 1731bfce281cSAl Viro struct mm_struct *mm = current->mm; 17323034097aSDavid Howells int ret; 17333034097aSDavid Howells 17343034097aSDavid Howells down_write(&mm->mmap_sem); 17353034097aSDavid Howells ret = do_munmap(mm, addr, len); 17363034097aSDavid Howells up_write(&mm->mmap_sem); 17373034097aSDavid Howells return ret; 17383034097aSDavid Howells } 1739a46ef99dSLinus Torvalds EXPORT_SYMBOL(vm_munmap); 1740a46ef99dSLinus Torvalds 1741a46ef99dSLinus Torvalds SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len) 1742a46ef99dSLinus Torvalds { 1743bfce281cSAl Viro return vm_munmap(addr, len); 1744a46ef99dSLinus Torvalds } 17453034097aSDavid Howells 17463034097aSDavid Howells /* 17478feae131SDavid Howells * release all the mappings made in a process's VM space 17483034097aSDavid Howells */ 17491da177e4SLinus Torvalds void exit_mmap(struct mm_struct *mm) 17501da177e4SLinus Torvalds { 17518feae131SDavid Howells struct vm_area_struct *vma; 17521da177e4SLinus Torvalds 17538feae131SDavid Howells if (!mm) 17548feae131SDavid Howells return; 17558feae131SDavid Howells 17568feae131SDavid Howells kenter(""); 17571da177e4SLinus Torvalds 17581da177e4SLinus Torvalds mm->total_vm = 0; 17591da177e4SLinus Torvalds 17608feae131SDavid Howells while ((vma = mm->mmap)) { 17618feae131SDavid Howells mm->mmap = vma->vm_next; 17628feae131SDavid Howells delete_vma_from_mm(vma); 17638feae131SDavid Howells delete_vma(mm, vma); 176404c34961SSteven J. Magnani cond_resched(); 17651da177e4SLinus Torvalds } 17661da177e4SLinus Torvalds 17678feae131SDavid Howells kleave(""); 17681da177e4SLinus Torvalds } 17691da177e4SLinus Torvalds 1770e4eb1ff6SLinus Torvalds unsigned long vm_brk(unsigned long addr, unsigned long len) 17711da177e4SLinus Torvalds { 17721da177e4SLinus Torvalds return -ENOMEM; 17731da177e4SLinus Torvalds } 17741da177e4SLinus Torvalds 17751da177e4SLinus Torvalds /* 17766fa5f80bSDavid Howells * expand (or shrink) an existing mapping, potentially moving it at the same 17776fa5f80bSDavid Howells * time (controlled by the MREMAP_MAYMOVE flag and available VM space) 17781da177e4SLinus Torvalds * 17796fa5f80bSDavid Howells * under NOMMU conditions, we only permit changing a mapping's size, and only 17808feae131SDavid Howells * as long as it stays within the region allocated by do_mmap_private() and the 17818feae131SDavid Howells * block is not shareable 17821da177e4SLinus Torvalds * 17836fa5f80bSDavid Howells * MREMAP_FIXED is not supported under NOMMU conditions 17841da177e4SLinus Torvalds */ 17854b377babSAl Viro static unsigned long do_mremap(unsigned long addr, 17861da177e4SLinus Torvalds unsigned long old_len, unsigned long new_len, 17871da177e4SLinus Torvalds unsigned long flags, unsigned long new_addr) 17881da177e4SLinus Torvalds { 17896fa5f80bSDavid Howells struct vm_area_struct *vma; 17901da177e4SLinus Torvalds 17911da177e4SLinus Torvalds /* insanity checks first */ 1792f67d9b15SBob Liu old_len = PAGE_ALIGN(old_len); 1793f67d9b15SBob Liu new_len = PAGE_ALIGN(new_len); 17948feae131SDavid Howells if (old_len == 0 || new_len == 0) 17951da177e4SLinus Torvalds return (unsigned long) -EINVAL; 17961da177e4SLinus Torvalds 17978feae131SDavid Howells if (addr & ~PAGE_MASK) 17988feae131SDavid Howells return -EINVAL; 17998feae131SDavid Howells 18001da177e4SLinus Torvalds if (flags & MREMAP_FIXED && new_addr != addr) 18011da177e4SLinus Torvalds return (unsigned long) -EINVAL; 18021da177e4SLinus Torvalds 18038feae131SDavid Howells vma = find_vma_exact(current->mm, addr, old_len); 18046fa5f80bSDavid Howells if (!vma) 18051da177e4SLinus Torvalds return (unsigned long) -EINVAL; 18061da177e4SLinus Torvalds 18076fa5f80bSDavid Howells if (vma->vm_end != vma->vm_start + old_len) 18081da177e4SLinus Torvalds return (unsigned long) -EFAULT; 18091da177e4SLinus Torvalds 18106fa5f80bSDavid Howells if (vma->vm_flags & VM_MAYSHARE) 18111da177e4SLinus Torvalds return (unsigned long) -EPERM; 18121da177e4SLinus Torvalds 18138feae131SDavid Howells if (new_len > vma->vm_region->vm_end - vma->vm_region->vm_start) 18141da177e4SLinus Torvalds return (unsigned long) -ENOMEM; 18151da177e4SLinus Torvalds 18161da177e4SLinus Torvalds /* all checks complete - do it */ 18176fa5f80bSDavid Howells vma->vm_end = vma->vm_start + new_len; 18186fa5f80bSDavid Howells return vma->vm_start; 18196fa5f80bSDavid Howells } 18206fa5f80bSDavid Howells 18216a6160a7SHeiko Carstens SYSCALL_DEFINE5(mremap, unsigned long, addr, unsigned long, old_len, 18226a6160a7SHeiko Carstens unsigned long, new_len, unsigned long, flags, 18236a6160a7SHeiko Carstens unsigned long, new_addr) 18246fa5f80bSDavid Howells { 18256fa5f80bSDavid Howells unsigned long ret; 18266fa5f80bSDavid Howells 18276fa5f80bSDavid Howells down_write(¤t->mm->mmap_sem); 18286fa5f80bSDavid Howells ret = do_mremap(addr, old_len, new_len, flags, new_addr); 18296fa5f80bSDavid Howells up_write(¤t->mm->mmap_sem); 18306fa5f80bSDavid Howells return ret; 18311da177e4SLinus Torvalds } 18321da177e4SLinus Torvalds 1833240aadeeSMichel Lespinasse struct page *follow_page_mask(struct vm_area_struct *vma, 1834240aadeeSMichel Lespinasse unsigned long address, unsigned int flags, 1835240aadeeSMichel Lespinasse unsigned int *page_mask) 18361da177e4SLinus Torvalds { 1837240aadeeSMichel Lespinasse *page_mask = 0; 18381da177e4SLinus Torvalds return NULL; 18391da177e4SLinus Torvalds } 18401da177e4SLinus Torvalds 18418f3b1327SBob Liu int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr, 18428f3b1327SBob Liu unsigned long pfn, unsigned long size, pgprot_t prot) 18431da177e4SLinus Torvalds { 18448f3b1327SBob Liu if (addr != (pfn << PAGE_SHIFT)) 18458f3b1327SBob Liu return -EINVAL; 18468f3b1327SBob Liu 1847314e51b9SKonstantin Khlebnikov vma->vm_flags |= VM_IO | VM_PFNMAP | VM_DONTEXPAND | VM_DONTDUMP; 184866aa2b4bSGreg Ungerer return 0; 18491da177e4SLinus Torvalds } 185022c4af40SLuke Yang EXPORT_SYMBOL(remap_pfn_range); 18511da177e4SLinus Torvalds 18523c0b9de6SLinus Torvalds int vm_iomap_memory(struct vm_area_struct *vma, phys_addr_t start, unsigned long len) 18533c0b9de6SLinus Torvalds { 18543c0b9de6SLinus Torvalds unsigned long pfn = start >> PAGE_SHIFT; 18553c0b9de6SLinus Torvalds unsigned long vm_len = vma->vm_end - vma->vm_start; 18563c0b9de6SLinus Torvalds 18573c0b9de6SLinus Torvalds pfn += vma->vm_pgoff; 18583c0b9de6SLinus Torvalds return io_remap_pfn_range(vma, vma->vm_start, pfn, vm_len, vma->vm_page_prot); 18593c0b9de6SLinus Torvalds } 18603c0b9de6SLinus Torvalds EXPORT_SYMBOL(vm_iomap_memory); 18613c0b9de6SLinus Torvalds 1862f905bc44SPaul Mundt int remap_vmalloc_range(struct vm_area_struct *vma, void *addr, 1863f905bc44SPaul Mundt unsigned long pgoff) 1864f905bc44SPaul Mundt { 1865f905bc44SPaul Mundt unsigned int size = vma->vm_end - vma->vm_start; 1866f905bc44SPaul Mundt 1867f905bc44SPaul Mundt if (!(vma->vm_flags & VM_USERMAP)) 1868f905bc44SPaul Mundt return -EINVAL; 1869f905bc44SPaul Mundt 1870f905bc44SPaul Mundt vma->vm_start = (unsigned long)(addr + (pgoff << PAGE_SHIFT)); 1871f905bc44SPaul Mundt vma->vm_end = vma->vm_start + size; 1872f905bc44SPaul Mundt 1873f905bc44SPaul Mundt return 0; 1874f905bc44SPaul Mundt } 1875f905bc44SPaul Mundt EXPORT_SYMBOL(remap_vmalloc_range); 1876f905bc44SPaul Mundt 18771da177e4SLinus Torvalds unsigned long arch_get_unmapped_area(struct file *file, unsigned long addr, 18781da177e4SLinus Torvalds unsigned long len, unsigned long pgoff, unsigned long flags) 18791da177e4SLinus Torvalds { 18801da177e4SLinus Torvalds return -ENOMEM; 18811da177e4SLinus Torvalds } 18821da177e4SLinus Torvalds 18831da177e4SLinus Torvalds void unmap_mapping_range(struct address_space *mapping, 18841da177e4SLinus Torvalds loff_t const holebegin, loff_t const holelen, 18851da177e4SLinus Torvalds int even_cows) 18861da177e4SLinus Torvalds { 18871da177e4SLinus Torvalds } 188822c4af40SLuke Yang EXPORT_SYMBOL(unmap_mapping_range); 18891da177e4SLinus Torvalds 18901da177e4SLinus Torvalds /* 18911da177e4SLinus Torvalds * Check that a process has enough memory to allocate a new virtual 18921da177e4SLinus Torvalds * mapping. 0 means there is enough memory for the allocation to 18931da177e4SLinus Torvalds * succeed and -ENOMEM implies there is not. 18941da177e4SLinus Torvalds * 18951da177e4SLinus Torvalds * We currently support three overcommit policies, which are set via the 18961da177e4SLinus Torvalds * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting 18971da177e4SLinus Torvalds * 18981da177e4SLinus Torvalds * Strict overcommit modes added 2002 Feb 26 by Alan Cox. 18991da177e4SLinus Torvalds * Additional code 2002 Jul 20 by Robert Love. 19001da177e4SLinus Torvalds * 19011da177e4SLinus Torvalds * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise. 19021da177e4SLinus Torvalds * 19031da177e4SLinus Torvalds * Note this is a helper function intended to be used by LSMs which 19041da177e4SLinus Torvalds * wish to use this logic. 19051da177e4SLinus Torvalds */ 190634b4e4aaSAlan Cox int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin) 19071da177e4SLinus Torvalds { 1908c9b1d098SAndrew Shewmaker unsigned long free, allowed, reserve; 19091da177e4SLinus Torvalds 19101da177e4SLinus Torvalds vm_acct_memory(pages); 19111da177e4SLinus Torvalds 19121da177e4SLinus Torvalds /* 19131da177e4SLinus Torvalds * Sometimes we want to use more memory than we have 19141da177e4SLinus Torvalds */ 19151da177e4SLinus Torvalds if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS) 19161da177e4SLinus Torvalds return 0; 19171da177e4SLinus Torvalds 19181da177e4SLinus Torvalds if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) { 1919c15bef30SDmitry Fink free = global_page_state(NR_FREE_PAGES); 1920c15bef30SDmitry Fink free += global_page_state(NR_FILE_PAGES); 19211da177e4SLinus Torvalds 1922c15bef30SDmitry Fink /* 1923c15bef30SDmitry Fink * shmem pages shouldn't be counted as free in this 1924c15bef30SDmitry Fink * case, they can't be purged, only swapped out, and 1925c15bef30SDmitry Fink * that won't affect the overall amount of available 1926c15bef30SDmitry Fink * memory in the system. 1927c15bef30SDmitry Fink */ 1928c15bef30SDmitry Fink free -= global_page_state(NR_SHMEM); 1929c15bef30SDmitry Fink 1930ec8acf20SShaohua Li free += get_nr_swap_pages(); 19311da177e4SLinus Torvalds 19321da177e4SLinus Torvalds /* 19331da177e4SLinus Torvalds * Any slabs which are created with the 19341da177e4SLinus Torvalds * SLAB_RECLAIM_ACCOUNT flag claim to have contents 19351da177e4SLinus Torvalds * which are reclaimable, under pressure. The dentry 19361da177e4SLinus Torvalds * cache and most inode caches should fall into this 19371da177e4SLinus Torvalds */ 1938972d1a7bSChristoph Lameter free += global_page_state(NR_SLAB_RECLAIMABLE); 19391da177e4SLinus Torvalds 19401da177e4SLinus Torvalds /* 1941c15bef30SDmitry Fink * Leave reserved pages. The pages are not for anonymous pages. 1942c15bef30SDmitry Fink */ 1943c15bef30SDmitry Fink if (free <= totalreserve_pages) 1944c15bef30SDmitry Fink goto error; 1945c15bef30SDmitry Fink else 1946c15bef30SDmitry Fink free -= totalreserve_pages; 1947c15bef30SDmitry Fink 1948c15bef30SDmitry Fink /* 19494eeab4f5SAndrew Shewmaker * Reserve some for root 19501da177e4SLinus Torvalds */ 19511da177e4SLinus Torvalds if (!cap_sys_admin) 19524eeab4f5SAndrew Shewmaker free -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10); 19531da177e4SLinus Torvalds 19541da177e4SLinus Torvalds if (free > pages) 19551da177e4SLinus Torvalds return 0; 19561da177e4SLinus Torvalds 1957d5ddc79bSHideo AOKI goto error; 19581da177e4SLinus Torvalds } 19591da177e4SLinus Torvalds 196000619bccSJerome Marchand allowed = vm_commit_limit(); 19611da177e4SLinus Torvalds /* 19624eeab4f5SAndrew Shewmaker * Reserve some 3% for root 19631da177e4SLinus Torvalds */ 19641da177e4SLinus Torvalds if (!cap_sys_admin) 19654eeab4f5SAndrew Shewmaker allowed -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10); 19661da177e4SLinus Torvalds 1967c9b1d098SAndrew Shewmaker /* 1968c9b1d098SAndrew Shewmaker * Don't let a single process grow so big a user can't recover 1969c9b1d098SAndrew Shewmaker */ 1970c9b1d098SAndrew Shewmaker if (mm) { 1971c9b1d098SAndrew Shewmaker reserve = sysctl_user_reserve_kbytes >> (PAGE_SHIFT - 10); 1972c9b1d098SAndrew Shewmaker allowed -= min(mm->total_vm / 32, reserve); 1973c9b1d098SAndrew Shewmaker } 19741da177e4SLinus Torvalds 197500a62ce9SKOSAKI Motohiro if (percpu_counter_read_positive(&vm_committed_as) < allowed) 19761da177e4SLinus Torvalds return 0; 197700a62ce9SKOSAKI Motohiro 1978d5ddc79bSHideo AOKI error: 19791da177e4SLinus Torvalds vm_unacct_memory(pages); 19801da177e4SLinus Torvalds 19811da177e4SLinus Torvalds return -ENOMEM; 19821da177e4SLinus Torvalds } 19831da177e4SLinus Torvalds 1984cae5d390SStephen Wilson int in_gate_area_no_mm(unsigned long addr) 19851da177e4SLinus Torvalds { 19861da177e4SLinus Torvalds return 0; 19871da177e4SLinus Torvalds } 1988b0e15190SDavid Howells 1989d0217ac0SNick Piggin int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf) 1990b0e15190SDavid Howells { 1991b0e15190SDavid Howells BUG(); 1992d0217ac0SNick Piggin return 0; 1993b0e15190SDavid Howells } 1994b5073173SPaul Mundt EXPORT_SYMBOL(filemap_fault); 19950ec76a11SDavid Howells 1996f1820361SKirill A. Shutemov void filemap_map_pages(struct vm_area_struct *vma, struct vm_fault *vmf) 1997f1820361SKirill A. Shutemov { 1998f1820361SKirill A. Shutemov BUG(); 1999f1820361SKirill A. Shutemov } 2000f1820361SKirill A. Shutemov EXPORT_SYMBOL(filemap_map_pages); 2001f1820361SKirill A. Shutemov 20020b173bc4SKonstantin Khlebnikov int generic_file_remap_pages(struct vm_area_struct *vma, unsigned long addr, 20030b173bc4SKonstantin Khlebnikov unsigned long size, pgoff_t pgoff) 20040b173bc4SKonstantin Khlebnikov { 20050b173bc4SKonstantin Khlebnikov BUG(); 20060b173bc4SKonstantin Khlebnikov return 0; 20070b173bc4SKonstantin Khlebnikov } 20080b173bc4SKonstantin Khlebnikov EXPORT_SYMBOL(generic_file_remap_pages); 20090b173bc4SKonstantin Khlebnikov 2010f55f199bSMike Frysinger static int __access_remote_vm(struct task_struct *tsk, struct mm_struct *mm, 2011f55f199bSMike Frysinger unsigned long addr, void *buf, int len, int write) 20120ec76a11SDavid Howells { 20130ec76a11SDavid Howells struct vm_area_struct *vma; 20140ec76a11SDavid Howells 20150ec76a11SDavid Howells down_read(&mm->mmap_sem); 20160ec76a11SDavid Howells 20170ec76a11SDavid Howells /* the access must start within one of the target process's mappings */ 20180159b141SDavid Howells vma = find_vma(mm, addr); 20190159b141SDavid Howells if (vma) { 20200ec76a11SDavid Howells /* don't overrun this mapping */ 20210ec76a11SDavid Howells if (addr + len >= vma->vm_end) 20220ec76a11SDavid Howells len = vma->vm_end - addr; 20230ec76a11SDavid Howells 20240ec76a11SDavid Howells /* only read or write mappings where it is permitted */ 2025d00c7b99SDavid Howells if (write && vma->vm_flags & VM_MAYWRITE) 20267959722bSJie Zhang copy_to_user_page(vma, NULL, addr, 20277959722bSJie Zhang (void *) addr, buf, len); 2028d00c7b99SDavid Howells else if (!write && vma->vm_flags & VM_MAYREAD) 20297959722bSJie Zhang copy_from_user_page(vma, NULL, addr, 20307959722bSJie Zhang buf, (void *) addr, len); 20310ec76a11SDavid Howells else 20320ec76a11SDavid Howells len = 0; 20330ec76a11SDavid Howells } else { 20340ec76a11SDavid Howells len = 0; 20350ec76a11SDavid Howells } 20360ec76a11SDavid Howells 20370ec76a11SDavid Howells up_read(&mm->mmap_sem); 2038f55f199bSMike Frysinger 2039f55f199bSMike Frysinger return len; 2040f55f199bSMike Frysinger } 2041f55f199bSMike Frysinger 2042f55f199bSMike Frysinger /** 2043f55f199bSMike Frysinger * @access_remote_vm - access another process' address space 2044f55f199bSMike Frysinger * @mm: the mm_struct of the target address space 2045f55f199bSMike Frysinger * @addr: start address to access 2046f55f199bSMike Frysinger * @buf: source or destination buffer 2047f55f199bSMike Frysinger * @len: number of bytes to transfer 2048f55f199bSMike Frysinger * @write: whether the access is a write 2049f55f199bSMike Frysinger * 2050f55f199bSMike Frysinger * The caller must hold a reference on @mm. 2051f55f199bSMike Frysinger */ 2052f55f199bSMike Frysinger int access_remote_vm(struct mm_struct *mm, unsigned long addr, 2053f55f199bSMike Frysinger void *buf, int len, int write) 2054f55f199bSMike Frysinger { 2055f55f199bSMike Frysinger return __access_remote_vm(NULL, mm, addr, buf, len, write); 2056f55f199bSMike Frysinger } 2057f55f199bSMike Frysinger 2058f55f199bSMike Frysinger /* 2059f55f199bSMike Frysinger * Access another process' address space. 2060f55f199bSMike Frysinger * - source/target buffer must be kernel space 2061f55f199bSMike Frysinger */ 2062f55f199bSMike Frysinger int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len, int write) 2063f55f199bSMike Frysinger { 2064f55f199bSMike Frysinger struct mm_struct *mm; 2065f55f199bSMike Frysinger 2066f55f199bSMike Frysinger if (addr + len < addr) 2067f55f199bSMike Frysinger return 0; 2068f55f199bSMike Frysinger 2069f55f199bSMike Frysinger mm = get_task_mm(tsk); 2070f55f199bSMike Frysinger if (!mm) 2071f55f199bSMike Frysinger return 0; 2072f55f199bSMike Frysinger 2073f55f199bSMike Frysinger len = __access_remote_vm(tsk, mm, addr, buf, len, write); 2074f55f199bSMike Frysinger 20750ec76a11SDavid Howells mmput(mm); 20760ec76a11SDavid Howells return len; 20770ec76a11SDavid Howells } 20787e660872SDavid Howells 20797e660872SDavid Howells /** 20807e660872SDavid Howells * nommu_shrink_inode_mappings - Shrink the shared mappings on an inode 20817e660872SDavid Howells * @inode: The inode to check 20827e660872SDavid Howells * @size: The current filesize of the inode 20837e660872SDavid Howells * @newsize: The proposed filesize of the inode 20847e660872SDavid Howells * 20857e660872SDavid Howells * Check the shared mappings on an inode on behalf of a shrinking truncate to 20867e660872SDavid Howells * make sure that that any outstanding VMAs aren't broken and then shrink the 20877e660872SDavid Howells * vm_regions that extend that beyond so that do_mmap_pgoff() doesn't 20887e660872SDavid Howells * automatically grant mappings that are too large. 20897e660872SDavid Howells */ 20907e660872SDavid Howells int nommu_shrink_inode_mappings(struct inode *inode, size_t size, 20917e660872SDavid Howells size_t newsize) 20927e660872SDavid Howells { 20937e660872SDavid Howells struct vm_area_struct *vma; 20947e660872SDavid Howells struct vm_region *region; 20957e660872SDavid Howells pgoff_t low, high; 20967e660872SDavid Howells size_t r_size, r_top; 20977e660872SDavid Howells 20987e660872SDavid Howells low = newsize >> PAGE_SHIFT; 20997e660872SDavid Howells high = (size + PAGE_SIZE - 1) >> PAGE_SHIFT; 21007e660872SDavid Howells 21017e660872SDavid Howells down_write(&nommu_region_sem); 2102918e556eSDavid Howells mutex_lock(&inode->i_mapping->i_mmap_mutex); 21037e660872SDavid Howells 21047e660872SDavid Howells /* search for VMAs that fall within the dead zone */ 21056b2dbba8SMichel Lespinasse vma_interval_tree_foreach(vma, &inode->i_mapping->i_mmap, low, high) { 21067e660872SDavid Howells /* found one - only interested if it's shared out of the page 21077e660872SDavid Howells * cache */ 21087e660872SDavid Howells if (vma->vm_flags & VM_SHARED) { 2109918e556eSDavid Howells mutex_unlock(&inode->i_mapping->i_mmap_mutex); 21107e660872SDavid Howells up_write(&nommu_region_sem); 21117e660872SDavid Howells return -ETXTBSY; /* not quite true, but near enough */ 21127e660872SDavid Howells } 21137e660872SDavid Howells } 21147e660872SDavid Howells 21157e660872SDavid Howells /* reduce any regions that overlap the dead zone - if in existence, 21167e660872SDavid Howells * these will be pointed to by VMAs that don't overlap the dead zone 21177e660872SDavid Howells * 21187e660872SDavid Howells * we don't check for any regions that start beyond the EOF as there 21197e660872SDavid Howells * shouldn't be any 21207e660872SDavid Howells */ 21216b2dbba8SMichel Lespinasse vma_interval_tree_foreach(vma, &inode->i_mapping->i_mmap, 21227e660872SDavid Howells 0, ULONG_MAX) { 21237e660872SDavid Howells if (!(vma->vm_flags & VM_SHARED)) 21247e660872SDavid Howells continue; 21257e660872SDavid Howells 21267e660872SDavid Howells region = vma->vm_region; 21277e660872SDavid Howells r_size = region->vm_top - region->vm_start; 21287e660872SDavid Howells r_top = (region->vm_pgoff << PAGE_SHIFT) + r_size; 21297e660872SDavid Howells 21307e660872SDavid Howells if (r_top > newsize) { 21317e660872SDavid Howells region->vm_top -= r_top - newsize; 21327e660872SDavid Howells if (region->vm_end > region->vm_top) 21337e660872SDavid Howells region->vm_end = region->vm_top; 21347e660872SDavid Howells } 21357e660872SDavid Howells } 21367e660872SDavid Howells 2137918e556eSDavid Howells mutex_unlock(&inode->i_mapping->i_mmap_mutex); 21387e660872SDavid Howells up_write(&nommu_region_sem); 21397e660872SDavid Howells return 0; 21407e660872SDavid Howells } 2141c9b1d098SAndrew Shewmaker 2142c9b1d098SAndrew Shewmaker /* 2143c9b1d098SAndrew Shewmaker * Initialise sysctl_user_reserve_kbytes. 2144c9b1d098SAndrew Shewmaker * 2145c9b1d098SAndrew Shewmaker * This is intended to prevent a user from starting a single memory hogging 2146c9b1d098SAndrew Shewmaker * process, such that they cannot recover (kill the hog) in OVERCOMMIT_NEVER 2147c9b1d098SAndrew Shewmaker * mode. 2148c9b1d098SAndrew Shewmaker * 2149c9b1d098SAndrew Shewmaker * The default value is min(3% of free memory, 128MB) 2150c9b1d098SAndrew Shewmaker * 128MB is enough to recover with sshd/login, bash, and top/kill. 2151c9b1d098SAndrew Shewmaker */ 2152c9b1d098SAndrew Shewmaker static int __meminit init_user_reserve(void) 2153c9b1d098SAndrew Shewmaker { 2154c9b1d098SAndrew Shewmaker unsigned long free_kbytes; 2155c9b1d098SAndrew Shewmaker 2156c9b1d098SAndrew Shewmaker free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10); 2157c9b1d098SAndrew Shewmaker 2158c9b1d098SAndrew Shewmaker sysctl_user_reserve_kbytes = min(free_kbytes / 32, 1UL << 17); 2159c9b1d098SAndrew Shewmaker return 0; 2160c9b1d098SAndrew Shewmaker } 2161c9b1d098SAndrew Shewmaker module_init(init_user_reserve) 21624eeab4f5SAndrew Shewmaker 21634eeab4f5SAndrew Shewmaker /* 21644eeab4f5SAndrew Shewmaker * Initialise sysctl_admin_reserve_kbytes. 21654eeab4f5SAndrew Shewmaker * 21664eeab4f5SAndrew Shewmaker * The purpose of sysctl_admin_reserve_kbytes is to allow the sys admin 21674eeab4f5SAndrew Shewmaker * to log in and kill a memory hogging process. 21684eeab4f5SAndrew Shewmaker * 21694eeab4f5SAndrew Shewmaker * Systems with more than 256MB will reserve 8MB, enough to recover 21704eeab4f5SAndrew Shewmaker * with sshd, bash, and top in OVERCOMMIT_GUESS. Smaller systems will 21714eeab4f5SAndrew Shewmaker * only reserve 3% of free pages by default. 21724eeab4f5SAndrew Shewmaker */ 21734eeab4f5SAndrew Shewmaker static int __meminit init_admin_reserve(void) 21744eeab4f5SAndrew Shewmaker { 21754eeab4f5SAndrew Shewmaker unsigned long free_kbytes; 21764eeab4f5SAndrew Shewmaker 21774eeab4f5SAndrew Shewmaker free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10); 21784eeab4f5SAndrew Shewmaker 21794eeab4f5SAndrew Shewmaker sysctl_admin_reserve_kbytes = min(free_kbytes / 32, 1UL << 13); 21804eeab4f5SAndrew Shewmaker return 0; 21814eeab4f5SAndrew Shewmaker } 21824eeab4f5SAndrew Shewmaker module_init(init_admin_reserve) 2183