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 16b1de0d13SMitchel Humpherys #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt 17b1de0d13SMitchel 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> 37b1de0d13SMitchel 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; 62944b6874SArnd Bergmann EXPORT_SYMBOL(high_memory); 631da177e4SLinus Torvalds struct page *mem_map; 641da177e4SLinus Torvalds unsigned long max_mapnr; 655b8bf307Sgchen gchen EXPORT_SYMBOL(max_mapnr); 664266c97aSHugh Dickins unsigned long highest_memmap_pfn; 6700a62ce9SKOSAKI Motohiro struct percpu_counter vm_committed_as; 681da177e4SLinus Torvalds int sysctl_overcommit_memory = OVERCOMMIT_GUESS; /* heuristic overcommit */ 691da177e4SLinus Torvalds int sysctl_overcommit_ratio = 50; /* default is 50% */ 7049f0ce5fSJerome Marchand unsigned long sysctl_overcommit_kbytes __read_mostly; 711da177e4SLinus Torvalds int sysctl_max_map_count = DEFAULT_MAX_MAP_COUNT; 72fc4d5c29SDavid Howells int sysctl_nr_trim_pages = CONFIG_NOMMU_INITIAL_TRIM_EXCESS; 73c9b1d098SAndrew Shewmaker unsigned long sysctl_user_reserve_kbytes __read_mostly = 1UL << 17; /* 128MB */ 744eeab4f5SAndrew Shewmaker unsigned long sysctl_admin_reserve_kbytes __read_mostly = 1UL << 13; /* 8MB */ 751da177e4SLinus Torvalds int heap_stack_gap = 0; 761da177e4SLinus Torvalds 7733e5d769SDavid Howells atomic_long_t mmap_pages_allocated; 788feae131SDavid Howells 79997071bcSK. Y. Srinivasan /* 80997071bcSK. Y. Srinivasan * The global memory commitment made in the system can be a metric 81997071bcSK. Y. Srinivasan * that can be used to drive ballooning decisions when Linux is hosted 82997071bcSK. Y. Srinivasan * as a guest. On Hyper-V, the host implements a policy engine for dynamically 83997071bcSK. Y. Srinivasan * balancing memory across competing virtual machines that are hosted. 84997071bcSK. Y. Srinivasan * Several metrics drive this policy engine including the guest reported 85997071bcSK. Y. Srinivasan * memory commitment. 86997071bcSK. Y. Srinivasan */ 87997071bcSK. Y. Srinivasan unsigned long vm_memory_committed(void) 88997071bcSK. Y. Srinivasan { 89997071bcSK. Y. Srinivasan return percpu_counter_read_positive(&vm_committed_as); 90997071bcSK. Y. Srinivasan } 91997071bcSK. Y. Srinivasan 92997071bcSK. Y. Srinivasan EXPORT_SYMBOL_GPL(vm_memory_committed); 93997071bcSK. Y. Srinivasan 941da177e4SLinus Torvalds EXPORT_SYMBOL(mem_map); 951da177e4SLinus Torvalds 968feae131SDavid Howells /* list of mapped, potentially shareable regions */ 978feae131SDavid Howells static struct kmem_cache *vm_region_jar; 988feae131SDavid Howells struct rb_root nommu_region_tree = RB_ROOT; 998feae131SDavid Howells DECLARE_RWSEM(nommu_region_sem); 1001da177e4SLinus Torvalds 101f0f37e2fSAlexey Dobriyan const struct vm_operations_struct generic_file_vm_ops = { 1021da177e4SLinus Torvalds }; 1031da177e4SLinus Torvalds 1041da177e4SLinus Torvalds /* 1051da177e4SLinus Torvalds * Return the total memory allocated for this pointer, not 1061da177e4SLinus Torvalds * just what the caller asked for. 1071da177e4SLinus Torvalds * 1081da177e4SLinus Torvalds * Doesn't have to be accurate, i.e. may have races. 1091da177e4SLinus Torvalds */ 1101da177e4SLinus Torvalds unsigned int kobjsize(const void *objp) 1111da177e4SLinus Torvalds { 1121da177e4SLinus Torvalds struct page *page; 1131da177e4SLinus Torvalds 1144016a139SMichael Hennerich /* 1154016a139SMichael Hennerich * If the object we have should not have ksize performed on it, 1164016a139SMichael Hennerich * return size of 0 1174016a139SMichael Hennerich */ 1185a1603beSPaul Mundt if (!objp || !virt_addr_valid(objp)) 1196cfd53fcSPaul Mundt return 0; 1206cfd53fcSPaul Mundt 1216cfd53fcSPaul Mundt page = virt_to_head_page(objp); 1226cfd53fcSPaul Mundt 1236cfd53fcSPaul Mundt /* 1246cfd53fcSPaul Mundt * If the allocator sets PageSlab, we know the pointer came from 1256cfd53fcSPaul Mundt * kmalloc(). 1266cfd53fcSPaul Mundt */ 1271da177e4SLinus Torvalds if (PageSlab(page)) 1281da177e4SLinus Torvalds return ksize(objp); 1291da177e4SLinus Torvalds 1306cfd53fcSPaul Mundt /* 131ab2e83eaSPaul Mundt * If it's not a compound page, see if we have a matching VMA 132ab2e83eaSPaul Mundt * region. This test is intentionally done in reverse order, 133ab2e83eaSPaul Mundt * so if there's no VMA, we still fall through and hand back 134ab2e83eaSPaul Mundt * PAGE_SIZE for 0-order pages. 135ab2e83eaSPaul Mundt */ 136ab2e83eaSPaul Mundt if (!PageCompound(page)) { 137ab2e83eaSPaul Mundt struct vm_area_struct *vma; 138ab2e83eaSPaul Mundt 139ab2e83eaSPaul Mundt vma = find_vma(current->mm, (unsigned long)objp); 140ab2e83eaSPaul Mundt if (vma) 141ab2e83eaSPaul Mundt return vma->vm_end - vma->vm_start; 142ab2e83eaSPaul Mundt } 143ab2e83eaSPaul Mundt 144ab2e83eaSPaul Mundt /* 1456cfd53fcSPaul Mundt * The ksize() function is only guaranteed to work for pointers 1465a1603beSPaul Mundt * returned by kmalloc(). So handle arbitrary pointers here. 1476cfd53fcSPaul Mundt */ 1485a1603beSPaul Mundt return PAGE_SIZE << compound_order(page); 1491da177e4SLinus Torvalds } 1501da177e4SLinus Torvalds 15128a35716SMichel Lespinasse long __get_user_pages(struct task_struct *tsk, struct mm_struct *mm, 15228a35716SMichel Lespinasse unsigned long start, unsigned long nr_pages, 15328a35716SMichel Lespinasse unsigned int foll_flags, struct page **pages, 15428a35716SMichel Lespinasse struct vm_area_struct **vmas, int *nonblocking) 1551da177e4SLinus Torvalds { 156910e46daSSonic Zhang struct vm_area_struct *vma; 1577b4d5b8bSDavid Howells unsigned long vm_flags; 1587b4d5b8bSDavid Howells int i; 1597b4d5b8bSDavid Howells 1607b4d5b8bSDavid Howells /* calculate required read or write permissions. 16158fa879eSHugh Dickins * If FOLL_FORCE is set, we only require the "MAY" flags. 1627b4d5b8bSDavid Howells */ 16358fa879eSHugh Dickins vm_flags = (foll_flags & FOLL_WRITE) ? 16458fa879eSHugh Dickins (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD); 16558fa879eSHugh Dickins vm_flags &= (foll_flags & FOLL_FORCE) ? 16658fa879eSHugh Dickins (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE); 1671da177e4SLinus Torvalds 1689d73777eSPeter Zijlstra for (i = 0; i < nr_pages; i++) { 1697561e8caSDavid Howells vma = find_vma(mm, start); 170910e46daSSonic Zhang if (!vma) 1717b4d5b8bSDavid Howells goto finish_or_fault; 1727b4d5b8bSDavid Howells 1737b4d5b8bSDavid Howells /* protect what we can, including chardevs */ 1741c3aff1cSHugh Dickins if ((vma->vm_flags & (VM_IO | VM_PFNMAP)) || 1751c3aff1cSHugh Dickins !(vm_flags & vma->vm_flags)) 1767b4d5b8bSDavid Howells goto finish_or_fault; 177910e46daSSonic Zhang 1781da177e4SLinus Torvalds if (pages) { 1791da177e4SLinus Torvalds pages[i] = virt_to_page(start); 1801da177e4SLinus Torvalds if (pages[i]) 1811da177e4SLinus Torvalds page_cache_get(pages[i]); 1821da177e4SLinus Torvalds } 1831da177e4SLinus Torvalds if (vmas) 184910e46daSSonic Zhang vmas[i] = vma; 185e1ee65d8SDavid Howells start = (start + PAGE_SIZE) & PAGE_MASK; 1861da177e4SLinus Torvalds } 1877b4d5b8bSDavid Howells 1887b4d5b8bSDavid Howells return i; 1897b4d5b8bSDavid Howells 1907b4d5b8bSDavid Howells finish_or_fault: 1917b4d5b8bSDavid Howells return i ? : -EFAULT; 1921da177e4SLinus Torvalds } 193b291f000SNick Piggin 194b291f000SNick Piggin /* 195b291f000SNick Piggin * get a list of pages in an address range belonging to the specified process 196b291f000SNick Piggin * and indicate the VMA that covers each page 197b291f000SNick Piggin * - this is potentially dodgy as we may end incrementing the page count of a 198b291f000SNick Piggin * slab page or a secondary page from a compound page 199b291f000SNick Piggin * - don't permit access to VMAs that don't support it, such as I/O mappings 200b291f000SNick Piggin */ 20128a35716SMichel Lespinasse long get_user_pages(struct task_struct *tsk, struct mm_struct *mm, 20228a35716SMichel Lespinasse unsigned long start, unsigned long nr_pages, 20328a35716SMichel Lespinasse int write, int force, struct page **pages, 20428a35716SMichel Lespinasse struct vm_area_struct **vmas) 205b291f000SNick Piggin { 206b291f000SNick Piggin int flags = 0; 207b291f000SNick Piggin 208b291f000SNick Piggin if (write) 20958fa879eSHugh Dickins flags |= FOLL_WRITE; 210b291f000SNick Piggin if (force) 21158fa879eSHugh Dickins flags |= FOLL_FORCE; 212b291f000SNick Piggin 21353a7706dSMichel Lespinasse return __get_user_pages(tsk, mm, start, nr_pages, flags, pages, vmas, 21453a7706dSMichel Lespinasse NULL); 215b291f000SNick Piggin } 21666aa2b4bSGreg Ungerer EXPORT_SYMBOL(get_user_pages); 21766aa2b4bSGreg Ungerer 218f0818f47SAndrea Arcangeli long get_user_pages_locked(struct task_struct *tsk, struct mm_struct *mm, 219f0818f47SAndrea Arcangeli unsigned long start, unsigned long nr_pages, 220f0818f47SAndrea Arcangeli int write, int force, struct page **pages, 221f0818f47SAndrea Arcangeli int *locked) 222f0818f47SAndrea Arcangeli { 223f0818f47SAndrea Arcangeli return get_user_pages(tsk, mm, start, nr_pages, write, force, 224f0818f47SAndrea Arcangeli pages, NULL); 225f0818f47SAndrea Arcangeli } 226f0818f47SAndrea Arcangeli EXPORT_SYMBOL(get_user_pages_locked); 227f0818f47SAndrea Arcangeli 2280fd71a56SAndrea Arcangeli long __get_user_pages_unlocked(struct task_struct *tsk, struct mm_struct *mm, 229f0818f47SAndrea Arcangeli unsigned long start, unsigned long nr_pages, 2300fd71a56SAndrea Arcangeli int write, int force, struct page **pages, 2310fd71a56SAndrea Arcangeli unsigned int gup_flags) 232f0818f47SAndrea Arcangeli { 233f0818f47SAndrea Arcangeli long ret; 234f0818f47SAndrea Arcangeli down_read(&mm->mmap_sem); 235f0818f47SAndrea Arcangeli ret = get_user_pages(tsk, mm, start, nr_pages, write, force, 236f0818f47SAndrea Arcangeli pages, NULL); 237f0818f47SAndrea Arcangeli up_read(&mm->mmap_sem); 238f0818f47SAndrea Arcangeli return ret; 239f0818f47SAndrea Arcangeli } 2400fd71a56SAndrea Arcangeli EXPORT_SYMBOL(__get_user_pages_unlocked); 2410fd71a56SAndrea Arcangeli 2420fd71a56SAndrea Arcangeli long get_user_pages_unlocked(struct task_struct *tsk, struct mm_struct *mm, 2430fd71a56SAndrea Arcangeli unsigned long start, unsigned long nr_pages, 2440fd71a56SAndrea Arcangeli int write, int force, struct page **pages) 2450fd71a56SAndrea Arcangeli { 2460fd71a56SAndrea Arcangeli return __get_user_pages_unlocked(tsk, mm, start, nr_pages, write, 2470fd71a56SAndrea Arcangeli force, pages, 0); 2480fd71a56SAndrea Arcangeli } 249f0818f47SAndrea Arcangeli EXPORT_SYMBOL(get_user_pages_unlocked); 250f0818f47SAndrea Arcangeli 251dfc2f91aSPaul Mundt /** 252dfc2f91aSPaul Mundt * follow_pfn - look up PFN at a user virtual address 253dfc2f91aSPaul Mundt * @vma: memory mapping 254dfc2f91aSPaul Mundt * @address: user virtual address 255dfc2f91aSPaul Mundt * @pfn: location to store found PFN 256dfc2f91aSPaul Mundt * 257dfc2f91aSPaul Mundt * Only IO mappings and raw PFN mappings are allowed. 258dfc2f91aSPaul Mundt * 259dfc2f91aSPaul Mundt * Returns zero and the pfn at @pfn on success, -ve otherwise. 260dfc2f91aSPaul Mundt */ 261dfc2f91aSPaul Mundt int follow_pfn(struct vm_area_struct *vma, unsigned long address, 262dfc2f91aSPaul Mundt unsigned long *pfn) 263dfc2f91aSPaul Mundt { 264dfc2f91aSPaul Mundt if (!(vma->vm_flags & (VM_IO | VM_PFNMAP))) 265dfc2f91aSPaul Mundt return -EINVAL; 266dfc2f91aSPaul Mundt 267dfc2f91aSPaul Mundt *pfn = address >> PAGE_SHIFT; 268dfc2f91aSPaul Mundt return 0; 269dfc2f91aSPaul Mundt } 270dfc2f91aSPaul Mundt EXPORT_SYMBOL(follow_pfn); 271dfc2f91aSPaul Mundt 272f1c4069eSJoonsoo Kim LIST_HEAD(vmap_area_list); 2731da177e4SLinus Torvalds 274b3bdda02SChristoph Lameter void vfree(const void *addr) 2751da177e4SLinus Torvalds { 2761da177e4SLinus Torvalds kfree(addr); 2771da177e4SLinus Torvalds } 278b5073173SPaul Mundt EXPORT_SYMBOL(vfree); 2791da177e4SLinus Torvalds 280dd0fc66fSAl Viro void *__vmalloc(unsigned long size, gfp_t gfp_mask, pgprot_t prot) 2811da177e4SLinus Torvalds { 2821da177e4SLinus Torvalds /* 2838518609dSRobert P. J. Day * You can't specify __GFP_HIGHMEM with kmalloc() since kmalloc() 2848518609dSRobert P. J. Day * returns only a logical address. 2851da177e4SLinus Torvalds */ 28684097518SNick Piggin return kmalloc(size, (gfp_mask | __GFP_COMP) & ~__GFP_HIGHMEM); 2871da177e4SLinus Torvalds } 288b5073173SPaul Mundt EXPORT_SYMBOL(__vmalloc); 2891da177e4SLinus Torvalds 290f905bc44SPaul Mundt void *vmalloc_user(unsigned long size) 291f905bc44SPaul Mundt { 292f905bc44SPaul Mundt void *ret; 293f905bc44SPaul Mundt 294f905bc44SPaul Mundt ret = __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO, 295f905bc44SPaul Mundt PAGE_KERNEL); 296f905bc44SPaul Mundt if (ret) { 297f905bc44SPaul Mundt struct vm_area_struct *vma; 298f905bc44SPaul Mundt 299f905bc44SPaul Mundt down_write(¤t->mm->mmap_sem); 300f905bc44SPaul Mundt vma = find_vma(current->mm, (unsigned long)ret); 301f905bc44SPaul Mundt if (vma) 302f905bc44SPaul Mundt vma->vm_flags |= VM_USERMAP; 303f905bc44SPaul Mundt up_write(¤t->mm->mmap_sem); 304f905bc44SPaul Mundt } 305f905bc44SPaul Mundt 306f905bc44SPaul Mundt return ret; 307f905bc44SPaul Mundt } 308f905bc44SPaul Mundt EXPORT_SYMBOL(vmalloc_user); 309f905bc44SPaul Mundt 310b3bdda02SChristoph Lameter struct page *vmalloc_to_page(const void *addr) 3111da177e4SLinus Torvalds { 3121da177e4SLinus Torvalds return virt_to_page(addr); 3131da177e4SLinus Torvalds } 314b5073173SPaul Mundt EXPORT_SYMBOL(vmalloc_to_page); 3151da177e4SLinus Torvalds 316b3bdda02SChristoph Lameter unsigned long vmalloc_to_pfn(const void *addr) 3171da177e4SLinus Torvalds { 3181da177e4SLinus Torvalds return page_to_pfn(virt_to_page(addr)); 3191da177e4SLinus Torvalds } 320b5073173SPaul Mundt EXPORT_SYMBOL(vmalloc_to_pfn); 3211da177e4SLinus Torvalds 3221da177e4SLinus Torvalds long vread(char *buf, char *addr, unsigned long count) 3231da177e4SLinus Torvalds { 3249bde916bSChen Gang /* Don't allow overflow */ 3259bde916bSChen Gang if ((unsigned long) buf + count < count) 3269bde916bSChen Gang count = -(unsigned long) buf; 3279bde916bSChen Gang 3281da177e4SLinus Torvalds memcpy(buf, addr, count); 3291da177e4SLinus Torvalds return count; 3301da177e4SLinus Torvalds } 3311da177e4SLinus Torvalds 3321da177e4SLinus Torvalds long vwrite(char *buf, char *addr, unsigned long count) 3331da177e4SLinus Torvalds { 3341da177e4SLinus Torvalds /* Don't allow overflow */ 3351da177e4SLinus Torvalds if ((unsigned long) addr + count < count) 3361da177e4SLinus Torvalds count = -(unsigned long) addr; 3371da177e4SLinus Torvalds 3381da177e4SLinus Torvalds memcpy(addr, buf, count); 339ac714904SChoi Gi-yong return count; 3401da177e4SLinus Torvalds } 3411da177e4SLinus Torvalds 3421da177e4SLinus Torvalds /* 3431da177e4SLinus Torvalds * vmalloc - allocate virtually continguos memory 3441da177e4SLinus Torvalds * 3451da177e4SLinus Torvalds * @size: allocation size 3461da177e4SLinus Torvalds * 3471da177e4SLinus Torvalds * Allocate enough pages to cover @size from the page level 3481da177e4SLinus Torvalds * allocator and map them into continguos kernel virtual space. 3491da177e4SLinus Torvalds * 350c1c8897fSMichael Opdenacker * For tight control over page level allocator and protection flags 3511da177e4SLinus Torvalds * use __vmalloc() instead. 3521da177e4SLinus Torvalds */ 3531da177e4SLinus Torvalds void *vmalloc(unsigned long size) 3541da177e4SLinus Torvalds { 3551da177e4SLinus Torvalds return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL); 3561da177e4SLinus Torvalds } 357f6138882SAndrew Morton EXPORT_SYMBOL(vmalloc); 358f6138882SAndrew Morton 359e1ca7788SDave Young /* 360e1ca7788SDave Young * vzalloc - allocate virtually continguos memory with zero fill 361e1ca7788SDave Young * 362e1ca7788SDave Young * @size: allocation size 363e1ca7788SDave Young * 364e1ca7788SDave Young * Allocate enough pages to cover @size from the page level 365e1ca7788SDave Young * allocator and map them into continguos kernel virtual space. 366e1ca7788SDave Young * The memory allocated is set to zero. 367e1ca7788SDave Young * 368e1ca7788SDave Young * For tight control over page level allocator and protection flags 369e1ca7788SDave Young * use __vmalloc() instead. 370e1ca7788SDave Young */ 371e1ca7788SDave Young void *vzalloc(unsigned long size) 372e1ca7788SDave Young { 373e1ca7788SDave Young return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO, 374e1ca7788SDave Young PAGE_KERNEL); 375e1ca7788SDave Young } 376e1ca7788SDave Young EXPORT_SYMBOL(vzalloc); 377e1ca7788SDave Young 378e1ca7788SDave Young /** 379e1ca7788SDave Young * vmalloc_node - allocate memory on a specific node 380e1ca7788SDave Young * @size: allocation size 381e1ca7788SDave Young * @node: numa node 382e1ca7788SDave Young * 383e1ca7788SDave Young * Allocate enough pages to cover @size from the page level 384e1ca7788SDave Young * allocator and map them into contiguous kernel virtual space. 385e1ca7788SDave Young * 386e1ca7788SDave Young * For tight control over page level allocator and protection flags 387e1ca7788SDave Young * use __vmalloc() instead. 388e1ca7788SDave Young */ 389f6138882SAndrew Morton void *vmalloc_node(unsigned long size, int node) 390f6138882SAndrew Morton { 391f6138882SAndrew Morton return vmalloc(size); 392f6138882SAndrew Morton } 3939a14f653SPaul Mundt EXPORT_SYMBOL(vmalloc_node); 394e1ca7788SDave Young 395e1ca7788SDave Young /** 396e1ca7788SDave Young * vzalloc_node - allocate memory on a specific node with zero fill 397e1ca7788SDave Young * @size: allocation size 398e1ca7788SDave Young * @node: numa node 399e1ca7788SDave Young * 400e1ca7788SDave Young * Allocate enough pages to cover @size from the page level 401e1ca7788SDave Young * allocator and map them into contiguous kernel virtual space. 402e1ca7788SDave Young * The memory allocated is set to zero. 403e1ca7788SDave Young * 404e1ca7788SDave Young * For tight control over page level allocator and protection flags 405e1ca7788SDave Young * use __vmalloc() instead. 406e1ca7788SDave Young */ 407e1ca7788SDave Young void *vzalloc_node(unsigned long size, int node) 408e1ca7788SDave Young { 409e1ca7788SDave Young return vzalloc(size); 410e1ca7788SDave Young } 411e1ca7788SDave Young EXPORT_SYMBOL(vzalloc_node); 4121da177e4SLinus Torvalds 4131af446edSPaul Mundt #ifndef PAGE_KERNEL_EXEC 4141af446edSPaul Mundt # define PAGE_KERNEL_EXEC PAGE_KERNEL 4151af446edSPaul Mundt #endif 4161af446edSPaul Mundt 4171af446edSPaul Mundt /** 4181af446edSPaul Mundt * vmalloc_exec - allocate virtually contiguous, executable memory 4191af446edSPaul Mundt * @size: allocation size 4201af446edSPaul Mundt * 4211af446edSPaul Mundt * Kernel-internal function to allocate enough pages to cover @size 4221af446edSPaul Mundt * the page level allocator and map them into contiguous and 4231af446edSPaul Mundt * executable kernel virtual space. 4241af446edSPaul Mundt * 4251af446edSPaul Mundt * For tight control over page level allocator and protection flags 4261af446edSPaul Mundt * use __vmalloc() instead. 4271af446edSPaul Mundt */ 4281af446edSPaul Mundt 4291af446edSPaul Mundt void *vmalloc_exec(unsigned long size) 4301af446edSPaul Mundt { 4311af446edSPaul Mundt return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC); 4321af446edSPaul Mundt } 4331af446edSPaul Mundt 434b5073173SPaul Mundt /** 435b5073173SPaul Mundt * vmalloc_32 - allocate virtually contiguous memory (32bit addressable) 4361da177e4SLinus Torvalds * @size: allocation size 4371da177e4SLinus Torvalds * 4381da177e4SLinus Torvalds * Allocate enough 32bit PA addressable pages to cover @size from the 4391da177e4SLinus Torvalds * page level allocator and map them into continguos kernel virtual space. 4401da177e4SLinus Torvalds */ 4411da177e4SLinus Torvalds void *vmalloc_32(unsigned long size) 4421da177e4SLinus Torvalds { 4431da177e4SLinus Torvalds return __vmalloc(size, GFP_KERNEL, PAGE_KERNEL); 4441da177e4SLinus Torvalds } 445b5073173SPaul Mundt EXPORT_SYMBOL(vmalloc_32); 446b5073173SPaul Mundt 447b5073173SPaul Mundt /** 448b5073173SPaul Mundt * vmalloc_32_user - allocate zeroed virtually contiguous 32bit memory 449b5073173SPaul Mundt * @size: allocation size 450b5073173SPaul Mundt * 451b5073173SPaul Mundt * The resulting memory area is 32bit addressable and zeroed so it can be 452b5073173SPaul Mundt * mapped to userspace without leaking data. 453f905bc44SPaul Mundt * 454f905bc44SPaul Mundt * VM_USERMAP is set on the corresponding VMA so that subsequent calls to 455f905bc44SPaul Mundt * remap_vmalloc_range() are permissible. 456b5073173SPaul Mundt */ 457b5073173SPaul Mundt void *vmalloc_32_user(unsigned long size) 458b5073173SPaul Mundt { 459f905bc44SPaul Mundt /* 460f905bc44SPaul Mundt * We'll have to sort out the ZONE_DMA bits for 64-bit, 461f905bc44SPaul Mundt * but for now this can simply use vmalloc_user() directly. 462f905bc44SPaul Mundt */ 463f905bc44SPaul Mundt return vmalloc_user(size); 464b5073173SPaul Mundt } 465b5073173SPaul Mundt EXPORT_SYMBOL(vmalloc_32_user); 4661da177e4SLinus Torvalds 4671da177e4SLinus Torvalds void *vmap(struct page **pages, unsigned int count, unsigned long flags, pgprot_t prot) 4681da177e4SLinus Torvalds { 4691da177e4SLinus Torvalds BUG(); 4701da177e4SLinus Torvalds return NULL; 4711da177e4SLinus Torvalds } 472b5073173SPaul Mundt EXPORT_SYMBOL(vmap); 4731da177e4SLinus Torvalds 474b3bdda02SChristoph Lameter void vunmap(const void *addr) 4751da177e4SLinus Torvalds { 4761da177e4SLinus Torvalds BUG(); 4771da177e4SLinus Torvalds } 478b5073173SPaul Mundt EXPORT_SYMBOL(vunmap); 4791da177e4SLinus Torvalds 480eb6434d9SPaul Mundt void *vm_map_ram(struct page **pages, unsigned int count, int node, pgprot_t prot) 481eb6434d9SPaul Mundt { 482eb6434d9SPaul Mundt BUG(); 483eb6434d9SPaul Mundt return NULL; 484eb6434d9SPaul Mundt } 485eb6434d9SPaul Mundt EXPORT_SYMBOL(vm_map_ram); 486eb6434d9SPaul Mundt 487eb6434d9SPaul Mundt void vm_unmap_ram(const void *mem, unsigned int count) 488eb6434d9SPaul Mundt { 489eb6434d9SPaul Mundt BUG(); 490eb6434d9SPaul Mundt } 491eb6434d9SPaul Mundt EXPORT_SYMBOL(vm_unmap_ram); 492eb6434d9SPaul Mundt 493eb6434d9SPaul Mundt void vm_unmap_aliases(void) 494eb6434d9SPaul Mundt { 495eb6434d9SPaul Mundt } 496eb6434d9SPaul Mundt EXPORT_SYMBOL_GPL(vm_unmap_aliases); 497eb6434d9SPaul Mundt 4981da177e4SLinus Torvalds /* 4991eeb66a1SChristoph Hellwig * Implement a stub for vmalloc_sync_all() if the architecture chose not to 5001eeb66a1SChristoph Hellwig * have one. 5011eeb66a1SChristoph Hellwig */ 5023b32123dSGideon Israel Dsouza void __weak vmalloc_sync_all(void) 5031eeb66a1SChristoph Hellwig { 5041eeb66a1SChristoph Hellwig } 5051eeb66a1SChristoph Hellwig 50629c185e5SPaul Mundt /** 50729c185e5SPaul Mundt * alloc_vm_area - allocate a range of kernel address space 50829c185e5SPaul Mundt * @size: size of the area 50929c185e5SPaul Mundt * 51029c185e5SPaul Mundt * Returns: NULL on failure, vm_struct on success 51129c185e5SPaul Mundt * 51229c185e5SPaul Mundt * This function reserves a range of kernel address space, and 51329c185e5SPaul Mundt * allocates pagetables to map that range. No actual mappings 51429c185e5SPaul Mundt * are created. If the kernel address space is not shared 51529c185e5SPaul Mundt * between processes, it syncs the pagetable across all 51629c185e5SPaul Mundt * processes. 51729c185e5SPaul Mundt */ 518cd12909cSDavid Vrabel struct vm_struct *alloc_vm_area(size_t size, pte_t **ptes) 51929c185e5SPaul Mundt { 52029c185e5SPaul Mundt BUG(); 52129c185e5SPaul Mundt return NULL; 52229c185e5SPaul Mundt } 52329c185e5SPaul Mundt EXPORT_SYMBOL_GPL(alloc_vm_area); 52429c185e5SPaul Mundt 52529c185e5SPaul Mundt void free_vm_area(struct vm_struct *area) 52629c185e5SPaul Mundt { 52729c185e5SPaul Mundt BUG(); 52829c185e5SPaul Mundt } 52929c185e5SPaul Mundt EXPORT_SYMBOL_GPL(free_vm_area); 53029c185e5SPaul Mundt 531b5073173SPaul Mundt int vm_insert_page(struct vm_area_struct *vma, unsigned long addr, 532b5073173SPaul Mundt struct page *page) 533b5073173SPaul Mundt { 534b5073173SPaul Mundt return -EINVAL; 535b5073173SPaul Mundt } 536b5073173SPaul Mundt EXPORT_SYMBOL(vm_insert_page); 537b5073173SPaul Mundt 5381eeb66a1SChristoph Hellwig /* 5391da177e4SLinus Torvalds * sys_brk() for the most part doesn't need the global kernel 5401da177e4SLinus Torvalds * lock, except when an application is doing something nasty 5411da177e4SLinus Torvalds * like trying to un-brk an area that has already been mapped 5421da177e4SLinus Torvalds * to a regular file. in this case, the unmapping will need 5431da177e4SLinus Torvalds * to invoke file system routines that need the global lock. 5441da177e4SLinus Torvalds */ 5456a6160a7SHeiko Carstens SYSCALL_DEFINE1(brk, unsigned long, brk) 5461da177e4SLinus Torvalds { 5471da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 5481da177e4SLinus Torvalds 5491da177e4SLinus Torvalds if (brk < mm->start_brk || brk > mm->context.end_brk) 5501da177e4SLinus Torvalds return mm->brk; 5511da177e4SLinus Torvalds 5521da177e4SLinus Torvalds if (mm->brk == brk) 5531da177e4SLinus Torvalds return mm->brk; 5541da177e4SLinus Torvalds 5551da177e4SLinus Torvalds /* 5561da177e4SLinus Torvalds * Always allow shrinking brk 5571da177e4SLinus Torvalds */ 5581da177e4SLinus Torvalds if (brk <= mm->brk) { 5591da177e4SLinus Torvalds mm->brk = brk; 5601da177e4SLinus Torvalds return brk; 5611da177e4SLinus Torvalds } 5621da177e4SLinus Torvalds 5631da177e4SLinus Torvalds /* 5641da177e4SLinus Torvalds * Ok, looks good - let it rip. 5651da177e4SLinus Torvalds */ 566cfe79c00SMike Frysinger flush_icache_range(mm->brk, brk); 5671da177e4SLinus Torvalds return mm->brk = brk; 5681da177e4SLinus Torvalds } 5691da177e4SLinus Torvalds 5708feae131SDavid Howells /* 5718feae131SDavid Howells * initialise the VMA and region record slabs 5728feae131SDavid Howells */ 5738feae131SDavid Howells void __init mmap_init(void) 5741da177e4SLinus Torvalds { 57500a62ce9SKOSAKI Motohiro int ret; 57600a62ce9SKOSAKI Motohiro 577908c7f19STejun Heo ret = percpu_counter_init(&vm_committed_as, 0, GFP_KERNEL); 57800a62ce9SKOSAKI Motohiro VM_BUG_ON(ret); 57933e5d769SDavid Howells vm_region_jar = KMEM_CACHE(vm_region, SLAB_PANIC); 5808feae131SDavid Howells } 5811da177e4SLinus Torvalds 5828feae131SDavid Howells /* 5838feae131SDavid Howells * validate the region tree 5848feae131SDavid Howells * - the caller must hold the region lock 5858feae131SDavid Howells */ 5868feae131SDavid Howells #ifdef CONFIG_DEBUG_NOMMU_REGIONS 5878feae131SDavid Howells static noinline void validate_nommu_regions(void) 5888feae131SDavid Howells { 5898feae131SDavid Howells struct vm_region *region, *last; 5908feae131SDavid Howells struct rb_node *p, *lastp; 5911da177e4SLinus Torvalds 5928feae131SDavid Howells lastp = rb_first(&nommu_region_tree); 5938feae131SDavid Howells if (!lastp) 5948feae131SDavid Howells return; 5958feae131SDavid Howells 5968feae131SDavid Howells last = rb_entry(lastp, struct vm_region, vm_rb); 59733e5d769SDavid Howells BUG_ON(unlikely(last->vm_end <= last->vm_start)); 59833e5d769SDavid Howells BUG_ON(unlikely(last->vm_top < last->vm_end)); 5998feae131SDavid Howells 6008feae131SDavid Howells while ((p = rb_next(lastp))) { 6018feae131SDavid Howells region = rb_entry(p, struct vm_region, vm_rb); 6028feae131SDavid Howells last = rb_entry(lastp, struct vm_region, vm_rb); 6038feae131SDavid Howells 60433e5d769SDavid Howells BUG_ON(unlikely(region->vm_end <= region->vm_start)); 60533e5d769SDavid Howells BUG_ON(unlikely(region->vm_top < region->vm_end)); 60633e5d769SDavid Howells BUG_ON(unlikely(region->vm_start < last->vm_top)); 6078feae131SDavid Howells 6088feae131SDavid Howells lastp = p; 6091da177e4SLinus Torvalds } 6101da177e4SLinus Torvalds } 6118feae131SDavid Howells #else 61233e5d769SDavid Howells static void validate_nommu_regions(void) 61333e5d769SDavid Howells { 61433e5d769SDavid Howells } 6158feae131SDavid Howells #endif 6168feae131SDavid Howells 6178feae131SDavid Howells /* 6188feae131SDavid Howells * add a region into the global tree 6198feae131SDavid Howells */ 6208feae131SDavid Howells static void add_nommu_region(struct vm_region *region) 6218feae131SDavid Howells { 6228feae131SDavid Howells struct vm_region *pregion; 6238feae131SDavid Howells struct rb_node **p, *parent; 6248feae131SDavid Howells 6258feae131SDavid Howells validate_nommu_regions(); 6268feae131SDavid Howells 6278feae131SDavid Howells parent = NULL; 6288feae131SDavid Howells p = &nommu_region_tree.rb_node; 6298feae131SDavid Howells while (*p) { 6308feae131SDavid Howells parent = *p; 6318feae131SDavid Howells pregion = rb_entry(parent, struct vm_region, vm_rb); 6328feae131SDavid Howells if (region->vm_start < pregion->vm_start) 6338feae131SDavid Howells p = &(*p)->rb_left; 6348feae131SDavid Howells else if (region->vm_start > pregion->vm_start) 6358feae131SDavid Howells p = &(*p)->rb_right; 6368feae131SDavid Howells else if (pregion == region) 6378feae131SDavid Howells return; 6388feae131SDavid Howells else 6398feae131SDavid Howells BUG(); 6408feae131SDavid Howells } 6418feae131SDavid Howells 6428feae131SDavid Howells rb_link_node(®ion->vm_rb, parent, p); 6438feae131SDavid Howells rb_insert_color(®ion->vm_rb, &nommu_region_tree); 6448feae131SDavid Howells 6458feae131SDavid Howells validate_nommu_regions(); 6468feae131SDavid Howells } 6478feae131SDavid Howells 6488feae131SDavid Howells /* 6498feae131SDavid Howells * delete a region from the global tree 6508feae131SDavid Howells */ 6518feae131SDavid Howells static void delete_nommu_region(struct vm_region *region) 6528feae131SDavid Howells { 6538feae131SDavid Howells BUG_ON(!nommu_region_tree.rb_node); 6548feae131SDavid Howells 6558feae131SDavid Howells validate_nommu_regions(); 6568feae131SDavid Howells rb_erase(®ion->vm_rb, &nommu_region_tree); 6578feae131SDavid Howells validate_nommu_regions(); 6588feae131SDavid Howells } 6598feae131SDavid Howells 6608feae131SDavid Howells /* 6618feae131SDavid Howells * free a contiguous series of pages 6628feae131SDavid Howells */ 6638feae131SDavid Howells static void free_page_series(unsigned long from, unsigned long to) 6648feae131SDavid Howells { 6658feae131SDavid Howells for (; from < to; from += PAGE_SIZE) { 6668feae131SDavid Howells struct page *page = virt_to_page(from); 6678feae131SDavid Howells 6688feae131SDavid Howells kdebug("- free %lx", from); 66933e5d769SDavid Howells atomic_long_dec(&mmap_pages_allocated); 6708feae131SDavid Howells if (page_count(page) != 1) 67133e5d769SDavid Howells kdebug("free page %p: refcount not one: %d", 67233e5d769SDavid Howells page, page_count(page)); 6738feae131SDavid Howells put_page(page); 6748feae131SDavid Howells } 6758feae131SDavid Howells } 6768feae131SDavid Howells 6778feae131SDavid Howells /* 6788feae131SDavid Howells * release a reference to a region 67933e5d769SDavid Howells * - the caller must hold the region semaphore for writing, which this releases 680dd8632a1SPaul Mundt * - the region may not have been added to the tree yet, in which case vm_top 6818feae131SDavid Howells * will equal vm_start 6828feae131SDavid Howells */ 6838feae131SDavid Howells static void __put_nommu_region(struct vm_region *region) 6848feae131SDavid Howells __releases(nommu_region_sem) 6858feae131SDavid Howells { 6861e2ae599SDavid Howells kenter("%p{%d}", region, region->vm_usage); 6878feae131SDavid Howells 6888feae131SDavid Howells BUG_ON(!nommu_region_tree.rb_node); 6898feae131SDavid Howells 6901e2ae599SDavid Howells if (--region->vm_usage == 0) { 691dd8632a1SPaul Mundt if (region->vm_top > region->vm_start) 6928feae131SDavid Howells delete_nommu_region(region); 6938feae131SDavid Howells up_write(&nommu_region_sem); 6948feae131SDavid Howells 6958feae131SDavid Howells if (region->vm_file) 6968feae131SDavid Howells fput(region->vm_file); 6978feae131SDavid Howells 6988feae131SDavid Howells /* IO memory and memory shared directly out of the pagecache 6998feae131SDavid Howells * from ramfs/tmpfs mustn't be released here */ 7008feae131SDavid Howells if (region->vm_flags & VM_MAPPED_COPY) { 7018feae131SDavid Howells kdebug("free series"); 702dd8632a1SPaul Mundt free_page_series(region->vm_start, region->vm_top); 7038feae131SDavid Howells } 7048feae131SDavid Howells kmem_cache_free(vm_region_jar, region); 7058feae131SDavid Howells } else { 7068feae131SDavid Howells up_write(&nommu_region_sem); 7078feae131SDavid Howells } 7088feae131SDavid Howells } 7098feae131SDavid Howells 7108feae131SDavid Howells /* 7118feae131SDavid Howells * release a reference to a region 7128feae131SDavid Howells */ 7138feae131SDavid Howells static void put_nommu_region(struct vm_region *region) 7148feae131SDavid Howells { 7158feae131SDavid Howells down_write(&nommu_region_sem); 7168feae131SDavid Howells __put_nommu_region(region); 7178feae131SDavid Howells } 7181da177e4SLinus Torvalds 7193034097aSDavid Howells /* 720eb8cdec4SBernd Schmidt * update protection on a vma 721eb8cdec4SBernd Schmidt */ 722eb8cdec4SBernd Schmidt static void protect_vma(struct vm_area_struct *vma, unsigned long flags) 723eb8cdec4SBernd Schmidt { 724eb8cdec4SBernd Schmidt #ifdef CONFIG_MPU 725eb8cdec4SBernd Schmidt struct mm_struct *mm = vma->vm_mm; 726eb8cdec4SBernd Schmidt long start = vma->vm_start & PAGE_MASK; 727eb8cdec4SBernd Schmidt while (start < vma->vm_end) { 728eb8cdec4SBernd Schmidt protect_page(mm, start, flags); 729eb8cdec4SBernd Schmidt start += PAGE_SIZE; 730eb8cdec4SBernd Schmidt } 731eb8cdec4SBernd Schmidt update_protections(mm); 732eb8cdec4SBernd Schmidt #endif 733eb8cdec4SBernd Schmidt } 734eb8cdec4SBernd Schmidt 735eb8cdec4SBernd Schmidt /* 7363034097aSDavid Howells * add a VMA into a process's mm_struct in the appropriate place in the list 7378feae131SDavid Howells * and tree and add to the address space's page tree also if not an anonymous 7388feae131SDavid Howells * page 7393034097aSDavid Howells * - should be called with mm->mmap_sem held writelocked 7403034097aSDavid Howells */ 7418feae131SDavid Howells static void add_vma_to_mm(struct mm_struct *mm, struct vm_area_struct *vma) 7423034097aSDavid Howells { 7436038def0SNamhyung Kim struct vm_area_struct *pvma, *prev; 7448feae131SDavid Howells struct address_space *mapping; 7456038def0SNamhyung Kim struct rb_node **p, *parent, *rb_prev; 7463034097aSDavid Howells 7478feae131SDavid Howells kenter(",%p", vma); 7488feae131SDavid Howells 7498feae131SDavid Howells BUG_ON(!vma->vm_region); 7508feae131SDavid Howells 7518feae131SDavid Howells mm->map_count++; 7528feae131SDavid Howells vma->vm_mm = mm; 7538feae131SDavid Howells 754eb8cdec4SBernd Schmidt protect_vma(vma, vma->vm_flags); 755eb8cdec4SBernd Schmidt 7568feae131SDavid Howells /* add the VMA to the mapping */ 7578feae131SDavid Howells if (vma->vm_file) { 7588feae131SDavid Howells mapping = vma->vm_file->f_mapping; 7598feae131SDavid Howells 76083cde9e8SDavidlohr Bueso i_mmap_lock_write(mapping); 7618feae131SDavid Howells flush_dcache_mmap_lock(mapping); 7626b2dbba8SMichel Lespinasse vma_interval_tree_insert(vma, &mapping->i_mmap); 7638feae131SDavid Howells flush_dcache_mmap_unlock(mapping); 76483cde9e8SDavidlohr Bueso i_mmap_unlock_write(mapping); 7658feae131SDavid Howells } 7668feae131SDavid Howells 7678feae131SDavid Howells /* add the VMA to the tree */ 7686038def0SNamhyung Kim parent = rb_prev = NULL; 7698feae131SDavid Howells p = &mm->mm_rb.rb_node; 7708feae131SDavid Howells while (*p) { 7718feae131SDavid Howells parent = *p; 7728feae131SDavid Howells pvma = rb_entry(parent, struct vm_area_struct, vm_rb); 7738feae131SDavid Howells 7748feae131SDavid Howells /* sort by: start addr, end addr, VMA struct addr in that order 7758feae131SDavid Howells * (the latter is necessary as we may get identical VMAs) */ 7768feae131SDavid Howells if (vma->vm_start < pvma->vm_start) 7778feae131SDavid Howells p = &(*p)->rb_left; 7786038def0SNamhyung Kim else if (vma->vm_start > pvma->vm_start) { 7796038def0SNamhyung Kim rb_prev = parent; 7808feae131SDavid Howells p = &(*p)->rb_right; 7816038def0SNamhyung Kim } else if (vma->vm_end < pvma->vm_end) 7828feae131SDavid Howells p = &(*p)->rb_left; 7836038def0SNamhyung Kim else if (vma->vm_end > pvma->vm_end) { 7846038def0SNamhyung Kim rb_prev = parent; 7858feae131SDavid Howells p = &(*p)->rb_right; 7866038def0SNamhyung Kim } else if (vma < pvma) 7878feae131SDavid Howells p = &(*p)->rb_left; 7886038def0SNamhyung Kim else if (vma > pvma) { 7896038def0SNamhyung Kim rb_prev = parent; 7908feae131SDavid Howells p = &(*p)->rb_right; 7916038def0SNamhyung Kim } else 7928feae131SDavid Howells BUG(); 7938feae131SDavid Howells } 7948feae131SDavid Howells 7958feae131SDavid Howells rb_link_node(&vma->vm_rb, parent, p); 7968feae131SDavid Howells rb_insert_color(&vma->vm_rb, &mm->mm_rb); 7978feae131SDavid Howells 7988feae131SDavid Howells /* add VMA to the VMA list also */ 7996038def0SNamhyung Kim prev = NULL; 8006038def0SNamhyung Kim if (rb_prev) 8016038def0SNamhyung Kim prev = rb_entry(rb_prev, struct vm_area_struct, vm_rb); 8023034097aSDavid Howells 8036038def0SNamhyung Kim __vma_link_list(mm, vma, prev, parent); 8048feae131SDavid Howells } 8058feae131SDavid Howells 8068feae131SDavid Howells /* 8078feae131SDavid Howells * delete a VMA from its owning mm_struct and address space 8088feae131SDavid Howells */ 8098feae131SDavid Howells static void delete_vma_from_mm(struct vm_area_struct *vma) 8108feae131SDavid Howells { 811615d6e87SDavidlohr Bueso int i; 8128feae131SDavid Howells struct address_space *mapping; 8138feae131SDavid Howells struct mm_struct *mm = vma->vm_mm; 814615d6e87SDavidlohr Bueso struct task_struct *curr = current; 8158feae131SDavid Howells 8168feae131SDavid Howells kenter("%p", vma); 8178feae131SDavid Howells 818eb8cdec4SBernd Schmidt protect_vma(vma, 0); 819eb8cdec4SBernd Schmidt 8208feae131SDavid Howells mm->map_count--; 821615d6e87SDavidlohr Bueso for (i = 0; i < VMACACHE_SIZE; i++) { 822615d6e87SDavidlohr Bueso /* if the vma is cached, invalidate the entire cache */ 823615d6e87SDavidlohr Bueso if (curr->vmacache[i] == vma) { 824e020d5bdSSteven Miao vmacache_invalidate(mm); 825615d6e87SDavidlohr Bueso break; 826615d6e87SDavidlohr Bueso } 827615d6e87SDavidlohr Bueso } 8288feae131SDavid Howells 8298feae131SDavid Howells /* remove the VMA from the mapping */ 8308feae131SDavid Howells if (vma->vm_file) { 8318feae131SDavid Howells mapping = vma->vm_file->f_mapping; 8328feae131SDavid Howells 83383cde9e8SDavidlohr Bueso i_mmap_lock_write(mapping); 8348feae131SDavid Howells flush_dcache_mmap_lock(mapping); 8356b2dbba8SMichel Lespinasse vma_interval_tree_remove(vma, &mapping->i_mmap); 8368feae131SDavid Howells flush_dcache_mmap_unlock(mapping); 83783cde9e8SDavidlohr Bueso i_mmap_unlock_write(mapping); 8388feae131SDavid Howells } 8398feae131SDavid Howells 8408feae131SDavid Howells /* remove from the MM's tree and list */ 8418feae131SDavid Howells rb_erase(&vma->vm_rb, &mm->mm_rb); 842b951bf2cSNamhyung Kim 843b951bf2cSNamhyung Kim if (vma->vm_prev) 844b951bf2cSNamhyung Kim vma->vm_prev->vm_next = vma->vm_next; 845b951bf2cSNamhyung Kim else 846b951bf2cSNamhyung Kim mm->mmap = vma->vm_next; 847b951bf2cSNamhyung Kim 848b951bf2cSNamhyung Kim if (vma->vm_next) 849b951bf2cSNamhyung Kim vma->vm_next->vm_prev = vma->vm_prev; 8508feae131SDavid Howells } 8518feae131SDavid Howells 8528feae131SDavid Howells /* 8538feae131SDavid Howells * destroy a VMA record 8548feae131SDavid Howells */ 8558feae131SDavid Howells static void delete_vma(struct mm_struct *mm, struct vm_area_struct *vma) 8568feae131SDavid Howells { 8578feae131SDavid Howells kenter("%p", vma); 8588feae131SDavid Howells if (vma->vm_ops && vma->vm_ops->close) 8598feae131SDavid Howells vma->vm_ops->close(vma); 860e9714acfSKonstantin Khlebnikov if (vma->vm_file) 8618feae131SDavid Howells fput(vma->vm_file); 8628feae131SDavid Howells put_nommu_region(vma->vm_region); 8638feae131SDavid Howells kmem_cache_free(vm_area_cachep, vma); 8643034097aSDavid Howells } 8653034097aSDavid Howells 8663034097aSDavid Howells /* 8673034097aSDavid Howells * look up the first VMA in which addr resides, NULL if none 8683034097aSDavid Howells * - should be called with mm->mmap_sem at least held readlocked 8693034097aSDavid Howells */ 8703034097aSDavid Howells struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr) 8713034097aSDavid Howells { 8728feae131SDavid Howells struct vm_area_struct *vma; 8733034097aSDavid Howells 8748feae131SDavid Howells /* check the cache first */ 875615d6e87SDavidlohr Bueso vma = vmacache_find(mm, addr); 876615d6e87SDavidlohr Bueso if (likely(vma)) 8778feae131SDavid Howells return vma; 8788feae131SDavid Howells 879e922c4c5SNamhyung Kim /* trawl the list (there may be multiple mappings in which addr 8808feae131SDavid Howells * resides) */ 881e922c4c5SNamhyung Kim for (vma = mm->mmap; vma; vma = vma->vm_next) { 8828feae131SDavid Howells if (vma->vm_start > addr) 8838feae131SDavid Howells return NULL; 8848feae131SDavid Howells if (vma->vm_end > addr) { 885615d6e87SDavidlohr Bueso vmacache_update(addr, vma); 8868feae131SDavid Howells return vma; 8873034097aSDavid Howells } 8888feae131SDavid Howells } 8893034097aSDavid Howells 8903034097aSDavid Howells return NULL; 8913034097aSDavid Howells } 8923034097aSDavid Howells EXPORT_SYMBOL(find_vma); 8933034097aSDavid Howells 8943034097aSDavid Howells /* 895930e652aSDavid Howells * find a VMA 896930e652aSDavid Howells * - we don't extend stack VMAs under NOMMU conditions 897930e652aSDavid Howells */ 898930e652aSDavid Howells struct vm_area_struct *find_extend_vma(struct mm_struct *mm, unsigned long addr) 899930e652aSDavid Howells { 9007561e8caSDavid Howells return find_vma(mm, addr); 901930e652aSDavid Howells } 902930e652aSDavid Howells 9038feae131SDavid Howells /* 9048feae131SDavid Howells * expand a stack to a given address 9058feae131SDavid Howells * - not supported under NOMMU conditions 9068feae131SDavid Howells */ 90757c8f63eSGreg Ungerer int expand_stack(struct vm_area_struct *vma, unsigned long address) 90857c8f63eSGreg Ungerer { 90957c8f63eSGreg Ungerer return -ENOMEM; 91057c8f63eSGreg Ungerer } 91157c8f63eSGreg Ungerer 912930e652aSDavid Howells /* 9136fa5f80bSDavid Howells * look up the first VMA exactly that exactly matches addr 9146fa5f80bSDavid Howells * - should be called with mm->mmap_sem at least held readlocked 9156fa5f80bSDavid Howells */ 9168feae131SDavid Howells static struct vm_area_struct *find_vma_exact(struct mm_struct *mm, 9178feae131SDavid Howells unsigned long addr, 9188feae131SDavid Howells unsigned long len) 9191da177e4SLinus Torvalds { 9201da177e4SLinus Torvalds struct vm_area_struct *vma; 9218feae131SDavid Howells unsigned long end = addr + len; 9221da177e4SLinus Torvalds 9238feae131SDavid Howells /* check the cache first */ 924615d6e87SDavidlohr Bueso vma = vmacache_find_exact(mm, addr, end); 925615d6e87SDavidlohr Bueso if (vma) 9261da177e4SLinus Torvalds return vma; 9278feae131SDavid Howells 928e922c4c5SNamhyung Kim /* trawl the list (there may be multiple mappings in which addr 9298feae131SDavid Howells * resides) */ 930e922c4c5SNamhyung Kim for (vma = mm->mmap; vma; vma = vma->vm_next) { 9318feae131SDavid Howells if (vma->vm_start < addr) 9328feae131SDavid Howells continue; 9338feae131SDavid Howells if (vma->vm_start > addr) 9348feae131SDavid Howells return NULL; 9358feae131SDavid Howells if (vma->vm_end == end) { 936615d6e87SDavidlohr Bueso vmacache_update(addr, vma); 9378feae131SDavid Howells return vma; 9388feae131SDavid Howells } 9391da177e4SLinus Torvalds } 9401da177e4SLinus Torvalds 9411da177e4SLinus Torvalds return NULL; 9421da177e4SLinus Torvalds } 9431da177e4SLinus Torvalds 9443034097aSDavid Howells /* 9451da177e4SLinus Torvalds * determine whether a mapping should be permitted and, if so, what sort of 9461da177e4SLinus Torvalds * mapping we're capable of supporting 9471da177e4SLinus Torvalds */ 9481da177e4SLinus Torvalds static int validate_mmap_request(struct file *file, 9491da177e4SLinus Torvalds unsigned long addr, 9501da177e4SLinus Torvalds unsigned long len, 9511da177e4SLinus Torvalds unsigned long prot, 9521da177e4SLinus Torvalds unsigned long flags, 9531da177e4SLinus Torvalds unsigned long pgoff, 9541da177e4SLinus Torvalds unsigned long *_capabilities) 9551da177e4SLinus Torvalds { 9568feae131SDavid Howells unsigned long capabilities, rlen; 9571da177e4SLinus Torvalds int ret; 9581da177e4SLinus Torvalds 9591da177e4SLinus Torvalds /* do the simple checks first */ 96006aab5a3SDavid Howells if (flags & MAP_FIXED) { 9611da177e4SLinus Torvalds printk(KERN_DEBUG 9621da177e4SLinus Torvalds "%d: Can't do fixed-address/overlay mmap of RAM\n", 9631da177e4SLinus Torvalds current->pid); 9641da177e4SLinus Torvalds return -EINVAL; 9651da177e4SLinus Torvalds } 9661da177e4SLinus Torvalds 9671da177e4SLinus Torvalds if ((flags & MAP_TYPE) != MAP_PRIVATE && 9681da177e4SLinus Torvalds (flags & MAP_TYPE) != MAP_SHARED) 9691da177e4SLinus Torvalds return -EINVAL; 9701da177e4SLinus Torvalds 971f81cff0dSMike Frysinger if (!len) 9721da177e4SLinus Torvalds return -EINVAL; 9731da177e4SLinus Torvalds 974f81cff0dSMike Frysinger /* Careful about overflows.. */ 9758feae131SDavid Howells rlen = PAGE_ALIGN(len); 9768feae131SDavid Howells if (!rlen || rlen > TASK_SIZE) 977f81cff0dSMike Frysinger return -ENOMEM; 978f81cff0dSMike Frysinger 9791da177e4SLinus Torvalds /* offset overflow? */ 9808feae131SDavid Howells if ((pgoff + (rlen >> PAGE_SHIFT)) < pgoff) 981f81cff0dSMike Frysinger return -EOVERFLOW; 9821da177e4SLinus Torvalds 9831da177e4SLinus Torvalds if (file) { 9841da177e4SLinus Torvalds /* files must support mmap */ 98572c2d531SAl Viro if (!file->f_op->mmap) 9861da177e4SLinus Torvalds return -ENODEV; 9871da177e4SLinus Torvalds 9881da177e4SLinus Torvalds /* work out if what we've got could possibly be shared 9891da177e4SLinus Torvalds * - we support chardevs that provide their own "memory" 9901da177e4SLinus Torvalds * - we support files/blockdevs that are memory backed 9911da177e4SLinus Torvalds */ 992b4caecd4SChristoph Hellwig if (file->f_op->mmap_capabilities) { 993b4caecd4SChristoph Hellwig capabilities = file->f_op->mmap_capabilities(file); 994b4caecd4SChristoph Hellwig } else { 9951da177e4SLinus Torvalds /* no explicit capabilities set, so assume some 9961da177e4SLinus Torvalds * defaults */ 997496ad9aaSAl Viro switch (file_inode(file)->i_mode & S_IFMT) { 9981da177e4SLinus Torvalds case S_IFREG: 9991da177e4SLinus Torvalds case S_IFBLK: 1000b4caecd4SChristoph Hellwig capabilities = NOMMU_MAP_COPY; 10011da177e4SLinus Torvalds break; 10021da177e4SLinus Torvalds 10031da177e4SLinus Torvalds case S_IFCHR: 10041da177e4SLinus Torvalds capabilities = 1005b4caecd4SChristoph Hellwig NOMMU_MAP_DIRECT | 1006b4caecd4SChristoph Hellwig NOMMU_MAP_READ | 1007b4caecd4SChristoph Hellwig NOMMU_MAP_WRITE; 10081da177e4SLinus Torvalds break; 10091da177e4SLinus Torvalds 10101da177e4SLinus Torvalds default: 10111da177e4SLinus Torvalds return -EINVAL; 10121da177e4SLinus Torvalds } 10131da177e4SLinus Torvalds } 10141da177e4SLinus Torvalds 10151da177e4SLinus Torvalds /* eliminate any capabilities that we can't support on this 10161da177e4SLinus Torvalds * device */ 10171da177e4SLinus Torvalds if (!file->f_op->get_unmapped_area) 1018b4caecd4SChristoph Hellwig capabilities &= ~NOMMU_MAP_DIRECT; 1019*6e242a1cSAl Viro if (!(file->f_mode & FMODE_CAN_READ)) 1020b4caecd4SChristoph Hellwig capabilities &= ~NOMMU_MAP_COPY; 10211da177e4SLinus Torvalds 102228d7a6aeSGraff Yang /* The file shall have been opened with read permission. */ 102328d7a6aeSGraff Yang if (!(file->f_mode & FMODE_READ)) 102428d7a6aeSGraff Yang return -EACCES; 102528d7a6aeSGraff Yang 10261da177e4SLinus Torvalds if (flags & MAP_SHARED) { 10271da177e4SLinus Torvalds /* do checks for writing, appending and locking */ 10281da177e4SLinus Torvalds if ((prot & PROT_WRITE) && 10291da177e4SLinus Torvalds !(file->f_mode & FMODE_WRITE)) 10301da177e4SLinus Torvalds return -EACCES; 10311da177e4SLinus Torvalds 1032496ad9aaSAl Viro if (IS_APPEND(file_inode(file)) && 10331da177e4SLinus Torvalds (file->f_mode & FMODE_WRITE)) 10341da177e4SLinus Torvalds return -EACCES; 10351da177e4SLinus Torvalds 1036d7a06983SJeff Layton if (locks_verify_locked(file)) 10371da177e4SLinus Torvalds return -EAGAIN; 10381da177e4SLinus Torvalds 1039b4caecd4SChristoph Hellwig if (!(capabilities & NOMMU_MAP_DIRECT)) 10401da177e4SLinus Torvalds return -ENODEV; 10411da177e4SLinus Torvalds 10421da177e4SLinus Torvalds /* we mustn't privatise shared mappings */ 1043b4caecd4SChristoph Hellwig capabilities &= ~NOMMU_MAP_COPY; 1044ac714904SChoi Gi-yong } else { 10451da177e4SLinus Torvalds /* we're going to read the file into private memory we 10461da177e4SLinus Torvalds * allocate */ 1047b4caecd4SChristoph Hellwig if (!(capabilities & NOMMU_MAP_COPY)) 10481da177e4SLinus Torvalds return -ENODEV; 10491da177e4SLinus Torvalds 10501da177e4SLinus Torvalds /* we don't permit a private writable mapping to be 10511da177e4SLinus Torvalds * shared with the backing device */ 10521da177e4SLinus Torvalds if (prot & PROT_WRITE) 1053b4caecd4SChristoph Hellwig capabilities &= ~NOMMU_MAP_DIRECT; 10541da177e4SLinus Torvalds } 10551da177e4SLinus Torvalds 1056b4caecd4SChristoph Hellwig if (capabilities & NOMMU_MAP_DIRECT) { 1057b4caecd4SChristoph Hellwig if (((prot & PROT_READ) && !(capabilities & NOMMU_MAP_READ)) || 1058b4caecd4SChristoph Hellwig ((prot & PROT_WRITE) && !(capabilities & NOMMU_MAP_WRITE)) || 1059b4caecd4SChristoph Hellwig ((prot & PROT_EXEC) && !(capabilities & NOMMU_MAP_EXEC)) 10603c7b2045SBernd Schmidt ) { 1061b4caecd4SChristoph Hellwig capabilities &= ~NOMMU_MAP_DIRECT; 10623c7b2045SBernd Schmidt if (flags & MAP_SHARED) { 10633c7b2045SBernd Schmidt printk(KERN_WARNING 10643c7b2045SBernd Schmidt "MAP_SHARED not completely supported on !MMU\n"); 10653c7b2045SBernd Schmidt return -EINVAL; 10663c7b2045SBernd Schmidt } 10673c7b2045SBernd Schmidt } 10683c7b2045SBernd Schmidt } 10693c7b2045SBernd Schmidt 10701da177e4SLinus Torvalds /* handle executable mappings and implied executable 10711da177e4SLinus Torvalds * mappings */ 1072e9536ae7SJosef Sipek if (file->f_path.mnt->mnt_flags & MNT_NOEXEC) { 10731da177e4SLinus Torvalds if (prot & PROT_EXEC) 10741da177e4SLinus Torvalds return -EPERM; 1075ac714904SChoi Gi-yong } else if ((prot & PROT_READ) && !(prot & PROT_EXEC)) { 10761da177e4SLinus Torvalds /* handle implication of PROT_EXEC by PROT_READ */ 10771da177e4SLinus Torvalds if (current->personality & READ_IMPLIES_EXEC) { 1078b4caecd4SChristoph Hellwig if (capabilities & NOMMU_MAP_EXEC) 10791da177e4SLinus Torvalds prot |= PROT_EXEC; 10801da177e4SLinus Torvalds } 1081ac714904SChoi Gi-yong } else if ((prot & PROT_READ) && 10821da177e4SLinus Torvalds (prot & PROT_EXEC) && 1083b4caecd4SChristoph Hellwig !(capabilities & NOMMU_MAP_EXEC) 10841da177e4SLinus Torvalds ) { 10851da177e4SLinus Torvalds /* backing file is not executable, try to copy */ 1086b4caecd4SChristoph Hellwig capabilities &= ~NOMMU_MAP_DIRECT; 10871da177e4SLinus Torvalds } 1088ac714904SChoi Gi-yong } else { 10891da177e4SLinus Torvalds /* anonymous mappings are always memory backed and can be 10901da177e4SLinus Torvalds * privately mapped 10911da177e4SLinus Torvalds */ 1092b4caecd4SChristoph Hellwig capabilities = NOMMU_MAP_COPY; 10931da177e4SLinus Torvalds 10941da177e4SLinus Torvalds /* handle PROT_EXEC implication by PROT_READ */ 10951da177e4SLinus Torvalds if ((prot & PROT_READ) && 10961da177e4SLinus Torvalds (current->personality & READ_IMPLIES_EXEC)) 10971da177e4SLinus Torvalds prot |= PROT_EXEC; 10981da177e4SLinus Torvalds } 10991da177e4SLinus Torvalds 11001da177e4SLinus Torvalds /* allow the security API to have its say */ 1101e5467859SAl Viro ret = security_mmap_addr(addr); 1102e5467859SAl Viro if (ret < 0) 1103e5467859SAl Viro return ret; 11041da177e4SLinus Torvalds 11051da177e4SLinus Torvalds /* looks okay */ 11061da177e4SLinus Torvalds *_capabilities = capabilities; 11071da177e4SLinus Torvalds return 0; 11081da177e4SLinus Torvalds } 11091da177e4SLinus Torvalds 11101da177e4SLinus Torvalds /* 11111da177e4SLinus Torvalds * we've determined that we can make the mapping, now translate what we 11121da177e4SLinus Torvalds * now know into VMA flags 11131da177e4SLinus Torvalds */ 11141da177e4SLinus Torvalds static unsigned long determine_vm_flags(struct file *file, 11151da177e4SLinus Torvalds unsigned long prot, 11161da177e4SLinus Torvalds unsigned long flags, 11171da177e4SLinus Torvalds unsigned long capabilities) 11181da177e4SLinus Torvalds { 11191da177e4SLinus Torvalds unsigned long vm_flags; 11201da177e4SLinus Torvalds 11211da177e4SLinus Torvalds vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags); 11221da177e4SLinus Torvalds /* vm_flags |= mm->def_flags; */ 11231da177e4SLinus Torvalds 1124b4caecd4SChristoph Hellwig if (!(capabilities & NOMMU_MAP_DIRECT)) { 11251da177e4SLinus Torvalds /* attempt to share read-only copies of mapped file chunks */ 11263c7b2045SBernd Schmidt vm_flags |= VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC; 11271da177e4SLinus Torvalds if (file && !(prot & PROT_WRITE)) 11281da177e4SLinus Torvalds vm_flags |= VM_MAYSHARE; 11293c7b2045SBernd Schmidt } else { 11301da177e4SLinus Torvalds /* overlay a shareable mapping on the backing device or inode 11311da177e4SLinus Torvalds * if possible - used for chardevs, ramfs/tmpfs/shmfs and 11321da177e4SLinus Torvalds * romfs/cramfs */ 1133b4caecd4SChristoph Hellwig vm_flags |= VM_MAYSHARE | (capabilities & NOMMU_VMFLAGS); 11341da177e4SLinus Torvalds if (flags & MAP_SHARED) 11353c7b2045SBernd Schmidt vm_flags |= VM_SHARED; 11361da177e4SLinus Torvalds } 11371da177e4SLinus Torvalds 11381da177e4SLinus Torvalds /* refuse to let anyone share private mappings with this process if 11391da177e4SLinus Torvalds * it's being traced - otherwise breakpoints set in it may interfere 11401da177e4SLinus Torvalds * with another untraced process 11411da177e4SLinus Torvalds */ 1142a288eeccSTejun Heo if ((flags & MAP_PRIVATE) && current->ptrace) 11431da177e4SLinus Torvalds vm_flags &= ~VM_MAYSHARE; 11441da177e4SLinus Torvalds 11451da177e4SLinus Torvalds return vm_flags; 11461da177e4SLinus Torvalds } 11471da177e4SLinus Torvalds 11481da177e4SLinus Torvalds /* 11498feae131SDavid Howells * set up a shared mapping on a file (the driver or filesystem provides and 11508feae131SDavid Howells * pins the storage) 11511da177e4SLinus Torvalds */ 11528feae131SDavid Howells static int do_mmap_shared_file(struct vm_area_struct *vma) 11531da177e4SLinus Torvalds { 11541da177e4SLinus Torvalds int ret; 11551da177e4SLinus Torvalds 11561da177e4SLinus Torvalds ret = vma->vm_file->f_op->mmap(vma->vm_file, vma); 1157dd8632a1SPaul Mundt if (ret == 0) { 1158dd8632a1SPaul Mundt vma->vm_region->vm_top = vma->vm_region->vm_end; 1159645d83c5SDavid Howells return 0; 1160dd8632a1SPaul Mundt } 11611da177e4SLinus Torvalds if (ret != -ENOSYS) 11621da177e4SLinus Torvalds return ret; 11631da177e4SLinus Torvalds 11643fa30460SDavid Howells /* getting -ENOSYS indicates that direct mmap isn't possible (as 11653fa30460SDavid Howells * opposed to tried but failed) so we can only give a suitable error as 11663fa30460SDavid Howells * it's not possible to make a private copy if MAP_SHARED was given */ 11671da177e4SLinus Torvalds return -ENODEV; 11681da177e4SLinus Torvalds } 11691da177e4SLinus Torvalds 11701da177e4SLinus Torvalds /* 11711da177e4SLinus Torvalds * set up a private mapping or an anonymous shared mapping 11721da177e4SLinus Torvalds */ 11738feae131SDavid Howells static int do_mmap_private(struct vm_area_struct *vma, 11748feae131SDavid Howells struct vm_region *region, 1175645d83c5SDavid Howells unsigned long len, 1176645d83c5SDavid Howells unsigned long capabilities) 11771da177e4SLinus Torvalds { 1178dbc8358cSJoonsoo Kim unsigned long total, point; 11791da177e4SLinus Torvalds void *base; 11808feae131SDavid Howells int ret, order; 11811da177e4SLinus Torvalds 11821da177e4SLinus Torvalds /* invoke the file's mapping function so that it can keep track of 11831da177e4SLinus Torvalds * shared mappings on devices or memory 11841da177e4SLinus Torvalds * - VM_MAYSHARE will be set if it may attempt to share 11851da177e4SLinus Torvalds */ 1186b4caecd4SChristoph Hellwig if (capabilities & NOMMU_MAP_DIRECT) { 11871da177e4SLinus Torvalds ret = vma->vm_file->f_op->mmap(vma->vm_file, vma); 1188dd8632a1SPaul Mundt if (ret == 0) { 11891da177e4SLinus Torvalds /* shouldn't return success if we're not sharing */ 1190dd8632a1SPaul Mundt BUG_ON(!(vma->vm_flags & VM_MAYSHARE)); 1191dd8632a1SPaul Mundt vma->vm_region->vm_top = vma->vm_region->vm_end; 1192645d83c5SDavid Howells return 0; 11931da177e4SLinus Torvalds } 1194dd8632a1SPaul Mundt if (ret != -ENOSYS) 1195dd8632a1SPaul Mundt return ret; 11961da177e4SLinus Torvalds 11971da177e4SLinus Torvalds /* getting an ENOSYS error indicates that direct mmap isn't 11981da177e4SLinus Torvalds * possible (as opposed to tried but failed) so we'll try to 11991da177e4SLinus Torvalds * make a private copy of the data and map that instead */ 12001da177e4SLinus Torvalds } 12011da177e4SLinus Torvalds 12028feae131SDavid Howells 12031da177e4SLinus Torvalds /* allocate some memory to hold the mapping 12041da177e4SLinus Torvalds * - note that this may not return a page-aligned address if the object 12051da177e4SLinus Torvalds * we're allocating is smaller than a page 12061da177e4SLinus Torvalds */ 1207f67d9b15SBob Liu order = get_order(len); 12088feae131SDavid Howells kdebug("alloc order %d for %lx", order, len); 12098feae131SDavid Howells 12108feae131SDavid Howells total = 1 << order; 1211f67d9b15SBob Liu point = len >> PAGE_SHIFT; 1212dd8632a1SPaul Mundt 1213dbc8358cSJoonsoo Kim /* we don't want to allocate a power-of-2 sized page set */ 1214dd8632a1SPaul Mundt if (sysctl_nr_trim_pages && total - point >= sysctl_nr_trim_pages) { 1215dbc8358cSJoonsoo Kim total = point; 1216dbc8358cSJoonsoo Kim kdebug("try to alloc exact %lu pages", total); 1217dd8632a1SPaul Mundt } 12188feae131SDavid Howells 1219da616534SJoonsoo Kim base = alloc_pages_exact(total << PAGE_SHIFT, GFP_KERNEL); 1220dbc8358cSJoonsoo Kim if (!base) 1221dbc8358cSJoonsoo Kim goto enomem; 12228feae131SDavid Howells 1223dbc8358cSJoonsoo Kim atomic_long_add(total, &mmap_pages_allocated); 1224dbc8358cSJoonsoo Kim 12258feae131SDavid Howells region->vm_flags = vma->vm_flags |= VM_MAPPED_COPY; 12268feae131SDavid Howells region->vm_start = (unsigned long) base; 1227f67d9b15SBob Liu region->vm_end = region->vm_start + len; 1228dd8632a1SPaul Mundt region->vm_top = region->vm_start + (total << PAGE_SHIFT); 12298feae131SDavid Howells 12308feae131SDavid Howells vma->vm_start = region->vm_start; 12318feae131SDavid Howells vma->vm_end = region->vm_start + len; 12321da177e4SLinus Torvalds 12331da177e4SLinus Torvalds if (vma->vm_file) { 12341da177e4SLinus Torvalds /* read the contents of a file into the copy */ 12351da177e4SLinus Torvalds mm_segment_t old_fs; 12361da177e4SLinus Torvalds loff_t fpos; 12371da177e4SLinus Torvalds 12381da177e4SLinus Torvalds fpos = vma->vm_pgoff; 12391da177e4SLinus Torvalds fpos <<= PAGE_SHIFT; 12401da177e4SLinus Torvalds 12411da177e4SLinus Torvalds old_fs = get_fs(); 12421da177e4SLinus Torvalds set_fs(KERNEL_DS); 1243*6e242a1cSAl Viro ret = __vfs_read(vma->vm_file, base, len, &fpos); 12441da177e4SLinus Torvalds set_fs(old_fs); 12451da177e4SLinus Torvalds 12461da177e4SLinus Torvalds if (ret < 0) 12471da177e4SLinus Torvalds goto error_free; 12481da177e4SLinus Torvalds 12491da177e4SLinus Torvalds /* clear the last little bit */ 1250f67d9b15SBob Liu if (ret < len) 1251f67d9b15SBob Liu memset(base + ret, 0, len - ret); 12521da177e4SLinus Torvalds 12531da177e4SLinus Torvalds } 12541da177e4SLinus Torvalds 12551da177e4SLinus Torvalds return 0; 12561da177e4SLinus Torvalds 12571da177e4SLinus Torvalds error_free: 12587223bb4aSNamhyung Kim free_page_series(region->vm_start, region->vm_top); 12598feae131SDavid Howells region->vm_start = vma->vm_start = 0; 12608feae131SDavid Howells region->vm_end = vma->vm_end = 0; 1261dd8632a1SPaul Mundt region->vm_top = 0; 12621da177e4SLinus Torvalds return ret; 12631da177e4SLinus Torvalds 12641da177e4SLinus Torvalds enomem: 1265b1de0d13SMitchel Humpherys pr_err("Allocation of length %lu from process %d (%s) failed\n", 126605ae6fa3SGreg Ungerer len, current->pid, current->comm); 12677bf02ea2SDavid Rientjes show_free_areas(0); 12681da177e4SLinus Torvalds return -ENOMEM; 12691da177e4SLinus Torvalds } 12701da177e4SLinus Torvalds 12711da177e4SLinus Torvalds /* 12721da177e4SLinus Torvalds * handle mapping creation for uClinux 12731da177e4SLinus Torvalds */ 1274e3fc629dSAl Viro unsigned long do_mmap_pgoff(struct file *file, 12751da177e4SLinus Torvalds unsigned long addr, 12761da177e4SLinus Torvalds unsigned long len, 12771da177e4SLinus Torvalds unsigned long prot, 12781da177e4SLinus Torvalds unsigned long flags, 1279bebeb3d6SMichel Lespinasse unsigned long pgoff, 128041badc15SMichel Lespinasse unsigned long *populate) 12811da177e4SLinus Torvalds { 12828feae131SDavid Howells struct vm_area_struct *vma; 12838feae131SDavid Howells struct vm_region *region; 12841da177e4SLinus Torvalds struct rb_node *rb; 12858feae131SDavid Howells unsigned long capabilities, vm_flags, result; 12861da177e4SLinus Torvalds int ret; 12871da177e4SLinus Torvalds 12888feae131SDavid Howells kenter(",%lx,%lx,%lx,%lx,%lx", addr, len, prot, flags, pgoff); 12898feae131SDavid Howells 129041badc15SMichel Lespinasse *populate = 0; 1291bebeb3d6SMichel Lespinasse 12921da177e4SLinus Torvalds /* decide whether we should attempt the mapping, and if so what sort of 12931da177e4SLinus Torvalds * mapping */ 12941da177e4SLinus Torvalds ret = validate_mmap_request(file, addr, len, prot, flags, pgoff, 12951da177e4SLinus Torvalds &capabilities); 12968feae131SDavid Howells if (ret < 0) { 12978feae131SDavid Howells kleave(" = %d [val]", ret); 12981da177e4SLinus Torvalds return ret; 12998feae131SDavid Howells } 13001da177e4SLinus Torvalds 130106aab5a3SDavid Howells /* we ignore the address hint */ 130206aab5a3SDavid Howells addr = 0; 1303f67d9b15SBob Liu len = PAGE_ALIGN(len); 130406aab5a3SDavid Howells 13051da177e4SLinus Torvalds /* we've determined that we can make the mapping, now translate what we 13061da177e4SLinus Torvalds * now know into VMA flags */ 13071da177e4SLinus Torvalds vm_flags = determine_vm_flags(file, prot, flags, capabilities); 13081da177e4SLinus Torvalds 13098feae131SDavid Howells /* we're going to need to record the mapping */ 13108feae131SDavid Howells region = kmem_cache_zalloc(vm_region_jar, GFP_KERNEL); 13118feae131SDavid Howells if (!region) 13128feae131SDavid Howells goto error_getting_region; 13131da177e4SLinus Torvalds 13148feae131SDavid Howells vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL); 13158feae131SDavid Howells if (!vma) 13168feae131SDavid Howells goto error_getting_vma; 13171da177e4SLinus Torvalds 13181e2ae599SDavid Howells region->vm_usage = 1; 13198feae131SDavid Howells region->vm_flags = vm_flags; 13208feae131SDavid Howells region->vm_pgoff = pgoff; 13218feae131SDavid Howells 13225beb4930SRik van Riel INIT_LIST_HEAD(&vma->anon_vma_chain); 13238feae131SDavid Howells vma->vm_flags = vm_flags; 13248feae131SDavid Howells vma->vm_pgoff = pgoff; 13258feae131SDavid Howells 13268feae131SDavid Howells if (file) { 1327cb0942b8SAl Viro region->vm_file = get_file(file); 1328cb0942b8SAl Viro vma->vm_file = get_file(file); 13298feae131SDavid Howells } 13308feae131SDavid Howells 13318feae131SDavid Howells down_write(&nommu_region_sem); 13328feae131SDavid Howells 13338feae131SDavid Howells /* if we want to share, we need to check for regions created by other 13341da177e4SLinus Torvalds * mmap() calls that overlap with our proposed mapping 13358feae131SDavid Howells * - we can only share with a superset match on most regular files 13361da177e4SLinus Torvalds * - shared mappings on character devices and memory backed files are 13371da177e4SLinus Torvalds * permitted to overlap inexactly as far as we are concerned for in 13381da177e4SLinus Torvalds * these cases, sharing is handled in the driver or filesystem rather 13391da177e4SLinus Torvalds * than here 13401da177e4SLinus Torvalds */ 13411da177e4SLinus Torvalds if (vm_flags & VM_MAYSHARE) { 13428feae131SDavid Howells struct vm_region *pregion; 13438feae131SDavid Howells unsigned long pglen, rpglen, pgend, rpgend, start; 13441da177e4SLinus Torvalds 13458feae131SDavid Howells pglen = (len + PAGE_SIZE - 1) >> PAGE_SHIFT; 13468feae131SDavid Howells pgend = pgoff + pglen; 1347165b2392SDavid Howells 13488feae131SDavid Howells for (rb = rb_first(&nommu_region_tree); rb; rb = rb_next(rb)) { 13498feae131SDavid Howells pregion = rb_entry(rb, struct vm_region, vm_rb); 13501da177e4SLinus Torvalds 13518feae131SDavid Howells if (!(pregion->vm_flags & VM_MAYSHARE)) 13521da177e4SLinus Torvalds continue; 13531da177e4SLinus Torvalds 13541da177e4SLinus Torvalds /* search for overlapping mappings on the same file */ 1355496ad9aaSAl Viro if (file_inode(pregion->vm_file) != 1356496ad9aaSAl Viro file_inode(file)) 13571da177e4SLinus Torvalds continue; 13581da177e4SLinus Torvalds 13598feae131SDavid Howells if (pregion->vm_pgoff >= pgend) 13601da177e4SLinus Torvalds continue; 13611da177e4SLinus Torvalds 13628feae131SDavid Howells rpglen = pregion->vm_end - pregion->vm_start; 13638feae131SDavid Howells rpglen = (rpglen + PAGE_SIZE - 1) >> PAGE_SHIFT; 13648feae131SDavid Howells rpgend = pregion->vm_pgoff + rpglen; 13658feae131SDavid Howells if (pgoff >= rpgend) 13661da177e4SLinus Torvalds continue; 13671da177e4SLinus Torvalds 13688feae131SDavid Howells /* handle inexactly overlapping matches between 13698feae131SDavid Howells * mappings */ 13708feae131SDavid Howells if ((pregion->vm_pgoff != pgoff || rpglen != pglen) && 13718feae131SDavid Howells !(pgoff >= pregion->vm_pgoff && pgend <= rpgend)) { 13728feae131SDavid Howells /* new mapping is not a subset of the region */ 1373b4caecd4SChristoph Hellwig if (!(capabilities & NOMMU_MAP_DIRECT)) 13741da177e4SLinus Torvalds goto sharing_violation; 13751da177e4SLinus Torvalds continue; 13761da177e4SLinus Torvalds } 13771da177e4SLinus Torvalds 13788feae131SDavid Howells /* we've found a region we can share */ 13791e2ae599SDavid Howells pregion->vm_usage++; 13808feae131SDavid Howells vma->vm_region = pregion; 13818feae131SDavid Howells start = pregion->vm_start; 13828feae131SDavid Howells start += (pgoff - pregion->vm_pgoff) << PAGE_SHIFT; 13838feae131SDavid Howells vma->vm_start = start; 13848feae131SDavid Howells vma->vm_end = start + len; 13851da177e4SLinus Torvalds 13868feae131SDavid Howells if (pregion->vm_flags & VM_MAPPED_COPY) { 13878feae131SDavid Howells kdebug("share copy"); 13888feae131SDavid Howells vma->vm_flags |= VM_MAPPED_COPY; 13898feae131SDavid Howells } else { 13908feae131SDavid Howells kdebug("share mmap"); 13918feae131SDavid Howells ret = do_mmap_shared_file(vma); 13928feae131SDavid Howells if (ret < 0) { 13938feae131SDavid Howells vma->vm_region = NULL; 13948feae131SDavid Howells vma->vm_start = 0; 13958feae131SDavid Howells vma->vm_end = 0; 13961e2ae599SDavid Howells pregion->vm_usage--; 13978feae131SDavid Howells pregion = NULL; 13988feae131SDavid Howells goto error_just_free; 13991da177e4SLinus Torvalds } 14008feae131SDavid Howells } 14018feae131SDavid Howells fput(region->vm_file); 14028feae131SDavid Howells kmem_cache_free(vm_region_jar, region); 14038feae131SDavid Howells region = pregion; 14048feae131SDavid Howells result = start; 14058feae131SDavid Howells goto share; 14068feae131SDavid Howells } 14071da177e4SLinus Torvalds 14081da177e4SLinus Torvalds /* obtain the address at which to make a shared mapping 14091da177e4SLinus Torvalds * - this is the hook for quasi-memory character devices to 14101da177e4SLinus Torvalds * tell us the location of a shared mapping 14111da177e4SLinus Torvalds */ 1412b4caecd4SChristoph Hellwig if (capabilities & NOMMU_MAP_DIRECT) { 14131da177e4SLinus Torvalds addr = file->f_op->get_unmapped_area(file, addr, len, 14141da177e4SLinus Torvalds pgoff, flags); 1415bb005a59SNamhyung Kim if (IS_ERR_VALUE(addr)) { 14161da177e4SLinus Torvalds ret = addr; 1417bb005a59SNamhyung Kim if (ret != -ENOSYS) 14188feae131SDavid Howells goto error_just_free; 14191da177e4SLinus Torvalds 14201da177e4SLinus Torvalds /* the driver refused to tell us where to site 14211da177e4SLinus Torvalds * the mapping so we'll have to attempt to copy 14221da177e4SLinus Torvalds * it */ 1423bb005a59SNamhyung Kim ret = -ENODEV; 1424b4caecd4SChristoph Hellwig if (!(capabilities & NOMMU_MAP_COPY)) 14258feae131SDavid Howells goto error_just_free; 14261da177e4SLinus Torvalds 1427b4caecd4SChristoph Hellwig capabilities &= ~NOMMU_MAP_DIRECT; 14288feae131SDavid Howells } else { 14298feae131SDavid Howells vma->vm_start = region->vm_start = addr; 14308feae131SDavid Howells vma->vm_end = region->vm_end = addr + len; 14311da177e4SLinus Torvalds } 14321da177e4SLinus Torvalds } 14331da177e4SLinus Torvalds } 14341da177e4SLinus Torvalds 14358feae131SDavid Howells vma->vm_region = region; 14361da177e4SLinus Torvalds 1437645d83c5SDavid Howells /* set up the mapping 1438b4caecd4SChristoph Hellwig * - the region is filled in if NOMMU_MAP_DIRECT is still set 1439645d83c5SDavid Howells */ 14401da177e4SLinus Torvalds if (file && vma->vm_flags & VM_SHARED) 14418feae131SDavid Howells ret = do_mmap_shared_file(vma); 14421da177e4SLinus Torvalds else 1443645d83c5SDavid Howells ret = do_mmap_private(vma, region, len, capabilities); 14441da177e4SLinus Torvalds if (ret < 0) 1445645d83c5SDavid Howells goto error_just_free; 1446645d83c5SDavid Howells add_nommu_region(region); 14478feae131SDavid Howells 1448ea637639SJie Zhang /* clear anonymous mappings that don't ask for uninitialized data */ 1449ea637639SJie Zhang if (!vma->vm_file && !(flags & MAP_UNINITIALIZED)) 1450ea637639SJie Zhang memset((void *)region->vm_start, 0, 1451ea637639SJie Zhang region->vm_end - region->vm_start); 1452ea637639SJie Zhang 14531da177e4SLinus Torvalds /* okay... we have a mapping; now we have to register it */ 14548feae131SDavid Howells result = vma->vm_start; 14551da177e4SLinus Torvalds 14561da177e4SLinus Torvalds current->mm->total_vm += len >> PAGE_SHIFT; 14571da177e4SLinus Torvalds 14588feae131SDavid Howells share: 14598feae131SDavid Howells add_vma_to_mm(current->mm, vma); 14601da177e4SLinus Torvalds 1461cfe79c00SMike Frysinger /* we flush the region from the icache only when the first executable 1462cfe79c00SMike Frysinger * mapping of it is made */ 1463cfe79c00SMike Frysinger if (vma->vm_flags & VM_EXEC && !region->vm_icache_flushed) { 1464cfe79c00SMike Frysinger flush_icache_range(region->vm_start, region->vm_end); 1465cfe79c00SMike Frysinger region->vm_icache_flushed = true; 1466cfe79c00SMike Frysinger } 14671da177e4SLinus Torvalds 1468cfe79c00SMike Frysinger up_write(&nommu_region_sem); 14691da177e4SLinus Torvalds 14708feae131SDavid Howells kleave(" = %lx", result); 14718feae131SDavid Howells return result; 14721da177e4SLinus Torvalds 14738feae131SDavid Howells error_just_free: 14748feae131SDavid Howells up_write(&nommu_region_sem); 14758feae131SDavid Howells error: 147689a86402SDavid Howells if (region->vm_file) 14778feae131SDavid Howells fput(region->vm_file); 14788feae131SDavid Howells kmem_cache_free(vm_region_jar, region); 147989a86402SDavid Howells if (vma->vm_file) 14808feae131SDavid Howells fput(vma->vm_file); 14818feae131SDavid Howells kmem_cache_free(vm_area_cachep, vma); 14828feae131SDavid Howells kleave(" = %d", ret); 14831da177e4SLinus Torvalds return ret; 14841da177e4SLinus Torvalds 14851da177e4SLinus Torvalds sharing_violation: 14868feae131SDavid Howells up_write(&nommu_region_sem); 14878feae131SDavid Howells printk(KERN_WARNING "Attempt to share mismatched mappings\n"); 14888feae131SDavid Howells ret = -EINVAL; 14898feae131SDavid Howells goto error; 14901da177e4SLinus Torvalds 14911da177e4SLinus Torvalds error_getting_vma: 14928feae131SDavid Howells kmem_cache_free(vm_region_jar, region); 14938feae131SDavid Howells printk(KERN_WARNING "Allocation of vma for %lu byte allocation" 14948feae131SDavid Howells " from process %d failed\n", 14951da177e4SLinus Torvalds len, current->pid); 14967bf02ea2SDavid Rientjes show_free_areas(0); 14971da177e4SLinus Torvalds return -ENOMEM; 14981da177e4SLinus Torvalds 14998feae131SDavid Howells error_getting_region: 15008feae131SDavid Howells printk(KERN_WARNING "Allocation of vm region for %lu byte allocation" 15018feae131SDavid Howells " from process %d failed\n", 15021da177e4SLinus Torvalds len, current->pid); 15037bf02ea2SDavid Rientjes show_free_areas(0); 15041da177e4SLinus Torvalds return -ENOMEM; 15051da177e4SLinus Torvalds } 15066be5ceb0SLinus Torvalds 150766f0dc48SHugh Dickins SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len, 150866f0dc48SHugh Dickins unsigned long, prot, unsigned long, flags, 150966f0dc48SHugh Dickins unsigned long, fd, unsigned long, pgoff) 151066f0dc48SHugh Dickins { 151166f0dc48SHugh Dickins struct file *file = NULL; 151266f0dc48SHugh Dickins unsigned long retval = -EBADF; 151366f0dc48SHugh Dickins 1514120a795dSAl Viro audit_mmap_fd(fd, flags); 151566f0dc48SHugh Dickins if (!(flags & MAP_ANONYMOUS)) { 151666f0dc48SHugh Dickins file = fget(fd); 151766f0dc48SHugh Dickins if (!file) 151866f0dc48SHugh Dickins goto out; 151966f0dc48SHugh Dickins } 152066f0dc48SHugh Dickins 152166f0dc48SHugh Dickins flags &= ~(MAP_EXECUTABLE | MAP_DENYWRITE); 152266f0dc48SHugh Dickins 1523ad1ed293SGreg Ungerer retval = vm_mmap_pgoff(file, addr, len, prot, flags, pgoff); 152466f0dc48SHugh Dickins 152566f0dc48SHugh Dickins if (file) 152666f0dc48SHugh Dickins fput(file); 152766f0dc48SHugh Dickins out: 152866f0dc48SHugh Dickins return retval; 152966f0dc48SHugh Dickins } 153066f0dc48SHugh Dickins 1531a4679373SChristoph Hellwig #ifdef __ARCH_WANT_SYS_OLD_MMAP 1532a4679373SChristoph Hellwig struct mmap_arg_struct { 1533a4679373SChristoph Hellwig unsigned long addr; 1534a4679373SChristoph Hellwig unsigned long len; 1535a4679373SChristoph Hellwig unsigned long prot; 1536a4679373SChristoph Hellwig unsigned long flags; 1537a4679373SChristoph Hellwig unsigned long fd; 1538a4679373SChristoph Hellwig unsigned long offset; 1539a4679373SChristoph Hellwig }; 1540a4679373SChristoph Hellwig 1541a4679373SChristoph Hellwig SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg) 1542a4679373SChristoph Hellwig { 1543a4679373SChristoph Hellwig struct mmap_arg_struct a; 1544a4679373SChristoph Hellwig 1545a4679373SChristoph Hellwig if (copy_from_user(&a, arg, sizeof(a))) 1546a4679373SChristoph Hellwig return -EFAULT; 1547a4679373SChristoph Hellwig if (a.offset & ~PAGE_MASK) 1548a4679373SChristoph Hellwig return -EINVAL; 1549a4679373SChristoph Hellwig 1550a4679373SChristoph Hellwig return sys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd, 1551a4679373SChristoph Hellwig a.offset >> PAGE_SHIFT); 1552a4679373SChristoph Hellwig } 1553a4679373SChristoph Hellwig #endif /* __ARCH_WANT_SYS_OLD_MMAP */ 1554a4679373SChristoph Hellwig 15551da177e4SLinus Torvalds /* 15568feae131SDavid Howells * split a vma into two pieces at address 'addr', a new vma is allocated either 15578feae131SDavid Howells * for the first part or the tail. 15581da177e4SLinus Torvalds */ 15598feae131SDavid Howells int split_vma(struct mm_struct *mm, struct vm_area_struct *vma, 15608feae131SDavid Howells unsigned long addr, int new_below) 15611da177e4SLinus Torvalds { 15628feae131SDavid Howells struct vm_area_struct *new; 15638feae131SDavid Howells struct vm_region *region; 15648feae131SDavid Howells unsigned long npages; 15651da177e4SLinus Torvalds 15668feae131SDavid Howells kenter(""); 15671da177e4SLinus Torvalds 1568779c1023SDavid Howells /* we're only permitted to split anonymous regions (these should have 1569779c1023SDavid Howells * only a single usage on the region) */ 1570779c1023SDavid Howells if (vma->vm_file) 15718feae131SDavid Howells return -ENOMEM; 15721da177e4SLinus Torvalds 15738feae131SDavid Howells if (mm->map_count >= sysctl_max_map_count) 15748feae131SDavid Howells return -ENOMEM; 15751da177e4SLinus Torvalds 15768feae131SDavid Howells region = kmem_cache_alloc(vm_region_jar, GFP_KERNEL); 15778feae131SDavid Howells if (!region) 15788feae131SDavid Howells return -ENOMEM; 15798feae131SDavid Howells 15808feae131SDavid Howells new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL); 15818feae131SDavid Howells if (!new) { 15828feae131SDavid Howells kmem_cache_free(vm_region_jar, region); 15838feae131SDavid Howells return -ENOMEM; 15841da177e4SLinus Torvalds } 15851da177e4SLinus Torvalds 15868feae131SDavid Howells /* most fields are the same, copy all, and then fixup */ 15878feae131SDavid Howells *new = *vma; 15888feae131SDavid Howells *region = *vma->vm_region; 15898feae131SDavid Howells new->vm_region = region; 15908feae131SDavid Howells 15918feae131SDavid Howells npages = (addr - vma->vm_start) >> PAGE_SHIFT; 15928feae131SDavid Howells 15938feae131SDavid Howells if (new_below) { 1594dd8632a1SPaul Mundt region->vm_top = region->vm_end = new->vm_end = addr; 15958feae131SDavid Howells } else { 15968feae131SDavid Howells region->vm_start = new->vm_start = addr; 15978feae131SDavid Howells region->vm_pgoff = new->vm_pgoff += npages; 15981da177e4SLinus Torvalds } 15998feae131SDavid Howells 16008feae131SDavid Howells if (new->vm_ops && new->vm_ops->open) 16018feae131SDavid Howells new->vm_ops->open(new); 16028feae131SDavid Howells 16038feae131SDavid Howells delete_vma_from_mm(vma); 16048feae131SDavid Howells down_write(&nommu_region_sem); 16058feae131SDavid Howells delete_nommu_region(vma->vm_region); 16068feae131SDavid Howells if (new_below) { 16078feae131SDavid Howells vma->vm_region->vm_start = vma->vm_start = addr; 16088feae131SDavid Howells vma->vm_region->vm_pgoff = vma->vm_pgoff += npages; 16098feae131SDavid Howells } else { 16108feae131SDavid Howells vma->vm_region->vm_end = vma->vm_end = addr; 1611dd8632a1SPaul Mundt vma->vm_region->vm_top = addr; 16128feae131SDavid Howells } 16138feae131SDavid Howells add_nommu_region(vma->vm_region); 16148feae131SDavid Howells add_nommu_region(new->vm_region); 16158feae131SDavid Howells up_write(&nommu_region_sem); 16168feae131SDavid Howells add_vma_to_mm(mm, vma); 16178feae131SDavid Howells add_vma_to_mm(mm, new); 16188feae131SDavid Howells return 0; 16198feae131SDavid Howells } 16208feae131SDavid Howells 16218feae131SDavid Howells /* 16228feae131SDavid Howells * shrink a VMA by removing the specified chunk from either the beginning or 16238feae131SDavid Howells * the end 16248feae131SDavid Howells */ 16258feae131SDavid Howells static int shrink_vma(struct mm_struct *mm, 16268feae131SDavid Howells struct vm_area_struct *vma, 16278feae131SDavid Howells unsigned long from, unsigned long to) 16288feae131SDavid Howells { 16298feae131SDavid Howells struct vm_region *region; 16308feae131SDavid Howells 16318feae131SDavid Howells kenter(""); 16328feae131SDavid Howells 16338feae131SDavid Howells /* adjust the VMA's pointers, which may reposition it in the MM's tree 16348feae131SDavid Howells * and list */ 16358feae131SDavid Howells delete_vma_from_mm(vma); 16368feae131SDavid Howells if (from > vma->vm_start) 16378feae131SDavid Howells vma->vm_end = from; 16388feae131SDavid Howells else 16398feae131SDavid Howells vma->vm_start = to; 16408feae131SDavid Howells add_vma_to_mm(mm, vma); 16418feae131SDavid Howells 16428feae131SDavid Howells /* cut the backing region down to size */ 16438feae131SDavid Howells region = vma->vm_region; 16441e2ae599SDavid Howells BUG_ON(region->vm_usage != 1); 16458feae131SDavid Howells 16468feae131SDavid Howells down_write(&nommu_region_sem); 16478feae131SDavid Howells delete_nommu_region(region); 1648dd8632a1SPaul Mundt if (from > region->vm_start) { 1649dd8632a1SPaul Mundt to = region->vm_top; 1650dd8632a1SPaul Mundt region->vm_top = region->vm_end = from; 1651dd8632a1SPaul Mundt } else { 16528feae131SDavid Howells region->vm_start = to; 1653dd8632a1SPaul Mundt } 16548feae131SDavid Howells add_nommu_region(region); 16558feae131SDavid Howells up_write(&nommu_region_sem); 16568feae131SDavid Howells 16578feae131SDavid Howells free_page_series(from, to); 16588feae131SDavid Howells return 0; 16591da177e4SLinus Torvalds } 16601da177e4SLinus Torvalds 16613034097aSDavid Howells /* 16623034097aSDavid Howells * release a mapping 16638feae131SDavid Howells * - under NOMMU conditions the chunk to be unmapped must be backed by a single 16648feae131SDavid Howells * VMA, though it need not cover the whole VMA 16653034097aSDavid Howells */ 16668feae131SDavid Howells int do_munmap(struct mm_struct *mm, unsigned long start, size_t len) 16671da177e4SLinus Torvalds { 16688feae131SDavid Howells struct vm_area_struct *vma; 1669f67d9b15SBob Liu unsigned long end; 16708feae131SDavid Howells int ret; 16711da177e4SLinus Torvalds 16728feae131SDavid Howells kenter(",%lx,%zx", start, len); 16731da177e4SLinus Torvalds 1674f67d9b15SBob Liu len = PAGE_ALIGN(len); 16758feae131SDavid Howells if (len == 0) 16761da177e4SLinus Torvalds return -EINVAL; 16771da177e4SLinus Torvalds 1678f67d9b15SBob Liu end = start + len; 1679f67d9b15SBob Liu 16808feae131SDavid Howells /* find the first potentially overlapping VMA */ 16818feae131SDavid Howells vma = find_vma(mm, start); 16828feae131SDavid Howells if (!vma) { 1683ac714904SChoi Gi-yong static int limit; 168433e5d769SDavid Howells if (limit < 5) { 16858feae131SDavid Howells printk(KERN_WARNING 168633e5d769SDavid Howells "munmap of memory not mmapped by process %d" 168733e5d769SDavid Howells " (%s): 0x%lx-0x%lx\n", 168833e5d769SDavid Howells current->pid, current->comm, 168933e5d769SDavid Howells start, start + len - 1); 169033e5d769SDavid Howells limit++; 169133e5d769SDavid Howells } 16928feae131SDavid Howells return -EINVAL; 16938feae131SDavid Howells } 16941da177e4SLinus Torvalds 16958feae131SDavid Howells /* we're allowed to split an anonymous VMA but not a file-backed one */ 16968feae131SDavid Howells if (vma->vm_file) { 16978feae131SDavid Howells do { 16988feae131SDavid Howells if (start > vma->vm_start) { 16998feae131SDavid Howells kleave(" = -EINVAL [miss]"); 17008feae131SDavid Howells return -EINVAL; 17018feae131SDavid Howells } 17028feae131SDavid Howells if (end == vma->vm_end) 17038feae131SDavid Howells goto erase_whole_vma; 1704d75a310cSNamhyung Kim vma = vma->vm_next; 1705d75a310cSNamhyung Kim } while (vma); 17068feae131SDavid Howells kleave(" = -EINVAL [split file]"); 17078feae131SDavid Howells return -EINVAL; 17088feae131SDavid Howells } else { 17098feae131SDavid Howells /* the chunk must be a subset of the VMA found */ 17108feae131SDavid Howells if (start == vma->vm_start && end == vma->vm_end) 17118feae131SDavid Howells goto erase_whole_vma; 17128feae131SDavid Howells if (start < vma->vm_start || end > vma->vm_end) { 17138feae131SDavid Howells kleave(" = -EINVAL [superset]"); 17148feae131SDavid Howells return -EINVAL; 17158feae131SDavid Howells } 17168feae131SDavid Howells if (start & ~PAGE_MASK) { 17178feae131SDavid Howells kleave(" = -EINVAL [unaligned start]"); 17188feae131SDavid Howells return -EINVAL; 17198feae131SDavid Howells } 17208feae131SDavid Howells if (end != vma->vm_end && end & ~PAGE_MASK) { 17218feae131SDavid Howells kleave(" = -EINVAL [unaligned split]"); 17228feae131SDavid Howells return -EINVAL; 17238feae131SDavid Howells } 17248feae131SDavid Howells if (start != vma->vm_start && end != vma->vm_end) { 17258feae131SDavid Howells ret = split_vma(mm, vma, start, 1); 17268feae131SDavid Howells if (ret < 0) { 17278feae131SDavid Howells kleave(" = %d [split]", ret); 17288feae131SDavid Howells return ret; 17298feae131SDavid Howells } 17308feae131SDavid Howells } 17318feae131SDavid Howells return shrink_vma(mm, vma, start, end); 17328feae131SDavid Howells } 17331da177e4SLinus Torvalds 17348feae131SDavid Howells erase_whole_vma: 17358feae131SDavid Howells delete_vma_from_mm(vma); 17368feae131SDavid Howells delete_vma(mm, vma); 17378feae131SDavid Howells kleave(" = 0"); 17381da177e4SLinus Torvalds return 0; 17391da177e4SLinus Torvalds } 1740b5073173SPaul Mundt EXPORT_SYMBOL(do_munmap); 17411da177e4SLinus Torvalds 1742bfce281cSAl Viro int vm_munmap(unsigned long addr, size_t len) 17433034097aSDavid Howells { 1744bfce281cSAl Viro struct mm_struct *mm = current->mm; 17453034097aSDavid Howells int ret; 17463034097aSDavid Howells 17473034097aSDavid Howells down_write(&mm->mmap_sem); 17483034097aSDavid Howells ret = do_munmap(mm, addr, len); 17493034097aSDavid Howells up_write(&mm->mmap_sem); 17503034097aSDavid Howells return ret; 17513034097aSDavid Howells } 1752a46ef99dSLinus Torvalds EXPORT_SYMBOL(vm_munmap); 1753a46ef99dSLinus Torvalds 1754a46ef99dSLinus Torvalds SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len) 1755a46ef99dSLinus Torvalds { 1756bfce281cSAl Viro return vm_munmap(addr, len); 1757a46ef99dSLinus Torvalds } 17583034097aSDavid Howells 17593034097aSDavid Howells /* 17608feae131SDavid Howells * release all the mappings made in a process's VM space 17613034097aSDavid Howells */ 17621da177e4SLinus Torvalds void exit_mmap(struct mm_struct *mm) 17631da177e4SLinus Torvalds { 17648feae131SDavid Howells struct vm_area_struct *vma; 17651da177e4SLinus Torvalds 17668feae131SDavid Howells if (!mm) 17678feae131SDavid Howells return; 17688feae131SDavid Howells 17698feae131SDavid Howells kenter(""); 17701da177e4SLinus Torvalds 17711da177e4SLinus Torvalds mm->total_vm = 0; 17721da177e4SLinus Torvalds 17738feae131SDavid Howells while ((vma = mm->mmap)) { 17748feae131SDavid Howells mm->mmap = vma->vm_next; 17758feae131SDavid Howells delete_vma_from_mm(vma); 17768feae131SDavid Howells delete_vma(mm, vma); 177704c34961SSteven J. Magnani cond_resched(); 17781da177e4SLinus Torvalds } 17791da177e4SLinus Torvalds 17808feae131SDavid Howells kleave(""); 17811da177e4SLinus Torvalds } 17821da177e4SLinus Torvalds 1783e4eb1ff6SLinus Torvalds unsigned long vm_brk(unsigned long addr, unsigned long len) 17841da177e4SLinus Torvalds { 17851da177e4SLinus Torvalds return -ENOMEM; 17861da177e4SLinus Torvalds } 17871da177e4SLinus Torvalds 17881da177e4SLinus Torvalds /* 17896fa5f80bSDavid Howells * expand (or shrink) an existing mapping, potentially moving it at the same 17906fa5f80bSDavid Howells * time (controlled by the MREMAP_MAYMOVE flag and available VM space) 17911da177e4SLinus Torvalds * 17926fa5f80bSDavid Howells * under NOMMU conditions, we only permit changing a mapping's size, and only 17938feae131SDavid Howells * as long as it stays within the region allocated by do_mmap_private() and the 17948feae131SDavid Howells * block is not shareable 17951da177e4SLinus Torvalds * 17966fa5f80bSDavid Howells * MREMAP_FIXED is not supported under NOMMU conditions 17971da177e4SLinus Torvalds */ 17984b377babSAl Viro static unsigned long do_mremap(unsigned long addr, 17991da177e4SLinus Torvalds unsigned long old_len, unsigned long new_len, 18001da177e4SLinus Torvalds unsigned long flags, unsigned long new_addr) 18011da177e4SLinus Torvalds { 18026fa5f80bSDavid Howells struct vm_area_struct *vma; 18031da177e4SLinus Torvalds 18041da177e4SLinus Torvalds /* insanity checks first */ 1805f67d9b15SBob Liu old_len = PAGE_ALIGN(old_len); 1806f67d9b15SBob Liu new_len = PAGE_ALIGN(new_len); 18078feae131SDavid Howells if (old_len == 0 || new_len == 0) 18081da177e4SLinus Torvalds return (unsigned long) -EINVAL; 18091da177e4SLinus Torvalds 18108feae131SDavid Howells if (addr & ~PAGE_MASK) 18118feae131SDavid Howells return -EINVAL; 18128feae131SDavid Howells 18131da177e4SLinus Torvalds if (flags & MREMAP_FIXED && new_addr != addr) 18141da177e4SLinus Torvalds return (unsigned long) -EINVAL; 18151da177e4SLinus Torvalds 18168feae131SDavid Howells vma = find_vma_exact(current->mm, addr, old_len); 18176fa5f80bSDavid Howells if (!vma) 18181da177e4SLinus Torvalds return (unsigned long) -EINVAL; 18191da177e4SLinus Torvalds 18206fa5f80bSDavid Howells if (vma->vm_end != vma->vm_start + old_len) 18211da177e4SLinus Torvalds return (unsigned long) -EFAULT; 18221da177e4SLinus Torvalds 18236fa5f80bSDavid Howells if (vma->vm_flags & VM_MAYSHARE) 18241da177e4SLinus Torvalds return (unsigned long) -EPERM; 18251da177e4SLinus Torvalds 18268feae131SDavid Howells if (new_len > vma->vm_region->vm_end - vma->vm_region->vm_start) 18271da177e4SLinus Torvalds return (unsigned long) -ENOMEM; 18281da177e4SLinus Torvalds 18291da177e4SLinus Torvalds /* all checks complete - do it */ 18306fa5f80bSDavid Howells vma->vm_end = vma->vm_start + new_len; 18316fa5f80bSDavid Howells return vma->vm_start; 18326fa5f80bSDavid Howells } 18336fa5f80bSDavid Howells 18346a6160a7SHeiko Carstens SYSCALL_DEFINE5(mremap, unsigned long, addr, unsigned long, old_len, 18356a6160a7SHeiko Carstens unsigned long, new_len, unsigned long, flags, 18366a6160a7SHeiko Carstens unsigned long, new_addr) 18376fa5f80bSDavid Howells { 18386fa5f80bSDavid Howells unsigned long ret; 18396fa5f80bSDavid Howells 18406fa5f80bSDavid Howells down_write(¤t->mm->mmap_sem); 18416fa5f80bSDavid Howells ret = do_mremap(addr, old_len, new_len, flags, new_addr); 18426fa5f80bSDavid Howells up_write(¤t->mm->mmap_sem); 18436fa5f80bSDavid Howells return ret; 18441da177e4SLinus Torvalds } 18451da177e4SLinus Torvalds 1846240aadeeSMichel Lespinasse struct page *follow_page_mask(struct vm_area_struct *vma, 1847240aadeeSMichel Lespinasse unsigned long address, unsigned int flags, 1848240aadeeSMichel Lespinasse unsigned int *page_mask) 18491da177e4SLinus Torvalds { 1850240aadeeSMichel Lespinasse *page_mask = 0; 18511da177e4SLinus Torvalds return NULL; 18521da177e4SLinus Torvalds } 18531da177e4SLinus Torvalds 18548f3b1327SBob Liu int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr, 18558f3b1327SBob Liu unsigned long pfn, unsigned long size, pgprot_t prot) 18561da177e4SLinus Torvalds { 18578f3b1327SBob Liu if (addr != (pfn << PAGE_SHIFT)) 18588f3b1327SBob Liu return -EINVAL; 18598f3b1327SBob Liu 1860314e51b9SKonstantin Khlebnikov vma->vm_flags |= VM_IO | VM_PFNMAP | VM_DONTEXPAND | VM_DONTDUMP; 186166aa2b4bSGreg Ungerer return 0; 18621da177e4SLinus Torvalds } 186322c4af40SLuke Yang EXPORT_SYMBOL(remap_pfn_range); 18641da177e4SLinus Torvalds 18653c0b9de6SLinus Torvalds int vm_iomap_memory(struct vm_area_struct *vma, phys_addr_t start, unsigned long len) 18663c0b9de6SLinus Torvalds { 18673c0b9de6SLinus Torvalds unsigned long pfn = start >> PAGE_SHIFT; 18683c0b9de6SLinus Torvalds unsigned long vm_len = vma->vm_end - vma->vm_start; 18693c0b9de6SLinus Torvalds 18703c0b9de6SLinus Torvalds pfn += vma->vm_pgoff; 18713c0b9de6SLinus Torvalds return io_remap_pfn_range(vma, vma->vm_start, pfn, vm_len, vma->vm_page_prot); 18723c0b9de6SLinus Torvalds } 18733c0b9de6SLinus Torvalds EXPORT_SYMBOL(vm_iomap_memory); 18743c0b9de6SLinus Torvalds 1875f905bc44SPaul Mundt int remap_vmalloc_range(struct vm_area_struct *vma, void *addr, 1876f905bc44SPaul Mundt unsigned long pgoff) 1877f905bc44SPaul Mundt { 1878f905bc44SPaul Mundt unsigned int size = vma->vm_end - vma->vm_start; 1879f905bc44SPaul Mundt 1880f905bc44SPaul Mundt if (!(vma->vm_flags & VM_USERMAP)) 1881f905bc44SPaul Mundt return -EINVAL; 1882f905bc44SPaul Mundt 1883f905bc44SPaul Mundt vma->vm_start = (unsigned long)(addr + (pgoff << PAGE_SHIFT)); 1884f905bc44SPaul Mundt vma->vm_end = vma->vm_start + size; 1885f905bc44SPaul Mundt 1886f905bc44SPaul Mundt return 0; 1887f905bc44SPaul Mundt } 1888f905bc44SPaul Mundt EXPORT_SYMBOL(remap_vmalloc_range); 1889f905bc44SPaul Mundt 18901da177e4SLinus Torvalds unsigned long arch_get_unmapped_area(struct file *file, unsigned long addr, 18911da177e4SLinus Torvalds unsigned long len, unsigned long pgoff, unsigned long flags) 18921da177e4SLinus Torvalds { 18931da177e4SLinus Torvalds return -ENOMEM; 18941da177e4SLinus Torvalds } 18951da177e4SLinus Torvalds 18961da177e4SLinus Torvalds void unmap_mapping_range(struct address_space *mapping, 18971da177e4SLinus Torvalds loff_t const holebegin, loff_t const holelen, 18981da177e4SLinus Torvalds int even_cows) 18991da177e4SLinus Torvalds { 19001da177e4SLinus Torvalds } 190122c4af40SLuke Yang EXPORT_SYMBOL(unmap_mapping_range); 19021da177e4SLinus Torvalds 19031da177e4SLinus Torvalds /* 19041da177e4SLinus Torvalds * Check that a process has enough memory to allocate a new virtual 19051da177e4SLinus Torvalds * mapping. 0 means there is enough memory for the allocation to 19061da177e4SLinus Torvalds * succeed and -ENOMEM implies there is not. 19071da177e4SLinus Torvalds * 19081da177e4SLinus Torvalds * We currently support three overcommit policies, which are set via the 19091da177e4SLinus Torvalds * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting 19101da177e4SLinus Torvalds * 19111da177e4SLinus Torvalds * Strict overcommit modes added 2002 Feb 26 by Alan Cox. 19121da177e4SLinus Torvalds * Additional code 2002 Jul 20 by Robert Love. 19131da177e4SLinus Torvalds * 19141da177e4SLinus Torvalds * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise. 19151da177e4SLinus Torvalds * 19161da177e4SLinus Torvalds * Note this is a helper function intended to be used by LSMs which 19171da177e4SLinus Torvalds * wish to use this logic. 19181da177e4SLinus Torvalds */ 191934b4e4aaSAlan Cox int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin) 19201da177e4SLinus Torvalds { 19218138a67aSRoman Gushchin long free, allowed, reserve; 19221da177e4SLinus Torvalds 19231da177e4SLinus Torvalds vm_acct_memory(pages); 19241da177e4SLinus Torvalds 19251da177e4SLinus Torvalds /* 19261da177e4SLinus Torvalds * Sometimes we want to use more memory than we have 19271da177e4SLinus Torvalds */ 19281da177e4SLinus Torvalds if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS) 19291da177e4SLinus Torvalds return 0; 19301da177e4SLinus Torvalds 19311da177e4SLinus Torvalds if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) { 1932c15bef30SDmitry Fink free = global_page_state(NR_FREE_PAGES); 1933c15bef30SDmitry Fink free += global_page_state(NR_FILE_PAGES); 19341da177e4SLinus Torvalds 1935c15bef30SDmitry Fink /* 1936c15bef30SDmitry Fink * shmem pages shouldn't be counted as free in this 1937c15bef30SDmitry Fink * case, they can't be purged, only swapped out, and 1938c15bef30SDmitry Fink * that won't affect the overall amount of available 1939c15bef30SDmitry Fink * memory in the system. 1940c15bef30SDmitry Fink */ 1941c15bef30SDmitry Fink free -= global_page_state(NR_SHMEM); 1942c15bef30SDmitry Fink 1943ec8acf20SShaohua Li free += get_nr_swap_pages(); 19441da177e4SLinus Torvalds 19451da177e4SLinus Torvalds /* 19461da177e4SLinus Torvalds * Any slabs which are created with the 19471da177e4SLinus Torvalds * SLAB_RECLAIM_ACCOUNT flag claim to have contents 19481da177e4SLinus Torvalds * which are reclaimable, under pressure. The dentry 19491da177e4SLinus Torvalds * cache and most inode caches should fall into this 19501da177e4SLinus Torvalds */ 1951972d1a7bSChristoph Lameter free += global_page_state(NR_SLAB_RECLAIMABLE); 19521da177e4SLinus Torvalds 19531da177e4SLinus Torvalds /* 1954c15bef30SDmitry Fink * Leave reserved pages. The pages are not for anonymous pages. 1955c15bef30SDmitry Fink */ 1956c15bef30SDmitry Fink if (free <= totalreserve_pages) 1957c15bef30SDmitry Fink goto error; 1958c15bef30SDmitry Fink else 1959c15bef30SDmitry Fink free -= totalreserve_pages; 1960c15bef30SDmitry Fink 1961c15bef30SDmitry Fink /* 19624eeab4f5SAndrew Shewmaker * Reserve some for root 19631da177e4SLinus Torvalds */ 19641da177e4SLinus Torvalds if (!cap_sys_admin) 19654eeab4f5SAndrew Shewmaker free -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10); 19661da177e4SLinus Torvalds 19671da177e4SLinus Torvalds if (free > pages) 19681da177e4SLinus Torvalds return 0; 19691da177e4SLinus Torvalds 1970d5ddc79bSHideo AOKI goto error; 19711da177e4SLinus Torvalds } 19721da177e4SLinus Torvalds 197300619bccSJerome Marchand allowed = vm_commit_limit(); 19741da177e4SLinus Torvalds /* 19754eeab4f5SAndrew Shewmaker * Reserve some 3% for root 19761da177e4SLinus Torvalds */ 19771da177e4SLinus Torvalds if (!cap_sys_admin) 19784eeab4f5SAndrew Shewmaker allowed -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10); 19791da177e4SLinus Torvalds 1980c9b1d098SAndrew Shewmaker /* 1981c9b1d098SAndrew Shewmaker * Don't let a single process grow so big a user can't recover 1982c9b1d098SAndrew Shewmaker */ 1983c9b1d098SAndrew Shewmaker if (mm) { 1984c9b1d098SAndrew Shewmaker reserve = sysctl_user_reserve_kbytes >> (PAGE_SHIFT - 10); 19858138a67aSRoman Gushchin allowed -= min_t(long, mm->total_vm / 32, reserve); 1986c9b1d098SAndrew Shewmaker } 19871da177e4SLinus Torvalds 198800a62ce9SKOSAKI Motohiro if (percpu_counter_read_positive(&vm_committed_as) < allowed) 19891da177e4SLinus Torvalds return 0; 199000a62ce9SKOSAKI Motohiro 1991d5ddc79bSHideo AOKI error: 19921da177e4SLinus Torvalds vm_unacct_memory(pages); 19931da177e4SLinus Torvalds 19941da177e4SLinus Torvalds return -ENOMEM; 19951da177e4SLinus Torvalds } 19961da177e4SLinus Torvalds 1997d0217ac0SNick Piggin int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf) 1998b0e15190SDavid Howells { 1999b0e15190SDavid Howells BUG(); 2000d0217ac0SNick Piggin return 0; 2001b0e15190SDavid Howells } 2002b5073173SPaul Mundt EXPORT_SYMBOL(filemap_fault); 20030ec76a11SDavid Howells 2004f1820361SKirill A. Shutemov void filemap_map_pages(struct vm_area_struct *vma, struct vm_fault *vmf) 2005f1820361SKirill A. Shutemov { 2006f1820361SKirill A. Shutemov BUG(); 2007f1820361SKirill A. Shutemov } 2008f1820361SKirill A. Shutemov EXPORT_SYMBOL(filemap_map_pages); 2009f1820361SKirill A. Shutemov 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); 21021acf2e04SDavidlohr Bueso i_mmap_lock_read(inode->i_mapping); 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) { 21091acf2e04SDavidlohr Bueso i_mmap_unlock_read(inode->i_mapping); 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 */ 21211acf2e04SDavidlohr Bueso vma_interval_tree_foreach(vma, &inode->i_mapping->i_mmap, 0, ULONG_MAX) { 21227e660872SDavid Howells if (!(vma->vm_flags & VM_SHARED)) 21237e660872SDavid Howells continue; 21247e660872SDavid Howells 21257e660872SDavid Howells region = vma->vm_region; 21267e660872SDavid Howells r_size = region->vm_top - region->vm_start; 21277e660872SDavid Howells r_top = (region->vm_pgoff << PAGE_SHIFT) + r_size; 21287e660872SDavid Howells 21297e660872SDavid Howells if (r_top > newsize) { 21307e660872SDavid Howells region->vm_top -= r_top - newsize; 21317e660872SDavid Howells if (region->vm_end > region->vm_top) 21327e660872SDavid Howells region->vm_end = region->vm_top; 21337e660872SDavid Howells } 21347e660872SDavid Howells } 21357e660872SDavid Howells 21361acf2e04SDavidlohr Bueso i_mmap_unlock_read(inode->i_mapping); 21377e660872SDavid Howells up_write(&nommu_region_sem); 21387e660872SDavid Howells return 0; 21397e660872SDavid Howells } 2140c9b1d098SAndrew Shewmaker 2141c9b1d098SAndrew Shewmaker /* 2142c9b1d098SAndrew Shewmaker * Initialise sysctl_user_reserve_kbytes. 2143c9b1d098SAndrew Shewmaker * 2144c9b1d098SAndrew Shewmaker * This is intended to prevent a user from starting a single memory hogging 2145c9b1d098SAndrew Shewmaker * process, such that they cannot recover (kill the hog) in OVERCOMMIT_NEVER 2146c9b1d098SAndrew Shewmaker * mode. 2147c9b1d098SAndrew Shewmaker * 2148c9b1d098SAndrew Shewmaker * The default value is min(3% of free memory, 128MB) 2149c9b1d098SAndrew Shewmaker * 128MB is enough to recover with sshd/login, bash, and top/kill. 2150c9b1d098SAndrew Shewmaker */ 2151c9b1d098SAndrew Shewmaker static int __meminit init_user_reserve(void) 2152c9b1d098SAndrew Shewmaker { 2153c9b1d098SAndrew Shewmaker unsigned long free_kbytes; 2154c9b1d098SAndrew Shewmaker 2155c9b1d098SAndrew Shewmaker free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10); 2156c9b1d098SAndrew Shewmaker 2157c9b1d098SAndrew Shewmaker sysctl_user_reserve_kbytes = min(free_kbytes / 32, 1UL << 17); 2158c9b1d098SAndrew Shewmaker return 0; 2159c9b1d098SAndrew Shewmaker } 2160c9b1d098SAndrew Shewmaker module_init(init_user_reserve) 21614eeab4f5SAndrew Shewmaker 21624eeab4f5SAndrew Shewmaker /* 21634eeab4f5SAndrew Shewmaker * Initialise sysctl_admin_reserve_kbytes. 21644eeab4f5SAndrew Shewmaker * 21654eeab4f5SAndrew Shewmaker * The purpose of sysctl_admin_reserve_kbytes is to allow the sys admin 21664eeab4f5SAndrew Shewmaker * to log in and kill a memory hogging process. 21674eeab4f5SAndrew Shewmaker * 21684eeab4f5SAndrew Shewmaker * Systems with more than 256MB will reserve 8MB, enough to recover 21694eeab4f5SAndrew Shewmaker * with sshd, bash, and top in OVERCOMMIT_GUESS. Smaller systems will 21704eeab4f5SAndrew Shewmaker * only reserve 3% of free pages by default. 21714eeab4f5SAndrew Shewmaker */ 21724eeab4f5SAndrew Shewmaker static int __meminit init_admin_reserve(void) 21734eeab4f5SAndrew Shewmaker { 21744eeab4f5SAndrew Shewmaker unsigned long free_kbytes; 21754eeab4f5SAndrew Shewmaker 21764eeab4f5SAndrew Shewmaker free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10); 21774eeab4f5SAndrew Shewmaker 21784eeab4f5SAndrew Shewmaker sysctl_admin_reserve_kbytes = min(free_kbytes / 32, 1UL << 13); 21794eeab4f5SAndrew Shewmaker return 0; 21804eeab4f5SAndrew Shewmaker } 21814eeab4f5SAndrew Shewmaker module_init(init_admin_reserve) 2182