11da177e4SLinus Torvalds /* 21da177e4SLinus Torvalds * mm/mmap.c 31da177e4SLinus Torvalds * 41da177e4SLinus Torvalds * Written by obz. 51da177e4SLinus Torvalds * 61da177e4SLinus Torvalds * Address space accounting code <alan@redhat.com> 71da177e4SLinus Torvalds */ 81da177e4SLinus Torvalds 91da177e4SLinus Torvalds #include <linux/slab.h> 104af3c9ccSAlexey Dobriyan #include <linux/backing-dev.h> 111da177e4SLinus Torvalds #include <linux/mm.h> 121da177e4SLinus Torvalds #include <linux/shm.h> 131da177e4SLinus Torvalds #include <linux/mman.h> 141da177e4SLinus Torvalds #include <linux/pagemap.h> 151da177e4SLinus Torvalds #include <linux/swap.h> 161da177e4SLinus Torvalds #include <linux/syscalls.h> 17c59ede7bSRandy.Dunlap #include <linux/capability.h> 181da177e4SLinus Torvalds #include <linux/init.h> 191da177e4SLinus Torvalds #include <linux/file.h> 201da177e4SLinus Torvalds #include <linux/fs.h> 211da177e4SLinus Torvalds #include <linux/personality.h> 221da177e4SLinus Torvalds #include <linux/security.h> 231da177e4SLinus Torvalds #include <linux/hugetlb.h> 241da177e4SLinus Torvalds #include <linux/profile.h> 251da177e4SLinus Torvalds #include <linux/module.h> 261da177e4SLinus Torvalds #include <linux/mount.h> 271da177e4SLinus Torvalds #include <linux/mempolicy.h> 281da177e4SLinus Torvalds #include <linux/rmap.h> 291da177e4SLinus Torvalds 301da177e4SLinus Torvalds #include <asm/uaccess.h> 311da177e4SLinus Torvalds #include <asm/cacheflush.h> 321da177e4SLinus Torvalds #include <asm/tlb.h> 33d6dd61c8SJeremy Fitzhardinge #include <asm/mmu_context.h> 341da177e4SLinus Torvalds 3542b77728SJan Beulich #include "internal.h" 3642b77728SJan Beulich 373a459756SKirill Korotaev #ifndef arch_mmap_check 383a459756SKirill Korotaev #define arch_mmap_check(addr, len, flags) (0) 393a459756SKirill Korotaev #endif 403a459756SKirill Korotaev 4108e7d9b5SMartin Schwidefsky #ifndef arch_rebalance_pgtables 4208e7d9b5SMartin Schwidefsky #define arch_rebalance_pgtables(addr, len) (addr) 4308e7d9b5SMartin Schwidefsky #endif 4408e7d9b5SMartin Schwidefsky 45e0da382cSHugh Dickins static void unmap_region(struct mm_struct *mm, 46e0da382cSHugh Dickins struct vm_area_struct *vma, struct vm_area_struct *prev, 47e0da382cSHugh Dickins unsigned long start, unsigned long end); 48e0da382cSHugh Dickins 491da177e4SLinus Torvalds /* 501da177e4SLinus Torvalds * WARNING: the debugging will use recursive algorithms so never enable this 511da177e4SLinus Torvalds * unless you know what you are doing. 521da177e4SLinus Torvalds */ 531da177e4SLinus Torvalds #undef DEBUG_MM_RB 541da177e4SLinus Torvalds 551da177e4SLinus Torvalds /* description of effects of mapping type and prot in current implementation. 561da177e4SLinus Torvalds * this is due to the limited x86 page protection hardware. The expected 571da177e4SLinus Torvalds * behavior is in parens: 581da177e4SLinus Torvalds * 591da177e4SLinus Torvalds * map_type prot 601da177e4SLinus Torvalds * PROT_NONE PROT_READ PROT_WRITE PROT_EXEC 611da177e4SLinus Torvalds * MAP_SHARED r: (no) no r: (yes) yes r: (no) yes r: (no) yes 621da177e4SLinus Torvalds * w: (no) no w: (no) no w: (yes) yes w: (no) no 631da177e4SLinus Torvalds * x: (no) no x: (no) yes x: (no) yes x: (yes) yes 641da177e4SLinus Torvalds * 651da177e4SLinus Torvalds * MAP_PRIVATE r: (no) no r: (yes) yes r: (no) yes r: (no) yes 661da177e4SLinus Torvalds * w: (no) no w: (no) no w: (copy) copy w: (no) no 671da177e4SLinus Torvalds * x: (no) no x: (no) yes x: (no) yes x: (yes) yes 681da177e4SLinus Torvalds * 691da177e4SLinus Torvalds */ 701da177e4SLinus Torvalds pgprot_t protection_map[16] = { 711da177e4SLinus Torvalds __P000, __P001, __P010, __P011, __P100, __P101, __P110, __P111, 721da177e4SLinus Torvalds __S000, __S001, __S010, __S011, __S100, __S101, __S110, __S111 731da177e4SLinus Torvalds }; 741da177e4SLinus Torvalds 75804af2cfSHugh Dickins pgprot_t vm_get_page_prot(unsigned long vm_flags) 76804af2cfSHugh Dickins { 77b845f313SDave Kleikamp return __pgprot(pgprot_val(protection_map[vm_flags & 78b845f313SDave Kleikamp (VM_READ|VM_WRITE|VM_EXEC|VM_SHARED)]) | 79b845f313SDave Kleikamp pgprot_val(arch_vm_get_page_prot(vm_flags))); 80804af2cfSHugh Dickins } 81804af2cfSHugh Dickins EXPORT_SYMBOL(vm_get_page_prot); 82804af2cfSHugh Dickins 831da177e4SLinus Torvalds int sysctl_overcommit_memory = OVERCOMMIT_GUESS; /* heuristic overcommit */ 841da177e4SLinus Torvalds int sysctl_overcommit_ratio = 50; /* default is 50% */ 85c3d8c141SChristoph Lameter int sysctl_max_map_count __read_mostly = DEFAULT_MAX_MAP_COUNT; 8680119ef5SAlan Cox atomic_long_t vm_committed_space = ATOMIC_LONG_INIT(0); 871da177e4SLinus Torvalds 881da177e4SLinus Torvalds /* 891da177e4SLinus Torvalds * Check that a process has enough memory to allocate a new virtual 901da177e4SLinus Torvalds * mapping. 0 means there is enough memory for the allocation to 911da177e4SLinus Torvalds * succeed and -ENOMEM implies there is not. 921da177e4SLinus Torvalds * 931da177e4SLinus Torvalds * We currently support three overcommit policies, which are set via the 941da177e4SLinus Torvalds * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting 951da177e4SLinus Torvalds * 961da177e4SLinus Torvalds * Strict overcommit modes added 2002 Feb 26 by Alan Cox. 971da177e4SLinus Torvalds * Additional code 2002 Jul 20 by Robert Love. 981da177e4SLinus Torvalds * 991da177e4SLinus Torvalds * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise. 1001da177e4SLinus Torvalds * 1011da177e4SLinus Torvalds * Note this is a helper function intended to be used by LSMs which 1021da177e4SLinus Torvalds * wish to use this logic. 1031da177e4SLinus Torvalds */ 10434b4e4aaSAlan Cox int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin) 1051da177e4SLinus Torvalds { 1061da177e4SLinus Torvalds unsigned long free, allowed; 1071da177e4SLinus Torvalds 1081da177e4SLinus Torvalds vm_acct_memory(pages); 1091da177e4SLinus Torvalds 1101da177e4SLinus Torvalds /* 1111da177e4SLinus Torvalds * Sometimes we want to use more memory than we have 1121da177e4SLinus Torvalds */ 1131da177e4SLinus Torvalds if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS) 1141da177e4SLinus Torvalds return 0; 1151da177e4SLinus Torvalds 1161da177e4SLinus Torvalds if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) { 1171da177e4SLinus Torvalds unsigned long n; 1181da177e4SLinus Torvalds 119347ce434SChristoph Lameter free = global_page_state(NR_FILE_PAGES); 1201da177e4SLinus Torvalds free += nr_swap_pages; 1211da177e4SLinus Torvalds 1221da177e4SLinus Torvalds /* 1231da177e4SLinus Torvalds * Any slabs which are created with the 1241da177e4SLinus Torvalds * SLAB_RECLAIM_ACCOUNT flag claim to have contents 1251da177e4SLinus Torvalds * which are reclaimable, under pressure. The dentry 1261da177e4SLinus Torvalds * cache and most inode caches should fall into this 1271da177e4SLinus Torvalds */ 128972d1a7bSChristoph Lameter free += global_page_state(NR_SLAB_RECLAIMABLE); 1291da177e4SLinus Torvalds 1301da177e4SLinus Torvalds /* 1311da177e4SLinus Torvalds * Leave the last 3% for root 1321da177e4SLinus Torvalds */ 1331da177e4SLinus Torvalds if (!cap_sys_admin) 1341da177e4SLinus Torvalds free -= free / 32; 1351da177e4SLinus Torvalds 1361da177e4SLinus Torvalds if (free > pages) 1371da177e4SLinus Torvalds return 0; 1381da177e4SLinus Torvalds 1391da177e4SLinus Torvalds /* 1401da177e4SLinus Torvalds * nr_free_pages() is very expensive on large systems, 1411da177e4SLinus Torvalds * only call if we're about to fail. 1421da177e4SLinus Torvalds */ 1431da177e4SLinus Torvalds n = nr_free_pages(); 1446d9f7839SHideo AOKI 1456d9f7839SHideo AOKI /* 1466d9f7839SHideo AOKI * Leave reserved pages. The pages are not for anonymous pages. 1476d9f7839SHideo AOKI */ 1486d9f7839SHideo AOKI if (n <= totalreserve_pages) 1496d9f7839SHideo AOKI goto error; 1506d9f7839SHideo AOKI else 1516d9f7839SHideo AOKI n -= totalreserve_pages; 1526d9f7839SHideo AOKI 1536d9f7839SHideo AOKI /* 1546d9f7839SHideo AOKI * Leave the last 3% for root 1556d9f7839SHideo AOKI */ 1561da177e4SLinus Torvalds if (!cap_sys_admin) 1571da177e4SLinus Torvalds n -= n / 32; 1581da177e4SLinus Torvalds free += n; 1591da177e4SLinus Torvalds 1601da177e4SLinus Torvalds if (free > pages) 1611da177e4SLinus Torvalds return 0; 1626d9f7839SHideo AOKI 1636d9f7839SHideo AOKI goto error; 1641da177e4SLinus Torvalds } 1651da177e4SLinus Torvalds 1661da177e4SLinus Torvalds allowed = (totalram_pages - hugetlb_total_pages()) 1671da177e4SLinus Torvalds * sysctl_overcommit_ratio / 100; 1681da177e4SLinus Torvalds /* 1691da177e4SLinus Torvalds * Leave the last 3% for root 1701da177e4SLinus Torvalds */ 1711da177e4SLinus Torvalds if (!cap_sys_admin) 1721da177e4SLinus Torvalds allowed -= allowed / 32; 1731da177e4SLinus Torvalds allowed += total_swap_pages; 1741da177e4SLinus Torvalds 1751da177e4SLinus Torvalds /* Don't let a single process grow too big: 1761da177e4SLinus Torvalds leave 3% of the size of this process for other processes */ 17734b4e4aaSAlan Cox allowed -= mm->total_vm / 32; 1781da177e4SLinus Torvalds 1792f60f8d3SSimon Derr /* 1802f60f8d3SSimon Derr * cast `allowed' as a signed long because vm_committed_space 1812f60f8d3SSimon Derr * sometimes has a negative value 1822f60f8d3SSimon Derr */ 18380119ef5SAlan Cox if (atomic_long_read(&vm_committed_space) < (long)allowed) 1841da177e4SLinus Torvalds return 0; 1856d9f7839SHideo AOKI error: 1861da177e4SLinus Torvalds vm_unacct_memory(pages); 1871da177e4SLinus Torvalds 1881da177e4SLinus Torvalds return -ENOMEM; 1891da177e4SLinus Torvalds } 1901da177e4SLinus Torvalds 1911da177e4SLinus Torvalds /* 1921da177e4SLinus Torvalds * Requires inode->i_mapping->i_mmap_lock 1931da177e4SLinus Torvalds */ 1941da177e4SLinus Torvalds static void __remove_shared_vm_struct(struct vm_area_struct *vma, 1951da177e4SLinus Torvalds struct file *file, struct address_space *mapping) 1961da177e4SLinus Torvalds { 1971da177e4SLinus Torvalds if (vma->vm_flags & VM_DENYWRITE) 198d3ac7f89SJosef "Jeff" Sipek atomic_inc(&file->f_path.dentry->d_inode->i_writecount); 1991da177e4SLinus Torvalds if (vma->vm_flags & VM_SHARED) 2001da177e4SLinus Torvalds mapping->i_mmap_writable--; 2011da177e4SLinus Torvalds 2021da177e4SLinus Torvalds flush_dcache_mmap_lock(mapping); 2031da177e4SLinus Torvalds if (unlikely(vma->vm_flags & VM_NONLINEAR)) 2041da177e4SLinus Torvalds list_del_init(&vma->shared.vm_set.list); 2051da177e4SLinus Torvalds else 2061da177e4SLinus Torvalds vma_prio_tree_remove(vma, &mapping->i_mmap); 2071da177e4SLinus Torvalds flush_dcache_mmap_unlock(mapping); 2081da177e4SLinus Torvalds } 2091da177e4SLinus Torvalds 2101da177e4SLinus Torvalds /* 211a8fb5618SHugh Dickins * Unlink a file-based vm structure from its prio_tree, to hide 212a8fb5618SHugh Dickins * vma from rmap and vmtruncate before freeing its page tables. 2131da177e4SLinus Torvalds */ 214a8fb5618SHugh Dickins void unlink_file_vma(struct vm_area_struct *vma) 2151da177e4SLinus Torvalds { 2161da177e4SLinus Torvalds struct file *file = vma->vm_file; 2171da177e4SLinus Torvalds 2181da177e4SLinus Torvalds if (file) { 2191da177e4SLinus Torvalds struct address_space *mapping = file->f_mapping; 2201da177e4SLinus Torvalds spin_lock(&mapping->i_mmap_lock); 2211da177e4SLinus Torvalds __remove_shared_vm_struct(vma, file, mapping); 2221da177e4SLinus Torvalds spin_unlock(&mapping->i_mmap_lock); 2231da177e4SLinus Torvalds } 224a8fb5618SHugh Dickins } 225a8fb5618SHugh Dickins 226a8fb5618SHugh Dickins /* 227a8fb5618SHugh Dickins * Close a vm structure and free it, returning the next. 228a8fb5618SHugh Dickins */ 229a8fb5618SHugh Dickins static struct vm_area_struct *remove_vma(struct vm_area_struct *vma) 230a8fb5618SHugh Dickins { 231a8fb5618SHugh Dickins struct vm_area_struct *next = vma->vm_next; 232a8fb5618SHugh Dickins 233a8fb5618SHugh Dickins might_sleep(); 2341da177e4SLinus Torvalds if (vma->vm_ops && vma->vm_ops->close) 2351da177e4SLinus Torvalds vma->vm_ops->close(vma); 236925d1c40SMatt Helsley if (vma->vm_file) { 237a8fb5618SHugh Dickins fput(vma->vm_file); 238925d1c40SMatt Helsley if (vma->vm_flags & VM_EXECUTABLE) 239925d1c40SMatt Helsley removed_exe_file_vma(vma->vm_mm); 240925d1c40SMatt Helsley } 241f0be3d32SLee Schermerhorn mpol_put(vma_policy(vma)); 2421da177e4SLinus Torvalds kmem_cache_free(vm_area_cachep, vma); 243a8fb5618SHugh Dickins return next; 2441da177e4SLinus Torvalds } 2451da177e4SLinus Torvalds 2461da177e4SLinus Torvalds asmlinkage unsigned long sys_brk(unsigned long brk) 2471da177e4SLinus Torvalds { 2481da177e4SLinus Torvalds unsigned long rlim, retval; 2491da177e4SLinus Torvalds unsigned long newbrk, oldbrk; 2501da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 251a5b4592cSJiri Kosina unsigned long min_brk; 2521da177e4SLinus Torvalds 2531da177e4SLinus Torvalds down_write(&mm->mmap_sem); 2541da177e4SLinus Torvalds 255a5b4592cSJiri Kosina #ifdef CONFIG_COMPAT_BRK 256a5b4592cSJiri Kosina min_brk = mm->end_code; 257a5b4592cSJiri Kosina #else 258a5b4592cSJiri Kosina min_brk = mm->start_brk; 259a5b4592cSJiri Kosina #endif 260a5b4592cSJiri Kosina if (brk < min_brk) 2611da177e4SLinus Torvalds goto out; 2621e624196SRam Gupta 2631e624196SRam Gupta /* 2641e624196SRam Gupta * Check against rlimit here. If this check is done later after the test 2651e624196SRam Gupta * of oldbrk with newbrk then it can escape the test and let the data 2661e624196SRam Gupta * segment grow beyond its set limit the in case where the limit is 2671e624196SRam Gupta * not page aligned -Ram Gupta 2681e624196SRam Gupta */ 2691e624196SRam Gupta rlim = current->signal->rlim[RLIMIT_DATA].rlim_cur; 270c1d171a0SJiri Kosina if (rlim < RLIM_INFINITY && (brk - mm->start_brk) + 271c1d171a0SJiri Kosina (mm->end_data - mm->start_data) > rlim) 2721e624196SRam Gupta goto out; 2731e624196SRam Gupta 2741da177e4SLinus Torvalds newbrk = PAGE_ALIGN(brk); 2751da177e4SLinus Torvalds oldbrk = PAGE_ALIGN(mm->brk); 2761da177e4SLinus Torvalds if (oldbrk == newbrk) 2771da177e4SLinus Torvalds goto set_brk; 2781da177e4SLinus Torvalds 2791da177e4SLinus Torvalds /* Always allow shrinking brk. */ 2801da177e4SLinus Torvalds if (brk <= mm->brk) { 2811da177e4SLinus Torvalds if (!do_munmap(mm, newbrk, oldbrk-newbrk)) 2821da177e4SLinus Torvalds goto set_brk; 2831da177e4SLinus Torvalds goto out; 2841da177e4SLinus Torvalds } 2851da177e4SLinus Torvalds 2861da177e4SLinus Torvalds /* Check against existing mmap mappings. */ 2871da177e4SLinus Torvalds if (find_vma_intersection(mm, oldbrk, newbrk+PAGE_SIZE)) 2881da177e4SLinus Torvalds goto out; 2891da177e4SLinus Torvalds 2901da177e4SLinus Torvalds /* Ok, looks good - let it rip. */ 2911da177e4SLinus Torvalds if (do_brk(oldbrk, newbrk-oldbrk) != oldbrk) 2921da177e4SLinus Torvalds goto out; 2931da177e4SLinus Torvalds set_brk: 2941da177e4SLinus Torvalds mm->brk = brk; 2951da177e4SLinus Torvalds out: 2961da177e4SLinus Torvalds retval = mm->brk; 2971da177e4SLinus Torvalds up_write(&mm->mmap_sem); 2981da177e4SLinus Torvalds return retval; 2991da177e4SLinus Torvalds } 3001da177e4SLinus Torvalds 3011da177e4SLinus Torvalds #ifdef DEBUG_MM_RB 3021da177e4SLinus Torvalds static int browse_rb(struct rb_root *root) 3031da177e4SLinus Torvalds { 3041da177e4SLinus Torvalds int i = 0, j; 3051da177e4SLinus Torvalds struct rb_node *nd, *pn = NULL; 3061da177e4SLinus Torvalds unsigned long prev = 0, pend = 0; 3071da177e4SLinus Torvalds 3081da177e4SLinus Torvalds for (nd = rb_first(root); nd; nd = rb_next(nd)) { 3091da177e4SLinus Torvalds struct vm_area_struct *vma; 3101da177e4SLinus Torvalds vma = rb_entry(nd, struct vm_area_struct, vm_rb); 3111da177e4SLinus Torvalds if (vma->vm_start < prev) 3121da177e4SLinus Torvalds printk("vm_start %lx prev %lx\n", vma->vm_start, prev), i = -1; 3131da177e4SLinus Torvalds if (vma->vm_start < pend) 3141da177e4SLinus Torvalds printk("vm_start %lx pend %lx\n", vma->vm_start, pend); 3151da177e4SLinus Torvalds if (vma->vm_start > vma->vm_end) 3161da177e4SLinus Torvalds printk("vm_end %lx < vm_start %lx\n", vma->vm_end, vma->vm_start); 3171da177e4SLinus Torvalds i++; 3181da177e4SLinus Torvalds pn = nd; 319d1af65d1SDavid Miller prev = vma->vm_start; 320d1af65d1SDavid Miller pend = vma->vm_end; 3211da177e4SLinus Torvalds } 3221da177e4SLinus Torvalds j = 0; 3231da177e4SLinus Torvalds for (nd = pn; nd; nd = rb_prev(nd)) { 3241da177e4SLinus Torvalds j++; 3251da177e4SLinus Torvalds } 3261da177e4SLinus Torvalds if (i != j) 3271da177e4SLinus Torvalds printk("backwards %d, forwards %d\n", j, i), i = 0; 3281da177e4SLinus Torvalds return i; 3291da177e4SLinus Torvalds } 3301da177e4SLinus Torvalds 3311da177e4SLinus Torvalds void validate_mm(struct mm_struct *mm) 3321da177e4SLinus Torvalds { 3331da177e4SLinus Torvalds int bug = 0; 3341da177e4SLinus Torvalds int i = 0; 3351da177e4SLinus Torvalds struct vm_area_struct *tmp = mm->mmap; 3361da177e4SLinus Torvalds while (tmp) { 3371da177e4SLinus Torvalds tmp = tmp->vm_next; 3381da177e4SLinus Torvalds i++; 3391da177e4SLinus Torvalds } 3401da177e4SLinus Torvalds if (i != mm->map_count) 3411da177e4SLinus Torvalds printk("map_count %d vm_next %d\n", mm->map_count, i), bug = 1; 3421da177e4SLinus Torvalds i = browse_rb(&mm->mm_rb); 3431da177e4SLinus Torvalds if (i != mm->map_count) 3441da177e4SLinus Torvalds printk("map_count %d rb %d\n", mm->map_count, i), bug = 1; 34546a350efSEric Sesterhenn BUG_ON(bug); 3461da177e4SLinus Torvalds } 3471da177e4SLinus Torvalds #else 3481da177e4SLinus Torvalds #define validate_mm(mm) do { } while (0) 3491da177e4SLinus Torvalds #endif 3501da177e4SLinus Torvalds 3511da177e4SLinus Torvalds static struct vm_area_struct * 3521da177e4SLinus Torvalds find_vma_prepare(struct mm_struct *mm, unsigned long addr, 3531da177e4SLinus Torvalds struct vm_area_struct **pprev, struct rb_node ***rb_link, 3541da177e4SLinus Torvalds struct rb_node ** rb_parent) 3551da177e4SLinus Torvalds { 3561da177e4SLinus Torvalds struct vm_area_struct * vma; 3571da177e4SLinus Torvalds struct rb_node ** __rb_link, * __rb_parent, * rb_prev; 3581da177e4SLinus Torvalds 3591da177e4SLinus Torvalds __rb_link = &mm->mm_rb.rb_node; 3601da177e4SLinus Torvalds rb_prev = __rb_parent = NULL; 3611da177e4SLinus Torvalds vma = NULL; 3621da177e4SLinus Torvalds 3631da177e4SLinus Torvalds while (*__rb_link) { 3641da177e4SLinus Torvalds struct vm_area_struct *vma_tmp; 3651da177e4SLinus Torvalds 3661da177e4SLinus Torvalds __rb_parent = *__rb_link; 3671da177e4SLinus Torvalds vma_tmp = rb_entry(__rb_parent, struct vm_area_struct, vm_rb); 3681da177e4SLinus Torvalds 3691da177e4SLinus Torvalds if (vma_tmp->vm_end > addr) { 3701da177e4SLinus Torvalds vma = vma_tmp; 3711da177e4SLinus Torvalds if (vma_tmp->vm_start <= addr) 3721da177e4SLinus Torvalds return vma; 3731da177e4SLinus Torvalds __rb_link = &__rb_parent->rb_left; 3741da177e4SLinus Torvalds } else { 3751da177e4SLinus Torvalds rb_prev = __rb_parent; 3761da177e4SLinus Torvalds __rb_link = &__rb_parent->rb_right; 3771da177e4SLinus Torvalds } 3781da177e4SLinus Torvalds } 3791da177e4SLinus Torvalds 3801da177e4SLinus Torvalds *pprev = NULL; 3811da177e4SLinus Torvalds if (rb_prev) 3821da177e4SLinus Torvalds *pprev = rb_entry(rb_prev, struct vm_area_struct, vm_rb); 3831da177e4SLinus Torvalds *rb_link = __rb_link; 3841da177e4SLinus Torvalds *rb_parent = __rb_parent; 3851da177e4SLinus Torvalds return vma; 3861da177e4SLinus Torvalds } 3871da177e4SLinus Torvalds 3881da177e4SLinus Torvalds static inline void 3891da177e4SLinus Torvalds __vma_link_list(struct mm_struct *mm, struct vm_area_struct *vma, 3901da177e4SLinus Torvalds struct vm_area_struct *prev, struct rb_node *rb_parent) 3911da177e4SLinus Torvalds { 3921da177e4SLinus Torvalds if (prev) { 3931da177e4SLinus Torvalds vma->vm_next = prev->vm_next; 3941da177e4SLinus Torvalds prev->vm_next = vma; 3951da177e4SLinus Torvalds } else { 3961da177e4SLinus Torvalds mm->mmap = vma; 3971da177e4SLinus Torvalds if (rb_parent) 3981da177e4SLinus Torvalds vma->vm_next = rb_entry(rb_parent, 3991da177e4SLinus Torvalds struct vm_area_struct, vm_rb); 4001da177e4SLinus Torvalds else 4011da177e4SLinus Torvalds vma->vm_next = NULL; 4021da177e4SLinus Torvalds } 4031da177e4SLinus Torvalds } 4041da177e4SLinus Torvalds 4051da177e4SLinus Torvalds void __vma_link_rb(struct mm_struct *mm, struct vm_area_struct *vma, 4061da177e4SLinus Torvalds struct rb_node **rb_link, struct rb_node *rb_parent) 4071da177e4SLinus Torvalds { 4081da177e4SLinus Torvalds rb_link_node(&vma->vm_rb, rb_parent, rb_link); 4091da177e4SLinus Torvalds rb_insert_color(&vma->vm_rb, &mm->mm_rb); 4101da177e4SLinus Torvalds } 4111da177e4SLinus Torvalds 4121da177e4SLinus Torvalds static inline void __vma_link_file(struct vm_area_struct *vma) 4131da177e4SLinus Torvalds { 4141da177e4SLinus Torvalds struct file * file; 4151da177e4SLinus Torvalds 4161da177e4SLinus Torvalds file = vma->vm_file; 4171da177e4SLinus Torvalds if (file) { 4181da177e4SLinus Torvalds struct address_space *mapping = file->f_mapping; 4191da177e4SLinus Torvalds 4201da177e4SLinus Torvalds if (vma->vm_flags & VM_DENYWRITE) 421d3ac7f89SJosef "Jeff" Sipek atomic_dec(&file->f_path.dentry->d_inode->i_writecount); 4221da177e4SLinus Torvalds if (vma->vm_flags & VM_SHARED) 4231da177e4SLinus Torvalds mapping->i_mmap_writable++; 4241da177e4SLinus Torvalds 4251da177e4SLinus Torvalds flush_dcache_mmap_lock(mapping); 4261da177e4SLinus Torvalds if (unlikely(vma->vm_flags & VM_NONLINEAR)) 4271da177e4SLinus Torvalds vma_nonlinear_insert(vma, &mapping->i_mmap_nonlinear); 4281da177e4SLinus Torvalds else 4291da177e4SLinus Torvalds vma_prio_tree_insert(vma, &mapping->i_mmap); 4301da177e4SLinus Torvalds flush_dcache_mmap_unlock(mapping); 4311da177e4SLinus Torvalds } 4321da177e4SLinus Torvalds } 4331da177e4SLinus Torvalds 4341da177e4SLinus Torvalds static void 4351da177e4SLinus Torvalds __vma_link(struct mm_struct *mm, struct vm_area_struct *vma, 4361da177e4SLinus Torvalds struct vm_area_struct *prev, struct rb_node **rb_link, 4371da177e4SLinus Torvalds struct rb_node *rb_parent) 4381da177e4SLinus Torvalds { 4391da177e4SLinus Torvalds __vma_link_list(mm, vma, prev, rb_parent); 4401da177e4SLinus Torvalds __vma_link_rb(mm, vma, rb_link, rb_parent); 4411da177e4SLinus Torvalds __anon_vma_link(vma); 4421da177e4SLinus Torvalds } 4431da177e4SLinus Torvalds 4441da177e4SLinus Torvalds static void vma_link(struct mm_struct *mm, struct vm_area_struct *vma, 4451da177e4SLinus Torvalds struct vm_area_struct *prev, struct rb_node **rb_link, 4461da177e4SLinus Torvalds struct rb_node *rb_parent) 4471da177e4SLinus Torvalds { 4481da177e4SLinus Torvalds struct address_space *mapping = NULL; 4491da177e4SLinus Torvalds 4501da177e4SLinus Torvalds if (vma->vm_file) 4511da177e4SLinus Torvalds mapping = vma->vm_file->f_mapping; 4521da177e4SLinus Torvalds 4531da177e4SLinus Torvalds if (mapping) { 4541da177e4SLinus Torvalds spin_lock(&mapping->i_mmap_lock); 4551da177e4SLinus Torvalds vma->vm_truncate_count = mapping->truncate_count; 4561da177e4SLinus Torvalds } 4571da177e4SLinus Torvalds anon_vma_lock(vma); 4581da177e4SLinus Torvalds 4591da177e4SLinus Torvalds __vma_link(mm, vma, prev, rb_link, rb_parent); 4601da177e4SLinus Torvalds __vma_link_file(vma); 4611da177e4SLinus Torvalds 4621da177e4SLinus Torvalds anon_vma_unlock(vma); 4631da177e4SLinus Torvalds if (mapping) 4641da177e4SLinus Torvalds spin_unlock(&mapping->i_mmap_lock); 4651da177e4SLinus Torvalds 4661da177e4SLinus Torvalds mm->map_count++; 4671da177e4SLinus Torvalds validate_mm(mm); 4681da177e4SLinus Torvalds } 4691da177e4SLinus Torvalds 4701da177e4SLinus Torvalds /* 4711da177e4SLinus Torvalds * Helper for vma_adjust in the split_vma insert case: 4721da177e4SLinus Torvalds * insert vm structure into list and rbtree and anon_vma, 4731da177e4SLinus Torvalds * but it has already been inserted into prio_tree earlier. 4741da177e4SLinus Torvalds */ 4751da177e4SLinus Torvalds static void 4761da177e4SLinus Torvalds __insert_vm_struct(struct mm_struct * mm, struct vm_area_struct * vma) 4771da177e4SLinus Torvalds { 4781da177e4SLinus Torvalds struct vm_area_struct * __vma, * prev; 4791da177e4SLinus Torvalds struct rb_node ** rb_link, * rb_parent; 4801da177e4SLinus Torvalds 4811da177e4SLinus Torvalds __vma = find_vma_prepare(mm, vma->vm_start,&prev, &rb_link, &rb_parent); 48246a350efSEric Sesterhenn BUG_ON(__vma && __vma->vm_start < vma->vm_end); 4831da177e4SLinus Torvalds __vma_link(mm, vma, prev, rb_link, rb_parent); 4841da177e4SLinus Torvalds mm->map_count++; 4851da177e4SLinus Torvalds } 4861da177e4SLinus Torvalds 4871da177e4SLinus Torvalds static inline void 4881da177e4SLinus Torvalds __vma_unlink(struct mm_struct *mm, struct vm_area_struct *vma, 4891da177e4SLinus Torvalds struct vm_area_struct *prev) 4901da177e4SLinus Torvalds { 4911da177e4SLinus Torvalds prev->vm_next = vma->vm_next; 4921da177e4SLinus Torvalds rb_erase(&vma->vm_rb, &mm->mm_rb); 4931da177e4SLinus Torvalds if (mm->mmap_cache == vma) 4941da177e4SLinus Torvalds mm->mmap_cache = prev; 4951da177e4SLinus Torvalds } 4961da177e4SLinus Torvalds 4971da177e4SLinus Torvalds /* 4981da177e4SLinus Torvalds * We cannot adjust vm_start, vm_end, vm_pgoff fields of a vma that 4991da177e4SLinus Torvalds * is already present in an i_mmap tree without adjusting the tree. 5001da177e4SLinus Torvalds * The following helper function should be used when such adjustments 5011da177e4SLinus Torvalds * are necessary. The "insert" vma (if any) is to be inserted 5021da177e4SLinus Torvalds * before we drop the necessary locks. 5031da177e4SLinus Torvalds */ 5041da177e4SLinus Torvalds void vma_adjust(struct vm_area_struct *vma, unsigned long start, 5051da177e4SLinus Torvalds unsigned long end, pgoff_t pgoff, struct vm_area_struct *insert) 5061da177e4SLinus Torvalds { 5071da177e4SLinus Torvalds struct mm_struct *mm = vma->vm_mm; 5081da177e4SLinus Torvalds struct vm_area_struct *next = vma->vm_next; 5091da177e4SLinus Torvalds struct vm_area_struct *importer = NULL; 5101da177e4SLinus Torvalds struct address_space *mapping = NULL; 5111da177e4SLinus Torvalds struct prio_tree_root *root = NULL; 5121da177e4SLinus Torvalds struct file *file = vma->vm_file; 5131da177e4SLinus Torvalds struct anon_vma *anon_vma = NULL; 5141da177e4SLinus Torvalds long adjust_next = 0; 5151da177e4SLinus Torvalds int remove_next = 0; 5161da177e4SLinus Torvalds 5171da177e4SLinus Torvalds if (next && !insert) { 5181da177e4SLinus Torvalds if (end >= next->vm_end) { 5191da177e4SLinus Torvalds /* 5201da177e4SLinus Torvalds * vma expands, overlapping all the next, and 5211da177e4SLinus Torvalds * perhaps the one after too (mprotect case 6). 5221da177e4SLinus Torvalds */ 5231da177e4SLinus Torvalds again: remove_next = 1 + (end > next->vm_end); 5241da177e4SLinus Torvalds end = next->vm_end; 5251da177e4SLinus Torvalds anon_vma = next->anon_vma; 5261da177e4SLinus Torvalds importer = vma; 5271da177e4SLinus Torvalds } else if (end > next->vm_start) { 5281da177e4SLinus Torvalds /* 5291da177e4SLinus Torvalds * vma expands, overlapping part of the next: 5301da177e4SLinus Torvalds * mprotect case 5 shifting the boundary up. 5311da177e4SLinus Torvalds */ 5321da177e4SLinus Torvalds adjust_next = (end - next->vm_start) >> PAGE_SHIFT; 5331da177e4SLinus Torvalds anon_vma = next->anon_vma; 5341da177e4SLinus Torvalds importer = vma; 5351da177e4SLinus Torvalds } else if (end < vma->vm_end) { 5361da177e4SLinus Torvalds /* 5371da177e4SLinus Torvalds * vma shrinks, and !insert tells it's not 5381da177e4SLinus Torvalds * split_vma inserting another: so it must be 5391da177e4SLinus Torvalds * mprotect case 4 shifting the boundary down. 5401da177e4SLinus Torvalds */ 5411da177e4SLinus Torvalds adjust_next = - ((vma->vm_end - end) >> PAGE_SHIFT); 5421da177e4SLinus Torvalds anon_vma = next->anon_vma; 5431da177e4SLinus Torvalds importer = next; 5441da177e4SLinus Torvalds } 5451da177e4SLinus Torvalds } 5461da177e4SLinus Torvalds 5471da177e4SLinus Torvalds if (file) { 5481da177e4SLinus Torvalds mapping = file->f_mapping; 5491da177e4SLinus Torvalds if (!(vma->vm_flags & VM_NONLINEAR)) 5501da177e4SLinus Torvalds root = &mapping->i_mmap; 5511da177e4SLinus Torvalds spin_lock(&mapping->i_mmap_lock); 5521da177e4SLinus Torvalds if (importer && 5531da177e4SLinus Torvalds vma->vm_truncate_count != next->vm_truncate_count) { 5541da177e4SLinus Torvalds /* 5551da177e4SLinus Torvalds * unmap_mapping_range might be in progress: 5561da177e4SLinus Torvalds * ensure that the expanding vma is rescanned. 5571da177e4SLinus Torvalds */ 5581da177e4SLinus Torvalds importer->vm_truncate_count = 0; 5591da177e4SLinus Torvalds } 5601da177e4SLinus Torvalds if (insert) { 5611da177e4SLinus Torvalds insert->vm_truncate_count = vma->vm_truncate_count; 5621da177e4SLinus Torvalds /* 5631da177e4SLinus Torvalds * Put into prio_tree now, so instantiated pages 5641da177e4SLinus Torvalds * are visible to arm/parisc __flush_dcache_page 5651da177e4SLinus Torvalds * throughout; but we cannot insert into address 5661da177e4SLinus Torvalds * space until vma start or end is updated. 5671da177e4SLinus Torvalds */ 5681da177e4SLinus Torvalds __vma_link_file(insert); 5691da177e4SLinus Torvalds } 5701da177e4SLinus Torvalds } 5711da177e4SLinus Torvalds 5721da177e4SLinus Torvalds /* 5731da177e4SLinus Torvalds * When changing only vma->vm_end, we don't really need 5741da177e4SLinus Torvalds * anon_vma lock: but is that case worth optimizing out? 5751da177e4SLinus Torvalds */ 5761da177e4SLinus Torvalds if (vma->anon_vma) 5771da177e4SLinus Torvalds anon_vma = vma->anon_vma; 5781da177e4SLinus Torvalds if (anon_vma) { 5791da177e4SLinus Torvalds spin_lock(&anon_vma->lock); 5801da177e4SLinus Torvalds /* 5811da177e4SLinus Torvalds * Easily overlooked: when mprotect shifts the boundary, 5821da177e4SLinus Torvalds * make sure the expanding vma has anon_vma set if the 5831da177e4SLinus Torvalds * shrinking vma had, to cover any anon pages imported. 5841da177e4SLinus Torvalds */ 5851da177e4SLinus Torvalds if (importer && !importer->anon_vma) { 5861da177e4SLinus Torvalds importer->anon_vma = anon_vma; 5871da177e4SLinus Torvalds __anon_vma_link(importer); 5881da177e4SLinus Torvalds } 5891da177e4SLinus Torvalds } 5901da177e4SLinus Torvalds 5911da177e4SLinus Torvalds if (root) { 5921da177e4SLinus Torvalds flush_dcache_mmap_lock(mapping); 5931da177e4SLinus Torvalds vma_prio_tree_remove(vma, root); 5941da177e4SLinus Torvalds if (adjust_next) 5951da177e4SLinus Torvalds vma_prio_tree_remove(next, root); 5961da177e4SLinus Torvalds } 5971da177e4SLinus Torvalds 5981da177e4SLinus Torvalds vma->vm_start = start; 5991da177e4SLinus Torvalds vma->vm_end = end; 6001da177e4SLinus Torvalds vma->vm_pgoff = pgoff; 6011da177e4SLinus Torvalds if (adjust_next) { 6021da177e4SLinus Torvalds next->vm_start += adjust_next << PAGE_SHIFT; 6031da177e4SLinus Torvalds next->vm_pgoff += adjust_next; 6041da177e4SLinus Torvalds } 6051da177e4SLinus Torvalds 6061da177e4SLinus Torvalds if (root) { 6071da177e4SLinus Torvalds if (adjust_next) 6081da177e4SLinus Torvalds vma_prio_tree_insert(next, root); 6091da177e4SLinus Torvalds vma_prio_tree_insert(vma, root); 6101da177e4SLinus Torvalds flush_dcache_mmap_unlock(mapping); 6111da177e4SLinus Torvalds } 6121da177e4SLinus Torvalds 6131da177e4SLinus Torvalds if (remove_next) { 6141da177e4SLinus Torvalds /* 6151da177e4SLinus Torvalds * vma_merge has merged next into vma, and needs 6161da177e4SLinus Torvalds * us to remove next before dropping the locks. 6171da177e4SLinus Torvalds */ 6181da177e4SLinus Torvalds __vma_unlink(mm, next, vma); 6191da177e4SLinus Torvalds if (file) 6201da177e4SLinus Torvalds __remove_shared_vm_struct(next, file, mapping); 6211da177e4SLinus Torvalds if (next->anon_vma) 6221da177e4SLinus Torvalds __anon_vma_merge(vma, next); 6231da177e4SLinus Torvalds } else if (insert) { 6241da177e4SLinus Torvalds /* 6251da177e4SLinus Torvalds * split_vma has split insert from vma, and needs 6261da177e4SLinus Torvalds * us to insert it before dropping the locks 6271da177e4SLinus Torvalds * (it may either follow vma or precede it). 6281da177e4SLinus Torvalds */ 6291da177e4SLinus Torvalds __insert_vm_struct(mm, insert); 6301da177e4SLinus Torvalds } 6311da177e4SLinus Torvalds 6321da177e4SLinus Torvalds if (anon_vma) 6331da177e4SLinus Torvalds spin_unlock(&anon_vma->lock); 6341da177e4SLinus Torvalds if (mapping) 6351da177e4SLinus Torvalds spin_unlock(&mapping->i_mmap_lock); 6361da177e4SLinus Torvalds 6371da177e4SLinus Torvalds if (remove_next) { 638925d1c40SMatt Helsley if (file) { 6391da177e4SLinus Torvalds fput(file); 640925d1c40SMatt Helsley if (next->vm_flags & VM_EXECUTABLE) 641925d1c40SMatt Helsley removed_exe_file_vma(mm); 642925d1c40SMatt Helsley } 6431da177e4SLinus Torvalds mm->map_count--; 644f0be3d32SLee Schermerhorn mpol_put(vma_policy(next)); 6451da177e4SLinus Torvalds kmem_cache_free(vm_area_cachep, next); 6461da177e4SLinus Torvalds /* 6471da177e4SLinus Torvalds * In mprotect's case 6 (see comments on vma_merge), 6481da177e4SLinus Torvalds * we must remove another next too. It would clutter 6491da177e4SLinus Torvalds * up the code too much to do both in one go. 6501da177e4SLinus Torvalds */ 6511da177e4SLinus Torvalds if (remove_next == 2) { 6521da177e4SLinus Torvalds next = vma->vm_next; 6531da177e4SLinus Torvalds goto again; 6541da177e4SLinus Torvalds } 6551da177e4SLinus Torvalds } 6561da177e4SLinus Torvalds 6571da177e4SLinus Torvalds validate_mm(mm); 6581da177e4SLinus Torvalds } 6591da177e4SLinus Torvalds 6601da177e4SLinus Torvalds /* 6611da177e4SLinus Torvalds * If the vma has a ->close operation then the driver probably needs to release 6621da177e4SLinus Torvalds * per-vma resources, so we don't attempt to merge those. 6631da177e4SLinus Torvalds */ 664a6f563dbSHugh Dickins #define VM_SPECIAL (VM_IO | VM_DONTEXPAND | VM_RESERVED | VM_PFNMAP) 6651da177e4SLinus Torvalds 6661da177e4SLinus Torvalds static inline int is_mergeable_vma(struct vm_area_struct *vma, 6671da177e4SLinus Torvalds struct file *file, unsigned long vm_flags) 6681da177e4SLinus Torvalds { 6691da177e4SLinus Torvalds if (vma->vm_flags != vm_flags) 6701da177e4SLinus Torvalds return 0; 6711da177e4SLinus Torvalds if (vma->vm_file != file) 6721da177e4SLinus Torvalds return 0; 6731da177e4SLinus Torvalds if (vma->vm_ops && vma->vm_ops->close) 6741da177e4SLinus Torvalds return 0; 6751da177e4SLinus Torvalds return 1; 6761da177e4SLinus Torvalds } 6771da177e4SLinus Torvalds 6781da177e4SLinus Torvalds static inline int is_mergeable_anon_vma(struct anon_vma *anon_vma1, 6791da177e4SLinus Torvalds struct anon_vma *anon_vma2) 6801da177e4SLinus Torvalds { 6811da177e4SLinus Torvalds return !anon_vma1 || !anon_vma2 || (anon_vma1 == anon_vma2); 6821da177e4SLinus Torvalds } 6831da177e4SLinus Torvalds 6841da177e4SLinus Torvalds /* 6851da177e4SLinus Torvalds * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff) 6861da177e4SLinus Torvalds * in front of (at a lower virtual address and file offset than) the vma. 6871da177e4SLinus Torvalds * 6881da177e4SLinus Torvalds * We cannot merge two vmas if they have differently assigned (non-NULL) 6891da177e4SLinus Torvalds * anon_vmas, nor if same anon_vma is assigned but offsets incompatible. 6901da177e4SLinus Torvalds * 6911da177e4SLinus Torvalds * We don't check here for the merged mmap wrapping around the end of pagecache 6921da177e4SLinus Torvalds * indices (16TB on ia32) because do_mmap_pgoff() does not permit mmap's which 6931da177e4SLinus Torvalds * wrap, nor mmaps which cover the final page at index -1UL. 6941da177e4SLinus Torvalds */ 6951da177e4SLinus Torvalds static int 6961da177e4SLinus Torvalds can_vma_merge_before(struct vm_area_struct *vma, unsigned long vm_flags, 6971da177e4SLinus Torvalds struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff) 6981da177e4SLinus Torvalds { 6991da177e4SLinus Torvalds if (is_mergeable_vma(vma, file, vm_flags) && 7001da177e4SLinus Torvalds is_mergeable_anon_vma(anon_vma, vma->anon_vma)) { 7011da177e4SLinus Torvalds if (vma->vm_pgoff == vm_pgoff) 7021da177e4SLinus Torvalds return 1; 7031da177e4SLinus Torvalds } 7041da177e4SLinus Torvalds return 0; 7051da177e4SLinus Torvalds } 7061da177e4SLinus Torvalds 7071da177e4SLinus Torvalds /* 7081da177e4SLinus Torvalds * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff) 7091da177e4SLinus Torvalds * beyond (at a higher virtual address and file offset than) the vma. 7101da177e4SLinus Torvalds * 7111da177e4SLinus Torvalds * We cannot merge two vmas if they have differently assigned (non-NULL) 7121da177e4SLinus Torvalds * anon_vmas, nor if same anon_vma is assigned but offsets incompatible. 7131da177e4SLinus Torvalds */ 7141da177e4SLinus Torvalds static int 7151da177e4SLinus Torvalds can_vma_merge_after(struct vm_area_struct *vma, unsigned long vm_flags, 7161da177e4SLinus Torvalds struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff) 7171da177e4SLinus Torvalds { 7181da177e4SLinus Torvalds if (is_mergeable_vma(vma, file, vm_flags) && 7191da177e4SLinus Torvalds is_mergeable_anon_vma(anon_vma, vma->anon_vma)) { 7201da177e4SLinus Torvalds pgoff_t vm_pglen; 7211da177e4SLinus Torvalds vm_pglen = (vma->vm_end - vma->vm_start) >> PAGE_SHIFT; 7221da177e4SLinus Torvalds if (vma->vm_pgoff + vm_pglen == vm_pgoff) 7231da177e4SLinus Torvalds return 1; 7241da177e4SLinus Torvalds } 7251da177e4SLinus Torvalds return 0; 7261da177e4SLinus Torvalds } 7271da177e4SLinus Torvalds 7281da177e4SLinus Torvalds /* 7291da177e4SLinus Torvalds * Given a mapping request (addr,end,vm_flags,file,pgoff), figure out 7301da177e4SLinus Torvalds * whether that can be merged with its predecessor or its successor. 7311da177e4SLinus Torvalds * Or both (it neatly fills a hole). 7321da177e4SLinus Torvalds * 7331da177e4SLinus Torvalds * In most cases - when called for mmap, brk or mremap - [addr,end) is 7341da177e4SLinus Torvalds * certain not to be mapped by the time vma_merge is called; but when 7351da177e4SLinus Torvalds * called for mprotect, it is certain to be already mapped (either at 7361da177e4SLinus Torvalds * an offset within prev, or at the start of next), and the flags of 7371da177e4SLinus Torvalds * this area are about to be changed to vm_flags - and the no-change 7381da177e4SLinus Torvalds * case has already been eliminated. 7391da177e4SLinus Torvalds * 7401da177e4SLinus Torvalds * The following mprotect cases have to be considered, where AAAA is 7411da177e4SLinus Torvalds * the area passed down from mprotect_fixup, never extending beyond one 7421da177e4SLinus Torvalds * vma, PPPPPP is the prev vma specified, and NNNNNN the next vma after: 7431da177e4SLinus Torvalds * 7441da177e4SLinus Torvalds * AAAA AAAA AAAA AAAA 7451da177e4SLinus Torvalds * PPPPPPNNNNNN PPPPPPNNNNNN PPPPPPNNNNNN PPPPNNNNXXXX 7461da177e4SLinus Torvalds * cannot merge might become might become might become 7471da177e4SLinus Torvalds * PPNNNNNNNNNN PPPPPPPPPPNN PPPPPPPPPPPP 6 or 7481da177e4SLinus Torvalds * mmap, brk or case 4 below case 5 below PPPPPPPPXXXX 7 or 7491da177e4SLinus Torvalds * mremap move: PPPPNNNNNNNN 8 7501da177e4SLinus Torvalds * AAAA 7511da177e4SLinus Torvalds * PPPP NNNN PPPPPPPPPPPP PPPPPPPPNNNN PPPPNNNNNNNN 7521da177e4SLinus Torvalds * might become case 1 below case 2 below case 3 below 7531da177e4SLinus Torvalds * 7541da177e4SLinus Torvalds * Odd one out? Case 8, because it extends NNNN but needs flags of XXXX: 7551da177e4SLinus Torvalds * mprotect_fixup updates vm_flags & vm_page_prot on successful return. 7561da177e4SLinus Torvalds */ 7571da177e4SLinus Torvalds struct vm_area_struct *vma_merge(struct mm_struct *mm, 7581da177e4SLinus Torvalds struct vm_area_struct *prev, unsigned long addr, 7591da177e4SLinus Torvalds unsigned long end, unsigned long vm_flags, 7601da177e4SLinus Torvalds struct anon_vma *anon_vma, struct file *file, 7611da177e4SLinus Torvalds pgoff_t pgoff, struct mempolicy *policy) 7621da177e4SLinus Torvalds { 7631da177e4SLinus Torvalds pgoff_t pglen = (end - addr) >> PAGE_SHIFT; 7641da177e4SLinus Torvalds struct vm_area_struct *area, *next; 7651da177e4SLinus Torvalds 7661da177e4SLinus Torvalds /* 7671da177e4SLinus Torvalds * We later require that vma->vm_flags == vm_flags, 7681da177e4SLinus Torvalds * so this tests vma->vm_flags & VM_SPECIAL, too. 7691da177e4SLinus Torvalds */ 7701da177e4SLinus Torvalds if (vm_flags & VM_SPECIAL) 7711da177e4SLinus Torvalds return NULL; 7721da177e4SLinus Torvalds 7731da177e4SLinus Torvalds if (prev) 7741da177e4SLinus Torvalds next = prev->vm_next; 7751da177e4SLinus Torvalds else 7761da177e4SLinus Torvalds next = mm->mmap; 7771da177e4SLinus Torvalds area = next; 7781da177e4SLinus Torvalds if (next && next->vm_end == end) /* cases 6, 7, 8 */ 7791da177e4SLinus Torvalds next = next->vm_next; 7801da177e4SLinus Torvalds 7811da177e4SLinus Torvalds /* 7821da177e4SLinus Torvalds * Can it merge with the predecessor? 7831da177e4SLinus Torvalds */ 7841da177e4SLinus Torvalds if (prev && prev->vm_end == addr && 7851da177e4SLinus Torvalds mpol_equal(vma_policy(prev), policy) && 7861da177e4SLinus Torvalds can_vma_merge_after(prev, vm_flags, 7871da177e4SLinus Torvalds anon_vma, file, pgoff)) { 7881da177e4SLinus Torvalds /* 7891da177e4SLinus Torvalds * OK, it can. Can we now merge in the successor as well? 7901da177e4SLinus Torvalds */ 7911da177e4SLinus Torvalds if (next && end == next->vm_start && 7921da177e4SLinus Torvalds mpol_equal(policy, vma_policy(next)) && 7931da177e4SLinus Torvalds can_vma_merge_before(next, vm_flags, 7941da177e4SLinus Torvalds anon_vma, file, pgoff+pglen) && 7951da177e4SLinus Torvalds is_mergeable_anon_vma(prev->anon_vma, 7961da177e4SLinus Torvalds next->anon_vma)) { 7971da177e4SLinus Torvalds /* cases 1, 6 */ 7981da177e4SLinus Torvalds vma_adjust(prev, prev->vm_start, 7991da177e4SLinus Torvalds next->vm_end, prev->vm_pgoff, NULL); 8001da177e4SLinus Torvalds } else /* cases 2, 5, 7 */ 8011da177e4SLinus Torvalds vma_adjust(prev, prev->vm_start, 8021da177e4SLinus Torvalds end, prev->vm_pgoff, NULL); 8031da177e4SLinus Torvalds return prev; 8041da177e4SLinus Torvalds } 8051da177e4SLinus Torvalds 8061da177e4SLinus Torvalds /* 8071da177e4SLinus Torvalds * Can this new request be merged in front of next? 8081da177e4SLinus Torvalds */ 8091da177e4SLinus Torvalds if (next && end == next->vm_start && 8101da177e4SLinus Torvalds mpol_equal(policy, vma_policy(next)) && 8111da177e4SLinus Torvalds can_vma_merge_before(next, vm_flags, 8121da177e4SLinus Torvalds anon_vma, file, pgoff+pglen)) { 8131da177e4SLinus Torvalds if (prev && addr < prev->vm_end) /* case 4 */ 8141da177e4SLinus Torvalds vma_adjust(prev, prev->vm_start, 8151da177e4SLinus Torvalds addr, prev->vm_pgoff, NULL); 8161da177e4SLinus Torvalds else /* cases 3, 8 */ 8171da177e4SLinus Torvalds vma_adjust(area, addr, next->vm_end, 8181da177e4SLinus Torvalds next->vm_pgoff - pglen, NULL); 8191da177e4SLinus Torvalds return area; 8201da177e4SLinus Torvalds } 8211da177e4SLinus Torvalds 8221da177e4SLinus Torvalds return NULL; 8231da177e4SLinus Torvalds } 8241da177e4SLinus Torvalds 8251da177e4SLinus Torvalds /* 8261da177e4SLinus Torvalds * find_mergeable_anon_vma is used by anon_vma_prepare, to check 8271da177e4SLinus Torvalds * neighbouring vmas for a suitable anon_vma, before it goes off 8281da177e4SLinus Torvalds * to allocate a new anon_vma. It checks because a repetitive 8291da177e4SLinus Torvalds * sequence of mprotects and faults may otherwise lead to distinct 8301da177e4SLinus Torvalds * anon_vmas being allocated, preventing vma merge in subsequent 8311da177e4SLinus Torvalds * mprotect. 8321da177e4SLinus Torvalds */ 8331da177e4SLinus Torvalds struct anon_vma *find_mergeable_anon_vma(struct vm_area_struct *vma) 8341da177e4SLinus Torvalds { 8351da177e4SLinus Torvalds struct vm_area_struct *near; 8361da177e4SLinus Torvalds unsigned long vm_flags; 8371da177e4SLinus Torvalds 8381da177e4SLinus Torvalds near = vma->vm_next; 8391da177e4SLinus Torvalds if (!near) 8401da177e4SLinus Torvalds goto try_prev; 8411da177e4SLinus Torvalds 8421da177e4SLinus Torvalds /* 8431da177e4SLinus Torvalds * Since only mprotect tries to remerge vmas, match flags 8441da177e4SLinus Torvalds * which might be mprotected into each other later on. 8451da177e4SLinus Torvalds * Neither mlock nor madvise tries to remerge at present, 8461da177e4SLinus Torvalds * so leave their flags as obstructing a merge. 8471da177e4SLinus Torvalds */ 8481da177e4SLinus Torvalds vm_flags = vma->vm_flags & ~(VM_READ|VM_WRITE|VM_EXEC); 8491da177e4SLinus Torvalds vm_flags |= near->vm_flags & (VM_READ|VM_WRITE|VM_EXEC); 8501da177e4SLinus Torvalds 8511da177e4SLinus Torvalds if (near->anon_vma && vma->vm_end == near->vm_start && 8521da177e4SLinus Torvalds mpol_equal(vma_policy(vma), vma_policy(near)) && 8531da177e4SLinus Torvalds can_vma_merge_before(near, vm_flags, 8541da177e4SLinus Torvalds NULL, vma->vm_file, vma->vm_pgoff + 8551da177e4SLinus Torvalds ((vma->vm_end - vma->vm_start) >> PAGE_SHIFT))) 8561da177e4SLinus Torvalds return near->anon_vma; 8571da177e4SLinus Torvalds try_prev: 8581da177e4SLinus Torvalds /* 8591da177e4SLinus Torvalds * It is potentially slow to have to call find_vma_prev here. 8601da177e4SLinus Torvalds * But it's only on the first write fault on the vma, not 8611da177e4SLinus Torvalds * every time, and we could devise a way to avoid it later 8621da177e4SLinus Torvalds * (e.g. stash info in next's anon_vma_node when assigning 8631da177e4SLinus Torvalds * an anon_vma, or when trying vma_merge). Another time. 8641da177e4SLinus Torvalds */ 86546a350efSEric Sesterhenn BUG_ON(find_vma_prev(vma->vm_mm, vma->vm_start, &near) != vma); 8661da177e4SLinus Torvalds if (!near) 8671da177e4SLinus Torvalds goto none; 8681da177e4SLinus Torvalds 8691da177e4SLinus Torvalds vm_flags = vma->vm_flags & ~(VM_READ|VM_WRITE|VM_EXEC); 8701da177e4SLinus Torvalds vm_flags |= near->vm_flags & (VM_READ|VM_WRITE|VM_EXEC); 8711da177e4SLinus Torvalds 8721da177e4SLinus Torvalds if (near->anon_vma && near->vm_end == vma->vm_start && 8731da177e4SLinus Torvalds mpol_equal(vma_policy(near), vma_policy(vma)) && 8741da177e4SLinus Torvalds can_vma_merge_after(near, vm_flags, 8751da177e4SLinus Torvalds NULL, vma->vm_file, vma->vm_pgoff)) 8761da177e4SLinus Torvalds return near->anon_vma; 8771da177e4SLinus Torvalds none: 8781da177e4SLinus Torvalds /* 8791da177e4SLinus Torvalds * There's no absolute need to look only at touching neighbours: 8801da177e4SLinus Torvalds * we could search further afield for "compatible" anon_vmas. 8811da177e4SLinus Torvalds * But it would probably just be a waste of time searching, 8821da177e4SLinus Torvalds * or lead to too many vmas hanging off the same anon_vma. 8831da177e4SLinus Torvalds * We're trying to allow mprotect remerging later on, 8841da177e4SLinus Torvalds * not trying to minimize memory used for anon_vmas. 8851da177e4SLinus Torvalds */ 8861da177e4SLinus Torvalds return NULL; 8871da177e4SLinus Torvalds } 8881da177e4SLinus Torvalds 8891da177e4SLinus Torvalds #ifdef CONFIG_PROC_FS 890ab50b8edSHugh Dickins void vm_stat_account(struct mm_struct *mm, unsigned long flags, 8911da177e4SLinus Torvalds struct file *file, long pages) 8921da177e4SLinus Torvalds { 8931da177e4SLinus Torvalds const unsigned long stack_flags 8941da177e4SLinus Torvalds = VM_STACK_FLAGS & (VM_GROWSUP|VM_GROWSDOWN); 8951da177e4SLinus Torvalds 8961da177e4SLinus Torvalds if (file) { 8971da177e4SLinus Torvalds mm->shared_vm += pages; 8981da177e4SLinus Torvalds if ((flags & (VM_EXEC|VM_WRITE)) == VM_EXEC) 8991da177e4SLinus Torvalds mm->exec_vm += pages; 9001da177e4SLinus Torvalds } else if (flags & stack_flags) 9011da177e4SLinus Torvalds mm->stack_vm += pages; 9021da177e4SLinus Torvalds if (flags & (VM_RESERVED|VM_IO)) 9031da177e4SLinus Torvalds mm->reserved_vm += pages; 9041da177e4SLinus Torvalds } 9051da177e4SLinus Torvalds #endif /* CONFIG_PROC_FS */ 9061da177e4SLinus Torvalds 9071da177e4SLinus Torvalds /* 9081da177e4SLinus Torvalds * The caller must hold down_write(current->mm->mmap_sem). 9091da177e4SLinus Torvalds */ 9101da177e4SLinus Torvalds 9111da177e4SLinus Torvalds unsigned long do_mmap_pgoff(struct file * file, unsigned long addr, 9121da177e4SLinus Torvalds unsigned long len, unsigned long prot, 9131da177e4SLinus Torvalds unsigned long flags, unsigned long pgoff) 9141da177e4SLinus Torvalds { 9151da177e4SLinus Torvalds struct mm_struct * mm = current->mm; 9161da177e4SLinus Torvalds struct inode *inode; 9171da177e4SLinus Torvalds unsigned int vm_flags; 9181da177e4SLinus Torvalds int error; 9191da177e4SLinus Torvalds int accountable = 1; 9200165ab44SMiklos Szeredi unsigned long reqprot = prot; 9211da177e4SLinus Torvalds 9221da177e4SLinus Torvalds /* 9231da177e4SLinus Torvalds * Does the application expect PROT_READ to imply PROT_EXEC? 9241da177e4SLinus Torvalds * 9251da177e4SLinus Torvalds * (the exception is when the underlying filesystem is noexec 9261da177e4SLinus Torvalds * mounted, in which case we dont add PROT_EXEC.) 9271da177e4SLinus Torvalds */ 9281da177e4SLinus Torvalds if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC)) 929d3ac7f89SJosef "Jeff" Sipek if (!(file && (file->f_path.mnt->mnt_flags & MNT_NOEXEC))) 9301da177e4SLinus Torvalds prot |= PROT_EXEC; 9311da177e4SLinus Torvalds 9321da177e4SLinus Torvalds if (!len) 9331da177e4SLinus Torvalds return -EINVAL; 9341da177e4SLinus Torvalds 9357cd94146SEric Paris if (!(flags & MAP_FIXED)) 9367cd94146SEric Paris addr = round_hint_to_min(addr); 9377cd94146SEric Paris 9383a459756SKirill Korotaev error = arch_mmap_check(addr, len, flags); 9393a459756SKirill Korotaev if (error) 9403a459756SKirill Korotaev return error; 9413a459756SKirill Korotaev 9421da177e4SLinus Torvalds /* Careful about overflows.. */ 9431da177e4SLinus Torvalds len = PAGE_ALIGN(len); 9441da177e4SLinus Torvalds if (!len || len > TASK_SIZE) 9451da177e4SLinus Torvalds return -ENOMEM; 9461da177e4SLinus Torvalds 9471da177e4SLinus Torvalds /* offset overflow? */ 9481da177e4SLinus Torvalds if ((pgoff + (len >> PAGE_SHIFT)) < pgoff) 9491da177e4SLinus Torvalds return -EOVERFLOW; 9501da177e4SLinus Torvalds 9511da177e4SLinus Torvalds /* Too many mappings? */ 9521da177e4SLinus Torvalds if (mm->map_count > sysctl_max_map_count) 9531da177e4SLinus Torvalds return -ENOMEM; 9541da177e4SLinus Torvalds 9551da177e4SLinus Torvalds /* Obtain the address to map to. we verify (or select) it and ensure 9561da177e4SLinus Torvalds * that it represents a valid section of the address space. 9571da177e4SLinus Torvalds */ 9581da177e4SLinus Torvalds addr = get_unmapped_area(file, addr, len, pgoff, flags); 9591da177e4SLinus Torvalds if (addr & ~PAGE_MASK) 9601da177e4SLinus Torvalds return addr; 9611da177e4SLinus Torvalds 9621da177e4SLinus Torvalds /* Do simple checking here so the lower-level routines won't have 9631da177e4SLinus Torvalds * to. we assume access permissions have been handled by the open 9641da177e4SLinus Torvalds * of the memory object, so we don't do any here. 9651da177e4SLinus Torvalds */ 9661da177e4SLinus Torvalds vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags) | 9671da177e4SLinus Torvalds mm->def_flags | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC; 9681da177e4SLinus Torvalds 9691da177e4SLinus Torvalds if (flags & MAP_LOCKED) { 9701da177e4SLinus Torvalds if (!can_do_mlock()) 9711da177e4SLinus Torvalds return -EPERM; 9721da177e4SLinus Torvalds vm_flags |= VM_LOCKED; 9731da177e4SLinus Torvalds } 9741da177e4SLinus Torvalds /* mlock MCL_FUTURE? */ 9751da177e4SLinus Torvalds if (vm_flags & VM_LOCKED) { 9761da177e4SLinus Torvalds unsigned long locked, lock_limit; 97793ea1d0aSChris Wright locked = len >> PAGE_SHIFT; 97893ea1d0aSChris Wright locked += mm->locked_vm; 9791da177e4SLinus Torvalds lock_limit = current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur; 98093ea1d0aSChris Wright lock_limit >>= PAGE_SHIFT; 9811da177e4SLinus Torvalds if (locked > lock_limit && !capable(CAP_IPC_LOCK)) 9821da177e4SLinus Torvalds return -EAGAIN; 9831da177e4SLinus Torvalds } 9841da177e4SLinus Torvalds 985d3ac7f89SJosef "Jeff" Sipek inode = file ? file->f_path.dentry->d_inode : NULL; 9861da177e4SLinus Torvalds 9871da177e4SLinus Torvalds if (file) { 9881da177e4SLinus Torvalds switch (flags & MAP_TYPE) { 9891da177e4SLinus Torvalds case MAP_SHARED: 9901da177e4SLinus Torvalds if ((prot&PROT_WRITE) && !(file->f_mode&FMODE_WRITE)) 9911da177e4SLinus Torvalds return -EACCES; 9921da177e4SLinus Torvalds 9931da177e4SLinus Torvalds /* 9941da177e4SLinus Torvalds * Make sure we don't allow writing to an append-only 9951da177e4SLinus Torvalds * file.. 9961da177e4SLinus Torvalds */ 9971da177e4SLinus Torvalds if (IS_APPEND(inode) && (file->f_mode & FMODE_WRITE)) 9981da177e4SLinus Torvalds return -EACCES; 9991da177e4SLinus Torvalds 10001da177e4SLinus Torvalds /* 10011da177e4SLinus Torvalds * Make sure there are no mandatory locks on the file. 10021da177e4SLinus Torvalds */ 10031da177e4SLinus Torvalds if (locks_verify_locked(inode)) 10041da177e4SLinus Torvalds return -EAGAIN; 10051da177e4SLinus Torvalds 10061da177e4SLinus Torvalds vm_flags |= VM_SHARED | VM_MAYSHARE; 10071da177e4SLinus Torvalds if (!(file->f_mode & FMODE_WRITE)) 10081da177e4SLinus Torvalds vm_flags &= ~(VM_MAYWRITE | VM_SHARED); 10091da177e4SLinus Torvalds 10101da177e4SLinus Torvalds /* fall through */ 10111da177e4SLinus Torvalds case MAP_PRIVATE: 10121da177e4SLinus Torvalds if (!(file->f_mode & FMODE_READ)) 10131da177e4SLinus Torvalds return -EACCES; 1014d3ac7f89SJosef "Jeff" Sipek if (file->f_path.mnt->mnt_flags & MNT_NOEXEC) { 101580c5606cSLinus Torvalds if (vm_flags & VM_EXEC) 101680c5606cSLinus Torvalds return -EPERM; 101780c5606cSLinus Torvalds vm_flags &= ~VM_MAYEXEC; 101880c5606cSLinus Torvalds } 101980c5606cSLinus Torvalds if (is_file_hugepages(file)) 102080c5606cSLinus Torvalds accountable = 0; 102180c5606cSLinus Torvalds 102280c5606cSLinus Torvalds if (!file->f_op || !file->f_op->mmap) 102380c5606cSLinus Torvalds return -ENODEV; 10241da177e4SLinus Torvalds break; 10251da177e4SLinus Torvalds 10261da177e4SLinus Torvalds default: 10271da177e4SLinus Torvalds return -EINVAL; 10281da177e4SLinus Torvalds } 10291da177e4SLinus Torvalds } else { 10301da177e4SLinus Torvalds switch (flags & MAP_TYPE) { 10311da177e4SLinus Torvalds case MAP_SHARED: 10321da177e4SLinus Torvalds vm_flags |= VM_SHARED | VM_MAYSHARE; 10331da177e4SLinus Torvalds break; 10341da177e4SLinus Torvalds case MAP_PRIVATE: 10351da177e4SLinus Torvalds /* 10361da177e4SLinus Torvalds * Set pgoff according to addr for anon_vma. 10371da177e4SLinus Torvalds */ 10381da177e4SLinus Torvalds pgoff = addr >> PAGE_SHIFT; 10391da177e4SLinus Torvalds break; 10401da177e4SLinus Torvalds default: 10411da177e4SLinus Torvalds return -EINVAL; 10421da177e4SLinus Torvalds } 10431da177e4SLinus Torvalds } 10441da177e4SLinus Torvalds 1045ed032189SEric Paris error = security_file_mmap(file, reqprot, prot, flags, addr, 0); 10461da177e4SLinus Torvalds if (error) 10471da177e4SLinus Torvalds return error; 10481da177e4SLinus Torvalds 10490165ab44SMiklos Szeredi return mmap_region(file, addr, len, flags, vm_flags, pgoff, 10500165ab44SMiklos Szeredi accountable); 10510165ab44SMiklos Szeredi } 10520165ab44SMiklos Szeredi EXPORT_SYMBOL(do_mmap_pgoff); 10530165ab44SMiklos Szeredi 10544e950f6fSAlexey Dobriyan /* 10554e950f6fSAlexey Dobriyan * Some shared mappigns will want the pages marked read-only 10564e950f6fSAlexey Dobriyan * to track write events. If so, we'll downgrade vm_page_prot 10574e950f6fSAlexey Dobriyan * to the private version (using protection_map[] without the 10584e950f6fSAlexey Dobriyan * VM_SHARED bit). 10594e950f6fSAlexey Dobriyan */ 10604e950f6fSAlexey Dobriyan int vma_wants_writenotify(struct vm_area_struct *vma) 10614e950f6fSAlexey Dobriyan { 10624e950f6fSAlexey Dobriyan unsigned int vm_flags = vma->vm_flags; 10634e950f6fSAlexey Dobriyan 10644e950f6fSAlexey Dobriyan /* If it was private or non-writable, the write bit is already clear */ 10654e950f6fSAlexey Dobriyan if ((vm_flags & (VM_WRITE|VM_SHARED)) != ((VM_WRITE|VM_SHARED))) 10664e950f6fSAlexey Dobriyan return 0; 10674e950f6fSAlexey Dobriyan 10684e950f6fSAlexey Dobriyan /* The backer wishes to know when pages are first written to? */ 10694e950f6fSAlexey Dobriyan if (vma->vm_ops && vma->vm_ops->page_mkwrite) 10704e950f6fSAlexey Dobriyan return 1; 10714e950f6fSAlexey Dobriyan 10724e950f6fSAlexey Dobriyan /* The open routine did something to the protections already? */ 10734e950f6fSAlexey Dobriyan if (pgprot_val(vma->vm_page_prot) != 10743ed75eb8SColy Li pgprot_val(vm_get_page_prot(vm_flags))) 10754e950f6fSAlexey Dobriyan return 0; 10764e950f6fSAlexey Dobriyan 10774e950f6fSAlexey Dobriyan /* Specialty mapping? */ 10784e950f6fSAlexey Dobriyan if (vm_flags & (VM_PFNMAP|VM_INSERTPAGE)) 10794e950f6fSAlexey Dobriyan return 0; 10804e950f6fSAlexey Dobriyan 10814e950f6fSAlexey Dobriyan /* Can the mapping track the dirty pages? */ 10824e950f6fSAlexey Dobriyan return vma->vm_file && vma->vm_file->f_mapping && 10834e950f6fSAlexey Dobriyan mapping_cap_account_dirty(vma->vm_file->f_mapping); 10844e950f6fSAlexey Dobriyan } 10854e950f6fSAlexey Dobriyan 10860165ab44SMiklos Szeredi unsigned long mmap_region(struct file *file, unsigned long addr, 10870165ab44SMiklos Szeredi unsigned long len, unsigned long flags, 10880165ab44SMiklos Szeredi unsigned int vm_flags, unsigned long pgoff, 10890165ab44SMiklos Szeredi int accountable) 10900165ab44SMiklos Szeredi { 10910165ab44SMiklos Szeredi struct mm_struct *mm = current->mm; 10920165ab44SMiklos Szeredi struct vm_area_struct *vma, *prev; 10930165ab44SMiklos Szeredi int correct_wcount = 0; 10940165ab44SMiklos Szeredi int error; 10950165ab44SMiklos Szeredi struct rb_node **rb_link, *rb_parent; 10960165ab44SMiklos Szeredi unsigned long charged = 0; 10970165ab44SMiklos Szeredi struct inode *inode = file ? file->f_path.dentry->d_inode : NULL; 10980165ab44SMiklos Szeredi 10991da177e4SLinus Torvalds /* Clear old maps */ 11001da177e4SLinus Torvalds error = -ENOMEM; 11011da177e4SLinus Torvalds munmap_back: 11021da177e4SLinus Torvalds vma = find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent); 11031da177e4SLinus Torvalds if (vma && vma->vm_start < addr + len) { 11041da177e4SLinus Torvalds if (do_munmap(mm, addr, len)) 11051da177e4SLinus Torvalds return -ENOMEM; 11061da177e4SLinus Torvalds goto munmap_back; 11071da177e4SLinus Torvalds } 11081da177e4SLinus Torvalds 11091da177e4SLinus Torvalds /* Check against address space limit. */ 1110119f657cSakpm@osdl.org if (!may_expand_vm(mm, len >> PAGE_SHIFT)) 11111da177e4SLinus Torvalds return -ENOMEM; 11121da177e4SLinus Torvalds 1113*cdfd4325SAndy Whitcroft if (flags & MAP_NORESERVE) 1114*cdfd4325SAndy Whitcroft vm_flags |= VM_NORESERVE; 1115*cdfd4325SAndy Whitcroft 11161da177e4SLinus Torvalds if (accountable && (!(flags & MAP_NORESERVE) || 11171da177e4SLinus Torvalds sysctl_overcommit_memory == OVERCOMMIT_NEVER)) { 11181da177e4SLinus Torvalds if (vm_flags & VM_SHARED) { 11191da177e4SLinus Torvalds /* Check memory availability in shmem_file_setup? */ 11201da177e4SLinus Torvalds vm_flags |= VM_ACCOUNT; 11211da177e4SLinus Torvalds } else if (vm_flags & VM_WRITE) { 11221da177e4SLinus Torvalds /* 11231da177e4SLinus Torvalds * Private writable mapping: check memory availability 11241da177e4SLinus Torvalds */ 11251da177e4SLinus Torvalds charged = len >> PAGE_SHIFT; 11261da177e4SLinus Torvalds if (security_vm_enough_memory(charged)) 11271da177e4SLinus Torvalds return -ENOMEM; 11281da177e4SLinus Torvalds vm_flags |= VM_ACCOUNT; 11291da177e4SLinus Torvalds } 11301da177e4SLinus Torvalds } 11311da177e4SLinus Torvalds 11321da177e4SLinus Torvalds /* 11331da177e4SLinus Torvalds * Can we just expand an old private anonymous mapping? 11341da177e4SLinus Torvalds * The VM_SHARED test is necessary because shmem_zero_setup 11351da177e4SLinus Torvalds * will create the file object for a shared anonymous map below. 11361da177e4SLinus Torvalds */ 11371da177e4SLinus Torvalds if (!file && !(vm_flags & VM_SHARED) && 11381da177e4SLinus Torvalds vma_merge(mm, prev, addr, addr + len, vm_flags, 11391da177e4SLinus Torvalds NULL, NULL, pgoff, NULL)) 11401da177e4SLinus Torvalds goto out; 11411da177e4SLinus Torvalds 11421da177e4SLinus Torvalds /* 11431da177e4SLinus Torvalds * Determine the object being mapped and call the appropriate 11441da177e4SLinus Torvalds * specific mapper. the address has already been validated, but 11451da177e4SLinus Torvalds * not unmapped, but the maps are removed from the list. 11461da177e4SLinus Torvalds */ 1147c5e3b83eSPekka Enberg vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL); 11481da177e4SLinus Torvalds if (!vma) { 11491da177e4SLinus Torvalds error = -ENOMEM; 11501da177e4SLinus Torvalds goto unacct_error; 11511da177e4SLinus Torvalds } 11521da177e4SLinus Torvalds 11531da177e4SLinus Torvalds vma->vm_mm = mm; 11541da177e4SLinus Torvalds vma->vm_start = addr; 11551da177e4SLinus Torvalds vma->vm_end = addr + len; 11561da177e4SLinus Torvalds vma->vm_flags = vm_flags; 11573ed75eb8SColy Li vma->vm_page_prot = vm_get_page_prot(vm_flags); 11581da177e4SLinus Torvalds vma->vm_pgoff = pgoff; 11591da177e4SLinus Torvalds 11601da177e4SLinus Torvalds if (file) { 11611da177e4SLinus Torvalds error = -EINVAL; 11621da177e4SLinus Torvalds if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP)) 11631da177e4SLinus Torvalds goto free_vma; 11641da177e4SLinus Torvalds if (vm_flags & VM_DENYWRITE) { 11651da177e4SLinus Torvalds error = deny_write_access(file); 11661da177e4SLinus Torvalds if (error) 11671da177e4SLinus Torvalds goto free_vma; 11681da177e4SLinus Torvalds correct_wcount = 1; 11691da177e4SLinus Torvalds } 11701da177e4SLinus Torvalds vma->vm_file = file; 11711da177e4SLinus Torvalds get_file(file); 11721da177e4SLinus Torvalds error = file->f_op->mmap(file, vma); 11731da177e4SLinus Torvalds if (error) 11741da177e4SLinus Torvalds goto unmap_and_free_vma; 1175925d1c40SMatt Helsley if (vm_flags & VM_EXECUTABLE) 1176925d1c40SMatt Helsley added_exe_file_vma(mm); 11771da177e4SLinus Torvalds } else if (vm_flags & VM_SHARED) { 11781da177e4SLinus Torvalds error = shmem_zero_setup(vma); 11791da177e4SLinus Torvalds if (error) 11801da177e4SLinus Torvalds goto free_vma; 11811da177e4SLinus Torvalds } 11821da177e4SLinus Torvalds 11831da177e4SLinus Torvalds /* We set VM_ACCOUNT in a shared mapping's vm_flags, to inform 11841da177e4SLinus Torvalds * shmem_zero_setup (perhaps called through /dev/zero's ->mmap) 11851da177e4SLinus Torvalds * that memory reservation must be checked; but that reservation 11861da177e4SLinus Torvalds * belongs to shared memory object, not to vma: so now clear it. 11871da177e4SLinus Torvalds */ 11881da177e4SLinus Torvalds if ((vm_flags & (VM_SHARED|VM_ACCOUNT)) == (VM_SHARED|VM_ACCOUNT)) 11891da177e4SLinus Torvalds vma->vm_flags &= ~VM_ACCOUNT; 11901da177e4SLinus Torvalds 11911da177e4SLinus Torvalds /* Can addr have changed?? 11921da177e4SLinus Torvalds * 11931da177e4SLinus Torvalds * Answer: Yes, several device drivers can do it in their 11941da177e4SLinus Torvalds * f_op->mmap method. -DaveM 11951da177e4SLinus Torvalds */ 11961da177e4SLinus Torvalds addr = vma->vm_start; 11971da177e4SLinus Torvalds pgoff = vma->vm_pgoff; 11981da177e4SLinus Torvalds vm_flags = vma->vm_flags; 11991da177e4SLinus Torvalds 1200d08b3851SPeter Zijlstra if (vma_wants_writenotify(vma)) 12011ddd439eSHugh Dickins vma->vm_page_prot = vm_get_page_prot(vm_flags & ~VM_SHARED); 1202d08b3851SPeter Zijlstra 12034d3d5b41SOleg Nesterov if (file && vma_merge(mm, prev, addr, vma->vm_end, 12041da177e4SLinus Torvalds vma->vm_flags, NULL, file, pgoff, vma_policy(vma))) { 1205f0be3d32SLee Schermerhorn mpol_put(vma_policy(vma)); 12061da177e4SLinus Torvalds kmem_cache_free(vm_area_cachep, vma); 12074d3d5b41SOleg Nesterov fput(file); 1208925d1c40SMatt Helsley if (vm_flags & VM_EXECUTABLE) 1209925d1c40SMatt Helsley removed_exe_file_vma(mm); 12104d3d5b41SOleg Nesterov } else { 12114d3d5b41SOleg Nesterov vma_link(mm, vma, prev, rb_link, rb_parent); 12124d3d5b41SOleg Nesterov file = vma->vm_file; 12131da177e4SLinus Torvalds } 12144d3d5b41SOleg Nesterov 12154d3d5b41SOleg Nesterov /* Once vma denies write, undo our temporary denial count */ 12164d3d5b41SOleg Nesterov if (correct_wcount) 12174d3d5b41SOleg Nesterov atomic_inc(&inode->i_writecount); 12181da177e4SLinus Torvalds out: 12191da177e4SLinus Torvalds mm->total_vm += len >> PAGE_SHIFT; 1220ab50b8edSHugh Dickins vm_stat_account(mm, vm_flags, file, len >> PAGE_SHIFT); 12211da177e4SLinus Torvalds if (vm_flags & VM_LOCKED) { 12221da177e4SLinus Torvalds mm->locked_vm += len >> PAGE_SHIFT; 12231da177e4SLinus Torvalds make_pages_present(addr, addr + len); 12241da177e4SLinus Torvalds } 122554cb8821SNick Piggin if ((flags & MAP_POPULATE) && !(flags & MAP_NONBLOCK)) 122654cb8821SNick Piggin make_pages_present(addr, addr + len); 12271da177e4SLinus Torvalds return addr; 12281da177e4SLinus Torvalds 12291da177e4SLinus Torvalds unmap_and_free_vma: 12301da177e4SLinus Torvalds if (correct_wcount) 12311da177e4SLinus Torvalds atomic_inc(&inode->i_writecount); 12321da177e4SLinus Torvalds vma->vm_file = NULL; 12331da177e4SLinus Torvalds fput(file); 12341da177e4SLinus Torvalds 12351da177e4SLinus Torvalds /* Undo any partial mapping done by a device driver. */ 1236e0da382cSHugh Dickins unmap_region(mm, vma, prev, vma->vm_start, vma->vm_end); 1237e0da382cSHugh Dickins charged = 0; 12381da177e4SLinus Torvalds free_vma: 12391da177e4SLinus Torvalds kmem_cache_free(vm_area_cachep, vma); 12401da177e4SLinus Torvalds unacct_error: 12411da177e4SLinus Torvalds if (charged) 12421da177e4SLinus Torvalds vm_unacct_memory(charged); 12431da177e4SLinus Torvalds return error; 12441da177e4SLinus Torvalds } 12451da177e4SLinus Torvalds 12461da177e4SLinus Torvalds /* Get an address range which is currently unmapped. 12471da177e4SLinus Torvalds * For shmat() with addr=0. 12481da177e4SLinus Torvalds * 12491da177e4SLinus Torvalds * Ugly calling convention alert: 12501da177e4SLinus Torvalds * Return value with the low bits set means error value, 12511da177e4SLinus Torvalds * ie 12521da177e4SLinus Torvalds * if (ret & ~PAGE_MASK) 12531da177e4SLinus Torvalds * error = ret; 12541da177e4SLinus Torvalds * 12551da177e4SLinus Torvalds * This function "knows" that -ENOMEM has the bits set. 12561da177e4SLinus Torvalds */ 12571da177e4SLinus Torvalds #ifndef HAVE_ARCH_UNMAPPED_AREA 12581da177e4SLinus Torvalds unsigned long 12591da177e4SLinus Torvalds arch_get_unmapped_area(struct file *filp, unsigned long addr, 12601da177e4SLinus Torvalds unsigned long len, unsigned long pgoff, unsigned long flags) 12611da177e4SLinus Torvalds { 12621da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 12631da177e4SLinus Torvalds struct vm_area_struct *vma; 12641da177e4SLinus Torvalds unsigned long start_addr; 12651da177e4SLinus Torvalds 12661da177e4SLinus Torvalds if (len > TASK_SIZE) 12671da177e4SLinus Torvalds return -ENOMEM; 12681da177e4SLinus Torvalds 126906abdfb4SBenjamin Herrenschmidt if (flags & MAP_FIXED) 127006abdfb4SBenjamin Herrenschmidt return addr; 127106abdfb4SBenjamin Herrenschmidt 12721da177e4SLinus Torvalds if (addr) { 12731da177e4SLinus Torvalds addr = PAGE_ALIGN(addr); 12741da177e4SLinus Torvalds vma = find_vma(mm, addr); 12751da177e4SLinus Torvalds if (TASK_SIZE - len >= addr && 12761da177e4SLinus Torvalds (!vma || addr + len <= vma->vm_start)) 12771da177e4SLinus Torvalds return addr; 12781da177e4SLinus Torvalds } 12791363c3cdSWolfgang Wander if (len > mm->cached_hole_size) { 12801da177e4SLinus Torvalds start_addr = addr = mm->free_area_cache; 12811363c3cdSWolfgang Wander } else { 12821363c3cdSWolfgang Wander start_addr = addr = TASK_UNMAPPED_BASE; 12831363c3cdSWolfgang Wander mm->cached_hole_size = 0; 12841363c3cdSWolfgang Wander } 12851da177e4SLinus Torvalds 12861da177e4SLinus Torvalds full_search: 12871da177e4SLinus Torvalds for (vma = find_vma(mm, addr); ; vma = vma->vm_next) { 12881da177e4SLinus Torvalds /* At this point: (!vma || addr < vma->vm_end). */ 12891da177e4SLinus Torvalds if (TASK_SIZE - len < addr) { 12901da177e4SLinus Torvalds /* 12911da177e4SLinus Torvalds * Start a new search - just in case we missed 12921da177e4SLinus Torvalds * some holes. 12931da177e4SLinus Torvalds */ 12941da177e4SLinus Torvalds if (start_addr != TASK_UNMAPPED_BASE) { 12951363c3cdSWolfgang Wander addr = TASK_UNMAPPED_BASE; 12961363c3cdSWolfgang Wander start_addr = addr; 12971363c3cdSWolfgang Wander mm->cached_hole_size = 0; 12981da177e4SLinus Torvalds goto full_search; 12991da177e4SLinus Torvalds } 13001da177e4SLinus Torvalds return -ENOMEM; 13011da177e4SLinus Torvalds } 13021da177e4SLinus Torvalds if (!vma || addr + len <= vma->vm_start) { 13031da177e4SLinus Torvalds /* 13041da177e4SLinus Torvalds * Remember the place where we stopped the search: 13051da177e4SLinus Torvalds */ 13061da177e4SLinus Torvalds mm->free_area_cache = addr + len; 13071da177e4SLinus Torvalds return addr; 13081da177e4SLinus Torvalds } 13091363c3cdSWolfgang Wander if (addr + mm->cached_hole_size < vma->vm_start) 13101363c3cdSWolfgang Wander mm->cached_hole_size = vma->vm_start - addr; 13111da177e4SLinus Torvalds addr = vma->vm_end; 13121da177e4SLinus Torvalds } 13131da177e4SLinus Torvalds } 13141da177e4SLinus Torvalds #endif 13151da177e4SLinus Torvalds 13161363c3cdSWolfgang Wander void arch_unmap_area(struct mm_struct *mm, unsigned long addr) 13171da177e4SLinus Torvalds { 13181da177e4SLinus Torvalds /* 13191da177e4SLinus Torvalds * Is this a new hole at the lowest possible address? 13201da177e4SLinus Torvalds */ 13211363c3cdSWolfgang Wander if (addr >= TASK_UNMAPPED_BASE && addr < mm->free_area_cache) { 13221363c3cdSWolfgang Wander mm->free_area_cache = addr; 13231363c3cdSWolfgang Wander mm->cached_hole_size = ~0UL; 13241363c3cdSWolfgang Wander } 13251da177e4SLinus Torvalds } 13261da177e4SLinus Torvalds 13271da177e4SLinus Torvalds /* 13281da177e4SLinus Torvalds * This mmap-allocator allocates new areas top-down from below the 13291da177e4SLinus Torvalds * stack's low limit (the base): 13301da177e4SLinus Torvalds */ 13311da177e4SLinus Torvalds #ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN 13321da177e4SLinus Torvalds unsigned long 13331da177e4SLinus Torvalds arch_get_unmapped_area_topdown(struct file *filp, const unsigned long addr0, 13341da177e4SLinus Torvalds const unsigned long len, const unsigned long pgoff, 13351da177e4SLinus Torvalds const unsigned long flags) 13361da177e4SLinus Torvalds { 13371da177e4SLinus Torvalds struct vm_area_struct *vma; 13381da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 13391da177e4SLinus Torvalds unsigned long addr = addr0; 13401da177e4SLinus Torvalds 13411da177e4SLinus Torvalds /* requested length too big for entire address space */ 13421da177e4SLinus Torvalds if (len > TASK_SIZE) 13431da177e4SLinus Torvalds return -ENOMEM; 13441da177e4SLinus Torvalds 134506abdfb4SBenjamin Herrenschmidt if (flags & MAP_FIXED) 134606abdfb4SBenjamin Herrenschmidt return addr; 134706abdfb4SBenjamin Herrenschmidt 13481da177e4SLinus Torvalds /* requesting a specific address */ 13491da177e4SLinus Torvalds if (addr) { 13501da177e4SLinus Torvalds addr = PAGE_ALIGN(addr); 13511da177e4SLinus Torvalds vma = find_vma(mm, addr); 13521da177e4SLinus Torvalds if (TASK_SIZE - len >= addr && 13531da177e4SLinus Torvalds (!vma || addr + len <= vma->vm_start)) 13541da177e4SLinus Torvalds return addr; 13551da177e4SLinus Torvalds } 13561da177e4SLinus Torvalds 13571363c3cdSWolfgang Wander /* check if free_area_cache is useful for us */ 13581363c3cdSWolfgang Wander if (len <= mm->cached_hole_size) { 13591363c3cdSWolfgang Wander mm->cached_hole_size = 0; 13601363c3cdSWolfgang Wander mm->free_area_cache = mm->mmap_base; 13611363c3cdSWolfgang Wander } 13621363c3cdSWolfgang Wander 13631da177e4SLinus Torvalds /* either no address requested or can't fit in requested address hole */ 13641da177e4SLinus Torvalds addr = mm->free_area_cache; 13651da177e4SLinus Torvalds 13661da177e4SLinus Torvalds /* make sure it can fit in the remaining address space */ 136749a43876SLinus Torvalds if (addr > len) { 13681da177e4SLinus Torvalds vma = find_vma(mm, addr-len); 13691da177e4SLinus Torvalds if (!vma || addr <= vma->vm_start) 13701da177e4SLinus Torvalds /* remember the address as a hint for next time */ 13711da177e4SLinus Torvalds return (mm->free_area_cache = addr-len); 13721da177e4SLinus Torvalds } 13731da177e4SLinus Torvalds 137473219d17SChris Wright if (mm->mmap_base < len) 137573219d17SChris Wright goto bottomup; 137673219d17SChris Wright 13771da177e4SLinus Torvalds addr = mm->mmap_base-len; 13781da177e4SLinus Torvalds 13791da177e4SLinus Torvalds do { 13801da177e4SLinus Torvalds /* 13811da177e4SLinus Torvalds * Lookup failure means no vma is above this address, 13821da177e4SLinus Torvalds * else if new region fits below vma->vm_start, 13831da177e4SLinus Torvalds * return with success: 13841da177e4SLinus Torvalds */ 13851da177e4SLinus Torvalds vma = find_vma(mm, addr); 13861da177e4SLinus Torvalds if (!vma || addr+len <= vma->vm_start) 13871da177e4SLinus Torvalds /* remember the address as a hint for next time */ 13881da177e4SLinus Torvalds return (mm->free_area_cache = addr); 13891da177e4SLinus Torvalds 13901363c3cdSWolfgang Wander /* remember the largest hole we saw so far */ 13911363c3cdSWolfgang Wander if (addr + mm->cached_hole_size < vma->vm_start) 13921363c3cdSWolfgang Wander mm->cached_hole_size = vma->vm_start - addr; 13931363c3cdSWolfgang Wander 13941da177e4SLinus Torvalds /* try just below the current vma->vm_start */ 13951da177e4SLinus Torvalds addr = vma->vm_start-len; 139649a43876SLinus Torvalds } while (len < vma->vm_start); 13971da177e4SLinus Torvalds 139873219d17SChris Wright bottomup: 13991da177e4SLinus Torvalds /* 14001da177e4SLinus Torvalds * A failed mmap() very likely causes application failure, 14011da177e4SLinus Torvalds * so fall back to the bottom-up function here. This scenario 14021da177e4SLinus Torvalds * can happen with large stack limits and large mmap() 14031da177e4SLinus Torvalds * allocations. 14041da177e4SLinus Torvalds */ 14051363c3cdSWolfgang Wander mm->cached_hole_size = ~0UL; 14061da177e4SLinus Torvalds mm->free_area_cache = TASK_UNMAPPED_BASE; 14071da177e4SLinus Torvalds addr = arch_get_unmapped_area(filp, addr0, len, pgoff, flags); 14081da177e4SLinus Torvalds /* 14091da177e4SLinus Torvalds * Restore the topdown base: 14101da177e4SLinus Torvalds */ 14111da177e4SLinus Torvalds mm->free_area_cache = mm->mmap_base; 14121363c3cdSWolfgang Wander mm->cached_hole_size = ~0UL; 14131da177e4SLinus Torvalds 14141da177e4SLinus Torvalds return addr; 14151da177e4SLinus Torvalds } 14161da177e4SLinus Torvalds #endif 14171da177e4SLinus Torvalds 14181363c3cdSWolfgang Wander void arch_unmap_area_topdown(struct mm_struct *mm, unsigned long addr) 14191da177e4SLinus Torvalds { 14201da177e4SLinus Torvalds /* 14211da177e4SLinus Torvalds * Is this a new hole at the highest possible address? 14221da177e4SLinus Torvalds */ 14231363c3cdSWolfgang Wander if (addr > mm->free_area_cache) 14241363c3cdSWolfgang Wander mm->free_area_cache = addr; 14251da177e4SLinus Torvalds 14261da177e4SLinus Torvalds /* dont allow allocations above current base */ 14271363c3cdSWolfgang Wander if (mm->free_area_cache > mm->mmap_base) 14281363c3cdSWolfgang Wander mm->free_area_cache = mm->mmap_base; 14291da177e4SLinus Torvalds } 14301da177e4SLinus Torvalds 14311da177e4SLinus Torvalds unsigned long 14321da177e4SLinus Torvalds get_unmapped_area(struct file *file, unsigned long addr, unsigned long len, 14331da177e4SLinus Torvalds unsigned long pgoff, unsigned long flags) 14341da177e4SLinus Torvalds { 143506abdfb4SBenjamin Herrenschmidt unsigned long (*get_area)(struct file *, unsigned long, 143606abdfb4SBenjamin Herrenschmidt unsigned long, unsigned long, unsigned long); 143707ab67c8SLinus Torvalds 143807ab67c8SLinus Torvalds get_area = current->mm->get_unmapped_area; 143907ab67c8SLinus Torvalds if (file && file->f_op && file->f_op->get_unmapped_area) 144007ab67c8SLinus Torvalds get_area = file->f_op->get_unmapped_area; 144107ab67c8SLinus Torvalds addr = get_area(file, addr, len, pgoff, flags); 144207ab67c8SLinus Torvalds if (IS_ERR_VALUE(addr)) 144307ab67c8SLinus Torvalds return addr; 144407ab67c8SLinus Torvalds 14451da177e4SLinus Torvalds if (addr > TASK_SIZE - len) 14461da177e4SLinus Torvalds return -ENOMEM; 14471da177e4SLinus Torvalds if (addr & ~PAGE_MASK) 14481da177e4SLinus Torvalds return -EINVAL; 144906abdfb4SBenjamin Herrenschmidt 145008e7d9b5SMartin Schwidefsky return arch_rebalance_pgtables(addr, len); 14511da177e4SLinus Torvalds } 14521da177e4SLinus Torvalds 14531da177e4SLinus Torvalds EXPORT_SYMBOL(get_unmapped_area); 14541da177e4SLinus Torvalds 14551da177e4SLinus Torvalds /* Look up the first VMA which satisfies addr < vm_end, NULL if none. */ 14561da177e4SLinus Torvalds struct vm_area_struct * find_vma(struct mm_struct * mm, unsigned long addr) 14571da177e4SLinus Torvalds { 14581da177e4SLinus Torvalds struct vm_area_struct *vma = NULL; 14591da177e4SLinus Torvalds 14601da177e4SLinus Torvalds if (mm) { 14611da177e4SLinus Torvalds /* Check the cache first. */ 14621da177e4SLinus Torvalds /* (Cache hit rate is typically around 35%.) */ 14631da177e4SLinus Torvalds vma = mm->mmap_cache; 14641da177e4SLinus Torvalds if (!(vma && vma->vm_end > addr && vma->vm_start <= addr)) { 14651da177e4SLinus Torvalds struct rb_node * rb_node; 14661da177e4SLinus Torvalds 14671da177e4SLinus Torvalds rb_node = mm->mm_rb.rb_node; 14681da177e4SLinus Torvalds vma = NULL; 14691da177e4SLinus Torvalds 14701da177e4SLinus Torvalds while (rb_node) { 14711da177e4SLinus Torvalds struct vm_area_struct * vma_tmp; 14721da177e4SLinus Torvalds 14731da177e4SLinus Torvalds vma_tmp = rb_entry(rb_node, 14741da177e4SLinus Torvalds struct vm_area_struct, vm_rb); 14751da177e4SLinus Torvalds 14761da177e4SLinus Torvalds if (vma_tmp->vm_end > addr) { 14771da177e4SLinus Torvalds vma = vma_tmp; 14781da177e4SLinus Torvalds if (vma_tmp->vm_start <= addr) 14791da177e4SLinus Torvalds break; 14801da177e4SLinus Torvalds rb_node = rb_node->rb_left; 14811da177e4SLinus Torvalds } else 14821da177e4SLinus Torvalds rb_node = rb_node->rb_right; 14831da177e4SLinus Torvalds } 14841da177e4SLinus Torvalds if (vma) 14851da177e4SLinus Torvalds mm->mmap_cache = vma; 14861da177e4SLinus Torvalds } 14871da177e4SLinus Torvalds } 14881da177e4SLinus Torvalds return vma; 14891da177e4SLinus Torvalds } 14901da177e4SLinus Torvalds 14911da177e4SLinus Torvalds EXPORT_SYMBOL(find_vma); 14921da177e4SLinus Torvalds 14931da177e4SLinus Torvalds /* Same as find_vma, but also return a pointer to the previous VMA in *pprev. */ 14941da177e4SLinus Torvalds struct vm_area_struct * 14951da177e4SLinus Torvalds find_vma_prev(struct mm_struct *mm, unsigned long addr, 14961da177e4SLinus Torvalds struct vm_area_struct **pprev) 14971da177e4SLinus Torvalds { 14981da177e4SLinus Torvalds struct vm_area_struct *vma = NULL, *prev = NULL; 14991da177e4SLinus Torvalds struct rb_node * rb_node; 15001da177e4SLinus Torvalds if (!mm) 15011da177e4SLinus Torvalds goto out; 15021da177e4SLinus Torvalds 15031da177e4SLinus Torvalds /* Guard against addr being lower than the first VMA */ 15041da177e4SLinus Torvalds vma = mm->mmap; 15051da177e4SLinus Torvalds 15061da177e4SLinus Torvalds /* Go through the RB tree quickly. */ 15071da177e4SLinus Torvalds rb_node = mm->mm_rb.rb_node; 15081da177e4SLinus Torvalds 15091da177e4SLinus Torvalds while (rb_node) { 15101da177e4SLinus Torvalds struct vm_area_struct *vma_tmp; 15111da177e4SLinus Torvalds vma_tmp = rb_entry(rb_node, struct vm_area_struct, vm_rb); 15121da177e4SLinus Torvalds 15131da177e4SLinus Torvalds if (addr < vma_tmp->vm_end) { 15141da177e4SLinus Torvalds rb_node = rb_node->rb_left; 15151da177e4SLinus Torvalds } else { 15161da177e4SLinus Torvalds prev = vma_tmp; 15171da177e4SLinus Torvalds if (!prev->vm_next || (addr < prev->vm_next->vm_end)) 15181da177e4SLinus Torvalds break; 15191da177e4SLinus Torvalds rb_node = rb_node->rb_right; 15201da177e4SLinus Torvalds } 15211da177e4SLinus Torvalds } 15221da177e4SLinus Torvalds 15231da177e4SLinus Torvalds out: 15241da177e4SLinus Torvalds *pprev = prev; 15251da177e4SLinus Torvalds return prev ? prev->vm_next : vma; 15261da177e4SLinus Torvalds } 15271da177e4SLinus Torvalds 15281da177e4SLinus Torvalds /* 15291da177e4SLinus Torvalds * Verify that the stack growth is acceptable and 15301da177e4SLinus Torvalds * update accounting. This is shared with both the 15311da177e4SLinus Torvalds * grow-up and grow-down cases. 15321da177e4SLinus Torvalds */ 15331da177e4SLinus Torvalds static int acct_stack_growth(struct vm_area_struct * vma, unsigned long size, unsigned long grow) 15341da177e4SLinus Torvalds { 15351da177e4SLinus Torvalds struct mm_struct *mm = vma->vm_mm; 15361da177e4SLinus Torvalds struct rlimit *rlim = current->signal->rlim; 15370d59a01bSAdam Litke unsigned long new_start; 15381da177e4SLinus Torvalds 15391da177e4SLinus Torvalds /* address space limit tests */ 1540119f657cSakpm@osdl.org if (!may_expand_vm(mm, grow)) 15411da177e4SLinus Torvalds return -ENOMEM; 15421da177e4SLinus Torvalds 15431da177e4SLinus Torvalds /* Stack limit test */ 15441da177e4SLinus Torvalds if (size > rlim[RLIMIT_STACK].rlim_cur) 15451da177e4SLinus Torvalds return -ENOMEM; 15461da177e4SLinus Torvalds 15471da177e4SLinus Torvalds /* mlock limit tests */ 15481da177e4SLinus Torvalds if (vma->vm_flags & VM_LOCKED) { 15491da177e4SLinus Torvalds unsigned long locked; 15501da177e4SLinus Torvalds unsigned long limit; 15511da177e4SLinus Torvalds locked = mm->locked_vm + grow; 15521da177e4SLinus Torvalds limit = rlim[RLIMIT_MEMLOCK].rlim_cur >> PAGE_SHIFT; 15531da177e4SLinus Torvalds if (locked > limit && !capable(CAP_IPC_LOCK)) 15541da177e4SLinus Torvalds return -ENOMEM; 15551da177e4SLinus Torvalds } 15561da177e4SLinus Torvalds 15570d59a01bSAdam Litke /* Check to ensure the stack will not grow into a hugetlb-only region */ 15580d59a01bSAdam Litke new_start = (vma->vm_flags & VM_GROWSUP) ? vma->vm_start : 15590d59a01bSAdam Litke vma->vm_end - size; 15600d59a01bSAdam Litke if (is_hugepage_only_range(vma->vm_mm, new_start, size)) 15610d59a01bSAdam Litke return -EFAULT; 15620d59a01bSAdam Litke 15631da177e4SLinus Torvalds /* 15641da177e4SLinus Torvalds * Overcommit.. This must be the final test, as it will 15651da177e4SLinus Torvalds * update security statistics. 15661da177e4SLinus Torvalds */ 15671da177e4SLinus Torvalds if (security_vm_enough_memory(grow)) 15681da177e4SLinus Torvalds return -ENOMEM; 15691da177e4SLinus Torvalds 15701da177e4SLinus Torvalds /* Ok, everything looks good - let it rip */ 15711da177e4SLinus Torvalds mm->total_vm += grow; 15721da177e4SLinus Torvalds if (vma->vm_flags & VM_LOCKED) 15731da177e4SLinus Torvalds mm->locked_vm += grow; 1574ab50b8edSHugh Dickins vm_stat_account(mm, vma->vm_flags, vma->vm_file, grow); 15751da177e4SLinus Torvalds return 0; 15761da177e4SLinus Torvalds } 15771da177e4SLinus Torvalds 157846dea3d0SHugh Dickins #if defined(CONFIG_STACK_GROWSUP) || defined(CONFIG_IA64) 15791da177e4SLinus Torvalds /* 158046dea3d0SHugh Dickins * PA-RISC uses this for its stack; IA64 for its Register Backing Store. 158146dea3d0SHugh Dickins * vma is the last one with address > vma->vm_end. Have to extend vma. 15821da177e4SLinus Torvalds */ 15839ab88515SMatthew Wilcox #ifndef CONFIG_IA64 158446dea3d0SHugh Dickins static inline 158546dea3d0SHugh Dickins #endif 158646dea3d0SHugh Dickins int expand_upwards(struct vm_area_struct *vma, unsigned long address) 15871da177e4SLinus Torvalds { 15881da177e4SLinus Torvalds int error; 15891da177e4SLinus Torvalds 15901da177e4SLinus Torvalds if (!(vma->vm_flags & VM_GROWSUP)) 15911da177e4SLinus Torvalds return -EFAULT; 15921da177e4SLinus Torvalds 15931da177e4SLinus Torvalds /* 15941da177e4SLinus Torvalds * We must make sure the anon_vma is allocated 15951da177e4SLinus Torvalds * so that the anon_vma locking is not a noop. 15961da177e4SLinus Torvalds */ 15971da177e4SLinus Torvalds if (unlikely(anon_vma_prepare(vma))) 15981da177e4SLinus Torvalds return -ENOMEM; 15991da177e4SLinus Torvalds anon_vma_lock(vma); 16001da177e4SLinus Torvalds 16011da177e4SLinus Torvalds /* 16021da177e4SLinus Torvalds * vma->vm_start/vm_end cannot change under us because the caller 16031da177e4SLinus Torvalds * is required to hold the mmap_sem in read mode. We need the 16041da177e4SLinus Torvalds * anon_vma lock to serialize against concurrent expand_stacks. 160506b32f3aSHelge Deller * Also guard against wrapping around to address 0. 16061da177e4SLinus Torvalds */ 160706b32f3aSHelge Deller if (address < PAGE_ALIGN(address+4)) 160806b32f3aSHelge Deller address = PAGE_ALIGN(address+4); 160906b32f3aSHelge Deller else { 161006b32f3aSHelge Deller anon_vma_unlock(vma); 161106b32f3aSHelge Deller return -ENOMEM; 161206b32f3aSHelge Deller } 16131da177e4SLinus Torvalds error = 0; 16141da177e4SLinus Torvalds 16151da177e4SLinus Torvalds /* Somebody else might have raced and expanded it already */ 16161da177e4SLinus Torvalds if (address > vma->vm_end) { 16171da177e4SLinus Torvalds unsigned long size, grow; 16181da177e4SLinus Torvalds 16191da177e4SLinus Torvalds size = address - vma->vm_start; 16201da177e4SLinus Torvalds grow = (address - vma->vm_end) >> PAGE_SHIFT; 16211da177e4SLinus Torvalds 16221da177e4SLinus Torvalds error = acct_stack_growth(vma, size, grow); 16231da177e4SLinus Torvalds if (!error) 16241da177e4SLinus Torvalds vma->vm_end = address; 16251da177e4SLinus Torvalds } 16261da177e4SLinus Torvalds anon_vma_unlock(vma); 16271da177e4SLinus Torvalds return error; 16281da177e4SLinus Torvalds } 162946dea3d0SHugh Dickins #endif /* CONFIG_STACK_GROWSUP || CONFIG_IA64 */ 163046dea3d0SHugh Dickins 16311da177e4SLinus Torvalds /* 16321da177e4SLinus Torvalds * vma is the first one with address < vma->vm_start. Have to extend vma. 16331da177e4SLinus Torvalds */ 1634b6a2fea3SOllie Wild static inline int expand_downwards(struct vm_area_struct *vma, 1635b6a2fea3SOllie Wild unsigned long address) 16361da177e4SLinus Torvalds { 16371da177e4SLinus Torvalds int error; 16381da177e4SLinus Torvalds 16391da177e4SLinus Torvalds /* 16401da177e4SLinus Torvalds * We must make sure the anon_vma is allocated 16411da177e4SLinus Torvalds * so that the anon_vma locking is not a noop. 16421da177e4SLinus Torvalds */ 16431da177e4SLinus Torvalds if (unlikely(anon_vma_prepare(vma))) 16441da177e4SLinus Torvalds return -ENOMEM; 16458869477aSEric Paris 16468869477aSEric Paris address &= PAGE_MASK; 164788c3f7a8SRichard Knutsson error = security_file_mmap(NULL, 0, 0, 0, address, 1); 16488869477aSEric Paris if (error) 16498869477aSEric Paris return error; 16508869477aSEric Paris 16511da177e4SLinus Torvalds anon_vma_lock(vma); 16521da177e4SLinus Torvalds 16531da177e4SLinus Torvalds /* 16541da177e4SLinus Torvalds * vma->vm_start/vm_end cannot change under us because the caller 16551da177e4SLinus Torvalds * is required to hold the mmap_sem in read mode. We need the 16561da177e4SLinus Torvalds * anon_vma lock to serialize against concurrent expand_stacks. 16571da177e4SLinus Torvalds */ 16581da177e4SLinus Torvalds 16591da177e4SLinus Torvalds /* Somebody else might have raced and expanded it already */ 16601da177e4SLinus Torvalds if (address < vma->vm_start) { 16611da177e4SLinus Torvalds unsigned long size, grow; 16621da177e4SLinus Torvalds 16631da177e4SLinus Torvalds size = vma->vm_end - address; 16641da177e4SLinus Torvalds grow = (vma->vm_start - address) >> PAGE_SHIFT; 16651da177e4SLinus Torvalds 16661da177e4SLinus Torvalds error = acct_stack_growth(vma, size, grow); 16671da177e4SLinus Torvalds if (!error) { 16681da177e4SLinus Torvalds vma->vm_start = address; 16691da177e4SLinus Torvalds vma->vm_pgoff -= grow; 16701da177e4SLinus Torvalds } 16711da177e4SLinus Torvalds } 16721da177e4SLinus Torvalds anon_vma_unlock(vma); 16731da177e4SLinus Torvalds return error; 16741da177e4SLinus Torvalds } 16751da177e4SLinus Torvalds 1676b6a2fea3SOllie Wild int expand_stack_downwards(struct vm_area_struct *vma, unsigned long address) 1677b6a2fea3SOllie Wild { 1678b6a2fea3SOllie Wild return expand_downwards(vma, address); 1679b6a2fea3SOllie Wild } 1680b6a2fea3SOllie Wild 1681b6a2fea3SOllie Wild #ifdef CONFIG_STACK_GROWSUP 1682b6a2fea3SOllie Wild int expand_stack(struct vm_area_struct *vma, unsigned long address) 1683b6a2fea3SOllie Wild { 1684b6a2fea3SOllie Wild return expand_upwards(vma, address); 1685b6a2fea3SOllie Wild } 1686b6a2fea3SOllie Wild 1687b6a2fea3SOllie Wild struct vm_area_struct * 1688b6a2fea3SOllie Wild find_extend_vma(struct mm_struct *mm, unsigned long addr) 1689b6a2fea3SOllie Wild { 1690b6a2fea3SOllie Wild struct vm_area_struct *vma, *prev; 1691b6a2fea3SOllie Wild 1692b6a2fea3SOllie Wild addr &= PAGE_MASK; 1693b6a2fea3SOllie Wild vma = find_vma_prev(mm, addr, &prev); 1694b6a2fea3SOllie Wild if (vma && (vma->vm_start <= addr)) 1695b6a2fea3SOllie Wild return vma; 1696b6a2fea3SOllie Wild if (!prev || expand_stack(prev, addr)) 1697b6a2fea3SOllie Wild return NULL; 1698b6a2fea3SOllie Wild if (prev->vm_flags & VM_LOCKED) 1699b6a2fea3SOllie Wild make_pages_present(addr, prev->vm_end); 1700b6a2fea3SOllie Wild return prev; 1701b6a2fea3SOllie Wild } 1702b6a2fea3SOllie Wild #else 1703b6a2fea3SOllie Wild int expand_stack(struct vm_area_struct *vma, unsigned long address) 1704b6a2fea3SOllie Wild { 1705b6a2fea3SOllie Wild return expand_downwards(vma, address); 1706b6a2fea3SOllie Wild } 1707b6a2fea3SOllie Wild 17081da177e4SLinus Torvalds struct vm_area_struct * 17091da177e4SLinus Torvalds find_extend_vma(struct mm_struct * mm, unsigned long addr) 17101da177e4SLinus Torvalds { 17111da177e4SLinus Torvalds struct vm_area_struct * vma; 17121da177e4SLinus Torvalds unsigned long start; 17131da177e4SLinus Torvalds 17141da177e4SLinus Torvalds addr &= PAGE_MASK; 17151da177e4SLinus Torvalds vma = find_vma(mm,addr); 17161da177e4SLinus Torvalds if (!vma) 17171da177e4SLinus Torvalds return NULL; 17181da177e4SLinus Torvalds if (vma->vm_start <= addr) 17191da177e4SLinus Torvalds return vma; 17201da177e4SLinus Torvalds if (!(vma->vm_flags & VM_GROWSDOWN)) 17211da177e4SLinus Torvalds return NULL; 17221da177e4SLinus Torvalds start = vma->vm_start; 17231da177e4SLinus Torvalds if (expand_stack(vma, addr)) 17241da177e4SLinus Torvalds return NULL; 1725b6a2fea3SOllie Wild if (vma->vm_flags & VM_LOCKED) 17261da177e4SLinus Torvalds make_pages_present(addr, start); 17271da177e4SLinus Torvalds return vma; 17281da177e4SLinus Torvalds } 17291da177e4SLinus Torvalds #endif 17301da177e4SLinus Torvalds 17312c0b3814SHugh Dickins /* 17322c0b3814SHugh Dickins * Ok - we have the memory areas we should free on the vma list, 17332c0b3814SHugh Dickins * so release them, and do the vma updates. 17341da177e4SLinus Torvalds * 17352c0b3814SHugh Dickins * Called with the mm semaphore held. 17361da177e4SLinus Torvalds */ 17372c0b3814SHugh Dickins static void remove_vma_list(struct mm_struct *mm, struct vm_area_struct *vma) 17381da177e4SLinus Torvalds { 1739365e9c87SHugh Dickins /* Update high watermark before we lower total_vm */ 1740365e9c87SHugh Dickins update_hiwater_vm(mm); 17412c0b3814SHugh Dickins do { 1742ab50b8edSHugh Dickins long nrpages = vma_pages(vma); 17431da177e4SLinus Torvalds 1744ab50b8edSHugh Dickins mm->total_vm -= nrpages; 1745ab50b8edSHugh Dickins if (vma->vm_flags & VM_LOCKED) 1746ab50b8edSHugh Dickins mm->locked_vm -= nrpages; 1747ab50b8edSHugh Dickins vm_stat_account(mm, vma->vm_flags, vma->vm_file, -nrpages); 1748a8fb5618SHugh Dickins vma = remove_vma(vma); 1749146425a3SHugh Dickins } while (vma); 17501da177e4SLinus Torvalds validate_mm(mm); 17511da177e4SLinus Torvalds } 17521da177e4SLinus Torvalds 17531da177e4SLinus Torvalds /* 17541da177e4SLinus Torvalds * Get rid of page table information in the indicated region. 17551da177e4SLinus Torvalds * 1756f10df686SPaolo 'Blaisorblade' Giarrusso * Called with the mm semaphore held. 17571da177e4SLinus Torvalds */ 17581da177e4SLinus Torvalds static void unmap_region(struct mm_struct *mm, 1759e0da382cSHugh Dickins struct vm_area_struct *vma, struct vm_area_struct *prev, 1760e0da382cSHugh Dickins unsigned long start, unsigned long end) 17611da177e4SLinus Torvalds { 1762e0da382cSHugh Dickins struct vm_area_struct *next = prev? prev->vm_next: mm->mmap; 17631da177e4SLinus Torvalds struct mmu_gather *tlb; 17641da177e4SLinus Torvalds unsigned long nr_accounted = 0; 17651da177e4SLinus Torvalds 17661da177e4SLinus Torvalds lru_add_drain(); 17671da177e4SLinus Torvalds tlb = tlb_gather_mmu(mm, 0); 1768365e9c87SHugh Dickins update_hiwater_rss(mm); 1769508034a3SHugh Dickins unmap_vmas(&tlb, vma, start, end, &nr_accounted, NULL); 17701da177e4SLinus Torvalds vm_unacct_memory(nr_accounted); 177142b77728SJan Beulich free_pgtables(tlb, vma, prev? prev->vm_end: FIRST_USER_ADDRESS, 1772e0da382cSHugh Dickins next? next->vm_start: 0); 17731da177e4SLinus Torvalds tlb_finish_mmu(tlb, start, end); 17741da177e4SLinus Torvalds } 17751da177e4SLinus Torvalds 17761da177e4SLinus Torvalds /* 17771da177e4SLinus Torvalds * Create a list of vma's touched by the unmap, removing them from the mm's 17781da177e4SLinus Torvalds * vma list as we go.. 17791da177e4SLinus Torvalds */ 17801da177e4SLinus Torvalds static void 17811da177e4SLinus Torvalds detach_vmas_to_be_unmapped(struct mm_struct *mm, struct vm_area_struct *vma, 17821da177e4SLinus Torvalds struct vm_area_struct *prev, unsigned long end) 17831da177e4SLinus Torvalds { 17841da177e4SLinus Torvalds struct vm_area_struct **insertion_point; 17851da177e4SLinus Torvalds struct vm_area_struct *tail_vma = NULL; 17861363c3cdSWolfgang Wander unsigned long addr; 17871da177e4SLinus Torvalds 17881da177e4SLinus Torvalds insertion_point = (prev ? &prev->vm_next : &mm->mmap); 17891da177e4SLinus Torvalds do { 17901da177e4SLinus Torvalds rb_erase(&vma->vm_rb, &mm->mm_rb); 17911da177e4SLinus Torvalds mm->map_count--; 17921da177e4SLinus Torvalds tail_vma = vma; 17931da177e4SLinus Torvalds vma = vma->vm_next; 17941da177e4SLinus Torvalds } while (vma && vma->vm_start < end); 17951da177e4SLinus Torvalds *insertion_point = vma; 17961da177e4SLinus Torvalds tail_vma->vm_next = NULL; 17971363c3cdSWolfgang Wander if (mm->unmap_area == arch_unmap_area) 17981363c3cdSWolfgang Wander addr = prev ? prev->vm_end : mm->mmap_base; 17991363c3cdSWolfgang Wander else 18001363c3cdSWolfgang Wander addr = vma ? vma->vm_start : mm->mmap_base; 18011363c3cdSWolfgang Wander mm->unmap_area(mm, addr); 18021da177e4SLinus Torvalds mm->mmap_cache = NULL; /* Kill the cache. */ 18031da177e4SLinus Torvalds } 18041da177e4SLinus Torvalds 18051da177e4SLinus Torvalds /* 18061da177e4SLinus Torvalds * Split a vma into two pieces at address 'addr', a new vma is allocated 180759c51591SMichael Opdenacker * either for the first part or the tail. 18081da177e4SLinus Torvalds */ 18091da177e4SLinus Torvalds int split_vma(struct mm_struct * mm, struct vm_area_struct * vma, 18101da177e4SLinus Torvalds unsigned long addr, int new_below) 18111da177e4SLinus Torvalds { 18121da177e4SLinus Torvalds struct mempolicy *pol; 18131da177e4SLinus Torvalds struct vm_area_struct *new; 18141da177e4SLinus Torvalds 18151da177e4SLinus Torvalds if (is_vm_hugetlb_page(vma) && (addr & ~HPAGE_MASK)) 18161da177e4SLinus Torvalds return -EINVAL; 18171da177e4SLinus Torvalds 18181da177e4SLinus Torvalds if (mm->map_count >= sysctl_max_map_count) 18191da177e4SLinus Torvalds return -ENOMEM; 18201da177e4SLinus Torvalds 1821e94b1766SChristoph Lameter new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL); 18221da177e4SLinus Torvalds if (!new) 18231da177e4SLinus Torvalds return -ENOMEM; 18241da177e4SLinus Torvalds 18251da177e4SLinus Torvalds /* most fields are the same, copy all, and then fixup */ 18261da177e4SLinus Torvalds *new = *vma; 18271da177e4SLinus Torvalds 18281da177e4SLinus Torvalds if (new_below) 18291da177e4SLinus Torvalds new->vm_end = addr; 18301da177e4SLinus Torvalds else { 18311da177e4SLinus Torvalds new->vm_start = addr; 18321da177e4SLinus Torvalds new->vm_pgoff += ((addr - vma->vm_start) >> PAGE_SHIFT); 18331da177e4SLinus Torvalds } 18341da177e4SLinus Torvalds 1835846a16bfSLee Schermerhorn pol = mpol_dup(vma_policy(vma)); 18361da177e4SLinus Torvalds if (IS_ERR(pol)) { 18371da177e4SLinus Torvalds kmem_cache_free(vm_area_cachep, new); 18381da177e4SLinus Torvalds return PTR_ERR(pol); 18391da177e4SLinus Torvalds } 18401da177e4SLinus Torvalds vma_set_policy(new, pol); 18411da177e4SLinus Torvalds 1842925d1c40SMatt Helsley if (new->vm_file) { 18431da177e4SLinus Torvalds get_file(new->vm_file); 1844925d1c40SMatt Helsley if (vma->vm_flags & VM_EXECUTABLE) 1845925d1c40SMatt Helsley added_exe_file_vma(mm); 1846925d1c40SMatt Helsley } 18471da177e4SLinus Torvalds 18481da177e4SLinus Torvalds if (new->vm_ops && new->vm_ops->open) 18491da177e4SLinus Torvalds new->vm_ops->open(new); 18501da177e4SLinus Torvalds 18511da177e4SLinus Torvalds if (new_below) 18521da177e4SLinus Torvalds vma_adjust(vma, addr, vma->vm_end, vma->vm_pgoff + 18531da177e4SLinus Torvalds ((addr - new->vm_start) >> PAGE_SHIFT), new); 18541da177e4SLinus Torvalds else 18551da177e4SLinus Torvalds vma_adjust(vma, vma->vm_start, addr, vma->vm_pgoff, new); 18561da177e4SLinus Torvalds 18571da177e4SLinus Torvalds return 0; 18581da177e4SLinus Torvalds } 18591da177e4SLinus Torvalds 18601da177e4SLinus Torvalds /* Munmap is split into 2 main parts -- this part which finds 18611da177e4SLinus Torvalds * what needs doing, and the areas themselves, which do the 18621da177e4SLinus Torvalds * work. This now handles partial unmappings. 18631da177e4SLinus Torvalds * Jeremy Fitzhardinge <jeremy@goop.org> 18641da177e4SLinus Torvalds */ 18651da177e4SLinus Torvalds int do_munmap(struct mm_struct *mm, unsigned long start, size_t len) 18661da177e4SLinus Torvalds { 18671da177e4SLinus Torvalds unsigned long end; 1868146425a3SHugh Dickins struct vm_area_struct *vma, *prev, *last; 18691da177e4SLinus Torvalds 18701da177e4SLinus Torvalds if ((start & ~PAGE_MASK) || start > TASK_SIZE || len > TASK_SIZE-start) 18711da177e4SLinus Torvalds return -EINVAL; 18721da177e4SLinus Torvalds 18731da177e4SLinus Torvalds if ((len = PAGE_ALIGN(len)) == 0) 18741da177e4SLinus Torvalds return -EINVAL; 18751da177e4SLinus Torvalds 18761da177e4SLinus Torvalds /* Find the first overlapping VMA */ 1877146425a3SHugh Dickins vma = find_vma_prev(mm, start, &prev); 1878146425a3SHugh Dickins if (!vma) 18791da177e4SLinus Torvalds return 0; 1880146425a3SHugh Dickins /* we have start < vma->vm_end */ 18811da177e4SLinus Torvalds 18821da177e4SLinus Torvalds /* if it doesn't overlap, we have nothing.. */ 18831da177e4SLinus Torvalds end = start + len; 1884146425a3SHugh Dickins if (vma->vm_start >= end) 18851da177e4SLinus Torvalds return 0; 18861da177e4SLinus Torvalds 18871da177e4SLinus Torvalds /* 18881da177e4SLinus Torvalds * If we need to split any vma, do it now to save pain later. 18891da177e4SLinus Torvalds * 18901da177e4SLinus Torvalds * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially 18911da177e4SLinus Torvalds * unmapped vm_area_struct will remain in use: so lower split_vma 18921da177e4SLinus Torvalds * places tmp vma above, and higher split_vma places tmp vma below. 18931da177e4SLinus Torvalds */ 1894146425a3SHugh Dickins if (start > vma->vm_start) { 1895146425a3SHugh Dickins int error = split_vma(mm, vma, start, 0); 18961da177e4SLinus Torvalds if (error) 18971da177e4SLinus Torvalds return error; 1898146425a3SHugh Dickins prev = vma; 18991da177e4SLinus Torvalds } 19001da177e4SLinus Torvalds 19011da177e4SLinus Torvalds /* Does it split the last one? */ 19021da177e4SLinus Torvalds last = find_vma(mm, end); 19031da177e4SLinus Torvalds if (last && end > last->vm_start) { 19041da177e4SLinus Torvalds int error = split_vma(mm, last, end, 1); 19051da177e4SLinus Torvalds if (error) 19061da177e4SLinus Torvalds return error; 19071da177e4SLinus Torvalds } 1908146425a3SHugh Dickins vma = prev? prev->vm_next: mm->mmap; 19091da177e4SLinus Torvalds 19101da177e4SLinus Torvalds /* 19111da177e4SLinus Torvalds * Remove the vma's, and unmap the actual pages 19121da177e4SLinus Torvalds */ 1913146425a3SHugh Dickins detach_vmas_to_be_unmapped(mm, vma, prev, end); 1914146425a3SHugh Dickins unmap_region(mm, vma, prev, start, end); 19151da177e4SLinus Torvalds 19161da177e4SLinus Torvalds /* Fix up all other VM information */ 19172c0b3814SHugh Dickins remove_vma_list(mm, vma); 19181da177e4SLinus Torvalds 19191da177e4SLinus Torvalds return 0; 19201da177e4SLinus Torvalds } 19211da177e4SLinus Torvalds 19221da177e4SLinus Torvalds EXPORT_SYMBOL(do_munmap); 19231da177e4SLinus Torvalds 19241da177e4SLinus Torvalds asmlinkage long sys_munmap(unsigned long addr, size_t len) 19251da177e4SLinus Torvalds { 19261da177e4SLinus Torvalds int ret; 19271da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 19281da177e4SLinus Torvalds 19291da177e4SLinus Torvalds profile_munmap(addr); 19301da177e4SLinus Torvalds 19311da177e4SLinus Torvalds down_write(&mm->mmap_sem); 19321da177e4SLinus Torvalds ret = do_munmap(mm, addr, len); 19331da177e4SLinus Torvalds up_write(&mm->mmap_sem); 19341da177e4SLinus Torvalds return ret; 19351da177e4SLinus Torvalds } 19361da177e4SLinus Torvalds 19371da177e4SLinus Torvalds static inline void verify_mm_writelocked(struct mm_struct *mm) 19381da177e4SLinus Torvalds { 1939a241ec65SPaul E. McKenney #ifdef CONFIG_DEBUG_VM 19401da177e4SLinus Torvalds if (unlikely(down_read_trylock(&mm->mmap_sem))) { 19411da177e4SLinus Torvalds WARN_ON(1); 19421da177e4SLinus Torvalds up_read(&mm->mmap_sem); 19431da177e4SLinus Torvalds } 19441da177e4SLinus Torvalds #endif 19451da177e4SLinus Torvalds } 19461da177e4SLinus Torvalds 19471da177e4SLinus Torvalds /* 19481da177e4SLinus Torvalds * this is really a simplified "do_mmap". it only handles 19491da177e4SLinus Torvalds * anonymous maps. eventually we may be able to do some 19501da177e4SLinus Torvalds * brk-specific accounting here. 19511da177e4SLinus Torvalds */ 19521da177e4SLinus Torvalds unsigned long do_brk(unsigned long addr, unsigned long len) 19531da177e4SLinus Torvalds { 19541da177e4SLinus Torvalds struct mm_struct * mm = current->mm; 19551da177e4SLinus Torvalds struct vm_area_struct * vma, * prev; 19561da177e4SLinus Torvalds unsigned long flags; 19571da177e4SLinus Torvalds struct rb_node ** rb_link, * rb_parent; 19581da177e4SLinus Torvalds pgoff_t pgoff = addr >> PAGE_SHIFT; 19593a459756SKirill Korotaev int error; 19601da177e4SLinus Torvalds 19611da177e4SLinus Torvalds len = PAGE_ALIGN(len); 19621da177e4SLinus Torvalds if (!len) 19631da177e4SLinus Torvalds return addr; 19641da177e4SLinus Torvalds 19651da177e4SLinus Torvalds if ((addr + len) > TASK_SIZE || (addr + len) < addr) 19661da177e4SLinus Torvalds return -EINVAL; 19671da177e4SLinus Torvalds 1968cd2579d7SHugh Dickins if (is_hugepage_only_range(mm, addr, len)) 1969cd2579d7SHugh Dickins return -EINVAL; 1970cb07c9a1SDavid Gibson 197188c3f7a8SRichard Knutsson error = security_file_mmap(NULL, 0, 0, 0, addr, 1); 19725a211a5dSEric Paris if (error) 19735a211a5dSEric Paris return error; 19745a211a5dSEric Paris 19753a459756SKirill Korotaev flags = VM_DATA_DEFAULT_FLAGS | VM_ACCOUNT | mm->def_flags; 19763a459756SKirill Korotaev 19773a459756SKirill Korotaev error = arch_mmap_check(addr, len, flags); 19783a459756SKirill Korotaev if (error) 19793a459756SKirill Korotaev return error; 19803a459756SKirill Korotaev 19811da177e4SLinus Torvalds /* 19821da177e4SLinus Torvalds * mlock MCL_FUTURE? 19831da177e4SLinus Torvalds */ 19841da177e4SLinus Torvalds if (mm->def_flags & VM_LOCKED) { 19851da177e4SLinus Torvalds unsigned long locked, lock_limit; 198693ea1d0aSChris Wright locked = len >> PAGE_SHIFT; 198793ea1d0aSChris Wright locked += mm->locked_vm; 19881da177e4SLinus Torvalds lock_limit = current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur; 198993ea1d0aSChris Wright lock_limit >>= PAGE_SHIFT; 19901da177e4SLinus Torvalds if (locked > lock_limit && !capable(CAP_IPC_LOCK)) 19911da177e4SLinus Torvalds return -EAGAIN; 19921da177e4SLinus Torvalds } 19931da177e4SLinus Torvalds 19941da177e4SLinus Torvalds /* 19951da177e4SLinus Torvalds * mm->mmap_sem is required to protect against another thread 19961da177e4SLinus Torvalds * changing the mappings in case we sleep. 19971da177e4SLinus Torvalds */ 19981da177e4SLinus Torvalds verify_mm_writelocked(mm); 19991da177e4SLinus Torvalds 20001da177e4SLinus Torvalds /* 20011da177e4SLinus Torvalds * Clear old maps. this also does some error checking for us 20021da177e4SLinus Torvalds */ 20031da177e4SLinus Torvalds munmap_back: 20041da177e4SLinus Torvalds vma = find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent); 20051da177e4SLinus Torvalds if (vma && vma->vm_start < addr + len) { 20061da177e4SLinus Torvalds if (do_munmap(mm, addr, len)) 20071da177e4SLinus Torvalds return -ENOMEM; 20081da177e4SLinus Torvalds goto munmap_back; 20091da177e4SLinus Torvalds } 20101da177e4SLinus Torvalds 20111da177e4SLinus Torvalds /* Check against address space limits *after* clearing old maps... */ 2012119f657cSakpm@osdl.org if (!may_expand_vm(mm, len >> PAGE_SHIFT)) 20131da177e4SLinus Torvalds return -ENOMEM; 20141da177e4SLinus Torvalds 20151da177e4SLinus Torvalds if (mm->map_count > sysctl_max_map_count) 20161da177e4SLinus Torvalds return -ENOMEM; 20171da177e4SLinus Torvalds 20181da177e4SLinus Torvalds if (security_vm_enough_memory(len >> PAGE_SHIFT)) 20191da177e4SLinus Torvalds return -ENOMEM; 20201da177e4SLinus Torvalds 20211da177e4SLinus Torvalds /* Can we just expand an old private anonymous mapping? */ 20221da177e4SLinus Torvalds if (vma_merge(mm, prev, addr, addr + len, flags, 20231da177e4SLinus Torvalds NULL, NULL, pgoff, NULL)) 20241da177e4SLinus Torvalds goto out; 20251da177e4SLinus Torvalds 20261da177e4SLinus Torvalds /* 20271da177e4SLinus Torvalds * create a vma struct for an anonymous mapping 20281da177e4SLinus Torvalds */ 2029c5e3b83eSPekka Enberg vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL); 20301da177e4SLinus Torvalds if (!vma) { 20311da177e4SLinus Torvalds vm_unacct_memory(len >> PAGE_SHIFT); 20321da177e4SLinus Torvalds return -ENOMEM; 20331da177e4SLinus Torvalds } 20341da177e4SLinus Torvalds 20351da177e4SLinus Torvalds vma->vm_mm = mm; 20361da177e4SLinus Torvalds vma->vm_start = addr; 20371da177e4SLinus Torvalds vma->vm_end = addr + len; 20381da177e4SLinus Torvalds vma->vm_pgoff = pgoff; 20391da177e4SLinus Torvalds vma->vm_flags = flags; 20403ed75eb8SColy Li vma->vm_page_prot = vm_get_page_prot(flags); 20411da177e4SLinus Torvalds vma_link(mm, vma, prev, rb_link, rb_parent); 20421da177e4SLinus Torvalds out: 20431da177e4SLinus Torvalds mm->total_vm += len >> PAGE_SHIFT; 20441da177e4SLinus Torvalds if (flags & VM_LOCKED) { 20451da177e4SLinus Torvalds mm->locked_vm += len >> PAGE_SHIFT; 20461da177e4SLinus Torvalds make_pages_present(addr, addr + len); 20471da177e4SLinus Torvalds } 20481da177e4SLinus Torvalds return addr; 20491da177e4SLinus Torvalds } 20501da177e4SLinus Torvalds 20511da177e4SLinus Torvalds EXPORT_SYMBOL(do_brk); 20521da177e4SLinus Torvalds 20531da177e4SLinus Torvalds /* Release all mmaps. */ 20541da177e4SLinus Torvalds void exit_mmap(struct mm_struct *mm) 20551da177e4SLinus Torvalds { 20561da177e4SLinus Torvalds struct mmu_gather *tlb; 2057e0da382cSHugh Dickins struct vm_area_struct *vma = mm->mmap; 20581da177e4SLinus Torvalds unsigned long nr_accounted = 0; 2059ee39b37bSHugh Dickins unsigned long end; 20601da177e4SLinus Torvalds 2061d6dd61c8SJeremy Fitzhardinge /* mm's last user has gone, and its about to be pulled down */ 2062d6dd61c8SJeremy Fitzhardinge arch_exit_mmap(mm); 2063d6dd61c8SJeremy Fitzhardinge 20641da177e4SLinus Torvalds lru_add_drain(); 20651da177e4SLinus Torvalds flush_cache_mm(mm); 2066e0da382cSHugh Dickins tlb = tlb_gather_mmu(mm, 1); 2067365e9c87SHugh Dickins /* Don't update_hiwater_rss(mm) here, do_exit already did */ 2068e0da382cSHugh Dickins /* Use -1 here to ensure all VMAs in the mm are unmapped */ 2069508034a3SHugh Dickins end = unmap_vmas(&tlb, vma, 0, -1, &nr_accounted, NULL); 20701da177e4SLinus Torvalds vm_unacct_memory(nr_accounted); 207142b77728SJan Beulich free_pgtables(tlb, vma, FIRST_USER_ADDRESS, 0); 2072ee39b37bSHugh Dickins tlb_finish_mmu(tlb, 0, end); 20731da177e4SLinus Torvalds 20741da177e4SLinus Torvalds /* 20758f4f8c16SHugh Dickins * Walk the list again, actually closing and freeing it, 20768f4f8c16SHugh Dickins * with preemption enabled, without holding any MM locks. 20771da177e4SLinus Torvalds */ 2078a8fb5618SHugh Dickins while (vma) 2079a8fb5618SHugh Dickins vma = remove_vma(vma); 2080e0da382cSHugh Dickins 2081e2cdef8cSHugh Dickins BUG_ON(mm->nr_ptes > (FIRST_USER_ADDRESS+PMD_SIZE-1)>>PMD_SHIFT); 20821da177e4SLinus Torvalds } 20831da177e4SLinus Torvalds 20841da177e4SLinus Torvalds /* Insert vm structure into process list sorted by address 20851da177e4SLinus Torvalds * and into the inode's i_mmap tree. If vm_file is non-NULL 20861da177e4SLinus Torvalds * then i_mmap_lock is taken here. 20871da177e4SLinus Torvalds */ 20881da177e4SLinus Torvalds int insert_vm_struct(struct mm_struct * mm, struct vm_area_struct * vma) 20891da177e4SLinus Torvalds { 20901da177e4SLinus Torvalds struct vm_area_struct * __vma, * prev; 20911da177e4SLinus Torvalds struct rb_node ** rb_link, * rb_parent; 20921da177e4SLinus Torvalds 20931da177e4SLinus Torvalds /* 20941da177e4SLinus Torvalds * The vm_pgoff of a purely anonymous vma should be irrelevant 20951da177e4SLinus Torvalds * until its first write fault, when page's anon_vma and index 20961da177e4SLinus Torvalds * are set. But now set the vm_pgoff it will almost certainly 20971da177e4SLinus Torvalds * end up with (unless mremap moves it elsewhere before that 20981da177e4SLinus Torvalds * first wfault), so /proc/pid/maps tells a consistent story. 20991da177e4SLinus Torvalds * 21001da177e4SLinus Torvalds * By setting it to reflect the virtual start address of the 21011da177e4SLinus Torvalds * vma, merges and splits can happen in a seamless way, just 21021da177e4SLinus Torvalds * using the existing file pgoff checks and manipulations. 21031da177e4SLinus Torvalds * Similarly in do_mmap_pgoff and in do_brk. 21041da177e4SLinus Torvalds */ 21051da177e4SLinus Torvalds if (!vma->vm_file) { 21061da177e4SLinus Torvalds BUG_ON(vma->anon_vma); 21071da177e4SLinus Torvalds vma->vm_pgoff = vma->vm_start >> PAGE_SHIFT; 21081da177e4SLinus Torvalds } 21091da177e4SLinus Torvalds __vma = find_vma_prepare(mm,vma->vm_start,&prev,&rb_link,&rb_parent); 21101da177e4SLinus Torvalds if (__vma && __vma->vm_start < vma->vm_end) 21111da177e4SLinus Torvalds return -ENOMEM; 21122fd4ef85SHugh Dickins if ((vma->vm_flags & VM_ACCOUNT) && 211334b4e4aaSAlan Cox security_vm_enough_memory_mm(mm, vma_pages(vma))) 21142fd4ef85SHugh Dickins return -ENOMEM; 21151da177e4SLinus Torvalds vma_link(mm, vma, prev, rb_link, rb_parent); 21161da177e4SLinus Torvalds return 0; 21171da177e4SLinus Torvalds } 21181da177e4SLinus Torvalds 21191da177e4SLinus Torvalds /* 21201da177e4SLinus Torvalds * Copy the vma structure to a new location in the same mm, 21211da177e4SLinus Torvalds * prior to moving page table entries, to effect an mremap move. 21221da177e4SLinus Torvalds */ 21231da177e4SLinus Torvalds struct vm_area_struct *copy_vma(struct vm_area_struct **vmap, 21241da177e4SLinus Torvalds unsigned long addr, unsigned long len, pgoff_t pgoff) 21251da177e4SLinus Torvalds { 21261da177e4SLinus Torvalds struct vm_area_struct *vma = *vmap; 21271da177e4SLinus Torvalds unsigned long vma_start = vma->vm_start; 21281da177e4SLinus Torvalds struct mm_struct *mm = vma->vm_mm; 21291da177e4SLinus Torvalds struct vm_area_struct *new_vma, *prev; 21301da177e4SLinus Torvalds struct rb_node **rb_link, *rb_parent; 21311da177e4SLinus Torvalds struct mempolicy *pol; 21321da177e4SLinus Torvalds 21331da177e4SLinus Torvalds /* 21341da177e4SLinus Torvalds * If anonymous vma has not yet been faulted, update new pgoff 21351da177e4SLinus Torvalds * to match new location, to increase its chance of merging. 21361da177e4SLinus Torvalds */ 21371da177e4SLinus Torvalds if (!vma->vm_file && !vma->anon_vma) 21381da177e4SLinus Torvalds pgoff = addr >> PAGE_SHIFT; 21391da177e4SLinus Torvalds 21401da177e4SLinus Torvalds find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent); 21411da177e4SLinus Torvalds new_vma = vma_merge(mm, prev, addr, addr + len, vma->vm_flags, 21421da177e4SLinus Torvalds vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma)); 21431da177e4SLinus Torvalds if (new_vma) { 21441da177e4SLinus Torvalds /* 21451da177e4SLinus Torvalds * Source vma may have been merged into new_vma 21461da177e4SLinus Torvalds */ 21471da177e4SLinus Torvalds if (vma_start >= new_vma->vm_start && 21481da177e4SLinus Torvalds vma_start < new_vma->vm_end) 21491da177e4SLinus Torvalds *vmap = new_vma; 21501da177e4SLinus Torvalds } else { 2151e94b1766SChristoph Lameter new_vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL); 21521da177e4SLinus Torvalds if (new_vma) { 21531da177e4SLinus Torvalds *new_vma = *vma; 2154846a16bfSLee Schermerhorn pol = mpol_dup(vma_policy(vma)); 21551da177e4SLinus Torvalds if (IS_ERR(pol)) { 21561da177e4SLinus Torvalds kmem_cache_free(vm_area_cachep, new_vma); 21571da177e4SLinus Torvalds return NULL; 21581da177e4SLinus Torvalds } 21591da177e4SLinus Torvalds vma_set_policy(new_vma, pol); 21601da177e4SLinus Torvalds new_vma->vm_start = addr; 21611da177e4SLinus Torvalds new_vma->vm_end = addr + len; 21621da177e4SLinus Torvalds new_vma->vm_pgoff = pgoff; 2163925d1c40SMatt Helsley if (new_vma->vm_file) { 21641da177e4SLinus Torvalds get_file(new_vma->vm_file); 2165925d1c40SMatt Helsley if (vma->vm_flags & VM_EXECUTABLE) 2166925d1c40SMatt Helsley added_exe_file_vma(mm); 2167925d1c40SMatt Helsley } 21681da177e4SLinus Torvalds if (new_vma->vm_ops && new_vma->vm_ops->open) 21691da177e4SLinus Torvalds new_vma->vm_ops->open(new_vma); 21701da177e4SLinus Torvalds vma_link(mm, new_vma, prev, rb_link, rb_parent); 21711da177e4SLinus Torvalds } 21721da177e4SLinus Torvalds } 21731da177e4SLinus Torvalds return new_vma; 21741da177e4SLinus Torvalds } 2175119f657cSakpm@osdl.org 2176119f657cSakpm@osdl.org /* 2177119f657cSakpm@osdl.org * Return true if the calling process may expand its vm space by the passed 2178119f657cSakpm@osdl.org * number of pages 2179119f657cSakpm@osdl.org */ 2180119f657cSakpm@osdl.org int may_expand_vm(struct mm_struct *mm, unsigned long npages) 2181119f657cSakpm@osdl.org { 2182119f657cSakpm@osdl.org unsigned long cur = mm->total_vm; /* pages */ 2183119f657cSakpm@osdl.org unsigned long lim; 2184119f657cSakpm@osdl.org 2185119f657cSakpm@osdl.org lim = current->signal->rlim[RLIMIT_AS].rlim_cur >> PAGE_SHIFT; 2186119f657cSakpm@osdl.org 2187119f657cSakpm@osdl.org if (cur + npages > lim) 2188119f657cSakpm@osdl.org return 0; 2189119f657cSakpm@osdl.org return 1; 2190119f657cSakpm@osdl.org } 2191fa5dc22fSRoland McGrath 2192fa5dc22fSRoland McGrath 2193b1d0e4f5SNick Piggin static int special_mapping_fault(struct vm_area_struct *vma, 2194b1d0e4f5SNick Piggin struct vm_fault *vmf) 2195fa5dc22fSRoland McGrath { 2196b1d0e4f5SNick Piggin pgoff_t pgoff; 2197fa5dc22fSRoland McGrath struct page **pages; 2198fa5dc22fSRoland McGrath 2199b1d0e4f5SNick Piggin /* 2200b1d0e4f5SNick Piggin * special mappings have no vm_file, and in that case, the mm 2201b1d0e4f5SNick Piggin * uses vm_pgoff internally. So we have to subtract it from here. 2202b1d0e4f5SNick Piggin * We are allowed to do this because we are the mm; do not copy 2203b1d0e4f5SNick Piggin * this code into drivers! 2204b1d0e4f5SNick Piggin */ 2205b1d0e4f5SNick Piggin pgoff = vmf->pgoff - vma->vm_pgoff; 2206fa5dc22fSRoland McGrath 2207b1d0e4f5SNick Piggin for (pages = vma->vm_private_data; pgoff && *pages; ++pages) 2208b1d0e4f5SNick Piggin pgoff--; 2209fa5dc22fSRoland McGrath 2210fa5dc22fSRoland McGrath if (*pages) { 2211fa5dc22fSRoland McGrath struct page *page = *pages; 2212fa5dc22fSRoland McGrath get_page(page); 2213b1d0e4f5SNick Piggin vmf->page = page; 2214b1d0e4f5SNick Piggin return 0; 2215fa5dc22fSRoland McGrath } 2216fa5dc22fSRoland McGrath 2217b1d0e4f5SNick Piggin return VM_FAULT_SIGBUS; 2218fa5dc22fSRoland McGrath } 2219fa5dc22fSRoland McGrath 2220fa5dc22fSRoland McGrath /* 2221fa5dc22fSRoland McGrath * Having a close hook prevents vma merging regardless of flags. 2222fa5dc22fSRoland McGrath */ 2223fa5dc22fSRoland McGrath static void special_mapping_close(struct vm_area_struct *vma) 2224fa5dc22fSRoland McGrath { 2225fa5dc22fSRoland McGrath } 2226fa5dc22fSRoland McGrath 2227fa5dc22fSRoland McGrath static struct vm_operations_struct special_mapping_vmops = { 2228fa5dc22fSRoland McGrath .close = special_mapping_close, 2229b1d0e4f5SNick Piggin .fault = special_mapping_fault, 2230fa5dc22fSRoland McGrath }; 2231fa5dc22fSRoland McGrath 2232fa5dc22fSRoland McGrath /* 2233fa5dc22fSRoland McGrath * Called with mm->mmap_sem held for writing. 2234fa5dc22fSRoland McGrath * Insert a new vma covering the given region, with the given flags. 2235fa5dc22fSRoland McGrath * Its pages are supplied by the given array of struct page *. 2236fa5dc22fSRoland McGrath * The array can be shorter than len >> PAGE_SHIFT if it's null-terminated. 2237fa5dc22fSRoland McGrath * The region past the last page supplied will always produce SIGBUS. 2238fa5dc22fSRoland McGrath * The array pointer and the pages it points to are assumed to stay alive 2239fa5dc22fSRoland McGrath * for as long as this mapping might exist. 2240fa5dc22fSRoland McGrath */ 2241fa5dc22fSRoland McGrath int install_special_mapping(struct mm_struct *mm, 2242fa5dc22fSRoland McGrath unsigned long addr, unsigned long len, 2243fa5dc22fSRoland McGrath unsigned long vm_flags, struct page **pages) 2244fa5dc22fSRoland McGrath { 2245fa5dc22fSRoland McGrath struct vm_area_struct *vma; 2246fa5dc22fSRoland McGrath 2247fa5dc22fSRoland McGrath vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL); 2248fa5dc22fSRoland McGrath if (unlikely(vma == NULL)) 2249fa5dc22fSRoland McGrath return -ENOMEM; 2250fa5dc22fSRoland McGrath 2251fa5dc22fSRoland McGrath vma->vm_mm = mm; 2252fa5dc22fSRoland McGrath vma->vm_start = addr; 2253fa5dc22fSRoland McGrath vma->vm_end = addr + len; 2254fa5dc22fSRoland McGrath 22552f98735cSNick Piggin vma->vm_flags = vm_flags | mm->def_flags | VM_DONTEXPAND; 22563ed75eb8SColy Li vma->vm_page_prot = vm_get_page_prot(vma->vm_flags); 2257fa5dc22fSRoland McGrath 2258fa5dc22fSRoland McGrath vma->vm_ops = &special_mapping_vmops; 2259fa5dc22fSRoland McGrath vma->vm_private_data = pages; 2260fa5dc22fSRoland McGrath 2261fa5dc22fSRoland McGrath if (unlikely(insert_vm_struct(mm, vma))) { 2262fa5dc22fSRoland McGrath kmem_cache_free(vm_area_cachep, vma); 2263fa5dc22fSRoland McGrath return -ENOMEM; 2264fa5dc22fSRoland McGrath } 2265fa5dc22fSRoland McGrath 2266fa5dc22fSRoland McGrath mm->total_vm += len >> PAGE_SHIFT; 2267fa5dc22fSRoland McGrath 2268fa5dc22fSRoland McGrath return 0; 2269fa5dc22fSRoland McGrath } 2270