11da177e4SLinus Torvalds /* 21da177e4SLinus Torvalds * mm/mmap.c 31da177e4SLinus Torvalds * 41da177e4SLinus Torvalds * Written by obz. 51da177e4SLinus Torvalds * 6046c6884SAlan Cox * Address space accounting code <alan@lxorguk.ukuu.org.uk> 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> 29cddb8a5cSAndrea Arcangeli #include <linux/mmu_notifier.h> 30cdd6c482SIngo Molnar #include <linux/perf_event.h> 31120a795dSAl Viro #include <linux/audit.h> 32b15d00b6SAndrea Arcangeli #include <linux/khugepaged.h> 331da177e4SLinus Torvalds 341da177e4SLinus Torvalds #include <asm/uaccess.h> 351da177e4SLinus Torvalds #include <asm/cacheflush.h> 361da177e4SLinus Torvalds #include <asm/tlb.h> 37d6dd61c8SJeremy Fitzhardinge #include <asm/mmu_context.h> 381da177e4SLinus Torvalds 3942b77728SJan Beulich #include "internal.h" 4042b77728SJan Beulich 413a459756SKirill Korotaev #ifndef arch_mmap_check 423a459756SKirill Korotaev #define arch_mmap_check(addr, len, flags) (0) 433a459756SKirill Korotaev #endif 443a459756SKirill Korotaev 4508e7d9b5SMartin Schwidefsky #ifndef arch_rebalance_pgtables 4608e7d9b5SMartin Schwidefsky #define arch_rebalance_pgtables(addr, len) (addr) 4708e7d9b5SMartin Schwidefsky #endif 4808e7d9b5SMartin Schwidefsky 49e0da382cSHugh Dickins static void unmap_region(struct mm_struct *mm, 50e0da382cSHugh Dickins struct vm_area_struct *vma, struct vm_area_struct *prev, 51e0da382cSHugh Dickins unsigned long start, unsigned long end); 52e0da382cSHugh Dickins 531da177e4SLinus Torvalds /* 541da177e4SLinus Torvalds * WARNING: the debugging will use recursive algorithms so never enable this 551da177e4SLinus Torvalds * unless you know what you are doing. 561da177e4SLinus Torvalds */ 571da177e4SLinus Torvalds #undef DEBUG_MM_RB 581da177e4SLinus Torvalds 591da177e4SLinus Torvalds /* description of effects of mapping type and prot in current implementation. 601da177e4SLinus Torvalds * this is due to the limited x86 page protection hardware. The expected 611da177e4SLinus Torvalds * behavior is in parens: 621da177e4SLinus Torvalds * 631da177e4SLinus Torvalds * map_type prot 641da177e4SLinus Torvalds * PROT_NONE PROT_READ PROT_WRITE PROT_EXEC 651da177e4SLinus Torvalds * MAP_SHARED r: (no) no r: (yes) yes r: (no) yes r: (no) yes 661da177e4SLinus Torvalds * w: (no) no w: (no) no w: (yes) yes w: (no) no 671da177e4SLinus Torvalds * x: (no) no x: (no) yes x: (no) yes x: (yes) yes 681da177e4SLinus Torvalds * 691da177e4SLinus Torvalds * MAP_PRIVATE r: (no) no r: (yes) yes r: (no) yes r: (no) yes 701da177e4SLinus Torvalds * w: (no) no w: (no) no w: (copy) copy w: (no) no 711da177e4SLinus Torvalds * x: (no) no x: (no) yes x: (no) yes x: (yes) yes 721da177e4SLinus Torvalds * 731da177e4SLinus Torvalds */ 741da177e4SLinus Torvalds pgprot_t protection_map[16] = { 751da177e4SLinus Torvalds __P000, __P001, __P010, __P011, __P100, __P101, __P110, __P111, 761da177e4SLinus Torvalds __S000, __S001, __S010, __S011, __S100, __S101, __S110, __S111 771da177e4SLinus Torvalds }; 781da177e4SLinus Torvalds 79804af2cfSHugh Dickins pgprot_t vm_get_page_prot(unsigned long vm_flags) 80804af2cfSHugh Dickins { 81b845f313SDave Kleikamp return __pgprot(pgprot_val(protection_map[vm_flags & 82b845f313SDave Kleikamp (VM_READ|VM_WRITE|VM_EXEC|VM_SHARED)]) | 83b845f313SDave Kleikamp pgprot_val(arch_vm_get_page_prot(vm_flags))); 84804af2cfSHugh Dickins } 85804af2cfSHugh Dickins EXPORT_SYMBOL(vm_get_page_prot); 86804af2cfSHugh Dickins 8734679d7eSShaohua Li int sysctl_overcommit_memory __read_mostly = OVERCOMMIT_GUESS; /* heuristic overcommit */ 8834679d7eSShaohua Li int sysctl_overcommit_ratio __read_mostly = 50; /* default is 50% */ 89c3d8c141SChristoph Lameter int sysctl_max_map_count __read_mostly = DEFAULT_MAX_MAP_COUNT; 9034679d7eSShaohua Li /* 9134679d7eSShaohua Li * Make sure vm_committed_as in one cacheline and not cacheline shared with 9234679d7eSShaohua Li * other variables. It can be updated by several CPUs frequently. 9334679d7eSShaohua Li */ 9434679d7eSShaohua Li struct percpu_counter vm_committed_as ____cacheline_aligned_in_smp; 951da177e4SLinus Torvalds 961da177e4SLinus Torvalds /* 971da177e4SLinus Torvalds * Check that a process has enough memory to allocate a new virtual 981da177e4SLinus Torvalds * mapping. 0 means there is enough memory for the allocation to 991da177e4SLinus Torvalds * succeed and -ENOMEM implies there is not. 1001da177e4SLinus Torvalds * 1011da177e4SLinus Torvalds * We currently support three overcommit policies, which are set via the 1021da177e4SLinus Torvalds * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting 1031da177e4SLinus Torvalds * 1041da177e4SLinus Torvalds * Strict overcommit modes added 2002 Feb 26 by Alan Cox. 1051da177e4SLinus Torvalds * Additional code 2002 Jul 20 by Robert Love. 1061da177e4SLinus Torvalds * 1071da177e4SLinus Torvalds * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise. 1081da177e4SLinus Torvalds * 1091da177e4SLinus Torvalds * Note this is a helper function intended to be used by LSMs which 1101da177e4SLinus Torvalds * wish to use this logic. 1111da177e4SLinus Torvalds */ 11234b4e4aaSAlan Cox int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin) 1131da177e4SLinus Torvalds { 1141da177e4SLinus Torvalds unsigned long free, allowed; 1151da177e4SLinus Torvalds 1161da177e4SLinus Torvalds vm_acct_memory(pages); 1171da177e4SLinus Torvalds 1181da177e4SLinus Torvalds /* 1191da177e4SLinus Torvalds * Sometimes we want to use more memory than we have 1201da177e4SLinus Torvalds */ 1211da177e4SLinus Torvalds if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS) 1221da177e4SLinus Torvalds return 0; 1231da177e4SLinus Torvalds 1241da177e4SLinus Torvalds if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) { 1251da177e4SLinus Torvalds unsigned long n; 1261da177e4SLinus Torvalds 127347ce434SChristoph Lameter free = global_page_state(NR_FILE_PAGES); 1281da177e4SLinus Torvalds free += nr_swap_pages; 1291da177e4SLinus Torvalds 1301da177e4SLinus Torvalds /* 1311da177e4SLinus Torvalds * Any slabs which are created with the 1321da177e4SLinus Torvalds * SLAB_RECLAIM_ACCOUNT flag claim to have contents 1331da177e4SLinus Torvalds * which are reclaimable, under pressure. The dentry 1341da177e4SLinus Torvalds * cache and most inode caches should fall into this 1351da177e4SLinus Torvalds */ 136972d1a7bSChristoph Lameter free += global_page_state(NR_SLAB_RECLAIMABLE); 1371da177e4SLinus Torvalds 1381da177e4SLinus Torvalds /* 1391da177e4SLinus Torvalds * Leave the last 3% for root 1401da177e4SLinus Torvalds */ 1411da177e4SLinus Torvalds if (!cap_sys_admin) 1421da177e4SLinus Torvalds free -= free / 32; 1431da177e4SLinus Torvalds 1441da177e4SLinus Torvalds if (free > pages) 1451da177e4SLinus Torvalds return 0; 1461da177e4SLinus Torvalds 1471da177e4SLinus Torvalds /* 1481da177e4SLinus Torvalds * nr_free_pages() is very expensive on large systems, 1491da177e4SLinus Torvalds * only call if we're about to fail. 1501da177e4SLinus Torvalds */ 1511da177e4SLinus Torvalds n = nr_free_pages(); 1526d9f7839SHideo AOKI 1536d9f7839SHideo AOKI /* 1546d9f7839SHideo AOKI * Leave reserved pages. The pages are not for anonymous pages. 1556d9f7839SHideo AOKI */ 1566d9f7839SHideo AOKI if (n <= totalreserve_pages) 1576d9f7839SHideo AOKI goto error; 1586d9f7839SHideo AOKI else 1596d9f7839SHideo AOKI n -= totalreserve_pages; 1606d9f7839SHideo AOKI 1616d9f7839SHideo AOKI /* 1626d9f7839SHideo AOKI * Leave the last 3% for root 1636d9f7839SHideo AOKI */ 1641da177e4SLinus Torvalds if (!cap_sys_admin) 1651da177e4SLinus Torvalds n -= n / 32; 1661da177e4SLinus Torvalds free += n; 1671da177e4SLinus Torvalds 1681da177e4SLinus Torvalds if (free > pages) 1691da177e4SLinus Torvalds return 0; 1706d9f7839SHideo AOKI 1716d9f7839SHideo AOKI goto error; 1721da177e4SLinus Torvalds } 1731da177e4SLinus Torvalds 1741da177e4SLinus Torvalds allowed = (totalram_pages - hugetlb_total_pages()) 1751da177e4SLinus Torvalds * sysctl_overcommit_ratio / 100; 1761da177e4SLinus Torvalds /* 1771da177e4SLinus Torvalds * Leave the last 3% for root 1781da177e4SLinus Torvalds */ 1791da177e4SLinus Torvalds if (!cap_sys_admin) 1801da177e4SLinus Torvalds allowed -= allowed / 32; 1811da177e4SLinus Torvalds allowed += total_swap_pages; 1821da177e4SLinus Torvalds 1831da177e4SLinus Torvalds /* Don't let a single process grow too big: 1841da177e4SLinus Torvalds leave 3% of the size of this process for other processes */ 185731572d3SAlan Cox if (mm) 18634b4e4aaSAlan Cox allowed -= mm->total_vm / 32; 1871da177e4SLinus Torvalds 18800a62ce9SKOSAKI Motohiro if (percpu_counter_read_positive(&vm_committed_as) < allowed) 1891da177e4SLinus Torvalds return 0; 1906d9f7839SHideo AOKI error: 1911da177e4SLinus Torvalds vm_unacct_memory(pages); 1921da177e4SLinus Torvalds 1931da177e4SLinus Torvalds return -ENOMEM; 1941da177e4SLinus Torvalds } 1951da177e4SLinus Torvalds 1961da177e4SLinus Torvalds /* 1971da177e4SLinus Torvalds * Requires inode->i_mapping->i_mmap_lock 1981da177e4SLinus Torvalds */ 1991da177e4SLinus Torvalds static void __remove_shared_vm_struct(struct vm_area_struct *vma, 2001da177e4SLinus Torvalds struct file *file, struct address_space *mapping) 2011da177e4SLinus Torvalds { 2021da177e4SLinus Torvalds if (vma->vm_flags & VM_DENYWRITE) 203d3ac7f89SJosef "Jeff" Sipek atomic_inc(&file->f_path.dentry->d_inode->i_writecount); 2041da177e4SLinus Torvalds if (vma->vm_flags & VM_SHARED) 2051da177e4SLinus Torvalds mapping->i_mmap_writable--; 2061da177e4SLinus Torvalds 2071da177e4SLinus Torvalds flush_dcache_mmap_lock(mapping); 2081da177e4SLinus Torvalds if (unlikely(vma->vm_flags & VM_NONLINEAR)) 2091da177e4SLinus Torvalds list_del_init(&vma->shared.vm_set.list); 2101da177e4SLinus Torvalds else 2111da177e4SLinus Torvalds vma_prio_tree_remove(vma, &mapping->i_mmap); 2121da177e4SLinus Torvalds flush_dcache_mmap_unlock(mapping); 2131da177e4SLinus Torvalds } 2141da177e4SLinus Torvalds 2151da177e4SLinus Torvalds /* 216a8fb5618SHugh Dickins * Unlink a file-based vm structure from its prio_tree, to hide 217a8fb5618SHugh Dickins * vma from rmap and vmtruncate before freeing its page tables. 2181da177e4SLinus Torvalds */ 219a8fb5618SHugh Dickins void unlink_file_vma(struct vm_area_struct *vma) 2201da177e4SLinus Torvalds { 2211da177e4SLinus Torvalds struct file *file = vma->vm_file; 2221da177e4SLinus Torvalds 2231da177e4SLinus Torvalds if (file) { 2241da177e4SLinus Torvalds struct address_space *mapping = file->f_mapping; 2251da177e4SLinus Torvalds spin_lock(&mapping->i_mmap_lock); 2261da177e4SLinus Torvalds __remove_shared_vm_struct(vma, file, mapping); 2271da177e4SLinus Torvalds spin_unlock(&mapping->i_mmap_lock); 2281da177e4SLinus Torvalds } 229a8fb5618SHugh Dickins } 230a8fb5618SHugh Dickins 231a8fb5618SHugh Dickins /* 232a8fb5618SHugh Dickins * Close a vm structure and free it, returning the next. 233a8fb5618SHugh Dickins */ 234a8fb5618SHugh Dickins static struct vm_area_struct *remove_vma(struct vm_area_struct *vma) 235a8fb5618SHugh Dickins { 236a8fb5618SHugh Dickins struct vm_area_struct *next = vma->vm_next; 237a8fb5618SHugh Dickins 238a8fb5618SHugh Dickins might_sleep(); 2391da177e4SLinus Torvalds if (vma->vm_ops && vma->vm_ops->close) 2401da177e4SLinus Torvalds vma->vm_ops->close(vma); 241925d1c40SMatt Helsley if (vma->vm_file) { 242a8fb5618SHugh Dickins fput(vma->vm_file); 243925d1c40SMatt Helsley if (vma->vm_flags & VM_EXECUTABLE) 244925d1c40SMatt Helsley removed_exe_file_vma(vma->vm_mm); 245925d1c40SMatt Helsley } 246f0be3d32SLee Schermerhorn mpol_put(vma_policy(vma)); 2471da177e4SLinus Torvalds kmem_cache_free(vm_area_cachep, vma); 248a8fb5618SHugh Dickins return next; 2491da177e4SLinus Torvalds } 2501da177e4SLinus Torvalds 2516a6160a7SHeiko Carstens SYSCALL_DEFINE1(brk, unsigned long, brk) 2521da177e4SLinus Torvalds { 2531da177e4SLinus Torvalds unsigned long rlim, retval; 2541da177e4SLinus Torvalds unsigned long newbrk, oldbrk; 2551da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 256a5b4592cSJiri Kosina unsigned long min_brk; 2571da177e4SLinus Torvalds 2581da177e4SLinus Torvalds down_write(&mm->mmap_sem); 2591da177e4SLinus Torvalds 260a5b4592cSJiri Kosina #ifdef CONFIG_COMPAT_BRK 2615520e894SJiri Kosina /* 2625520e894SJiri Kosina * CONFIG_COMPAT_BRK can still be overridden by setting 2635520e894SJiri Kosina * randomize_va_space to 2, which will still cause mm->start_brk 2645520e894SJiri Kosina * to be arbitrarily shifted 2655520e894SJiri Kosina */ 2664471a675SJiri Kosina if (current->brk_randomized) 2675520e894SJiri Kosina min_brk = mm->start_brk; 2685520e894SJiri Kosina else 2695520e894SJiri Kosina min_brk = mm->end_data; 270a5b4592cSJiri Kosina #else 271a5b4592cSJiri Kosina min_brk = mm->start_brk; 272a5b4592cSJiri Kosina #endif 273a5b4592cSJiri Kosina if (brk < min_brk) 2741da177e4SLinus Torvalds goto out; 2751e624196SRam Gupta 2761e624196SRam Gupta /* 2771e624196SRam Gupta * Check against rlimit here. If this check is done later after the test 2781e624196SRam Gupta * of oldbrk with newbrk then it can escape the test and let the data 2791e624196SRam Gupta * segment grow beyond its set limit the in case where the limit is 2801e624196SRam Gupta * not page aligned -Ram Gupta 2811e624196SRam Gupta */ 28259e99e5bSJiri Slaby rlim = rlimit(RLIMIT_DATA); 283c1d171a0SJiri Kosina if (rlim < RLIM_INFINITY && (brk - mm->start_brk) + 284c1d171a0SJiri Kosina (mm->end_data - mm->start_data) > rlim) 2851e624196SRam Gupta goto out; 2861e624196SRam Gupta 2871da177e4SLinus Torvalds newbrk = PAGE_ALIGN(brk); 2881da177e4SLinus Torvalds oldbrk = PAGE_ALIGN(mm->brk); 2891da177e4SLinus Torvalds if (oldbrk == newbrk) 2901da177e4SLinus Torvalds goto set_brk; 2911da177e4SLinus Torvalds 2921da177e4SLinus Torvalds /* Always allow shrinking brk. */ 2931da177e4SLinus Torvalds if (brk <= mm->brk) { 2941da177e4SLinus Torvalds if (!do_munmap(mm, newbrk, oldbrk-newbrk)) 2951da177e4SLinus Torvalds goto set_brk; 2961da177e4SLinus Torvalds goto out; 2971da177e4SLinus Torvalds } 2981da177e4SLinus Torvalds 2991da177e4SLinus Torvalds /* Check against existing mmap mappings. */ 3001da177e4SLinus Torvalds if (find_vma_intersection(mm, oldbrk, newbrk+PAGE_SIZE)) 3011da177e4SLinus Torvalds goto out; 3021da177e4SLinus Torvalds 3031da177e4SLinus Torvalds /* Ok, looks good - let it rip. */ 3041da177e4SLinus Torvalds if (do_brk(oldbrk, newbrk-oldbrk) != oldbrk) 3051da177e4SLinus Torvalds goto out; 3061da177e4SLinus Torvalds set_brk: 3071da177e4SLinus Torvalds mm->brk = brk; 3081da177e4SLinus Torvalds out: 3091da177e4SLinus Torvalds retval = mm->brk; 3101da177e4SLinus Torvalds up_write(&mm->mmap_sem); 3111da177e4SLinus Torvalds return retval; 3121da177e4SLinus Torvalds } 3131da177e4SLinus Torvalds 3141da177e4SLinus Torvalds #ifdef DEBUG_MM_RB 3151da177e4SLinus Torvalds static int browse_rb(struct rb_root *root) 3161da177e4SLinus Torvalds { 3171da177e4SLinus Torvalds int i = 0, j; 3181da177e4SLinus Torvalds struct rb_node *nd, *pn = NULL; 3191da177e4SLinus Torvalds unsigned long prev = 0, pend = 0; 3201da177e4SLinus Torvalds 3211da177e4SLinus Torvalds for (nd = rb_first(root); nd; nd = rb_next(nd)) { 3221da177e4SLinus Torvalds struct vm_area_struct *vma; 3231da177e4SLinus Torvalds vma = rb_entry(nd, struct vm_area_struct, vm_rb); 3241da177e4SLinus Torvalds if (vma->vm_start < prev) 3251da177e4SLinus Torvalds printk("vm_start %lx prev %lx\n", vma->vm_start, prev), i = -1; 3261da177e4SLinus Torvalds if (vma->vm_start < pend) 3271da177e4SLinus Torvalds printk("vm_start %lx pend %lx\n", vma->vm_start, pend); 3281da177e4SLinus Torvalds if (vma->vm_start > vma->vm_end) 3291da177e4SLinus Torvalds printk("vm_end %lx < vm_start %lx\n", vma->vm_end, vma->vm_start); 3301da177e4SLinus Torvalds i++; 3311da177e4SLinus Torvalds pn = nd; 332d1af65d1SDavid Miller prev = vma->vm_start; 333d1af65d1SDavid Miller pend = vma->vm_end; 3341da177e4SLinus Torvalds } 3351da177e4SLinus Torvalds j = 0; 3361da177e4SLinus Torvalds for (nd = pn; nd; nd = rb_prev(nd)) { 3371da177e4SLinus Torvalds j++; 3381da177e4SLinus Torvalds } 3391da177e4SLinus Torvalds if (i != j) 3401da177e4SLinus Torvalds printk("backwards %d, forwards %d\n", j, i), i = 0; 3411da177e4SLinus Torvalds return i; 3421da177e4SLinus Torvalds } 3431da177e4SLinus Torvalds 3441da177e4SLinus Torvalds void validate_mm(struct mm_struct *mm) 3451da177e4SLinus Torvalds { 3461da177e4SLinus Torvalds int bug = 0; 3471da177e4SLinus Torvalds int i = 0; 3481da177e4SLinus Torvalds struct vm_area_struct *tmp = mm->mmap; 3491da177e4SLinus Torvalds while (tmp) { 3501da177e4SLinus Torvalds tmp = tmp->vm_next; 3511da177e4SLinus Torvalds i++; 3521da177e4SLinus Torvalds } 3531da177e4SLinus Torvalds if (i != mm->map_count) 3541da177e4SLinus Torvalds printk("map_count %d vm_next %d\n", mm->map_count, i), bug = 1; 3551da177e4SLinus Torvalds i = browse_rb(&mm->mm_rb); 3561da177e4SLinus Torvalds if (i != mm->map_count) 3571da177e4SLinus Torvalds printk("map_count %d rb %d\n", mm->map_count, i), bug = 1; 35846a350efSEric Sesterhenn BUG_ON(bug); 3591da177e4SLinus Torvalds } 3601da177e4SLinus Torvalds #else 3611da177e4SLinus Torvalds #define validate_mm(mm) do { } while (0) 3621da177e4SLinus Torvalds #endif 3631da177e4SLinus Torvalds 3641da177e4SLinus Torvalds static struct vm_area_struct * 3651da177e4SLinus Torvalds find_vma_prepare(struct mm_struct *mm, unsigned long addr, 3661da177e4SLinus Torvalds struct vm_area_struct **pprev, struct rb_node ***rb_link, 3671da177e4SLinus Torvalds struct rb_node ** rb_parent) 3681da177e4SLinus Torvalds { 3691da177e4SLinus Torvalds struct vm_area_struct * vma; 3701da177e4SLinus Torvalds struct rb_node ** __rb_link, * __rb_parent, * rb_prev; 3711da177e4SLinus Torvalds 3721da177e4SLinus Torvalds __rb_link = &mm->mm_rb.rb_node; 3731da177e4SLinus Torvalds rb_prev = __rb_parent = NULL; 3741da177e4SLinus Torvalds vma = NULL; 3751da177e4SLinus Torvalds 3761da177e4SLinus Torvalds while (*__rb_link) { 3771da177e4SLinus Torvalds struct vm_area_struct *vma_tmp; 3781da177e4SLinus Torvalds 3791da177e4SLinus Torvalds __rb_parent = *__rb_link; 3801da177e4SLinus Torvalds vma_tmp = rb_entry(__rb_parent, struct vm_area_struct, vm_rb); 3811da177e4SLinus Torvalds 3821da177e4SLinus Torvalds if (vma_tmp->vm_end > addr) { 3831da177e4SLinus Torvalds vma = vma_tmp; 3841da177e4SLinus Torvalds if (vma_tmp->vm_start <= addr) 385dfe195fbSBenny Halevy break; 3861da177e4SLinus Torvalds __rb_link = &__rb_parent->rb_left; 3871da177e4SLinus Torvalds } else { 3881da177e4SLinus Torvalds rb_prev = __rb_parent; 3891da177e4SLinus Torvalds __rb_link = &__rb_parent->rb_right; 3901da177e4SLinus Torvalds } 3911da177e4SLinus Torvalds } 3921da177e4SLinus Torvalds 3931da177e4SLinus Torvalds *pprev = NULL; 3941da177e4SLinus Torvalds if (rb_prev) 3951da177e4SLinus Torvalds *pprev = rb_entry(rb_prev, struct vm_area_struct, vm_rb); 3961da177e4SLinus Torvalds *rb_link = __rb_link; 3971da177e4SLinus Torvalds *rb_parent = __rb_parent; 3981da177e4SLinus Torvalds return vma; 3991da177e4SLinus Torvalds } 4001da177e4SLinus Torvalds 4011da177e4SLinus Torvalds static inline void 4021da177e4SLinus Torvalds __vma_link_list(struct mm_struct *mm, struct vm_area_struct *vma, 4031da177e4SLinus Torvalds struct vm_area_struct *prev, struct rb_node *rb_parent) 4041da177e4SLinus Torvalds { 405297c5eeeSLinus Torvalds struct vm_area_struct *next; 406297c5eeeSLinus Torvalds 407297c5eeeSLinus Torvalds vma->vm_prev = prev; 4081da177e4SLinus Torvalds if (prev) { 409297c5eeeSLinus Torvalds next = prev->vm_next; 4101da177e4SLinus Torvalds prev->vm_next = vma; 4111da177e4SLinus Torvalds } else { 4121da177e4SLinus Torvalds mm->mmap = vma; 4131da177e4SLinus Torvalds if (rb_parent) 414297c5eeeSLinus Torvalds next = rb_entry(rb_parent, 4151da177e4SLinus Torvalds struct vm_area_struct, vm_rb); 4161da177e4SLinus Torvalds else 417297c5eeeSLinus Torvalds next = NULL; 4181da177e4SLinus Torvalds } 419297c5eeeSLinus Torvalds vma->vm_next = next; 420297c5eeeSLinus Torvalds if (next) 421297c5eeeSLinus Torvalds next->vm_prev = vma; 4221da177e4SLinus Torvalds } 4231da177e4SLinus Torvalds 4241da177e4SLinus Torvalds void __vma_link_rb(struct mm_struct *mm, struct vm_area_struct *vma, 4251da177e4SLinus Torvalds struct rb_node **rb_link, struct rb_node *rb_parent) 4261da177e4SLinus Torvalds { 4271da177e4SLinus Torvalds rb_link_node(&vma->vm_rb, rb_parent, rb_link); 4281da177e4SLinus Torvalds rb_insert_color(&vma->vm_rb, &mm->mm_rb); 4291da177e4SLinus Torvalds } 4301da177e4SLinus Torvalds 431cb8f488cSDenys Vlasenko static void __vma_link_file(struct vm_area_struct *vma) 4321da177e4SLinus Torvalds { 4331da177e4SLinus Torvalds struct file *file; 4341da177e4SLinus Torvalds 4351da177e4SLinus Torvalds file = vma->vm_file; 4361da177e4SLinus Torvalds if (file) { 4371da177e4SLinus Torvalds struct address_space *mapping = file->f_mapping; 4381da177e4SLinus Torvalds 4391da177e4SLinus Torvalds if (vma->vm_flags & VM_DENYWRITE) 440d3ac7f89SJosef "Jeff" Sipek atomic_dec(&file->f_path.dentry->d_inode->i_writecount); 4411da177e4SLinus Torvalds if (vma->vm_flags & VM_SHARED) 4421da177e4SLinus Torvalds mapping->i_mmap_writable++; 4431da177e4SLinus Torvalds 4441da177e4SLinus Torvalds flush_dcache_mmap_lock(mapping); 4451da177e4SLinus Torvalds if (unlikely(vma->vm_flags & VM_NONLINEAR)) 4461da177e4SLinus Torvalds vma_nonlinear_insert(vma, &mapping->i_mmap_nonlinear); 4471da177e4SLinus Torvalds else 4481da177e4SLinus Torvalds vma_prio_tree_insert(vma, &mapping->i_mmap); 4491da177e4SLinus Torvalds flush_dcache_mmap_unlock(mapping); 4501da177e4SLinus Torvalds } 4511da177e4SLinus Torvalds } 4521da177e4SLinus Torvalds 4531da177e4SLinus Torvalds static void 4541da177e4SLinus Torvalds __vma_link(struct mm_struct *mm, struct vm_area_struct *vma, 4551da177e4SLinus Torvalds struct vm_area_struct *prev, struct rb_node **rb_link, 4561da177e4SLinus Torvalds struct rb_node *rb_parent) 4571da177e4SLinus Torvalds { 4581da177e4SLinus Torvalds __vma_link_list(mm, vma, prev, rb_parent); 4591da177e4SLinus Torvalds __vma_link_rb(mm, vma, rb_link, rb_parent); 4601da177e4SLinus Torvalds } 4611da177e4SLinus Torvalds 4621da177e4SLinus Torvalds static void vma_link(struct mm_struct *mm, struct vm_area_struct *vma, 4631da177e4SLinus Torvalds struct vm_area_struct *prev, struct rb_node **rb_link, 4641da177e4SLinus Torvalds struct rb_node *rb_parent) 4651da177e4SLinus Torvalds { 4661da177e4SLinus Torvalds struct address_space *mapping = NULL; 4671da177e4SLinus Torvalds 4681da177e4SLinus Torvalds if (vma->vm_file) 4691da177e4SLinus Torvalds mapping = vma->vm_file->f_mapping; 4701da177e4SLinus Torvalds 4711da177e4SLinus Torvalds if (mapping) { 4721da177e4SLinus Torvalds spin_lock(&mapping->i_mmap_lock); 4731da177e4SLinus Torvalds vma->vm_truncate_count = mapping->truncate_count; 4741da177e4SLinus Torvalds } 4751da177e4SLinus Torvalds 4761da177e4SLinus Torvalds __vma_link(mm, vma, prev, rb_link, rb_parent); 4771da177e4SLinus Torvalds __vma_link_file(vma); 4781da177e4SLinus Torvalds 4791da177e4SLinus Torvalds if (mapping) 4801da177e4SLinus Torvalds spin_unlock(&mapping->i_mmap_lock); 4811da177e4SLinus Torvalds 4821da177e4SLinus Torvalds mm->map_count++; 4831da177e4SLinus Torvalds validate_mm(mm); 4841da177e4SLinus Torvalds } 4851da177e4SLinus Torvalds 4861da177e4SLinus Torvalds /* 4871da177e4SLinus Torvalds * Helper for vma_adjust in the split_vma insert case: 4881da177e4SLinus Torvalds * insert vm structure into list and rbtree and anon_vma, 4891da177e4SLinus Torvalds * but it has already been inserted into prio_tree earlier. 4901da177e4SLinus Torvalds */ 49148aae425SZhenwenXu static void __insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma) 4921da177e4SLinus Torvalds { 4931da177e4SLinus Torvalds struct vm_area_struct *__vma, *prev; 4941da177e4SLinus Torvalds struct rb_node **rb_link, *rb_parent; 4951da177e4SLinus Torvalds 4961da177e4SLinus Torvalds __vma = find_vma_prepare(mm, vma->vm_start,&prev, &rb_link, &rb_parent); 49746a350efSEric Sesterhenn BUG_ON(__vma && __vma->vm_start < vma->vm_end); 4981da177e4SLinus Torvalds __vma_link(mm, vma, prev, rb_link, rb_parent); 4991da177e4SLinus Torvalds mm->map_count++; 5001da177e4SLinus Torvalds } 5011da177e4SLinus Torvalds 5021da177e4SLinus Torvalds static inline void 5031da177e4SLinus Torvalds __vma_unlink(struct mm_struct *mm, struct vm_area_struct *vma, 5041da177e4SLinus Torvalds struct vm_area_struct *prev) 5051da177e4SLinus Torvalds { 506297c5eeeSLinus Torvalds struct vm_area_struct *next = vma->vm_next; 507297c5eeeSLinus Torvalds 508297c5eeeSLinus Torvalds prev->vm_next = next; 509297c5eeeSLinus Torvalds if (next) 510297c5eeeSLinus Torvalds next->vm_prev = prev; 5111da177e4SLinus Torvalds rb_erase(&vma->vm_rb, &mm->mm_rb); 5121da177e4SLinus Torvalds if (mm->mmap_cache == vma) 5131da177e4SLinus Torvalds mm->mmap_cache = prev; 5141da177e4SLinus Torvalds } 5151da177e4SLinus Torvalds 5161da177e4SLinus Torvalds /* 5171da177e4SLinus Torvalds * We cannot adjust vm_start, vm_end, vm_pgoff fields of a vma that 5181da177e4SLinus Torvalds * is already present in an i_mmap tree without adjusting the tree. 5191da177e4SLinus Torvalds * The following helper function should be used when such adjustments 5201da177e4SLinus Torvalds * are necessary. The "insert" vma (if any) is to be inserted 5211da177e4SLinus Torvalds * before we drop the necessary locks. 5221da177e4SLinus Torvalds */ 5235beb4930SRik van Riel int vma_adjust(struct vm_area_struct *vma, unsigned long start, 5241da177e4SLinus Torvalds unsigned long end, pgoff_t pgoff, struct vm_area_struct *insert) 5251da177e4SLinus Torvalds { 5261da177e4SLinus Torvalds struct mm_struct *mm = vma->vm_mm; 5271da177e4SLinus Torvalds struct vm_area_struct *next = vma->vm_next; 5281da177e4SLinus Torvalds struct vm_area_struct *importer = NULL; 5291da177e4SLinus Torvalds struct address_space *mapping = NULL; 5301da177e4SLinus Torvalds struct prio_tree_root *root = NULL; 531012f1800SRik van Riel struct anon_vma *anon_vma = NULL; 5321da177e4SLinus Torvalds struct file *file = vma->vm_file; 5331da177e4SLinus Torvalds long adjust_next = 0; 5341da177e4SLinus Torvalds int remove_next = 0; 5351da177e4SLinus Torvalds 5361da177e4SLinus Torvalds if (next && !insert) { 537287d97acSLinus Torvalds struct vm_area_struct *exporter = NULL; 538287d97acSLinus Torvalds 5391da177e4SLinus Torvalds if (end >= next->vm_end) { 5401da177e4SLinus Torvalds /* 5411da177e4SLinus Torvalds * vma expands, overlapping all the next, and 5421da177e4SLinus Torvalds * perhaps the one after too (mprotect case 6). 5431da177e4SLinus Torvalds */ 5441da177e4SLinus Torvalds again: remove_next = 1 + (end > next->vm_end); 5451da177e4SLinus Torvalds end = next->vm_end; 546287d97acSLinus Torvalds exporter = next; 5471da177e4SLinus Torvalds importer = vma; 5481da177e4SLinus Torvalds } else if (end > next->vm_start) { 5491da177e4SLinus Torvalds /* 5501da177e4SLinus Torvalds * vma expands, overlapping part of the next: 5511da177e4SLinus Torvalds * mprotect case 5 shifting the boundary up. 5521da177e4SLinus Torvalds */ 5531da177e4SLinus Torvalds adjust_next = (end - next->vm_start) >> PAGE_SHIFT; 554287d97acSLinus Torvalds exporter = next; 5551da177e4SLinus Torvalds importer = vma; 5561da177e4SLinus Torvalds } else if (end < vma->vm_end) { 5571da177e4SLinus Torvalds /* 5581da177e4SLinus Torvalds * vma shrinks, and !insert tells it's not 5591da177e4SLinus Torvalds * split_vma inserting another: so it must be 5601da177e4SLinus Torvalds * mprotect case 4 shifting the boundary down. 5611da177e4SLinus Torvalds */ 5621da177e4SLinus Torvalds adjust_next = - ((vma->vm_end - end) >> PAGE_SHIFT); 563287d97acSLinus Torvalds exporter = vma; 5641da177e4SLinus Torvalds importer = next; 5651da177e4SLinus Torvalds } 5661da177e4SLinus Torvalds 5675beb4930SRik van Riel /* 5685beb4930SRik van Riel * Easily overlooked: when mprotect shifts the boundary, 5695beb4930SRik van Riel * make sure the expanding vma has anon_vma set if the 5705beb4930SRik van Riel * shrinking vma had, to cover any anon pages imported. 5715beb4930SRik van Riel */ 572287d97acSLinus Torvalds if (exporter && exporter->anon_vma && !importer->anon_vma) { 573287d97acSLinus Torvalds if (anon_vma_clone(importer, exporter)) 5745beb4930SRik van Riel return -ENOMEM; 575287d97acSLinus Torvalds importer->anon_vma = exporter->anon_vma; 5765beb4930SRik van Riel } 5775beb4930SRik van Riel } 5785beb4930SRik van Riel 5791da177e4SLinus Torvalds if (file) { 5801da177e4SLinus Torvalds mapping = file->f_mapping; 5811da177e4SLinus Torvalds if (!(vma->vm_flags & VM_NONLINEAR)) 5821da177e4SLinus Torvalds root = &mapping->i_mmap; 5831da177e4SLinus Torvalds spin_lock(&mapping->i_mmap_lock); 5841da177e4SLinus Torvalds if (importer && 5851da177e4SLinus Torvalds vma->vm_truncate_count != next->vm_truncate_count) { 5861da177e4SLinus Torvalds /* 5871da177e4SLinus Torvalds * unmap_mapping_range might be in progress: 5881da177e4SLinus Torvalds * ensure that the expanding vma is rescanned. 5891da177e4SLinus Torvalds */ 5901da177e4SLinus Torvalds importer->vm_truncate_count = 0; 5911da177e4SLinus Torvalds } 5921da177e4SLinus Torvalds if (insert) { 5931da177e4SLinus Torvalds insert->vm_truncate_count = vma->vm_truncate_count; 5941da177e4SLinus Torvalds /* 5951da177e4SLinus Torvalds * Put into prio_tree now, so instantiated pages 5961da177e4SLinus Torvalds * are visible to arm/parisc __flush_dcache_page 5971da177e4SLinus Torvalds * throughout; but we cannot insert into address 5981da177e4SLinus Torvalds * space until vma start or end is updated. 5991da177e4SLinus Torvalds */ 6001da177e4SLinus Torvalds __vma_link_file(insert); 6011da177e4SLinus Torvalds } 6021da177e4SLinus Torvalds } 6031da177e4SLinus Torvalds 60494fcc585SAndrea Arcangeli vma_adjust_trans_huge(vma, start, end, adjust_next); 60594fcc585SAndrea Arcangeli 606012f1800SRik van Riel /* 607012f1800SRik van Riel * When changing only vma->vm_end, we don't really need anon_vma 608012f1800SRik van Riel * lock. This is a fairly rare case by itself, but the anon_vma 609012f1800SRik van Riel * lock may be shared between many sibling processes. Skipping 610012f1800SRik van Riel * the lock for brk adjustments makes a difference sometimes. 611012f1800SRik van Riel */ 612*5f70b962SShaohua Li if (vma->anon_vma && (importer || start != vma->vm_start)) { 613012f1800SRik van Riel anon_vma = vma->anon_vma; 614012f1800SRik van Riel anon_vma_lock(anon_vma); 615012f1800SRik van Riel } 616012f1800SRik van Riel 6171da177e4SLinus Torvalds if (root) { 6181da177e4SLinus Torvalds flush_dcache_mmap_lock(mapping); 6191da177e4SLinus Torvalds vma_prio_tree_remove(vma, root); 6201da177e4SLinus Torvalds if (adjust_next) 6211da177e4SLinus Torvalds vma_prio_tree_remove(next, root); 6221da177e4SLinus Torvalds } 6231da177e4SLinus Torvalds 6241da177e4SLinus Torvalds vma->vm_start = start; 6251da177e4SLinus Torvalds vma->vm_end = end; 6261da177e4SLinus Torvalds vma->vm_pgoff = pgoff; 6271da177e4SLinus Torvalds if (adjust_next) { 6281da177e4SLinus Torvalds next->vm_start += adjust_next << PAGE_SHIFT; 6291da177e4SLinus Torvalds next->vm_pgoff += adjust_next; 6301da177e4SLinus Torvalds } 6311da177e4SLinus Torvalds 6321da177e4SLinus Torvalds if (root) { 6331da177e4SLinus Torvalds if (adjust_next) 6341da177e4SLinus Torvalds vma_prio_tree_insert(next, root); 6351da177e4SLinus Torvalds vma_prio_tree_insert(vma, root); 6361da177e4SLinus Torvalds flush_dcache_mmap_unlock(mapping); 6371da177e4SLinus Torvalds } 6381da177e4SLinus Torvalds 6391da177e4SLinus Torvalds if (remove_next) { 6401da177e4SLinus Torvalds /* 6411da177e4SLinus Torvalds * vma_merge has merged next into vma, and needs 6421da177e4SLinus Torvalds * us to remove next before dropping the locks. 6431da177e4SLinus Torvalds */ 6441da177e4SLinus Torvalds __vma_unlink(mm, next, vma); 6451da177e4SLinus Torvalds if (file) 6461da177e4SLinus Torvalds __remove_shared_vm_struct(next, file, mapping); 6471da177e4SLinus Torvalds } else if (insert) { 6481da177e4SLinus Torvalds /* 6491da177e4SLinus Torvalds * split_vma has split insert from vma, and needs 6501da177e4SLinus Torvalds * us to insert it before dropping the locks 6511da177e4SLinus Torvalds * (it may either follow vma or precede it). 6521da177e4SLinus Torvalds */ 6531da177e4SLinus Torvalds __insert_vm_struct(mm, insert); 6541da177e4SLinus Torvalds } 6551da177e4SLinus Torvalds 656012f1800SRik van Riel if (anon_vma) 657012f1800SRik van Riel anon_vma_unlock(anon_vma); 6581da177e4SLinus Torvalds if (mapping) 6591da177e4SLinus Torvalds spin_unlock(&mapping->i_mmap_lock); 6601da177e4SLinus Torvalds 6611da177e4SLinus Torvalds if (remove_next) { 662925d1c40SMatt Helsley if (file) { 6631da177e4SLinus Torvalds fput(file); 664925d1c40SMatt Helsley if (next->vm_flags & VM_EXECUTABLE) 665925d1c40SMatt Helsley removed_exe_file_vma(mm); 666925d1c40SMatt Helsley } 6675beb4930SRik van Riel if (next->anon_vma) 6685beb4930SRik van Riel anon_vma_merge(vma, next); 6691da177e4SLinus Torvalds mm->map_count--; 670f0be3d32SLee Schermerhorn mpol_put(vma_policy(next)); 6711da177e4SLinus Torvalds kmem_cache_free(vm_area_cachep, next); 6721da177e4SLinus Torvalds /* 6731da177e4SLinus Torvalds * In mprotect's case 6 (see comments on vma_merge), 6741da177e4SLinus Torvalds * we must remove another next too. It would clutter 6751da177e4SLinus Torvalds * up the code too much to do both in one go. 6761da177e4SLinus Torvalds */ 6771da177e4SLinus Torvalds if (remove_next == 2) { 6781da177e4SLinus Torvalds next = vma->vm_next; 6791da177e4SLinus Torvalds goto again; 6801da177e4SLinus Torvalds } 6811da177e4SLinus Torvalds } 6821da177e4SLinus Torvalds 6831da177e4SLinus Torvalds validate_mm(mm); 6845beb4930SRik van Riel 6855beb4930SRik van Riel return 0; 6861da177e4SLinus Torvalds } 6871da177e4SLinus Torvalds 6881da177e4SLinus Torvalds /* 6891da177e4SLinus Torvalds * If the vma has a ->close operation then the driver probably needs to release 6901da177e4SLinus Torvalds * per-vma resources, so we don't attempt to merge those. 6911da177e4SLinus Torvalds */ 6921da177e4SLinus Torvalds static inline int is_mergeable_vma(struct vm_area_struct *vma, 6931da177e4SLinus Torvalds struct file *file, unsigned long vm_flags) 6941da177e4SLinus Torvalds { 6958314c4f2SHugh Dickins /* VM_CAN_NONLINEAR may get set later by f_op->mmap() */ 6968314c4f2SHugh Dickins if ((vma->vm_flags ^ vm_flags) & ~VM_CAN_NONLINEAR) 6971da177e4SLinus Torvalds return 0; 6981da177e4SLinus Torvalds if (vma->vm_file != file) 6991da177e4SLinus Torvalds return 0; 7001da177e4SLinus Torvalds if (vma->vm_ops && vma->vm_ops->close) 7011da177e4SLinus Torvalds return 0; 7021da177e4SLinus Torvalds return 1; 7031da177e4SLinus Torvalds } 7041da177e4SLinus Torvalds 7051da177e4SLinus Torvalds static inline int is_mergeable_anon_vma(struct anon_vma *anon_vma1, 7061da177e4SLinus Torvalds struct anon_vma *anon_vma2) 7071da177e4SLinus Torvalds { 7081da177e4SLinus Torvalds return !anon_vma1 || !anon_vma2 || (anon_vma1 == anon_vma2); 7091da177e4SLinus Torvalds } 7101da177e4SLinus Torvalds 7111da177e4SLinus Torvalds /* 7121da177e4SLinus Torvalds * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff) 7131da177e4SLinus Torvalds * in front of (at a lower virtual address and file offset than) the vma. 7141da177e4SLinus Torvalds * 7151da177e4SLinus Torvalds * We cannot merge two vmas if they have differently assigned (non-NULL) 7161da177e4SLinus Torvalds * anon_vmas, nor if same anon_vma is assigned but offsets incompatible. 7171da177e4SLinus Torvalds * 7181da177e4SLinus Torvalds * We don't check here for the merged mmap wrapping around the end of pagecache 7191da177e4SLinus Torvalds * indices (16TB on ia32) because do_mmap_pgoff() does not permit mmap's which 7201da177e4SLinus Torvalds * wrap, nor mmaps which cover the final page at index -1UL. 7211da177e4SLinus Torvalds */ 7221da177e4SLinus Torvalds static int 7231da177e4SLinus Torvalds can_vma_merge_before(struct vm_area_struct *vma, unsigned long vm_flags, 7241da177e4SLinus Torvalds struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff) 7251da177e4SLinus Torvalds { 7261da177e4SLinus Torvalds if (is_mergeable_vma(vma, file, vm_flags) && 7271da177e4SLinus Torvalds is_mergeable_anon_vma(anon_vma, vma->anon_vma)) { 7281da177e4SLinus Torvalds if (vma->vm_pgoff == vm_pgoff) 7291da177e4SLinus Torvalds return 1; 7301da177e4SLinus Torvalds } 7311da177e4SLinus Torvalds return 0; 7321da177e4SLinus Torvalds } 7331da177e4SLinus Torvalds 7341da177e4SLinus Torvalds /* 7351da177e4SLinus Torvalds * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff) 7361da177e4SLinus Torvalds * beyond (at a higher virtual address and file offset than) the vma. 7371da177e4SLinus Torvalds * 7381da177e4SLinus Torvalds * We cannot merge two vmas if they have differently assigned (non-NULL) 7391da177e4SLinus Torvalds * anon_vmas, nor if same anon_vma is assigned but offsets incompatible. 7401da177e4SLinus Torvalds */ 7411da177e4SLinus Torvalds static int 7421da177e4SLinus Torvalds can_vma_merge_after(struct vm_area_struct *vma, unsigned long vm_flags, 7431da177e4SLinus Torvalds struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff) 7441da177e4SLinus Torvalds { 7451da177e4SLinus Torvalds if (is_mergeable_vma(vma, file, vm_flags) && 7461da177e4SLinus Torvalds is_mergeable_anon_vma(anon_vma, vma->anon_vma)) { 7471da177e4SLinus Torvalds pgoff_t vm_pglen; 7481da177e4SLinus Torvalds vm_pglen = (vma->vm_end - vma->vm_start) >> PAGE_SHIFT; 7491da177e4SLinus Torvalds if (vma->vm_pgoff + vm_pglen == vm_pgoff) 7501da177e4SLinus Torvalds return 1; 7511da177e4SLinus Torvalds } 7521da177e4SLinus Torvalds return 0; 7531da177e4SLinus Torvalds } 7541da177e4SLinus Torvalds 7551da177e4SLinus Torvalds /* 7561da177e4SLinus Torvalds * Given a mapping request (addr,end,vm_flags,file,pgoff), figure out 7571da177e4SLinus Torvalds * whether that can be merged with its predecessor or its successor. 7581da177e4SLinus Torvalds * Or both (it neatly fills a hole). 7591da177e4SLinus Torvalds * 7601da177e4SLinus Torvalds * In most cases - when called for mmap, brk or mremap - [addr,end) is 7611da177e4SLinus Torvalds * certain not to be mapped by the time vma_merge is called; but when 7621da177e4SLinus Torvalds * called for mprotect, it is certain to be already mapped (either at 7631da177e4SLinus Torvalds * an offset within prev, or at the start of next), and the flags of 7641da177e4SLinus Torvalds * this area are about to be changed to vm_flags - and the no-change 7651da177e4SLinus Torvalds * case has already been eliminated. 7661da177e4SLinus Torvalds * 7671da177e4SLinus Torvalds * The following mprotect cases have to be considered, where AAAA is 7681da177e4SLinus Torvalds * the area passed down from mprotect_fixup, never extending beyond one 7691da177e4SLinus Torvalds * vma, PPPPPP is the prev vma specified, and NNNNNN the next vma after: 7701da177e4SLinus Torvalds * 7711da177e4SLinus Torvalds * AAAA AAAA AAAA AAAA 7721da177e4SLinus Torvalds * PPPPPPNNNNNN PPPPPPNNNNNN PPPPPPNNNNNN PPPPNNNNXXXX 7731da177e4SLinus Torvalds * cannot merge might become might become might become 7741da177e4SLinus Torvalds * PPNNNNNNNNNN PPPPPPPPPPNN PPPPPPPPPPPP 6 or 7751da177e4SLinus Torvalds * mmap, brk or case 4 below case 5 below PPPPPPPPXXXX 7 or 7761da177e4SLinus Torvalds * mremap move: PPPPNNNNNNNN 8 7771da177e4SLinus Torvalds * AAAA 7781da177e4SLinus Torvalds * PPPP NNNN PPPPPPPPPPPP PPPPPPPPNNNN PPPPNNNNNNNN 7791da177e4SLinus Torvalds * might become case 1 below case 2 below case 3 below 7801da177e4SLinus Torvalds * 7811da177e4SLinus Torvalds * Odd one out? Case 8, because it extends NNNN but needs flags of XXXX: 7821da177e4SLinus Torvalds * mprotect_fixup updates vm_flags & vm_page_prot on successful return. 7831da177e4SLinus Torvalds */ 7841da177e4SLinus Torvalds struct vm_area_struct *vma_merge(struct mm_struct *mm, 7851da177e4SLinus Torvalds struct vm_area_struct *prev, unsigned long addr, 7861da177e4SLinus Torvalds unsigned long end, unsigned long vm_flags, 7871da177e4SLinus Torvalds struct anon_vma *anon_vma, struct file *file, 7881da177e4SLinus Torvalds pgoff_t pgoff, struct mempolicy *policy) 7891da177e4SLinus Torvalds { 7901da177e4SLinus Torvalds pgoff_t pglen = (end - addr) >> PAGE_SHIFT; 7911da177e4SLinus Torvalds struct vm_area_struct *area, *next; 7925beb4930SRik van Riel int err; 7931da177e4SLinus Torvalds 7941da177e4SLinus Torvalds /* 7951da177e4SLinus Torvalds * We later require that vma->vm_flags == vm_flags, 7961da177e4SLinus Torvalds * so this tests vma->vm_flags & VM_SPECIAL, too. 7971da177e4SLinus Torvalds */ 7981da177e4SLinus Torvalds if (vm_flags & VM_SPECIAL) 7991da177e4SLinus Torvalds return NULL; 8001da177e4SLinus Torvalds 8011da177e4SLinus Torvalds if (prev) 8021da177e4SLinus Torvalds next = prev->vm_next; 8031da177e4SLinus Torvalds else 8041da177e4SLinus Torvalds next = mm->mmap; 8051da177e4SLinus Torvalds area = next; 8061da177e4SLinus Torvalds if (next && next->vm_end == end) /* cases 6, 7, 8 */ 8071da177e4SLinus Torvalds next = next->vm_next; 8081da177e4SLinus Torvalds 8091da177e4SLinus Torvalds /* 8101da177e4SLinus Torvalds * Can it merge with the predecessor? 8111da177e4SLinus Torvalds */ 8121da177e4SLinus Torvalds if (prev && prev->vm_end == addr && 8131da177e4SLinus Torvalds mpol_equal(vma_policy(prev), policy) && 8141da177e4SLinus Torvalds can_vma_merge_after(prev, vm_flags, 8151da177e4SLinus Torvalds anon_vma, file, pgoff)) { 8161da177e4SLinus Torvalds /* 8171da177e4SLinus Torvalds * OK, it can. Can we now merge in the successor as well? 8181da177e4SLinus Torvalds */ 8191da177e4SLinus Torvalds if (next && end == next->vm_start && 8201da177e4SLinus Torvalds mpol_equal(policy, vma_policy(next)) && 8211da177e4SLinus Torvalds can_vma_merge_before(next, vm_flags, 8221da177e4SLinus Torvalds anon_vma, file, pgoff+pglen) && 8231da177e4SLinus Torvalds is_mergeable_anon_vma(prev->anon_vma, 8241da177e4SLinus Torvalds next->anon_vma)) { 8251da177e4SLinus Torvalds /* cases 1, 6 */ 8265beb4930SRik van Riel err = vma_adjust(prev, prev->vm_start, 8271da177e4SLinus Torvalds next->vm_end, prev->vm_pgoff, NULL); 8281da177e4SLinus Torvalds } else /* cases 2, 5, 7 */ 8295beb4930SRik van Riel err = vma_adjust(prev, prev->vm_start, 8301da177e4SLinus Torvalds end, prev->vm_pgoff, NULL); 8315beb4930SRik van Riel if (err) 8325beb4930SRik van Riel return NULL; 833b15d00b6SAndrea Arcangeli khugepaged_enter_vma_merge(prev); 8341da177e4SLinus Torvalds return prev; 8351da177e4SLinus Torvalds } 8361da177e4SLinus Torvalds 8371da177e4SLinus Torvalds /* 8381da177e4SLinus Torvalds * Can this new request be merged in front of next? 8391da177e4SLinus Torvalds */ 8401da177e4SLinus Torvalds if (next && end == next->vm_start && 8411da177e4SLinus Torvalds mpol_equal(policy, vma_policy(next)) && 8421da177e4SLinus Torvalds can_vma_merge_before(next, vm_flags, 8431da177e4SLinus Torvalds anon_vma, file, pgoff+pglen)) { 8441da177e4SLinus Torvalds if (prev && addr < prev->vm_end) /* case 4 */ 8455beb4930SRik van Riel err = vma_adjust(prev, prev->vm_start, 8461da177e4SLinus Torvalds addr, prev->vm_pgoff, NULL); 8471da177e4SLinus Torvalds else /* cases 3, 8 */ 8485beb4930SRik van Riel err = vma_adjust(area, addr, next->vm_end, 8491da177e4SLinus Torvalds next->vm_pgoff - pglen, NULL); 8505beb4930SRik van Riel if (err) 8515beb4930SRik van Riel return NULL; 852b15d00b6SAndrea Arcangeli khugepaged_enter_vma_merge(area); 8531da177e4SLinus Torvalds return area; 8541da177e4SLinus Torvalds } 8551da177e4SLinus Torvalds 8561da177e4SLinus Torvalds return NULL; 8571da177e4SLinus Torvalds } 8581da177e4SLinus Torvalds 8591da177e4SLinus Torvalds /* 860d0e9fe17SLinus Torvalds * Rough compatbility check to quickly see if it's even worth looking 861d0e9fe17SLinus Torvalds * at sharing an anon_vma. 862d0e9fe17SLinus Torvalds * 863d0e9fe17SLinus Torvalds * They need to have the same vm_file, and the flags can only differ 864d0e9fe17SLinus Torvalds * in things that mprotect may change. 865d0e9fe17SLinus Torvalds * 866d0e9fe17SLinus Torvalds * NOTE! The fact that we share an anon_vma doesn't _have_ to mean that 867d0e9fe17SLinus Torvalds * we can merge the two vma's. For example, we refuse to merge a vma if 868d0e9fe17SLinus Torvalds * there is a vm_ops->close() function, because that indicates that the 869d0e9fe17SLinus Torvalds * driver is doing some kind of reference counting. But that doesn't 870d0e9fe17SLinus Torvalds * really matter for the anon_vma sharing case. 871d0e9fe17SLinus Torvalds */ 872d0e9fe17SLinus Torvalds static int anon_vma_compatible(struct vm_area_struct *a, struct vm_area_struct *b) 873d0e9fe17SLinus Torvalds { 874d0e9fe17SLinus Torvalds return a->vm_end == b->vm_start && 875d0e9fe17SLinus Torvalds mpol_equal(vma_policy(a), vma_policy(b)) && 876d0e9fe17SLinus Torvalds a->vm_file == b->vm_file && 877d0e9fe17SLinus Torvalds !((a->vm_flags ^ b->vm_flags) & ~(VM_READ|VM_WRITE|VM_EXEC)) && 878d0e9fe17SLinus Torvalds b->vm_pgoff == a->vm_pgoff + ((b->vm_start - a->vm_start) >> PAGE_SHIFT); 879d0e9fe17SLinus Torvalds } 880d0e9fe17SLinus Torvalds 881d0e9fe17SLinus Torvalds /* 882d0e9fe17SLinus Torvalds * Do some basic sanity checking to see if we can re-use the anon_vma 883d0e9fe17SLinus Torvalds * from 'old'. The 'a'/'b' vma's are in VM order - one of them will be 884d0e9fe17SLinus Torvalds * the same as 'old', the other will be the new one that is trying 885d0e9fe17SLinus Torvalds * to share the anon_vma. 886d0e9fe17SLinus Torvalds * 887d0e9fe17SLinus Torvalds * NOTE! This runs with mm_sem held for reading, so it is possible that 888d0e9fe17SLinus Torvalds * the anon_vma of 'old' is concurrently in the process of being set up 889d0e9fe17SLinus Torvalds * by another page fault trying to merge _that_. But that's ok: if it 890d0e9fe17SLinus Torvalds * is being set up, that automatically means that it will be a singleton 891d0e9fe17SLinus Torvalds * acceptable for merging, so we can do all of this optimistically. But 892d0e9fe17SLinus Torvalds * we do that ACCESS_ONCE() to make sure that we never re-load the pointer. 893d0e9fe17SLinus Torvalds * 894d0e9fe17SLinus Torvalds * IOW: that the "list_is_singular()" test on the anon_vma_chain only 895d0e9fe17SLinus Torvalds * matters for the 'stable anon_vma' case (ie the thing we want to avoid 896d0e9fe17SLinus Torvalds * is to return an anon_vma that is "complex" due to having gone through 897d0e9fe17SLinus Torvalds * a fork). 898d0e9fe17SLinus Torvalds * 899d0e9fe17SLinus Torvalds * We also make sure that the two vma's are compatible (adjacent, 900d0e9fe17SLinus Torvalds * and with the same memory policies). That's all stable, even with just 901d0e9fe17SLinus Torvalds * a read lock on the mm_sem. 902d0e9fe17SLinus Torvalds */ 903d0e9fe17SLinus Torvalds static struct anon_vma *reusable_anon_vma(struct vm_area_struct *old, struct vm_area_struct *a, struct vm_area_struct *b) 904d0e9fe17SLinus Torvalds { 905d0e9fe17SLinus Torvalds if (anon_vma_compatible(a, b)) { 906d0e9fe17SLinus Torvalds struct anon_vma *anon_vma = ACCESS_ONCE(old->anon_vma); 907d0e9fe17SLinus Torvalds 908d0e9fe17SLinus Torvalds if (anon_vma && list_is_singular(&old->anon_vma_chain)) 909d0e9fe17SLinus Torvalds return anon_vma; 910d0e9fe17SLinus Torvalds } 911d0e9fe17SLinus Torvalds return NULL; 912d0e9fe17SLinus Torvalds } 913d0e9fe17SLinus Torvalds 914d0e9fe17SLinus Torvalds /* 9151da177e4SLinus Torvalds * find_mergeable_anon_vma is used by anon_vma_prepare, to check 9161da177e4SLinus Torvalds * neighbouring vmas for a suitable anon_vma, before it goes off 9171da177e4SLinus Torvalds * to allocate a new anon_vma. It checks because a repetitive 9181da177e4SLinus Torvalds * sequence of mprotects and faults may otherwise lead to distinct 9191da177e4SLinus Torvalds * anon_vmas being allocated, preventing vma merge in subsequent 9201da177e4SLinus Torvalds * mprotect. 9211da177e4SLinus Torvalds */ 9221da177e4SLinus Torvalds struct anon_vma *find_mergeable_anon_vma(struct vm_area_struct *vma) 9231da177e4SLinus Torvalds { 924d0e9fe17SLinus Torvalds struct anon_vma *anon_vma; 9251da177e4SLinus Torvalds struct vm_area_struct *near; 9261da177e4SLinus Torvalds 9271da177e4SLinus Torvalds near = vma->vm_next; 9281da177e4SLinus Torvalds if (!near) 9291da177e4SLinus Torvalds goto try_prev; 9301da177e4SLinus Torvalds 931d0e9fe17SLinus Torvalds anon_vma = reusable_anon_vma(near, vma, near); 932d0e9fe17SLinus Torvalds if (anon_vma) 933d0e9fe17SLinus Torvalds return anon_vma; 9341da177e4SLinus Torvalds try_prev: 9351da177e4SLinus Torvalds /* 9361da177e4SLinus Torvalds * It is potentially slow to have to call find_vma_prev here. 9371da177e4SLinus Torvalds * But it's only on the first write fault on the vma, not 9381da177e4SLinus Torvalds * every time, and we could devise a way to avoid it later 9391da177e4SLinus Torvalds * (e.g. stash info in next's anon_vma_node when assigning 9401da177e4SLinus Torvalds * an anon_vma, or when trying vma_merge). Another time. 9411da177e4SLinus Torvalds */ 94246a350efSEric Sesterhenn BUG_ON(find_vma_prev(vma->vm_mm, vma->vm_start, &near) != vma); 9431da177e4SLinus Torvalds if (!near) 9441da177e4SLinus Torvalds goto none; 9451da177e4SLinus Torvalds 946d0e9fe17SLinus Torvalds anon_vma = reusable_anon_vma(near, near, vma); 947d0e9fe17SLinus Torvalds if (anon_vma) 948d0e9fe17SLinus Torvalds return anon_vma; 9491da177e4SLinus Torvalds none: 9501da177e4SLinus Torvalds /* 9511da177e4SLinus Torvalds * There's no absolute need to look only at touching neighbours: 9521da177e4SLinus Torvalds * we could search further afield for "compatible" anon_vmas. 9531da177e4SLinus Torvalds * But it would probably just be a waste of time searching, 9541da177e4SLinus Torvalds * or lead to too many vmas hanging off the same anon_vma. 9551da177e4SLinus Torvalds * We're trying to allow mprotect remerging later on, 9561da177e4SLinus Torvalds * not trying to minimize memory used for anon_vmas. 9571da177e4SLinus Torvalds */ 9581da177e4SLinus Torvalds return NULL; 9591da177e4SLinus Torvalds } 9601da177e4SLinus Torvalds 9611da177e4SLinus Torvalds #ifdef CONFIG_PROC_FS 962ab50b8edSHugh Dickins void vm_stat_account(struct mm_struct *mm, unsigned long flags, 9631da177e4SLinus Torvalds struct file *file, long pages) 9641da177e4SLinus Torvalds { 9651da177e4SLinus Torvalds const unsigned long stack_flags 9661da177e4SLinus Torvalds = VM_STACK_FLAGS & (VM_GROWSUP|VM_GROWSDOWN); 9671da177e4SLinus Torvalds 9681da177e4SLinus Torvalds if (file) { 9691da177e4SLinus Torvalds mm->shared_vm += pages; 9701da177e4SLinus Torvalds if ((flags & (VM_EXEC|VM_WRITE)) == VM_EXEC) 9711da177e4SLinus Torvalds mm->exec_vm += pages; 9721da177e4SLinus Torvalds } else if (flags & stack_flags) 9731da177e4SLinus Torvalds mm->stack_vm += pages; 9741da177e4SLinus Torvalds if (flags & (VM_RESERVED|VM_IO)) 9751da177e4SLinus Torvalds mm->reserved_vm += pages; 9761da177e4SLinus Torvalds } 9771da177e4SLinus Torvalds #endif /* CONFIG_PROC_FS */ 9781da177e4SLinus Torvalds 9791da177e4SLinus Torvalds /* 98027f5de79SJianjun Kong * The caller must hold down_write(¤t->mm->mmap_sem). 9811da177e4SLinus Torvalds */ 9821da177e4SLinus Torvalds 9831da177e4SLinus Torvalds unsigned long do_mmap_pgoff(struct file *file, unsigned long addr, 9841da177e4SLinus Torvalds unsigned long len, unsigned long prot, 9851da177e4SLinus Torvalds unsigned long flags, unsigned long pgoff) 9861da177e4SLinus Torvalds { 9871da177e4SLinus Torvalds struct mm_struct * mm = current->mm; 9881da177e4SLinus Torvalds struct inode *inode; 9891da177e4SLinus Torvalds unsigned int vm_flags; 9901da177e4SLinus Torvalds int error; 9910165ab44SMiklos Szeredi unsigned long reqprot = prot; 9921da177e4SLinus Torvalds 9931da177e4SLinus Torvalds /* 9941da177e4SLinus Torvalds * Does the application expect PROT_READ to imply PROT_EXEC? 9951da177e4SLinus Torvalds * 9961da177e4SLinus Torvalds * (the exception is when the underlying filesystem is noexec 9971da177e4SLinus Torvalds * mounted, in which case we dont add PROT_EXEC.) 9981da177e4SLinus Torvalds */ 9991da177e4SLinus Torvalds if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC)) 1000d3ac7f89SJosef "Jeff" Sipek if (!(file && (file->f_path.mnt->mnt_flags & MNT_NOEXEC))) 10011da177e4SLinus Torvalds prot |= PROT_EXEC; 10021da177e4SLinus Torvalds 10031da177e4SLinus Torvalds if (!len) 10041da177e4SLinus Torvalds return -EINVAL; 10051da177e4SLinus Torvalds 10067cd94146SEric Paris if (!(flags & MAP_FIXED)) 10077cd94146SEric Paris addr = round_hint_to_min(addr); 10087cd94146SEric Paris 10091da177e4SLinus Torvalds /* Careful about overflows.. */ 10101da177e4SLinus Torvalds len = PAGE_ALIGN(len); 10119206de95SAl Viro if (!len) 10121da177e4SLinus Torvalds return -ENOMEM; 10131da177e4SLinus Torvalds 10141da177e4SLinus Torvalds /* offset overflow? */ 10151da177e4SLinus Torvalds if ((pgoff + (len >> PAGE_SHIFT)) < pgoff) 10161da177e4SLinus Torvalds return -EOVERFLOW; 10171da177e4SLinus Torvalds 10181da177e4SLinus Torvalds /* Too many mappings? */ 10191da177e4SLinus Torvalds if (mm->map_count > sysctl_max_map_count) 10201da177e4SLinus Torvalds return -ENOMEM; 10211da177e4SLinus Torvalds 10221da177e4SLinus Torvalds /* Obtain the address to map to. we verify (or select) it and ensure 10231da177e4SLinus Torvalds * that it represents a valid section of the address space. 10241da177e4SLinus Torvalds */ 10251da177e4SLinus Torvalds addr = get_unmapped_area(file, addr, len, pgoff, flags); 10261da177e4SLinus Torvalds if (addr & ~PAGE_MASK) 10271da177e4SLinus Torvalds return addr; 10281da177e4SLinus Torvalds 10291da177e4SLinus Torvalds /* Do simple checking here so the lower-level routines won't have 10301da177e4SLinus Torvalds * to. we assume access permissions have been handled by the open 10311da177e4SLinus Torvalds * of the memory object, so we don't do any here. 10321da177e4SLinus Torvalds */ 10331da177e4SLinus Torvalds vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags) | 10341da177e4SLinus Torvalds mm->def_flags | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC; 10351da177e4SLinus Torvalds 1036cdf7b341SHuang Shijie if (flags & MAP_LOCKED) 10371da177e4SLinus Torvalds if (!can_do_mlock()) 10381da177e4SLinus Torvalds return -EPERM; 1039ba470de4SRik van Riel 10401da177e4SLinus Torvalds /* mlock MCL_FUTURE? */ 10411da177e4SLinus Torvalds if (vm_flags & VM_LOCKED) { 10421da177e4SLinus Torvalds unsigned long locked, lock_limit; 104393ea1d0aSChris Wright locked = len >> PAGE_SHIFT; 104493ea1d0aSChris Wright locked += mm->locked_vm; 104559e99e5bSJiri Slaby lock_limit = rlimit(RLIMIT_MEMLOCK); 104693ea1d0aSChris Wright lock_limit >>= PAGE_SHIFT; 10471da177e4SLinus Torvalds if (locked > lock_limit && !capable(CAP_IPC_LOCK)) 10481da177e4SLinus Torvalds return -EAGAIN; 10491da177e4SLinus Torvalds } 10501da177e4SLinus Torvalds 1051d3ac7f89SJosef "Jeff" Sipek inode = file ? file->f_path.dentry->d_inode : NULL; 10521da177e4SLinus Torvalds 10531da177e4SLinus Torvalds if (file) { 10541da177e4SLinus Torvalds switch (flags & MAP_TYPE) { 10551da177e4SLinus Torvalds case MAP_SHARED: 10561da177e4SLinus Torvalds if ((prot&PROT_WRITE) && !(file->f_mode&FMODE_WRITE)) 10571da177e4SLinus Torvalds return -EACCES; 10581da177e4SLinus Torvalds 10591da177e4SLinus Torvalds /* 10601da177e4SLinus Torvalds * Make sure we don't allow writing to an append-only 10611da177e4SLinus Torvalds * file.. 10621da177e4SLinus Torvalds */ 10631da177e4SLinus Torvalds if (IS_APPEND(inode) && (file->f_mode & FMODE_WRITE)) 10641da177e4SLinus Torvalds return -EACCES; 10651da177e4SLinus Torvalds 10661da177e4SLinus Torvalds /* 10671da177e4SLinus Torvalds * Make sure there are no mandatory locks on the file. 10681da177e4SLinus Torvalds */ 10691da177e4SLinus Torvalds if (locks_verify_locked(inode)) 10701da177e4SLinus Torvalds return -EAGAIN; 10711da177e4SLinus Torvalds 10721da177e4SLinus Torvalds vm_flags |= VM_SHARED | VM_MAYSHARE; 10731da177e4SLinus Torvalds if (!(file->f_mode & FMODE_WRITE)) 10741da177e4SLinus Torvalds vm_flags &= ~(VM_MAYWRITE | VM_SHARED); 10751da177e4SLinus Torvalds 10761da177e4SLinus Torvalds /* fall through */ 10771da177e4SLinus Torvalds case MAP_PRIVATE: 10781da177e4SLinus Torvalds if (!(file->f_mode & FMODE_READ)) 10791da177e4SLinus Torvalds return -EACCES; 1080d3ac7f89SJosef "Jeff" Sipek if (file->f_path.mnt->mnt_flags & MNT_NOEXEC) { 108180c5606cSLinus Torvalds if (vm_flags & VM_EXEC) 108280c5606cSLinus Torvalds return -EPERM; 108380c5606cSLinus Torvalds vm_flags &= ~VM_MAYEXEC; 108480c5606cSLinus Torvalds } 108580c5606cSLinus Torvalds 108680c5606cSLinus Torvalds if (!file->f_op || !file->f_op->mmap) 108780c5606cSLinus Torvalds return -ENODEV; 10881da177e4SLinus Torvalds break; 10891da177e4SLinus Torvalds 10901da177e4SLinus Torvalds default: 10911da177e4SLinus Torvalds return -EINVAL; 10921da177e4SLinus Torvalds } 10931da177e4SLinus Torvalds } else { 10941da177e4SLinus Torvalds switch (flags & MAP_TYPE) { 10951da177e4SLinus Torvalds case MAP_SHARED: 1096ce363942STejun Heo /* 1097ce363942STejun Heo * Ignore pgoff. 1098ce363942STejun Heo */ 1099ce363942STejun Heo pgoff = 0; 11001da177e4SLinus Torvalds vm_flags |= VM_SHARED | VM_MAYSHARE; 11011da177e4SLinus Torvalds break; 11021da177e4SLinus Torvalds case MAP_PRIVATE: 11031da177e4SLinus Torvalds /* 11041da177e4SLinus Torvalds * Set pgoff according to addr for anon_vma. 11051da177e4SLinus Torvalds */ 11061da177e4SLinus Torvalds pgoff = addr >> PAGE_SHIFT; 11071da177e4SLinus Torvalds break; 11081da177e4SLinus Torvalds default: 11091da177e4SLinus Torvalds return -EINVAL; 11101da177e4SLinus Torvalds } 11111da177e4SLinus Torvalds } 11121da177e4SLinus Torvalds 1113ed032189SEric Paris error = security_file_mmap(file, reqprot, prot, flags, addr, 0); 11141da177e4SLinus Torvalds if (error) 11151da177e4SLinus Torvalds return error; 11161da177e4SLinus Torvalds 11175a6fe125SMel Gorman return mmap_region(file, addr, len, flags, vm_flags, pgoff); 11180165ab44SMiklos Szeredi } 11190165ab44SMiklos Szeredi EXPORT_SYMBOL(do_mmap_pgoff); 11200165ab44SMiklos Szeredi 112166f0dc48SHugh Dickins SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len, 112266f0dc48SHugh Dickins unsigned long, prot, unsigned long, flags, 112366f0dc48SHugh Dickins unsigned long, fd, unsigned long, pgoff) 112466f0dc48SHugh Dickins { 112566f0dc48SHugh Dickins struct file *file = NULL; 112666f0dc48SHugh Dickins unsigned long retval = -EBADF; 112766f0dc48SHugh Dickins 112866f0dc48SHugh Dickins if (!(flags & MAP_ANONYMOUS)) { 1129120a795dSAl Viro audit_mmap_fd(fd, flags); 113066f0dc48SHugh Dickins if (unlikely(flags & MAP_HUGETLB)) 113166f0dc48SHugh Dickins return -EINVAL; 113266f0dc48SHugh Dickins file = fget(fd); 113366f0dc48SHugh Dickins if (!file) 113466f0dc48SHugh Dickins goto out; 113566f0dc48SHugh Dickins } else if (flags & MAP_HUGETLB) { 113666f0dc48SHugh Dickins struct user_struct *user = NULL; 113766f0dc48SHugh Dickins /* 113866f0dc48SHugh Dickins * VM_NORESERVE is used because the reservations will be 113966f0dc48SHugh Dickins * taken when vm_ops->mmap() is called 114066f0dc48SHugh Dickins * A dummy user value is used because we are not locking 114166f0dc48SHugh Dickins * memory so no accounting is necessary 114266f0dc48SHugh Dickins */ 114366f0dc48SHugh Dickins len = ALIGN(len, huge_page_size(&default_hstate)); 114466f0dc48SHugh Dickins file = hugetlb_file_setup(HUGETLB_ANON_FILE, len, VM_NORESERVE, 114566f0dc48SHugh Dickins &user, HUGETLB_ANONHUGE_INODE); 114666f0dc48SHugh Dickins if (IS_ERR(file)) 114766f0dc48SHugh Dickins return PTR_ERR(file); 114866f0dc48SHugh Dickins } 114966f0dc48SHugh Dickins 115066f0dc48SHugh Dickins flags &= ~(MAP_EXECUTABLE | MAP_DENYWRITE); 115166f0dc48SHugh Dickins 115266f0dc48SHugh Dickins down_write(¤t->mm->mmap_sem); 115366f0dc48SHugh Dickins retval = do_mmap_pgoff(file, addr, len, prot, flags, pgoff); 115466f0dc48SHugh Dickins up_write(¤t->mm->mmap_sem); 115566f0dc48SHugh Dickins 115666f0dc48SHugh Dickins if (file) 115766f0dc48SHugh Dickins fput(file); 115866f0dc48SHugh Dickins out: 115966f0dc48SHugh Dickins return retval; 116066f0dc48SHugh Dickins } 116166f0dc48SHugh Dickins 1162a4679373SChristoph Hellwig #ifdef __ARCH_WANT_SYS_OLD_MMAP 1163a4679373SChristoph Hellwig struct mmap_arg_struct { 1164a4679373SChristoph Hellwig unsigned long addr; 1165a4679373SChristoph Hellwig unsigned long len; 1166a4679373SChristoph Hellwig unsigned long prot; 1167a4679373SChristoph Hellwig unsigned long flags; 1168a4679373SChristoph Hellwig unsigned long fd; 1169a4679373SChristoph Hellwig unsigned long offset; 1170a4679373SChristoph Hellwig }; 1171a4679373SChristoph Hellwig 1172a4679373SChristoph Hellwig SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg) 1173a4679373SChristoph Hellwig { 1174a4679373SChristoph Hellwig struct mmap_arg_struct a; 1175a4679373SChristoph Hellwig 1176a4679373SChristoph Hellwig if (copy_from_user(&a, arg, sizeof(a))) 1177a4679373SChristoph Hellwig return -EFAULT; 1178a4679373SChristoph Hellwig if (a.offset & ~PAGE_MASK) 1179a4679373SChristoph Hellwig return -EINVAL; 1180a4679373SChristoph Hellwig 1181a4679373SChristoph Hellwig return sys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd, 1182a4679373SChristoph Hellwig a.offset >> PAGE_SHIFT); 1183a4679373SChristoph Hellwig } 1184a4679373SChristoph Hellwig #endif /* __ARCH_WANT_SYS_OLD_MMAP */ 1185a4679373SChristoph Hellwig 11864e950f6fSAlexey Dobriyan /* 11874e950f6fSAlexey Dobriyan * Some shared mappigns will want the pages marked read-only 11884e950f6fSAlexey Dobriyan * to track write events. If so, we'll downgrade vm_page_prot 11894e950f6fSAlexey Dobriyan * to the private version (using protection_map[] without the 11904e950f6fSAlexey Dobriyan * VM_SHARED bit). 11914e950f6fSAlexey Dobriyan */ 11924e950f6fSAlexey Dobriyan int vma_wants_writenotify(struct vm_area_struct *vma) 11934e950f6fSAlexey Dobriyan { 11944e950f6fSAlexey Dobriyan unsigned int vm_flags = vma->vm_flags; 11954e950f6fSAlexey Dobriyan 11964e950f6fSAlexey Dobriyan /* If it was private or non-writable, the write bit is already clear */ 11974e950f6fSAlexey Dobriyan if ((vm_flags & (VM_WRITE|VM_SHARED)) != ((VM_WRITE|VM_SHARED))) 11984e950f6fSAlexey Dobriyan return 0; 11994e950f6fSAlexey Dobriyan 12004e950f6fSAlexey Dobriyan /* The backer wishes to know when pages are first written to? */ 12014e950f6fSAlexey Dobriyan if (vma->vm_ops && vma->vm_ops->page_mkwrite) 12024e950f6fSAlexey Dobriyan return 1; 12034e950f6fSAlexey Dobriyan 12044e950f6fSAlexey Dobriyan /* The open routine did something to the protections already? */ 12054e950f6fSAlexey Dobriyan if (pgprot_val(vma->vm_page_prot) != 12063ed75eb8SColy Li pgprot_val(vm_get_page_prot(vm_flags))) 12074e950f6fSAlexey Dobriyan return 0; 12084e950f6fSAlexey Dobriyan 12094e950f6fSAlexey Dobriyan /* Specialty mapping? */ 12104e950f6fSAlexey Dobriyan if (vm_flags & (VM_PFNMAP|VM_INSERTPAGE)) 12114e950f6fSAlexey Dobriyan return 0; 12124e950f6fSAlexey Dobriyan 12134e950f6fSAlexey Dobriyan /* Can the mapping track the dirty pages? */ 12144e950f6fSAlexey Dobriyan return vma->vm_file && vma->vm_file->f_mapping && 12154e950f6fSAlexey Dobriyan mapping_cap_account_dirty(vma->vm_file->f_mapping); 12164e950f6fSAlexey Dobriyan } 12174e950f6fSAlexey Dobriyan 1218fc8744adSLinus Torvalds /* 1219fc8744adSLinus Torvalds * We account for memory if it's a private writeable mapping, 12205a6fe125SMel Gorman * not hugepages and VM_NORESERVE wasn't set. 1221fc8744adSLinus Torvalds */ 12225a6fe125SMel Gorman static inline int accountable_mapping(struct file *file, unsigned int vm_flags) 1223fc8744adSLinus Torvalds { 12245a6fe125SMel Gorman /* 12255a6fe125SMel Gorman * hugetlb has its own accounting separate from the core VM 12265a6fe125SMel Gorman * VM_HUGETLB may not be set yet so we cannot check for that flag. 12275a6fe125SMel Gorman */ 12285a6fe125SMel Gorman if (file && is_file_hugepages(file)) 12295a6fe125SMel Gorman return 0; 12305a6fe125SMel Gorman 1231fc8744adSLinus Torvalds return (vm_flags & (VM_NORESERVE | VM_SHARED | VM_WRITE)) == VM_WRITE; 1232fc8744adSLinus Torvalds } 1233fc8744adSLinus Torvalds 12340165ab44SMiklos Szeredi unsigned long mmap_region(struct file *file, unsigned long addr, 12350165ab44SMiklos Szeredi unsigned long len, unsigned long flags, 12365a6fe125SMel Gorman unsigned int vm_flags, unsigned long pgoff) 12370165ab44SMiklos Szeredi { 12380165ab44SMiklos Szeredi struct mm_struct *mm = current->mm; 12390165ab44SMiklos Szeredi struct vm_area_struct *vma, *prev; 12400165ab44SMiklos Szeredi int correct_wcount = 0; 12410165ab44SMiklos Szeredi int error; 12420165ab44SMiklos Szeredi struct rb_node **rb_link, *rb_parent; 12430165ab44SMiklos Szeredi unsigned long charged = 0; 12440165ab44SMiklos Szeredi struct inode *inode = file ? file->f_path.dentry->d_inode : NULL; 12450165ab44SMiklos Szeredi 12461da177e4SLinus Torvalds /* Clear old maps */ 12471da177e4SLinus Torvalds error = -ENOMEM; 12481da177e4SLinus Torvalds munmap_back: 12491da177e4SLinus Torvalds vma = find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent); 12501da177e4SLinus Torvalds if (vma && vma->vm_start < addr + len) { 12511da177e4SLinus Torvalds if (do_munmap(mm, addr, len)) 12521da177e4SLinus Torvalds return -ENOMEM; 12531da177e4SLinus Torvalds goto munmap_back; 12541da177e4SLinus Torvalds } 12551da177e4SLinus Torvalds 12561da177e4SLinus Torvalds /* Check against address space limit. */ 1257119f657cSakpm@osdl.org if (!may_expand_vm(mm, len >> PAGE_SHIFT)) 12581da177e4SLinus Torvalds return -ENOMEM; 12591da177e4SLinus Torvalds 1260fc8744adSLinus Torvalds /* 1261fc8744adSLinus Torvalds * Set 'VM_NORESERVE' if we should not account for the 12625a6fe125SMel Gorman * memory use of this mapping. 1263fc8744adSLinus Torvalds */ 12645a6fe125SMel Gorman if ((flags & MAP_NORESERVE)) { 12655a6fe125SMel Gorman /* We honor MAP_NORESERVE if allowed to overcommit */ 12665a6fe125SMel Gorman if (sysctl_overcommit_memory != OVERCOMMIT_NEVER) 1267fc8744adSLinus Torvalds vm_flags |= VM_NORESERVE; 12685a6fe125SMel Gorman 12695a6fe125SMel Gorman /* hugetlb applies strict overcommit unless MAP_NORESERVE */ 12705a6fe125SMel Gorman if (file && is_file_hugepages(file)) 1271cdfd4325SAndy Whitcroft vm_flags |= VM_NORESERVE; 12725a6fe125SMel Gorman } 1273cdfd4325SAndy Whitcroft 12741da177e4SLinus Torvalds /* 12751da177e4SLinus Torvalds * Private writable mapping: check memory availability 12761da177e4SLinus Torvalds */ 12775a6fe125SMel Gorman if (accountable_mapping(file, vm_flags)) { 12781da177e4SLinus Torvalds charged = len >> PAGE_SHIFT; 12791da177e4SLinus Torvalds if (security_vm_enough_memory(charged)) 12801da177e4SLinus Torvalds return -ENOMEM; 12811da177e4SLinus Torvalds vm_flags |= VM_ACCOUNT; 12821da177e4SLinus Torvalds } 12831da177e4SLinus Torvalds 12841da177e4SLinus Torvalds /* 1285de33c8dbSLinus Torvalds * Can we just expand an old mapping? 12861da177e4SLinus Torvalds */ 1287de33c8dbSLinus Torvalds vma = vma_merge(mm, prev, addr, addr + len, vm_flags, NULL, file, pgoff, NULL); 1288ba470de4SRik van Riel if (vma) 12891da177e4SLinus Torvalds goto out; 12901da177e4SLinus Torvalds 12911da177e4SLinus Torvalds /* 12921da177e4SLinus Torvalds * Determine the object being mapped and call the appropriate 12931da177e4SLinus Torvalds * specific mapper. the address has already been validated, but 12941da177e4SLinus Torvalds * not unmapped, but the maps are removed from the list. 12951da177e4SLinus Torvalds */ 1296c5e3b83eSPekka Enberg vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL); 12971da177e4SLinus Torvalds if (!vma) { 12981da177e4SLinus Torvalds error = -ENOMEM; 12991da177e4SLinus Torvalds goto unacct_error; 13001da177e4SLinus Torvalds } 13011da177e4SLinus Torvalds 13021da177e4SLinus Torvalds vma->vm_mm = mm; 13031da177e4SLinus Torvalds vma->vm_start = addr; 13041da177e4SLinus Torvalds vma->vm_end = addr + len; 13051da177e4SLinus Torvalds vma->vm_flags = vm_flags; 13063ed75eb8SColy Li vma->vm_page_prot = vm_get_page_prot(vm_flags); 13071da177e4SLinus Torvalds vma->vm_pgoff = pgoff; 13085beb4930SRik van Riel INIT_LIST_HEAD(&vma->anon_vma_chain); 13091da177e4SLinus Torvalds 13101da177e4SLinus Torvalds if (file) { 13111da177e4SLinus Torvalds error = -EINVAL; 13121da177e4SLinus Torvalds if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP)) 13131da177e4SLinus Torvalds goto free_vma; 13141da177e4SLinus Torvalds if (vm_flags & VM_DENYWRITE) { 13151da177e4SLinus Torvalds error = deny_write_access(file); 13161da177e4SLinus Torvalds if (error) 13171da177e4SLinus Torvalds goto free_vma; 13181da177e4SLinus Torvalds correct_wcount = 1; 13191da177e4SLinus Torvalds } 13201da177e4SLinus Torvalds vma->vm_file = file; 13211da177e4SLinus Torvalds get_file(file); 13221da177e4SLinus Torvalds error = file->f_op->mmap(file, vma); 13231da177e4SLinus Torvalds if (error) 13241da177e4SLinus Torvalds goto unmap_and_free_vma; 1325925d1c40SMatt Helsley if (vm_flags & VM_EXECUTABLE) 1326925d1c40SMatt Helsley added_exe_file_vma(mm); 13271da177e4SLinus Torvalds 13281da177e4SLinus Torvalds /* Can addr have changed?? 13291da177e4SLinus Torvalds * 13301da177e4SLinus Torvalds * Answer: Yes, several device drivers can do it in their 13311da177e4SLinus Torvalds * f_op->mmap method. -DaveM 13321da177e4SLinus Torvalds */ 13331da177e4SLinus Torvalds addr = vma->vm_start; 13341da177e4SLinus Torvalds pgoff = vma->vm_pgoff; 13351da177e4SLinus Torvalds vm_flags = vma->vm_flags; 1336f8dbf0a7SHuang Shijie } else if (vm_flags & VM_SHARED) { 1337f8dbf0a7SHuang Shijie error = shmem_zero_setup(vma); 1338f8dbf0a7SHuang Shijie if (error) 1339f8dbf0a7SHuang Shijie goto free_vma; 1340f8dbf0a7SHuang Shijie } 13411da177e4SLinus Torvalds 1342c9d0bf24SMagnus Damm if (vma_wants_writenotify(vma)) { 1343c9d0bf24SMagnus Damm pgprot_t pprot = vma->vm_page_prot; 1344c9d0bf24SMagnus Damm 1345c9d0bf24SMagnus Damm /* Can vma->vm_page_prot have changed?? 1346c9d0bf24SMagnus Damm * 1347c9d0bf24SMagnus Damm * Answer: Yes, drivers may have changed it in their 1348c9d0bf24SMagnus Damm * f_op->mmap method. 1349c9d0bf24SMagnus Damm * 1350c9d0bf24SMagnus Damm * Ensures that vmas marked as uncached stay that way. 1351c9d0bf24SMagnus Damm */ 13521ddd439eSHugh Dickins vma->vm_page_prot = vm_get_page_prot(vm_flags & ~VM_SHARED); 1353c9d0bf24SMagnus Damm if (pgprot_val(pprot) == pgprot_val(pgprot_noncached(pprot))) 1354c9d0bf24SMagnus Damm vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot); 1355c9d0bf24SMagnus Damm } 1356d08b3851SPeter Zijlstra 13574d3d5b41SOleg Nesterov vma_link(mm, vma, prev, rb_link, rb_parent); 13584d3d5b41SOleg Nesterov file = vma->vm_file; 13594d3d5b41SOleg Nesterov 13604d3d5b41SOleg Nesterov /* Once vma denies write, undo our temporary denial count */ 13614d3d5b41SOleg Nesterov if (correct_wcount) 13624d3d5b41SOleg Nesterov atomic_inc(&inode->i_writecount); 13631da177e4SLinus Torvalds out: 1364cdd6c482SIngo Molnar perf_event_mmap(vma); 13650a4a9391SPeter Zijlstra 13661da177e4SLinus Torvalds mm->total_vm += len >> PAGE_SHIFT; 1367ab50b8edSHugh Dickins vm_stat_account(mm, vm_flags, file, len >> PAGE_SHIFT); 13681da177e4SLinus Torvalds if (vm_flags & VM_LOCKED) { 136906f9d8c2SKOSAKI Motohiro if (!mlock_vma_pages_range(vma, addr, addr + len)) 137006f9d8c2SKOSAKI Motohiro mm->locked_vm += (len >> PAGE_SHIFT); 1371ba470de4SRik van Riel } else if ((flags & MAP_POPULATE) && !(flags & MAP_NONBLOCK)) 137254cb8821SNick Piggin make_pages_present(addr, addr + len); 13731da177e4SLinus Torvalds return addr; 13741da177e4SLinus Torvalds 13751da177e4SLinus Torvalds unmap_and_free_vma: 13761da177e4SLinus Torvalds if (correct_wcount) 13771da177e4SLinus Torvalds atomic_inc(&inode->i_writecount); 13781da177e4SLinus Torvalds vma->vm_file = NULL; 13791da177e4SLinus Torvalds fput(file); 13801da177e4SLinus Torvalds 13811da177e4SLinus Torvalds /* Undo any partial mapping done by a device driver. */ 1382e0da382cSHugh Dickins unmap_region(mm, vma, prev, vma->vm_start, vma->vm_end); 1383e0da382cSHugh Dickins charged = 0; 13841da177e4SLinus Torvalds free_vma: 13851da177e4SLinus Torvalds kmem_cache_free(vm_area_cachep, vma); 13861da177e4SLinus Torvalds unacct_error: 13871da177e4SLinus Torvalds if (charged) 13881da177e4SLinus Torvalds vm_unacct_memory(charged); 13891da177e4SLinus Torvalds return error; 13901da177e4SLinus Torvalds } 13911da177e4SLinus Torvalds 13921da177e4SLinus Torvalds /* Get an address range which is currently unmapped. 13931da177e4SLinus Torvalds * For shmat() with addr=0. 13941da177e4SLinus Torvalds * 13951da177e4SLinus Torvalds * Ugly calling convention alert: 13961da177e4SLinus Torvalds * Return value with the low bits set means error value, 13971da177e4SLinus Torvalds * ie 13981da177e4SLinus Torvalds * if (ret & ~PAGE_MASK) 13991da177e4SLinus Torvalds * error = ret; 14001da177e4SLinus Torvalds * 14011da177e4SLinus Torvalds * This function "knows" that -ENOMEM has the bits set. 14021da177e4SLinus Torvalds */ 14031da177e4SLinus Torvalds #ifndef HAVE_ARCH_UNMAPPED_AREA 14041da177e4SLinus Torvalds unsigned long 14051da177e4SLinus Torvalds arch_get_unmapped_area(struct file *filp, unsigned long addr, 14061da177e4SLinus Torvalds unsigned long len, unsigned long pgoff, unsigned long flags) 14071da177e4SLinus Torvalds { 14081da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 14091da177e4SLinus Torvalds struct vm_area_struct *vma; 14101da177e4SLinus Torvalds unsigned long start_addr; 14111da177e4SLinus Torvalds 14121da177e4SLinus Torvalds if (len > TASK_SIZE) 14131da177e4SLinus Torvalds return -ENOMEM; 14141da177e4SLinus Torvalds 141506abdfb4SBenjamin Herrenschmidt if (flags & MAP_FIXED) 141606abdfb4SBenjamin Herrenschmidt return addr; 141706abdfb4SBenjamin Herrenschmidt 14181da177e4SLinus Torvalds if (addr) { 14191da177e4SLinus Torvalds addr = PAGE_ALIGN(addr); 14201da177e4SLinus Torvalds vma = find_vma(mm, addr); 14211da177e4SLinus Torvalds if (TASK_SIZE - len >= addr && 14221da177e4SLinus Torvalds (!vma || addr + len <= vma->vm_start)) 14231da177e4SLinus Torvalds return addr; 14241da177e4SLinus Torvalds } 14251363c3cdSWolfgang Wander if (len > mm->cached_hole_size) { 14261da177e4SLinus Torvalds start_addr = addr = mm->free_area_cache; 14271363c3cdSWolfgang Wander } else { 14281363c3cdSWolfgang Wander start_addr = addr = TASK_UNMAPPED_BASE; 14291363c3cdSWolfgang Wander mm->cached_hole_size = 0; 14301363c3cdSWolfgang Wander } 14311da177e4SLinus Torvalds 14321da177e4SLinus Torvalds full_search: 14331da177e4SLinus Torvalds for (vma = find_vma(mm, addr); ; vma = vma->vm_next) { 14341da177e4SLinus Torvalds /* At this point: (!vma || addr < vma->vm_end). */ 14351da177e4SLinus Torvalds if (TASK_SIZE - len < addr) { 14361da177e4SLinus Torvalds /* 14371da177e4SLinus Torvalds * Start a new search - just in case we missed 14381da177e4SLinus Torvalds * some holes. 14391da177e4SLinus Torvalds */ 14401da177e4SLinus Torvalds if (start_addr != TASK_UNMAPPED_BASE) { 14411363c3cdSWolfgang Wander addr = TASK_UNMAPPED_BASE; 14421363c3cdSWolfgang Wander start_addr = addr; 14431363c3cdSWolfgang Wander mm->cached_hole_size = 0; 14441da177e4SLinus Torvalds goto full_search; 14451da177e4SLinus Torvalds } 14461da177e4SLinus Torvalds return -ENOMEM; 14471da177e4SLinus Torvalds } 14481da177e4SLinus Torvalds if (!vma || addr + len <= vma->vm_start) { 14491da177e4SLinus Torvalds /* 14501da177e4SLinus Torvalds * Remember the place where we stopped the search: 14511da177e4SLinus Torvalds */ 14521da177e4SLinus Torvalds mm->free_area_cache = addr + len; 14531da177e4SLinus Torvalds return addr; 14541da177e4SLinus Torvalds } 14551363c3cdSWolfgang Wander if (addr + mm->cached_hole_size < vma->vm_start) 14561363c3cdSWolfgang Wander mm->cached_hole_size = vma->vm_start - addr; 14571da177e4SLinus Torvalds addr = vma->vm_end; 14581da177e4SLinus Torvalds } 14591da177e4SLinus Torvalds } 14601da177e4SLinus Torvalds #endif 14611da177e4SLinus Torvalds 14621363c3cdSWolfgang Wander void arch_unmap_area(struct mm_struct *mm, unsigned long addr) 14631da177e4SLinus Torvalds { 14641da177e4SLinus Torvalds /* 14651da177e4SLinus Torvalds * Is this a new hole at the lowest possible address? 14661da177e4SLinus Torvalds */ 14671363c3cdSWolfgang Wander if (addr >= TASK_UNMAPPED_BASE && addr < mm->free_area_cache) { 14681363c3cdSWolfgang Wander mm->free_area_cache = addr; 14691363c3cdSWolfgang Wander mm->cached_hole_size = ~0UL; 14701363c3cdSWolfgang Wander } 14711da177e4SLinus Torvalds } 14721da177e4SLinus Torvalds 14731da177e4SLinus Torvalds /* 14741da177e4SLinus Torvalds * This mmap-allocator allocates new areas top-down from below the 14751da177e4SLinus Torvalds * stack's low limit (the base): 14761da177e4SLinus Torvalds */ 14771da177e4SLinus Torvalds #ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN 14781da177e4SLinus Torvalds unsigned long 14791da177e4SLinus Torvalds arch_get_unmapped_area_topdown(struct file *filp, const unsigned long addr0, 14801da177e4SLinus Torvalds const unsigned long len, const unsigned long pgoff, 14811da177e4SLinus Torvalds const unsigned long flags) 14821da177e4SLinus Torvalds { 14831da177e4SLinus Torvalds struct vm_area_struct *vma; 14841da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 14851da177e4SLinus Torvalds unsigned long addr = addr0; 14861da177e4SLinus Torvalds 14871da177e4SLinus Torvalds /* requested length too big for entire address space */ 14881da177e4SLinus Torvalds if (len > TASK_SIZE) 14891da177e4SLinus Torvalds return -ENOMEM; 14901da177e4SLinus Torvalds 149106abdfb4SBenjamin Herrenschmidt if (flags & MAP_FIXED) 149206abdfb4SBenjamin Herrenschmidt return addr; 149306abdfb4SBenjamin Herrenschmidt 14941da177e4SLinus Torvalds /* requesting a specific address */ 14951da177e4SLinus Torvalds if (addr) { 14961da177e4SLinus Torvalds addr = PAGE_ALIGN(addr); 14971da177e4SLinus Torvalds vma = find_vma(mm, addr); 14981da177e4SLinus Torvalds if (TASK_SIZE - len >= addr && 14991da177e4SLinus Torvalds (!vma || addr + len <= vma->vm_start)) 15001da177e4SLinus Torvalds return addr; 15011da177e4SLinus Torvalds } 15021da177e4SLinus Torvalds 15031363c3cdSWolfgang Wander /* check if free_area_cache is useful for us */ 15041363c3cdSWolfgang Wander if (len <= mm->cached_hole_size) { 15051363c3cdSWolfgang Wander mm->cached_hole_size = 0; 15061363c3cdSWolfgang Wander mm->free_area_cache = mm->mmap_base; 15071363c3cdSWolfgang Wander } 15081363c3cdSWolfgang Wander 15091da177e4SLinus Torvalds /* either no address requested or can't fit in requested address hole */ 15101da177e4SLinus Torvalds addr = mm->free_area_cache; 15111da177e4SLinus Torvalds 15121da177e4SLinus Torvalds /* make sure it can fit in the remaining address space */ 151349a43876SLinus Torvalds if (addr > len) { 15141da177e4SLinus Torvalds vma = find_vma(mm, addr-len); 15151da177e4SLinus Torvalds if (!vma || addr <= vma->vm_start) 15161da177e4SLinus Torvalds /* remember the address as a hint for next time */ 15171da177e4SLinus Torvalds return (mm->free_area_cache = addr-len); 15181da177e4SLinus Torvalds } 15191da177e4SLinus Torvalds 152073219d17SChris Wright if (mm->mmap_base < len) 152173219d17SChris Wright goto bottomup; 152273219d17SChris Wright 15231da177e4SLinus Torvalds addr = mm->mmap_base-len; 15241da177e4SLinus Torvalds 15251da177e4SLinus Torvalds do { 15261da177e4SLinus Torvalds /* 15271da177e4SLinus Torvalds * Lookup failure means no vma is above this address, 15281da177e4SLinus Torvalds * else if new region fits below vma->vm_start, 15291da177e4SLinus Torvalds * return with success: 15301da177e4SLinus Torvalds */ 15311da177e4SLinus Torvalds vma = find_vma(mm, addr); 15321da177e4SLinus Torvalds if (!vma || addr+len <= vma->vm_start) 15331da177e4SLinus Torvalds /* remember the address as a hint for next time */ 15341da177e4SLinus Torvalds return (mm->free_area_cache = addr); 15351da177e4SLinus Torvalds 15361363c3cdSWolfgang Wander /* remember the largest hole we saw so far */ 15371363c3cdSWolfgang Wander if (addr + mm->cached_hole_size < vma->vm_start) 15381363c3cdSWolfgang Wander mm->cached_hole_size = vma->vm_start - addr; 15391363c3cdSWolfgang Wander 15401da177e4SLinus Torvalds /* try just below the current vma->vm_start */ 15411da177e4SLinus Torvalds addr = vma->vm_start-len; 154249a43876SLinus Torvalds } while (len < vma->vm_start); 15431da177e4SLinus Torvalds 154473219d17SChris Wright bottomup: 15451da177e4SLinus Torvalds /* 15461da177e4SLinus Torvalds * A failed mmap() very likely causes application failure, 15471da177e4SLinus Torvalds * so fall back to the bottom-up function here. This scenario 15481da177e4SLinus Torvalds * can happen with large stack limits and large mmap() 15491da177e4SLinus Torvalds * allocations. 15501da177e4SLinus Torvalds */ 15511363c3cdSWolfgang Wander mm->cached_hole_size = ~0UL; 15521da177e4SLinus Torvalds mm->free_area_cache = TASK_UNMAPPED_BASE; 15531da177e4SLinus Torvalds addr = arch_get_unmapped_area(filp, addr0, len, pgoff, flags); 15541da177e4SLinus Torvalds /* 15551da177e4SLinus Torvalds * Restore the topdown base: 15561da177e4SLinus Torvalds */ 15571da177e4SLinus Torvalds mm->free_area_cache = mm->mmap_base; 15581363c3cdSWolfgang Wander mm->cached_hole_size = ~0UL; 15591da177e4SLinus Torvalds 15601da177e4SLinus Torvalds return addr; 15611da177e4SLinus Torvalds } 15621da177e4SLinus Torvalds #endif 15631da177e4SLinus Torvalds 15641363c3cdSWolfgang Wander void arch_unmap_area_topdown(struct mm_struct *mm, unsigned long addr) 15651da177e4SLinus Torvalds { 15661da177e4SLinus Torvalds /* 15671da177e4SLinus Torvalds * Is this a new hole at the highest possible address? 15681da177e4SLinus Torvalds */ 15691363c3cdSWolfgang Wander if (addr > mm->free_area_cache) 15701363c3cdSWolfgang Wander mm->free_area_cache = addr; 15711da177e4SLinus Torvalds 15721da177e4SLinus Torvalds /* dont allow allocations above current base */ 15731363c3cdSWolfgang Wander if (mm->free_area_cache > mm->mmap_base) 15741363c3cdSWolfgang Wander mm->free_area_cache = mm->mmap_base; 15751da177e4SLinus Torvalds } 15761da177e4SLinus Torvalds 15771da177e4SLinus Torvalds unsigned long 15781da177e4SLinus Torvalds get_unmapped_area(struct file *file, unsigned long addr, unsigned long len, 15791da177e4SLinus Torvalds unsigned long pgoff, unsigned long flags) 15801da177e4SLinus Torvalds { 158106abdfb4SBenjamin Herrenschmidt unsigned long (*get_area)(struct file *, unsigned long, 158206abdfb4SBenjamin Herrenschmidt unsigned long, unsigned long, unsigned long); 158307ab67c8SLinus Torvalds 15849206de95SAl Viro unsigned long error = arch_mmap_check(addr, len, flags); 15859206de95SAl Viro if (error) 15869206de95SAl Viro return error; 15879206de95SAl Viro 15889206de95SAl Viro /* Careful about overflows.. */ 15899206de95SAl Viro if (len > TASK_SIZE) 15909206de95SAl Viro return -ENOMEM; 15919206de95SAl Viro 159207ab67c8SLinus Torvalds get_area = current->mm->get_unmapped_area; 159307ab67c8SLinus Torvalds if (file && file->f_op && file->f_op->get_unmapped_area) 159407ab67c8SLinus Torvalds get_area = file->f_op->get_unmapped_area; 159507ab67c8SLinus Torvalds addr = get_area(file, addr, len, pgoff, flags); 159607ab67c8SLinus Torvalds if (IS_ERR_VALUE(addr)) 159707ab67c8SLinus Torvalds return addr; 159807ab67c8SLinus Torvalds 15991da177e4SLinus Torvalds if (addr > TASK_SIZE - len) 16001da177e4SLinus Torvalds return -ENOMEM; 16011da177e4SLinus Torvalds if (addr & ~PAGE_MASK) 16021da177e4SLinus Torvalds return -EINVAL; 160306abdfb4SBenjamin Herrenschmidt 160408e7d9b5SMartin Schwidefsky return arch_rebalance_pgtables(addr, len); 16051da177e4SLinus Torvalds } 16061da177e4SLinus Torvalds 16071da177e4SLinus Torvalds EXPORT_SYMBOL(get_unmapped_area); 16081da177e4SLinus Torvalds 16091da177e4SLinus Torvalds /* Look up the first VMA which satisfies addr < vm_end, NULL if none. */ 16101da177e4SLinus Torvalds struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr) 16111da177e4SLinus Torvalds { 16121da177e4SLinus Torvalds struct vm_area_struct *vma = NULL; 16131da177e4SLinus Torvalds 16141da177e4SLinus Torvalds if (mm) { 16151da177e4SLinus Torvalds /* Check the cache first. */ 16161da177e4SLinus Torvalds /* (Cache hit rate is typically around 35%.) */ 16171da177e4SLinus Torvalds vma = mm->mmap_cache; 16181da177e4SLinus Torvalds if (!(vma && vma->vm_end > addr && vma->vm_start <= addr)) { 16191da177e4SLinus Torvalds struct rb_node * rb_node; 16201da177e4SLinus Torvalds 16211da177e4SLinus Torvalds rb_node = mm->mm_rb.rb_node; 16221da177e4SLinus Torvalds vma = NULL; 16231da177e4SLinus Torvalds 16241da177e4SLinus Torvalds while (rb_node) { 16251da177e4SLinus Torvalds struct vm_area_struct * vma_tmp; 16261da177e4SLinus Torvalds 16271da177e4SLinus Torvalds vma_tmp = rb_entry(rb_node, 16281da177e4SLinus Torvalds struct vm_area_struct, vm_rb); 16291da177e4SLinus Torvalds 16301da177e4SLinus Torvalds if (vma_tmp->vm_end > addr) { 16311da177e4SLinus Torvalds vma = vma_tmp; 16321da177e4SLinus Torvalds if (vma_tmp->vm_start <= addr) 16331da177e4SLinus Torvalds break; 16341da177e4SLinus Torvalds rb_node = rb_node->rb_left; 16351da177e4SLinus Torvalds } else 16361da177e4SLinus Torvalds rb_node = rb_node->rb_right; 16371da177e4SLinus Torvalds } 16381da177e4SLinus Torvalds if (vma) 16391da177e4SLinus Torvalds mm->mmap_cache = vma; 16401da177e4SLinus Torvalds } 16411da177e4SLinus Torvalds } 16421da177e4SLinus Torvalds return vma; 16431da177e4SLinus Torvalds } 16441da177e4SLinus Torvalds 16451da177e4SLinus Torvalds EXPORT_SYMBOL(find_vma); 16461da177e4SLinus Torvalds 16471da177e4SLinus Torvalds /* Same as find_vma, but also return a pointer to the previous VMA in *pprev. */ 16481da177e4SLinus Torvalds struct vm_area_struct * 16491da177e4SLinus Torvalds find_vma_prev(struct mm_struct *mm, unsigned long addr, 16501da177e4SLinus Torvalds struct vm_area_struct **pprev) 16511da177e4SLinus Torvalds { 16521da177e4SLinus Torvalds struct vm_area_struct *vma = NULL, *prev = NULL; 16531da177e4SLinus Torvalds struct rb_node *rb_node; 16541da177e4SLinus Torvalds if (!mm) 16551da177e4SLinus Torvalds goto out; 16561da177e4SLinus Torvalds 16571da177e4SLinus Torvalds /* Guard against addr being lower than the first VMA */ 16581da177e4SLinus Torvalds vma = mm->mmap; 16591da177e4SLinus Torvalds 16601da177e4SLinus Torvalds /* Go through the RB tree quickly. */ 16611da177e4SLinus Torvalds rb_node = mm->mm_rb.rb_node; 16621da177e4SLinus Torvalds 16631da177e4SLinus Torvalds while (rb_node) { 16641da177e4SLinus Torvalds struct vm_area_struct *vma_tmp; 16651da177e4SLinus Torvalds vma_tmp = rb_entry(rb_node, struct vm_area_struct, vm_rb); 16661da177e4SLinus Torvalds 16671da177e4SLinus Torvalds if (addr < vma_tmp->vm_end) { 16681da177e4SLinus Torvalds rb_node = rb_node->rb_left; 16691da177e4SLinus Torvalds } else { 16701da177e4SLinus Torvalds prev = vma_tmp; 16711da177e4SLinus Torvalds if (!prev->vm_next || (addr < prev->vm_next->vm_end)) 16721da177e4SLinus Torvalds break; 16731da177e4SLinus Torvalds rb_node = rb_node->rb_right; 16741da177e4SLinus Torvalds } 16751da177e4SLinus Torvalds } 16761da177e4SLinus Torvalds 16771da177e4SLinus Torvalds out: 16781da177e4SLinus Torvalds *pprev = prev; 16791da177e4SLinus Torvalds return prev ? prev->vm_next : vma; 16801da177e4SLinus Torvalds } 16811da177e4SLinus Torvalds 16821da177e4SLinus Torvalds /* 16831da177e4SLinus Torvalds * Verify that the stack growth is acceptable and 16841da177e4SLinus Torvalds * update accounting. This is shared with both the 16851da177e4SLinus Torvalds * grow-up and grow-down cases. 16861da177e4SLinus Torvalds */ 16871da177e4SLinus Torvalds static int acct_stack_growth(struct vm_area_struct *vma, unsigned long size, unsigned long grow) 16881da177e4SLinus Torvalds { 16891da177e4SLinus Torvalds struct mm_struct *mm = vma->vm_mm; 16901da177e4SLinus Torvalds struct rlimit *rlim = current->signal->rlim; 16910d59a01bSAdam Litke unsigned long new_start; 16921da177e4SLinus Torvalds 16931da177e4SLinus Torvalds /* address space limit tests */ 1694119f657cSakpm@osdl.org if (!may_expand_vm(mm, grow)) 16951da177e4SLinus Torvalds return -ENOMEM; 16961da177e4SLinus Torvalds 16971da177e4SLinus Torvalds /* Stack limit test */ 169859e99e5bSJiri Slaby if (size > ACCESS_ONCE(rlim[RLIMIT_STACK].rlim_cur)) 16991da177e4SLinus Torvalds return -ENOMEM; 17001da177e4SLinus Torvalds 17011da177e4SLinus Torvalds /* mlock limit tests */ 17021da177e4SLinus Torvalds if (vma->vm_flags & VM_LOCKED) { 17031da177e4SLinus Torvalds unsigned long locked; 17041da177e4SLinus Torvalds unsigned long limit; 17051da177e4SLinus Torvalds locked = mm->locked_vm + grow; 170659e99e5bSJiri Slaby limit = ACCESS_ONCE(rlim[RLIMIT_MEMLOCK].rlim_cur); 170759e99e5bSJiri Slaby limit >>= PAGE_SHIFT; 17081da177e4SLinus Torvalds if (locked > limit && !capable(CAP_IPC_LOCK)) 17091da177e4SLinus Torvalds return -ENOMEM; 17101da177e4SLinus Torvalds } 17111da177e4SLinus Torvalds 17120d59a01bSAdam Litke /* Check to ensure the stack will not grow into a hugetlb-only region */ 17130d59a01bSAdam Litke new_start = (vma->vm_flags & VM_GROWSUP) ? vma->vm_start : 17140d59a01bSAdam Litke vma->vm_end - size; 17150d59a01bSAdam Litke if (is_hugepage_only_range(vma->vm_mm, new_start, size)) 17160d59a01bSAdam Litke return -EFAULT; 17170d59a01bSAdam Litke 17181da177e4SLinus Torvalds /* 17191da177e4SLinus Torvalds * Overcommit.. This must be the final test, as it will 17201da177e4SLinus Torvalds * update security statistics. 17211da177e4SLinus Torvalds */ 172205fa199dSHugh Dickins if (security_vm_enough_memory_mm(mm, grow)) 17231da177e4SLinus Torvalds return -ENOMEM; 17241da177e4SLinus Torvalds 17251da177e4SLinus Torvalds /* Ok, everything looks good - let it rip */ 17261da177e4SLinus Torvalds mm->total_vm += grow; 17271da177e4SLinus Torvalds if (vma->vm_flags & VM_LOCKED) 17281da177e4SLinus Torvalds mm->locked_vm += grow; 1729ab50b8edSHugh Dickins vm_stat_account(mm, vma->vm_flags, vma->vm_file, grow); 17301da177e4SLinus Torvalds return 0; 17311da177e4SLinus Torvalds } 17321da177e4SLinus Torvalds 173346dea3d0SHugh Dickins #if defined(CONFIG_STACK_GROWSUP) || defined(CONFIG_IA64) 17341da177e4SLinus Torvalds /* 173546dea3d0SHugh Dickins * PA-RISC uses this for its stack; IA64 for its Register Backing Store. 173646dea3d0SHugh Dickins * vma is the last one with address > vma->vm_end. Have to extend vma. 17371da177e4SLinus Torvalds */ 173846dea3d0SHugh Dickins int expand_upwards(struct vm_area_struct *vma, unsigned long address) 17391da177e4SLinus Torvalds { 17401da177e4SLinus Torvalds int error; 17411da177e4SLinus Torvalds 17421da177e4SLinus Torvalds if (!(vma->vm_flags & VM_GROWSUP)) 17431da177e4SLinus Torvalds return -EFAULT; 17441da177e4SLinus Torvalds 17451da177e4SLinus Torvalds /* 17461da177e4SLinus Torvalds * We must make sure the anon_vma is allocated 17471da177e4SLinus Torvalds * so that the anon_vma locking is not a noop. 17481da177e4SLinus Torvalds */ 17491da177e4SLinus Torvalds if (unlikely(anon_vma_prepare(vma))) 17501da177e4SLinus Torvalds return -ENOMEM; 1751bb4a340eSRik van Riel vma_lock_anon_vma(vma); 17521da177e4SLinus Torvalds 17531da177e4SLinus Torvalds /* 17541da177e4SLinus Torvalds * vma->vm_start/vm_end cannot change under us because the caller 17551da177e4SLinus Torvalds * is required to hold the mmap_sem in read mode. We need the 17561da177e4SLinus Torvalds * anon_vma lock to serialize against concurrent expand_stacks. 175706b32f3aSHelge Deller * Also guard against wrapping around to address 0. 17581da177e4SLinus Torvalds */ 175906b32f3aSHelge Deller if (address < PAGE_ALIGN(address+4)) 176006b32f3aSHelge Deller address = PAGE_ALIGN(address+4); 176106b32f3aSHelge Deller else { 1762bb4a340eSRik van Riel vma_unlock_anon_vma(vma); 176306b32f3aSHelge Deller return -ENOMEM; 176406b32f3aSHelge Deller } 17651da177e4SLinus Torvalds error = 0; 17661da177e4SLinus Torvalds 17671da177e4SLinus Torvalds /* Somebody else might have raced and expanded it already */ 17681da177e4SLinus Torvalds if (address > vma->vm_end) { 17691da177e4SLinus Torvalds unsigned long size, grow; 17701da177e4SLinus Torvalds 17711da177e4SLinus Torvalds size = address - vma->vm_start; 17721da177e4SLinus Torvalds grow = (address - vma->vm_end) >> PAGE_SHIFT; 17731da177e4SLinus Torvalds 177442c36f63SHugh Dickins error = -ENOMEM; 177542c36f63SHugh Dickins if (vma->vm_pgoff + (size >> PAGE_SHIFT) >= vma->vm_pgoff) { 17761da177e4SLinus Torvalds error = acct_stack_growth(vma, size, grow); 17773af9e859SEric B Munson if (!error) { 17781da177e4SLinus Torvalds vma->vm_end = address; 17793af9e859SEric B Munson perf_event_mmap(vma); 17803af9e859SEric B Munson } 17811da177e4SLinus Torvalds } 178242c36f63SHugh Dickins } 1783bb4a340eSRik van Riel vma_unlock_anon_vma(vma); 1784b15d00b6SAndrea Arcangeli khugepaged_enter_vma_merge(vma); 17851da177e4SLinus Torvalds return error; 17861da177e4SLinus Torvalds } 178746dea3d0SHugh Dickins #endif /* CONFIG_STACK_GROWSUP || CONFIG_IA64 */ 178846dea3d0SHugh Dickins 17891da177e4SLinus Torvalds /* 17901da177e4SLinus Torvalds * vma is the first one with address < vma->vm_start. Have to extend vma. 17911da177e4SLinus Torvalds */ 1792cb8f488cSDenys Vlasenko static int expand_downwards(struct vm_area_struct *vma, 1793b6a2fea3SOllie Wild unsigned long address) 17941da177e4SLinus Torvalds { 17951da177e4SLinus Torvalds int error; 17961da177e4SLinus Torvalds 17971da177e4SLinus Torvalds /* 17981da177e4SLinus Torvalds * We must make sure the anon_vma is allocated 17991da177e4SLinus Torvalds * so that the anon_vma locking is not a noop. 18001da177e4SLinus Torvalds */ 18011da177e4SLinus Torvalds if (unlikely(anon_vma_prepare(vma))) 18021da177e4SLinus Torvalds return -ENOMEM; 18038869477aSEric Paris 18048869477aSEric Paris address &= PAGE_MASK; 180588c3f7a8SRichard Knutsson error = security_file_mmap(NULL, 0, 0, 0, address, 1); 18068869477aSEric Paris if (error) 18078869477aSEric Paris return error; 18088869477aSEric Paris 1809bb4a340eSRik van Riel vma_lock_anon_vma(vma); 18101da177e4SLinus Torvalds 18111da177e4SLinus Torvalds /* 18121da177e4SLinus Torvalds * vma->vm_start/vm_end cannot change under us because the caller 18131da177e4SLinus Torvalds * is required to hold the mmap_sem in read mode. We need the 18141da177e4SLinus Torvalds * anon_vma lock to serialize against concurrent expand_stacks. 18151da177e4SLinus Torvalds */ 18161da177e4SLinus Torvalds 18171da177e4SLinus Torvalds /* Somebody else might have raced and expanded it already */ 18181da177e4SLinus Torvalds if (address < vma->vm_start) { 18191da177e4SLinus Torvalds unsigned long size, grow; 18201da177e4SLinus Torvalds 18211da177e4SLinus Torvalds size = vma->vm_end - address; 18221da177e4SLinus Torvalds grow = (vma->vm_start - address) >> PAGE_SHIFT; 18231da177e4SLinus Torvalds 1824a626ca6aSLinus Torvalds error = -ENOMEM; 1825a626ca6aSLinus Torvalds if (grow <= vma->vm_pgoff) { 18261da177e4SLinus Torvalds error = acct_stack_growth(vma, size, grow); 18271da177e4SLinus Torvalds if (!error) { 18281da177e4SLinus Torvalds vma->vm_start = address; 18291da177e4SLinus Torvalds vma->vm_pgoff -= grow; 18303af9e859SEric B Munson perf_event_mmap(vma); 18311da177e4SLinus Torvalds } 18321da177e4SLinus Torvalds } 1833a626ca6aSLinus Torvalds } 1834bb4a340eSRik van Riel vma_unlock_anon_vma(vma); 1835b15d00b6SAndrea Arcangeli khugepaged_enter_vma_merge(vma); 18361da177e4SLinus Torvalds return error; 18371da177e4SLinus Torvalds } 18381da177e4SLinus Torvalds 1839b6a2fea3SOllie Wild int expand_stack_downwards(struct vm_area_struct *vma, unsigned long address) 1840b6a2fea3SOllie Wild { 1841b6a2fea3SOllie Wild return expand_downwards(vma, address); 1842b6a2fea3SOllie Wild } 1843b6a2fea3SOllie Wild 1844b6a2fea3SOllie Wild #ifdef CONFIG_STACK_GROWSUP 1845b6a2fea3SOllie Wild int expand_stack(struct vm_area_struct *vma, unsigned long address) 1846b6a2fea3SOllie Wild { 1847b6a2fea3SOllie Wild return expand_upwards(vma, address); 1848b6a2fea3SOllie Wild } 1849b6a2fea3SOllie Wild 1850b6a2fea3SOllie Wild struct vm_area_struct * 1851b6a2fea3SOllie Wild find_extend_vma(struct mm_struct *mm, unsigned long addr) 1852b6a2fea3SOllie Wild { 1853b6a2fea3SOllie Wild struct vm_area_struct *vma, *prev; 1854b6a2fea3SOllie Wild 1855b6a2fea3SOllie Wild addr &= PAGE_MASK; 1856b6a2fea3SOllie Wild vma = find_vma_prev(mm, addr, &prev); 1857b6a2fea3SOllie Wild if (vma && (vma->vm_start <= addr)) 1858b6a2fea3SOllie Wild return vma; 18591c127185SDenys Vlasenko if (!prev || expand_stack(prev, addr)) 1860b6a2fea3SOllie Wild return NULL; 1861ba470de4SRik van Riel if (prev->vm_flags & VM_LOCKED) { 1862c58267c3SKOSAKI Motohiro mlock_vma_pages_range(prev, addr, prev->vm_end); 1863ba470de4SRik van Riel } 1864b6a2fea3SOllie Wild return prev; 1865b6a2fea3SOllie Wild } 1866b6a2fea3SOllie Wild #else 1867b6a2fea3SOllie Wild int expand_stack(struct vm_area_struct *vma, unsigned long address) 1868b6a2fea3SOllie Wild { 1869b6a2fea3SOllie Wild return expand_downwards(vma, address); 1870b6a2fea3SOllie Wild } 1871b6a2fea3SOllie Wild 18721da177e4SLinus Torvalds struct vm_area_struct * 18731da177e4SLinus Torvalds find_extend_vma(struct mm_struct * mm, unsigned long addr) 18741da177e4SLinus Torvalds { 18751da177e4SLinus Torvalds struct vm_area_struct * vma; 18761da177e4SLinus Torvalds unsigned long start; 18771da177e4SLinus Torvalds 18781da177e4SLinus Torvalds addr &= PAGE_MASK; 18791da177e4SLinus Torvalds vma = find_vma(mm,addr); 18801da177e4SLinus Torvalds if (!vma) 18811da177e4SLinus Torvalds return NULL; 18821da177e4SLinus Torvalds if (vma->vm_start <= addr) 18831da177e4SLinus Torvalds return vma; 18841da177e4SLinus Torvalds if (!(vma->vm_flags & VM_GROWSDOWN)) 18851da177e4SLinus Torvalds return NULL; 18861da177e4SLinus Torvalds start = vma->vm_start; 18871da177e4SLinus Torvalds if (expand_stack(vma, addr)) 18881da177e4SLinus Torvalds return NULL; 1889ba470de4SRik van Riel if (vma->vm_flags & VM_LOCKED) { 1890c58267c3SKOSAKI Motohiro mlock_vma_pages_range(vma, addr, start); 1891ba470de4SRik van Riel } 18921da177e4SLinus Torvalds return vma; 18931da177e4SLinus Torvalds } 18941da177e4SLinus Torvalds #endif 18951da177e4SLinus Torvalds 18962c0b3814SHugh Dickins /* 18972c0b3814SHugh Dickins * Ok - we have the memory areas we should free on the vma list, 18982c0b3814SHugh Dickins * so release them, and do the vma updates. 18991da177e4SLinus Torvalds * 19002c0b3814SHugh Dickins * Called with the mm semaphore held. 19011da177e4SLinus Torvalds */ 19022c0b3814SHugh Dickins static void remove_vma_list(struct mm_struct *mm, struct vm_area_struct *vma) 19031da177e4SLinus Torvalds { 1904365e9c87SHugh Dickins /* Update high watermark before we lower total_vm */ 1905365e9c87SHugh Dickins update_hiwater_vm(mm); 19062c0b3814SHugh Dickins do { 1907ab50b8edSHugh Dickins long nrpages = vma_pages(vma); 19081da177e4SLinus Torvalds 1909ab50b8edSHugh Dickins mm->total_vm -= nrpages; 1910ab50b8edSHugh Dickins vm_stat_account(mm, vma->vm_flags, vma->vm_file, -nrpages); 1911a8fb5618SHugh Dickins vma = remove_vma(vma); 1912146425a3SHugh Dickins } while (vma); 19131da177e4SLinus Torvalds validate_mm(mm); 19141da177e4SLinus Torvalds } 19151da177e4SLinus Torvalds 19161da177e4SLinus Torvalds /* 19171da177e4SLinus Torvalds * Get rid of page table information in the indicated region. 19181da177e4SLinus Torvalds * 1919f10df686SPaolo 'Blaisorblade' Giarrusso * Called with the mm semaphore held. 19201da177e4SLinus Torvalds */ 19211da177e4SLinus Torvalds static void unmap_region(struct mm_struct *mm, 1922e0da382cSHugh Dickins struct vm_area_struct *vma, struct vm_area_struct *prev, 1923e0da382cSHugh Dickins unsigned long start, unsigned long end) 19241da177e4SLinus Torvalds { 1925e0da382cSHugh Dickins struct vm_area_struct *next = prev? prev->vm_next: mm->mmap; 19261da177e4SLinus Torvalds struct mmu_gather *tlb; 19271da177e4SLinus Torvalds unsigned long nr_accounted = 0; 19281da177e4SLinus Torvalds 19291da177e4SLinus Torvalds lru_add_drain(); 19301da177e4SLinus Torvalds tlb = tlb_gather_mmu(mm, 0); 1931365e9c87SHugh Dickins update_hiwater_rss(mm); 1932508034a3SHugh Dickins unmap_vmas(&tlb, vma, start, end, &nr_accounted, NULL); 19331da177e4SLinus Torvalds vm_unacct_memory(nr_accounted); 193442b77728SJan Beulich free_pgtables(tlb, vma, prev? prev->vm_end: FIRST_USER_ADDRESS, 1935e0da382cSHugh Dickins next? next->vm_start: 0); 19361da177e4SLinus Torvalds tlb_finish_mmu(tlb, start, end); 19371da177e4SLinus Torvalds } 19381da177e4SLinus Torvalds 19391da177e4SLinus Torvalds /* 19401da177e4SLinus Torvalds * Create a list of vma's touched by the unmap, removing them from the mm's 19411da177e4SLinus Torvalds * vma list as we go.. 19421da177e4SLinus Torvalds */ 19431da177e4SLinus Torvalds static void 19441da177e4SLinus Torvalds detach_vmas_to_be_unmapped(struct mm_struct *mm, struct vm_area_struct *vma, 19451da177e4SLinus Torvalds struct vm_area_struct *prev, unsigned long end) 19461da177e4SLinus Torvalds { 19471da177e4SLinus Torvalds struct vm_area_struct **insertion_point; 19481da177e4SLinus Torvalds struct vm_area_struct *tail_vma = NULL; 19491363c3cdSWolfgang Wander unsigned long addr; 19501da177e4SLinus Torvalds 19511da177e4SLinus Torvalds insertion_point = (prev ? &prev->vm_next : &mm->mmap); 1952297c5eeeSLinus Torvalds vma->vm_prev = NULL; 19531da177e4SLinus Torvalds do { 19541da177e4SLinus Torvalds rb_erase(&vma->vm_rb, &mm->mm_rb); 19551da177e4SLinus Torvalds mm->map_count--; 19561da177e4SLinus Torvalds tail_vma = vma; 19571da177e4SLinus Torvalds vma = vma->vm_next; 19581da177e4SLinus Torvalds } while (vma && vma->vm_start < end); 19591da177e4SLinus Torvalds *insertion_point = vma; 1960297c5eeeSLinus Torvalds if (vma) 1961297c5eeeSLinus Torvalds vma->vm_prev = prev; 19621da177e4SLinus Torvalds tail_vma->vm_next = NULL; 19631363c3cdSWolfgang Wander if (mm->unmap_area == arch_unmap_area) 19641363c3cdSWolfgang Wander addr = prev ? prev->vm_end : mm->mmap_base; 19651363c3cdSWolfgang Wander else 19661363c3cdSWolfgang Wander addr = vma ? vma->vm_start : mm->mmap_base; 19671363c3cdSWolfgang Wander mm->unmap_area(mm, addr); 19681da177e4SLinus Torvalds mm->mmap_cache = NULL; /* Kill the cache. */ 19691da177e4SLinus Torvalds } 19701da177e4SLinus Torvalds 19711da177e4SLinus Torvalds /* 1972659ace58SKOSAKI Motohiro * __split_vma() bypasses sysctl_max_map_count checking. We use this on the 1973659ace58SKOSAKI Motohiro * munmap path where it doesn't make sense to fail. 19741da177e4SLinus Torvalds */ 1975659ace58SKOSAKI Motohiro static int __split_vma(struct mm_struct * mm, struct vm_area_struct * vma, 19761da177e4SLinus Torvalds unsigned long addr, int new_below) 19771da177e4SLinus Torvalds { 19781da177e4SLinus Torvalds struct mempolicy *pol; 19791da177e4SLinus Torvalds struct vm_area_struct *new; 19805beb4930SRik van Riel int err = -ENOMEM; 19811da177e4SLinus Torvalds 1982a5516438SAndi Kleen if (is_vm_hugetlb_page(vma) && (addr & 1983a5516438SAndi Kleen ~(huge_page_mask(hstate_vma(vma))))) 19841da177e4SLinus Torvalds return -EINVAL; 19851da177e4SLinus Torvalds 1986e94b1766SChristoph Lameter new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL); 19871da177e4SLinus Torvalds if (!new) 19885beb4930SRik van Riel goto out_err; 19891da177e4SLinus Torvalds 19901da177e4SLinus Torvalds /* most fields are the same, copy all, and then fixup */ 19911da177e4SLinus Torvalds *new = *vma; 19921da177e4SLinus Torvalds 19935beb4930SRik van Riel INIT_LIST_HEAD(&new->anon_vma_chain); 19945beb4930SRik van Riel 19951da177e4SLinus Torvalds if (new_below) 19961da177e4SLinus Torvalds new->vm_end = addr; 19971da177e4SLinus Torvalds else { 19981da177e4SLinus Torvalds new->vm_start = addr; 19991da177e4SLinus Torvalds new->vm_pgoff += ((addr - vma->vm_start) >> PAGE_SHIFT); 20001da177e4SLinus Torvalds } 20011da177e4SLinus Torvalds 2002846a16bfSLee Schermerhorn pol = mpol_dup(vma_policy(vma)); 20031da177e4SLinus Torvalds if (IS_ERR(pol)) { 20045beb4930SRik van Riel err = PTR_ERR(pol); 20055beb4930SRik van Riel goto out_free_vma; 20061da177e4SLinus Torvalds } 20071da177e4SLinus Torvalds vma_set_policy(new, pol); 20081da177e4SLinus Torvalds 20095beb4930SRik van Riel if (anon_vma_clone(new, vma)) 20105beb4930SRik van Riel goto out_free_mpol; 20115beb4930SRik van Riel 2012925d1c40SMatt Helsley if (new->vm_file) { 20131da177e4SLinus Torvalds get_file(new->vm_file); 2014925d1c40SMatt Helsley if (vma->vm_flags & VM_EXECUTABLE) 2015925d1c40SMatt Helsley added_exe_file_vma(mm); 2016925d1c40SMatt Helsley } 20171da177e4SLinus Torvalds 20181da177e4SLinus Torvalds if (new->vm_ops && new->vm_ops->open) 20191da177e4SLinus Torvalds new->vm_ops->open(new); 20201da177e4SLinus Torvalds 20211da177e4SLinus Torvalds if (new_below) 20225beb4930SRik van Riel err = vma_adjust(vma, addr, vma->vm_end, vma->vm_pgoff + 20231da177e4SLinus Torvalds ((addr - new->vm_start) >> PAGE_SHIFT), new); 20241da177e4SLinus Torvalds else 20255beb4930SRik van Riel err = vma_adjust(vma, vma->vm_start, addr, vma->vm_pgoff, new); 20261da177e4SLinus Torvalds 20275beb4930SRik van Riel /* Success. */ 20285beb4930SRik van Riel if (!err) 20291da177e4SLinus Torvalds return 0; 20305beb4930SRik van Riel 20315beb4930SRik van Riel /* Clean everything up if vma_adjust failed. */ 203258927533SRik van Riel if (new->vm_ops && new->vm_ops->close) 20335beb4930SRik van Riel new->vm_ops->close(new); 20345beb4930SRik van Riel if (new->vm_file) { 20355beb4930SRik van Riel if (vma->vm_flags & VM_EXECUTABLE) 20365beb4930SRik van Riel removed_exe_file_vma(mm); 20375beb4930SRik van Riel fput(new->vm_file); 20385beb4930SRik van Riel } 20392aeadc30SAndrea Arcangeli unlink_anon_vmas(new); 20405beb4930SRik van Riel out_free_mpol: 20415beb4930SRik van Riel mpol_put(pol); 20425beb4930SRik van Riel out_free_vma: 20435beb4930SRik van Riel kmem_cache_free(vm_area_cachep, new); 20445beb4930SRik van Riel out_err: 20455beb4930SRik van Riel return err; 20461da177e4SLinus Torvalds } 20471da177e4SLinus Torvalds 2048659ace58SKOSAKI Motohiro /* 2049659ace58SKOSAKI Motohiro * Split a vma into two pieces at address 'addr', a new vma is allocated 2050659ace58SKOSAKI Motohiro * either for the first part or the tail. 2051659ace58SKOSAKI Motohiro */ 2052659ace58SKOSAKI Motohiro int split_vma(struct mm_struct *mm, struct vm_area_struct *vma, 2053659ace58SKOSAKI Motohiro unsigned long addr, int new_below) 2054659ace58SKOSAKI Motohiro { 2055659ace58SKOSAKI Motohiro if (mm->map_count >= sysctl_max_map_count) 2056659ace58SKOSAKI Motohiro return -ENOMEM; 2057659ace58SKOSAKI Motohiro 2058659ace58SKOSAKI Motohiro return __split_vma(mm, vma, addr, new_below); 2059659ace58SKOSAKI Motohiro } 2060659ace58SKOSAKI Motohiro 20611da177e4SLinus Torvalds /* Munmap is split into 2 main parts -- this part which finds 20621da177e4SLinus Torvalds * what needs doing, and the areas themselves, which do the 20631da177e4SLinus Torvalds * work. This now handles partial unmappings. 20641da177e4SLinus Torvalds * Jeremy Fitzhardinge <jeremy@goop.org> 20651da177e4SLinus Torvalds */ 20661da177e4SLinus Torvalds int do_munmap(struct mm_struct *mm, unsigned long start, size_t len) 20671da177e4SLinus Torvalds { 20681da177e4SLinus Torvalds unsigned long end; 2069146425a3SHugh Dickins struct vm_area_struct *vma, *prev, *last; 20701da177e4SLinus Torvalds 20711da177e4SLinus Torvalds if ((start & ~PAGE_MASK) || start > TASK_SIZE || len > TASK_SIZE-start) 20721da177e4SLinus Torvalds return -EINVAL; 20731da177e4SLinus Torvalds 20741da177e4SLinus Torvalds if ((len = PAGE_ALIGN(len)) == 0) 20751da177e4SLinus Torvalds return -EINVAL; 20761da177e4SLinus Torvalds 20771da177e4SLinus Torvalds /* Find the first overlapping VMA */ 2078146425a3SHugh Dickins vma = find_vma_prev(mm, start, &prev); 2079146425a3SHugh Dickins if (!vma) 20801da177e4SLinus Torvalds return 0; 2081146425a3SHugh Dickins /* we have start < vma->vm_end */ 20821da177e4SLinus Torvalds 20831da177e4SLinus Torvalds /* if it doesn't overlap, we have nothing.. */ 20841da177e4SLinus Torvalds end = start + len; 2085146425a3SHugh Dickins if (vma->vm_start >= end) 20861da177e4SLinus Torvalds return 0; 20871da177e4SLinus Torvalds 20881da177e4SLinus Torvalds /* 20891da177e4SLinus Torvalds * If we need to split any vma, do it now to save pain later. 20901da177e4SLinus Torvalds * 20911da177e4SLinus Torvalds * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially 20921da177e4SLinus Torvalds * unmapped vm_area_struct will remain in use: so lower split_vma 20931da177e4SLinus Torvalds * places tmp vma above, and higher split_vma places tmp vma below. 20941da177e4SLinus Torvalds */ 2095146425a3SHugh Dickins if (start > vma->vm_start) { 2096659ace58SKOSAKI Motohiro int error; 2097659ace58SKOSAKI Motohiro 2098659ace58SKOSAKI Motohiro /* 2099659ace58SKOSAKI Motohiro * Make sure that map_count on return from munmap() will 2100659ace58SKOSAKI Motohiro * not exceed its limit; but let map_count go just above 2101659ace58SKOSAKI Motohiro * its limit temporarily, to help free resources as expected. 2102659ace58SKOSAKI Motohiro */ 2103659ace58SKOSAKI Motohiro if (end < vma->vm_end && mm->map_count >= sysctl_max_map_count) 2104659ace58SKOSAKI Motohiro return -ENOMEM; 2105659ace58SKOSAKI Motohiro 2106659ace58SKOSAKI Motohiro error = __split_vma(mm, vma, start, 0); 21071da177e4SLinus Torvalds if (error) 21081da177e4SLinus Torvalds return error; 2109146425a3SHugh Dickins prev = vma; 21101da177e4SLinus Torvalds } 21111da177e4SLinus Torvalds 21121da177e4SLinus Torvalds /* Does it split the last one? */ 21131da177e4SLinus Torvalds last = find_vma(mm, end); 21141da177e4SLinus Torvalds if (last && end > last->vm_start) { 2115659ace58SKOSAKI Motohiro int error = __split_vma(mm, last, end, 1); 21161da177e4SLinus Torvalds if (error) 21171da177e4SLinus Torvalds return error; 21181da177e4SLinus Torvalds } 2119146425a3SHugh Dickins vma = prev? prev->vm_next: mm->mmap; 21201da177e4SLinus Torvalds 21211da177e4SLinus Torvalds /* 2122ba470de4SRik van Riel * unlock any mlock()ed ranges before detaching vmas 2123ba470de4SRik van Riel */ 2124ba470de4SRik van Riel if (mm->locked_vm) { 2125ba470de4SRik van Riel struct vm_area_struct *tmp = vma; 2126ba470de4SRik van Riel while (tmp && tmp->vm_start < end) { 2127ba470de4SRik van Riel if (tmp->vm_flags & VM_LOCKED) { 2128ba470de4SRik van Riel mm->locked_vm -= vma_pages(tmp); 2129ba470de4SRik van Riel munlock_vma_pages_all(tmp); 2130ba470de4SRik van Riel } 2131ba470de4SRik van Riel tmp = tmp->vm_next; 2132ba470de4SRik van Riel } 2133ba470de4SRik van Riel } 2134ba470de4SRik van Riel 2135ba470de4SRik van Riel /* 21361da177e4SLinus Torvalds * Remove the vma's, and unmap the actual pages 21371da177e4SLinus Torvalds */ 2138146425a3SHugh Dickins detach_vmas_to_be_unmapped(mm, vma, prev, end); 2139146425a3SHugh Dickins unmap_region(mm, vma, prev, start, end); 21401da177e4SLinus Torvalds 21411da177e4SLinus Torvalds /* Fix up all other VM information */ 21422c0b3814SHugh Dickins remove_vma_list(mm, vma); 21431da177e4SLinus Torvalds 21441da177e4SLinus Torvalds return 0; 21451da177e4SLinus Torvalds } 21461da177e4SLinus Torvalds 21471da177e4SLinus Torvalds EXPORT_SYMBOL(do_munmap); 21481da177e4SLinus Torvalds 21496a6160a7SHeiko Carstens SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len) 21501da177e4SLinus Torvalds { 21511da177e4SLinus Torvalds int ret; 21521da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 21531da177e4SLinus Torvalds 21541da177e4SLinus Torvalds profile_munmap(addr); 21551da177e4SLinus Torvalds 21561da177e4SLinus Torvalds down_write(&mm->mmap_sem); 21571da177e4SLinus Torvalds ret = do_munmap(mm, addr, len); 21581da177e4SLinus Torvalds up_write(&mm->mmap_sem); 21591da177e4SLinus Torvalds return ret; 21601da177e4SLinus Torvalds } 21611da177e4SLinus Torvalds 21621da177e4SLinus Torvalds static inline void verify_mm_writelocked(struct mm_struct *mm) 21631da177e4SLinus Torvalds { 2164a241ec65SPaul E. McKenney #ifdef CONFIG_DEBUG_VM 21651da177e4SLinus Torvalds if (unlikely(down_read_trylock(&mm->mmap_sem))) { 21661da177e4SLinus Torvalds WARN_ON(1); 21671da177e4SLinus Torvalds up_read(&mm->mmap_sem); 21681da177e4SLinus Torvalds } 21691da177e4SLinus Torvalds #endif 21701da177e4SLinus Torvalds } 21711da177e4SLinus Torvalds 21721da177e4SLinus Torvalds /* 21731da177e4SLinus Torvalds * this is really a simplified "do_mmap". it only handles 21741da177e4SLinus Torvalds * anonymous maps. eventually we may be able to do some 21751da177e4SLinus Torvalds * brk-specific accounting here. 21761da177e4SLinus Torvalds */ 21771da177e4SLinus Torvalds unsigned long do_brk(unsigned long addr, unsigned long len) 21781da177e4SLinus Torvalds { 21791da177e4SLinus Torvalds struct mm_struct * mm = current->mm; 21801da177e4SLinus Torvalds struct vm_area_struct * vma, * prev; 21811da177e4SLinus Torvalds unsigned long flags; 21821da177e4SLinus Torvalds struct rb_node ** rb_link, * rb_parent; 21831da177e4SLinus Torvalds pgoff_t pgoff = addr >> PAGE_SHIFT; 21843a459756SKirill Korotaev int error; 21851da177e4SLinus Torvalds 21861da177e4SLinus Torvalds len = PAGE_ALIGN(len); 21871da177e4SLinus Torvalds if (!len) 21881da177e4SLinus Torvalds return addr; 21891da177e4SLinus Torvalds 219088c3f7a8SRichard Knutsson error = security_file_mmap(NULL, 0, 0, 0, addr, 1); 21915a211a5dSEric Paris if (error) 21925a211a5dSEric Paris return error; 21935a211a5dSEric Paris 21943a459756SKirill Korotaev flags = VM_DATA_DEFAULT_FLAGS | VM_ACCOUNT | mm->def_flags; 21953a459756SKirill Korotaev 21962c6a1016SAl Viro error = get_unmapped_area(NULL, addr, len, 0, MAP_FIXED); 21972c6a1016SAl Viro if (error & ~PAGE_MASK) 21983a459756SKirill Korotaev return error; 21993a459756SKirill Korotaev 22001da177e4SLinus Torvalds /* 22011da177e4SLinus Torvalds * mlock MCL_FUTURE? 22021da177e4SLinus Torvalds */ 22031da177e4SLinus Torvalds if (mm->def_flags & VM_LOCKED) { 22041da177e4SLinus Torvalds unsigned long locked, lock_limit; 220593ea1d0aSChris Wright locked = len >> PAGE_SHIFT; 220693ea1d0aSChris Wright locked += mm->locked_vm; 220759e99e5bSJiri Slaby lock_limit = rlimit(RLIMIT_MEMLOCK); 220893ea1d0aSChris Wright lock_limit >>= PAGE_SHIFT; 22091da177e4SLinus Torvalds if (locked > lock_limit && !capable(CAP_IPC_LOCK)) 22101da177e4SLinus Torvalds return -EAGAIN; 22111da177e4SLinus Torvalds } 22121da177e4SLinus Torvalds 22131da177e4SLinus Torvalds /* 22141da177e4SLinus Torvalds * mm->mmap_sem is required to protect against another thread 22151da177e4SLinus Torvalds * changing the mappings in case we sleep. 22161da177e4SLinus Torvalds */ 22171da177e4SLinus Torvalds verify_mm_writelocked(mm); 22181da177e4SLinus Torvalds 22191da177e4SLinus Torvalds /* 22201da177e4SLinus Torvalds * Clear old maps. this also does some error checking for us 22211da177e4SLinus Torvalds */ 22221da177e4SLinus Torvalds munmap_back: 22231da177e4SLinus Torvalds vma = find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent); 22241da177e4SLinus Torvalds if (vma && vma->vm_start < addr + len) { 22251da177e4SLinus Torvalds if (do_munmap(mm, addr, len)) 22261da177e4SLinus Torvalds return -ENOMEM; 22271da177e4SLinus Torvalds goto munmap_back; 22281da177e4SLinus Torvalds } 22291da177e4SLinus Torvalds 22301da177e4SLinus Torvalds /* Check against address space limits *after* clearing old maps... */ 2231119f657cSakpm@osdl.org if (!may_expand_vm(mm, len >> PAGE_SHIFT)) 22321da177e4SLinus Torvalds return -ENOMEM; 22331da177e4SLinus Torvalds 22341da177e4SLinus Torvalds if (mm->map_count > sysctl_max_map_count) 22351da177e4SLinus Torvalds return -ENOMEM; 22361da177e4SLinus Torvalds 22371da177e4SLinus Torvalds if (security_vm_enough_memory(len >> PAGE_SHIFT)) 22381da177e4SLinus Torvalds return -ENOMEM; 22391da177e4SLinus Torvalds 22401da177e4SLinus Torvalds /* Can we just expand an old private anonymous mapping? */ 2241ba470de4SRik van Riel vma = vma_merge(mm, prev, addr, addr + len, flags, 2242ba470de4SRik van Riel NULL, NULL, pgoff, NULL); 2243ba470de4SRik van Riel if (vma) 22441da177e4SLinus Torvalds goto out; 22451da177e4SLinus Torvalds 22461da177e4SLinus Torvalds /* 22471da177e4SLinus Torvalds * create a vma struct for an anonymous mapping 22481da177e4SLinus Torvalds */ 2249c5e3b83eSPekka Enberg vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL); 22501da177e4SLinus Torvalds if (!vma) { 22511da177e4SLinus Torvalds vm_unacct_memory(len >> PAGE_SHIFT); 22521da177e4SLinus Torvalds return -ENOMEM; 22531da177e4SLinus Torvalds } 22541da177e4SLinus Torvalds 22555beb4930SRik van Riel INIT_LIST_HEAD(&vma->anon_vma_chain); 22561da177e4SLinus Torvalds vma->vm_mm = mm; 22571da177e4SLinus Torvalds vma->vm_start = addr; 22581da177e4SLinus Torvalds vma->vm_end = addr + len; 22591da177e4SLinus Torvalds vma->vm_pgoff = pgoff; 22601da177e4SLinus Torvalds vma->vm_flags = flags; 22613ed75eb8SColy Li vma->vm_page_prot = vm_get_page_prot(flags); 22621da177e4SLinus Torvalds vma_link(mm, vma, prev, rb_link, rb_parent); 22631da177e4SLinus Torvalds out: 22643af9e859SEric B Munson perf_event_mmap(vma); 22651da177e4SLinus Torvalds mm->total_vm += len >> PAGE_SHIFT; 22661da177e4SLinus Torvalds if (flags & VM_LOCKED) { 2267ba470de4SRik van Riel if (!mlock_vma_pages_range(vma, addr, addr + len)) 2268ba470de4SRik van Riel mm->locked_vm += (len >> PAGE_SHIFT); 22691da177e4SLinus Torvalds } 22701da177e4SLinus Torvalds return addr; 22711da177e4SLinus Torvalds } 22721da177e4SLinus Torvalds 22731da177e4SLinus Torvalds EXPORT_SYMBOL(do_brk); 22741da177e4SLinus Torvalds 22751da177e4SLinus Torvalds /* Release all mmaps. */ 22761da177e4SLinus Torvalds void exit_mmap(struct mm_struct *mm) 22771da177e4SLinus Torvalds { 22781da177e4SLinus Torvalds struct mmu_gather *tlb; 2279ba470de4SRik van Riel struct vm_area_struct *vma; 22801da177e4SLinus Torvalds unsigned long nr_accounted = 0; 2281ee39b37bSHugh Dickins unsigned long end; 22821da177e4SLinus Torvalds 2283d6dd61c8SJeremy Fitzhardinge /* mm's last user has gone, and its about to be pulled down */ 2284cddb8a5cSAndrea Arcangeli mmu_notifier_release(mm); 2285d6dd61c8SJeremy Fitzhardinge 2286ba470de4SRik van Riel if (mm->locked_vm) { 2287ba470de4SRik van Riel vma = mm->mmap; 2288ba470de4SRik van Riel while (vma) { 2289ba470de4SRik van Riel if (vma->vm_flags & VM_LOCKED) 2290ba470de4SRik van Riel munlock_vma_pages_all(vma); 2291ba470de4SRik van Riel vma = vma->vm_next; 2292ba470de4SRik van Riel } 2293ba470de4SRik van Riel } 22949480c53eSJeremy Fitzhardinge 22959480c53eSJeremy Fitzhardinge arch_exit_mmap(mm); 22969480c53eSJeremy Fitzhardinge 2297ba470de4SRik van Riel vma = mm->mmap; 22989480c53eSJeremy Fitzhardinge if (!vma) /* Can happen if dup_mmap() received an OOM */ 22999480c53eSJeremy Fitzhardinge return; 23009480c53eSJeremy Fitzhardinge 23011da177e4SLinus Torvalds lru_add_drain(); 23021da177e4SLinus Torvalds flush_cache_mm(mm); 2303e0da382cSHugh Dickins tlb = tlb_gather_mmu(mm, 1); 2304901608d9SOleg Nesterov /* update_hiwater_rss(mm) here? but nobody should be looking */ 2305e0da382cSHugh Dickins /* Use -1 here to ensure all VMAs in the mm are unmapped */ 2306508034a3SHugh Dickins end = unmap_vmas(&tlb, vma, 0, -1, &nr_accounted, NULL); 23071da177e4SLinus Torvalds vm_unacct_memory(nr_accounted); 23089ba69294SHugh Dickins 230942b77728SJan Beulich free_pgtables(tlb, vma, FIRST_USER_ADDRESS, 0); 2310ee39b37bSHugh Dickins tlb_finish_mmu(tlb, 0, end); 23111da177e4SLinus Torvalds 23121da177e4SLinus Torvalds /* 23138f4f8c16SHugh Dickins * Walk the list again, actually closing and freeing it, 23148f4f8c16SHugh Dickins * with preemption enabled, without holding any MM locks. 23151da177e4SLinus Torvalds */ 2316a8fb5618SHugh Dickins while (vma) 2317a8fb5618SHugh Dickins vma = remove_vma(vma); 2318e0da382cSHugh Dickins 2319e2cdef8cSHugh Dickins BUG_ON(mm->nr_ptes > (FIRST_USER_ADDRESS+PMD_SIZE-1)>>PMD_SHIFT); 23201da177e4SLinus Torvalds } 23211da177e4SLinus Torvalds 23221da177e4SLinus Torvalds /* Insert vm structure into process list sorted by address 23231da177e4SLinus Torvalds * and into the inode's i_mmap tree. If vm_file is non-NULL 23241da177e4SLinus Torvalds * then i_mmap_lock is taken here. 23251da177e4SLinus Torvalds */ 23261da177e4SLinus Torvalds int insert_vm_struct(struct mm_struct * mm, struct vm_area_struct * vma) 23271da177e4SLinus Torvalds { 23281da177e4SLinus Torvalds struct vm_area_struct * __vma, * prev; 23291da177e4SLinus Torvalds struct rb_node ** rb_link, * rb_parent; 23301da177e4SLinus Torvalds 23311da177e4SLinus Torvalds /* 23321da177e4SLinus Torvalds * The vm_pgoff of a purely anonymous vma should be irrelevant 23331da177e4SLinus Torvalds * until its first write fault, when page's anon_vma and index 23341da177e4SLinus Torvalds * are set. But now set the vm_pgoff it will almost certainly 23351da177e4SLinus Torvalds * end up with (unless mremap moves it elsewhere before that 23361da177e4SLinus Torvalds * first wfault), so /proc/pid/maps tells a consistent story. 23371da177e4SLinus Torvalds * 23381da177e4SLinus Torvalds * By setting it to reflect the virtual start address of the 23391da177e4SLinus Torvalds * vma, merges and splits can happen in a seamless way, just 23401da177e4SLinus Torvalds * using the existing file pgoff checks and manipulations. 23411da177e4SLinus Torvalds * Similarly in do_mmap_pgoff and in do_brk. 23421da177e4SLinus Torvalds */ 23431da177e4SLinus Torvalds if (!vma->vm_file) { 23441da177e4SLinus Torvalds BUG_ON(vma->anon_vma); 23451da177e4SLinus Torvalds vma->vm_pgoff = vma->vm_start >> PAGE_SHIFT; 23461da177e4SLinus Torvalds } 23471da177e4SLinus Torvalds __vma = find_vma_prepare(mm,vma->vm_start,&prev,&rb_link,&rb_parent); 23481da177e4SLinus Torvalds if (__vma && __vma->vm_start < vma->vm_end) 23491da177e4SLinus Torvalds return -ENOMEM; 23502fd4ef85SHugh Dickins if ((vma->vm_flags & VM_ACCOUNT) && 235134b4e4aaSAlan Cox security_vm_enough_memory_mm(mm, vma_pages(vma))) 23522fd4ef85SHugh Dickins return -ENOMEM; 23531da177e4SLinus Torvalds vma_link(mm, vma, prev, rb_link, rb_parent); 23541da177e4SLinus Torvalds return 0; 23551da177e4SLinus Torvalds } 23561da177e4SLinus Torvalds 23571da177e4SLinus Torvalds /* 23581da177e4SLinus Torvalds * Copy the vma structure to a new location in the same mm, 23591da177e4SLinus Torvalds * prior to moving page table entries, to effect an mremap move. 23601da177e4SLinus Torvalds */ 23611da177e4SLinus Torvalds struct vm_area_struct *copy_vma(struct vm_area_struct **vmap, 23621da177e4SLinus Torvalds unsigned long addr, unsigned long len, pgoff_t pgoff) 23631da177e4SLinus Torvalds { 23641da177e4SLinus Torvalds struct vm_area_struct *vma = *vmap; 23651da177e4SLinus Torvalds unsigned long vma_start = vma->vm_start; 23661da177e4SLinus Torvalds struct mm_struct *mm = vma->vm_mm; 23671da177e4SLinus Torvalds struct vm_area_struct *new_vma, *prev; 23681da177e4SLinus Torvalds struct rb_node **rb_link, *rb_parent; 23691da177e4SLinus Torvalds struct mempolicy *pol; 23701da177e4SLinus Torvalds 23711da177e4SLinus Torvalds /* 23721da177e4SLinus Torvalds * If anonymous vma has not yet been faulted, update new pgoff 23731da177e4SLinus Torvalds * to match new location, to increase its chance of merging. 23741da177e4SLinus Torvalds */ 23751da177e4SLinus Torvalds if (!vma->vm_file && !vma->anon_vma) 23761da177e4SLinus Torvalds pgoff = addr >> PAGE_SHIFT; 23771da177e4SLinus Torvalds 23781da177e4SLinus Torvalds find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent); 23791da177e4SLinus Torvalds new_vma = vma_merge(mm, prev, addr, addr + len, vma->vm_flags, 23801da177e4SLinus Torvalds vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma)); 23811da177e4SLinus Torvalds if (new_vma) { 23821da177e4SLinus Torvalds /* 23831da177e4SLinus Torvalds * Source vma may have been merged into new_vma 23841da177e4SLinus Torvalds */ 23851da177e4SLinus Torvalds if (vma_start >= new_vma->vm_start && 23861da177e4SLinus Torvalds vma_start < new_vma->vm_end) 23871da177e4SLinus Torvalds *vmap = new_vma; 23881da177e4SLinus Torvalds } else { 2389e94b1766SChristoph Lameter new_vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL); 23901da177e4SLinus Torvalds if (new_vma) { 23911da177e4SLinus Torvalds *new_vma = *vma; 2392846a16bfSLee Schermerhorn pol = mpol_dup(vma_policy(vma)); 23935beb4930SRik van Riel if (IS_ERR(pol)) 23945beb4930SRik van Riel goto out_free_vma; 23955beb4930SRik van Riel INIT_LIST_HEAD(&new_vma->anon_vma_chain); 23965beb4930SRik van Riel if (anon_vma_clone(new_vma, vma)) 23975beb4930SRik van Riel goto out_free_mempol; 23981da177e4SLinus Torvalds vma_set_policy(new_vma, pol); 23991da177e4SLinus Torvalds new_vma->vm_start = addr; 24001da177e4SLinus Torvalds new_vma->vm_end = addr + len; 24011da177e4SLinus Torvalds new_vma->vm_pgoff = pgoff; 2402925d1c40SMatt Helsley if (new_vma->vm_file) { 24031da177e4SLinus Torvalds get_file(new_vma->vm_file); 2404925d1c40SMatt Helsley if (vma->vm_flags & VM_EXECUTABLE) 2405925d1c40SMatt Helsley added_exe_file_vma(mm); 2406925d1c40SMatt Helsley } 24071da177e4SLinus Torvalds if (new_vma->vm_ops && new_vma->vm_ops->open) 24081da177e4SLinus Torvalds new_vma->vm_ops->open(new_vma); 24091da177e4SLinus Torvalds vma_link(mm, new_vma, prev, rb_link, rb_parent); 24101da177e4SLinus Torvalds } 24111da177e4SLinus Torvalds } 24121da177e4SLinus Torvalds return new_vma; 24135beb4930SRik van Riel 24145beb4930SRik van Riel out_free_mempol: 24155beb4930SRik van Riel mpol_put(pol); 24165beb4930SRik van Riel out_free_vma: 24175beb4930SRik van Riel kmem_cache_free(vm_area_cachep, new_vma); 24185beb4930SRik van Riel return NULL; 24191da177e4SLinus Torvalds } 2420119f657cSakpm@osdl.org 2421119f657cSakpm@osdl.org /* 2422119f657cSakpm@osdl.org * Return true if the calling process may expand its vm space by the passed 2423119f657cSakpm@osdl.org * number of pages 2424119f657cSakpm@osdl.org */ 2425119f657cSakpm@osdl.org int may_expand_vm(struct mm_struct *mm, unsigned long npages) 2426119f657cSakpm@osdl.org { 2427119f657cSakpm@osdl.org unsigned long cur = mm->total_vm; /* pages */ 2428119f657cSakpm@osdl.org unsigned long lim; 2429119f657cSakpm@osdl.org 243059e99e5bSJiri Slaby lim = rlimit(RLIMIT_AS) >> PAGE_SHIFT; 2431119f657cSakpm@osdl.org 2432119f657cSakpm@osdl.org if (cur + npages > lim) 2433119f657cSakpm@osdl.org return 0; 2434119f657cSakpm@osdl.org return 1; 2435119f657cSakpm@osdl.org } 2436fa5dc22fSRoland McGrath 2437fa5dc22fSRoland McGrath 2438b1d0e4f5SNick Piggin static int special_mapping_fault(struct vm_area_struct *vma, 2439b1d0e4f5SNick Piggin struct vm_fault *vmf) 2440fa5dc22fSRoland McGrath { 2441b1d0e4f5SNick Piggin pgoff_t pgoff; 2442fa5dc22fSRoland McGrath struct page **pages; 2443fa5dc22fSRoland McGrath 2444b1d0e4f5SNick Piggin /* 2445b1d0e4f5SNick Piggin * special mappings have no vm_file, and in that case, the mm 2446b1d0e4f5SNick Piggin * uses vm_pgoff internally. So we have to subtract it from here. 2447b1d0e4f5SNick Piggin * We are allowed to do this because we are the mm; do not copy 2448b1d0e4f5SNick Piggin * this code into drivers! 2449b1d0e4f5SNick Piggin */ 2450b1d0e4f5SNick Piggin pgoff = vmf->pgoff - vma->vm_pgoff; 2451fa5dc22fSRoland McGrath 2452b1d0e4f5SNick Piggin for (pages = vma->vm_private_data; pgoff && *pages; ++pages) 2453b1d0e4f5SNick Piggin pgoff--; 2454fa5dc22fSRoland McGrath 2455fa5dc22fSRoland McGrath if (*pages) { 2456fa5dc22fSRoland McGrath struct page *page = *pages; 2457fa5dc22fSRoland McGrath get_page(page); 2458b1d0e4f5SNick Piggin vmf->page = page; 2459b1d0e4f5SNick Piggin return 0; 2460fa5dc22fSRoland McGrath } 2461fa5dc22fSRoland McGrath 2462b1d0e4f5SNick Piggin return VM_FAULT_SIGBUS; 2463fa5dc22fSRoland McGrath } 2464fa5dc22fSRoland McGrath 2465fa5dc22fSRoland McGrath /* 2466fa5dc22fSRoland McGrath * Having a close hook prevents vma merging regardless of flags. 2467fa5dc22fSRoland McGrath */ 2468fa5dc22fSRoland McGrath static void special_mapping_close(struct vm_area_struct *vma) 2469fa5dc22fSRoland McGrath { 2470fa5dc22fSRoland McGrath } 2471fa5dc22fSRoland McGrath 2472f0f37e2fSAlexey Dobriyan static const struct vm_operations_struct special_mapping_vmops = { 2473fa5dc22fSRoland McGrath .close = special_mapping_close, 2474b1d0e4f5SNick Piggin .fault = special_mapping_fault, 2475fa5dc22fSRoland McGrath }; 2476fa5dc22fSRoland McGrath 2477fa5dc22fSRoland McGrath /* 2478fa5dc22fSRoland McGrath * Called with mm->mmap_sem held for writing. 2479fa5dc22fSRoland McGrath * Insert a new vma covering the given region, with the given flags. 2480fa5dc22fSRoland McGrath * Its pages are supplied by the given array of struct page *. 2481fa5dc22fSRoland McGrath * The array can be shorter than len >> PAGE_SHIFT if it's null-terminated. 2482fa5dc22fSRoland McGrath * The region past the last page supplied will always produce SIGBUS. 2483fa5dc22fSRoland McGrath * The array pointer and the pages it points to are assumed to stay alive 2484fa5dc22fSRoland McGrath * for as long as this mapping might exist. 2485fa5dc22fSRoland McGrath */ 2486fa5dc22fSRoland McGrath int install_special_mapping(struct mm_struct *mm, 2487fa5dc22fSRoland McGrath unsigned long addr, unsigned long len, 2488fa5dc22fSRoland McGrath unsigned long vm_flags, struct page **pages) 2489fa5dc22fSRoland McGrath { 2490462e635eSTavis Ormandy int ret; 2491fa5dc22fSRoland McGrath struct vm_area_struct *vma; 2492fa5dc22fSRoland McGrath 2493fa5dc22fSRoland McGrath vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL); 2494fa5dc22fSRoland McGrath if (unlikely(vma == NULL)) 2495fa5dc22fSRoland McGrath return -ENOMEM; 2496fa5dc22fSRoland McGrath 24975beb4930SRik van Riel INIT_LIST_HEAD(&vma->anon_vma_chain); 2498fa5dc22fSRoland McGrath vma->vm_mm = mm; 2499fa5dc22fSRoland McGrath vma->vm_start = addr; 2500fa5dc22fSRoland McGrath vma->vm_end = addr + len; 2501fa5dc22fSRoland McGrath 25022f98735cSNick Piggin vma->vm_flags = vm_flags | mm->def_flags | VM_DONTEXPAND; 25033ed75eb8SColy Li vma->vm_page_prot = vm_get_page_prot(vma->vm_flags); 2504fa5dc22fSRoland McGrath 2505fa5dc22fSRoland McGrath vma->vm_ops = &special_mapping_vmops; 2506fa5dc22fSRoland McGrath vma->vm_private_data = pages; 2507fa5dc22fSRoland McGrath 2508462e635eSTavis Ormandy ret = security_file_mmap(NULL, 0, 0, 0, vma->vm_start, 1); 2509462e635eSTavis Ormandy if (ret) 2510462e635eSTavis Ormandy goto out; 2511462e635eSTavis Ormandy 2512462e635eSTavis Ormandy ret = insert_vm_struct(mm, vma); 2513462e635eSTavis Ormandy if (ret) 2514462e635eSTavis Ormandy goto out; 2515fa5dc22fSRoland McGrath 2516fa5dc22fSRoland McGrath mm->total_vm += len >> PAGE_SHIFT; 2517fa5dc22fSRoland McGrath 2518cdd6c482SIngo Molnar perf_event_mmap(vma); 2519089dd79dSPeter Zijlstra 2520fa5dc22fSRoland McGrath return 0; 2521462e635eSTavis Ormandy 2522462e635eSTavis Ormandy out: 2523462e635eSTavis Ormandy kmem_cache_free(vm_area_cachep, vma); 2524462e635eSTavis Ormandy return ret; 2525fa5dc22fSRoland McGrath } 25267906d00cSAndrea Arcangeli 25277906d00cSAndrea Arcangeli static DEFINE_MUTEX(mm_all_locks_mutex); 25287906d00cSAndrea Arcangeli 2529454ed842SPeter Zijlstra static void vm_lock_anon_vma(struct mm_struct *mm, struct anon_vma *anon_vma) 25307906d00cSAndrea Arcangeli { 2531012f1800SRik van Riel if (!test_bit(0, (unsigned long *) &anon_vma->root->head.next)) { 25327906d00cSAndrea Arcangeli /* 25337906d00cSAndrea Arcangeli * The LSB of head.next can't change from under us 25347906d00cSAndrea Arcangeli * because we hold the mm_all_locks_mutex. 25357906d00cSAndrea Arcangeli */ 2536012f1800SRik van Riel spin_lock_nest_lock(&anon_vma->root->lock, &mm->mmap_sem); 25377906d00cSAndrea Arcangeli /* 25387906d00cSAndrea Arcangeli * We can safely modify head.next after taking the 2539012f1800SRik van Riel * anon_vma->root->lock. If some other vma in this mm shares 25407906d00cSAndrea Arcangeli * the same anon_vma we won't take it again. 25417906d00cSAndrea Arcangeli * 25427906d00cSAndrea Arcangeli * No need of atomic instructions here, head.next 25437906d00cSAndrea Arcangeli * can't change from under us thanks to the 2544012f1800SRik van Riel * anon_vma->root->lock. 25457906d00cSAndrea Arcangeli */ 25467906d00cSAndrea Arcangeli if (__test_and_set_bit(0, (unsigned long *) 2547012f1800SRik van Riel &anon_vma->root->head.next)) 25487906d00cSAndrea Arcangeli BUG(); 25497906d00cSAndrea Arcangeli } 25507906d00cSAndrea Arcangeli } 25517906d00cSAndrea Arcangeli 2552454ed842SPeter Zijlstra static void vm_lock_mapping(struct mm_struct *mm, struct address_space *mapping) 25537906d00cSAndrea Arcangeli { 25547906d00cSAndrea Arcangeli if (!test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) { 25557906d00cSAndrea Arcangeli /* 25567906d00cSAndrea Arcangeli * AS_MM_ALL_LOCKS can't change from under us because 25577906d00cSAndrea Arcangeli * we hold the mm_all_locks_mutex. 25587906d00cSAndrea Arcangeli * 25597906d00cSAndrea Arcangeli * Operations on ->flags have to be atomic because 25607906d00cSAndrea Arcangeli * even if AS_MM_ALL_LOCKS is stable thanks to the 25617906d00cSAndrea Arcangeli * mm_all_locks_mutex, there may be other cpus 25627906d00cSAndrea Arcangeli * changing other bitflags in parallel to us. 25637906d00cSAndrea Arcangeli */ 25647906d00cSAndrea Arcangeli if (test_and_set_bit(AS_MM_ALL_LOCKS, &mapping->flags)) 25657906d00cSAndrea Arcangeli BUG(); 2566454ed842SPeter Zijlstra spin_lock_nest_lock(&mapping->i_mmap_lock, &mm->mmap_sem); 25677906d00cSAndrea Arcangeli } 25687906d00cSAndrea Arcangeli } 25697906d00cSAndrea Arcangeli 25707906d00cSAndrea Arcangeli /* 25717906d00cSAndrea Arcangeli * This operation locks against the VM for all pte/vma/mm related 25727906d00cSAndrea Arcangeli * operations that could ever happen on a certain mm. This includes 25737906d00cSAndrea Arcangeli * vmtruncate, try_to_unmap, and all page faults. 25747906d00cSAndrea Arcangeli * 25757906d00cSAndrea Arcangeli * The caller must take the mmap_sem in write mode before calling 25767906d00cSAndrea Arcangeli * mm_take_all_locks(). The caller isn't allowed to release the 25777906d00cSAndrea Arcangeli * mmap_sem until mm_drop_all_locks() returns. 25787906d00cSAndrea Arcangeli * 25797906d00cSAndrea Arcangeli * mmap_sem in write mode is required in order to block all operations 25807906d00cSAndrea Arcangeli * that could modify pagetables and free pages without need of 25817906d00cSAndrea Arcangeli * altering the vma layout (for example populate_range() with 25827906d00cSAndrea Arcangeli * nonlinear vmas). It's also needed in write mode to avoid new 25837906d00cSAndrea Arcangeli * anon_vmas to be associated with existing vmas. 25847906d00cSAndrea Arcangeli * 25857906d00cSAndrea Arcangeli * A single task can't take more than one mm_take_all_locks() in a row 25867906d00cSAndrea Arcangeli * or it would deadlock. 25877906d00cSAndrea Arcangeli * 25887906d00cSAndrea Arcangeli * The LSB in anon_vma->head.next and the AS_MM_ALL_LOCKS bitflag in 25897906d00cSAndrea Arcangeli * mapping->flags avoid to take the same lock twice, if more than one 25907906d00cSAndrea Arcangeli * vma in this mm is backed by the same anon_vma or address_space. 25917906d00cSAndrea Arcangeli * 25927906d00cSAndrea Arcangeli * We can take all the locks in random order because the VM code 25937906d00cSAndrea Arcangeli * taking i_mmap_lock or anon_vma->lock outside the mmap_sem never 25947906d00cSAndrea Arcangeli * takes more than one of them in a row. Secondly we're protected 25957906d00cSAndrea Arcangeli * against a concurrent mm_take_all_locks() by the mm_all_locks_mutex. 25967906d00cSAndrea Arcangeli * 25977906d00cSAndrea Arcangeli * mm_take_all_locks() and mm_drop_all_locks are expensive operations 25987906d00cSAndrea Arcangeli * that may have to take thousand of locks. 25997906d00cSAndrea Arcangeli * 26007906d00cSAndrea Arcangeli * mm_take_all_locks() can fail if it's interrupted by signals. 26017906d00cSAndrea Arcangeli */ 26027906d00cSAndrea Arcangeli int mm_take_all_locks(struct mm_struct *mm) 26037906d00cSAndrea Arcangeli { 26047906d00cSAndrea Arcangeli struct vm_area_struct *vma; 26055beb4930SRik van Riel struct anon_vma_chain *avc; 26067906d00cSAndrea Arcangeli int ret = -EINTR; 26077906d00cSAndrea Arcangeli 26087906d00cSAndrea Arcangeli BUG_ON(down_read_trylock(&mm->mmap_sem)); 26097906d00cSAndrea Arcangeli 26107906d00cSAndrea Arcangeli mutex_lock(&mm_all_locks_mutex); 26117906d00cSAndrea Arcangeli 26127906d00cSAndrea Arcangeli for (vma = mm->mmap; vma; vma = vma->vm_next) { 26137906d00cSAndrea Arcangeli if (signal_pending(current)) 26147906d00cSAndrea Arcangeli goto out_unlock; 26157906d00cSAndrea Arcangeli if (vma->vm_file && vma->vm_file->f_mapping) 2616454ed842SPeter Zijlstra vm_lock_mapping(mm, vma->vm_file->f_mapping); 26177906d00cSAndrea Arcangeli } 26187cd5a02fSPeter Zijlstra 26197cd5a02fSPeter Zijlstra for (vma = mm->mmap; vma; vma = vma->vm_next) { 26207cd5a02fSPeter Zijlstra if (signal_pending(current)) 26217cd5a02fSPeter Zijlstra goto out_unlock; 26227cd5a02fSPeter Zijlstra if (vma->anon_vma) 26235beb4930SRik van Riel list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) 26245beb4930SRik van Riel vm_lock_anon_vma(mm, avc->anon_vma); 26257cd5a02fSPeter Zijlstra } 26267cd5a02fSPeter Zijlstra 26277906d00cSAndrea Arcangeli ret = 0; 26287906d00cSAndrea Arcangeli 26297906d00cSAndrea Arcangeli out_unlock: 26307906d00cSAndrea Arcangeli if (ret) 26317906d00cSAndrea Arcangeli mm_drop_all_locks(mm); 26327906d00cSAndrea Arcangeli 26337906d00cSAndrea Arcangeli return ret; 26347906d00cSAndrea Arcangeli } 26357906d00cSAndrea Arcangeli 26367906d00cSAndrea Arcangeli static void vm_unlock_anon_vma(struct anon_vma *anon_vma) 26377906d00cSAndrea Arcangeli { 2638012f1800SRik van Riel if (test_bit(0, (unsigned long *) &anon_vma->root->head.next)) { 26397906d00cSAndrea Arcangeli /* 26407906d00cSAndrea Arcangeli * The LSB of head.next can't change to 0 from under 26417906d00cSAndrea Arcangeli * us because we hold the mm_all_locks_mutex. 26427906d00cSAndrea Arcangeli * 26437906d00cSAndrea Arcangeli * We must however clear the bitflag before unlocking 26447906d00cSAndrea Arcangeli * the vma so the users using the anon_vma->head will 26457906d00cSAndrea Arcangeli * never see our bitflag. 26467906d00cSAndrea Arcangeli * 26477906d00cSAndrea Arcangeli * No need of atomic instructions here, head.next 26487906d00cSAndrea Arcangeli * can't change from under us until we release the 2649012f1800SRik van Riel * anon_vma->root->lock. 26507906d00cSAndrea Arcangeli */ 26517906d00cSAndrea Arcangeli if (!__test_and_clear_bit(0, (unsigned long *) 2652012f1800SRik van Riel &anon_vma->root->head.next)) 26537906d00cSAndrea Arcangeli BUG(); 2654cba48b98SRik van Riel anon_vma_unlock(anon_vma); 26557906d00cSAndrea Arcangeli } 26567906d00cSAndrea Arcangeli } 26577906d00cSAndrea Arcangeli 26587906d00cSAndrea Arcangeli static void vm_unlock_mapping(struct address_space *mapping) 26597906d00cSAndrea Arcangeli { 26607906d00cSAndrea Arcangeli if (test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) { 26617906d00cSAndrea Arcangeli /* 26627906d00cSAndrea Arcangeli * AS_MM_ALL_LOCKS can't change to 0 from under us 26637906d00cSAndrea Arcangeli * because we hold the mm_all_locks_mutex. 26647906d00cSAndrea Arcangeli */ 26657906d00cSAndrea Arcangeli spin_unlock(&mapping->i_mmap_lock); 26667906d00cSAndrea Arcangeli if (!test_and_clear_bit(AS_MM_ALL_LOCKS, 26677906d00cSAndrea Arcangeli &mapping->flags)) 26687906d00cSAndrea Arcangeli BUG(); 26697906d00cSAndrea Arcangeli } 26707906d00cSAndrea Arcangeli } 26717906d00cSAndrea Arcangeli 26727906d00cSAndrea Arcangeli /* 26737906d00cSAndrea Arcangeli * The mmap_sem cannot be released by the caller until 26747906d00cSAndrea Arcangeli * mm_drop_all_locks() returns. 26757906d00cSAndrea Arcangeli */ 26767906d00cSAndrea Arcangeli void mm_drop_all_locks(struct mm_struct *mm) 26777906d00cSAndrea Arcangeli { 26787906d00cSAndrea Arcangeli struct vm_area_struct *vma; 26795beb4930SRik van Riel struct anon_vma_chain *avc; 26807906d00cSAndrea Arcangeli 26817906d00cSAndrea Arcangeli BUG_ON(down_read_trylock(&mm->mmap_sem)); 26827906d00cSAndrea Arcangeli BUG_ON(!mutex_is_locked(&mm_all_locks_mutex)); 26837906d00cSAndrea Arcangeli 26847906d00cSAndrea Arcangeli for (vma = mm->mmap; vma; vma = vma->vm_next) { 26857906d00cSAndrea Arcangeli if (vma->anon_vma) 26865beb4930SRik van Riel list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) 26875beb4930SRik van Riel vm_unlock_anon_vma(avc->anon_vma); 26887906d00cSAndrea Arcangeli if (vma->vm_file && vma->vm_file->f_mapping) 26897906d00cSAndrea Arcangeli vm_unlock_mapping(vma->vm_file->f_mapping); 26907906d00cSAndrea Arcangeli } 26917906d00cSAndrea Arcangeli 26927906d00cSAndrea Arcangeli mutex_unlock(&mm_all_locks_mutex); 26937906d00cSAndrea Arcangeli } 26948feae131SDavid Howells 26958feae131SDavid Howells /* 26968feae131SDavid Howells * initialise the VMA slab 26978feae131SDavid Howells */ 26988feae131SDavid Howells void __init mmap_init(void) 26998feae131SDavid Howells { 270000a62ce9SKOSAKI Motohiro int ret; 270100a62ce9SKOSAKI Motohiro 270200a62ce9SKOSAKI Motohiro ret = percpu_counter_init(&vm_committed_as, 0); 270300a62ce9SKOSAKI Motohiro VM_BUG_ON(ret); 27048feae131SDavid Howells } 2705