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> 236146f0d5SMimi Zohar #include <linux/ima.h> 241da177e4SLinus Torvalds #include <linux/hugetlb.h> 251da177e4SLinus Torvalds #include <linux/profile.h> 261da177e4SLinus Torvalds #include <linux/module.h> 271da177e4SLinus Torvalds #include <linux/mount.h> 281da177e4SLinus Torvalds #include <linux/mempolicy.h> 291da177e4SLinus Torvalds #include <linux/rmap.h> 30*9ba69294SHugh Dickins #include <linux/ksm.h> 31cddb8a5cSAndrea Arcangeli #include <linux/mmu_notifier.h> 32cdd6c482SIngo Molnar #include <linux/perf_event.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 871da177e4SLinus Torvalds int sysctl_overcommit_memory = OVERCOMMIT_GUESS; /* heuristic overcommit */ 881da177e4SLinus Torvalds int sysctl_overcommit_ratio = 50; /* default is 50% */ 89c3d8c141SChristoph Lameter int sysctl_max_map_count __read_mostly = DEFAULT_MAX_MAP_COUNT; 9000a62ce9SKOSAKI Motohiro struct percpu_counter vm_committed_as; 911da177e4SLinus Torvalds 921da177e4SLinus Torvalds /* 931da177e4SLinus Torvalds * Check that a process has enough memory to allocate a new virtual 941da177e4SLinus Torvalds * mapping. 0 means there is enough memory for the allocation to 951da177e4SLinus Torvalds * succeed and -ENOMEM implies there is not. 961da177e4SLinus Torvalds * 971da177e4SLinus Torvalds * We currently support three overcommit policies, which are set via the 981da177e4SLinus Torvalds * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting 991da177e4SLinus Torvalds * 1001da177e4SLinus Torvalds * Strict overcommit modes added 2002 Feb 26 by Alan Cox. 1011da177e4SLinus Torvalds * Additional code 2002 Jul 20 by Robert Love. 1021da177e4SLinus Torvalds * 1031da177e4SLinus Torvalds * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise. 1041da177e4SLinus Torvalds * 1051da177e4SLinus Torvalds * Note this is a helper function intended to be used by LSMs which 1061da177e4SLinus Torvalds * wish to use this logic. 1071da177e4SLinus Torvalds */ 10834b4e4aaSAlan Cox int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin) 1091da177e4SLinus Torvalds { 1101da177e4SLinus Torvalds unsigned long free, allowed; 1111da177e4SLinus Torvalds 1121da177e4SLinus Torvalds vm_acct_memory(pages); 1131da177e4SLinus Torvalds 1141da177e4SLinus Torvalds /* 1151da177e4SLinus Torvalds * Sometimes we want to use more memory than we have 1161da177e4SLinus Torvalds */ 1171da177e4SLinus Torvalds if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS) 1181da177e4SLinus Torvalds return 0; 1191da177e4SLinus Torvalds 1201da177e4SLinus Torvalds if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) { 1211da177e4SLinus Torvalds unsigned long n; 1221da177e4SLinus Torvalds 123347ce434SChristoph Lameter free = global_page_state(NR_FILE_PAGES); 1241da177e4SLinus Torvalds free += nr_swap_pages; 1251da177e4SLinus Torvalds 1261da177e4SLinus Torvalds /* 1271da177e4SLinus Torvalds * Any slabs which are created with the 1281da177e4SLinus Torvalds * SLAB_RECLAIM_ACCOUNT flag claim to have contents 1291da177e4SLinus Torvalds * which are reclaimable, under pressure. The dentry 1301da177e4SLinus Torvalds * cache and most inode caches should fall into this 1311da177e4SLinus Torvalds */ 132972d1a7bSChristoph Lameter free += global_page_state(NR_SLAB_RECLAIMABLE); 1331da177e4SLinus Torvalds 1341da177e4SLinus Torvalds /* 1351da177e4SLinus Torvalds * Leave the last 3% for root 1361da177e4SLinus Torvalds */ 1371da177e4SLinus Torvalds if (!cap_sys_admin) 1381da177e4SLinus Torvalds free -= free / 32; 1391da177e4SLinus Torvalds 1401da177e4SLinus Torvalds if (free > pages) 1411da177e4SLinus Torvalds return 0; 1421da177e4SLinus Torvalds 1431da177e4SLinus Torvalds /* 1441da177e4SLinus Torvalds * nr_free_pages() is very expensive on large systems, 1451da177e4SLinus Torvalds * only call if we're about to fail. 1461da177e4SLinus Torvalds */ 1471da177e4SLinus Torvalds n = nr_free_pages(); 1486d9f7839SHideo AOKI 1496d9f7839SHideo AOKI /* 1506d9f7839SHideo AOKI * Leave reserved pages. The pages are not for anonymous pages. 1516d9f7839SHideo AOKI */ 1526d9f7839SHideo AOKI if (n <= totalreserve_pages) 1536d9f7839SHideo AOKI goto error; 1546d9f7839SHideo AOKI else 1556d9f7839SHideo AOKI n -= totalreserve_pages; 1566d9f7839SHideo AOKI 1576d9f7839SHideo AOKI /* 1586d9f7839SHideo AOKI * Leave the last 3% for root 1596d9f7839SHideo AOKI */ 1601da177e4SLinus Torvalds if (!cap_sys_admin) 1611da177e4SLinus Torvalds n -= n / 32; 1621da177e4SLinus Torvalds free += n; 1631da177e4SLinus Torvalds 1641da177e4SLinus Torvalds if (free > pages) 1651da177e4SLinus Torvalds return 0; 1666d9f7839SHideo AOKI 1676d9f7839SHideo AOKI goto error; 1681da177e4SLinus Torvalds } 1691da177e4SLinus Torvalds 1701da177e4SLinus Torvalds allowed = (totalram_pages - hugetlb_total_pages()) 1711da177e4SLinus Torvalds * sysctl_overcommit_ratio / 100; 1721da177e4SLinus Torvalds /* 1731da177e4SLinus Torvalds * Leave the last 3% for root 1741da177e4SLinus Torvalds */ 1751da177e4SLinus Torvalds if (!cap_sys_admin) 1761da177e4SLinus Torvalds allowed -= allowed / 32; 1771da177e4SLinus Torvalds allowed += total_swap_pages; 1781da177e4SLinus Torvalds 1791da177e4SLinus Torvalds /* Don't let a single process grow too big: 1801da177e4SLinus Torvalds leave 3% of the size of this process for other processes */ 181731572d3SAlan Cox if (mm) 18234b4e4aaSAlan Cox allowed -= mm->total_vm / 32; 1831da177e4SLinus Torvalds 18400a62ce9SKOSAKI Motohiro if (percpu_counter_read_positive(&vm_committed_as) < allowed) 1851da177e4SLinus Torvalds return 0; 1866d9f7839SHideo AOKI error: 1871da177e4SLinus Torvalds vm_unacct_memory(pages); 1881da177e4SLinus Torvalds 1891da177e4SLinus Torvalds return -ENOMEM; 1901da177e4SLinus Torvalds } 1911da177e4SLinus Torvalds 1921da177e4SLinus Torvalds /* 1931da177e4SLinus Torvalds * Requires inode->i_mapping->i_mmap_lock 1941da177e4SLinus Torvalds */ 1951da177e4SLinus Torvalds static void __remove_shared_vm_struct(struct vm_area_struct *vma, 1961da177e4SLinus Torvalds struct file *file, struct address_space *mapping) 1971da177e4SLinus Torvalds { 1981da177e4SLinus Torvalds if (vma->vm_flags & VM_DENYWRITE) 199d3ac7f89SJosef "Jeff" Sipek atomic_inc(&file->f_path.dentry->d_inode->i_writecount); 2001da177e4SLinus Torvalds if (vma->vm_flags & VM_SHARED) 2011da177e4SLinus Torvalds mapping->i_mmap_writable--; 2021da177e4SLinus Torvalds 2031da177e4SLinus Torvalds flush_dcache_mmap_lock(mapping); 2041da177e4SLinus Torvalds if (unlikely(vma->vm_flags & VM_NONLINEAR)) 2051da177e4SLinus Torvalds list_del_init(&vma->shared.vm_set.list); 2061da177e4SLinus Torvalds else 2071da177e4SLinus Torvalds vma_prio_tree_remove(vma, &mapping->i_mmap); 2081da177e4SLinus Torvalds flush_dcache_mmap_unlock(mapping); 2091da177e4SLinus Torvalds } 2101da177e4SLinus Torvalds 2111da177e4SLinus Torvalds /* 212a8fb5618SHugh Dickins * Unlink a file-based vm structure from its prio_tree, to hide 213a8fb5618SHugh Dickins * vma from rmap and vmtruncate before freeing its page tables. 2141da177e4SLinus Torvalds */ 215a8fb5618SHugh Dickins void unlink_file_vma(struct vm_area_struct *vma) 2161da177e4SLinus Torvalds { 2171da177e4SLinus Torvalds struct file *file = vma->vm_file; 2181da177e4SLinus Torvalds 2191da177e4SLinus Torvalds if (file) { 2201da177e4SLinus Torvalds struct address_space *mapping = file->f_mapping; 2211da177e4SLinus Torvalds spin_lock(&mapping->i_mmap_lock); 2221da177e4SLinus Torvalds __remove_shared_vm_struct(vma, file, mapping); 2231da177e4SLinus Torvalds spin_unlock(&mapping->i_mmap_lock); 2241da177e4SLinus Torvalds } 225a8fb5618SHugh Dickins } 226a8fb5618SHugh Dickins 227a8fb5618SHugh Dickins /* 228a8fb5618SHugh Dickins * Close a vm structure and free it, returning the next. 229a8fb5618SHugh Dickins */ 230a8fb5618SHugh Dickins static struct vm_area_struct *remove_vma(struct vm_area_struct *vma) 231a8fb5618SHugh Dickins { 232a8fb5618SHugh Dickins struct vm_area_struct *next = vma->vm_next; 233a8fb5618SHugh Dickins 234a8fb5618SHugh Dickins might_sleep(); 2351da177e4SLinus Torvalds if (vma->vm_ops && vma->vm_ops->close) 2361da177e4SLinus Torvalds vma->vm_ops->close(vma); 237925d1c40SMatt Helsley if (vma->vm_file) { 238a8fb5618SHugh Dickins fput(vma->vm_file); 239925d1c40SMatt Helsley if (vma->vm_flags & VM_EXECUTABLE) 240925d1c40SMatt Helsley removed_exe_file_vma(vma->vm_mm); 241925d1c40SMatt Helsley } 242f0be3d32SLee Schermerhorn mpol_put(vma_policy(vma)); 2431da177e4SLinus Torvalds kmem_cache_free(vm_area_cachep, vma); 244a8fb5618SHugh Dickins return next; 2451da177e4SLinus Torvalds } 2461da177e4SLinus Torvalds 2476a6160a7SHeiko Carstens SYSCALL_DEFINE1(brk, unsigned long, brk) 2481da177e4SLinus Torvalds { 2491da177e4SLinus Torvalds unsigned long rlim, retval; 2501da177e4SLinus Torvalds unsigned long newbrk, oldbrk; 2511da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 252a5b4592cSJiri Kosina unsigned long min_brk; 2531da177e4SLinus Torvalds 2541da177e4SLinus Torvalds down_write(&mm->mmap_sem); 2551da177e4SLinus Torvalds 256a5b4592cSJiri Kosina #ifdef CONFIG_COMPAT_BRK 257a5b4592cSJiri Kosina min_brk = mm->end_code; 258a5b4592cSJiri Kosina #else 259a5b4592cSJiri Kosina min_brk = mm->start_brk; 260a5b4592cSJiri Kosina #endif 261a5b4592cSJiri Kosina if (brk < min_brk) 2621da177e4SLinus Torvalds goto out; 2631e624196SRam Gupta 2641e624196SRam Gupta /* 2651e624196SRam Gupta * Check against rlimit here. If this check is done later after the test 2661e624196SRam Gupta * of oldbrk with newbrk then it can escape the test and let the data 2671e624196SRam Gupta * segment grow beyond its set limit the in case where the limit is 2681e624196SRam Gupta * not page aligned -Ram Gupta 2691e624196SRam Gupta */ 2701e624196SRam Gupta rlim = current->signal->rlim[RLIMIT_DATA].rlim_cur; 271c1d171a0SJiri Kosina if (rlim < RLIM_INFINITY && (brk - mm->start_brk) + 272c1d171a0SJiri Kosina (mm->end_data - mm->start_data) > rlim) 2731e624196SRam Gupta goto out; 2741e624196SRam Gupta 2751da177e4SLinus Torvalds newbrk = PAGE_ALIGN(brk); 2761da177e4SLinus Torvalds oldbrk = PAGE_ALIGN(mm->brk); 2771da177e4SLinus Torvalds if (oldbrk == newbrk) 2781da177e4SLinus Torvalds goto set_brk; 2791da177e4SLinus Torvalds 2801da177e4SLinus Torvalds /* Always allow shrinking brk. */ 2811da177e4SLinus Torvalds if (brk <= mm->brk) { 2821da177e4SLinus Torvalds if (!do_munmap(mm, newbrk, oldbrk-newbrk)) 2831da177e4SLinus Torvalds goto set_brk; 2841da177e4SLinus Torvalds goto out; 2851da177e4SLinus Torvalds } 2861da177e4SLinus Torvalds 2871da177e4SLinus Torvalds /* Check against existing mmap mappings. */ 2881da177e4SLinus Torvalds if (find_vma_intersection(mm, oldbrk, newbrk+PAGE_SIZE)) 2891da177e4SLinus Torvalds goto out; 2901da177e4SLinus Torvalds 2911da177e4SLinus Torvalds /* Ok, looks good - let it rip. */ 2921da177e4SLinus Torvalds if (do_brk(oldbrk, newbrk-oldbrk) != oldbrk) 2931da177e4SLinus Torvalds goto out; 2941da177e4SLinus Torvalds set_brk: 2951da177e4SLinus Torvalds mm->brk = brk; 2961da177e4SLinus Torvalds out: 2971da177e4SLinus Torvalds retval = mm->brk; 2981da177e4SLinus Torvalds up_write(&mm->mmap_sem); 2991da177e4SLinus Torvalds return retval; 3001da177e4SLinus Torvalds } 3011da177e4SLinus Torvalds 3021da177e4SLinus Torvalds #ifdef DEBUG_MM_RB 3031da177e4SLinus Torvalds static int browse_rb(struct rb_root *root) 3041da177e4SLinus Torvalds { 3051da177e4SLinus Torvalds int i = 0, j; 3061da177e4SLinus Torvalds struct rb_node *nd, *pn = NULL; 3071da177e4SLinus Torvalds unsigned long prev = 0, pend = 0; 3081da177e4SLinus Torvalds 3091da177e4SLinus Torvalds for (nd = rb_first(root); nd; nd = rb_next(nd)) { 3101da177e4SLinus Torvalds struct vm_area_struct *vma; 3111da177e4SLinus Torvalds vma = rb_entry(nd, struct vm_area_struct, vm_rb); 3121da177e4SLinus Torvalds if (vma->vm_start < prev) 3131da177e4SLinus Torvalds printk("vm_start %lx prev %lx\n", vma->vm_start, prev), i = -1; 3141da177e4SLinus Torvalds if (vma->vm_start < pend) 3151da177e4SLinus Torvalds printk("vm_start %lx pend %lx\n", vma->vm_start, pend); 3161da177e4SLinus Torvalds if (vma->vm_start > vma->vm_end) 3171da177e4SLinus Torvalds printk("vm_end %lx < vm_start %lx\n", vma->vm_end, vma->vm_start); 3181da177e4SLinus Torvalds i++; 3191da177e4SLinus Torvalds pn = nd; 320d1af65d1SDavid Miller prev = vma->vm_start; 321d1af65d1SDavid Miller pend = vma->vm_end; 3221da177e4SLinus Torvalds } 3231da177e4SLinus Torvalds j = 0; 3241da177e4SLinus Torvalds for (nd = pn; nd; nd = rb_prev(nd)) { 3251da177e4SLinus Torvalds j++; 3261da177e4SLinus Torvalds } 3271da177e4SLinus Torvalds if (i != j) 3281da177e4SLinus Torvalds printk("backwards %d, forwards %d\n", j, i), i = 0; 3291da177e4SLinus Torvalds return i; 3301da177e4SLinus Torvalds } 3311da177e4SLinus Torvalds 3321da177e4SLinus Torvalds void validate_mm(struct mm_struct *mm) 3331da177e4SLinus Torvalds { 3341da177e4SLinus Torvalds int bug = 0; 3351da177e4SLinus Torvalds int i = 0; 3361da177e4SLinus Torvalds struct vm_area_struct *tmp = mm->mmap; 3371da177e4SLinus Torvalds while (tmp) { 3381da177e4SLinus Torvalds tmp = tmp->vm_next; 3391da177e4SLinus Torvalds i++; 3401da177e4SLinus Torvalds } 3411da177e4SLinus Torvalds if (i != mm->map_count) 3421da177e4SLinus Torvalds printk("map_count %d vm_next %d\n", mm->map_count, i), bug = 1; 3431da177e4SLinus Torvalds i = browse_rb(&mm->mm_rb); 3441da177e4SLinus Torvalds if (i != mm->map_count) 3451da177e4SLinus Torvalds printk("map_count %d rb %d\n", mm->map_count, i), bug = 1; 34646a350efSEric Sesterhenn BUG_ON(bug); 3471da177e4SLinus Torvalds } 3481da177e4SLinus Torvalds #else 3491da177e4SLinus Torvalds #define validate_mm(mm) do { } while (0) 3501da177e4SLinus Torvalds #endif 3511da177e4SLinus Torvalds 3521da177e4SLinus Torvalds static struct vm_area_struct * 3531da177e4SLinus Torvalds find_vma_prepare(struct mm_struct *mm, unsigned long addr, 3541da177e4SLinus Torvalds struct vm_area_struct **pprev, struct rb_node ***rb_link, 3551da177e4SLinus Torvalds struct rb_node ** rb_parent) 3561da177e4SLinus Torvalds { 3571da177e4SLinus Torvalds struct vm_area_struct * vma; 3581da177e4SLinus Torvalds struct rb_node ** __rb_link, * __rb_parent, * rb_prev; 3591da177e4SLinus Torvalds 3601da177e4SLinus Torvalds __rb_link = &mm->mm_rb.rb_node; 3611da177e4SLinus Torvalds rb_prev = __rb_parent = NULL; 3621da177e4SLinus Torvalds vma = NULL; 3631da177e4SLinus Torvalds 3641da177e4SLinus Torvalds while (*__rb_link) { 3651da177e4SLinus Torvalds struct vm_area_struct *vma_tmp; 3661da177e4SLinus Torvalds 3671da177e4SLinus Torvalds __rb_parent = *__rb_link; 3681da177e4SLinus Torvalds vma_tmp = rb_entry(__rb_parent, struct vm_area_struct, vm_rb); 3691da177e4SLinus Torvalds 3701da177e4SLinus Torvalds if (vma_tmp->vm_end > addr) { 3711da177e4SLinus Torvalds vma = vma_tmp; 3721da177e4SLinus Torvalds if (vma_tmp->vm_start <= addr) 373dfe195fbSBenny Halevy break; 3741da177e4SLinus Torvalds __rb_link = &__rb_parent->rb_left; 3751da177e4SLinus Torvalds } else { 3761da177e4SLinus Torvalds rb_prev = __rb_parent; 3771da177e4SLinus Torvalds __rb_link = &__rb_parent->rb_right; 3781da177e4SLinus Torvalds } 3791da177e4SLinus Torvalds } 3801da177e4SLinus Torvalds 3811da177e4SLinus Torvalds *pprev = NULL; 3821da177e4SLinus Torvalds if (rb_prev) 3831da177e4SLinus Torvalds *pprev = rb_entry(rb_prev, struct vm_area_struct, vm_rb); 3841da177e4SLinus Torvalds *rb_link = __rb_link; 3851da177e4SLinus Torvalds *rb_parent = __rb_parent; 3861da177e4SLinus Torvalds return vma; 3871da177e4SLinus Torvalds } 3881da177e4SLinus Torvalds 3891da177e4SLinus Torvalds static inline void 3901da177e4SLinus Torvalds __vma_link_list(struct mm_struct *mm, struct vm_area_struct *vma, 3911da177e4SLinus Torvalds struct vm_area_struct *prev, struct rb_node *rb_parent) 3921da177e4SLinus Torvalds { 3931da177e4SLinus Torvalds if (prev) { 3941da177e4SLinus Torvalds vma->vm_next = prev->vm_next; 3951da177e4SLinus Torvalds prev->vm_next = vma; 3961da177e4SLinus Torvalds } else { 3971da177e4SLinus Torvalds mm->mmap = vma; 3981da177e4SLinus Torvalds if (rb_parent) 3991da177e4SLinus Torvalds vma->vm_next = rb_entry(rb_parent, 4001da177e4SLinus Torvalds struct vm_area_struct, vm_rb); 4011da177e4SLinus Torvalds else 4021da177e4SLinus Torvalds vma->vm_next = NULL; 4031da177e4SLinus Torvalds } 4041da177e4SLinus Torvalds } 4051da177e4SLinus Torvalds 4061da177e4SLinus Torvalds void __vma_link_rb(struct mm_struct *mm, struct vm_area_struct *vma, 4071da177e4SLinus Torvalds struct rb_node **rb_link, struct rb_node *rb_parent) 4081da177e4SLinus Torvalds { 4091da177e4SLinus Torvalds rb_link_node(&vma->vm_rb, rb_parent, rb_link); 4101da177e4SLinus Torvalds rb_insert_color(&vma->vm_rb, &mm->mm_rb); 4111da177e4SLinus Torvalds } 4121da177e4SLinus Torvalds 413cb8f488cSDenys Vlasenko static void __vma_link_file(struct vm_area_struct *vma) 4141da177e4SLinus Torvalds { 4151da177e4SLinus Torvalds struct file *file; 4161da177e4SLinus Torvalds 4171da177e4SLinus Torvalds file = vma->vm_file; 4181da177e4SLinus Torvalds if (file) { 4191da177e4SLinus Torvalds struct address_space *mapping = file->f_mapping; 4201da177e4SLinus Torvalds 4211da177e4SLinus Torvalds if (vma->vm_flags & VM_DENYWRITE) 422d3ac7f89SJosef "Jeff" Sipek atomic_dec(&file->f_path.dentry->d_inode->i_writecount); 4231da177e4SLinus Torvalds if (vma->vm_flags & VM_SHARED) 4241da177e4SLinus Torvalds mapping->i_mmap_writable++; 4251da177e4SLinus Torvalds 4261da177e4SLinus Torvalds flush_dcache_mmap_lock(mapping); 4271da177e4SLinus Torvalds if (unlikely(vma->vm_flags & VM_NONLINEAR)) 4281da177e4SLinus Torvalds vma_nonlinear_insert(vma, &mapping->i_mmap_nonlinear); 4291da177e4SLinus Torvalds else 4301da177e4SLinus Torvalds vma_prio_tree_insert(vma, &mapping->i_mmap); 4311da177e4SLinus Torvalds flush_dcache_mmap_unlock(mapping); 4321da177e4SLinus Torvalds } 4331da177e4SLinus Torvalds } 4341da177e4SLinus Torvalds 4351da177e4SLinus Torvalds static void 4361da177e4SLinus Torvalds __vma_link(struct mm_struct *mm, struct vm_area_struct *vma, 4371da177e4SLinus Torvalds struct vm_area_struct *prev, struct rb_node **rb_link, 4381da177e4SLinus Torvalds struct rb_node *rb_parent) 4391da177e4SLinus Torvalds { 4401da177e4SLinus Torvalds __vma_link_list(mm, vma, prev, rb_parent); 4411da177e4SLinus Torvalds __vma_link_rb(mm, vma, rb_link, rb_parent); 4421da177e4SLinus Torvalds __anon_vma_link(vma); 4431da177e4SLinus Torvalds } 4441da177e4SLinus Torvalds 4451da177e4SLinus Torvalds static void vma_link(struct mm_struct *mm, struct vm_area_struct *vma, 4461da177e4SLinus Torvalds struct vm_area_struct *prev, struct rb_node **rb_link, 4471da177e4SLinus Torvalds struct rb_node *rb_parent) 4481da177e4SLinus Torvalds { 4491da177e4SLinus Torvalds struct address_space *mapping = NULL; 4501da177e4SLinus Torvalds 4511da177e4SLinus Torvalds if (vma->vm_file) 4521da177e4SLinus Torvalds mapping = vma->vm_file->f_mapping; 4531da177e4SLinus Torvalds 4541da177e4SLinus Torvalds if (mapping) { 4551da177e4SLinus Torvalds spin_lock(&mapping->i_mmap_lock); 4561da177e4SLinus Torvalds vma->vm_truncate_count = mapping->truncate_count; 4571da177e4SLinus Torvalds } 4581da177e4SLinus Torvalds anon_vma_lock(vma); 4591da177e4SLinus Torvalds 4601da177e4SLinus Torvalds __vma_link(mm, vma, prev, rb_link, rb_parent); 4611da177e4SLinus Torvalds __vma_link_file(vma); 4621da177e4SLinus Torvalds 4631da177e4SLinus Torvalds anon_vma_unlock(vma); 4641da177e4SLinus Torvalds if (mapping) 4651da177e4SLinus Torvalds spin_unlock(&mapping->i_mmap_lock); 4661da177e4SLinus Torvalds 4671da177e4SLinus Torvalds mm->map_count++; 4681da177e4SLinus Torvalds validate_mm(mm); 4691da177e4SLinus Torvalds } 4701da177e4SLinus Torvalds 4711da177e4SLinus Torvalds /* 4721da177e4SLinus Torvalds * Helper for vma_adjust in the split_vma insert case: 4731da177e4SLinus Torvalds * insert vm structure into list and rbtree and anon_vma, 4741da177e4SLinus Torvalds * but it has already been inserted into prio_tree earlier. 4751da177e4SLinus Torvalds */ 47648aae425SZhenwenXu static void __insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma) 4771da177e4SLinus Torvalds { 4781da177e4SLinus Torvalds struct vm_area_struct *__vma, *prev; 4791da177e4SLinus Torvalds struct rb_node **rb_link, *rb_parent; 4801da177e4SLinus Torvalds 4811da177e4SLinus Torvalds __vma = find_vma_prepare(mm, vma->vm_start,&prev, &rb_link, &rb_parent); 48246a350efSEric Sesterhenn BUG_ON(__vma && __vma->vm_start < vma->vm_end); 4831da177e4SLinus Torvalds __vma_link(mm, vma, prev, rb_link, rb_parent); 4841da177e4SLinus Torvalds mm->map_count++; 4851da177e4SLinus Torvalds } 4861da177e4SLinus Torvalds 4871da177e4SLinus Torvalds static inline void 4881da177e4SLinus Torvalds __vma_unlink(struct mm_struct *mm, struct vm_area_struct *vma, 4891da177e4SLinus Torvalds struct vm_area_struct *prev) 4901da177e4SLinus Torvalds { 4911da177e4SLinus Torvalds prev->vm_next = vma->vm_next; 4921da177e4SLinus Torvalds rb_erase(&vma->vm_rb, &mm->mm_rb); 4931da177e4SLinus Torvalds if (mm->mmap_cache == vma) 4941da177e4SLinus Torvalds mm->mmap_cache = prev; 4951da177e4SLinus Torvalds } 4961da177e4SLinus Torvalds 4971da177e4SLinus Torvalds /* 4981da177e4SLinus Torvalds * We cannot adjust vm_start, vm_end, vm_pgoff fields of a vma that 4991da177e4SLinus Torvalds * is already present in an i_mmap tree without adjusting the tree. 5001da177e4SLinus Torvalds * The following helper function should be used when such adjustments 5011da177e4SLinus Torvalds * are necessary. The "insert" vma (if any) is to be inserted 5021da177e4SLinus Torvalds * before we drop the necessary locks. 5031da177e4SLinus Torvalds */ 5041da177e4SLinus Torvalds void vma_adjust(struct vm_area_struct *vma, unsigned long start, 5051da177e4SLinus Torvalds unsigned long end, pgoff_t pgoff, struct vm_area_struct *insert) 5061da177e4SLinus Torvalds { 5071da177e4SLinus Torvalds struct mm_struct *mm = vma->vm_mm; 5081da177e4SLinus Torvalds struct vm_area_struct *next = vma->vm_next; 5091da177e4SLinus Torvalds struct vm_area_struct *importer = NULL; 5101da177e4SLinus Torvalds struct address_space *mapping = NULL; 5111da177e4SLinus Torvalds struct prio_tree_root *root = NULL; 5121da177e4SLinus Torvalds struct file *file = vma->vm_file; 5131da177e4SLinus Torvalds struct anon_vma *anon_vma = NULL; 5141da177e4SLinus Torvalds long adjust_next = 0; 5151da177e4SLinus Torvalds int remove_next = 0; 5161da177e4SLinus Torvalds 5171da177e4SLinus Torvalds if (next && !insert) { 5181da177e4SLinus Torvalds if (end >= next->vm_end) { 5191da177e4SLinus Torvalds /* 5201da177e4SLinus Torvalds * vma expands, overlapping all the next, and 5211da177e4SLinus Torvalds * perhaps the one after too (mprotect case 6). 5221da177e4SLinus Torvalds */ 5231da177e4SLinus Torvalds again: remove_next = 1 + (end > next->vm_end); 5241da177e4SLinus Torvalds end = next->vm_end; 5251da177e4SLinus Torvalds anon_vma = next->anon_vma; 5261da177e4SLinus Torvalds importer = vma; 5271da177e4SLinus Torvalds } else if (end > next->vm_start) { 5281da177e4SLinus Torvalds /* 5291da177e4SLinus Torvalds * vma expands, overlapping part of the next: 5301da177e4SLinus Torvalds * mprotect case 5 shifting the boundary up. 5311da177e4SLinus Torvalds */ 5321da177e4SLinus Torvalds adjust_next = (end - next->vm_start) >> PAGE_SHIFT; 5331da177e4SLinus Torvalds anon_vma = next->anon_vma; 5341da177e4SLinus Torvalds importer = vma; 5351da177e4SLinus Torvalds } else if (end < vma->vm_end) { 5361da177e4SLinus Torvalds /* 5371da177e4SLinus Torvalds * vma shrinks, and !insert tells it's not 5381da177e4SLinus Torvalds * split_vma inserting another: so it must be 5391da177e4SLinus Torvalds * mprotect case 4 shifting the boundary down. 5401da177e4SLinus Torvalds */ 5411da177e4SLinus Torvalds adjust_next = - ((vma->vm_end - end) >> PAGE_SHIFT); 5421da177e4SLinus Torvalds anon_vma = next->anon_vma; 5431da177e4SLinus Torvalds importer = next; 5441da177e4SLinus Torvalds } 5451da177e4SLinus Torvalds } 5461da177e4SLinus Torvalds 5471da177e4SLinus Torvalds if (file) { 5481da177e4SLinus Torvalds mapping = file->f_mapping; 5491da177e4SLinus Torvalds if (!(vma->vm_flags & VM_NONLINEAR)) 5501da177e4SLinus Torvalds root = &mapping->i_mmap; 5511da177e4SLinus Torvalds spin_lock(&mapping->i_mmap_lock); 5521da177e4SLinus Torvalds if (importer && 5531da177e4SLinus Torvalds vma->vm_truncate_count != next->vm_truncate_count) { 5541da177e4SLinus Torvalds /* 5551da177e4SLinus Torvalds * unmap_mapping_range might be in progress: 5561da177e4SLinus Torvalds * ensure that the expanding vma is rescanned. 5571da177e4SLinus Torvalds */ 5581da177e4SLinus Torvalds importer->vm_truncate_count = 0; 5591da177e4SLinus Torvalds } 5601da177e4SLinus Torvalds if (insert) { 5611da177e4SLinus Torvalds insert->vm_truncate_count = vma->vm_truncate_count; 5621da177e4SLinus Torvalds /* 5631da177e4SLinus Torvalds * Put into prio_tree now, so instantiated pages 5641da177e4SLinus Torvalds * are visible to arm/parisc __flush_dcache_page 5651da177e4SLinus Torvalds * throughout; but we cannot insert into address 5661da177e4SLinus Torvalds * space until vma start or end is updated. 5671da177e4SLinus Torvalds */ 5681da177e4SLinus Torvalds __vma_link_file(insert); 5691da177e4SLinus Torvalds } 5701da177e4SLinus Torvalds } 5711da177e4SLinus Torvalds 5721da177e4SLinus Torvalds /* 5731da177e4SLinus Torvalds * When changing only vma->vm_end, we don't really need 5741da177e4SLinus Torvalds * anon_vma lock: but is that case worth optimizing out? 5751da177e4SLinus Torvalds */ 5761da177e4SLinus Torvalds if (vma->anon_vma) 5771da177e4SLinus Torvalds anon_vma = vma->anon_vma; 5781da177e4SLinus Torvalds if (anon_vma) { 5791da177e4SLinus Torvalds spin_lock(&anon_vma->lock); 5801da177e4SLinus Torvalds /* 5811da177e4SLinus Torvalds * Easily overlooked: when mprotect shifts the boundary, 5821da177e4SLinus Torvalds * make sure the expanding vma has anon_vma set if the 5831da177e4SLinus Torvalds * shrinking vma had, to cover any anon pages imported. 5841da177e4SLinus Torvalds */ 5851da177e4SLinus Torvalds if (importer && !importer->anon_vma) { 5861da177e4SLinus Torvalds importer->anon_vma = anon_vma; 5871da177e4SLinus Torvalds __anon_vma_link(importer); 5881da177e4SLinus Torvalds } 5891da177e4SLinus Torvalds } 5901da177e4SLinus Torvalds 5911da177e4SLinus Torvalds if (root) { 5921da177e4SLinus Torvalds flush_dcache_mmap_lock(mapping); 5931da177e4SLinus Torvalds vma_prio_tree_remove(vma, root); 5941da177e4SLinus Torvalds if (adjust_next) 5951da177e4SLinus Torvalds vma_prio_tree_remove(next, root); 5961da177e4SLinus Torvalds } 5971da177e4SLinus Torvalds 5981da177e4SLinus Torvalds vma->vm_start = start; 5991da177e4SLinus Torvalds vma->vm_end = end; 6001da177e4SLinus Torvalds vma->vm_pgoff = pgoff; 6011da177e4SLinus Torvalds if (adjust_next) { 6021da177e4SLinus Torvalds next->vm_start += adjust_next << PAGE_SHIFT; 6031da177e4SLinus Torvalds next->vm_pgoff += adjust_next; 6041da177e4SLinus Torvalds } 6051da177e4SLinus Torvalds 6061da177e4SLinus Torvalds if (root) { 6071da177e4SLinus Torvalds if (adjust_next) 6081da177e4SLinus Torvalds vma_prio_tree_insert(next, root); 6091da177e4SLinus Torvalds vma_prio_tree_insert(vma, root); 6101da177e4SLinus Torvalds flush_dcache_mmap_unlock(mapping); 6111da177e4SLinus Torvalds } 6121da177e4SLinus Torvalds 6131da177e4SLinus Torvalds if (remove_next) { 6141da177e4SLinus Torvalds /* 6151da177e4SLinus Torvalds * vma_merge has merged next into vma, and needs 6161da177e4SLinus Torvalds * us to remove next before dropping the locks. 6171da177e4SLinus Torvalds */ 6181da177e4SLinus Torvalds __vma_unlink(mm, next, vma); 6191da177e4SLinus Torvalds if (file) 6201da177e4SLinus Torvalds __remove_shared_vm_struct(next, file, mapping); 6211da177e4SLinus Torvalds if (next->anon_vma) 6221da177e4SLinus Torvalds __anon_vma_merge(vma, next); 6231da177e4SLinus Torvalds } else if (insert) { 6241da177e4SLinus Torvalds /* 6251da177e4SLinus Torvalds * split_vma has split insert from vma, and needs 6261da177e4SLinus Torvalds * us to insert it before dropping the locks 6271da177e4SLinus Torvalds * (it may either follow vma or precede it). 6281da177e4SLinus Torvalds */ 6291da177e4SLinus Torvalds __insert_vm_struct(mm, insert); 6301da177e4SLinus Torvalds } 6311da177e4SLinus Torvalds 6321da177e4SLinus Torvalds if (anon_vma) 6331da177e4SLinus Torvalds spin_unlock(&anon_vma->lock); 6341da177e4SLinus Torvalds if (mapping) 6351da177e4SLinus Torvalds spin_unlock(&mapping->i_mmap_lock); 6361da177e4SLinus Torvalds 6371da177e4SLinus Torvalds if (remove_next) { 638925d1c40SMatt Helsley if (file) { 6391da177e4SLinus Torvalds fput(file); 640925d1c40SMatt Helsley if (next->vm_flags & VM_EXECUTABLE) 641925d1c40SMatt Helsley removed_exe_file_vma(mm); 642925d1c40SMatt Helsley } 6431da177e4SLinus Torvalds mm->map_count--; 644f0be3d32SLee Schermerhorn mpol_put(vma_policy(next)); 6451da177e4SLinus Torvalds kmem_cache_free(vm_area_cachep, next); 6461da177e4SLinus Torvalds /* 6471da177e4SLinus Torvalds * In mprotect's case 6 (see comments on vma_merge), 6481da177e4SLinus Torvalds * we must remove another next too. It would clutter 6491da177e4SLinus Torvalds * up the code too much to do both in one go. 6501da177e4SLinus Torvalds */ 6511da177e4SLinus Torvalds if (remove_next == 2) { 6521da177e4SLinus Torvalds next = vma->vm_next; 6531da177e4SLinus Torvalds goto again; 6541da177e4SLinus Torvalds } 6551da177e4SLinus Torvalds } 6561da177e4SLinus Torvalds 6571da177e4SLinus Torvalds validate_mm(mm); 6581da177e4SLinus Torvalds } 6591da177e4SLinus Torvalds 66033bfad54SLinus Torvalds /* Flags that can be inherited from an existing mapping when merging */ 66133bfad54SLinus Torvalds #define VM_MERGEABLE_FLAGS (VM_CAN_NONLINEAR) 66233bfad54SLinus Torvalds 6631da177e4SLinus Torvalds /* 6641da177e4SLinus Torvalds * If the vma has a ->close operation then the driver probably needs to release 6651da177e4SLinus Torvalds * per-vma resources, so we don't attempt to merge those. 6661da177e4SLinus Torvalds */ 6671da177e4SLinus Torvalds static inline int is_mergeable_vma(struct vm_area_struct *vma, 6681da177e4SLinus Torvalds struct file *file, unsigned long vm_flags) 6691da177e4SLinus Torvalds { 67033bfad54SLinus Torvalds if ((vma->vm_flags ^ vm_flags) & ~VM_MERGEABLE_FLAGS) 6711da177e4SLinus Torvalds return 0; 6721da177e4SLinus Torvalds if (vma->vm_file != file) 6731da177e4SLinus Torvalds return 0; 6741da177e4SLinus Torvalds if (vma->vm_ops && vma->vm_ops->close) 6751da177e4SLinus Torvalds return 0; 6761da177e4SLinus Torvalds return 1; 6771da177e4SLinus Torvalds } 6781da177e4SLinus Torvalds 6791da177e4SLinus Torvalds static inline int is_mergeable_anon_vma(struct anon_vma *anon_vma1, 6801da177e4SLinus Torvalds struct anon_vma *anon_vma2) 6811da177e4SLinus Torvalds { 6821da177e4SLinus Torvalds return !anon_vma1 || !anon_vma2 || (anon_vma1 == anon_vma2); 6831da177e4SLinus Torvalds } 6841da177e4SLinus Torvalds 6851da177e4SLinus Torvalds /* 6861da177e4SLinus Torvalds * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff) 6871da177e4SLinus Torvalds * in front of (at a lower virtual address and file offset than) the vma. 6881da177e4SLinus Torvalds * 6891da177e4SLinus Torvalds * We cannot merge two vmas if they have differently assigned (non-NULL) 6901da177e4SLinus Torvalds * anon_vmas, nor if same anon_vma is assigned but offsets incompatible. 6911da177e4SLinus Torvalds * 6921da177e4SLinus Torvalds * We don't check here for the merged mmap wrapping around the end of pagecache 6931da177e4SLinus Torvalds * indices (16TB on ia32) because do_mmap_pgoff() does not permit mmap's which 6941da177e4SLinus Torvalds * wrap, nor mmaps which cover the final page at index -1UL. 6951da177e4SLinus Torvalds */ 6961da177e4SLinus Torvalds static int 6971da177e4SLinus Torvalds can_vma_merge_before(struct vm_area_struct *vma, unsigned long vm_flags, 6981da177e4SLinus Torvalds struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff) 6991da177e4SLinus Torvalds { 7001da177e4SLinus Torvalds if (is_mergeable_vma(vma, file, vm_flags) && 7011da177e4SLinus Torvalds is_mergeable_anon_vma(anon_vma, vma->anon_vma)) { 7021da177e4SLinus Torvalds if (vma->vm_pgoff == vm_pgoff) 7031da177e4SLinus Torvalds return 1; 7041da177e4SLinus Torvalds } 7051da177e4SLinus Torvalds return 0; 7061da177e4SLinus Torvalds } 7071da177e4SLinus Torvalds 7081da177e4SLinus Torvalds /* 7091da177e4SLinus Torvalds * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff) 7101da177e4SLinus Torvalds * beyond (at a higher virtual address and file offset than) the vma. 7111da177e4SLinus Torvalds * 7121da177e4SLinus Torvalds * We cannot merge two vmas if they have differently assigned (non-NULL) 7131da177e4SLinus Torvalds * anon_vmas, nor if same anon_vma is assigned but offsets incompatible. 7141da177e4SLinus Torvalds */ 7151da177e4SLinus Torvalds static int 7161da177e4SLinus Torvalds can_vma_merge_after(struct vm_area_struct *vma, unsigned long vm_flags, 7171da177e4SLinus Torvalds struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff) 7181da177e4SLinus Torvalds { 7191da177e4SLinus Torvalds if (is_mergeable_vma(vma, file, vm_flags) && 7201da177e4SLinus Torvalds is_mergeable_anon_vma(anon_vma, vma->anon_vma)) { 7211da177e4SLinus Torvalds pgoff_t vm_pglen; 7221da177e4SLinus Torvalds vm_pglen = (vma->vm_end - vma->vm_start) >> PAGE_SHIFT; 7231da177e4SLinus Torvalds if (vma->vm_pgoff + vm_pglen == vm_pgoff) 7241da177e4SLinus Torvalds return 1; 7251da177e4SLinus Torvalds } 7261da177e4SLinus Torvalds return 0; 7271da177e4SLinus Torvalds } 7281da177e4SLinus Torvalds 7291da177e4SLinus Torvalds /* 7301da177e4SLinus Torvalds * Given a mapping request (addr,end,vm_flags,file,pgoff), figure out 7311da177e4SLinus Torvalds * whether that can be merged with its predecessor or its successor. 7321da177e4SLinus Torvalds * Or both (it neatly fills a hole). 7331da177e4SLinus Torvalds * 7341da177e4SLinus Torvalds * In most cases - when called for mmap, brk or mremap - [addr,end) is 7351da177e4SLinus Torvalds * certain not to be mapped by the time vma_merge is called; but when 7361da177e4SLinus Torvalds * called for mprotect, it is certain to be already mapped (either at 7371da177e4SLinus Torvalds * an offset within prev, or at the start of next), and the flags of 7381da177e4SLinus Torvalds * this area are about to be changed to vm_flags - and the no-change 7391da177e4SLinus Torvalds * case has already been eliminated. 7401da177e4SLinus Torvalds * 7411da177e4SLinus Torvalds * The following mprotect cases have to be considered, where AAAA is 7421da177e4SLinus Torvalds * the area passed down from mprotect_fixup, never extending beyond one 7431da177e4SLinus Torvalds * vma, PPPPPP is the prev vma specified, and NNNNNN the next vma after: 7441da177e4SLinus Torvalds * 7451da177e4SLinus Torvalds * AAAA AAAA AAAA AAAA 7461da177e4SLinus Torvalds * PPPPPPNNNNNN PPPPPPNNNNNN PPPPPPNNNNNN PPPPNNNNXXXX 7471da177e4SLinus Torvalds * cannot merge might become might become might become 7481da177e4SLinus Torvalds * PPNNNNNNNNNN PPPPPPPPPPNN PPPPPPPPPPPP 6 or 7491da177e4SLinus Torvalds * mmap, brk or case 4 below case 5 below PPPPPPPPXXXX 7 or 7501da177e4SLinus Torvalds * mremap move: PPPPNNNNNNNN 8 7511da177e4SLinus Torvalds * AAAA 7521da177e4SLinus Torvalds * PPPP NNNN PPPPPPPPPPPP PPPPPPPPNNNN PPPPNNNNNNNN 7531da177e4SLinus Torvalds * might become case 1 below case 2 below case 3 below 7541da177e4SLinus Torvalds * 7551da177e4SLinus Torvalds * Odd one out? Case 8, because it extends NNNN but needs flags of XXXX: 7561da177e4SLinus Torvalds * mprotect_fixup updates vm_flags & vm_page_prot on successful return. 7571da177e4SLinus Torvalds */ 7581da177e4SLinus Torvalds struct vm_area_struct *vma_merge(struct mm_struct *mm, 7591da177e4SLinus Torvalds struct vm_area_struct *prev, unsigned long addr, 7601da177e4SLinus Torvalds unsigned long end, unsigned long vm_flags, 7611da177e4SLinus Torvalds struct anon_vma *anon_vma, struct file *file, 7621da177e4SLinus Torvalds pgoff_t pgoff, struct mempolicy *policy) 7631da177e4SLinus Torvalds { 7641da177e4SLinus Torvalds pgoff_t pglen = (end - addr) >> PAGE_SHIFT; 7651da177e4SLinus Torvalds struct vm_area_struct *area, *next; 7661da177e4SLinus Torvalds 7671da177e4SLinus Torvalds /* 7681da177e4SLinus Torvalds * We later require that vma->vm_flags == vm_flags, 7691da177e4SLinus Torvalds * so this tests vma->vm_flags & VM_SPECIAL, too. 7701da177e4SLinus Torvalds */ 7711da177e4SLinus Torvalds if (vm_flags & VM_SPECIAL) 7721da177e4SLinus Torvalds return NULL; 7731da177e4SLinus Torvalds 7741da177e4SLinus Torvalds if (prev) 7751da177e4SLinus Torvalds next = prev->vm_next; 7761da177e4SLinus Torvalds else 7771da177e4SLinus Torvalds next = mm->mmap; 7781da177e4SLinus Torvalds area = next; 7791da177e4SLinus Torvalds if (next && next->vm_end == end) /* cases 6, 7, 8 */ 7801da177e4SLinus Torvalds next = next->vm_next; 7811da177e4SLinus Torvalds 7821da177e4SLinus Torvalds /* 7831da177e4SLinus Torvalds * Can it merge with the predecessor? 7841da177e4SLinus Torvalds */ 7851da177e4SLinus Torvalds if (prev && prev->vm_end == addr && 7861da177e4SLinus Torvalds mpol_equal(vma_policy(prev), policy) && 7871da177e4SLinus Torvalds can_vma_merge_after(prev, vm_flags, 7881da177e4SLinus Torvalds anon_vma, file, pgoff)) { 7891da177e4SLinus Torvalds /* 7901da177e4SLinus Torvalds * OK, it can. Can we now merge in the successor as well? 7911da177e4SLinus Torvalds */ 7921da177e4SLinus Torvalds if (next && end == next->vm_start && 7931da177e4SLinus Torvalds mpol_equal(policy, vma_policy(next)) && 7941da177e4SLinus Torvalds can_vma_merge_before(next, vm_flags, 7951da177e4SLinus Torvalds anon_vma, file, pgoff+pglen) && 7961da177e4SLinus Torvalds is_mergeable_anon_vma(prev->anon_vma, 7971da177e4SLinus Torvalds next->anon_vma)) { 7981da177e4SLinus Torvalds /* cases 1, 6 */ 7991da177e4SLinus Torvalds vma_adjust(prev, prev->vm_start, 8001da177e4SLinus Torvalds next->vm_end, prev->vm_pgoff, NULL); 8011da177e4SLinus Torvalds } else /* cases 2, 5, 7 */ 8021da177e4SLinus Torvalds vma_adjust(prev, prev->vm_start, 8031da177e4SLinus Torvalds end, prev->vm_pgoff, NULL); 8041da177e4SLinus Torvalds return prev; 8051da177e4SLinus Torvalds } 8061da177e4SLinus Torvalds 8071da177e4SLinus Torvalds /* 8081da177e4SLinus Torvalds * Can this new request be merged in front of next? 8091da177e4SLinus Torvalds */ 8101da177e4SLinus Torvalds if (next && end == next->vm_start && 8111da177e4SLinus Torvalds mpol_equal(policy, vma_policy(next)) && 8121da177e4SLinus Torvalds can_vma_merge_before(next, vm_flags, 8131da177e4SLinus Torvalds anon_vma, file, pgoff+pglen)) { 8141da177e4SLinus Torvalds if (prev && addr < prev->vm_end) /* case 4 */ 8151da177e4SLinus Torvalds vma_adjust(prev, prev->vm_start, 8161da177e4SLinus Torvalds addr, prev->vm_pgoff, NULL); 8171da177e4SLinus Torvalds else /* cases 3, 8 */ 8181da177e4SLinus Torvalds vma_adjust(area, addr, next->vm_end, 8191da177e4SLinus Torvalds next->vm_pgoff - pglen, NULL); 8201da177e4SLinus Torvalds return area; 8211da177e4SLinus Torvalds } 8221da177e4SLinus Torvalds 8231da177e4SLinus Torvalds return NULL; 8241da177e4SLinus Torvalds } 8251da177e4SLinus Torvalds 8261da177e4SLinus Torvalds /* 8271da177e4SLinus Torvalds * find_mergeable_anon_vma is used by anon_vma_prepare, to check 8281da177e4SLinus Torvalds * neighbouring vmas for a suitable anon_vma, before it goes off 8291da177e4SLinus Torvalds * to allocate a new anon_vma. It checks because a repetitive 8301da177e4SLinus Torvalds * sequence of mprotects and faults may otherwise lead to distinct 8311da177e4SLinus Torvalds * anon_vmas being allocated, preventing vma merge in subsequent 8321da177e4SLinus Torvalds * mprotect. 8331da177e4SLinus Torvalds */ 8341da177e4SLinus Torvalds struct anon_vma *find_mergeable_anon_vma(struct vm_area_struct *vma) 8351da177e4SLinus Torvalds { 8361da177e4SLinus Torvalds struct vm_area_struct *near; 8371da177e4SLinus Torvalds unsigned long vm_flags; 8381da177e4SLinus Torvalds 8391da177e4SLinus Torvalds near = vma->vm_next; 8401da177e4SLinus Torvalds if (!near) 8411da177e4SLinus Torvalds goto try_prev; 8421da177e4SLinus Torvalds 8431da177e4SLinus Torvalds /* 8441da177e4SLinus Torvalds * Since only mprotect tries to remerge vmas, match flags 8451da177e4SLinus Torvalds * which might be mprotected into each other later on. 8461da177e4SLinus Torvalds * Neither mlock nor madvise tries to remerge at present, 8471da177e4SLinus Torvalds * so leave their flags as obstructing a merge. 8481da177e4SLinus Torvalds */ 8491da177e4SLinus Torvalds vm_flags = vma->vm_flags & ~(VM_READ|VM_WRITE|VM_EXEC); 8501da177e4SLinus Torvalds vm_flags |= near->vm_flags & (VM_READ|VM_WRITE|VM_EXEC); 8511da177e4SLinus Torvalds 8521da177e4SLinus Torvalds if (near->anon_vma && vma->vm_end == near->vm_start && 8531da177e4SLinus Torvalds mpol_equal(vma_policy(vma), vma_policy(near)) && 8541da177e4SLinus Torvalds can_vma_merge_before(near, vm_flags, 8551da177e4SLinus Torvalds NULL, vma->vm_file, vma->vm_pgoff + 8561da177e4SLinus Torvalds ((vma->vm_end - vma->vm_start) >> PAGE_SHIFT))) 8571da177e4SLinus Torvalds return near->anon_vma; 8581da177e4SLinus Torvalds try_prev: 8591da177e4SLinus Torvalds /* 8601da177e4SLinus Torvalds * It is potentially slow to have to call find_vma_prev here. 8611da177e4SLinus Torvalds * But it's only on the first write fault on the vma, not 8621da177e4SLinus Torvalds * every time, and we could devise a way to avoid it later 8631da177e4SLinus Torvalds * (e.g. stash info in next's anon_vma_node when assigning 8641da177e4SLinus Torvalds * an anon_vma, or when trying vma_merge). Another time. 8651da177e4SLinus Torvalds */ 86646a350efSEric Sesterhenn BUG_ON(find_vma_prev(vma->vm_mm, vma->vm_start, &near) != vma); 8671da177e4SLinus Torvalds if (!near) 8681da177e4SLinus Torvalds goto none; 8691da177e4SLinus Torvalds 8701da177e4SLinus Torvalds vm_flags = vma->vm_flags & ~(VM_READ|VM_WRITE|VM_EXEC); 8711da177e4SLinus Torvalds vm_flags |= near->vm_flags & (VM_READ|VM_WRITE|VM_EXEC); 8721da177e4SLinus Torvalds 8731da177e4SLinus Torvalds if (near->anon_vma && near->vm_end == vma->vm_start && 8741da177e4SLinus Torvalds mpol_equal(vma_policy(near), vma_policy(vma)) && 8751da177e4SLinus Torvalds can_vma_merge_after(near, vm_flags, 8761da177e4SLinus Torvalds NULL, vma->vm_file, vma->vm_pgoff)) 8771da177e4SLinus Torvalds return near->anon_vma; 8781da177e4SLinus Torvalds none: 8791da177e4SLinus Torvalds /* 8801da177e4SLinus Torvalds * There's no absolute need to look only at touching neighbours: 8811da177e4SLinus Torvalds * we could search further afield for "compatible" anon_vmas. 8821da177e4SLinus Torvalds * But it would probably just be a waste of time searching, 8831da177e4SLinus Torvalds * or lead to too many vmas hanging off the same anon_vma. 8841da177e4SLinus Torvalds * We're trying to allow mprotect remerging later on, 8851da177e4SLinus Torvalds * not trying to minimize memory used for anon_vmas. 8861da177e4SLinus Torvalds */ 8871da177e4SLinus Torvalds return NULL; 8881da177e4SLinus Torvalds } 8891da177e4SLinus Torvalds 8901da177e4SLinus Torvalds #ifdef CONFIG_PROC_FS 891ab50b8edSHugh Dickins void vm_stat_account(struct mm_struct *mm, unsigned long flags, 8921da177e4SLinus Torvalds struct file *file, long pages) 8931da177e4SLinus Torvalds { 8941da177e4SLinus Torvalds const unsigned long stack_flags 8951da177e4SLinus Torvalds = VM_STACK_FLAGS & (VM_GROWSUP|VM_GROWSDOWN); 8961da177e4SLinus Torvalds 8971da177e4SLinus Torvalds if (file) { 8981da177e4SLinus Torvalds mm->shared_vm += pages; 8991da177e4SLinus Torvalds if ((flags & (VM_EXEC|VM_WRITE)) == VM_EXEC) 9001da177e4SLinus Torvalds mm->exec_vm += pages; 9011da177e4SLinus Torvalds } else if (flags & stack_flags) 9021da177e4SLinus Torvalds mm->stack_vm += pages; 9031da177e4SLinus Torvalds if (flags & (VM_RESERVED|VM_IO)) 9041da177e4SLinus Torvalds mm->reserved_vm += pages; 9051da177e4SLinus Torvalds } 9061da177e4SLinus Torvalds #endif /* CONFIG_PROC_FS */ 9071da177e4SLinus Torvalds 9081da177e4SLinus Torvalds /* 90927f5de79SJianjun Kong * The caller must hold down_write(¤t->mm->mmap_sem). 9101da177e4SLinus Torvalds */ 9111da177e4SLinus Torvalds 9121da177e4SLinus Torvalds unsigned long do_mmap_pgoff(struct file *file, unsigned long addr, 9131da177e4SLinus Torvalds unsigned long len, unsigned long prot, 9141da177e4SLinus Torvalds unsigned long flags, unsigned long pgoff) 9151da177e4SLinus Torvalds { 9161da177e4SLinus Torvalds struct mm_struct * mm = current->mm; 9171da177e4SLinus Torvalds struct inode *inode; 9181da177e4SLinus Torvalds unsigned int vm_flags; 9191da177e4SLinus Torvalds int error; 9200165ab44SMiklos Szeredi unsigned long reqprot = prot; 9211da177e4SLinus Torvalds 9221da177e4SLinus Torvalds /* 9231da177e4SLinus Torvalds * Does the application expect PROT_READ to imply PROT_EXEC? 9241da177e4SLinus Torvalds * 9251da177e4SLinus Torvalds * (the exception is when the underlying filesystem is noexec 9261da177e4SLinus Torvalds * mounted, in which case we dont add PROT_EXEC.) 9271da177e4SLinus Torvalds */ 9281da177e4SLinus Torvalds if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC)) 929d3ac7f89SJosef "Jeff" Sipek if (!(file && (file->f_path.mnt->mnt_flags & MNT_NOEXEC))) 9301da177e4SLinus Torvalds prot |= PROT_EXEC; 9311da177e4SLinus Torvalds 9321da177e4SLinus Torvalds if (!len) 9331da177e4SLinus Torvalds return -EINVAL; 9341da177e4SLinus Torvalds 9357cd94146SEric Paris if (!(flags & MAP_FIXED)) 9367cd94146SEric Paris addr = round_hint_to_min(addr); 9377cd94146SEric Paris 9383a459756SKirill Korotaev error = arch_mmap_check(addr, len, flags); 9393a459756SKirill Korotaev if (error) 9403a459756SKirill Korotaev return error; 9413a459756SKirill Korotaev 9421da177e4SLinus Torvalds /* Careful about overflows.. */ 9431da177e4SLinus Torvalds len = PAGE_ALIGN(len); 9441da177e4SLinus Torvalds if (!len || len > TASK_SIZE) 9451da177e4SLinus Torvalds return -ENOMEM; 9461da177e4SLinus Torvalds 9471da177e4SLinus Torvalds /* offset overflow? */ 9481da177e4SLinus Torvalds if ((pgoff + (len >> PAGE_SHIFT)) < pgoff) 9491da177e4SLinus Torvalds return -EOVERFLOW; 9501da177e4SLinus Torvalds 9511da177e4SLinus Torvalds /* Too many mappings? */ 9521da177e4SLinus Torvalds if (mm->map_count > sysctl_max_map_count) 9531da177e4SLinus Torvalds return -ENOMEM; 9541da177e4SLinus Torvalds 9551da177e4SLinus Torvalds /* Obtain the address to map to. we verify (or select) it and ensure 9561da177e4SLinus Torvalds * that it represents a valid section of the address space. 9571da177e4SLinus Torvalds */ 9581da177e4SLinus Torvalds addr = get_unmapped_area(file, addr, len, pgoff, flags); 9591da177e4SLinus Torvalds if (addr & ~PAGE_MASK) 9601da177e4SLinus Torvalds return addr; 9611da177e4SLinus Torvalds 9621da177e4SLinus Torvalds /* Do simple checking here so the lower-level routines won't have 9631da177e4SLinus Torvalds * to. we assume access permissions have been handled by the open 9641da177e4SLinus Torvalds * of the memory object, so we don't do any here. 9651da177e4SLinus Torvalds */ 9661da177e4SLinus Torvalds vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags) | 9671da177e4SLinus Torvalds mm->def_flags | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC; 9681da177e4SLinus Torvalds 9691da177e4SLinus Torvalds if (flags & MAP_LOCKED) { 9701da177e4SLinus Torvalds if (!can_do_mlock()) 9711da177e4SLinus Torvalds return -EPERM; 9721da177e4SLinus Torvalds vm_flags |= VM_LOCKED; 9731da177e4SLinus Torvalds } 974ba470de4SRik van Riel 9751da177e4SLinus Torvalds /* mlock MCL_FUTURE? */ 9761da177e4SLinus Torvalds if (vm_flags & VM_LOCKED) { 9771da177e4SLinus Torvalds unsigned long locked, lock_limit; 97893ea1d0aSChris Wright locked = len >> PAGE_SHIFT; 97993ea1d0aSChris Wright locked += mm->locked_vm; 9801da177e4SLinus Torvalds lock_limit = current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur; 98193ea1d0aSChris Wright lock_limit >>= PAGE_SHIFT; 9821da177e4SLinus Torvalds if (locked > lock_limit && !capable(CAP_IPC_LOCK)) 9831da177e4SLinus Torvalds return -EAGAIN; 9841da177e4SLinus Torvalds } 9851da177e4SLinus Torvalds 986d3ac7f89SJosef "Jeff" Sipek inode = file ? file->f_path.dentry->d_inode : NULL; 9871da177e4SLinus Torvalds 9881da177e4SLinus Torvalds if (file) { 9891da177e4SLinus Torvalds switch (flags & MAP_TYPE) { 9901da177e4SLinus Torvalds case MAP_SHARED: 9911da177e4SLinus Torvalds if ((prot&PROT_WRITE) && !(file->f_mode&FMODE_WRITE)) 9921da177e4SLinus Torvalds return -EACCES; 9931da177e4SLinus Torvalds 9941da177e4SLinus Torvalds /* 9951da177e4SLinus Torvalds * Make sure we don't allow writing to an append-only 9961da177e4SLinus Torvalds * file.. 9971da177e4SLinus Torvalds */ 9981da177e4SLinus Torvalds if (IS_APPEND(inode) && (file->f_mode & FMODE_WRITE)) 9991da177e4SLinus Torvalds return -EACCES; 10001da177e4SLinus Torvalds 10011da177e4SLinus Torvalds /* 10021da177e4SLinus Torvalds * Make sure there are no mandatory locks on the file. 10031da177e4SLinus Torvalds */ 10041da177e4SLinus Torvalds if (locks_verify_locked(inode)) 10051da177e4SLinus Torvalds return -EAGAIN; 10061da177e4SLinus Torvalds 10071da177e4SLinus Torvalds vm_flags |= VM_SHARED | VM_MAYSHARE; 10081da177e4SLinus Torvalds if (!(file->f_mode & FMODE_WRITE)) 10091da177e4SLinus Torvalds vm_flags &= ~(VM_MAYWRITE | VM_SHARED); 10101da177e4SLinus Torvalds 10111da177e4SLinus Torvalds /* fall through */ 10121da177e4SLinus Torvalds case MAP_PRIVATE: 10131da177e4SLinus Torvalds if (!(file->f_mode & FMODE_READ)) 10141da177e4SLinus Torvalds return -EACCES; 1015d3ac7f89SJosef "Jeff" Sipek if (file->f_path.mnt->mnt_flags & MNT_NOEXEC) { 101680c5606cSLinus Torvalds if (vm_flags & VM_EXEC) 101780c5606cSLinus Torvalds return -EPERM; 101880c5606cSLinus Torvalds vm_flags &= ~VM_MAYEXEC; 101980c5606cSLinus Torvalds } 102080c5606cSLinus Torvalds 102180c5606cSLinus Torvalds if (!file->f_op || !file->f_op->mmap) 102280c5606cSLinus Torvalds return -ENODEV; 10231da177e4SLinus Torvalds break; 10241da177e4SLinus Torvalds 10251da177e4SLinus Torvalds default: 10261da177e4SLinus Torvalds return -EINVAL; 10271da177e4SLinus Torvalds } 10281da177e4SLinus Torvalds } else { 10291da177e4SLinus Torvalds switch (flags & MAP_TYPE) { 10301da177e4SLinus Torvalds case MAP_SHARED: 1031ce363942STejun Heo /* 1032ce363942STejun Heo * Ignore pgoff. 1033ce363942STejun Heo */ 1034ce363942STejun Heo pgoff = 0; 10351da177e4SLinus Torvalds vm_flags |= VM_SHARED | VM_MAYSHARE; 10361da177e4SLinus Torvalds break; 10371da177e4SLinus Torvalds case MAP_PRIVATE: 10381da177e4SLinus Torvalds /* 10391da177e4SLinus Torvalds * Set pgoff according to addr for anon_vma. 10401da177e4SLinus Torvalds */ 10411da177e4SLinus Torvalds pgoff = addr >> PAGE_SHIFT; 10421da177e4SLinus Torvalds break; 10431da177e4SLinus Torvalds default: 10441da177e4SLinus Torvalds return -EINVAL; 10451da177e4SLinus Torvalds } 10461da177e4SLinus Torvalds } 10471da177e4SLinus Torvalds 1048ed032189SEric Paris error = security_file_mmap(file, reqprot, prot, flags, addr, 0); 10491da177e4SLinus Torvalds if (error) 10501da177e4SLinus Torvalds return error; 10516146f0d5SMimi Zohar error = ima_file_mmap(file, prot); 10526146f0d5SMimi Zohar if (error) 10536146f0d5SMimi Zohar return error; 10541da177e4SLinus Torvalds 10555a6fe125SMel Gorman return mmap_region(file, addr, len, flags, vm_flags, pgoff); 10560165ab44SMiklos Szeredi } 10570165ab44SMiklos Szeredi EXPORT_SYMBOL(do_mmap_pgoff); 10580165ab44SMiklos Szeredi 10594e950f6fSAlexey Dobriyan /* 10604e950f6fSAlexey Dobriyan * Some shared mappigns will want the pages marked read-only 10614e950f6fSAlexey Dobriyan * to track write events. If so, we'll downgrade vm_page_prot 10624e950f6fSAlexey Dobriyan * to the private version (using protection_map[] without the 10634e950f6fSAlexey Dobriyan * VM_SHARED bit). 10644e950f6fSAlexey Dobriyan */ 10654e950f6fSAlexey Dobriyan int vma_wants_writenotify(struct vm_area_struct *vma) 10664e950f6fSAlexey Dobriyan { 10674e950f6fSAlexey Dobriyan unsigned int vm_flags = vma->vm_flags; 10684e950f6fSAlexey Dobriyan 10694e950f6fSAlexey Dobriyan /* If it was private or non-writable, the write bit is already clear */ 10704e950f6fSAlexey Dobriyan if ((vm_flags & (VM_WRITE|VM_SHARED)) != ((VM_WRITE|VM_SHARED))) 10714e950f6fSAlexey Dobriyan return 0; 10724e950f6fSAlexey Dobriyan 10734e950f6fSAlexey Dobriyan /* The backer wishes to know when pages are first written to? */ 10744e950f6fSAlexey Dobriyan if (vma->vm_ops && vma->vm_ops->page_mkwrite) 10754e950f6fSAlexey Dobriyan return 1; 10764e950f6fSAlexey Dobriyan 10774e950f6fSAlexey Dobriyan /* The open routine did something to the protections already? */ 10784e950f6fSAlexey Dobriyan if (pgprot_val(vma->vm_page_prot) != 10793ed75eb8SColy Li pgprot_val(vm_get_page_prot(vm_flags))) 10804e950f6fSAlexey Dobriyan return 0; 10814e950f6fSAlexey Dobriyan 10824e950f6fSAlexey Dobriyan /* Specialty mapping? */ 10834e950f6fSAlexey Dobriyan if (vm_flags & (VM_PFNMAP|VM_INSERTPAGE)) 10844e950f6fSAlexey Dobriyan return 0; 10854e950f6fSAlexey Dobriyan 10864e950f6fSAlexey Dobriyan /* Can the mapping track the dirty pages? */ 10874e950f6fSAlexey Dobriyan return vma->vm_file && vma->vm_file->f_mapping && 10884e950f6fSAlexey Dobriyan mapping_cap_account_dirty(vma->vm_file->f_mapping); 10894e950f6fSAlexey Dobriyan } 10904e950f6fSAlexey Dobriyan 1091fc8744adSLinus Torvalds /* 1092fc8744adSLinus Torvalds * We account for memory if it's a private writeable mapping, 10935a6fe125SMel Gorman * not hugepages and VM_NORESERVE wasn't set. 1094fc8744adSLinus Torvalds */ 10955a6fe125SMel Gorman static inline int accountable_mapping(struct file *file, unsigned int vm_flags) 1096fc8744adSLinus Torvalds { 10975a6fe125SMel Gorman /* 10985a6fe125SMel Gorman * hugetlb has its own accounting separate from the core VM 10995a6fe125SMel Gorman * VM_HUGETLB may not be set yet so we cannot check for that flag. 11005a6fe125SMel Gorman */ 11015a6fe125SMel Gorman if (file && is_file_hugepages(file)) 11025a6fe125SMel Gorman return 0; 11035a6fe125SMel Gorman 1104fc8744adSLinus Torvalds return (vm_flags & (VM_NORESERVE | VM_SHARED | VM_WRITE)) == VM_WRITE; 1105fc8744adSLinus Torvalds } 1106fc8744adSLinus Torvalds 11070165ab44SMiklos Szeredi unsigned long mmap_region(struct file *file, unsigned long addr, 11080165ab44SMiklos Szeredi unsigned long len, unsigned long flags, 11095a6fe125SMel Gorman unsigned int vm_flags, unsigned long pgoff) 11100165ab44SMiklos Szeredi { 11110165ab44SMiklos Szeredi struct mm_struct *mm = current->mm; 11120165ab44SMiklos Szeredi struct vm_area_struct *vma, *prev; 11130165ab44SMiklos Szeredi int correct_wcount = 0; 11140165ab44SMiklos Szeredi int error; 11150165ab44SMiklos Szeredi struct rb_node **rb_link, *rb_parent; 11160165ab44SMiklos Szeredi unsigned long charged = 0; 11170165ab44SMiklos Szeredi struct inode *inode = file ? file->f_path.dentry->d_inode : NULL; 11180165ab44SMiklos Szeredi 11191da177e4SLinus Torvalds /* Clear old maps */ 11201da177e4SLinus Torvalds error = -ENOMEM; 11211da177e4SLinus Torvalds munmap_back: 11221da177e4SLinus Torvalds vma = find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent); 11231da177e4SLinus Torvalds if (vma && vma->vm_start < addr + len) { 11241da177e4SLinus Torvalds if (do_munmap(mm, addr, len)) 11251da177e4SLinus Torvalds return -ENOMEM; 11261da177e4SLinus Torvalds goto munmap_back; 11271da177e4SLinus Torvalds } 11281da177e4SLinus Torvalds 11291da177e4SLinus Torvalds /* Check against address space limit. */ 1130119f657cSakpm@osdl.org if (!may_expand_vm(mm, len >> PAGE_SHIFT)) 11311da177e4SLinus Torvalds return -ENOMEM; 11321da177e4SLinus Torvalds 1133fc8744adSLinus Torvalds /* 1134fc8744adSLinus Torvalds * Set 'VM_NORESERVE' if we should not account for the 11355a6fe125SMel Gorman * memory use of this mapping. 1136fc8744adSLinus Torvalds */ 11375a6fe125SMel Gorman if ((flags & MAP_NORESERVE)) { 11385a6fe125SMel Gorman /* We honor MAP_NORESERVE if allowed to overcommit */ 11395a6fe125SMel Gorman if (sysctl_overcommit_memory != OVERCOMMIT_NEVER) 1140fc8744adSLinus Torvalds vm_flags |= VM_NORESERVE; 11415a6fe125SMel Gorman 11425a6fe125SMel Gorman /* hugetlb applies strict overcommit unless MAP_NORESERVE */ 11435a6fe125SMel Gorman if (file && is_file_hugepages(file)) 1144cdfd4325SAndy Whitcroft vm_flags |= VM_NORESERVE; 11455a6fe125SMel Gorman } 1146cdfd4325SAndy Whitcroft 11471da177e4SLinus Torvalds /* 11481da177e4SLinus Torvalds * Private writable mapping: check memory availability 11491da177e4SLinus Torvalds */ 11505a6fe125SMel Gorman if (accountable_mapping(file, vm_flags)) { 11511da177e4SLinus Torvalds charged = len >> PAGE_SHIFT; 11521da177e4SLinus Torvalds if (security_vm_enough_memory(charged)) 11531da177e4SLinus Torvalds return -ENOMEM; 11541da177e4SLinus Torvalds vm_flags |= VM_ACCOUNT; 11551da177e4SLinus Torvalds } 11561da177e4SLinus Torvalds 11571da177e4SLinus Torvalds /* 1158de33c8dbSLinus Torvalds * Can we just expand an old mapping? 11591da177e4SLinus Torvalds */ 1160de33c8dbSLinus Torvalds vma = vma_merge(mm, prev, addr, addr + len, vm_flags, NULL, file, pgoff, NULL); 1161ba470de4SRik van Riel if (vma) 11621da177e4SLinus Torvalds goto out; 11631da177e4SLinus Torvalds 11641da177e4SLinus Torvalds /* 11651da177e4SLinus Torvalds * Determine the object being mapped and call the appropriate 11661da177e4SLinus Torvalds * specific mapper. the address has already been validated, but 11671da177e4SLinus Torvalds * not unmapped, but the maps are removed from the list. 11681da177e4SLinus Torvalds */ 1169c5e3b83eSPekka Enberg vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL); 11701da177e4SLinus Torvalds if (!vma) { 11711da177e4SLinus Torvalds error = -ENOMEM; 11721da177e4SLinus Torvalds goto unacct_error; 11731da177e4SLinus Torvalds } 11741da177e4SLinus Torvalds 11751da177e4SLinus Torvalds vma->vm_mm = mm; 11761da177e4SLinus Torvalds vma->vm_start = addr; 11771da177e4SLinus Torvalds vma->vm_end = addr + len; 11781da177e4SLinus Torvalds vma->vm_flags = vm_flags; 11793ed75eb8SColy Li vma->vm_page_prot = vm_get_page_prot(vm_flags); 11801da177e4SLinus Torvalds vma->vm_pgoff = pgoff; 11811da177e4SLinus Torvalds 11821da177e4SLinus Torvalds if (file) { 11831da177e4SLinus Torvalds error = -EINVAL; 11841da177e4SLinus Torvalds if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP)) 11851da177e4SLinus Torvalds goto free_vma; 11861da177e4SLinus Torvalds if (vm_flags & VM_DENYWRITE) { 11871da177e4SLinus Torvalds error = deny_write_access(file); 11881da177e4SLinus Torvalds if (error) 11891da177e4SLinus Torvalds goto free_vma; 11901da177e4SLinus Torvalds correct_wcount = 1; 11911da177e4SLinus Torvalds } 11921da177e4SLinus Torvalds vma->vm_file = file; 11931da177e4SLinus Torvalds get_file(file); 11941da177e4SLinus Torvalds error = file->f_op->mmap(file, vma); 11951da177e4SLinus Torvalds if (error) 11961da177e4SLinus Torvalds goto unmap_and_free_vma; 1197925d1c40SMatt Helsley if (vm_flags & VM_EXECUTABLE) 1198925d1c40SMatt Helsley added_exe_file_vma(mm); 11991da177e4SLinus Torvalds } else if (vm_flags & VM_SHARED) { 12001da177e4SLinus Torvalds error = shmem_zero_setup(vma); 12011da177e4SLinus Torvalds if (error) 12021da177e4SLinus Torvalds goto free_vma; 12031da177e4SLinus Torvalds } 12041da177e4SLinus Torvalds 12051da177e4SLinus Torvalds /* Can addr have changed?? 12061da177e4SLinus Torvalds * 12071da177e4SLinus Torvalds * Answer: Yes, several device drivers can do it in their 12081da177e4SLinus Torvalds * f_op->mmap method. -DaveM 12091da177e4SLinus Torvalds */ 12101da177e4SLinus Torvalds addr = vma->vm_start; 12111da177e4SLinus Torvalds pgoff = vma->vm_pgoff; 12121da177e4SLinus Torvalds vm_flags = vma->vm_flags; 12131da177e4SLinus Torvalds 1214d08b3851SPeter Zijlstra if (vma_wants_writenotify(vma)) 12151ddd439eSHugh Dickins vma->vm_page_prot = vm_get_page_prot(vm_flags & ~VM_SHARED); 1216d08b3851SPeter Zijlstra 12174d3d5b41SOleg Nesterov vma_link(mm, vma, prev, rb_link, rb_parent); 12184d3d5b41SOleg Nesterov file = vma->vm_file; 12194d3d5b41SOleg Nesterov 12204d3d5b41SOleg Nesterov /* Once vma denies write, undo our temporary denial count */ 12214d3d5b41SOleg Nesterov if (correct_wcount) 12224d3d5b41SOleg Nesterov atomic_inc(&inode->i_writecount); 12231da177e4SLinus Torvalds out: 1224cdd6c482SIngo Molnar perf_event_mmap(vma); 12250a4a9391SPeter Zijlstra 12261da177e4SLinus Torvalds mm->total_vm += len >> PAGE_SHIFT; 1227ab50b8edSHugh Dickins vm_stat_account(mm, vm_flags, file, len >> PAGE_SHIFT); 12281da177e4SLinus Torvalds if (vm_flags & VM_LOCKED) { 1229ba470de4SRik van Riel /* 1230ba470de4SRik van Riel * makes pages present; downgrades, drops, reacquires mmap_sem 1231ba470de4SRik van Riel */ 1232ba470de4SRik van Riel long nr_pages = mlock_vma_pages_range(vma, addr, addr + len); 1233ba470de4SRik van Riel if (nr_pages < 0) 1234ba470de4SRik van Riel return nr_pages; /* vma gone! */ 1235ba470de4SRik van Riel mm->locked_vm += (len >> PAGE_SHIFT) - nr_pages; 1236ba470de4SRik van Riel } else if ((flags & MAP_POPULATE) && !(flags & MAP_NONBLOCK)) 123754cb8821SNick Piggin make_pages_present(addr, addr + len); 12381da177e4SLinus Torvalds return addr; 12391da177e4SLinus Torvalds 12401da177e4SLinus Torvalds unmap_and_free_vma: 12411da177e4SLinus Torvalds if (correct_wcount) 12421da177e4SLinus Torvalds atomic_inc(&inode->i_writecount); 12431da177e4SLinus Torvalds vma->vm_file = NULL; 12441da177e4SLinus Torvalds fput(file); 12451da177e4SLinus Torvalds 12461da177e4SLinus Torvalds /* Undo any partial mapping done by a device driver. */ 1247e0da382cSHugh Dickins unmap_region(mm, vma, prev, vma->vm_start, vma->vm_end); 1248e0da382cSHugh Dickins charged = 0; 12491da177e4SLinus Torvalds free_vma: 12501da177e4SLinus Torvalds kmem_cache_free(vm_area_cachep, vma); 12511da177e4SLinus Torvalds unacct_error: 12521da177e4SLinus Torvalds if (charged) 12531da177e4SLinus Torvalds vm_unacct_memory(charged); 12541da177e4SLinus Torvalds return error; 12551da177e4SLinus Torvalds } 12561da177e4SLinus Torvalds 12571da177e4SLinus Torvalds /* Get an address range which is currently unmapped. 12581da177e4SLinus Torvalds * For shmat() with addr=0. 12591da177e4SLinus Torvalds * 12601da177e4SLinus Torvalds * Ugly calling convention alert: 12611da177e4SLinus Torvalds * Return value with the low bits set means error value, 12621da177e4SLinus Torvalds * ie 12631da177e4SLinus Torvalds * if (ret & ~PAGE_MASK) 12641da177e4SLinus Torvalds * error = ret; 12651da177e4SLinus Torvalds * 12661da177e4SLinus Torvalds * This function "knows" that -ENOMEM has the bits set. 12671da177e4SLinus Torvalds */ 12681da177e4SLinus Torvalds #ifndef HAVE_ARCH_UNMAPPED_AREA 12691da177e4SLinus Torvalds unsigned long 12701da177e4SLinus Torvalds arch_get_unmapped_area(struct file *filp, unsigned long addr, 12711da177e4SLinus Torvalds unsigned long len, unsigned long pgoff, unsigned long flags) 12721da177e4SLinus Torvalds { 12731da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 12741da177e4SLinus Torvalds struct vm_area_struct *vma; 12751da177e4SLinus Torvalds unsigned long start_addr; 12761da177e4SLinus Torvalds 12771da177e4SLinus Torvalds if (len > TASK_SIZE) 12781da177e4SLinus Torvalds return -ENOMEM; 12791da177e4SLinus Torvalds 128006abdfb4SBenjamin Herrenschmidt if (flags & MAP_FIXED) 128106abdfb4SBenjamin Herrenschmidt return addr; 128206abdfb4SBenjamin Herrenschmidt 12831da177e4SLinus Torvalds if (addr) { 12841da177e4SLinus Torvalds addr = PAGE_ALIGN(addr); 12851da177e4SLinus Torvalds vma = find_vma(mm, addr); 12861da177e4SLinus Torvalds if (TASK_SIZE - len >= addr && 12871da177e4SLinus Torvalds (!vma || addr + len <= vma->vm_start)) 12881da177e4SLinus Torvalds return addr; 12891da177e4SLinus Torvalds } 12901363c3cdSWolfgang Wander if (len > mm->cached_hole_size) { 12911da177e4SLinus Torvalds start_addr = addr = mm->free_area_cache; 12921363c3cdSWolfgang Wander } else { 12931363c3cdSWolfgang Wander start_addr = addr = TASK_UNMAPPED_BASE; 12941363c3cdSWolfgang Wander mm->cached_hole_size = 0; 12951363c3cdSWolfgang Wander } 12961da177e4SLinus Torvalds 12971da177e4SLinus Torvalds full_search: 12981da177e4SLinus Torvalds for (vma = find_vma(mm, addr); ; vma = vma->vm_next) { 12991da177e4SLinus Torvalds /* At this point: (!vma || addr < vma->vm_end). */ 13001da177e4SLinus Torvalds if (TASK_SIZE - len < addr) { 13011da177e4SLinus Torvalds /* 13021da177e4SLinus Torvalds * Start a new search - just in case we missed 13031da177e4SLinus Torvalds * some holes. 13041da177e4SLinus Torvalds */ 13051da177e4SLinus Torvalds if (start_addr != TASK_UNMAPPED_BASE) { 13061363c3cdSWolfgang Wander addr = TASK_UNMAPPED_BASE; 13071363c3cdSWolfgang Wander start_addr = addr; 13081363c3cdSWolfgang Wander mm->cached_hole_size = 0; 13091da177e4SLinus Torvalds goto full_search; 13101da177e4SLinus Torvalds } 13111da177e4SLinus Torvalds return -ENOMEM; 13121da177e4SLinus Torvalds } 13131da177e4SLinus Torvalds if (!vma || addr + len <= vma->vm_start) { 13141da177e4SLinus Torvalds /* 13151da177e4SLinus Torvalds * Remember the place where we stopped the search: 13161da177e4SLinus Torvalds */ 13171da177e4SLinus Torvalds mm->free_area_cache = addr + len; 13181da177e4SLinus Torvalds return addr; 13191da177e4SLinus Torvalds } 13201363c3cdSWolfgang Wander if (addr + mm->cached_hole_size < vma->vm_start) 13211363c3cdSWolfgang Wander mm->cached_hole_size = vma->vm_start - addr; 13221da177e4SLinus Torvalds addr = vma->vm_end; 13231da177e4SLinus Torvalds } 13241da177e4SLinus Torvalds } 13251da177e4SLinus Torvalds #endif 13261da177e4SLinus Torvalds 13271363c3cdSWolfgang Wander void arch_unmap_area(struct mm_struct *mm, unsigned long addr) 13281da177e4SLinus Torvalds { 13291da177e4SLinus Torvalds /* 13301da177e4SLinus Torvalds * Is this a new hole at the lowest possible address? 13311da177e4SLinus Torvalds */ 13321363c3cdSWolfgang Wander if (addr >= TASK_UNMAPPED_BASE && addr < mm->free_area_cache) { 13331363c3cdSWolfgang Wander mm->free_area_cache = addr; 13341363c3cdSWolfgang Wander mm->cached_hole_size = ~0UL; 13351363c3cdSWolfgang Wander } 13361da177e4SLinus Torvalds } 13371da177e4SLinus Torvalds 13381da177e4SLinus Torvalds /* 13391da177e4SLinus Torvalds * This mmap-allocator allocates new areas top-down from below the 13401da177e4SLinus Torvalds * stack's low limit (the base): 13411da177e4SLinus Torvalds */ 13421da177e4SLinus Torvalds #ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN 13431da177e4SLinus Torvalds unsigned long 13441da177e4SLinus Torvalds arch_get_unmapped_area_topdown(struct file *filp, const unsigned long addr0, 13451da177e4SLinus Torvalds const unsigned long len, const unsigned long pgoff, 13461da177e4SLinus Torvalds const unsigned long flags) 13471da177e4SLinus Torvalds { 13481da177e4SLinus Torvalds struct vm_area_struct *vma; 13491da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 13501da177e4SLinus Torvalds unsigned long addr = addr0; 13511da177e4SLinus Torvalds 13521da177e4SLinus Torvalds /* requested length too big for entire address space */ 13531da177e4SLinus Torvalds if (len > TASK_SIZE) 13541da177e4SLinus Torvalds return -ENOMEM; 13551da177e4SLinus Torvalds 135606abdfb4SBenjamin Herrenschmidt if (flags & MAP_FIXED) 135706abdfb4SBenjamin Herrenschmidt return addr; 135806abdfb4SBenjamin Herrenschmidt 13591da177e4SLinus Torvalds /* requesting a specific address */ 13601da177e4SLinus Torvalds if (addr) { 13611da177e4SLinus Torvalds addr = PAGE_ALIGN(addr); 13621da177e4SLinus Torvalds vma = find_vma(mm, addr); 13631da177e4SLinus Torvalds if (TASK_SIZE - len >= addr && 13641da177e4SLinus Torvalds (!vma || addr + len <= vma->vm_start)) 13651da177e4SLinus Torvalds return addr; 13661da177e4SLinus Torvalds } 13671da177e4SLinus Torvalds 13681363c3cdSWolfgang Wander /* check if free_area_cache is useful for us */ 13691363c3cdSWolfgang Wander if (len <= mm->cached_hole_size) { 13701363c3cdSWolfgang Wander mm->cached_hole_size = 0; 13711363c3cdSWolfgang Wander mm->free_area_cache = mm->mmap_base; 13721363c3cdSWolfgang Wander } 13731363c3cdSWolfgang Wander 13741da177e4SLinus Torvalds /* either no address requested or can't fit in requested address hole */ 13751da177e4SLinus Torvalds addr = mm->free_area_cache; 13761da177e4SLinus Torvalds 13771da177e4SLinus Torvalds /* make sure it can fit in the remaining address space */ 137849a43876SLinus Torvalds if (addr > len) { 13791da177e4SLinus Torvalds vma = find_vma(mm, addr-len); 13801da177e4SLinus Torvalds if (!vma || addr <= vma->vm_start) 13811da177e4SLinus Torvalds /* remember the address as a hint for next time */ 13821da177e4SLinus Torvalds return (mm->free_area_cache = addr-len); 13831da177e4SLinus Torvalds } 13841da177e4SLinus Torvalds 138573219d17SChris Wright if (mm->mmap_base < len) 138673219d17SChris Wright goto bottomup; 138773219d17SChris Wright 13881da177e4SLinus Torvalds addr = mm->mmap_base-len; 13891da177e4SLinus Torvalds 13901da177e4SLinus Torvalds do { 13911da177e4SLinus Torvalds /* 13921da177e4SLinus Torvalds * Lookup failure means no vma is above this address, 13931da177e4SLinus Torvalds * else if new region fits below vma->vm_start, 13941da177e4SLinus Torvalds * return with success: 13951da177e4SLinus Torvalds */ 13961da177e4SLinus Torvalds vma = find_vma(mm, addr); 13971da177e4SLinus Torvalds if (!vma || addr+len <= vma->vm_start) 13981da177e4SLinus Torvalds /* remember the address as a hint for next time */ 13991da177e4SLinus Torvalds return (mm->free_area_cache = addr); 14001da177e4SLinus Torvalds 14011363c3cdSWolfgang Wander /* remember the largest hole we saw so far */ 14021363c3cdSWolfgang Wander if (addr + mm->cached_hole_size < vma->vm_start) 14031363c3cdSWolfgang Wander mm->cached_hole_size = vma->vm_start - addr; 14041363c3cdSWolfgang Wander 14051da177e4SLinus Torvalds /* try just below the current vma->vm_start */ 14061da177e4SLinus Torvalds addr = vma->vm_start-len; 140749a43876SLinus Torvalds } while (len < vma->vm_start); 14081da177e4SLinus Torvalds 140973219d17SChris Wright bottomup: 14101da177e4SLinus Torvalds /* 14111da177e4SLinus Torvalds * A failed mmap() very likely causes application failure, 14121da177e4SLinus Torvalds * so fall back to the bottom-up function here. This scenario 14131da177e4SLinus Torvalds * can happen with large stack limits and large mmap() 14141da177e4SLinus Torvalds * allocations. 14151da177e4SLinus Torvalds */ 14161363c3cdSWolfgang Wander mm->cached_hole_size = ~0UL; 14171da177e4SLinus Torvalds mm->free_area_cache = TASK_UNMAPPED_BASE; 14181da177e4SLinus Torvalds addr = arch_get_unmapped_area(filp, addr0, len, pgoff, flags); 14191da177e4SLinus Torvalds /* 14201da177e4SLinus Torvalds * Restore the topdown base: 14211da177e4SLinus Torvalds */ 14221da177e4SLinus Torvalds mm->free_area_cache = mm->mmap_base; 14231363c3cdSWolfgang Wander mm->cached_hole_size = ~0UL; 14241da177e4SLinus Torvalds 14251da177e4SLinus Torvalds return addr; 14261da177e4SLinus Torvalds } 14271da177e4SLinus Torvalds #endif 14281da177e4SLinus Torvalds 14291363c3cdSWolfgang Wander void arch_unmap_area_topdown(struct mm_struct *mm, unsigned long addr) 14301da177e4SLinus Torvalds { 14311da177e4SLinus Torvalds /* 14321da177e4SLinus Torvalds * Is this a new hole at the highest possible address? 14331da177e4SLinus Torvalds */ 14341363c3cdSWolfgang Wander if (addr > mm->free_area_cache) 14351363c3cdSWolfgang Wander mm->free_area_cache = addr; 14361da177e4SLinus Torvalds 14371da177e4SLinus Torvalds /* dont allow allocations above current base */ 14381363c3cdSWolfgang Wander if (mm->free_area_cache > mm->mmap_base) 14391363c3cdSWolfgang Wander mm->free_area_cache = mm->mmap_base; 14401da177e4SLinus Torvalds } 14411da177e4SLinus Torvalds 14421da177e4SLinus Torvalds unsigned long 14431da177e4SLinus Torvalds get_unmapped_area(struct file *file, unsigned long addr, unsigned long len, 14441da177e4SLinus Torvalds unsigned long pgoff, unsigned long flags) 14451da177e4SLinus Torvalds { 144606abdfb4SBenjamin Herrenschmidt unsigned long (*get_area)(struct file *, unsigned long, 144706abdfb4SBenjamin Herrenschmidt unsigned long, unsigned long, unsigned long); 144807ab67c8SLinus Torvalds 144907ab67c8SLinus Torvalds get_area = current->mm->get_unmapped_area; 145007ab67c8SLinus Torvalds if (file && file->f_op && file->f_op->get_unmapped_area) 145107ab67c8SLinus Torvalds get_area = file->f_op->get_unmapped_area; 145207ab67c8SLinus Torvalds addr = get_area(file, addr, len, pgoff, flags); 145307ab67c8SLinus Torvalds if (IS_ERR_VALUE(addr)) 145407ab67c8SLinus Torvalds return addr; 145507ab67c8SLinus Torvalds 14561da177e4SLinus Torvalds if (addr > TASK_SIZE - len) 14571da177e4SLinus Torvalds return -ENOMEM; 14581da177e4SLinus Torvalds if (addr & ~PAGE_MASK) 14591da177e4SLinus Torvalds return -EINVAL; 146006abdfb4SBenjamin Herrenschmidt 146108e7d9b5SMartin Schwidefsky return arch_rebalance_pgtables(addr, len); 14621da177e4SLinus Torvalds } 14631da177e4SLinus Torvalds 14641da177e4SLinus Torvalds EXPORT_SYMBOL(get_unmapped_area); 14651da177e4SLinus Torvalds 14661da177e4SLinus Torvalds /* Look up the first VMA which satisfies addr < vm_end, NULL if none. */ 14671da177e4SLinus Torvalds struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr) 14681da177e4SLinus Torvalds { 14691da177e4SLinus Torvalds struct vm_area_struct *vma = NULL; 14701da177e4SLinus Torvalds 14711da177e4SLinus Torvalds if (mm) { 14721da177e4SLinus Torvalds /* Check the cache first. */ 14731da177e4SLinus Torvalds /* (Cache hit rate is typically around 35%.) */ 14741da177e4SLinus Torvalds vma = mm->mmap_cache; 14751da177e4SLinus Torvalds if (!(vma && vma->vm_end > addr && vma->vm_start <= addr)) { 14761da177e4SLinus Torvalds struct rb_node * rb_node; 14771da177e4SLinus Torvalds 14781da177e4SLinus Torvalds rb_node = mm->mm_rb.rb_node; 14791da177e4SLinus Torvalds vma = NULL; 14801da177e4SLinus Torvalds 14811da177e4SLinus Torvalds while (rb_node) { 14821da177e4SLinus Torvalds struct vm_area_struct * vma_tmp; 14831da177e4SLinus Torvalds 14841da177e4SLinus Torvalds vma_tmp = rb_entry(rb_node, 14851da177e4SLinus Torvalds struct vm_area_struct, vm_rb); 14861da177e4SLinus Torvalds 14871da177e4SLinus Torvalds if (vma_tmp->vm_end > addr) { 14881da177e4SLinus Torvalds vma = vma_tmp; 14891da177e4SLinus Torvalds if (vma_tmp->vm_start <= addr) 14901da177e4SLinus Torvalds break; 14911da177e4SLinus Torvalds rb_node = rb_node->rb_left; 14921da177e4SLinus Torvalds } else 14931da177e4SLinus Torvalds rb_node = rb_node->rb_right; 14941da177e4SLinus Torvalds } 14951da177e4SLinus Torvalds if (vma) 14961da177e4SLinus Torvalds mm->mmap_cache = vma; 14971da177e4SLinus Torvalds } 14981da177e4SLinus Torvalds } 14991da177e4SLinus Torvalds return vma; 15001da177e4SLinus Torvalds } 15011da177e4SLinus Torvalds 15021da177e4SLinus Torvalds EXPORT_SYMBOL(find_vma); 15031da177e4SLinus Torvalds 15041da177e4SLinus Torvalds /* Same as find_vma, but also return a pointer to the previous VMA in *pprev. */ 15051da177e4SLinus Torvalds struct vm_area_struct * 15061da177e4SLinus Torvalds find_vma_prev(struct mm_struct *mm, unsigned long addr, 15071da177e4SLinus Torvalds struct vm_area_struct **pprev) 15081da177e4SLinus Torvalds { 15091da177e4SLinus Torvalds struct vm_area_struct *vma = NULL, *prev = NULL; 15101da177e4SLinus Torvalds struct rb_node *rb_node; 15111da177e4SLinus Torvalds if (!mm) 15121da177e4SLinus Torvalds goto out; 15131da177e4SLinus Torvalds 15141da177e4SLinus Torvalds /* Guard against addr being lower than the first VMA */ 15151da177e4SLinus Torvalds vma = mm->mmap; 15161da177e4SLinus Torvalds 15171da177e4SLinus Torvalds /* Go through the RB tree quickly. */ 15181da177e4SLinus Torvalds rb_node = mm->mm_rb.rb_node; 15191da177e4SLinus Torvalds 15201da177e4SLinus Torvalds while (rb_node) { 15211da177e4SLinus Torvalds struct vm_area_struct *vma_tmp; 15221da177e4SLinus Torvalds vma_tmp = rb_entry(rb_node, struct vm_area_struct, vm_rb); 15231da177e4SLinus Torvalds 15241da177e4SLinus Torvalds if (addr < vma_tmp->vm_end) { 15251da177e4SLinus Torvalds rb_node = rb_node->rb_left; 15261da177e4SLinus Torvalds } else { 15271da177e4SLinus Torvalds prev = vma_tmp; 15281da177e4SLinus Torvalds if (!prev->vm_next || (addr < prev->vm_next->vm_end)) 15291da177e4SLinus Torvalds break; 15301da177e4SLinus Torvalds rb_node = rb_node->rb_right; 15311da177e4SLinus Torvalds } 15321da177e4SLinus Torvalds } 15331da177e4SLinus Torvalds 15341da177e4SLinus Torvalds out: 15351da177e4SLinus Torvalds *pprev = prev; 15361da177e4SLinus Torvalds return prev ? prev->vm_next : vma; 15371da177e4SLinus Torvalds } 15381da177e4SLinus Torvalds 15391da177e4SLinus Torvalds /* 15401da177e4SLinus Torvalds * Verify that the stack growth is acceptable and 15411da177e4SLinus Torvalds * update accounting. This is shared with both the 15421da177e4SLinus Torvalds * grow-up and grow-down cases. 15431da177e4SLinus Torvalds */ 15441da177e4SLinus Torvalds static int acct_stack_growth(struct vm_area_struct *vma, unsigned long size, unsigned long grow) 15451da177e4SLinus Torvalds { 15461da177e4SLinus Torvalds struct mm_struct *mm = vma->vm_mm; 15471da177e4SLinus Torvalds struct rlimit *rlim = current->signal->rlim; 15480d59a01bSAdam Litke unsigned long new_start; 15491da177e4SLinus Torvalds 15501da177e4SLinus Torvalds /* address space limit tests */ 1551119f657cSakpm@osdl.org if (!may_expand_vm(mm, grow)) 15521da177e4SLinus Torvalds return -ENOMEM; 15531da177e4SLinus Torvalds 15541da177e4SLinus Torvalds /* Stack limit test */ 15551da177e4SLinus Torvalds if (size > rlim[RLIMIT_STACK].rlim_cur) 15561da177e4SLinus Torvalds return -ENOMEM; 15571da177e4SLinus Torvalds 15581da177e4SLinus Torvalds /* mlock limit tests */ 15591da177e4SLinus Torvalds if (vma->vm_flags & VM_LOCKED) { 15601da177e4SLinus Torvalds unsigned long locked; 15611da177e4SLinus Torvalds unsigned long limit; 15621da177e4SLinus Torvalds locked = mm->locked_vm + grow; 15631da177e4SLinus Torvalds limit = rlim[RLIMIT_MEMLOCK].rlim_cur >> PAGE_SHIFT; 15641da177e4SLinus Torvalds if (locked > limit && !capable(CAP_IPC_LOCK)) 15651da177e4SLinus Torvalds return -ENOMEM; 15661da177e4SLinus Torvalds } 15671da177e4SLinus Torvalds 15680d59a01bSAdam Litke /* Check to ensure the stack will not grow into a hugetlb-only region */ 15690d59a01bSAdam Litke new_start = (vma->vm_flags & VM_GROWSUP) ? vma->vm_start : 15700d59a01bSAdam Litke vma->vm_end - size; 15710d59a01bSAdam Litke if (is_hugepage_only_range(vma->vm_mm, new_start, size)) 15720d59a01bSAdam Litke return -EFAULT; 15730d59a01bSAdam Litke 15741da177e4SLinus Torvalds /* 15751da177e4SLinus Torvalds * Overcommit.. This must be the final test, as it will 15761da177e4SLinus Torvalds * update security statistics. 15771da177e4SLinus Torvalds */ 157805fa199dSHugh Dickins if (security_vm_enough_memory_mm(mm, grow)) 15791da177e4SLinus Torvalds return -ENOMEM; 15801da177e4SLinus Torvalds 15811da177e4SLinus Torvalds /* Ok, everything looks good - let it rip */ 15821da177e4SLinus Torvalds mm->total_vm += grow; 15831da177e4SLinus Torvalds if (vma->vm_flags & VM_LOCKED) 15841da177e4SLinus Torvalds mm->locked_vm += grow; 1585ab50b8edSHugh Dickins vm_stat_account(mm, vma->vm_flags, vma->vm_file, grow); 15861da177e4SLinus Torvalds return 0; 15871da177e4SLinus Torvalds } 15881da177e4SLinus Torvalds 158946dea3d0SHugh Dickins #if defined(CONFIG_STACK_GROWSUP) || defined(CONFIG_IA64) 15901da177e4SLinus Torvalds /* 159146dea3d0SHugh Dickins * PA-RISC uses this for its stack; IA64 for its Register Backing Store. 159246dea3d0SHugh Dickins * vma is the last one with address > vma->vm_end. Have to extend vma. 15931da177e4SLinus Torvalds */ 15949ab88515SMatthew Wilcox #ifndef CONFIG_IA64 1595cb8f488cSDenys Vlasenko static 159646dea3d0SHugh Dickins #endif 159746dea3d0SHugh Dickins int expand_upwards(struct vm_area_struct *vma, unsigned long address) 15981da177e4SLinus Torvalds { 15991da177e4SLinus Torvalds int error; 16001da177e4SLinus Torvalds 16011da177e4SLinus Torvalds if (!(vma->vm_flags & VM_GROWSUP)) 16021da177e4SLinus Torvalds return -EFAULT; 16031da177e4SLinus Torvalds 16041da177e4SLinus Torvalds /* 16051da177e4SLinus Torvalds * We must make sure the anon_vma is allocated 16061da177e4SLinus Torvalds * so that the anon_vma locking is not a noop. 16071da177e4SLinus Torvalds */ 16081da177e4SLinus Torvalds if (unlikely(anon_vma_prepare(vma))) 16091da177e4SLinus Torvalds return -ENOMEM; 16101da177e4SLinus Torvalds anon_vma_lock(vma); 16111da177e4SLinus Torvalds 16121da177e4SLinus Torvalds /* 16131da177e4SLinus Torvalds * vma->vm_start/vm_end cannot change under us because the caller 16141da177e4SLinus Torvalds * is required to hold the mmap_sem in read mode. We need the 16151da177e4SLinus Torvalds * anon_vma lock to serialize against concurrent expand_stacks. 161606b32f3aSHelge Deller * Also guard against wrapping around to address 0. 16171da177e4SLinus Torvalds */ 161806b32f3aSHelge Deller if (address < PAGE_ALIGN(address+4)) 161906b32f3aSHelge Deller address = PAGE_ALIGN(address+4); 162006b32f3aSHelge Deller else { 162106b32f3aSHelge Deller anon_vma_unlock(vma); 162206b32f3aSHelge Deller return -ENOMEM; 162306b32f3aSHelge Deller } 16241da177e4SLinus Torvalds error = 0; 16251da177e4SLinus Torvalds 16261da177e4SLinus Torvalds /* Somebody else might have raced and expanded it already */ 16271da177e4SLinus Torvalds if (address > vma->vm_end) { 16281da177e4SLinus Torvalds unsigned long size, grow; 16291da177e4SLinus Torvalds 16301da177e4SLinus Torvalds size = address - vma->vm_start; 16311da177e4SLinus Torvalds grow = (address - vma->vm_end) >> PAGE_SHIFT; 16321da177e4SLinus Torvalds 16331da177e4SLinus Torvalds error = acct_stack_growth(vma, size, grow); 16341da177e4SLinus Torvalds if (!error) 16351da177e4SLinus Torvalds vma->vm_end = address; 16361da177e4SLinus Torvalds } 16371da177e4SLinus Torvalds anon_vma_unlock(vma); 16381da177e4SLinus Torvalds return error; 16391da177e4SLinus Torvalds } 164046dea3d0SHugh Dickins #endif /* CONFIG_STACK_GROWSUP || CONFIG_IA64 */ 164146dea3d0SHugh Dickins 16421da177e4SLinus Torvalds /* 16431da177e4SLinus Torvalds * vma is the first one with address < vma->vm_start. Have to extend vma. 16441da177e4SLinus Torvalds */ 1645cb8f488cSDenys Vlasenko static int expand_downwards(struct vm_area_struct *vma, 1646b6a2fea3SOllie Wild unsigned long address) 16471da177e4SLinus Torvalds { 16481da177e4SLinus Torvalds int error; 16491da177e4SLinus Torvalds 16501da177e4SLinus Torvalds /* 16511da177e4SLinus Torvalds * We must make sure the anon_vma is allocated 16521da177e4SLinus Torvalds * so that the anon_vma locking is not a noop. 16531da177e4SLinus Torvalds */ 16541da177e4SLinus Torvalds if (unlikely(anon_vma_prepare(vma))) 16551da177e4SLinus Torvalds return -ENOMEM; 16568869477aSEric Paris 16578869477aSEric Paris address &= PAGE_MASK; 165888c3f7a8SRichard Knutsson error = security_file_mmap(NULL, 0, 0, 0, address, 1); 16598869477aSEric Paris if (error) 16608869477aSEric Paris return error; 16618869477aSEric Paris 16621da177e4SLinus Torvalds anon_vma_lock(vma); 16631da177e4SLinus Torvalds 16641da177e4SLinus Torvalds /* 16651da177e4SLinus Torvalds * vma->vm_start/vm_end cannot change under us because the caller 16661da177e4SLinus Torvalds * is required to hold the mmap_sem in read mode. We need the 16671da177e4SLinus Torvalds * anon_vma lock to serialize against concurrent expand_stacks. 16681da177e4SLinus Torvalds */ 16691da177e4SLinus Torvalds 16701da177e4SLinus Torvalds /* Somebody else might have raced and expanded it already */ 16711da177e4SLinus Torvalds if (address < vma->vm_start) { 16721da177e4SLinus Torvalds unsigned long size, grow; 16731da177e4SLinus Torvalds 16741da177e4SLinus Torvalds size = vma->vm_end - address; 16751da177e4SLinus Torvalds grow = (vma->vm_start - address) >> PAGE_SHIFT; 16761da177e4SLinus Torvalds 16771da177e4SLinus Torvalds error = acct_stack_growth(vma, size, grow); 16781da177e4SLinus Torvalds if (!error) { 16791da177e4SLinus Torvalds vma->vm_start = address; 16801da177e4SLinus Torvalds vma->vm_pgoff -= grow; 16811da177e4SLinus Torvalds } 16821da177e4SLinus Torvalds } 16831da177e4SLinus Torvalds anon_vma_unlock(vma); 16841da177e4SLinus Torvalds return error; 16851da177e4SLinus Torvalds } 16861da177e4SLinus Torvalds 1687b6a2fea3SOllie Wild int expand_stack_downwards(struct vm_area_struct *vma, unsigned long address) 1688b6a2fea3SOllie Wild { 1689b6a2fea3SOllie Wild return expand_downwards(vma, address); 1690b6a2fea3SOllie Wild } 1691b6a2fea3SOllie Wild 1692b6a2fea3SOllie Wild #ifdef CONFIG_STACK_GROWSUP 1693b6a2fea3SOllie Wild int expand_stack(struct vm_area_struct *vma, unsigned long address) 1694b6a2fea3SOllie Wild { 1695b6a2fea3SOllie Wild return expand_upwards(vma, address); 1696b6a2fea3SOllie Wild } 1697b6a2fea3SOllie Wild 1698b6a2fea3SOllie Wild struct vm_area_struct * 1699b6a2fea3SOllie Wild find_extend_vma(struct mm_struct *mm, unsigned long addr) 1700b6a2fea3SOllie Wild { 1701b6a2fea3SOllie Wild struct vm_area_struct *vma, *prev; 1702b6a2fea3SOllie Wild 1703b6a2fea3SOllie Wild addr &= PAGE_MASK; 1704b6a2fea3SOllie Wild vma = find_vma_prev(mm, addr, &prev); 1705b6a2fea3SOllie Wild if (vma && (vma->vm_start <= addr)) 1706b6a2fea3SOllie Wild return vma; 17071c127185SDenys Vlasenko if (!prev || expand_stack(prev, addr)) 1708b6a2fea3SOllie Wild return NULL; 1709ba470de4SRik van Riel if (prev->vm_flags & VM_LOCKED) { 1710ba470de4SRik van Riel if (mlock_vma_pages_range(prev, addr, prev->vm_end) < 0) 1711ba470de4SRik van Riel return NULL; /* vma gone! */ 1712ba470de4SRik van Riel } 1713b6a2fea3SOllie Wild return prev; 1714b6a2fea3SOllie Wild } 1715b6a2fea3SOllie Wild #else 1716b6a2fea3SOllie Wild int expand_stack(struct vm_area_struct *vma, unsigned long address) 1717b6a2fea3SOllie Wild { 1718b6a2fea3SOllie Wild return expand_downwards(vma, address); 1719b6a2fea3SOllie Wild } 1720b6a2fea3SOllie Wild 17211da177e4SLinus Torvalds struct vm_area_struct * 17221da177e4SLinus Torvalds find_extend_vma(struct mm_struct * mm, unsigned long addr) 17231da177e4SLinus Torvalds { 17241da177e4SLinus Torvalds struct vm_area_struct * vma; 17251da177e4SLinus Torvalds unsigned long start; 17261da177e4SLinus Torvalds 17271da177e4SLinus Torvalds addr &= PAGE_MASK; 17281da177e4SLinus Torvalds vma = find_vma(mm,addr); 17291da177e4SLinus Torvalds if (!vma) 17301da177e4SLinus Torvalds return NULL; 17311da177e4SLinus Torvalds if (vma->vm_start <= addr) 17321da177e4SLinus Torvalds return vma; 17331da177e4SLinus Torvalds if (!(vma->vm_flags & VM_GROWSDOWN)) 17341da177e4SLinus Torvalds return NULL; 17351da177e4SLinus Torvalds start = vma->vm_start; 17361da177e4SLinus Torvalds if (expand_stack(vma, addr)) 17371da177e4SLinus Torvalds return NULL; 1738ba470de4SRik van Riel if (vma->vm_flags & VM_LOCKED) { 1739ba470de4SRik van Riel if (mlock_vma_pages_range(vma, addr, start) < 0) 1740ba470de4SRik van Riel return NULL; /* vma gone! */ 1741ba470de4SRik van Riel } 17421da177e4SLinus Torvalds return vma; 17431da177e4SLinus Torvalds } 17441da177e4SLinus Torvalds #endif 17451da177e4SLinus Torvalds 17462c0b3814SHugh Dickins /* 17472c0b3814SHugh Dickins * Ok - we have the memory areas we should free on the vma list, 17482c0b3814SHugh Dickins * so release them, and do the vma updates. 17491da177e4SLinus Torvalds * 17502c0b3814SHugh Dickins * Called with the mm semaphore held. 17511da177e4SLinus Torvalds */ 17522c0b3814SHugh Dickins static void remove_vma_list(struct mm_struct *mm, struct vm_area_struct *vma) 17531da177e4SLinus Torvalds { 1754365e9c87SHugh Dickins /* Update high watermark before we lower total_vm */ 1755365e9c87SHugh Dickins update_hiwater_vm(mm); 17562c0b3814SHugh Dickins do { 1757ab50b8edSHugh Dickins long nrpages = vma_pages(vma); 17581da177e4SLinus Torvalds 1759ab50b8edSHugh Dickins mm->total_vm -= nrpages; 1760ab50b8edSHugh Dickins vm_stat_account(mm, vma->vm_flags, vma->vm_file, -nrpages); 1761a8fb5618SHugh Dickins vma = remove_vma(vma); 1762146425a3SHugh Dickins } while (vma); 17631da177e4SLinus Torvalds validate_mm(mm); 17641da177e4SLinus Torvalds } 17651da177e4SLinus Torvalds 17661da177e4SLinus Torvalds /* 17671da177e4SLinus Torvalds * Get rid of page table information in the indicated region. 17681da177e4SLinus Torvalds * 1769f10df686SPaolo 'Blaisorblade' Giarrusso * Called with the mm semaphore held. 17701da177e4SLinus Torvalds */ 17711da177e4SLinus Torvalds static void unmap_region(struct mm_struct *mm, 1772e0da382cSHugh Dickins struct vm_area_struct *vma, struct vm_area_struct *prev, 1773e0da382cSHugh Dickins unsigned long start, unsigned long end) 17741da177e4SLinus Torvalds { 1775e0da382cSHugh Dickins struct vm_area_struct *next = prev? prev->vm_next: mm->mmap; 17761da177e4SLinus Torvalds struct mmu_gather *tlb; 17771da177e4SLinus Torvalds unsigned long nr_accounted = 0; 17781da177e4SLinus Torvalds 17791da177e4SLinus Torvalds lru_add_drain(); 17801da177e4SLinus Torvalds tlb = tlb_gather_mmu(mm, 0); 1781365e9c87SHugh Dickins update_hiwater_rss(mm); 1782508034a3SHugh Dickins unmap_vmas(&tlb, vma, start, end, &nr_accounted, NULL); 17831da177e4SLinus Torvalds vm_unacct_memory(nr_accounted); 178442b77728SJan Beulich free_pgtables(tlb, vma, prev? prev->vm_end: FIRST_USER_ADDRESS, 1785e0da382cSHugh Dickins next? next->vm_start: 0); 17861da177e4SLinus Torvalds tlb_finish_mmu(tlb, start, end); 17871da177e4SLinus Torvalds } 17881da177e4SLinus Torvalds 17891da177e4SLinus Torvalds /* 17901da177e4SLinus Torvalds * Create a list of vma's touched by the unmap, removing them from the mm's 17911da177e4SLinus Torvalds * vma list as we go.. 17921da177e4SLinus Torvalds */ 17931da177e4SLinus Torvalds static void 17941da177e4SLinus Torvalds detach_vmas_to_be_unmapped(struct mm_struct *mm, struct vm_area_struct *vma, 17951da177e4SLinus Torvalds struct vm_area_struct *prev, unsigned long end) 17961da177e4SLinus Torvalds { 17971da177e4SLinus Torvalds struct vm_area_struct **insertion_point; 17981da177e4SLinus Torvalds struct vm_area_struct *tail_vma = NULL; 17991363c3cdSWolfgang Wander unsigned long addr; 18001da177e4SLinus Torvalds 18011da177e4SLinus Torvalds insertion_point = (prev ? &prev->vm_next : &mm->mmap); 18021da177e4SLinus Torvalds do { 18031da177e4SLinus Torvalds rb_erase(&vma->vm_rb, &mm->mm_rb); 18041da177e4SLinus Torvalds mm->map_count--; 18051da177e4SLinus Torvalds tail_vma = vma; 18061da177e4SLinus Torvalds vma = vma->vm_next; 18071da177e4SLinus Torvalds } while (vma && vma->vm_start < end); 18081da177e4SLinus Torvalds *insertion_point = vma; 18091da177e4SLinus Torvalds tail_vma->vm_next = NULL; 18101363c3cdSWolfgang Wander if (mm->unmap_area == arch_unmap_area) 18111363c3cdSWolfgang Wander addr = prev ? prev->vm_end : mm->mmap_base; 18121363c3cdSWolfgang Wander else 18131363c3cdSWolfgang Wander addr = vma ? vma->vm_start : mm->mmap_base; 18141363c3cdSWolfgang Wander mm->unmap_area(mm, addr); 18151da177e4SLinus Torvalds mm->mmap_cache = NULL; /* Kill the cache. */ 18161da177e4SLinus Torvalds } 18171da177e4SLinus Torvalds 18181da177e4SLinus Torvalds /* 18191da177e4SLinus Torvalds * Split a vma into two pieces at address 'addr', a new vma is allocated 182059c51591SMichael Opdenacker * either for the first part or the tail. 18211da177e4SLinus Torvalds */ 18221da177e4SLinus Torvalds int split_vma(struct mm_struct * mm, struct vm_area_struct * vma, 18231da177e4SLinus Torvalds unsigned long addr, int new_below) 18241da177e4SLinus Torvalds { 18251da177e4SLinus Torvalds struct mempolicy *pol; 18261da177e4SLinus Torvalds struct vm_area_struct *new; 18271da177e4SLinus Torvalds 1828a5516438SAndi Kleen if (is_vm_hugetlb_page(vma) && (addr & 1829a5516438SAndi Kleen ~(huge_page_mask(hstate_vma(vma))))) 18301da177e4SLinus Torvalds return -EINVAL; 18311da177e4SLinus Torvalds 18321da177e4SLinus Torvalds if (mm->map_count >= sysctl_max_map_count) 18331da177e4SLinus Torvalds return -ENOMEM; 18341da177e4SLinus Torvalds 1835e94b1766SChristoph Lameter new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL); 18361da177e4SLinus Torvalds if (!new) 18371da177e4SLinus Torvalds return -ENOMEM; 18381da177e4SLinus Torvalds 18391da177e4SLinus Torvalds /* most fields are the same, copy all, and then fixup */ 18401da177e4SLinus Torvalds *new = *vma; 18411da177e4SLinus Torvalds 18421da177e4SLinus Torvalds if (new_below) 18431da177e4SLinus Torvalds new->vm_end = addr; 18441da177e4SLinus Torvalds else { 18451da177e4SLinus Torvalds new->vm_start = addr; 18461da177e4SLinus Torvalds new->vm_pgoff += ((addr - vma->vm_start) >> PAGE_SHIFT); 18471da177e4SLinus Torvalds } 18481da177e4SLinus Torvalds 1849846a16bfSLee Schermerhorn pol = mpol_dup(vma_policy(vma)); 18501da177e4SLinus Torvalds if (IS_ERR(pol)) { 18511da177e4SLinus Torvalds kmem_cache_free(vm_area_cachep, new); 18521da177e4SLinus Torvalds return PTR_ERR(pol); 18531da177e4SLinus Torvalds } 18541da177e4SLinus Torvalds vma_set_policy(new, pol); 18551da177e4SLinus Torvalds 1856925d1c40SMatt Helsley if (new->vm_file) { 18571da177e4SLinus Torvalds get_file(new->vm_file); 1858925d1c40SMatt Helsley if (vma->vm_flags & VM_EXECUTABLE) 1859925d1c40SMatt Helsley added_exe_file_vma(mm); 1860925d1c40SMatt Helsley } 18611da177e4SLinus Torvalds 18621da177e4SLinus Torvalds if (new->vm_ops && new->vm_ops->open) 18631da177e4SLinus Torvalds new->vm_ops->open(new); 18641da177e4SLinus Torvalds 18651da177e4SLinus Torvalds if (new_below) 18661da177e4SLinus Torvalds vma_adjust(vma, addr, vma->vm_end, vma->vm_pgoff + 18671da177e4SLinus Torvalds ((addr - new->vm_start) >> PAGE_SHIFT), new); 18681da177e4SLinus Torvalds else 18691da177e4SLinus Torvalds vma_adjust(vma, vma->vm_start, addr, vma->vm_pgoff, new); 18701da177e4SLinus Torvalds 18711da177e4SLinus Torvalds return 0; 18721da177e4SLinus Torvalds } 18731da177e4SLinus Torvalds 18741da177e4SLinus Torvalds /* Munmap is split into 2 main parts -- this part which finds 18751da177e4SLinus Torvalds * what needs doing, and the areas themselves, which do the 18761da177e4SLinus Torvalds * work. This now handles partial unmappings. 18771da177e4SLinus Torvalds * Jeremy Fitzhardinge <jeremy@goop.org> 18781da177e4SLinus Torvalds */ 18791da177e4SLinus Torvalds int do_munmap(struct mm_struct *mm, unsigned long start, size_t len) 18801da177e4SLinus Torvalds { 18811da177e4SLinus Torvalds unsigned long end; 1882146425a3SHugh Dickins struct vm_area_struct *vma, *prev, *last; 18831da177e4SLinus Torvalds 18841da177e4SLinus Torvalds if ((start & ~PAGE_MASK) || start > TASK_SIZE || len > TASK_SIZE-start) 18851da177e4SLinus Torvalds return -EINVAL; 18861da177e4SLinus Torvalds 18871da177e4SLinus Torvalds if ((len = PAGE_ALIGN(len)) == 0) 18881da177e4SLinus Torvalds return -EINVAL; 18891da177e4SLinus Torvalds 18901da177e4SLinus Torvalds /* Find the first overlapping VMA */ 1891146425a3SHugh Dickins vma = find_vma_prev(mm, start, &prev); 1892146425a3SHugh Dickins if (!vma) 18931da177e4SLinus Torvalds return 0; 1894146425a3SHugh Dickins /* we have start < vma->vm_end */ 18951da177e4SLinus Torvalds 18961da177e4SLinus Torvalds /* if it doesn't overlap, we have nothing.. */ 18971da177e4SLinus Torvalds end = start + len; 1898146425a3SHugh Dickins if (vma->vm_start >= end) 18991da177e4SLinus Torvalds return 0; 19001da177e4SLinus Torvalds 19011da177e4SLinus Torvalds /* 19021da177e4SLinus Torvalds * If we need to split any vma, do it now to save pain later. 19031da177e4SLinus Torvalds * 19041da177e4SLinus Torvalds * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially 19051da177e4SLinus Torvalds * unmapped vm_area_struct will remain in use: so lower split_vma 19061da177e4SLinus Torvalds * places tmp vma above, and higher split_vma places tmp vma below. 19071da177e4SLinus Torvalds */ 1908146425a3SHugh Dickins if (start > vma->vm_start) { 1909146425a3SHugh Dickins int error = split_vma(mm, vma, start, 0); 19101da177e4SLinus Torvalds if (error) 19111da177e4SLinus Torvalds return error; 1912146425a3SHugh Dickins prev = vma; 19131da177e4SLinus Torvalds } 19141da177e4SLinus Torvalds 19151da177e4SLinus Torvalds /* Does it split the last one? */ 19161da177e4SLinus Torvalds last = find_vma(mm, end); 19171da177e4SLinus Torvalds if (last && end > last->vm_start) { 19181da177e4SLinus Torvalds int error = split_vma(mm, last, end, 1); 19191da177e4SLinus Torvalds if (error) 19201da177e4SLinus Torvalds return error; 19211da177e4SLinus Torvalds } 1922146425a3SHugh Dickins vma = prev? prev->vm_next: mm->mmap; 19231da177e4SLinus Torvalds 19241da177e4SLinus Torvalds /* 1925ba470de4SRik van Riel * unlock any mlock()ed ranges before detaching vmas 1926ba470de4SRik van Riel */ 1927ba470de4SRik van Riel if (mm->locked_vm) { 1928ba470de4SRik van Riel struct vm_area_struct *tmp = vma; 1929ba470de4SRik van Riel while (tmp && tmp->vm_start < end) { 1930ba470de4SRik van Riel if (tmp->vm_flags & VM_LOCKED) { 1931ba470de4SRik van Riel mm->locked_vm -= vma_pages(tmp); 1932ba470de4SRik van Riel munlock_vma_pages_all(tmp); 1933ba470de4SRik van Riel } 1934ba470de4SRik van Riel tmp = tmp->vm_next; 1935ba470de4SRik van Riel } 1936ba470de4SRik van Riel } 1937ba470de4SRik van Riel 1938ba470de4SRik van Riel /* 19391da177e4SLinus Torvalds * Remove the vma's, and unmap the actual pages 19401da177e4SLinus Torvalds */ 1941146425a3SHugh Dickins detach_vmas_to_be_unmapped(mm, vma, prev, end); 1942146425a3SHugh Dickins unmap_region(mm, vma, prev, start, end); 19431da177e4SLinus Torvalds 19441da177e4SLinus Torvalds /* Fix up all other VM information */ 19452c0b3814SHugh Dickins remove_vma_list(mm, vma); 19461da177e4SLinus Torvalds 19471da177e4SLinus Torvalds return 0; 19481da177e4SLinus Torvalds } 19491da177e4SLinus Torvalds 19501da177e4SLinus Torvalds EXPORT_SYMBOL(do_munmap); 19511da177e4SLinus Torvalds 19526a6160a7SHeiko Carstens SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len) 19531da177e4SLinus Torvalds { 19541da177e4SLinus Torvalds int ret; 19551da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 19561da177e4SLinus Torvalds 19571da177e4SLinus Torvalds profile_munmap(addr); 19581da177e4SLinus Torvalds 19591da177e4SLinus Torvalds down_write(&mm->mmap_sem); 19601da177e4SLinus Torvalds ret = do_munmap(mm, addr, len); 19611da177e4SLinus Torvalds up_write(&mm->mmap_sem); 19621da177e4SLinus Torvalds return ret; 19631da177e4SLinus Torvalds } 19641da177e4SLinus Torvalds 19651da177e4SLinus Torvalds static inline void verify_mm_writelocked(struct mm_struct *mm) 19661da177e4SLinus Torvalds { 1967a241ec65SPaul E. McKenney #ifdef CONFIG_DEBUG_VM 19681da177e4SLinus Torvalds if (unlikely(down_read_trylock(&mm->mmap_sem))) { 19691da177e4SLinus Torvalds WARN_ON(1); 19701da177e4SLinus Torvalds up_read(&mm->mmap_sem); 19711da177e4SLinus Torvalds } 19721da177e4SLinus Torvalds #endif 19731da177e4SLinus Torvalds } 19741da177e4SLinus Torvalds 19751da177e4SLinus Torvalds /* 19761da177e4SLinus Torvalds * this is really a simplified "do_mmap". it only handles 19771da177e4SLinus Torvalds * anonymous maps. eventually we may be able to do some 19781da177e4SLinus Torvalds * brk-specific accounting here. 19791da177e4SLinus Torvalds */ 19801da177e4SLinus Torvalds unsigned long do_brk(unsigned long addr, unsigned long len) 19811da177e4SLinus Torvalds { 19821da177e4SLinus Torvalds struct mm_struct * mm = current->mm; 19831da177e4SLinus Torvalds struct vm_area_struct * vma, * prev; 19841da177e4SLinus Torvalds unsigned long flags; 19851da177e4SLinus Torvalds struct rb_node ** rb_link, * rb_parent; 19861da177e4SLinus Torvalds pgoff_t pgoff = addr >> PAGE_SHIFT; 19873a459756SKirill Korotaev int error; 19881da177e4SLinus Torvalds 19891da177e4SLinus Torvalds len = PAGE_ALIGN(len); 19901da177e4SLinus Torvalds if (!len) 19911da177e4SLinus Torvalds return addr; 19921da177e4SLinus Torvalds 19931da177e4SLinus Torvalds if ((addr + len) > TASK_SIZE || (addr + len) < addr) 19941da177e4SLinus Torvalds return -EINVAL; 19951da177e4SLinus Torvalds 1996cd2579d7SHugh Dickins if (is_hugepage_only_range(mm, addr, len)) 1997cd2579d7SHugh Dickins return -EINVAL; 1998cb07c9a1SDavid Gibson 199988c3f7a8SRichard Knutsson error = security_file_mmap(NULL, 0, 0, 0, addr, 1); 20005a211a5dSEric Paris if (error) 20015a211a5dSEric Paris return error; 20025a211a5dSEric Paris 20033a459756SKirill Korotaev flags = VM_DATA_DEFAULT_FLAGS | VM_ACCOUNT | mm->def_flags; 20043a459756SKirill Korotaev 20053a459756SKirill Korotaev error = arch_mmap_check(addr, len, flags); 20063a459756SKirill Korotaev if (error) 20073a459756SKirill Korotaev return error; 20083a459756SKirill Korotaev 20091da177e4SLinus Torvalds /* 20101da177e4SLinus Torvalds * mlock MCL_FUTURE? 20111da177e4SLinus Torvalds */ 20121da177e4SLinus Torvalds if (mm->def_flags & VM_LOCKED) { 20131da177e4SLinus Torvalds unsigned long locked, lock_limit; 201493ea1d0aSChris Wright locked = len >> PAGE_SHIFT; 201593ea1d0aSChris Wright locked += mm->locked_vm; 20161da177e4SLinus Torvalds lock_limit = current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur; 201793ea1d0aSChris Wright lock_limit >>= PAGE_SHIFT; 20181da177e4SLinus Torvalds if (locked > lock_limit && !capable(CAP_IPC_LOCK)) 20191da177e4SLinus Torvalds return -EAGAIN; 20201da177e4SLinus Torvalds } 20211da177e4SLinus Torvalds 20221da177e4SLinus Torvalds /* 20231da177e4SLinus Torvalds * mm->mmap_sem is required to protect against another thread 20241da177e4SLinus Torvalds * changing the mappings in case we sleep. 20251da177e4SLinus Torvalds */ 20261da177e4SLinus Torvalds verify_mm_writelocked(mm); 20271da177e4SLinus Torvalds 20281da177e4SLinus Torvalds /* 20291da177e4SLinus Torvalds * Clear old maps. this also does some error checking for us 20301da177e4SLinus Torvalds */ 20311da177e4SLinus Torvalds munmap_back: 20321da177e4SLinus Torvalds vma = find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent); 20331da177e4SLinus Torvalds if (vma && vma->vm_start < addr + len) { 20341da177e4SLinus Torvalds if (do_munmap(mm, addr, len)) 20351da177e4SLinus Torvalds return -ENOMEM; 20361da177e4SLinus Torvalds goto munmap_back; 20371da177e4SLinus Torvalds } 20381da177e4SLinus Torvalds 20391da177e4SLinus Torvalds /* Check against address space limits *after* clearing old maps... */ 2040119f657cSakpm@osdl.org if (!may_expand_vm(mm, len >> PAGE_SHIFT)) 20411da177e4SLinus Torvalds return -ENOMEM; 20421da177e4SLinus Torvalds 20431da177e4SLinus Torvalds if (mm->map_count > sysctl_max_map_count) 20441da177e4SLinus Torvalds return -ENOMEM; 20451da177e4SLinus Torvalds 20461da177e4SLinus Torvalds if (security_vm_enough_memory(len >> PAGE_SHIFT)) 20471da177e4SLinus Torvalds return -ENOMEM; 20481da177e4SLinus Torvalds 20491da177e4SLinus Torvalds /* Can we just expand an old private anonymous mapping? */ 2050ba470de4SRik van Riel vma = vma_merge(mm, prev, addr, addr + len, flags, 2051ba470de4SRik van Riel NULL, NULL, pgoff, NULL); 2052ba470de4SRik van Riel if (vma) 20531da177e4SLinus Torvalds goto out; 20541da177e4SLinus Torvalds 20551da177e4SLinus Torvalds /* 20561da177e4SLinus Torvalds * create a vma struct for an anonymous mapping 20571da177e4SLinus Torvalds */ 2058c5e3b83eSPekka Enberg vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL); 20591da177e4SLinus Torvalds if (!vma) { 20601da177e4SLinus Torvalds vm_unacct_memory(len >> PAGE_SHIFT); 20611da177e4SLinus Torvalds return -ENOMEM; 20621da177e4SLinus Torvalds } 20631da177e4SLinus Torvalds 20641da177e4SLinus Torvalds vma->vm_mm = mm; 20651da177e4SLinus Torvalds vma->vm_start = addr; 20661da177e4SLinus Torvalds vma->vm_end = addr + len; 20671da177e4SLinus Torvalds vma->vm_pgoff = pgoff; 20681da177e4SLinus Torvalds vma->vm_flags = flags; 20693ed75eb8SColy Li vma->vm_page_prot = vm_get_page_prot(flags); 20701da177e4SLinus Torvalds vma_link(mm, vma, prev, rb_link, rb_parent); 20711da177e4SLinus Torvalds out: 20721da177e4SLinus Torvalds mm->total_vm += len >> PAGE_SHIFT; 20731da177e4SLinus Torvalds if (flags & VM_LOCKED) { 2074ba470de4SRik van Riel if (!mlock_vma_pages_range(vma, addr, addr + len)) 2075ba470de4SRik van Riel mm->locked_vm += (len >> PAGE_SHIFT); 20761da177e4SLinus Torvalds } 20771da177e4SLinus Torvalds return addr; 20781da177e4SLinus Torvalds } 20791da177e4SLinus Torvalds 20801da177e4SLinus Torvalds EXPORT_SYMBOL(do_brk); 20811da177e4SLinus Torvalds 20821da177e4SLinus Torvalds /* Release all mmaps. */ 20831da177e4SLinus Torvalds void exit_mmap(struct mm_struct *mm) 20841da177e4SLinus Torvalds { 20851da177e4SLinus Torvalds struct mmu_gather *tlb; 2086ba470de4SRik van Riel struct vm_area_struct *vma; 20871da177e4SLinus Torvalds unsigned long nr_accounted = 0; 2088ee39b37bSHugh Dickins unsigned long end; 20891da177e4SLinus Torvalds 2090d6dd61c8SJeremy Fitzhardinge /* mm's last user has gone, and its about to be pulled down */ 2091cddb8a5cSAndrea Arcangeli mmu_notifier_release(mm); 2092d6dd61c8SJeremy Fitzhardinge 2093ba470de4SRik van Riel if (mm->locked_vm) { 2094ba470de4SRik van Riel vma = mm->mmap; 2095ba470de4SRik van Riel while (vma) { 2096ba470de4SRik van Riel if (vma->vm_flags & VM_LOCKED) 2097ba470de4SRik van Riel munlock_vma_pages_all(vma); 2098ba470de4SRik van Riel vma = vma->vm_next; 2099ba470de4SRik van Riel } 2100ba470de4SRik van Riel } 21019480c53eSJeremy Fitzhardinge 21029480c53eSJeremy Fitzhardinge arch_exit_mmap(mm); 21039480c53eSJeremy Fitzhardinge 2104ba470de4SRik van Riel vma = mm->mmap; 21059480c53eSJeremy Fitzhardinge if (!vma) /* Can happen if dup_mmap() received an OOM */ 21069480c53eSJeremy Fitzhardinge return; 21079480c53eSJeremy Fitzhardinge 21081da177e4SLinus Torvalds lru_add_drain(); 21091da177e4SLinus Torvalds flush_cache_mm(mm); 2110e0da382cSHugh Dickins tlb = tlb_gather_mmu(mm, 1); 2111901608d9SOleg Nesterov /* update_hiwater_rss(mm) here? but nobody should be looking */ 2112e0da382cSHugh Dickins /* Use -1 here to ensure all VMAs in the mm are unmapped */ 2113508034a3SHugh Dickins end = unmap_vmas(&tlb, vma, 0, -1, &nr_accounted, NULL); 21141da177e4SLinus Torvalds vm_unacct_memory(nr_accounted); 2115*9ba69294SHugh Dickins 2116*9ba69294SHugh Dickins /* 2117*9ba69294SHugh Dickins * For KSM to handle OOM without deadlock when it's breaking COW in a 2118*9ba69294SHugh Dickins * likely victim of the OOM killer, we must serialize with ksm_exit() 2119*9ba69294SHugh Dickins * after freeing mm's pages but before freeing its page tables. 2120*9ba69294SHugh Dickins */ 2121*9ba69294SHugh Dickins ksm_exit(mm, &tlb, end); 2122*9ba69294SHugh Dickins 212342b77728SJan Beulich free_pgtables(tlb, vma, FIRST_USER_ADDRESS, 0); 2124ee39b37bSHugh Dickins tlb_finish_mmu(tlb, 0, end); 21251da177e4SLinus Torvalds 21261da177e4SLinus Torvalds /* 21278f4f8c16SHugh Dickins * Walk the list again, actually closing and freeing it, 21288f4f8c16SHugh Dickins * with preemption enabled, without holding any MM locks. 21291da177e4SLinus Torvalds */ 2130a8fb5618SHugh Dickins while (vma) 2131a8fb5618SHugh Dickins vma = remove_vma(vma); 2132e0da382cSHugh Dickins 2133e2cdef8cSHugh Dickins BUG_ON(mm->nr_ptes > (FIRST_USER_ADDRESS+PMD_SIZE-1)>>PMD_SHIFT); 21341da177e4SLinus Torvalds } 21351da177e4SLinus Torvalds 21361da177e4SLinus Torvalds /* Insert vm structure into process list sorted by address 21371da177e4SLinus Torvalds * and into the inode's i_mmap tree. If vm_file is non-NULL 21381da177e4SLinus Torvalds * then i_mmap_lock is taken here. 21391da177e4SLinus Torvalds */ 21401da177e4SLinus Torvalds int insert_vm_struct(struct mm_struct * mm, struct vm_area_struct * vma) 21411da177e4SLinus Torvalds { 21421da177e4SLinus Torvalds struct vm_area_struct * __vma, * prev; 21431da177e4SLinus Torvalds struct rb_node ** rb_link, * rb_parent; 21441da177e4SLinus Torvalds 21451da177e4SLinus Torvalds /* 21461da177e4SLinus Torvalds * The vm_pgoff of a purely anonymous vma should be irrelevant 21471da177e4SLinus Torvalds * until its first write fault, when page's anon_vma and index 21481da177e4SLinus Torvalds * are set. But now set the vm_pgoff it will almost certainly 21491da177e4SLinus Torvalds * end up with (unless mremap moves it elsewhere before that 21501da177e4SLinus Torvalds * first wfault), so /proc/pid/maps tells a consistent story. 21511da177e4SLinus Torvalds * 21521da177e4SLinus Torvalds * By setting it to reflect the virtual start address of the 21531da177e4SLinus Torvalds * vma, merges and splits can happen in a seamless way, just 21541da177e4SLinus Torvalds * using the existing file pgoff checks and manipulations. 21551da177e4SLinus Torvalds * Similarly in do_mmap_pgoff and in do_brk. 21561da177e4SLinus Torvalds */ 21571da177e4SLinus Torvalds if (!vma->vm_file) { 21581da177e4SLinus Torvalds BUG_ON(vma->anon_vma); 21591da177e4SLinus Torvalds vma->vm_pgoff = vma->vm_start >> PAGE_SHIFT; 21601da177e4SLinus Torvalds } 21611da177e4SLinus Torvalds __vma = find_vma_prepare(mm,vma->vm_start,&prev,&rb_link,&rb_parent); 21621da177e4SLinus Torvalds if (__vma && __vma->vm_start < vma->vm_end) 21631da177e4SLinus Torvalds return -ENOMEM; 21642fd4ef85SHugh Dickins if ((vma->vm_flags & VM_ACCOUNT) && 216534b4e4aaSAlan Cox security_vm_enough_memory_mm(mm, vma_pages(vma))) 21662fd4ef85SHugh Dickins return -ENOMEM; 21671da177e4SLinus Torvalds vma_link(mm, vma, prev, rb_link, rb_parent); 21681da177e4SLinus Torvalds return 0; 21691da177e4SLinus Torvalds } 21701da177e4SLinus Torvalds 21711da177e4SLinus Torvalds /* 21721da177e4SLinus Torvalds * Copy the vma structure to a new location in the same mm, 21731da177e4SLinus Torvalds * prior to moving page table entries, to effect an mremap move. 21741da177e4SLinus Torvalds */ 21751da177e4SLinus Torvalds struct vm_area_struct *copy_vma(struct vm_area_struct **vmap, 21761da177e4SLinus Torvalds unsigned long addr, unsigned long len, pgoff_t pgoff) 21771da177e4SLinus Torvalds { 21781da177e4SLinus Torvalds struct vm_area_struct *vma = *vmap; 21791da177e4SLinus Torvalds unsigned long vma_start = vma->vm_start; 21801da177e4SLinus Torvalds struct mm_struct *mm = vma->vm_mm; 21811da177e4SLinus Torvalds struct vm_area_struct *new_vma, *prev; 21821da177e4SLinus Torvalds struct rb_node **rb_link, *rb_parent; 21831da177e4SLinus Torvalds struct mempolicy *pol; 21841da177e4SLinus Torvalds 21851da177e4SLinus Torvalds /* 21861da177e4SLinus Torvalds * If anonymous vma has not yet been faulted, update new pgoff 21871da177e4SLinus Torvalds * to match new location, to increase its chance of merging. 21881da177e4SLinus Torvalds */ 21891da177e4SLinus Torvalds if (!vma->vm_file && !vma->anon_vma) 21901da177e4SLinus Torvalds pgoff = addr >> PAGE_SHIFT; 21911da177e4SLinus Torvalds 21921da177e4SLinus Torvalds find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent); 21931da177e4SLinus Torvalds new_vma = vma_merge(mm, prev, addr, addr + len, vma->vm_flags, 21941da177e4SLinus Torvalds vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma)); 21951da177e4SLinus Torvalds if (new_vma) { 21961da177e4SLinus Torvalds /* 21971da177e4SLinus Torvalds * Source vma may have been merged into new_vma 21981da177e4SLinus Torvalds */ 21991da177e4SLinus Torvalds if (vma_start >= new_vma->vm_start && 22001da177e4SLinus Torvalds vma_start < new_vma->vm_end) 22011da177e4SLinus Torvalds *vmap = new_vma; 22021da177e4SLinus Torvalds } else { 2203e94b1766SChristoph Lameter new_vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL); 22041da177e4SLinus Torvalds if (new_vma) { 22051da177e4SLinus Torvalds *new_vma = *vma; 2206846a16bfSLee Schermerhorn pol = mpol_dup(vma_policy(vma)); 22071da177e4SLinus Torvalds if (IS_ERR(pol)) { 22081da177e4SLinus Torvalds kmem_cache_free(vm_area_cachep, new_vma); 22091da177e4SLinus Torvalds return NULL; 22101da177e4SLinus Torvalds } 22111da177e4SLinus Torvalds vma_set_policy(new_vma, pol); 22121da177e4SLinus Torvalds new_vma->vm_start = addr; 22131da177e4SLinus Torvalds new_vma->vm_end = addr + len; 22141da177e4SLinus Torvalds new_vma->vm_pgoff = pgoff; 2215925d1c40SMatt Helsley if (new_vma->vm_file) { 22161da177e4SLinus Torvalds get_file(new_vma->vm_file); 2217925d1c40SMatt Helsley if (vma->vm_flags & VM_EXECUTABLE) 2218925d1c40SMatt Helsley added_exe_file_vma(mm); 2219925d1c40SMatt Helsley } 22201da177e4SLinus Torvalds if (new_vma->vm_ops && new_vma->vm_ops->open) 22211da177e4SLinus Torvalds new_vma->vm_ops->open(new_vma); 22221da177e4SLinus Torvalds vma_link(mm, new_vma, prev, rb_link, rb_parent); 22231da177e4SLinus Torvalds } 22241da177e4SLinus Torvalds } 22251da177e4SLinus Torvalds return new_vma; 22261da177e4SLinus Torvalds } 2227119f657cSakpm@osdl.org 2228119f657cSakpm@osdl.org /* 2229119f657cSakpm@osdl.org * Return true if the calling process may expand its vm space by the passed 2230119f657cSakpm@osdl.org * number of pages 2231119f657cSakpm@osdl.org */ 2232119f657cSakpm@osdl.org int may_expand_vm(struct mm_struct *mm, unsigned long npages) 2233119f657cSakpm@osdl.org { 2234119f657cSakpm@osdl.org unsigned long cur = mm->total_vm; /* pages */ 2235119f657cSakpm@osdl.org unsigned long lim; 2236119f657cSakpm@osdl.org 2237119f657cSakpm@osdl.org lim = current->signal->rlim[RLIMIT_AS].rlim_cur >> PAGE_SHIFT; 2238119f657cSakpm@osdl.org 2239119f657cSakpm@osdl.org if (cur + npages > lim) 2240119f657cSakpm@osdl.org return 0; 2241119f657cSakpm@osdl.org return 1; 2242119f657cSakpm@osdl.org } 2243fa5dc22fSRoland McGrath 2244fa5dc22fSRoland McGrath 2245b1d0e4f5SNick Piggin static int special_mapping_fault(struct vm_area_struct *vma, 2246b1d0e4f5SNick Piggin struct vm_fault *vmf) 2247fa5dc22fSRoland McGrath { 2248b1d0e4f5SNick Piggin pgoff_t pgoff; 2249fa5dc22fSRoland McGrath struct page **pages; 2250fa5dc22fSRoland McGrath 2251b1d0e4f5SNick Piggin /* 2252b1d0e4f5SNick Piggin * special mappings have no vm_file, and in that case, the mm 2253b1d0e4f5SNick Piggin * uses vm_pgoff internally. So we have to subtract it from here. 2254b1d0e4f5SNick Piggin * We are allowed to do this because we are the mm; do not copy 2255b1d0e4f5SNick Piggin * this code into drivers! 2256b1d0e4f5SNick Piggin */ 2257b1d0e4f5SNick Piggin pgoff = vmf->pgoff - vma->vm_pgoff; 2258fa5dc22fSRoland McGrath 2259b1d0e4f5SNick Piggin for (pages = vma->vm_private_data; pgoff && *pages; ++pages) 2260b1d0e4f5SNick Piggin pgoff--; 2261fa5dc22fSRoland McGrath 2262fa5dc22fSRoland McGrath if (*pages) { 2263fa5dc22fSRoland McGrath struct page *page = *pages; 2264fa5dc22fSRoland McGrath get_page(page); 2265b1d0e4f5SNick Piggin vmf->page = page; 2266b1d0e4f5SNick Piggin return 0; 2267fa5dc22fSRoland McGrath } 2268fa5dc22fSRoland McGrath 2269b1d0e4f5SNick Piggin return VM_FAULT_SIGBUS; 2270fa5dc22fSRoland McGrath } 2271fa5dc22fSRoland McGrath 2272fa5dc22fSRoland McGrath /* 2273fa5dc22fSRoland McGrath * Having a close hook prevents vma merging regardless of flags. 2274fa5dc22fSRoland McGrath */ 2275fa5dc22fSRoland McGrath static void special_mapping_close(struct vm_area_struct *vma) 2276fa5dc22fSRoland McGrath { 2277fa5dc22fSRoland McGrath } 2278fa5dc22fSRoland McGrath 2279fa5dc22fSRoland McGrath static struct vm_operations_struct special_mapping_vmops = { 2280fa5dc22fSRoland McGrath .close = special_mapping_close, 2281b1d0e4f5SNick Piggin .fault = special_mapping_fault, 2282fa5dc22fSRoland McGrath }; 2283fa5dc22fSRoland McGrath 2284fa5dc22fSRoland McGrath /* 2285fa5dc22fSRoland McGrath * Called with mm->mmap_sem held for writing. 2286fa5dc22fSRoland McGrath * Insert a new vma covering the given region, with the given flags. 2287fa5dc22fSRoland McGrath * Its pages are supplied by the given array of struct page *. 2288fa5dc22fSRoland McGrath * The array can be shorter than len >> PAGE_SHIFT if it's null-terminated. 2289fa5dc22fSRoland McGrath * The region past the last page supplied will always produce SIGBUS. 2290fa5dc22fSRoland McGrath * The array pointer and the pages it points to are assumed to stay alive 2291fa5dc22fSRoland McGrath * for as long as this mapping might exist. 2292fa5dc22fSRoland McGrath */ 2293fa5dc22fSRoland McGrath int install_special_mapping(struct mm_struct *mm, 2294fa5dc22fSRoland McGrath unsigned long addr, unsigned long len, 2295fa5dc22fSRoland McGrath unsigned long vm_flags, struct page **pages) 2296fa5dc22fSRoland McGrath { 2297fa5dc22fSRoland McGrath struct vm_area_struct *vma; 2298fa5dc22fSRoland McGrath 2299fa5dc22fSRoland McGrath vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL); 2300fa5dc22fSRoland McGrath if (unlikely(vma == NULL)) 2301fa5dc22fSRoland McGrath return -ENOMEM; 2302fa5dc22fSRoland McGrath 2303fa5dc22fSRoland McGrath vma->vm_mm = mm; 2304fa5dc22fSRoland McGrath vma->vm_start = addr; 2305fa5dc22fSRoland McGrath vma->vm_end = addr + len; 2306fa5dc22fSRoland McGrath 23072f98735cSNick Piggin vma->vm_flags = vm_flags | mm->def_flags | VM_DONTEXPAND; 23083ed75eb8SColy Li vma->vm_page_prot = vm_get_page_prot(vma->vm_flags); 2309fa5dc22fSRoland McGrath 2310fa5dc22fSRoland McGrath vma->vm_ops = &special_mapping_vmops; 2311fa5dc22fSRoland McGrath vma->vm_private_data = pages; 2312fa5dc22fSRoland McGrath 2313fa5dc22fSRoland McGrath if (unlikely(insert_vm_struct(mm, vma))) { 2314fa5dc22fSRoland McGrath kmem_cache_free(vm_area_cachep, vma); 2315fa5dc22fSRoland McGrath return -ENOMEM; 2316fa5dc22fSRoland McGrath } 2317fa5dc22fSRoland McGrath 2318fa5dc22fSRoland McGrath mm->total_vm += len >> PAGE_SHIFT; 2319fa5dc22fSRoland McGrath 2320cdd6c482SIngo Molnar perf_event_mmap(vma); 2321089dd79dSPeter Zijlstra 2322fa5dc22fSRoland McGrath return 0; 2323fa5dc22fSRoland McGrath } 23247906d00cSAndrea Arcangeli 23257906d00cSAndrea Arcangeli static DEFINE_MUTEX(mm_all_locks_mutex); 23267906d00cSAndrea Arcangeli 2327454ed842SPeter Zijlstra static void vm_lock_anon_vma(struct mm_struct *mm, struct anon_vma *anon_vma) 23287906d00cSAndrea Arcangeli { 23297906d00cSAndrea Arcangeli if (!test_bit(0, (unsigned long *) &anon_vma->head.next)) { 23307906d00cSAndrea Arcangeli /* 23317906d00cSAndrea Arcangeli * The LSB of head.next can't change from under us 23327906d00cSAndrea Arcangeli * because we hold the mm_all_locks_mutex. 23337906d00cSAndrea Arcangeli */ 2334454ed842SPeter Zijlstra spin_lock_nest_lock(&anon_vma->lock, &mm->mmap_sem); 23357906d00cSAndrea Arcangeli /* 23367906d00cSAndrea Arcangeli * We can safely modify head.next after taking the 23377906d00cSAndrea Arcangeli * anon_vma->lock. If some other vma in this mm shares 23387906d00cSAndrea Arcangeli * the same anon_vma we won't take it again. 23397906d00cSAndrea Arcangeli * 23407906d00cSAndrea Arcangeli * No need of atomic instructions here, head.next 23417906d00cSAndrea Arcangeli * can't change from under us thanks to the 23427906d00cSAndrea Arcangeli * anon_vma->lock. 23437906d00cSAndrea Arcangeli */ 23447906d00cSAndrea Arcangeli if (__test_and_set_bit(0, (unsigned long *) 23457906d00cSAndrea Arcangeli &anon_vma->head.next)) 23467906d00cSAndrea Arcangeli BUG(); 23477906d00cSAndrea Arcangeli } 23487906d00cSAndrea Arcangeli } 23497906d00cSAndrea Arcangeli 2350454ed842SPeter Zijlstra static void vm_lock_mapping(struct mm_struct *mm, struct address_space *mapping) 23517906d00cSAndrea Arcangeli { 23527906d00cSAndrea Arcangeli if (!test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) { 23537906d00cSAndrea Arcangeli /* 23547906d00cSAndrea Arcangeli * AS_MM_ALL_LOCKS can't change from under us because 23557906d00cSAndrea Arcangeli * we hold the mm_all_locks_mutex. 23567906d00cSAndrea Arcangeli * 23577906d00cSAndrea Arcangeli * Operations on ->flags have to be atomic because 23587906d00cSAndrea Arcangeli * even if AS_MM_ALL_LOCKS is stable thanks to the 23597906d00cSAndrea Arcangeli * mm_all_locks_mutex, there may be other cpus 23607906d00cSAndrea Arcangeli * changing other bitflags in parallel to us. 23617906d00cSAndrea Arcangeli */ 23627906d00cSAndrea Arcangeli if (test_and_set_bit(AS_MM_ALL_LOCKS, &mapping->flags)) 23637906d00cSAndrea Arcangeli BUG(); 2364454ed842SPeter Zijlstra spin_lock_nest_lock(&mapping->i_mmap_lock, &mm->mmap_sem); 23657906d00cSAndrea Arcangeli } 23667906d00cSAndrea Arcangeli } 23677906d00cSAndrea Arcangeli 23687906d00cSAndrea Arcangeli /* 23697906d00cSAndrea Arcangeli * This operation locks against the VM for all pte/vma/mm related 23707906d00cSAndrea Arcangeli * operations that could ever happen on a certain mm. This includes 23717906d00cSAndrea Arcangeli * vmtruncate, try_to_unmap, and all page faults. 23727906d00cSAndrea Arcangeli * 23737906d00cSAndrea Arcangeli * The caller must take the mmap_sem in write mode before calling 23747906d00cSAndrea Arcangeli * mm_take_all_locks(). The caller isn't allowed to release the 23757906d00cSAndrea Arcangeli * mmap_sem until mm_drop_all_locks() returns. 23767906d00cSAndrea Arcangeli * 23777906d00cSAndrea Arcangeli * mmap_sem in write mode is required in order to block all operations 23787906d00cSAndrea Arcangeli * that could modify pagetables and free pages without need of 23797906d00cSAndrea Arcangeli * altering the vma layout (for example populate_range() with 23807906d00cSAndrea Arcangeli * nonlinear vmas). It's also needed in write mode to avoid new 23817906d00cSAndrea Arcangeli * anon_vmas to be associated with existing vmas. 23827906d00cSAndrea Arcangeli * 23837906d00cSAndrea Arcangeli * A single task can't take more than one mm_take_all_locks() in a row 23847906d00cSAndrea Arcangeli * or it would deadlock. 23857906d00cSAndrea Arcangeli * 23867906d00cSAndrea Arcangeli * The LSB in anon_vma->head.next and the AS_MM_ALL_LOCKS bitflag in 23877906d00cSAndrea Arcangeli * mapping->flags avoid to take the same lock twice, if more than one 23887906d00cSAndrea Arcangeli * vma in this mm is backed by the same anon_vma or address_space. 23897906d00cSAndrea Arcangeli * 23907906d00cSAndrea Arcangeli * We can take all the locks in random order because the VM code 23917906d00cSAndrea Arcangeli * taking i_mmap_lock or anon_vma->lock outside the mmap_sem never 23927906d00cSAndrea Arcangeli * takes more than one of them in a row. Secondly we're protected 23937906d00cSAndrea Arcangeli * against a concurrent mm_take_all_locks() by the mm_all_locks_mutex. 23947906d00cSAndrea Arcangeli * 23957906d00cSAndrea Arcangeli * mm_take_all_locks() and mm_drop_all_locks are expensive operations 23967906d00cSAndrea Arcangeli * that may have to take thousand of locks. 23977906d00cSAndrea Arcangeli * 23987906d00cSAndrea Arcangeli * mm_take_all_locks() can fail if it's interrupted by signals. 23997906d00cSAndrea Arcangeli */ 24007906d00cSAndrea Arcangeli int mm_take_all_locks(struct mm_struct *mm) 24017906d00cSAndrea Arcangeli { 24027906d00cSAndrea Arcangeli struct vm_area_struct *vma; 24037906d00cSAndrea Arcangeli int ret = -EINTR; 24047906d00cSAndrea Arcangeli 24057906d00cSAndrea Arcangeli BUG_ON(down_read_trylock(&mm->mmap_sem)); 24067906d00cSAndrea Arcangeli 24077906d00cSAndrea Arcangeli mutex_lock(&mm_all_locks_mutex); 24087906d00cSAndrea Arcangeli 24097906d00cSAndrea Arcangeli for (vma = mm->mmap; vma; vma = vma->vm_next) { 24107906d00cSAndrea Arcangeli if (signal_pending(current)) 24117906d00cSAndrea Arcangeli goto out_unlock; 24127906d00cSAndrea Arcangeli if (vma->vm_file && vma->vm_file->f_mapping) 2413454ed842SPeter Zijlstra vm_lock_mapping(mm, vma->vm_file->f_mapping); 24147906d00cSAndrea Arcangeli } 24157cd5a02fSPeter Zijlstra 24167cd5a02fSPeter Zijlstra for (vma = mm->mmap; vma; vma = vma->vm_next) { 24177cd5a02fSPeter Zijlstra if (signal_pending(current)) 24187cd5a02fSPeter Zijlstra goto out_unlock; 24197cd5a02fSPeter Zijlstra if (vma->anon_vma) 24207cd5a02fSPeter Zijlstra vm_lock_anon_vma(mm, vma->anon_vma); 24217cd5a02fSPeter Zijlstra } 24227cd5a02fSPeter Zijlstra 24237906d00cSAndrea Arcangeli ret = 0; 24247906d00cSAndrea Arcangeli 24257906d00cSAndrea Arcangeli out_unlock: 24267906d00cSAndrea Arcangeli if (ret) 24277906d00cSAndrea Arcangeli mm_drop_all_locks(mm); 24287906d00cSAndrea Arcangeli 24297906d00cSAndrea Arcangeli return ret; 24307906d00cSAndrea Arcangeli } 24317906d00cSAndrea Arcangeli 24327906d00cSAndrea Arcangeli static void vm_unlock_anon_vma(struct anon_vma *anon_vma) 24337906d00cSAndrea Arcangeli { 24347906d00cSAndrea Arcangeli if (test_bit(0, (unsigned long *) &anon_vma->head.next)) { 24357906d00cSAndrea Arcangeli /* 24367906d00cSAndrea Arcangeli * The LSB of head.next can't change to 0 from under 24377906d00cSAndrea Arcangeli * us because we hold the mm_all_locks_mutex. 24387906d00cSAndrea Arcangeli * 24397906d00cSAndrea Arcangeli * We must however clear the bitflag before unlocking 24407906d00cSAndrea Arcangeli * the vma so the users using the anon_vma->head will 24417906d00cSAndrea Arcangeli * never see our bitflag. 24427906d00cSAndrea Arcangeli * 24437906d00cSAndrea Arcangeli * No need of atomic instructions here, head.next 24447906d00cSAndrea Arcangeli * can't change from under us until we release the 24457906d00cSAndrea Arcangeli * anon_vma->lock. 24467906d00cSAndrea Arcangeli */ 24477906d00cSAndrea Arcangeli if (!__test_and_clear_bit(0, (unsigned long *) 24487906d00cSAndrea Arcangeli &anon_vma->head.next)) 24497906d00cSAndrea Arcangeli BUG(); 24507906d00cSAndrea Arcangeli spin_unlock(&anon_vma->lock); 24517906d00cSAndrea Arcangeli } 24527906d00cSAndrea Arcangeli } 24537906d00cSAndrea Arcangeli 24547906d00cSAndrea Arcangeli static void vm_unlock_mapping(struct address_space *mapping) 24557906d00cSAndrea Arcangeli { 24567906d00cSAndrea Arcangeli if (test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) { 24577906d00cSAndrea Arcangeli /* 24587906d00cSAndrea Arcangeli * AS_MM_ALL_LOCKS can't change to 0 from under us 24597906d00cSAndrea Arcangeli * because we hold the mm_all_locks_mutex. 24607906d00cSAndrea Arcangeli */ 24617906d00cSAndrea Arcangeli spin_unlock(&mapping->i_mmap_lock); 24627906d00cSAndrea Arcangeli if (!test_and_clear_bit(AS_MM_ALL_LOCKS, 24637906d00cSAndrea Arcangeli &mapping->flags)) 24647906d00cSAndrea Arcangeli BUG(); 24657906d00cSAndrea Arcangeli } 24667906d00cSAndrea Arcangeli } 24677906d00cSAndrea Arcangeli 24687906d00cSAndrea Arcangeli /* 24697906d00cSAndrea Arcangeli * The mmap_sem cannot be released by the caller until 24707906d00cSAndrea Arcangeli * mm_drop_all_locks() returns. 24717906d00cSAndrea Arcangeli */ 24727906d00cSAndrea Arcangeli void mm_drop_all_locks(struct mm_struct *mm) 24737906d00cSAndrea Arcangeli { 24747906d00cSAndrea Arcangeli struct vm_area_struct *vma; 24757906d00cSAndrea Arcangeli 24767906d00cSAndrea Arcangeli BUG_ON(down_read_trylock(&mm->mmap_sem)); 24777906d00cSAndrea Arcangeli BUG_ON(!mutex_is_locked(&mm_all_locks_mutex)); 24787906d00cSAndrea Arcangeli 24797906d00cSAndrea Arcangeli for (vma = mm->mmap; vma; vma = vma->vm_next) { 24807906d00cSAndrea Arcangeli if (vma->anon_vma) 24817906d00cSAndrea Arcangeli vm_unlock_anon_vma(vma->anon_vma); 24827906d00cSAndrea Arcangeli if (vma->vm_file && vma->vm_file->f_mapping) 24837906d00cSAndrea Arcangeli vm_unlock_mapping(vma->vm_file->f_mapping); 24847906d00cSAndrea Arcangeli } 24857906d00cSAndrea Arcangeli 24867906d00cSAndrea Arcangeli mutex_unlock(&mm_all_locks_mutex); 24877906d00cSAndrea Arcangeli } 24888feae131SDavid Howells 24898feae131SDavid Howells /* 24908feae131SDavid Howells * initialise the VMA slab 24918feae131SDavid Howells */ 24928feae131SDavid Howells void __init mmap_init(void) 24938feae131SDavid Howells { 249400a62ce9SKOSAKI Motohiro int ret; 249500a62ce9SKOSAKI Motohiro 249600a62ce9SKOSAKI Motohiro ret = percpu_counter_init(&vm_committed_as, 0); 249700a62ce9SKOSAKI Motohiro VM_BUG_ON(ret); 24988feae131SDavid Howells } 2499