11da177e4SLinus Torvalds /* 21da177e4SLinus Torvalds * mm/mmap.c 31da177e4SLinus Torvalds * 41da177e4SLinus Torvalds * Written by obz. 51da177e4SLinus Torvalds * 6046c6884SAlan Cox * Address space accounting code <alan@lxorguk.ukuu.org.uk> 71da177e4SLinus Torvalds */ 81da177e4SLinus Torvalds 91da177e4SLinus Torvalds #include <linux/slab.h> 104af3c9ccSAlexey Dobriyan #include <linux/backing-dev.h> 111da177e4SLinus Torvalds #include <linux/mm.h> 121da177e4SLinus Torvalds #include <linux/shm.h> 131da177e4SLinus Torvalds #include <linux/mman.h> 141da177e4SLinus Torvalds #include <linux/pagemap.h> 151da177e4SLinus Torvalds #include <linux/swap.h> 161da177e4SLinus Torvalds #include <linux/syscalls.h> 17c59ede7bSRandy.Dunlap #include <linux/capability.h> 181da177e4SLinus Torvalds #include <linux/init.h> 191da177e4SLinus Torvalds #include <linux/file.h> 201da177e4SLinus Torvalds #include <linux/fs.h> 211da177e4SLinus Torvalds #include <linux/personality.h> 221da177e4SLinus Torvalds #include <linux/security.h> 231da177e4SLinus Torvalds #include <linux/hugetlb.h> 241da177e4SLinus Torvalds #include <linux/profile.h> 251da177e4SLinus Torvalds #include <linux/module.h> 261da177e4SLinus Torvalds #include <linux/mount.h> 271da177e4SLinus Torvalds #include <linux/mempolicy.h> 281da177e4SLinus Torvalds #include <linux/rmap.h> 29cddb8a5cSAndrea Arcangeli #include <linux/mmu_notifier.h> 30cdd6c482SIngo Molnar #include <linux/perf_event.h> 311da177e4SLinus Torvalds 321da177e4SLinus Torvalds #include <asm/uaccess.h> 331da177e4SLinus Torvalds #include <asm/cacheflush.h> 341da177e4SLinus Torvalds #include <asm/tlb.h> 35d6dd61c8SJeremy Fitzhardinge #include <asm/mmu_context.h> 361da177e4SLinus Torvalds 3742b77728SJan Beulich #include "internal.h" 3842b77728SJan Beulich 393a459756SKirill Korotaev #ifndef arch_mmap_check 403a459756SKirill Korotaev #define arch_mmap_check(addr, len, flags) (0) 413a459756SKirill Korotaev #endif 423a459756SKirill Korotaev 4308e7d9b5SMartin Schwidefsky #ifndef arch_rebalance_pgtables 4408e7d9b5SMartin Schwidefsky #define arch_rebalance_pgtables(addr, len) (addr) 4508e7d9b5SMartin Schwidefsky #endif 4608e7d9b5SMartin Schwidefsky 47e0da382cSHugh Dickins static void unmap_region(struct mm_struct *mm, 48e0da382cSHugh Dickins struct vm_area_struct *vma, struct vm_area_struct *prev, 49e0da382cSHugh Dickins unsigned long start, unsigned long end); 50e0da382cSHugh Dickins 511da177e4SLinus Torvalds /* 521da177e4SLinus Torvalds * WARNING: the debugging will use recursive algorithms so never enable this 531da177e4SLinus Torvalds * unless you know what you are doing. 541da177e4SLinus Torvalds */ 551da177e4SLinus Torvalds #undef DEBUG_MM_RB 561da177e4SLinus Torvalds 571da177e4SLinus Torvalds /* description of effects of mapping type and prot in current implementation. 581da177e4SLinus Torvalds * this is due to the limited x86 page protection hardware. The expected 591da177e4SLinus Torvalds * behavior is in parens: 601da177e4SLinus Torvalds * 611da177e4SLinus Torvalds * map_type prot 621da177e4SLinus Torvalds * PROT_NONE PROT_READ PROT_WRITE PROT_EXEC 631da177e4SLinus Torvalds * MAP_SHARED r: (no) no r: (yes) yes r: (no) yes r: (no) yes 641da177e4SLinus Torvalds * w: (no) no w: (no) no w: (yes) yes w: (no) no 651da177e4SLinus Torvalds * x: (no) no x: (no) yes x: (no) yes x: (yes) yes 661da177e4SLinus Torvalds * 671da177e4SLinus Torvalds * MAP_PRIVATE r: (no) no r: (yes) yes r: (no) yes r: (no) yes 681da177e4SLinus Torvalds * w: (no) no w: (no) no w: (copy) copy w: (no) no 691da177e4SLinus Torvalds * x: (no) no x: (no) yes x: (no) yes x: (yes) yes 701da177e4SLinus Torvalds * 711da177e4SLinus Torvalds */ 721da177e4SLinus Torvalds pgprot_t protection_map[16] = { 731da177e4SLinus Torvalds __P000, __P001, __P010, __P011, __P100, __P101, __P110, __P111, 741da177e4SLinus Torvalds __S000, __S001, __S010, __S011, __S100, __S101, __S110, __S111 751da177e4SLinus Torvalds }; 761da177e4SLinus Torvalds 77804af2cfSHugh Dickins pgprot_t vm_get_page_prot(unsigned long vm_flags) 78804af2cfSHugh Dickins { 79b845f313SDave Kleikamp return __pgprot(pgprot_val(protection_map[vm_flags & 80b845f313SDave Kleikamp (VM_READ|VM_WRITE|VM_EXEC|VM_SHARED)]) | 81b845f313SDave Kleikamp pgprot_val(arch_vm_get_page_prot(vm_flags))); 82804af2cfSHugh Dickins } 83804af2cfSHugh Dickins EXPORT_SYMBOL(vm_get_page_prot); 84804af2cfSHugh Dickins 851da177e4SLinus Torvalds int sysctl_overcommit_memory = OVERCOMMIT_GUESS; /* heuristic overcommit */ 861da177e4SLinus Torvalds int sysctl_overcommit_ratio = 50; /* default is 50% */ 87c3d8c141SChristoph Lameter int sysctl_max_map_count __read_mostly = DEFAULT_MAX_MAP_COUNT; 8800a62ce9SKOSAKI Motohiro struct percpu_counter vm_committed_as; 891da177e4SLinus Torvalds 901da177e4SLinus Torvalds /* 911da177e4SLinus Torvalds * Check that a process has enough memory to allocate a new virtual 921da177e4SLinus Torvalds * mapping. 0 means there is enough memory for the allocation to 931da177e4SLinus Torvalds * succeed and -ENOMEM implies there is not. 941da177e4SLinus Torvalds * 951da177e4SLinus Torvalds * We currently support three overcommit policies, which are set via the 961da177e4SLinus Torvalds * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting 971da177e4SLinus Torvalds * 981da177e4SLinus Torvalds * Strict overcommit modes added 2002 Feb 26 by Alan Cox. 991da177e4SLinus Torvalds * Additional code 2002 Jul 20 by Robert Love. 1001da177e4SLinus Torvalds * 1011da177e4SLinus Torvalds * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise. 1021da177e4SLinus Torvalds * 1031da177e4SLinus Torvalds * Note this is a helper function intended to be used by LSMs which 1041da177e4SLinus Torvalds * wish to use this logic. 1051da177e4SLinus Torvalds */ 10634b4e4aaSAlan Cox int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin) 1071da177e4SLinus Torvalds { 1081da177e4SLinus Torvalds unsigned long free, allowed; 1091da177e4SLinus Torvalds 1101da177e4SLinus Torvalds vm_acct_memory(pages); 1111da177e4SLinus Torvalds 1121da177e4SLinus Torvalds /* 1131da177e4SLinus Torvalds * Sometimes we want to use more memory than we have 1141da177e4SLinus Torvalds */ 1151da177e4SLinus Torvalds if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS) 1161da177e4SLinus Torvalds return 0; 1171da177e4SLinus Torvalds 1181da177e4SLinus Torvalds if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) { 1191da177e4SLinus Torvalds unsigned long n; 1201da177e4SLinus Torvalds 121347ce434SChristoph Lameter free = global_page_state(NR_FILE_PAGES); 1221da177e4SLinus Torvalds free += nr_swap_pages; 1231da177e4SLinus Torvalds 1241da177e4SLinus Torvalds /* 1251da177e4SLinus Torvalds * Any slabs which are created with the 1261da177e4SLinus Torvalds * SLAB_RECLAIM_ACCOUNT flag claim to have contents 1271da177e4SLinus Torvalds * which are reclaimable, under pressure. The dentry 1281da177e4SLinus Torvalds * cache and most inode caches should fall into this 1291da177e4SLinus Torvalds */ 130972d1a7bSChristoph Lameter free += global_page_state(NR_SLAB_RECLAIMABLE); 1311da177e4SLinus Torvalds 1321da177e4SLinus Torvalds /* 1331da177e4SLinus Torvalds * Leave the last 3% for root 1341da177e4SLinus Torvalds */ 1351da177e4SLinus Torvalds if (!cap_sys_admin) 1361da177e4SLinus Torvalds free -= free / 32; 1371da177e4SLinus Torvalds 1381da177e4SLinus Torvalds if (free > pages) 1391da177e4SLinus Torvalds return 0; 1401da177e4SLinus Torvalds 1411da177e4SLinus Torvalds /* 1421da177e4SLinus Torvalds * nr_free_pages() is very expensive on large systems, 1431da177e4SLinus Torvalds * only call if we're about to fail. 1441da177e4SLinus Torvalds */ 1451da177e4SLinus Torvalds n = nr_free_pages(); 1466d9f7839SHideo AOKI 1476d9f7839SHideo AOKI /* 1486d9f7839SHideo AOKI * Leave reserved pages. The pages are not for anonymous pages. 1496d9f7839SHideo AOKI */ 1506d9f7839SHideo AOKI if (n <= totalreserve_pages) 1516d9f7839SHideo AOKI goto error; 1526d9f7839SHideo AOKI else 1536d9f7839SHideo AOKI n -= totalreserve_pages; 1546d9f7839SHideo AOKI 1556d9f7839SHideo AOKI /* 1566d9f7839SHideo AOKI * Leave the last 3% for root 1576d9f7839SHideo AOKI */ 1581da177e4SLinus Torvalds if (!cap_sys_admin) 1591da177e4SLinus Torvalds n -= n / 32; 1601da177e4SLinus Torvalds free += n; 1611da177e4SLinus Torvalds 1621da177e4SLinus Torvalds if (free > pages) 1631da177e4SLinus Torvalds return 0; 1646d9f7839SHideo AOKI 1656d9f7839SHideo AOKI goto error; 1661da177e4SLinus Torvalds } 1671da177e4SLinus Torvalds 1681da177e4SLinus Torvalds allowed = (totalram_pages - hugetlb_total_pages()) 1691da177e4SLinus Torvalds * sysctl_overcommit_ratio / 100; 1701da177e4SLinus Torvalds /* 1711da177e4SLinus Torvalds * Leave the last 3% for root 1721da177e4SLinus Torvalds */ 1731da177e4SLinus Torvalds if (!cap_sys_admin) 1741da177e4SLinus Torvalds allowed -= allowed / 32; 1751da177e4SLinus Torvalds allowed += total_swap_pages; 1761da177e4SLinus Torvalds 1771da177e4SLinus Torvalds /* Don't let a single process grow too big: 1781da177e4SLinus Torvalds leave 3% of the size of this process for other processes */ 179731572d3SAlan Cox if (mm) 18034b4e4aaSAlan Cox allowed -= mm->total_vm / 32; 1811da177e4SLinus Torvalds 18200a62ce9SKOSAKI Motohiro if (percpu_counter_read_positive(&vm_committed_as) < allowed) 1831da177e4SLinus Torvalds return 0; 1846d9f7839SHideo AOKI error: 1851da177e4SLinus Torvalds vm_unacct_memory(pages); 1861da177e4SLinus Torvalds 1871da177e4SLinus Torvalds return -ENOMEM; 1881da177e4SLinus Torvalds } 1891da177e4SLinus Torvalds 1901da177e4SLinus Torvalds /* 1911da177e4SLinus Torvalds * Requires inode->i_mapping->i_mmap_lock 1921da177e4SLinus Torvalds */ 1931da177e4SLinus Torvalds static void __remove_shared_vm_struct(struct vm_area_struct *vma, 1941da177e4SLinus Torvalds struct file *file, struct address_space *mapping) 1951da177e4SLinus Torvalds { 1961da177e4SLinus Torvalds if (vma->vm_flags & VM_DENYWRITE) 197d3ac7f89SJosef "Jeff" Sipek atomic_inc(&file->f_path.dentry->d_inode->i_writecount); 1981da177e4SLinus Torvalds if (vma->vm_flags & VM_SHARED) 1991da177e4SLinus Torvalds mapping->i_mmap_writable--; 2001da177e4SLinus Torvalds 2011da177e4SLinus Torvalds flush_dcache_mmap_lock(mapping); 2021da177e4SLinus Torvalds if (unlikely(vma->vm_flags & VM_NONLINEAR)) 2031da177e4SLinus Torvalds list_del_init(&vma->shared.vm_set.list); 2041da177e4SLinus Torvalds else 2051da177e4SLinus Torvalds vma_prio_tree_remove(vma, &mapping->i_mmap); 2061da177e4SLinus Torvalds flush_dcache_mmap_unlock(mapping); 2071da177e4SLinus Torvalds } 2081da177e4SLinus Torvalds 2091da177e4SLinus Torvalds /* 210a8fb5618SHugh Dickins * Unlink a file-based vm structure from its prio_tree, to hide 211a8fb5618SHugh Dickins * vma from rmap and vmtruncate before freeing its page tables. 2121da177e4SLinus Torvalds */ 213a8fb5618SHugh Dickins void unlink_file_vma(struct vm_area_struct *vma) 2141da177e4SLinus Torvalds { 2151da177e4SLinus Torvalds struct file *file = vma->vm_file; 2161da177e4SLinus Torvalds 2171da177e4SLinus Torvalds if (file) { 2181da177e4SLinus Torvalds struct address_space *mapping = file->f_mapping; 2191da177e4SLinus Torvalds spin_lock(&mapping->i_mmap_lock); 2201da177e4SLinus Torvalds __remove_shared_vm_struct(vma, file, mapping); 2211da177e4SLinus Torvalds spin_unlock(&mapping->i_mmap_lock); 2221da177e4SLinus Torvalds } 223a8fb5618SHugh Dickins } 224a8fb5618SHugh Dickins 225a8fb5618SHugh Dickins /* 226a8fb5618SHugh Dickins * Close a vm structure and free it, returning the next. 227a8fb5618SHugh Dickins */ 228a8fb5618SHugh Dickins static struct vm_area_struct *remove_vma(struct vm_area_struct *vma) 229a8fb5618SHugh Dickins { 230a8fb5618SHugh Dickins struct vm_area_struct *next = vma->vm_next; 231a8fb5618SHugh Dickins 232a8fb5618SHugh Dickins might_sleep(); 2331da177e4SLinus Torvalds if (vma->vm_ops && vma->vm_ops->close) 2341da177e4SLinus Torvalds vma->vm_ops->close(vma); 235925d1c40SMatt Helsley if (vma->vm_file) { 236a8fb5618SHugh Dickins fput(vma->vm_file); 237925d1c40SMatt Helsley if (vma->vm_flags & VM_EXECUTABLE) 238925d1c40SMatt Helsley removed_exe_file_vma(vma->vm_mm); 239925d1c40SMatt Helsley } 240f0be3d32SLee Schermerhorn mpol_put(vma_policy(vma)); 2411da177e4SLinus Torvalds kmem_cache_free(vm_area_cachep, vma); 242a8fb5618SHugh Dickins return next; 2431da177e4SLinus Torvalds } 2441da177e4SLinus Torvalds 2456a6160a7SHeiko Carstens SYSCALL_DEFINE1(brk, unsigned long, brk) 2461da177e4SLinus Torvalds { 2471da177e4SLinus Torvalds unsigned long rlim, retval; 2481da177e4SLinus Torvalds unsigned long newbrk, oldbrk; 2491da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 250a5b4592cSJiri Kosina unsigned long min_brk; 2511da177e4SLinus Torvalds 2521da177e4SLinus Torvalds down_write(&mm->mmap_sem); 2531da177e4SLinus Torvalds 254a5b4592cSJiri Kosina #ifdef CONFIG_COMPAT_BRK 255a5b4592cSJiri Kosina min_brk = mm->end_code; 256a5b4592cSJiri Kosina #else 257a5b4592cSJiri Kosina min_brk = mm->start_brk; 258a5b4592cSJiri Kosina #endif 259a5b4592cSJiri Kosina if (brk < min_brk) 2601da177e4SLinus Torvalds goto out; 2611e624196SRam Gupta 2621e624196SRam Gupta /* 2631e624196SRam Gupta * Check against rlimit here. If this check is done later after the test 2641e624196SRam Gupta * of oldbrk with newbrk then it can escape the test and let the data 2651e624196SRam Gupta * segment grow beyond its set limit the in case where the limit is 2661e624196SRam Gupta * not page aligned -Ram Gupta 2671e624196SRam Gupta */ 26859e99e5bSJiri Slaby rlim = rlimit(RLIMIT_DATA); 269c1d171a0SJiri Kosina if (rlim < RLIM_INFINITY && (brk - mm->start_brk) + 270c1d171a0SJiri Kosina (mm->end_data - mm->start_data) > rlim) 2711e624196SRam Gupta goto out; 2721e624196SRam Gupta 2731da177e4SLinus Torvalds newbrk = PAGE_ALIGN(brk); 2741da177e4SLinus Torvalds oldbrk = PAGE_ALIGN(mm->brk); 2751da177e4SLinus Torvalds if (oldbrk == newbrk) 2761da177e4SLinus Torvalds goto set_brk; 2771da177e4SLinus Torvalds 2781da177e4SLinus Torvalds /* Always allow shrinking brk. */ 2791da177e4SLinus Torvalds if (brk <= mm->brk) { 2801da177e4SLinus Torvalds if (!do_munmap(mm, newbrk, oldbrk-newbrk)) 2811da177e4SLinus Torvalds goto set_brk; 2821da177e4SLinus Torvalds goto out; 2831da177e4SLinus Torvalds } 2841da177e4SLinus Torvalds 2851da177e4SLinus Torvalds /* Check against existing mmap mappings. */ 2861da177e4SLinus Torvalds if (find_vma_intersection(mm, oldbrk, newbrk+PAGE_SIZE)) 2871da177e4SLinus Torvalds goto out; 2881da177e4SLinus Torvalds 2891da177e4SLinus Torvalds /* Ok, looks good - let it rip. */ 2901da177e4SLinus Torvalds if (do_brk(oldbrk, newbrk-oldbrk) != oldbrk) 2911da177e4SLinus Torvalds goto out; 2921da177e4SLinus Torvalds set_brk: 2931da177e4SLinus Torvalds mm->brk = brk; 2941da177e4SLinus Torvalds out: 2951da177e4SLinus Torvalds retval = mm->brk; 2961da177e4SLinus Torvalds up_write(&mm->mmap_sem); 2971da177e4SLinus Torvalds return retval; 2981da177e4SLinus Torvalds } 2991da177e4SLinus Torvalds 3001da177e4SLinus Torvalds #ifdef DEBUG_MM_RB 3011da177e4SLinus Torvalds static int browse_rb(struct rb_root *root) 3021da177e4SLinus Torvalds { 3031da177e4SLinus Torvalds int i = 0, j; 3041da177e4SLinus Torvalds struct rb_node *nd, *pn = NULL; 3051da177e4SLinus Torvalds unsigned long prev = 0, pend = 0; 3061da177e4SLinus Torvalds 3071da177e4SLinus Torvalds for (nd = rb_first(root); nd; nd = rb_next(nd)) { 3081da177e4SLinus Torvalds struct vm_area_struct *vma; 3091da177e4SLinus Torvalds vma = rb_entry(nd, struct vm_area_struct, vm_rb); 3101da177e4SLinus Torvalds if (vma->vm_start < prev) 3111da177e4SLinus Torvalds printk("vm_start %lx prev %lx\n", vma->vm_start, prev), i = -1; 3121da177e4SLinus Torvalds if (vma->vm_start < pend) 3131da177e4SLinus Torvalds printk("vm_start %lx pend %lx\n", vma->vm_start, pend); 3141da177e4SLinus Torvalds if (vma->vm_start > vma->vm_end) 3151da177e4SLinus Torvalds printk("vm_end %lx < vm_start %lx\n", vma->vm_end, vma->vm_start); 3161da177e4SLinus Torvalds i++; 3171da177e4SLinus Torvalds pn = nd; 318d1af65d1SDavid Miller prev = vma->vm_start; 319d1af65d1SDavid Miller pend = vma->vm_end; 3201da177e4SLinus Torvalds } 3211da177e4SLinus Torvalds j = 0; 3221da177e4SLinus Torvalds for (nd = pn; nd; nd = rb_prev(nd)) { 3231da177e4SLinus Torvalds j++; 3241da177e4SLinus Torvalds } 3251da177e4SLinus Torvalds if (i != j) 3261da177e4SLinus Torvalds printk("backwards %d, forwards %d\n", j, i), i = 0; 3271da177e4SLinus Torvalds return i; 3281da177e4SLinus Torvalds } 3291da177e4SLinus Torvalds 3301da177e4SLinus Torvalds void validate_mm(struct mm_struct *mm) 3311da177e4SLinus Torvalds { 3321da177e4SLinus Torvalds int bug = 0; 3331da177e4SLinus Torvalds int i = 0; 3341da177e4SLinus Torvalds struct vm_area_struct *tmp = mm->mmap; 3351da177e4SLinus Torvalds while (tmp) { 3361da177e4SLinus Torvalds tmp = tmp->vm_next; 3371da177e4SLinus Torvalds i++; 3381da177e4SLinus Torvalds } 3391da177e4SLinus Torvalds if (i != mm->map_count) 3401da177e4SLinus Torvalds printk("map_count %d vm_next %d\n", mm->map_count, i), bug = 1; 3411da177e4SLinus Torvalds i = browse_rb(&mm->mm_rb); 3421da177e4SLinus Torvalds if (i != mm->map_count) 3431da177e4SLinus Torvalds printk("map_count %d rb %d\n", mm->map_count, i), bug = 1; 34446a350efSEric Sesterhenn BUG_ON(bug); 3451da177e4SLinus Torvalds } 3461da177e4SLinus Torvalds #else 3471da177e4SLinus Torvalds #define validate_mm(mm) do { } while (0) 3481da177e4SLinus Torvalds #endif 3491da177e4SLinus Torvalds 3501da177e4SLinus Torvalds static struct vm_area_struct * 3511da177e4SLinus Torvalds find_vma_prepare(struct mm_struct *mm, unsigned long addr, 3521da177e4SLinus Torvalds struct vm_area_struct **pprev, struct rb_node ***rb_link, 3531da177e4SLinus Torvalds struct rb_node ** rb_parent) 3541da177e4SLinus Torvalds { 3551da177e4SLinus Torvalds struct vm_area_struct * vma; 3561da177e4SLinus Torvalds struct rb_node ** __rb_link, * __rb_parent, * rb_prev; 3571da177e4SLinus Torvalds 3581da177e4SLinus Torvalds __rb_link = &mm->mm_rb.rb_node; 3591da177e4SLinus Torvalds rb_prev = __rb_parent = NULL; 3601da177e4SLinus Torvalds vma = NULL; 3611da177e4SLinus Torvalds 3621da177e4SLinus Torvalds while (*__rb_link) { 3631da177e4SLinus Torvalds struct vm_area_struct *vma_tmp; 3641da177e4SLinus Torvalds 3651da177e4SLinus Torvalds __rb_parent = *__rb_link; 3661da177e4SLinus Torvalds vma_tmp = rb_entry(__rb_parent, struct vm_area_struct, vm_rb); 3671da177e4SLinus Torvalds 3681da177e4SLinus Torvalds if (vma_tmp->vm_end > addr) { 3691da177e4SLinus Torvalds vma = vma_tmp; 3701da177e4SLinus Torvalds if (vma_tmp->vm_start <= addr) 371dfe195fbSBenny Halevy break; 3721da177e4SLinus Torvalds __rb_link = &__rb_parent->rb_left; 3731da177e4SLinus Torvalds } else { 3741da177e4SLinus Torvalds rb_prev = __rb_parent; 3751da177e4SLinus Torvalds __rb_link = &__rb_parent->rb_right; 3761da177e4SLinus Torvalds } 3771da177e4SLinus Torvalds } 3781da177e4SLinus Torvalds 3791da177e4SLinus Torvalds *pprev = NULL; 3801da177e4SLinus Torvalds if (rb_prev) 3811da177e4SLinus Torvalds *pprev = rb_entry(rb_prev, struct vm_area_struct, vm_rb); 3821da177e4SLinus Torvalds *rb_link = __rb_link; 3831da177e4SLinus Torvalds *rb_parent = __rb_parent; 3841da177e4SLinus Torvalds return vma; 3851da177e4SLinus Torvalds } 3861da177e4SLinus Torvalds 3871da177e4SLinus Torvalds static inline void 3881da177e4SLinus Torvalds __vma_link_list(struct mm_struct *mm, struct vm_area_struct *vma, 3891da177e4SLinus Torvalds struct vm_area_struct *prev, struct rb_node *rb_parent) 3901da177e4SLinus Torvalds { 391297c5eeeSLinus Torvalds struct vm_area_struct *next; 392297c5eeeSLinus Torvalds 393297c5eeeSLinus Torvalds vma->vm_prev = prev; 3941da177e4SLinus Torvalds if (prev) { 395297c5eeeSLinus Torvalds next = prev->vm_next; 3961da177e4SLinus Torvalds prev->vm_next = vma; 3971da177e4SLinus Torvalds } else { 3981da177e4SLinus Torvalds mm->mmap = vma; 3991da177e4SLinus Torvalds if (rb_parent) 400297c5eeeSLinus Torvalds next = rb_entry(rb_parent, 4011da177e4SLinus Torvalds struct vm_area_struct, vm_rb); 4021da177e4SLinus Torvalds else 403297c5eeeSLinus Torvalds next = NULL; 4041da177e4SLinus Torvalds } 405297c5eeeSLinus Torvalds vma->vm_next = next; 406297c5eeeSLinus Torvalds if (next) 407297c5eeeSLinus Torvalds next->vm_prev = vma; 4081da177e4SLinus Torvalds } 4091da177e4SLinus Torvalds 4101da177e4SLinus Torvalds void __vma_link_rb(struct mm_struct *mm, struct vm_area_struct *vma, 4111da177e4SLinus Torvalds struct rb_node **rb_link, struct rb_node *rb_parent) 4121da177e4SLinus Torvalds { 4131da177e4SLinus Torvalds rb_link_node(&vma->vm_rb, rb_parent, rb_link); 4141da177e4SLinus Torvalds rb_insert_color(&vma->vm_rb, &mm->mm_rb); 4151da177e4SLinus Torvalds } 4161da177e4SLinus Torvalds 417cb8f488cSDenys Vlasenko static void __vma_link_file(struct vm_area_struct *vma) 4181da177e4SLinus Torvalds { 4191da177e4SLinus Torvalds struct file *file; 4201da177e4SLinus Torvalds 4211da177e4SLinus Torvalds file = vma->vm_file; 4221da177e4SLinus Torvalds if (file) { 4231da177e4SLinus Torvalds struct address_space *mapping = file->f_mapping; 4241da177e4SLinus Torvalds 4251da177e4SLinus Torvalds if (vma->vm_flags & VM_DENYWRITE) 426d3ac7f89SJosef "Jeff" Sipek atomic_dec(&file->f_path.dentry->d_inode->i_writecount); 4271da177e4SLinus Torvalds if (vma->vm_flags & VM_SHARED) 4281da177e4SLinus Torvalds mapping->i_mmap_writable++; 4291da177e4SLinus Torvalds 4301da177e4SLinus Torvalds flush_dcache_mmap_lock(mapping); 4311da177e4SLinus Torvalds if (unlikely(vma->vm_flags & VM_NONLINEAR)) 4321da177e4SLinus Torvalds vma_nonlinear_insert(vma, &mapping->i_mmap_nonlinear); 4331da177e4SLinus Torvalds else 4341da177e4SLinus Torvalds vma_prio_tree_insert(vma, &mapping->i_mmap); 4351da177e4SLinus Torvalds flush_dcache_mmap_unlock(mapping); 4361da177e4SLinus Torvalds } 4371da177e4SLinus Torvalds } 4381da177e4SLinus Torvalds 4391da177e4SLinus Torvalds static void 4401da177e4SLinus Torvalds __vma_link(struct mm_struct *mm, struct vm_area_struct *vma, 4411da177e4SLinus Torvalds struct vm_area_struct *prev, struct rb_node **rb_link, 4421da177e4SLinus Torvalds struct rb_node *rb_parent) 4431da177e4SLinus Torvalds { 4441da177e4SLinus Torvalds __vma_link_list(mm, vma, prev, rb_parent); 4451da177e4SLinus Torvalds __vma_link_rb(mm, vma, rb_link, rb_parent); 4461da177e4SLinus Torvalds } 4471da177e4SLinus Torvalds 4481da177e4SLinus Torvalds static void vma_link(struct mm_struct *mm, struct vm_area_struct *vma, 4491da177e4SLinus Torvalds struct vm_area_struct *prev, struct rb_node **rb_link, 4501da177e4SLinus Torvalds struct rb_node *rb_parent) 4511da177e4SLinus Torvalds { 4521da177e4SLinus Torvalds struct address_space *mapping = NULL; 4531da177e4SLinus Torvalds 4541da177e4SLinus Torvalds if (vma->vm_file) 4551da177e4SLinus Torvalds mapping = vma->vm_file->f_mapping; 4561da177e4SLinus Torvalds 4571da177e4SLinus Torvalds if (mapping) { 4581da177e4SLinus Torvalds spin_lock(&mapping->i_mmap_lock); 4591da177e4SLinus Torvalds vma->vm_truncate_count = mapping->truncate_count; 4601da177e4SLinus Torvalds } 4611da177e4SLinus Torvalds 4621da177e4SLinus Torvalds __vma_link(mm, vma, prev, rb_link, rb_parent); 4631da177e4SLinus Torvalds __vma_link_file(vma); 4641da177e4SLinus Torvalds 4651da177e4SLinus Torvalds if (mapping) 4661da177e4SLinus Torvalds spin_unlock(&mapping->i_mmap_lock); 4671da177e4SLinus Torvalds 4681da177e4SLinus Torvalds mm->map_count++; 4691da177e4SLinus Torvalds validate_mm(mm); 4701da177e4SLinus Torvalds } 4711da177e4SLinus Torvalds 4721da177e4SLinus Torvalds /* 4731da177e4SLinus Torvalds * Helper for vma_adjust in the split_vma insert case: 4741da177e4SLinus Torvalds * insert vm structure into list and rbtree and anon_vma, 4751da177e4SLinus Torvalds * but it has already been inserted into prio_tree earlier. 4761da177e4SLinus Torvalds */ 47748aae425SZhenwenXu static void __insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma) 4781da177e4SLinus Torvalds { 4791da177e4SLinus Torvalds struct vm_area_struct *__vma, *prev; 4801da177e4SLinus Torvalds struct rb_node **rb_link, *rb_parent; 4811da177e4SLinus Torvalds 4821da177e4SLinus Torvalds __vma = find_vma_prepare(mm, vma->vm_start,&prev, &rb_link, &rb_parent); 48346a350efSEric Sesterhenn BUG_ON(__vma && __vma->vm_start < vma->vm_end); 4841da177e4SLinus Torvalds __vma_link(mm, vma, prev, rb_link, rb_parent); 4851da177e4SLinus Torvalds mm->map_count++; 4861da177e4SLinus Torvalds } 4871da177e4SLinus Torvalds 4881da177e4SLinus Torvalds static inline void 4891da177e4SLinus Torvalds __vma_unlink(struct mm_struct *mm, struct vm_area_struct *vma, 4901da177e4SLinus Torvalds struct vm_area_struct *prev) 4911da177e4SLinus Torvalds { 492297c5eeeSLinus Torvalds struct vm_area_struct *next = vma->vm_next; 493297c5eeeSLinus Torvalds 494297c5eeeSLinus Torvalds prev->vm_next = next; 495297c5eeeSLinus Torvalds if (next) 496297c5eeeSLinus Torvalds next->vm_prev = prev; 4971da177e4SLinus Torvalds rb_erase(&vma->vm_rb, &mm->mm_rb); 4981da177e4SLinus Torvalds if (mm->mmap_cache == vma) 4991da177e4SLinus Torvalds mm->mmap_cache = prev; 5001da177e4SLinus Torvalds } 5011da177e4SLinus Torvalds 5021da177e4SLinus Torvalds /* 5031da177e4SLinus Torvalds * We cannot adjust vm_start, vm_end, vm_pgoff fields of a vma that 5041da177e4SLinus Torvalds * is already present in an i_mmap tree without adjusting the tree. 5051da177e4SLinus Torvalds * The following helper function should be used when such adjustments 5061da177e4SLinus Torvalds * are necessary. The "insert" vma (if any) is to be inserted 5071da177e4SLinus Torvalds * before we drop the necessary locks. 5081da177e4SLinus Torvalds */ 5095beb4930SRik van Riel int vma_adjust(struct vm_area_struct *vma, unsigned long start, 5101da177e4SLinus Torvalds unsigned long end, pgoff_t pgoff, struct vm_area_struct *insert) 5111da177e4SLinus Torvalds { 5121da177e4SLinus Torvalds struct mm_struct *mm = vma->vm_mm; 5131da177e4SLinus Torvalds struct vm_area_struct *next = vma->vm_next; 5141da177e4SLinus Torvalds struct vm_area_struct *importer = NULL; 5151da177e4SLinus Torvalds struct address_space *mapping = NULL; 5161da177e4SLinus Torvalds struct prio_tree_root *root = NULL; 517012f1800SRik van Riel struct anon_vma *anon_vma = NULL; 5181da177e4SLinus Torvalds struct file *file = vma->vm_file; 5191da177e4SLinus Torvalds long adjust_next = 0; 5201da177e4SLinus Torvalds int remove_next = 0; 5211da177e4SLinus Torvalds 5221da177e4SLinus Torvalds if (next && !insert) { 523287d97acSLinus Torvalds struct vm_area_struct *exporter = NULL; 524287d97acSLinus Torvalds 5251da177e4SLinus Torvalds if (end >= next->vm_end) { 5261da177e4SLinus Torvalds /* 5271da177e4SLinus Torvalds * vma expands, overlapping all the next, and 5281da177e4SLinus Torvalds * perhaps the one after too (mprotect case 6). 5291da177e4SLinus Torvalds */ 5301da177e4SLinus Torvalds again: remove_next = 1 + (end > next->vm_end); 5311da177e4SLinus Torvalds end = next->vm_end; 532287d97acSLinus Torvalds exporter = next; 5331da177e4SLinus Torvalds importer = vma; 5341da177e4SLinus Torvalds } else if (end > next->vm_start) { 5351da177e4SLinus Torvalds /* 5361da177e4SLinus Torvalds * vma expands, overlapping part of the next: 5371da177e4SLinus Torvalds * mprotect case 5 shifting the boundary up. 5381da177e4SLinus Torvalds */ 5391da177e4SLinus Torvalds adjust_next = (end - next->vm_start) >> PAGE_SHIFT; 540287d97acSLinus Torvalds exporter = next; 5411da177e4SLinus Torvalds importer = vma; 5421da177e4SLinus Torvalds } else if (end < vma->vm_end) { 5431da177e4SLinus Torvalds /* 5441da177e4SLinus Torvalds * vma shrinks, and !insert tells it's not 5451da177e4SLinus Torvalds * split_vma inserting another: so it must be 5461da177e4SLinus Torvalds * mprotect case 4 shifting the boundary down. 5471da177e4SLinus Torvalds */ 5481da177e4SLinus Torvalds adjust_next = - ((vma->vm_end - end) >> PAGE_SHIFT); 549287d97acSLinus Torvalds exporter = vma; 5501da177e4SLinus Torvalds importer = next; 5511da177e4SLinus Torvalds } 5521da177e4SLinus Torvalds 5535beb4930SRik van Riel /* 5545beb4930SRik van Riel * Easily overlooked: when mprotect shifts the boundary, 5555beb4930SRik van Riel * make sure the expanding vma has anon_vma set if the 5565beb4930SRik van Riel * shrinking vma had, to cover any anon pages imported. 5575beb4930SRik van Riel */ 558287d97acSLinus Torvalds if (exporter && exporter->anon_vma && !importer->anon_vma) { 559287d97acSLinus Torvalds if (anon_vma_clone(importer, exporter)) 5605beb4930SRik van Riel return -ENOMEM; 561287d97acSLinus Torvalds importer->anon_vma = exporter->anon_vma; 5625beb4930SRik van Riel } 5635beb4930SRik van Riel } 5645beb4930SRik van Riel 5651da177e4SLinus Torvalds if (file) { 5661da177e4SLinus Torvalds mapping = file->f_mapping; 5671da177e4SLinus Torvalds if (!(vma->vm_flags & VM_NONLINEAR)) 5681da177e4SLinus Torvalds root = &mapping->i_mmap; 5691da177e4SLinus Torvalds spin_lock(&mapping->i_mmap_lock); 5701da177e4SLinus Torvalds if (importer && 5711da177e4SLinus Torvalds vma->vm_truncate_count != next->vm_truncate_count) { 5721da177e4SLinus Torvalds /* 5731da177e4SLinus Torvalds * unmap_mapping_range might be in progress: 5741da177e4SLinus Torvalds * ensure that the expanding vma is rescanned. 5751da177e4SLinus Torvalds */ 5761da177e4SLinus Torvalds importer->vm_truncate_count = 0; 5771da177e4SLinus Torvalds } 5781da177e4SLinus Torvalds if (insert) { 5791da177e4SLinus Torvalds insert->vm_truncate_count = vma->vm_truncate_count; 5801da177e4SLinus Torvalds /* 5811da177e4SLinus Torvalds * Put into prio_tree now, so instantiated pages 5821da177e4SLinus Torvalds * are visible to arm/parisc __flush_dcache_page 5831da177e4SLinus Torvalds * throughout; but we cannot insert into address 5841da177e4SLinus Torvalds * space until vma start or end is updated. 5851da177e4SLinus Torvalds */ 5861da177e4SLinus Torvalds __vma_link_file(insert); 5871da177e4SLinus Torvalds } 5881da177e4SLinus Torvalds } 5891da177e4SLinus Torvalds 590012f1800SRik van Riel /* 591012f1800SRik van Riel * When changing only vma->vm_end, we don't really need anon_vma 592012f1800SRik van Riel * lock. This is a fairly rare case by itself, but the anon_vma 593012f1800SRik van Riel * lock may be shared between many sibling processes. Skipping 594012f1800SRik van Riel * the lock for brk adjustments makes a difference sometimes. 595012f1800SRik van Riel */ 596012f1800SRik van Riel if (vma->anon_vma && (insert || importer || start != vma->vm_start)) { 597012f1800SRik van Riel anon_vma = vma->anon_vma; 598012f1800SRik van Riel anon_vma_lock(anon_vma); 599012f1800SRik van Riel } 600012f1800SRik van Riel 6011da177e4SLinus Torvalds if (root) { 6021da177e4SLinus Torvalds flush_dcache_mmap_lock(mapping); 6031da177e4SLinus Torvalds vma_prio_tree_remove(vma, root); 6041da177e4SLinus Torvalds if (adjust_next) 6051da177e4SLinus Torvalds vma_prio_tree_remove(next, root); 6061da177e4SLinus Torvalds } 6071da177e4SLinus Torvalds 6081da177e4SLinus Torvalds vma->vm_start = start; 6091da177e4SLinus Torvalds vma->vm_end = end; 6101da177e4SLinus Torvalds vma->vm_pgoff = pgoff; 6111da177e4SLinus Torvalds if (adjust_next) { 6121da177e4SLinus Torvalds next->vm_start += adjust_next << PAGE_SHIFT; 6131da177e4SLinus Torvalds next->vm_pgoff += adjust_next; 6141da177e4SLinus Torvalds } 6151da177e4SLinus Torvalds 6161da177e4SLinus Torvalds if (root) { 6171da177e4SLinus Torvalds if (adjust_next) 6181da177e4SLinus Torvalds vma_prio_tree_insert(next, root); 6191da177e4SLinus Torvalds vma_prio_tree_insert(vma, root); 6201da177e4SLinus Torvalds flush_dcache_mmap_unlock(mapping); 6211da177e4SLinus Torvalds } 6221da177e4SLinus Torvalds 6231da177e4SLinus Torvalds if (remove_next) { 6241da177e4SLinus Torvalds /* 6251da177e4SLinus Torvalds * vma_merge has merged next into vma, and needs 6261da177e4SLinus Torvalds * us to remove next before dropping the locks. 6271da177e4SLinus Torvalds */ 6281da177e4SLinus Torvalds __vma_unlink(mm, next, vma); 6291da177e4SLinus Torvalds if (file) 6301da177e4SLinus Torvalds __remove_shared_vm_struct(next, file, mapping); 6311da177e4SLinus Torvalds } else if (insert) { 6321da177e4SLinus Torvalds /* 6331da177e4SLinus Torvalds * split_vma has split insert from vma, and needs 6341da177e4SLinus Torvalds * us to insert it before dropping the locks 6351da177e4SLinus Torvalds * (it may either follow vma or precede it). 6361da177e4SLinus Torvalds */ 6371da177e4SLinus Torvalds __insert_vm_struct(mm, insert); 6381da177e4SLinus Torvalds } 6391da177e4SLinus Torvalds 640012f1800SRik van Riel if (anon_vma) 641012f1800SRik van Riel anon_vma_unlock(anon_vma); 6421da177e4SLinus Torvalds if (mapping) 6431da177e4SLinus Torvalds spin_unlock(&mapping->i_mmap_lock); 6441da177e4SLinus Torvalds 6451da177e4SLinus Torvalds if (remove_next) { 646925d1c40SMatt Helsley if (file) { 6471da177e4SLinus Torvalds fput(file); 648925d1c40SMatt Helsley if (next->vm_flags & VM_EXECUTABLE) 649925d1c40SMatt Helsley removed_exe_file_vma(mm); 650925d1c40SMatt Helsley } 6515beb4930SRik van Riel if (next->anon_vma) 6525beb4930SRik van Riel anon_vma_merge(vma, next); 6531da177e4SLinus Torvalds mm->map_count--; 654f0be3d32SLee Schermerhorn mpol_put(vma_policy(next)); 6551da177e4SLinus Torvalds kmem_cache_free(vm_area_cachep, next); 6561da177e4SLinus Torvalds /* 6571da177e4SLinus Torvalds * In mprotect's case 6 (see comments on vma_merge), 6581da177e4SLinus Torvalds * we must remove another next too. It would clutter 6591da177e4SLinus Torvalds * up the code too much to do both in one go. 6601da177e4SLinus Torvalds */ 6611da177e4SLinus Torvalds if (remove_next == 2) { 6621da177e4SLinus Torvalds next = vma->vm_next; 6631da177e4SLinus Torvalds goto again; 6641da177e4SLinus Torvalds } 6651da177e4SLinus Torvalds } 6661da177e4SLinus Torvalds 6671da177e4SLinus Torvalds validate_mm(mm); 6685beb4930SRik van Riel 6695beb4930SRik van Riel return 0; 6701da177e4SLinus Torvalds } 6711da177e4SLinus Torvalds 6721da177e4SLinus Torvalds /* 6731da177e4SLinus Torvalds * If the vma has a ->close operation then the driver probably needs to release 6741da177e4SLinus Torvalds * per-vma resources, so we don't attempt to merge those. 6751da177e4SLinus Torvalds */ 6761da177e4SLinus Torvalds static inline int is_mergeable_vma(struct vm_area_struct *vma, 6771da177e4SLinus Torvalds struct file *file, unsigned long vm_flags) 6781da177e4SLinus Torvalds { 6798314c4f2SHugh Dickins /* VM_CAN_NONLINEAR may get set later by f_op->mmap() */ 6808314c4f2SHugh Dickins if ((vma->vm_flags ^ vm_flags) & ~VM_CAN_NONLINEAR) 6811da177e4SLinus Torvalds return 0; 6821da177e4SLinus Torvalds if (vma->vm_file != file) 6831da177e4SLinus Torvalds return 0; 6841da177e4SLinus Torvalds if (vma->vm_ops && vma->vm_ops->close) 6851da177e4SLinus Torvalds return 0; 6861da177e4SLinus Torvalds return 1; 6871da177e4SLinus Torvalds } 6881da177e4SLinus Torvalds 6891da177e4SLinus Torvalds static inline int is_mergeable_anon_vma(struct anon_vma *anon_vma1, 6901da177e4SLinus Torvalds struct anon_vma *anon_vma2) 6911da177e4SLinus Torvalds { 6921da177e4SLinus Torvalds return !anon_vma1 || !anon_vma2 || (anon_vma1 == anon_vma2); 6931da177e4SLinus Torvalds } 6941da177e4SLinus Torvalds 6951da177e4SLinus Torvalds /* 6961da177e4SLinus Torvalds * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff) 6971da177e4SLinus Torvalds * in front of (at a lower virtual address and file offset than) the vma. 6981da177e4SLinus Torvalds * 6991da177e4SLinus Torvalds * We cannot merge two vmas if they have differently assigned (non-NULL) 7001da177e4SLinus Torvalds * anon_vmas, nor if same anon_vma is assigned but offsets incompatible. 7011da177e4SLinus Torvalds * 7021da177e4SLinus Torvalds * We don't check here for the merged mmap wrapping around the end of pagecache 7031da177e4SLinus Torvalds * indices (16TB on ia32) because do_mmap_pgoff() does not permit mmap's which 7041da177e4SLinus Torvalds * wrap, nor mmaps which cover the final page at index -1UL. 7051da177e4SLinus Torvalds */ 7061da177e4SLinus Torvalds static int 7071da177e4SLinus Torvalds can_vma_merge_before(struct vm_area_struct *vma, unsigned long vm_flags, 7081da177e4SLinus Torvalds struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff) 7091da177e4SLinus Torvalds { 7101da177e4SLinus Torvalds if (is_mergeable_vma(vma, file, vm_flags) && 7111da177e4SLinus Torvalds is_mergeable_anon_vma(anon_vma, vma->anon_vma)) { 7121da177e4SLinus Torvalds if (vma->vm_pgoff == vm_pgoff) 7131da177e4SLinus Torvalds return 1; 7141da177e4SLinus Torvalds } 7151da177e4SLinus Torvalds return 0; 7161da177e4SLinus Torvalds } 7171da177e4SLinus Torvalds 7181da177e4SLinus Torvalds /* 7191da177e4SLinus Torvalds * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff) 7201da177e4SLinus Torvalds * beyond (at a higher virtual address and file offset than) the vma. 7211da177e4SLinus Torvalds * 7221da177e4SLinus Torvalds * We cannot merge two vmas if they have differently assigned (non-NULL) 7231da177e4SLinus Torvalds * anon_vmas, nor if same anon_vma is assigned but offsets incompatible. 7241da177e4SLinus Torvalds */ 7251da177e4SLinus Torvalds static int 7261da177e4SLinus Torvalds can_vma_merge_after(struct vm_area_struct *vma, unsigned long vm_flags, 7271da177e4SLinus Torvalds struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff) 7281da177e4SLinus Torvalds { 7291da177e4SLinus Torvalds if (is_mergeable_vma(vma, file, vm_flags) && 7301da177e4SLinus Torvalds is_mergeable_anon_vma(anon_vma, vma->anon_vma)) { 7311da177e4SLinus Torvalds pgoff_t vm_pglen; 7321da177e4SLinus Torvalds vm_pglen = (vma->vm_end - vma->vm_start) >> PAGE_SHIFT; 7331da177e4SLinus Torvalds if (vma->vm_pgoff + vm_pglen == vm_pgoff) 7341da177e4SLinus Torvalds return 1; 7351da177e4SLinus Torvalds } 7361da177e4SLinus Torvalds return 0; 7371da177e4SLinus Torvalds } 7381da177e4SLinus Torvalds 7391da177e4SLinus Torvalds /* 7401da177e4SLinus Torvalds * Given a mapping request (addr,end,vm_flags,file,pgoff), figure out 7411da177e4SLinus Torvalds * whether that can be merged with its predecessor or its successor. 7421da177e4SLinus Torvalds * Or both (it neatly fills a hole). 7431da177e4SLinus Torvalds * 7441da177e4SLinus Torvalds * In most cases - when called for mmap, brk or mremap - [addr,end) is 7451da177e4SLinus Torvalds * certain not to be mapped by the time vma_merge is called; but when 7461da177e4SLinus Torvalds * called for mprotect, it is certain to be already mapped (either at 7471da177e4SLinus Torvalds * an offset within prev, or at the start of next), and the flags of 7481da177e4SLinus Torvalds * this area are about to be changed to vm_flags - and the no-change 7491da177e4SLinus Torvalds * case has already been eliminated. 7501da177e4SLinus Torvalds * 7511da177e4SLinus Torvalds * The following mprotect cases have to be considered, where AAAA is 7521da177e4SLinus Torvalds * the area passed down from mprotect_fixup, never extending beyond one 7531da177e4SLinus Torvalds * vma, PPPPPP is the prev vma specified, and NNNNNN the next vma after: 7541da177e4SLinus Torvalds * 7551da177e4SLinus Torvalds * AAAA AAAA AAAA AAAA 7561da177e4SLinus Torvalds * PPPPPPNNNNNN PPPPPPNNNNNN PPPPPPNNNNNN PPPPNNNNXXXX 7571da177e4SLinus Torvalds * cannot merge might become might become might become 7581da177e4SLinus Torvalds * PPNNNNNNNNNN PPPPPPPPPPNN PPPPPPPPPPPP 6 or 7591da177e4SLinus Torvalds * mmap, brk or case 4 below case 5 below PPPPPPPPXXXX 7 or 7601da177e4SLinus Torvalds * mremap move: PPPPNNNNNNNN 8 7611da177e4SLinus Torvalds * AAAA 7621da177e4SLinus Torvalds * PPPP NNNN PPPPPPPPPPPP PPPPPPPPNNNN PPPPNNNNNNNN 7631da177e4SLinus Torvalds * might become case 1 below case 2 below case 3 below 7641da177e4SLinus Torvalds * 7651da177e4SLinus Torvalds * Odd one out? Case 8, because it extends NNNN but needs flags of XXXX: 7661da177e4SLinus Torvalds * mprotect_fixup updates vm_flags & vm_page_prot on successful return. 7671da177e4SLinus Torvalds */ 7681da177e4SLinus Torvalds struct vm_area_struct *vma_merge(struct mm_struct *mm, 7691da177e4SLinus Torvalds struct vm_area_struct *prev, unsigned long addr, 7701da177e4SLinus Torvalds unsigned long end, unsigned long vm_flags, 7711da177e4SLinus Torvalds struct anon_vma *anon_vma, struct file *file, 7721da177e4SLinus Torvalds pgoff_t pgoff, struct mempolicy *policy) 7731da177e4SLinus Torvalds { 7741da177e4SLinus Torvalds pgoff_t pglen = (end - addr) >> PAGE_SHIFT; 7751da177e4SLinus Torvalds struct vm_area_struct *area, *next; 7765beb4930SRik van Riel int err; 7771da177e4SLinus Torvalds 7781da177e4SLinus Torvalds /* 7791da177e4SLinus Torvalds * We later require that vma->vm_flags == vm_flags, 7801da177e4SLinus Torvalds * so this tests vma->vm_flags & VM_SPECIAL, too. 7811da177e4SLinus Torvalds */ 7821da177e4SLinus Torvalds if (vm_flags & VM_SPECIAL) 7831da177e4SLinus Torvalds return NULL; 7841da177e4SLinus Torvalds 7851da177e4SLinus Torvalds if (prev) 7861da177e4SLinus Torvalds next = prev->vm_next; 7871da177e4SLinus Torvalds else 7881da177e4SLinus Torvalds next = mm->mmap; 7891da177e4SLinus Torvalds area = next; 7901da177e4SLinus Torvalds if (next && next->vm_end == end) /* cases 6, 7, 8 */ 7911da177e4SLinus Torvalds next = next->vm_next; 7921da177e4SLinus Torvalds 7931da177e4SLinus Torvalds /* 7941da177e4SLinus Torvalds * Can it merge with the predecessor? 7951da177e4SLinus Torvalds */ 7961da177e4SLinus Torvalds if (prev && prev->vm_end == addr && 7971da177e4SLinus Torvalds mpol_equal(vma_policy(prev), policy) && 7981da177e4SLinus Torvalds can_vma_merge_after(prev, vm_flags, 7991da177e4SLinus Torvalds anon_vma, file, pgoff)) { 8001da177e4SLinus Torvalds /* 8011da177e4SLinus Torvalds * OK, it can. Can we now merge in the successor as well? 8021da177e4SLinus Torvalds */ 8031da177e4SLinus Torvalds if (next && end == next->vm_start && 8041da177e4SLinus Torvalds mpol_equal(policy, vma_policy(next)) && 8051da177e4SLinus Torvalds can_vma_merge_before(next, vm_flags, 8061da177e4SLinus Torvalds anon_vma, file, pgoff+pglen) && 8071da177e4SLinus Torvalds is_mergeable_anon_vma(prev->anon_vma, 8081da177e4SLinus Torvalds next->anon_vma)) { 8091da177e4SLinus Torvalds /* cases 1, 6 */ 8105beb4930SRik van Riel err = vma_adjust(prev, prev->vm_start, 8111da177e4SLinus Torvalds next->vm_end, prev->vm_pgoff, NULL); 8121da177e4SLinus Torvalds } else /* cases 2, 5, 7 */ 8135beb4930SRik van Riel err = vma_adjust(prev, prev->vm_start, 8141da177e4SLinus Torvalds end, prev->vm_pgoff, NULL); 8155beb4930SRik van Riel if (err) 8165beb4930SRik van Riel return NULL; 8171da177e4SLinus Torvalds return prev; 8181da177e4SLinus Torvalds } 8191da177e4SLinus Torvalds 8201da177e4SLinus Torvalds /* 8211da177e4SLinus Torvalds * Can this new request be merged in front of next? 8221da177e4SLinus Torvalds */ 8231da177e4SLinus Torvalds if (next && end == next->vm_start && 8241da177e4SLinus Torvalds mpol_equal(policy, vma_policy(next)) && 8251da177e4SLinus Torvalds can_vma_merge_before(next, vm_flags, 8261da177e4SLinus Torvalds anon_vma, file, pgoff+pglen)) { 8271da177e4SLinus Torvalds if (prev && addr < prev->vm_end) /* case 4 */ 8285beb4930SRik van Riel err = vma_adjust(prev, prev->vm_start, 8291da177e4SLinus Torvalds addr, prev->vm_pgoff, NULL); 8301da177e4SLinus Torvalds else /* cases 3, 8 */ 8315beb4930SRik van Riel err = vma_adjust(area, addr, next->vm_end, 8321da177e4SLinus Torvalds next->vm_pgoff - pglen, NULL); 8335beb4930SRik van Riel if (err) 8345beb4930SRik van Riel return NULL; 8351da177e4SLinus Torvalds return area; 8361da177e4SLinus Torvalds } 8371da177e4SLinus Torvalds 8381da177e4SLinus Torvalds return NULL; 8391da177e4SLinus Torvalds } 8401da177e4SLinus Torvalds 8411da177e4SLinus Torvalds /* 842d0e9fe17SLinus Torvalds * Rough compatbility check to quickly see if it's even worth looking 843d0e9fe17SLinus Torvalds * at sharing an anon_vma. 844d0e9fe17SLinus Torvalds * 845d0e9fe17SLinus Torvalds * They need to have the same vm_file, and the flags can only differ 846d0e9fe17SLinus Torvalds * in things that mprotect may change. 847d0e9fe17SLinus Torvalds * 848d0e9fe17SLinus Torvalds * NOTE! The fact that we share an anon_vma doesn't _have_ to mean that 849d0e9fe17SLinus Torvalds * we can merge the two vma's. For example, we refuse to merge a vma if 850d0e9fe17SLinus Torvalds * there is a vm_ops->close() function, because that indicates that the 851d0e9fe17SLinus Torvalds * driver is doing some kind of reference counting. But that doesn't 852d0e9fe17SLinus Torvalds * really matter for the anon_vma sharing case. 853d0e9fe17SLinus Torvalds */ 854d0e9fe17SLinus Torvalds static int anon_vma_compatible(struct vm_area_struct *a, struct vm_area_struct *b) 855d0e9fe17SLinus Torvalds { 856d0e9fe17SLinus Torvalds return a->vm_end == b->vm_start && 857d0e9fe17SLinus Torvalds mpol_equal(vma_policy(a), vma_policy(b)) && 858d0e9fe17SLinus Torvalds a->vm_file == b->vm_file && 859d0e9fe17SLinus Torvalds !((a->vm_flags ^ b->vm_flags) & ~(VM_READ|VM_WRITE|VM_EXEC)) && 860d0e9fe17SLinus Torvalds b->vm_pgoff == a->vm_pgoff + ((b->vm_start - a->vm_start) >> PAGE_SHIFT); 861d0e9fe17SLinus Torvalds } 862d0e9fe17SLinus Torvalds 863d0e9fe17SLinus Torvalds /* 864d0e9fe17SLinus Torvalds * Do some basic sanity checking to see if we can re-use the anon_vma 865d0e9fe17SLinus Torvalds * from 'old'. The 'a'/'b' vma's are in VM order - one of them will be 866d0e9fe17SLinus Torvalds * the same as 'old', the other will be the new one that is trying 867d0e9fe17SLinus Torvalds * to share the anon_vma. 868d0e9fe17SLinus Torvalds * 869d0e9fe17SLinus Torvalds * NOTE! This runs with mm_sem held for reading, so it is possible that 870d0e9fe17SLinus Torvalds * the anon_vma of 'old' is concurrently in the process of being set up 871d0e9fe17SLinus Torvalds * by another page fault trying to merge _that_. But that's ok: if it 872d0e9fe17SLinus Torvalds * is being set up, that automatically means that it will be a singleton 873d0e9fe17SLinus Torvalds * acceptable for merging, so we can do all of this optimistically. But 874d0e9fe17SLinus Torvalds * we do that ACCESS_ONCE() to make sure that we never re-load the pointer. 875d0e9fe17SLinus Torvalds * 876d0e9fe17SLinus Torvalds * IOW: that the "list_is_singular()" test on the anon_vma_chain only 877d0e9fe17SLinus Torvalds * matters for the 'stable anon_vma' case (ie the thing we want to avoid 878d0e9fe17SLinus Torvalds * is to return an anon_vma that is "complex" due to having gone through 879d0e9fe17SLinus Torvalds * a fork). 880d0e9fe17SLinus Torvalds * 881d0e9fe17SLinus Torvalds * We also make sure that the two vma's are compatible (adjacent, 882d0e9fe17SLinus Torvalds * and with the same memory policies). That's all stable, even with just 883d0e9fe17SLinus Torvalds * a read lock on the mm_sem. 884d0e9fe17SLinus Torvalds */ 885d0e9fe17SLinus Torvalds static struct anon_vma *reusable_anon_vma(struct vm_area_struct *old, struct vm_area_struct *a, struct vm_area_struct *b) 886d0e9fe17SLinus Torvalds { 887d0e9fe17SLinus Torvalds if (anon_vma_compatible(a, b)) { 888d0e9fe17SLinus Torvalds struct anon_vma *anon_vma = ACCESS_ONCE(old->anon_vma); 889d0e9fe17SLinus Torvalds 890d0e9fe17SLinus Torvalds if (anon_vma && list_is_singular(&old->anon_vma_chain)) 891d0e9fe17SLinus Torvalds return anon_vma; 892d0e9fe17SLinus Torvalds } 893d0e9fe17SLinus Torvalds return NULL; 894d0e9fe17SLinus Torvalds } 895d0e9fe17SLinus Torvalds 896d0e9fe17SLinus Torvalds /* 8971da177e4SLinus Torvalds * find_mergeable_anon_vma is used by anon_vma_prepare, to check 8981da177e4SLinus Torvalds * neighbouring vmas for a suitable anon_vma, before it goes off 8991da177e4SLinus Torvalds * to allocate a new anon_vma. It checks because a repetitive 9001da177e4SLinus Torvalds * sequence of mprotects and faults may otherwise lead to distinct 9011da177e4SLinus Torvalds * anon_vmas being allocated, preventing vma merge in subsequent 9021da177e4SLinus Torvalds * mprotect. 9031da177e4SLinus Torvalds */ 9041da177e4SLinus Torvalds struct anon_vma *find_mergeable_anon_vma(struct vm_area_struct *vma) 9051da177e4SLinus Torvalds { 906d0e9fe17SLinus Torvalds struct anon_vma *anon_vma; 9071da177e4SLinus Torvalds struct vm_area_struct *near; 9081da177e4SLinus Torvalds 9091da177e4SLinus Torvalds near = vma->vm_next; 9101da177e4SLinus Torvalds if (!near) 9111da177e4SLinus Torvalds goto try_prev; 9121da177e4SLinus Torvalds 913d0e9fe17SLinus Torvalds anon_vma = reusable_anon_vma(near, vma, near); 914d0e9fe17SLinus Torvalds if (anon_vma) 915d0e9fe17SLinus Torvalds return anon_vma; 9161da177e4SLinus Torvalds try_prev: 9171da177e4SLinus Torvalds /* 9181da177e4SLinus Torvalds * It is potentially slow to have to call find_vma_prev here. 9191da177e4SLinus Torvalds * But it's only on the first write fault on the vma, not 9201da177e4SLinus Torvalds * every time, and we could devise a way to avoid it later 9211da177e4SLinus Torvalds * (e.g. stash info in next's anon_vma_node when assigning 9221da177e4SLinus Torvalds * an anon_vma, or when trying vma_merge). Another time. 9231da177e4SLinus Torvalds */ 92446a350efSEric Sesterhenn BUG_ON(find_vma_prev(vma->vm_mm, vma->vm_start, &near) != vma); 9251da177e4SLinus Torvalds if (!near) 9261da177e4SLinus Torvalds goto none; 9271da177e4SLinus Torvalds 928d0e9fe17SLinus Torvalds anon_vma = reusable_anon_vma(near, near, vma); 929d0e9fe17SLinus Torvalds if (anon_vma) 930d0e9fe17SLinus Torvalds return anon_vma; 9311da177e4SLinus Torvalds none: 9321da177e4SLinus Torvalds /* 9331da177e4SLinus Torvalds * There's no absolute need to look only at touching neighbours: 9341da177e4SLinus Torvalds * we could search further afield for "compatible" anon_vmas. 9351da177e4SLinus Torvalds * But it would probably just be a waste of time searching, 9361da177e4SLinus Torvalds * or lead to too many vmas hanging off the same anon_vma. 9371da177e4SLinus Torvalds * We're trying to allow mprotect remerging later on, 9381da177e4SLinus Torvalds * not trying to minimize memory used for anon_vmas. 9391da177e4SLinus Torvalds */ 9401da177e4SLinus Torvalds return NULL; 9411da177e4SLinus Torvalds } 9421da177e4SLinus Torvalds 9431da177e4SLinus Torvalds #ifdef CONFIG_PROC_FS 944ab50b8edSHugh Dickins void vm_stat_account(struct mm_struct *mm, unsigned long flags, 9451da177e4SLinus Torvalds struct file *file, long pages) 9461da177e4SLinus Torvalds { 9471da177e4SLinus Torvalds const unsigned long stack_flags 9481da177e4SLinus Torvalds = VM_STACK_FLAGS & (VM_GROWSUP|VM_GROWSDOWN); 9491da177e4SLinus Torvalds 9501da177e4SLinus Torvalds if (file) { 9511da177e4SLinus Torvalds mm->shared_vm += pages; 9521da177e4SLinus Torvalds if ((flags & (VM_EXEC|VM_WRITE)) == VM_EXEC) 9531da177e4SLinus Torvalds mm->exec_vm += pages; 9541da177e4SLinus Torvalds } else if (flags & stack_flags) 9551da177e4SLinus Torvalds mm->stack_vm += pages; 9561da177e4SLinus Torvalds if (flags & (VM_RESERVED|VM_IO)) 9571da177e4SLinus Torvalds mm->reserved_vm += pages; 9581da177e4SLinus Torvalds } 9591da177e4SLinus Torvalds #endif /* CONFIG_PROC_FS */ 9601da177e4SLinus Torvalds 9611da177e4SLinus Torvalds /* 96227f5de79SJianjun Kong * The caller must hold down_write(¤t->mm->mmap_sem). 9631da177e4SLinus Torvalds */ 9641da177e4SLinus Torvalds 9651da177e4SLinus Torvalds unsigned long do_mmap_pgoff(struct file *file, unsigned long addr, 9661da177e4SLinus Torvalds unsigned long len, unsigned long prot, 9671da177e4SLinus Torvalds unsigned long flags, unsigned long pgoff) 9681da177e4SLinus Torvalds { 9691da177e4SLinus Torvalds struct mm_struct * mm = current->mm; 9701da177e4SLinus Torvalds struct inode *inode; 9711da177e4SLinus Torvalds unsigned int vm_flags; 9721da177e4SLinus Torvalds int error; 9730165ab44SMiklos Szeredi unsigned long reqprot = prot; 9741da177e4SLinus Torvalds 9751da177e4SLinus Torvalds /* 9761da177e4SLinus Torvalds * Does the application expect PROT_READ to imply PROT_EXEC? 9771da177e4SLinus Torvalds * 9781da177e4SLinus Torvalds * (the exception is when the underlying filesystem is noexec 9791da177e4SLinus Torvalds * mounted, in which case we dont add PROT_EXEC.) 9801da177e4SLinus Torvalds */ 9811da177e4SLinus Torvalds if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC)) 982d3ac7f89SJosef "Jeff" Sipek if (!(file && (file->f_path.mnt->mnt_flags & MNT_NOEXEC))) 9831da177e4SLinus Torvalds prot |= PROT_EXEC; 9841da177e4SLinus Torvalds 9851da177e4SLinus Torvalds if (!len) 9861da177e4SLinus Torvalds return -EINVAL; 9871da177e4SLinus Torvalds 9887cd94146SEric Paris if (!(flags & MAP_FIXED)) 9897cd94146SEric Paris addr = round_hint_to_min(addr); 9907cd94146SEric Paris 9911da177e4SLinus Torvalds /* Careful about overflows.. */ 9921da177e4SLinus Torvalds len = PAGE_ALIGN(len); 9939206de95SAl Viro if (!len) 9941da177e4SLinus Torvalds return -ENOMEM; 9951da177e4SLinus Torvalds 9961da177e4SLinus Torvalds /* offset overflow? */ 9971da177e4SLinus Torvalds if ((pgoff + (len >> PAGE_SHIFT)) < pgoff) 9981da177e4SLinus Torvalds return -EOVERFLOW; 9991da177e4SLinus Torvalds 10001da177e4SLinus Torvalds /* Too many mappings? */ 10011da177e4SLinus Torvalds if (mm->map_count > sysctl_max_map_count) 10021da177e4SLinus Torvalds return -ENOMEM; 10031da177e4SLinus Torvalds 10041da177e4SLinus Torvalds /* Obtain the address to map to. we verify (or select) it and ensure 10051da177e4SLinus Torvalds * that it represents a valid section of the address space. 10061da177e4SLinus Torvalds */ 10071da177e4SLinus Torvalds addr = get_unmapped_area(file, addr, len, pgoff, flags); 10081da177e4SLinus Torvalds if (addr & ~PAGE_MASK) 10091da177e4SLinus Torvalds return addr; 10101da177e4SLinus Torvalds 10111da177e4SLinus Torvalds /* Do simple checking here so the lower-level routines won't have 10121da177e4SLinus Torvalds * to. we assume access permissions have been handled by the open 10131da177e4SLinus Torvalds * of the memory object, so we don't do any here. 10141da177e4SLinus Torvalds */ 10151da177e4SLinus Torvalds vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags) | 10161da177e4SLinus Torvalds mm->def_flags | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC; 10171da177e4SLinus Torvalds 1018cdf7b341SHuang Shijie if (flags & MAP_LOCKED) 10191da177e4SLinus Torvalds if (!can_do_mlock()) 10201da177e4SLinus Torvalds return -EPERM; 1021ba470de4SRik van Riel 10221da177e4SLinus Torvalds /* mlock MCL_FUTURE? */ 10231da177e4SLinus Torvalds if (vm_flags & VM_LOCKED) { 10241da177e4SLinus Torvalds unsigned long locked, lock_limit; 102593ea1d0aSChris Wright locked = len >> PAGE_SHIFT; 102693ea1d0aSChris Wright locked += mm->locked_vm; 102759e99e5bSJiri Slaby lock_limit = rlimit(RLIMIT_MEMLOCK); 102893ea1d0aSChris Wright lock_limit >>= PAGE_SHIFT; 10291da177e4SLinus Torvalds if (locked > lock_limit && !capable(CAP_IPC_LOCK)) 10301da177e4SLinus Torvalds return -EAGAIN; 10311da177e4SLinus Torvalds } 10321da177e4SLinus Torvalds 1033d3ac7f89SJosef "Jeff" Sipek inode = file ? file->f_path.dentry->d_inode : NULL; 10341da177e4SLinus Torvalds 10351da177e4SLinus Torvalds if (file) { 10361da177e4SLinus Torvalds switch (flags & MAP_TYPE) { 10371da177e4SLinus Torvalds case MAP_SHARED: 10381da177e4SLinus Torvalds if ((prot&PROT_WRITE) && !(file->f_mode&FMODE_WRITE)) 10391da177e4SLinus Torvalds return -EACCES; 10401da177e4SLinus Torvalds 10411da177e4SLinus Torvalds /* 10421da177e4SLinus Torvalds * Make sure we don't allow writing to an append-only 10431da177e4SLinus Torvalds * file.. 10441da177e4SLinus Torvalds */ 10451da177e4SLinus Torvalds if (IS_APPEND(inode) && (file->f_mode & FMODE_WRITE)) 10461da177e4SLinus Torvalds return -EACCES; 10471da177e4SLinus Torvalds 10481da177e4SLinus Torvalds /* 10491da177e4SLinus Torvalds * Make sure there are no mandatory locks on the file. 10501da177e4SLinus Torvalds */ 10511da177e4SLinus Torvalds if (locks_verify_locked(inode)) 10521da177e4SLinus Torvalds return -EAGAIN; 10531da177e4SLinus Torvalds 10541da177e4SLinus Torvalds vm_flags |= VM_SHARED | VM_MAYSHARE; 10551da177e4SLinus Torvalds if (!(file->f_mode & FMODE_WRITE)) 10561da177e4SLinus Torvalds vm_flags &= ~(VM_MAYWRITE | VM_SHARED); 10571da177e4SLinus Torvalds 10581da177e4SLinus Torvalds /* fall through */ 10591da177e4SLinus Torvalds case MAP_PRIVATE: 10601da177e4SLinus Torvalds if (!(file->f_mode & FMODE_READ)) 10611da177e4SLinus Torvalds return -EACCES; 1062d3ac7f89SJosef "Jeff" Sipek if (file->f_path.mnt->mnt_flags & MNT_NOEXEC) { 106380c5606cSLinus Torvalds if (vm_flags & VM_EXEC) 106480c5606cSLinus Torvalds return -EPERM; 106580c5606cSLinus Torvalds vm_flags &= ~VM_MAYEXEC; 106680c5606cSLinus Torvalds } 106780c5606cSLinus Torvalds 106880c5606cSLinus Torvalds if (!file->f_op || !file->f_op->mmap) 106980c5606cSLinus Torvalds return -ENODEV; 10701da177e4SLinus Torvalds break; 10711da177e4SLinus Torvalds 10721da177e4SLinus Torvalds default: 10731da177e4SLinus Torvalds return -EINVAL; 10741da177e4SLinus Torvalds } 10751da177e4SLinus Torvalds } else { 10761da177e4SLinus Torvalds switch (flags & MAP_TYPE) { 10771da177e4SLinus Torvalds case MAP_SHARED: 1078ce363942STejun Heo /* 1079ce363942STejun Heo * Ignore pgoff. 1080ce363942STejun Heo */ 1081ce363942STejun Heo pgoff = 0; 10821da177e4SLinus Torvalds vm_flags |= VM_SHARED | VM_MAYSHARE; 10831da177e4SLinus Torvalds break; 10841da177e4SLinus Torvalds case MAP_PRIVATE: 10851da177e4SLinus Torvalds /* 10861da177e4SLinus Torvalds * Set pgoff according to addr for anon_vma. 10871da177e4SLinus Torvalds */ 10881da177e4SLinus Torvalds pgoff = addr >> PAGE_SHIFT; 10891da177e4SLinus Torvalds break; 10901da177e4SLinus Torvalds default: 10911da177e4SLinus Torvalds return -EINVAL; 10921da177e4SLinus Torvalds } 10931da177e4SLinus Torvalds } 10941da177e4SLinus Torvalds 1095ed032189SEric Paris error = security_file_mmap(file, reqprot, prot, flags, addr, 0); 10961da177e4SLinus Torvalds if (error) 10971da177e4SLinus Torvalds return error; 10981da177e4SLinus Torvalds 10995a6fe125SMel Gorman return mmap_region(file, addr, len, flags, vm_flags, pgoff); 11000165ab44SMiklos Szeredi } 11010165ab44SMiklos Szeredi EXPORT_SYMBOL(do_mmap_pgoff); 11020165ab44SMiklos Szeredi 110366f0dc48SHugh Dickins SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len, 110466f0dc48SHugh Dickins unsigned long, prot, unsigned long, flags, 110566f0dc48SHugh Dickins unsigned long, fd, unsigned long, pgoff) 110666f0dc48SHugh Dickins { 110766f0dc48SHugh Dickins struct file *file = NULL; 110866f0dc48SHugh Dickins unsigned long retval = -EBADF; 110966f0dc48SHugh Dickins 111066f0dc48SHugh Dickins if (!(flags & MAP_ANONYMOUS)) { 111166f0dc48SHugh Dickins if (unlikely(flags & MAP_HUGETLB)) 111266f0dc48SHugh Dickins return -EINVAL; 111366f0dc48SHugh Dickins file = fget(fd); 111466f0dc48SHugh Dickins if (!file) 111566f0dc48SHugh Dickins goto out; 111666f0dc48SHugh Dickins } else if (flags & MAP_HUGETLB) { 111766f0dc48SHugh Dickins struct user_struct *user = NULL; 111866f0dc48SHugh Dickins /* 111966f0dc48SHugh Dickins * VM_NORESERVE is used because the reservations will be 112066f0dc48SHugh Dickins * taken when vm_ops->mmap() is called 112166f0dc48SHugh Dickins * A dummy user value is used because we are not locking 112266f0dc48SHugh Dickins * memory so no accounting is necessary 112366f0dc48SHugh Dickins */ 112466f0dc48SHugh Dickins len = ALIGN(len, huge_page_size(&default_hstate)); 112566f0dc48SHugh Dickins file = hugetlb_file_setup(HUGETLB_ANON_FILE, len, VM_NORESERVE, 112666f0dc48SHugh Dickins &user, HUGETLB_ANONHUGE_INODE); 112766f0dc48SHugh Dickins if (IS_ERR(file)) 112866f0dc48SHugh Dickins return PTR_ERR(file); 112966f0dc48SHugh Dickins } 113066f0dc48SHugh Dickins 113166f0dc48SHugh Dickins flags &= ~(MAP_EXECUTABLE | MAP_DENYWRITE); 113266f0dc48SHugh Dickins 113366f0dc48SHugh Dickins down_write(¤t->mm->mmap_sem); 113466f0dc48SHugh Dickins retval = do_mmap_pgoff(file, addr, len, prot, flags, pgoff); 113566f0dc48SHugh Dickins up_write(¤t->mm->mmap_sem); 113666f0dc48SHugh Dickins 113766f0dc48SHugh Dickins if (file) 113866f0dc48SHugh Dickins fput(file); 113966f0dc48SHugh Dickins out: 114066f0dc48SHugh Dickins return retval; 114166f0dc48SHugh Dickins } 114266f0dc48SHugh Dickins 1143a4679373SChristoph Hellwig #ifdef __ARCH_WANT_SYS_OLD_MMAP 1144a4679373SChristoph Hellwig struct mmap_arg_struct { 1145a4679373SChristoph Hellwig unsigned long addr; 1146a4679373SChristoph Hellwig unsigned long len; 1147a4679373SChristoph Hellwig unsigned long prot; 1148a4679373SChristoph Hellwig unsigned long flags; 1149a4679373SChristoph Hellwig unsigned long fd; 1150a4679373SChristoph Hellwig unsigned long offset; 1151a4679373SChristoph Hellwig }; 1152a4679373SChristoph Hellwig 1153a4679373SChristoph Hellwig SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg) 1154a4679373SChristoph Hellwig { 1155a4679373SChristoph Hellwig struct mmap_arg_struct a; 1156a4679373SChristoph Hellwig 1157a4679373SChristoph Hellwig if (copy_from_user(&a, arg, sizeof(a))) 1158a4679373SChristoph Hellwig return -EFAULT; 1159a4679373SChristoph Hellwig if (a.offset & ~PAGE_MASK) 1160a4679373SChristoph Hellwig return -EINVAL; 1161a4679373SChristoph Hellwig 1162a4679373SChristoph Hellwig return sys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd, 1163a4679373SChristoph Hellwig a.offset >> PAGE_SHIFT); 1164a4679373SChristoph Hellwig } 1165a4679373SChristoph Hellwig #endif /* __ARCH_WANT_SYS_OLD_MMAP */ 1166a4679373SChristoph Hellwig 11674e950f6fSAlexey Dobriyan /* 11684e950f6fSAlexey Dobriyan * Some shared mappigns will want the pages marked read-only 11694e950f6fSAlexey Dobriyan * to track write events. If so, we'll downgrade vm_page_prot 11704e950f6fSAlexey Dobriyan * to the private version (using protection_map[] without the 11714e950f6fSAlexey Dobriyan * VM_SHARED bit). 11724e950f6fSAlexey Dobriyan */ 11734e950f6fSAlexey Dobriyan int vma_wants_writenotify(struct vm_area_struct *vma) 11744e950f6fSAlexey Dobriyan { 11754e950f6fSAlexey Dobriyan unsigned int vm_flags = vma->vm_flags; 11764e950f6fSAlexey Dobriyan 11774e950f6fSAlexey Dobriyan /* If it was private or non-writable, the write bit is already clear */ 11784e950f6fSAlexey Dobriyan if ((vm_flags & (VM_WRITE|VM_SHARED)) != ((VM_WRITE|VM_SHARED))) 11794e950f6fSAlexey Dobriyan return 0; 11804e950f6fSAlexey Dobriyan 11814e950f6fSAlexey Dobriyan /* The backer wishes to know when pages are first written to? */ 11824e950f6fSAlexey Dobriyan if (vma->vm_ops && vma->vm_ops->page_mkwrite) 11834e950f6fSAlexey Dobriyan return 1; 11844e950f6fSAlexey Dobriyan 11854e950f6fSAlexey Dobriyan /* The open routine did something to the protections already? */ 11864e950f6fSAlexey Dobriyan if (pgprot_val(vma->vm_page_prot) != 11873ed75eb8SColy Li pgprot_val(vm_get_page_prot(vm_flags))) 11884e950f6fSAlexey Dobriyan return 0; 11894e950f6fSAlexey Dobriyan 11904e950f6fSAlexey Dobriyan /* Specialty mapping? */ 11914e950f6fSAlexey Dobriyan if (vm_flags & (VM_PFNMAP|VM_INSERTPAGE)) 11924e950f6fSAlexey Dobriyan return 0; 11934e950f6fSAlexey Dobriyan 11944e950f6fSAlexey Dobriyan /* Can the mapping track the dirty pages? */ 11954e950f6fSAlexey Dobriyan return vma->vm_file && vma->vm_file->f_mapping && 11964e950f6fSAlexey Dobriyan mapping_cap_account_dirty(vma->vm_file->f_mapping); 11974e950f6fSAlexey Dobriyan } 11984e950f6fSAlexey Dobriyan 1199fc8744adSLinus Torvalds /* 1200fc8744adSLinus Torvalds * We account for memory if it's a private writeable mapping, 12015a6fe125SMel Gorman * not hugepages and VM_NORESERVE wasn't set. 1202fc8744adSLinus Torvalds */ 12035a6fe125SMel Gorman static inline int accountable_mapping(struct file *file, unsigned int vm_flags) 1204fc8744adSLinus Torvalds { 12055a6fe125SMel Gorman /* 12065a6fe125SMel Gorman * hugetlb has its own accounting separate from the core VM 12075a6fe125SMel Gorman * VM_HUGETLB may not be set yet so we cannot check for that flag. 12085a6fe125SMel Gorman */ 12095a6fe125SMel Gorman if (file && is_file_hugepages(file)) 12105a6fe125SMel Gorman return 0; 12115a6fe125SMel Gorman 1212fc8744adSLinus Torvalds return (vm_flags & (VM_NORESERVE | VM_SHARED | VM_WRITE)) == VM_WRITE; 1213fc8744adSLinus Torvalds } 1214fc8744adSLinus Torvalds 12150165ab44SMiklos Szeredi unsigned long mmap_region(struct file *file, unsigned long addr, 12160165ab44SMiklos Szeredi unsigned long len, unsigned long flags, 12175a6fe125SMel Gorman unsigned int vm_flags, unsigned long pgoff) 12180165ab44SMiklos Szeredi { 12190165ab44SMiklos Szeredi struct mm_struct *mm = current->mm; 12200165ab44SMiklos Szeredi struct vm_area_struct *vma, *prev; 12210165ab44SMiklos Szeredi int correct_wcount = 0; 12220165ab44SMiklos Szeredi int error; 12230165ab44SMiklos Szeredi struct rb_node **rb_link, *rb_parent; 12240165ab44SMiklos Szeredi unsigned long charged = 0; 12250165ab44SMiklos Szeredi struct inode *inode = file ? file->f_path.dentry->d_inode : NULL; 12260165ab44SMiklos Szeredi 12271da177e4SLinus Torvalds /* Clear old maps */ 12281da177e4SLinus Torvalds error = -ENOMEM; 12291da177e4SLinus Torvalds munmap_back: 12301da177e4SLinus Torvalds vma = find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent); 12311da177e4SLinus Torvalds if (vma && vma->vm_start < addr + len) { 12321da177e4SLinus Torvalds if (do_munmap(mm, addr, len)) 12331da177e4SLinus Torvalds return -ENOMEM; 12341da177e4SLinus Torvalds goto munmap_back; 12351da177e4SLinus Torvalds } 12361da177e4SLinus Torvalds 12371da177e4SLinus Torvalds /* Check against address space limit. */ 1238119f657cSakpm@osdl.org if (!may_expand_vm(mm, len >> PAGE_SHIFT)) 12391da177e4SLinus Torvalds return -ENOMEM; 12401da177e4SLinus Torvalds 1241fc8744adSLinus Torvalds /* 1242fc8744adSLinus Torvalds * Set 'VM_NORESERVE' if we should not account for the 12435a6fe125SMel Gorman * memory use of this mapping. 1244fc8744adSLinus Torvalds */ 12455a6fe125SMel Gorman if ((flags & MAP_NORESERVE)) { 12465a6fe125SMel Gorman /* We honor MAP_NORESERVE if allowed to overcommit */ 12475a6fe125SMel Gorman if (sysctl_overcommit_memory != OVERCOMMIT_NEVER) 1248fc8744adSLinus Torvalds vm_flags |= VM_NORESERVE; 12495a6fe125SMel Gorman 12505a6fe125SMel Gorman /* hugetlb applies strict overcommit unless MAP_NORESERVE */ 12515a6fe125SMel Gorman if (file && is_file_hugepages(file)) 1252cdfd4325SAndy Whitcroft vm_flags |= VM_NORESERVE; 12535a6fe125SMel Gorman } 1254cdfd4325SAndy Whitcroft 12551da177e4SLinus Torvalds /* 12561da177e4SLinus Torvalds * Private writable mapping: check memory availability 12571da177e4SLinus Torvalds */ 12585a6fe125SMel Gorman if (accountable_mapping(file, vm_flags)) { 12591da177e4SLinus Torvalds charged = len >> PAGE_SHIFT; 12601da177e4SLinus Torvalds if (security_vm_enough_memory(charged)) 12611da177e4SLinus Torvalds return -ENOMEM; 12621da177e4SLinus Torvalds vm_flags |= VM_ACCOUNT; 12631da177e4SLinus Torvalds } 12641da177e4SLinus Torvalds 12651da177e4SLinus Torvalds /* 1266de33c8dbSLinus Torvalds * Can we just expand an old mapping? 12671da177e4SLinus Torvalds */ 1268de33c8dbSLinus Torvalds vma = vma_merge(mm, prev, addr, addr + len, vm_flags, NULL, file, pgoff, NULL); 1269ba470de4SRik van Riel if (vma) 12701da177e4SLinus Torvalds goto out; 12711da177e4SLinus Torvalds 12721da177e4SLinus Torvalds /* 12731da177e4SLinus Torvalds * Determine the object being mapped and call the appropriate 12741da177e4SLinus Torvalds * specific mapper. the address has already been validated, but 12751da177e4SLinus Torvalds * not unmapped, but the maps are removed from the list. 12761da177e4SLinus Torvalds */ 1277c5e3b83eSPekka Enberg vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL); 12781da177e4SLinus Torvalds if (!vma) { 12791da177e4SLinus Torvalds error = -ENOMEM; 12801da177e4SLinus Torvalds goto unacct_error; 12811da177e4SLinus Torvalds } 12821da177e4SLinus Torvalds 12831da177e4SLinus Torvalds vma->vm_mm = mm; 12841da177e4SLinus Torvalds vma->vm_start = addr; 12851da177e4SLinus Torvalds vma->vm_end = addr + len; 12861da177e4SLinus Torvalds vma->vm_flags = vm_flags; 12873ed75eb8SColy Li vma->vm_page_prot = vm_get_page_prot(vm_flags); 12881da177e4SLinus Torvalds vma->vm_pgoff = pgoff; 12895beb4930SRik van Riel INIT_LIST_HEAD(&vma->anon_vma_chain); 12901da177e4SLinus Torvalds 12911da177e4SLinus Torvalds if (file) { 12921da177e4SLinus Torvalds error = -EINVAL; 12931da177e4SLinus Torvalds if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP)) 12941da177e4SLinus Torvalds goto free_vma; 12951da177e4SLinus Torvalds if (vm_flags & VM_DENYWRITE) { 12961da177e4SLinus Torvalds error = deny_write_access(file); 12971da177e4SLinus Torvalds if (error) 12981da177e4SLinus Torvalds goto free_vma; 12991da177e4SLinus Torvalds correct_wcount = 1; 13001da177e4SLinus Torvalds } 13011da177e4SLinus Torvalds vma->vm_file = file; 13021da177e4SLinus Torvalds get_file(file); 13031da177e4SLinus Torvalds error = file->f_op->mmap(file, vma); 13041da177e4SLinus Torvalds if (error) 13051da177e4SLinus Torvalds goto unmap_and_free_vma; 1306925d1c40SMatt Helsley if (vm_flags & VM_EXECUTABLE) 1307925d1c40SMatt Helsley added_exe_file_vma(mm); 13081da177e4SLinus Torvalds 13091da177e4SLinus Torvalds /* Can addr have changed?? 13101da177e4SLinus Torvalds * 13111da177e4SLinus Torvalds * Answer: Yes, several device drivers can do it in their 13121da177e4SLinus Torvalds * f_op->mmap method. -DaveM 13131da177e4SLinus Torvalds */ 13141da177e4SLinus Torvalds addr = vma->vm_start; 13151da177e4SLinus Torvalds pgoff = vma->vm_pgoff; 13161da177e4SLinus Torvalds vm_flags = vma->vm_flags; 1317f8dbf0a7SHuang Shijie } else if (vm_flags & VM_SHARED) { 1318f8dbf0a7SHuang Shijie error = shmem_zero_setup(vma); 1319f8dbf0a7SHuang Shijie if (error) 1320f8dbf0a7SHuang Shijie goto free_vma; 1321f8dbf0a7SHuang Shijie } 13221da177e4SLinus Torvalds 1323c9d0bf24SMagnus Damm if (vma_wants_writenotify(vma)) { 1324c9d0bf24SMagnus Damm pgprot_t pprot = vma->vm_page_prot; 1325c9d0bf24SMagnus Damm 1326c9d0bf24SMagnus Damm /* Can vma->vm_page_prot have changed?? 1327c9d0bf24SMagnus Damm * 1328c9d0bf24SMagnus Damm * Answer: Yes, drivers may have changed it in their 1329c9d0bf24SMagnus Damm * f_op->mmap method. 1330c9d0bf24SMagnus Damm * 1331c9d0bf24SMagnus Damm * Ensures that vmas marked as uncached stay that way. 1332c9d0bf24SMagnus Damm */ 13331ddd439eSHugh Dickins vma->vm_page_prot = vm_get_page_prot(vm_flags & ~VM_SHARED); 1334c9d0bf24SMagnus Damm if (pgprot_val(pprot) == pgprot_val(pgprot_noncached(pprot))) 1335c9d0bf24SMagnus Damm vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot); 1336c9d0bf24SMagnus Damm } 1337d08b3851SPeter Zijlstra 13384d3d5b41SOleg Nesterov vma_link(mm, vma, prev, rb_link, rb_parent); 13394d3d5b41SOleg Nesterov file = vma->vm_file; 13404d3d5b41SOleg Nesterov 13414d3d5b41SOleg Nesterov /* Once vma denies write, undo our temporary denial count */ 13424d3d5b41SOleg Nesterov if (correct_wcount) 13434d3d5b41SOleg Nesterov atomic_inc(&inode->i_writecount); 13441da177e4SLinus Torvalds out: 1345cdd6c482SIngo Molnar perf_event_mmap(vma); 13460a4a9391SPeter Zijlstra 13471da177e4SLinus Torvalds mm->total_vm += len >> PAGE_SHIFT; 1348ab50b8edSHugh Dickins vm_stat_account(mm, vm_flags, file, len >> PAGE_SHIFT); 13491da177e4SLinus Torvalds if (vm_flags & VM_LOCKED) { 135006f9d8c2SKOSAKI Motohiro if (!mlock_vma_pages_range(vma, addr, addr + len)) 135106f9d8c2SKOSAKI Motohiro mm->locked_vm += (len >> PAGE_SHIFT); 1352ba470de4SRik van Riel } else if ((flags & MAP_POPULATE) && !(flags & MAP_NONBLOCK)) 135354cb8821SNick Piggin make_pages_present(addr, addr + len); 13541da177e4SLinus Torvalds return addr; 13551da177e4SLinus Torvalds 13561da177e4SLinus Torvalds unmap_and_free_vma: 13571da177e4SLinus Torvalds if (correct_wcount) 13581da177e4SLinus Torvalds atomic_inc(&inode->i_writecount); 13591da177e4SLinus Torvalds vma->vm_file = NULL; 13601da177e4SLinus Torvalds fput(file); 13611da177e4SLinus Torvalds 13621da177e4SLinus Torvalds /* Undo any partial mapping done by a device driver. */ 1363e0da382cSHugh Dickins unmap_region(mm, vma, prev, vma->vm_start, vma->vm_end); 1364e0da382cSHugh Dickins charged = 0; 13651da177e4SLinus Torvalds free_vma: 13661da177e4SLinus Torvalds kmem_cache_free(vm_area_cachep, vma); 13671da177e4SLinus Torvalds unacct_error: 13681da177e4SLinus Torvalds if (charged) 13691da177e4SLinus Torvalds vm_unacct_memory(charged); 13701da177e4SLinus Torvalds return error; 13711da177e4SLinus Torvalds } 13721da177e4SLinus Torvalds 13731da177e4SLinus Torvalds /* Get an address range which is currently unmapped. 13741da177e4SLinus Torvalds * For shmat() with addr=0. 13751da177e4SLinus Torvalds * 13761da177e4SLinus Torvalds * Ugly calling convention alert: 13771da177e4SLinus Torvalds * Return value with the low bits set means error value, 13781da177e4SLinus Torvalds * ie 13791da177e4SLinus Torvalds * if (ret & ~PAGE_MASK) 13801da177e4SLinus Torvalds * error = ret; 13811da177e4SLinus Torvalds * 13821da177e4SLinus Torvalds * This function "knows" that -ENOMEM has the bits set. 13831da177e4SLinus Torvalds */ 13841da177e4SLinus Torvalds #ifndef HAVE_ARCH_UNMAPPED_AREA 13851da177e4SLinus Torvalds unsigned long 13861da177e4SLinus Torvalds arch_get_unmapped_area(struct file *filp, unsigned long addr, 13871da177e4SLinus Torvalds unsigned long len, unsigned long pgoff, unsigned long flags) 13881da177e4SLinus Torvalds { 13891da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 13901da177e4SLinus Torvalds struct vm_area_struct *vma; 13911da177e4SLinus Torvalds unsigned long start_addr; 13921da177e4SLinus Torvalds 13931da177e4SLinus Torvalds if (len > TASK_SIZE) 13941da177e4SLinus Torvalds return -ENOMEM; 13951da177e4SLinus Torvalds 139606abdfb4SBenjamin Herrenschmidt if (flags & MAP_FIXED) 139706abdfb4SBenjamin Herrenschmidt return addr; 139806abdfb4SBenjamin Herrenschmidt 13991da177e4SLinus Torvalds if (addr) { 14001da177e4SLinus Torvalds addr = PAGE_ALIGN(addr); 14011da177e4SLinus Torvalds vma = find_vma(mm, addr); 14021da177e4SLinus Torvalds if (TASK_SIZE - len >= addr && 14031da177e4SLinus Torvalds (!vma || addr + len <= vma->vm_start)) 14041da177e4SLinus Torvalds return addr; 14051da177e4SLinus Torvalds } 14061363c3cdSWolfgang Wander if (len > mm->cached_hole_size) { 14071da177e4SLinus Torvalds start_addr = addr = mm->free_area_cache; 14081363c3cdSWolfgang Wander } else { 14091363c3cdSWolfgang Wander start_addr = addr = TASK_UNMAPPED_BASE; 14101363c3cdSWolfgang Wander mm->cached_hole_size = 0; 14111363c3cdSWolfgang Wander } 14121da177e4SLinus Torvalds 14131da177e4SLinus Torvalds full_search: 14141da177e4SLinus Torvalds for (vma = find_vma(mm, addr); ; vma = vma->vm_next) { 14151da177e4SLinus Torvalds /* At this point: (!vma || addr < vma->vm_end). */ 14161da177e4SLinus Torvalds if (TASK_SIZE - len < addr) { 14171da177e4SLinus Torvalds /* 14181da177e4SLinus Torvalds * Start a new search - just in case we missed 14191da177e4SLinus Torvalds * some holes. 14201da177e4SLinus Torvalds */ 14211da177e4SLinus Torvalds if (start_addr != TASK_UNMAPPED_BASE) { 14221363c3cdSWolfgang Wander addr = TASK_UNMAPPED_BASE; 14231363c3cdSWolfgang Wander start_addr = addr; 14241363c3cdSWolfgang Wander mm->cached_hole_size = 0; 14251da177e4SLinus Torvalds goto full_search; 14261da177e4SLinus Torvalds } 14271da177e4SLinus Torvalds return -ENOMEM; 14281da177e4SLinus Torvalds } 14291da177e4SLinus Torvalds if (!vma || addr + len <= vma->vm_start) { 14301da177e4SLinus Torvalds /* 14311da177e4SLinus Torvalds * Remember the place where we stopped the search: 14321da177e4SLinus Torvalds */ 14331da177e4SLinus Torvalds mm->free_area_cache = addr + len; 14341da177e4SLinus Torvalds return addr; 14351da177e4SLinus Torvalds } 14361363c3cdSWolfgang Wander if (addr + mm->cached_hole_size < vma->vm_start) 14371363c3cdSWolfgang Wander mm->cached_hole_size = vma->vm_start - addr; 14381da177e4SLinus Torvalds addr = vma->vm_end; 14391da177e4SLinus Torvalds } 14401da177e4SLinus Torvalds } 14411da177e4SLinus Torvalds #endif 14421da177e4SLinus Torvalds 14431363c3cdSWolfgang Wander void arch_unmap_area(struct mm_struct *mm, unsigned long addr) 14441da177e4SLinus Torvalds { 14451da177e4SLinus Torvalds /* 14461da177e4SLinus Torvalds * Is this a new hole at the lowest possible address? 14471da177e4SLinus Torvalds */ 14481363c3cdSWolfgang Wander if (addr >= TASK_UNMAPPED_BASE && addr < mm->free_area_cache) { 14491363c3cdSWolfgang Wander mm->free_area_cache = addr; 14501363c3cdSWolfgang Wander mm->cached_hole_size = ~0UL; 14511363c3cdSWolfgang Wander } 14521da177e4SLinus Torvalds } 14531da177e4SLinus Torvalds 14541da177e4SLinus Torvalds /* 14551da177e4SLinus Torvalds * This mmap-allocator allocates new areas top-down from below the 14561da177e4SLinus Torvalds * stack's low limit (the base): 14571da177e4SLinus Torvalds */ 14581da177e4SLinus Torvalds #ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN 14591da177e4SLinus Torvalds unsigned long 14601da177e4SLinus Torvalds arch_get_unmapped_area_topdown(struct file *filp, const unsigned long addr0, 14611da177e4SLinus Torvalds const unsigned long len, const unsigned long pgoff, 14621da177e4SLinus Torvalds const unsigned long flags) 14631da177e4SLinus Torvalds { 14641da177e4SLinus Torvalds struct vm_area_struct *vma; 14651da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 14661da177e4SLinus Torvalds unsigned long addr = addr0; 14671da177e4SLinus Torvalds 14681da177e4SLinus Torvalds /* requested length too big for entire address space */ 14691da177e4SLinus Torvalds if (len > TASK_SIZE) 14701da177e4SLinus Torvalds return -ENOMEM; 14711da177e4SLinus Torvalds 147206abdfb4SBenjamin Herrenschmidt if (flags & MAP_FIXED) 147306abdfb4SBenjamin Herrenschmidt return addr; 147406abdfb4SBenjamin Herrenschmidt 14751da177e4SLinus Torvalds /* requesting a specific address */ 14761da177e4SLinus Torvalds if (addr) { 14771da177e4SLinus Torvalds addr = PAGE_ALIGN(addr); 14781da177e4SLinus Torvalds vma = find_vma(mm, addr); 14791da177e4SLinus Torvalds if (TASK_SIZE - len >= addr && 14801da177e4SLinus Torvalds (!vma || addr + len <= vma->vm_start)) 14811da177e4SLinus Torvalds return addr; 14821da177e4SLinus Torvalds } 14831da177e4SLinus Torvalds 14841363c3cdSWolfgang Wander /* check if free_area_cache is useful for us */ 14851363c3cdSWolfgang Wander if (len <= mm->cached_hole_size) { 14861363c3cdSWolfgang Wander mm->cached_hole_size = 0; 14871363c3cdSWolfgang Wander mm->free_area_cache = mm->mmap_base; 14881363c3cdSWolfgang Wander } 14891363c3cdSWolfgang Wander 14901da177e4SLinus Torvalds /* either no address requested or can't fit in requested address hole */ 14911da177e4SLinus Torvalds addr = mm->free_area_cache; 14921da177e4SLinus Torvalds 14931da177e4SLinus Torvalds /* make sure it can fit in the remaining address space */ 149449a43876SLinus Torvalds if (addr > len) { 14951da177e4SLinus Torvalds vma = find_vma(mm, addr-len); 14961da177e4SLinus Torvalds if (!vma || addr <= vma->vm_start) 14971da177e4SLinus Torvalds /* remember the address as a hint for next time */ 14981da177e4SLinus Torvalds return (mm->free_area_cache = addr-len); 14991da177e4SLinus Torvalds } 15001da177e4SLinus Torvalds 150173219d17SChris Wright if (mm->mmap_base < len) 150273219d17SChris Wright goto bottomup; 150373219d17SChris Wright 15041da177e4SLinus Torvalds addr = mm->mmap_base-len; 15051da177e4SLinus Torvalds 15061da177e4SLinus Torvalds do { 15071da177e4SLinus Torvalds /* 15081da177e4SLinus Torvalds * Lookup failure means no vma is above this address, 15091da177e4SLinus Torvalds * else if new region fits below vma->vm_start, 15101da177e4SLinus Torvalds * return with success: 15111da177e4SLinus Torvalds */ 15121da177e4SLinus Torvalds vma = find_vma(mm, addr); 15131da177e4SLinus Torvalds if (!vma || addr+len <= vma->vm_start) 15141da177e4SLinus Torvalds /* remember the address as a hint for next time */ 15151da177e4SLinus Torvalds return (mm->free_area_cache = addr); 15161da177e4SLinus Torvalds 15171363c3cdSWolfgang Wander /* remember the largest hole we saw so far */ 15181363c3cdSWolfgang Wander if (addr + mm->cached_hole_size < vma->vm_start) 15191363c3cdSWolfgang Wander mm->cached_hole_size = vma->vm_start - addr; 15201363c3cdSWolfgang Wander 15211da177e4SLinus Torvalds /* try just below the current vma->vm_start */ 15221da177e4SLinus Torvalds addr = vma->vm_start-len; 152349a43876SLinus Torvalds } while (len < vma->vm_start); 15241da177e4SLinus Torvalds 152573219d17SChris Wright bottomup: 15261da177e4SLinus Torvalds /* 15271da177e4SLinus Torvalds * A failed mmap() very likely causes application failure, 15281da177e4SLinus Torvalds * so fall back to the bottom-up function here. This scenario 15291da177e4SLinus Torvalds * can happen with large stack limits and large mmap() 15301da177e4SLinus Torvalds * allocations. 15311da177e4SLinus Torvalds */ 15321363c3cdSWolfgang Wander mm->cached_hole_size = ~0UL; 15331da177e4SLinus Torvalds mm->free_area_cache = TASK_UNMAPPED_BASE; 15341da177e4SLinus Torvalds addr = arch_get_unmapped_area(filp, addr0, len, pgoff, flags); 15351da177e4SLinus Torvalds /* 15361da177e4SLinus Torvalds * Restore the topdown base: 15371da177e4SLinus Torvalds */ 15381da177e4SLinus Torvalds mm->free_area_cache = mm->mmap_base; 15391363c3cdSWolfgang Wander mm->cached_hole_size = ~0UL; 15401da177e4SLinus Torvalds 15411da177e4SLinus Torvalds return addr; 15421da177e4SLinus Torvalds } 15431da177e4SLinus Torvalds #endif 15441da177e4SLinus Torvalds 15451363c3cdSWolfgang Wander void arch_unmap_area_topdown(struct mm_struct *mm, unsigned long addr) 15461da177e4SLinus Torvalds { 15471da177e4SLinus Torvalds /* 15481da177e4SLinus Torvalds * Is this a new hole at the highest possible address? 15491da177e4SLinus Torvalds */ 15501363c3cdSWolfgang Wander if (addr > mm->free_area_cache) 15511363c3cdSWolfgang Wander mm->free_area_cache = addr; 15521da177e4SLinus Torvalds 15531da177e4SLinus Torvalds /* dont allow allocations above current base */ 15541363c3cdSWolfgang Wander if (mm->free_area_cache > mm->mmap_base) 15551363c3cdSWolfgang Wander mm->free_area_cache = mm->mmap_base; 15561da177e4SLinus Torvalds } 15571da177e4SLinus Torvalds 15581da177e4SLinus Torvalds unsigned long 15591da177e4SLinus Torvalds get_unmapped_area(struct file *file, unsigned long addr, unsigned long len, 15601da177e4SLinus Torvalds unsigned long pgoff, unsigned long flags) 15611da177e4SLinus Torvalds { 156206abdfb4SBenjamin Herrenschmidt unsigned long (*get_area)(struct file *, unsigned long, 156306abdfb4SBenjamin Herrenschmidt unsigned long, unsigned long, unsigned long); 156407ab67c8SLinus Torvalds 15659206de95SAl Viro unsigned long error = arch_mmap_check(addr, len, flags); 15669206de95SAl Viro if (error) 15679206de95SAl Viro return error; 15689206de95SAl Viro 15699206de95SAl Viro /* Careful about overflows.. */ 15709206de95SAl Viro if (len > TASK_SIZE) 15719206de95SAl Viro return -ENOMEM; 15729206de95SAl Viro 157307ab67c8SLinus Torvalds get_area = current->mm->get_unmapped_area; 157407ab67c8SLinus Torvalds if (file && file->f_op && file->f_op->get_unmapped_area) 157507ab67c8SLinus Torvalds get_area = file->f_op->get_unmapped_area; 157607ab67c8SLinus Torvalds addr = get_area(file, addr, len, pgoff, flags); 157707ab67c8SLinus Torvalds if (IS_ERR_VALUE(addr)) 157807ab67c8SLinus Torvalds return addr; 157907ab67c8SLinus Torvalds 15801da177e4SLinus Torvalds if (addr > TASK_SIZE - len) 15811da177e4SLinus Torvalds return -ENOMEM; 15821da177e4SLinus Torvalds if (addr & ~PAGE_MASK) 15831da177e4SLinus Torvalds return -EINVAL; 158406abdfb4SBenjamin Herrenschmidt 158508e7d9b5SMartin Schwidefsky return arch_rebalance_pgtables(addr, len); 15861da177e4SLinus Torvalds } 15871da177e4SLinus Torvalds 15881da177e4SLinus Torvalds EXPORT_SYMBOL(get_unmapped_area); 15891da177e4SLinus Torvalds 15901da177e4SLinus Torvalds /* Look up the first VMA which satisfies addr < vm_end, NULL if none. */ 15911da177e4SLinus Torvalds struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr) 15921da177e4SLinus Torvalds { 15931da177e4SLinus Torvalds struct vm_area_struct *vma = NULL; 15941da177e4SLinus Torvalds 15951da177e4SLinus Torvalds if (mm) { 15961da177e4SLinus Torvalds /* Check the cache first. */ 15971da177e4SLinus Torvalds /* (Cache hit rate is typically around 35%.) */ 15981da177e4SLinus Torvalds vma = mm->mmap_cache; 15991da177e4SLinus Torvalds if (!(vma && vma->vm_end > addr && vma->vm_start <= addr)) { 16001da177e4SLinus Torvalds struct rb_node * rb_node; 16011da177e4SLinus Torvalds 16021da177e4SLinus Torvalds rb_node = mm->mm_rb.rb_node; 16031da177e4SLinus Torvalds vma = NULL; 16041da177e4SLinus Torvalds 16051da177e4SLinus Torvalds while (rb_node) { 16061da177e4SLinus Torvalds struct vm_area_struct * vma_tmp; 16071da177e4SLinus Torvalds 16081da177e4SLinus Torvalds vma_tmp = rb_entry(rb_node, 16091da177e4SLinus Torvalds struct vm_area_struct, vm_rb); 16101da177e4SLinus Torvalds 16111da177e4SLinus Torvalds if (vma_tmp->vm_end > addr) { 16121da177e4SLinus Torvalds vma = vma_tmp; 16131da177e4SLinus Torvalds if (vma_tmp->vm_start <= addr) 16141da177e4SLinus Torvalds break; 16151da177e4SLinus Torvalds rb_node = rb_node->rb_left; 16161da177e4SLinus Torvalds } else 16171da177e4SLinus Torvalds rb_node = rb_node->rb_right; 16181da177e4SLinus Torvalds } 16191da177e4SLinus Torvalds if (vma) 16201da177e4SLinus Torvalds mm->mmap_cache = vma; 16211da177e4SLinus Torvalds } 16221da177e4SLinus Torvalds } 16231da177e4SLinus Torvalds return vma; 16241da177e4SLinus Torvalds } 16251da177e4SLinus Torvalds 16261da177e4SLinus Torvalds EXPORT_SYMBOL(find_vma); 16271da177e4SLinus Torvalds 16281da177e4SLinus Torvalds /* Same as find_vma, but also return a pointer to the previous VMA in *pprev. */ 16291da177e4SLinus Torvalds struct vm_area_struct * 16301da177e4SLinus Torvalds find_vma_prev(struct mm_struct *mm, unsigned long addr, 16311da177e4SLinus Torvalds struct vm_area_struct **pprev) 16321da177e4SLinus Torvalds { 16331da177e4SLinus Torvalds struct vm_area_struct *vma = NULL, *prev = NULL; 16341da177e4SLinus Torvalds struct rb_node *rb_node; 16351da177e4SLinus Torvalds if (!mm) 16361da177e4SLinus Torvalds goto out; 16371da177e4SLinus Torvalds 16381da177e4SLinus Torvalds /* Guard against addr being lower than the first VMA */ 16391da177e4SLinus Torvalds vma = mm->mmap; 16401da177e4SLinus Torvalds 16411da177e4SLinus Torvalds /* Go through the RB tree quickly. */ 16421da177e4SLinus Torvalds rb_node = mm->mm_rb.rb_node; 16431da177e4SLinus Torvalds 16441da177e4SLinus Torvalds while (rb_node) { 16451da177e4SLinus Torvalds struct vm_area_struct *vma_tmp; 16461da177e4SLinus Torvalds vma_tmp = rb_entry(rb_node, struct vm_area_struct, vm_rb); 16471da177e4SLinus Torvalds 16481da177e4SLinus Torvalds if (addr < vma_tmp->vm_end) { 16491da177e4SLinus Torvalds rb_node = rb_node->rb_left; 16501da177e4SLinus Torvalds } else { 16511da177e4SLinus Torvalds prev = vma_tmp; 16521da177e4SLinus Torvalds if (!prev->vm_next || (addr < prev->vm_next->vm_end)) 16531da177e4SLinus Torvalds break; 16541da177e4SLinus Torvalds rb_node = rb_node->rb_right; 16551da177e4SLinus Torvalds } 16561da177e4SLinus Torvalds } 16571da177e4SLinus Torvalds 16581da177e4SLinus Torvalds out: 16591da177e4SLinus Torvalds *pprev = prev; 16601da177e4SLinus Torvalds return prev ? prev->vm_next : vma; 16611da177e4SLinus Torvalds } 16621da177e4SLinus Torvalds 16631da177e4SLinus Torvalds /* 16641da177e4SLinus Torvalds * Verify that the stack growth is acceptable and 16651da177e4SLinus Torvalds * update accounting. This is shared with both the 16661da177e4SLinus Torvalds * grow-up and grow-down cases. 16671da177e4SLinus Torvalds */ 16681da177e4SLinus Torvalds static int acct_stack_growth(struct vm_area_struct *vma, unsigned long size, unsigned long grow) 16691da177e4SLinus Torvalds { 16701da177e4SLinus Torvalds struct mm_struct *mm = vma->vm_mm; 16711da177e4SLinus Torvalds struct rlimit *rlim = current->signal->rlim; 16720d59a01bSAdam Litke unsigned long new_start; 16731da177e4SLinus Torvalds 16741da177e4SLinus Torvalds /* address space limit tests */ 1675119f657cSakpm@osdl.org if (!may_expand_vm(mm, grow)) 16761da177e4SLinus Torvalds return -ENOMEM; 16771da177e4SLinus Torvalds 16781da177e4SLinus Torvalds /* Stack limit test */ 167959e99e5bSJiri Slaby if (size > ACCESS_ONCE(rlim[RLIMIT_STACK].rlim_cur)) 16801da177e4SLinus Torvalds return -ENOMEM; 16811da177e4SLinus Torvalds 16821da177e4SLinus Torvalds /* mlock limit tests */ 16831da177e4SLinus Torvalds if (vma->vm_flags & VM_LOCKED) { 16841da177e4SLinus Torvalds unsigned long locked; 16851da177e4SLinus Torvalds unsigned long limit; 16861da177e4SLinus Torvalds locked = mm->locked_vm + grow; 168759e99e5bSJiri Slaby limit = ACCESS_ONCE(rlim[RLIMIT_MEMLOCK].rlim_cur); 168859e99e5bSJiri Slaby limit >>= PAGE_SHIFT; 16891da177e4SLinus Torvalds if (locked > limit && !capable(CAP_IPC_LOCK)) 16901da177e4SLinus Torvalds return -ENOMEM; 16911da177e4SLinus Torvalds } 16921da177e4SLinus Torvalds 16930d59a01bSAdam Litke /* Check to ensure the stack will not grow into a hugetlb-only region */ 16940d59a01bSAdam Litke new_start = (vma->vm_flags & VM_GROWSUP) ? vma->vm_start : 16950d59a01bSAdam Litke vma->vm_end - size; 16960d59a01bSAdam Litke if (is_hugepage_only_range(vma->vm_mm, new_start, size)) 16970d59a01bSAdam Litke return -EFAULT; 16980d59a01bSAdam Litke 16991da177e4SLinus Torvalds /* 17001da177e4SLinus Torvalds * Overcommit.. This must be the final test, as it will 17011da177e4SLinus Torvalds * update security statistics. 17021da177e4SLinus Torvalds */ 170305fa199dSHugh Dickins if (security_vm_enough_memory_mm(mm, grow)) 17041da177e4SLinus Torvalds return -ENOMEM; 17051da177e4SLinus Torvalds 17061da177e4SLinus Torvalds /* Ok, everything looks good - let it rip */ 17071da177e4SLinus Torvalds mm->total_vm += grow; 17081da177e4SLinus Torvalds if (vma->vm_flags & VM_LOCKED) 17091da177e4SLinus Torvalds mm->locked_vm += grow; 1710ab50b8edSHugh Dickins vm_stat_account(mm, vma->vm_flags, vma->vm_file, grow); 17111da177e4SLinus Torvalds return 0; 17121da177e4SLinus Torvalds } 17131da177e4SLinus Torvalds 171446dea3d0SHugh Dickins #if defined(CONFIG_STACK_GROWSUP) || defined(CONFIG_IA64) 17151da177e4SLinus Torvalds /* 171646dea3d0SHugh Dickins * PA-RISC uses this for its stack; IA64 for its Register Backing Store. 171746dea3d0SHugh Dickins * vma is the last one with address > vma->vm_end. Have to extend vma. 17181da177e4SLinus Torvalds */ 171946dea3d0SHugh Dickins int expand_upwards(struct vm_area_struct *vma, unsigned long address) 17201da177e4SLinus Torvalds { 17211da177e4SLinus Torvalds int error; 17221da177e4SLinus Torvalds 17231da177e4SLinus Torvalds if (!(vma->vm_flags & VM_GROWSUP)) 17241da177e4SLinus Torvalds return -EFAULT; 17251da177e4SLinus Torvalds 17261da177e4SLinus Torvalds /* 17271da177e4SLinus Torvalds * We must make sure the anon_vma is allocated 17281da177e4SLinus Torvalds * so that the anon_vma locking is not a noop. 17291da177e4SLinus Torvalds */ 17301da177e4SLinus Torvalds if (unlikely(anon_vma_prepare(vma))) 17311da177e4SLinus Torvalds return -ENOMEM; 1732bb4a340eSRik van Riel vma_lock_anon_vma(vma); 17331da177e4SLinus Torvalds 17341da177e4SLinus Torvalds /* 17351da177e4SLinus Torvalds * vma->vm_start/vm_end cannot change under us because the caller 17361da177e4SLinus Torvalds * is required to hold the mmap_sem in read mode. We need the 17371da177e4SLinus Torvalds * anon_vma lock to serialize against concurrent expand_stacks. 173806b32f3aSHelge Deller * Also guard against wrapping around to address 0. 17391da177e4SLinus Torvalds */ 174006b32f3aSHelge Deller if (address < PAGE_ALIGN(address+4)) 174106b32f3aSHelge Deller address = PAGE_ALIGN(address+4); 174206b32f3aSHelge Deller else { 1743bb4a340eSRik van Riel vma_unlock_anon_vma(vma); 174406b32f3aSHelge Deller return -ENOMEM; 174506b32f3aSHelge Deller } 17461da177e4SLinus Torvalds error = 0; 17471da177e4SLinus Torvalds 17481da177e4SLinus Torvalds /* Somebody else might have raced and expanded it already */ 17491da177e4SLinus Torvalds if (address > vma->vm_end) { 17501da177e4SLinus Torvalds unsigned long size, grow; 17511da177e4SLinus Torvalds 17521da177e4SLinus Torvalds size = address - vma->vm_start; 17531da177e4SLinus Torvalds grow = (address - vma->vm_end) >> PAGE_SHIFT; 17541da177e4SLinus Torvalds 17551da177e4SLinus Torvalds error = acct_stack_growth(vma, size, grow); 17563af9e859SEric B Munson if (!error) { 17571da177e4SLinus Torvalds vma->vm_end = address; 17583af9e859SEric B Munson perf_event_mmap(vma); 17593af9e859SEric B Munson } 17601da177e4SLinus Torvalds } 1761bb4a340eSRik van Riel vma_unlock_anon_vma(vma); 17621da177e4SLinus Torvalds return error; 17631da177e4SLinus Torvalds } 176446dea3d0SHugh Dickins #endif /* CONFIG_STACK_GROWSUP || CONFIG_IA64 */ 176546dea3d0SHugh Dickins 17661da177e4SLinus Torvalds /* 17671da177e4SLinus Torvalds * vma is the first one with address < vma->vm_start. Have to extend vma. 17681da177e4SLinus Torvalds */ 1769cb8f488cSDenys Vlasenko static int expand_downwards(struct vm_area_struct *vma, 1770b6a2fea3SOllie Wild unsigned long address) 17711da177e4SLinus Torvalds { 17721da177e4SLinus Torvalds int error; 17731da177e4SLinus Torvalds 17741da177e4SLinus Torvalds /* 17751da177e4SLinus Torvalds * We must make sure the anon_vma is allocated 17761da177e4SLinus Torvalds * so that the anon_vma locking is not a noop. 17771da177e4SLinus Torvalds */ 17781da177e4SLinus Torvalds if (unlikely(anon_vma_prepare(vma))) 17791da177e4SLinus Torvalds return -ENOMEM; 17808869477aSEric Paris 17818869477aSEric Paris address &= PAGE_MASK; 178288c3f7a8SRichard Knutsson error = security_file_mmap(NULL, 0, 0, 0, address, 1); 17838869477aSEric Paris if (error) 17848869477aSEric Paris return error; 17858869477aSEric Paris 1786bb4a340eSRik van Riel vma_lock_anon_vma(vma); 17871da177e4SLinus Torvalds 17881da177e4SLinus Torvalds /* 17891da177e4SLinus Torvalds * vma->vm_start/vm_end cannot change under us because the caller 17901da177e4SLinus Torvalds * is required to hold the mmap_sem in read mode. We need the 17911da177e4SLinus Torvalds * anon_vma lock to serialize against concurrent expand_stacks. 17921da177e4SLinus Torvalds */ 17931da177e4SLinus Torvalds 17941da177e4SLinus Torvalds /* Somebody else might have raced and expanded it already */ 17951da177e4SLinus Torvalds if (address < vma->vm_start) { 17961da177e4SLinus Torvalds unsigned long size, grow; 17971da177e4SLinus Torvalds 17981da177e4SLinus Torvalds size = vma->vm_end - address; 17991da177e4SLinus Torvalds grow = (vma->vm_start - address) >> PAGE_SHIFT; 18001da177e4SLinus Torvalds 18011da177e4SLinus Torvalds error = acct_stack_growth(vma, size, grow); 18021da177e4SLinus Torvalds if (!error) { 18031da177e4SLinus Torvalds vma->vm_start = address; 18041da177e4SLinus Torvalds vma->vm_pgoff -= grow; 18053af9e859SEric B Munson perf_event_mmap(vma); 18061da177e4SLinus Torvalds } 18071da177e4SLinus Torvalds } 1808bb4a340eSRik van Riel vma_unlock_anon_vma(vma); 18091da177e4SLinus Torvalds return error; 18101da177e4SLinus Torvalds } 18111da177e4SLinus Torvalds 1812b6a2fea3SOllie Wild int expand_stack_downwards(struct vm_area_struct *vma, unsigned long address) 1813b6a2fea3SOllie Wild { 1814b6a2fea3SOllie Wild return expand_downwards(vma, address); 1815b6a2fea3SOllie Wild } 1816b6a2fea3SOllie Wild 1817b6a2fea3SOllie Wild #ifdef CONFIG_STACK_GROWSUP 1818b6a2fea3SOllie Wild int expand_stack(struct vm_area_struct *vma, unsigned long address) 1819b6a2fea3SOllie Wild { 1820b6a2fea3SOllie Wild return expand_upwards(vma, address); 1821b6a2fea3SOllie Wild } 1822b6a2fea3SOllie Wild 1823b6a2fea3SOllie Wild struct vm_area_struct * 1824b6a2fea3SOllie Wild find_extend_vma(struct mm_struct *mm, unsigned long addr) 1825b6a2fea3SOllie Wild { 1826b6a2fea3SOllie Wild struct vm_area_struct *vma, *prev; 1827b6a2fea3SOllie Wild 1828b6a2fea3SOllie Wild addr &= PAGE_MASK; 1829b6a2fea3SOllie Wild vma = find_vma_prev(mm, addr, &prev); 1830b6a2fea3SOllie Wild if (vma && (vma->vm_start <= addr)) 1831b6a2fea3SOllie Wild return vma; 18321c127185SDenys Vlasenko if (!prev || expand_stack(prev, addr)) 1833b6a2fea3SOllie Wild return NULL; 1834ba470de4SRik van Riel if (prev->vm_flags & VM_LOCKED) { 1835c58267c3SKOSAKI Motohiro mlock_vma_pages_range(prev, addr, prev->vm_end); 1836ba470de4SRik van Riel } 1837b6a2fea3SOllie Wild return prev; 1838b6a2fea3SOllie Wild } 1839b6a2fea3SOllie Wild #else 1840b6a2fea3SOllie Wild int expand_stack(struct vm_area_struct *vma, unsigned long address) 1841b6a2fea3SOllie Wild { 1842b6a2fea3SOllie Wild return expand_downwards(vma, address); 1843b6a2fea3SOllie Wild } 1844b6a2fea3SOllie Wild 18451da177e4SLinus Torvalds struct vm_area_struct * 18461da177e4SLinus Torvalds find_extend_vma(struct mm_struct * mm, unsigned long addr) 18471da177e4SLinus Torvalds { 18481da177e4SLinus Torvalds struct vm_area_struct * vma; 18491da177e4SLinus Torvalds unsigned long start; 18501da177e4SLinus Torvalds 18511da177e4SLinus Torvalds addr &= PAGE_MASK; 18521da177e4SLinus Torvalds vma = find_vma(mm,addr); 18531da177e4SLinus Torvalds if (!vma) 18541da177e4SLinus Torvalds return NULL; 18551da177e4SLinus Torvalds if (vma->vm_start <= addr) 18561da177e4SLinus Torvalds return vma; 18571da177e4SLinus Torvalds if (!(vma->vm_flags & VM_GROWSDOWN)) 18581da177e4SLinus Torvalds return NULL; 18591da177e4SLinus Torvalds start = vma->vm_start; 18601da177e4SLinus Torvalds if (expand_stack(vma, addr)) 18611da177e4SLinus Torvalds return NULL; 1862ba470de4SRik van Riel if (vma->vm_flags & VM_LOCKED) { 1863c58267c3SKOSAKI Motohiro mlock_vma_pages_range(vma, addr, start); 1864ba470de4SRik van Riel } 18651da177e4SLinus Torvalds return vma; 18661da177e4SLinus Torvalds } 18671da177e4SLinus Torvalds #endif 18681da177e4SLinus Torvalds 18692c0b3814SHugh Dickins /* 18702c0b3814SHugh Dickins * Ok - we have the memory areas we should free on the vma list, 18712c0b3814SHugh Dickins * so release them, and do the vma updates. 18721da177e4SLinus Torvalds * 18732c0b3814SHugh Dickins * Called with the mm semaphore held. 18741da177e4SLinus Torvalds */ 18752c0b3814SHugh Dickins static void remove_vma_list(struct mm_struct *mm, struct vm_area_struct *vma) 18761da177e4SLinus Torvalds { 1877365e9c87SHugh Dickins /* Update high watermark before we lower total_vm */ 1878365e9c87SHugh Dickins update_hiwater_vm(mm); 18792c0b3814SHugh Dickins do { 1880ab50b8edSHugh Dickins long nrpages = vma_pages(vma); 18811da177e4SLinus Torvalds 1882ab50b8edSHugh Dickins mm->total_vm -= nrpages; 1883ab50b8edSHugh Dickins vm_stat_account(mm, vma->vm_flags, vma->vm_file, -nrpages); 1884a8fb5618SHugh Dickins vma = remove_vma(vma); 1885146425a3SHugh Dickins } while (vma); 18861da177e4SLinus Torvalds validate_mm(mm); 18871da177e4SLinus Torvalds } 18881da177e4SLinus Torvalds 18891da177e4SLinus Torvalds /* 18901da177e4SLinus Torvalds * Get rid of page table information in the indicated region. 18911da177e4SLinus Torvalds * 1892f10df686SPaolo 'Blaisorblade' Giarrusso * Called with the mm semaphore held. 18931da177e4SLinus Torvalds */ 18941da177e4SLinus Torvalds static void unmap_region(struct mm_struct *mm, 1895e0da382cSHugh Dickins struct vm_area_struct *vma, struct vm_area_struct *prev, 1896e0da382cSHugh Dickins unsigned long start, unsigned long end) 18971da177e4SLinus Torvalds { 1898e0da382cSHugh Dickins struct vm_area_struct *next = prev? prev->vm_next: mm->mmap; 18991da177e4SLinus Torvalds struct mmu_gather *tlb; 19001da177e4SLinus Torvalds unsigned long nr_accounted = 0; 19011da177e4SLinus Torvalds 19021da177e4SLinus Torvalds lru_add_drain(); 19031da177e4SLinus Torvalds tlb = tlb_gather_mmu(mm, 0); 1904365e9c87SHugh Dickins update_hiwater_rss(mm); 1905508034a3SHugh Dickins unmap_vmas(&tlb, vma, start, end, &nr_accounted, NULL); 19061da177e4SLinus Torvalds vm_unacct_memory(nr_accounted); 190742b77728SJan Beulich free_pgtables(tlb, vma, prev? prev->vm_end: FIRST_USER_ADDRESS, 1908e0da382cSHugh Dickins next? next->vm_start: 0); 19091da177e4SLinus Torvalds tlb_finish_mmu(tlb, start, end); 19101da177e4SLinus Torvalds } 19111da177e4SLinus Torvalds 19121da177e4SLinus Torvalds /* 19131da177e4SLinus Torvalds * Create a list of vma's touched by the unmap, removing them from the mm's 19141da177e4SLinus Torvalds * vma list as we go.. 19151da177e4SLinus Torvalds */ 19161da177e4SLinus Torvalds static void 19171da177e4SLinus Torvalds detach_vmas_to_be_unmapped(struct mm_struct *mm, struct vm_area_struct *vma, 19181da177e4SLinus Torvalds struct vm_area_struct *prev, unsigned long end) 19191da177e4SLinus Torvalds { 19201da177e4SLinus Torvalds struct vm_area_struct **insertion_point; 19211da177e4SLinus Torvalds struct vm_area_struct *tail_vma = NULL; 19221363c3cdSWolfgang Wander unsigned long addr; 19231da177e4SLinus Torvalds 19241da177e4SLinus Torvalds insertion_point = (prev ? &prev->vm_next : &mm->mmap); 1925297c5eeeSLinus Torvalds vma->vm_prev = NULL; 19261da177e4SLinus Torvalds do { 19271da177e4SLinus Torvalds rb_erase(&vma->vm_rb, &mm->mm_rb); 19281da177e4SLinus Torvalds mm->map_count--; 19291da177e4SLinus Torvalds tail_vma = vma; 19301da177e4SLinus Torvalds vma = vma->vm_next; 19311da177e4SLinus Torvalds } while (vma && vma->vm_start < end); 19321da177e4SLinus Torvalds *insertion_point = vma; 1933297c5eeeSLinus Torvalds if (vma) 1934297c5eeeSLinus Torvalds vma->vm_prev = prev; 19351da177e4SLinus Torvalds tail_vma->vm_next = NULL; 19361363c3cdSWolfgang Wander if (mm->unmap_area == arch_unmap_area) 19371363c3cdSWolfgang Wander addr = prev ? prev->vm_end : mm->mmap_base; 19381363c3cdSWolfgang Wander else 19391363c3cdSWolfgang Wander addr = vma ? vma->vm_start : mm->mmap_base; 19401363c3cdSWolfgang Wander mm->unmap_area(mm, addr); 19411da177e4SLinus Torvalds mm->mmap_cache = NULL; /* Kill the cache. */ 19421da177e4SLinus Torvalds } 19431da177e4SLinus Torvalds 19441da177e4SLinus Torvalds /* 1945659ace58SKOSAKI Motohiro * __split_vma() bypasses sysctl_max_map_count checking. We use this on the 1946659ace58SKOSAKI Motohiro * munmap path where it doesn't make sense to fail. 19471da177e4SLinus Torvalds */ 1948659ace58SKOSAKI Motohiro static int __split_vma(struct mm_struct * mm, struct vm_area_struct * vma, 19491da177e4SLinus Torvalds unsigned long addr, int new_below) 19501da177e4SLinus Torvalds { 19511da177e4SLinus Torvalds struct mempolicy *pol; 19521da177e4SLinus Torvalds struct vm_area_struct *new; 19535beb4930SRik van Riel int err = -ENOMEM; 19541da177e4SLinus Torvalds 1955a5516438SAndi Kleen if (is_vm_hugetlb_page(vma) && (addr & 1956a5516438SAndi Kleen ~(huge_page_mask(hstate_vma(vma))))) 19571da177e4SLinus Torvalds return -EINVAL; 19581da177e4SLinus Torvalds 1959e94b1766SChristoph Lameter new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL); 19601da177e4SLinus Torvalds if (!new) 19615beb4930SRik van Riel goto out_err; 19621da177e4SLinus Torvalds 19631da177e4SLinus Torvalds /* most fields are the same, copy all, and then fixup */ 19641da177e4SLinus Torvalds *new = *vma; 19651da177e4SLinus Torvalds 19665beb4930SRik van Riel INIT_LIST_HEAD(&new->anon_vma_chain); 19675beb4930SRik van Riel 19681da177e4SLinus Torvalds if (new_below) 19691da177e4SLinus Torvalds new->vm_end = addr; 19701da177e4SLinus Torvalds else { 19711da177e4SLinus Torvalds new->vm_start = addr; 19721da177e4SLinus Torvalds new->vm_pgoff += ((addr - vma->vm_start) >> PAGE_SHIFT); 19731da177e4SLinus Torvalds } 19741da177e4SLinus Torvalds 1975846a16bfSLee Schermerhorn pol = mpol_dup(vma_policy(vma)); 19761da177e4SLinus Torvalds if (IS_ERR(pol)) { 19775beb4930SRik van Riel err = PTR_ERR(pol); 19785beb4930SRik van Riel goto out_free_vma; 19791da177e4SLinus Torvalds } 19801da177e4SLinus Torvalds vma_set_policy(new, pol); 19811da177e4SLinus Torvalds 19825beb4930SRik van Riel if (anon_vma_clone(new, vma)) 19835beb4930SRik van Riel goto out_free_mpol; 19845beb4930SRik van Riel 1985925d1c40SMatt Helsley if (new->vm_file) { 19861da177e4SLinus Torvalds get_file(new->vm_file); 1987925d1c40SMatt Helsley if (vma->vm_flags & VM_EXECUTABLE) 1988925d1c40SMatt Helsley added_exe_file_vma(mm); 1989925d1c40SMatt Helsley } 19901da177e4SLinus Torvalds 19911da177e4SLinus Torvalds if (new->vm_ops && new->vm_ops->open) 19921da177e4SLinus Torvalds new->vm_ops->open(new); 19931da177e4SLinus Torvalds 19941da177e4SLinus Torvalds if (new_below) 19955beb4930SRik van Riel err = vma_adjust(vma, addr, vma->vm_end, vma->vm_pgoff + 19961da177e4SLinus Torvalds ((addr - new->vm_start) >> PAGE_SHIFT), new); 19971da177e4SLinus Torvalds else 19985beb4930SRik van Riel err = vma_adjust(vma, vma->vm_start, addr, vma->vm_pgoff, new); 19991da177e4SLinus Torvalds 20005beb4930SRik van Riel /* Success. */ 20015beb4930SRik van Riel if (!err) 20021da177e4SLinus Torvalds return 0; 20035beb4930SRik van Riel 20045beb4930SRik van Riel /* Clean everything up if vma_adjust failed. */ 200558927533SRik van Riel if (new->vm_ops && new->vm_ops->close) 20065beb4930SRik van Riel new->vm_ops->close(new); 20075beb4930SRik van Riel if (new->vm_file) { 20085beb4930SRik van Riel if (vma->vm_flags & VM_EXECUTABLE) 20095beb4930SRik van Riel removed_exe_file_vma(mm); 20105beb4930SRik van Riel fput(new->vm_file); 20115beb4930SRik van Riel } 2012*2aeadc30SAndrea Arcangeli unlink_anon_vmas(new); 20135beb4930SRik van Riel out_free_mpol: 20145beb4930SRik van Riel mpol_put(pol); 20155beb4930SRik van Riel out_free_vma: 20165beb4930SRik van Riel kmem_cache_free(vm_area_cachep, new); 20175beb4930SRik van Riel out_err: 20185beb4930SRik van Riel return err; 20191da177e4SLinus Torvalds } 20201da177e4SLinus Torvalds 2021659ace58SKOSAKI Motohiro /* 2022659ace58SKOSAKI Motohiro * Split a vma into two pieces at address 'addr', a new vma is allocated 2023659ace58SKOSAKI Motohiro * either for the first part or the tail. 2024659ace58SKOSAKI Motohiro */ 2025659ace58SKOSAKI Motohiro int split_vma(struct mm_struct *mm, struct vm_area_struct *vma, 2026659ace58SKOSAKI Motohiro unsigned long addr, int new_below) 2027659ace58SKOSAKI Motohiro { 2028659ace58SKOSAKI Motohiro if (mm->map_count >= sysctl_max_map_count) 2029659ace58SKOSAKI Motohiro return -ENOMEM; 2030659ace58SKOSAKI Motohiro 2031659ace58SKOSAKI Motohiro return __split_vma(mm, vma, addr, new_below); 2032659ace58SKOSAKI Motohiro } 2033659ace58SKOSAKI Motohiro 20341da177e4SLinus Torvalds /* Munmap is split into 2 main parts -- this part which finds 20351da177e4SLinus Torvalds * what needs doing, and the areas themselves, which do the 20361da177e4SLinus Torvalds * work. This now handles partial unmappings. 20371da177e4SLinus Torvalds * Jeremy Fitzhardinge <jeremy@goop.org> 20381da177e4SLinus Torvalds */ 20391da177e4SLinus Torvalds int do_munmap(struct mm_struct *mm, unsigned long start, size_t len) 20401da177e4SLinus Torvalds { 20411da177e4SLinus Torvalds unsigned long end; 2042146425a3SHugh Dickins struct vm_area_struct *vma, *prev, *last; 20431da177e4SLinus Torvalds 20441da177e4SLinus Torvalds if ((start & ~PAGE_MASK) || start > TASK_SIZE || len > TASK_SIZE-start) 20451da177e4SLinus Torvalds return -EINVAL; 20461da177e4SLinus Torvalds 20471da177e4SLinus Torvalds if ((len = PAGE_ALIGN(len)) == 0) 20481da177e4SLinus Torvalds return -EINVAL; 20491da177e4SLinus Torvalds 20501da177e4SLinus Torvalds /* Find the first overlapping VMA */ 2051146425a3SHugh Dickins vma = find_vma_prev(mm, start, &prev); 2052146425a3SHugh Dickins if (!vma) 20531da177e4SLinus Torvalds return 0; 2054146425a3SHugh Dickins /* we have start < vma->vm_end */ 20551da177e4SLinus Torvalds 20561da177e4SLinus Torvalds /* if it doesn't overlap, we have nothing.. */ 20571da177e4SLinus Torvalds end = start + len; 2058146425a3SHugh Dickins if (vma->vm_start >= end) 20591da177e4SLinus Torvalds return 0; 20601da177e4SLinus Torvalds 20611da177e4SLinus Torvalds /* 20621da177e4SLinus Torvalds * If we need to split any vma, do it now to save pain later. 20631da177e4SLinus Torvalds * 20641da177e4SLinus Torvalds * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially 20651da177e4SLinus Torvalds * unmapped vm_area_struct will remain in use: so lower split_vma 20661da177e4SLinus Torvalds * places tmp vma above, and higher split_vma places tmp vma below. 20671da177e4SLinus Torvalds */ 2068146425a3SHugh Dickins if (start > vma->vm_start) { 2069659ace58SKOSAKI Motohiro int error; 2070659ace58SKOSAKI Motohiro 2071659ace58SKOSAKI Motohiro /* 2072659ace58SKOSAKI Motohiro * Make sure that map_count on return from munmap() will 2073659ace58SKOSAKI Motohiro * not exceed its limit; but let map_count go just above 2074659ace58SKOSAKI Motohiro * its limit temporarily, to help free resources as expected. 2075659ace58SKOSAKI Motohiro */ 2076659ace58SKOSAKI Motohiro if (end < vma->vm_end && mm->map_count >= sysctl_max_map_count) 2077659ace58SKOSAKI Motohiro return -ENOMEM; 2078659ace58SKOSAKI Motohiro 2079659ace58SKOSAKI Motohiro error = __split_vma(mm, vma, start, 0); 20801da177e4SLinus Torvalds if (error) 20811da177e4SLinus Torvalds return error; 2082146425a3SHugh Dickins prev = vma; 20831da177e4SLinus Torvalds } 20841da177e4SLinus Torvalds 20851da177e4SLinus Torvalds /* Does it split the last one? */ 20861da177e4SLinus Torvalds last = find_vma(mm, end); 20871da177e4SLinus Torvalds if (last && end > last->vm_start) { 2088659ace58SKOSAKI Motohiro int error = __split_vma(mm, last, end, 1); 20891da177e4SLinus Torvalds if (error) 20901da177e4SLinus Torvalds return error; 20911da177e4SLinus Torvalds } 2092146425a3SHugh Dickins vma = prev? prev->vm_next: mm->mmap; 20931da177e4SLinus Torvalds 20941da177e4SLinus Torvalds /* 2095ba470de4SRik van Riel * unlock any mlock()ed ranges before detaching vmas 2096ba470de4SRik van Riel */ 2097ba470de4SRik van Riel if (mm->locked_vm) { 2098ba470de4SRik van Riel struct vm_area_struct *tmp = vma; 2099ba470de4SRik van Riel while (tmp && tmp->vm_start < end) { 2100ba470de4SRik van Riel if (tmp->vm_flags & VM_LOCKED) { 2101ba470de4SRik van Riel mm->locked_vm -= vma_pages(tmp); 2102ba470de4SRik van Riel munlock_vma_pages_all(tmp); 2103ba470de4SRik van Riel } 2104ba470de4SRik van Riel tmp = tmp->vm_next; 2105ba470de4SRik van Riel } 2106ba470de4SRik van Riel } 2107ba470de4SRik van Riel 2108ba470de4SRik van Riel /* 21091da177e4SLinus Torvalds * Remove the vma's, and unmap the actual pages 21101da177e4SLinus Torvalds */ 2111146425a3SHugh Dickins detach_vmas_to_be_unmapped(mm, vma, prev, end); 2112146425a3SHugh Dickins unmap_region(mm, vma, prev, start, end); 21131da177e4SLinus Torvalds 21141da177e4SLinus Torvalds /* Fix up all other VM information */ 21152c0b3814SHugh Dickins remove_vma_list(mm, vma); 21161da177e4SLinus Torvalds 21171da177e4SLinus Torvalds return 0; 21181da177e4SLinus Torvalds } 21191da177e4SLinus Torvalds 21201da177e4SLinus Torvalds EXPORT_SYMBOL(do_munmap); 21211da177e4SLinus Torvalds 21226a6160a7SHeiko Carstens SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len) 21231da177e4SLinus Torvalds { 21241da177e4SLinus Torvalds int ret; 21251da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 21261da177e4SLinus Torvalds 21271da177e4SLinus Torvalds profile_munmap(addr); 21281da177e4SLinus Torvalds 21291da177e4SLinus Torvalds down_write(&mm->mmap_sem); 21301da177e4SLinus Torvalds ret = do_munmap(mm, addr, len); 21311da177e4SLinus Torvalds up_write(&mm->mmap_sem); 21321da177e4SLinus Torvalds return ret; 21331da177e4SLinus Torvalds } 21341da177e4SLinus Torvalds 21351da177e4SLinus Torvalds static inline void verify_mm_writelocked(struct mm_struct *mm) 21361da177e4SLinus Torvalds { 2137a241ec65SPaul E. McKenney #ifdef CONFIG_DEBUG_VM 21381da177e4SLinus Torvalds if (unlikely(down_read_trylock(&mm->mmap_sem))) { 21391da177e4SLinus Torvalds WARN_ON(1); 21401da177e4SLinus Torvalds up_read(&mm->mmap_sem); 21411da177e4SLinus Torvalds } 21421da177e4SLinus Torvalds #endif 21431da177e4SLinus Torvalds } 21441da177e4SLinus Torvalds 21451da177e4SLinus Torvalds /* 21461da177e4SLinus Torvalds * this is really a simplified "do_mmap". it only handles 21471da177e4SLinus Torvalds * anonymous maps. eventually we may be able to do some 21481da177e4SLinus Torvalds * brk-specific accounting here. 21491da177e4SLinus Torvalds */ 21501da177e4SLinus Torvalds unsigned long do_brk(unsigned long addr, unsigned long len) 21511da177e4SLinus Torvalds { 21521da177e4SLinus Torvalds struct mm_struct * mm = current->mm; 21531da177e4SLinus Torvalds struct vm_area_struct * vma, * prev; 21541da177e4SLinus Torvalds unsigned long flags; 21551da177e4SLinus Torvalds struct rb_node ** rb_link, * rb_parent; 21561da177e4SLinus Torvalds pgoff_t pgoff = addr >> PAGE_SHIFT; 21573a459756SKirill Korotaev int error; 21581da177e4SLinus Torvalds 21591da177e4SLinus Torvalds len = PAGE_ALIGN(len); 21601da177e4SLinus Torvalds if (!len) 21611da177e4SLinus Torvalds return addr; 21621da177e4SLinus Torvalds 216388c3f7a8SRichard Knutsson error = security_file_mmap(NULL, 0, 0, 0, addr, 1); 21645a211a5dSEric Paris if (error) 21655a211a5dSEric Paris return error; 21665a211a5dSEric Paris 21673a459756SKirill Korotaev flags = VM_DATA_DEFAULT_FLAGS | VM_ACCOUNT | mm->def_flags; 21683a459756SKirill Korotaev 21692c6a1016SAl Viro error = get_unmapped_area(NULL, addr, len, 0, MAP_FIXED); 21702c6a1016SAl Viro if (error & ~PAGE_MASK) 21713a459756SKirill Korotaev return error; 21723a459756SKirill Korotaev 21731da177e4SLinus Torvalds /* 21741da177e4SLinus Torvalds * mlock MCL_FUTURE? 21751da177e4SLinus Torvalds */ 21761da177e4SLinus Torvalds if (mm->def_flags & VM_LOCKED) { 21771da177e4SLinus Torvalds unsigned long locked, lock_limit; 217893ea1d0aSChris Wright locked = len >> PAGE_SHIFT; 217993ea1d0aSChris Wright locked += mm->locked_vm; 218059e99e5bSJiri Slaby lock_limit = rlimit(RLIMIT_MEMLOCK); 218193ea1d0aSChris Wright lock_limit >>= PAGE_SHIFT; 21821da177e4SLinus Torvalds if (locked > lock_limit && !capable(CAP_IPC_LOCK)) 21831da177e4SLinus Torvalds return -EAGAIN; 21841da177e4SLinus Torvalds } 21851da177e4SLinus Torvalds 21861da177e4SLinus Torvalds /* 21871da177e4SLinus Torvalds * mm->mmap_sem is required to protect against another thread 21881da177e4SLinus Torvalds * changing the mappings in case we sleep. 21891da177e4SLinus Torvalds */ 21901da177e4SLinus Torvalds verify_mm_writelocked(mm); 21911da177e4SLinus Torvalds 21921da177e4SLinus Torvalds /* 21931da177e4SLinus Torvalds * Clear old maps. this also does some error checking for us 21941da177e4SLinus Torvalds */ 21951da177e4SLinus Torvalds munmap_back: 21961da177e4SLinus Torvalds vma = find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent); 21971da177e4SLinus Torvalds if (vma && vma->vm_start < addr + len) { 21981da177e4SLinus Torvalds if (do_munmap(mm, addr, len)) 21991da177e4SLinus Torvalds return -ENOMEM; 22001da177e4SLinus Torvalds goto munmap_back; 22011da177e4SLinus Torvalds } 22021da177e4SLinus Torvalds 22031da177e4SLinus Torvalds /* Check against address space limits *after* clearing old maps... */ 2204119f657cSakpm@osdl.org if (!may_expand_vm(mm, len >> PAGE_SHIFT)) 22051da177e4SLinus Torvalds return -ENOMEM; 22061da177e4SLinus Torvalds 22071da177e4SLinus Torvalds if (mm->map_count > sysctl_max_map_count) 22081da177e4SLinus Torvalds return -ENOMEM; 22091da177e4SLinus Torvalds 22101da177e4SLinus Torvalds if (security_vm_enough_memory(len >> PAGE_SHIFT)) 22111da177e4SLinus Torvalds return -ENOMEM; 22121da177e4SLinus Torvalds 22131da177e4SLinus Torvalds /* Can we just expand an old private anonymous mapping? */ 2214ba470de4SRik van Riel vma = vma_merge(mm, prev, addr, addr + len, flags, 2215ba470de4SRik van Riel NULL, NULL, pgoff, NULL); 2216ba470de4SRik van Riel if (vma) 22171da177e4SLinus Torvalds goto out; 22181da177e4SLinus Torvalds 22191da177e4SLinus Torvalds /* 22201da177e4SLinus Torvalds * create a vma struct for an anonymous mapping 22211da177e4SLinus Torvalds */ 2222c5e3b83eSPekka Enberg vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL); 22231da177e4SLinus Torvalds if (!vma) { 22241da177e4SLinus Torvalds vm_unacct_memory(len >> PAGE_SHIFT); 22251da177e4SLinus Torvalds return -ENOMEM; 22261da177e4SLinus Torvalds } 22271da177e4SLinus Torvalds 22285beb4930SRik van Riel INIT_LIST_HEAD(&vma->anon_vma_chain); 22291da177e4SLinus Torvalds vma->vm_mm = mm; 22301da177e4SLinus Torvalds vma->vm_start = addr; 22311da177e4SLinus Torvalds vma->vm_end = addr + len; 22321da177e4SLinus Torvalds vma->vm_pgoff = pgoff; 22331da177e4SLinus Torvalds vma->vm_flags = flags; 22343ed75eb8SColy Li vma->vm_page_prot = vm_get_page_prot(flags); 22351da177e4SLinus Torvalds vma_link(mm, vma, prev, rb_link, rb_parent); 22361da177e4SLinus Torvalds out: 22373af9e859SEric B Munson perf_event_mmap(vma); 22381da177e4SLinus Torvalds mm->total_vm += len >> PAGE_SHIFT; 22391da177e4SLinus Torvalds if (flags & VM_LOCKED) { 2240ba470de4SRik van Riel if (!mlock_vma_pages_range(vma, addr, addr + len)) 2241ba470de4SRik van Riel mm->locked_vm += (len >> PAGE_SHIFT); 22421da177e4SLinus Torvalds } 22431da177e4SLinus Torvalds return addr; 22441da177e4SLinus Torvalds } 22451da177e4SLinus Torvalds 22461da177e4SLinus Torvalds EXPORT_SYMBOL(do_brk); 22471da177e4SLinus Torvalds 22481da177e4SLinus Torvalds /* Release all mmaps. */ 22491da177e4SLinus Torvalds void exit_mmap(struct mm_struct *mm) 22501da177e4SLinus Torvalds { 22511da177e4SLinus Torvalds struct mmu_gather *tlb; 2252ba470de4SRik van Riel struct vm_area_struct *vma; 22531da177e4SLinus Torvalds unsigned long nr_accounted = 0; 2254ee39b37bSHugh Dickins unsigned long end; 22551da177e4SLinus Torvalds 2256d6dd61c8SJeremy Fitzhardinge /* mm's last user has gone, and its about to be pulled down */ 2257cddb8a5cSAndrea Arcangeli mmu_notifier_release(mm); 2258d6dd61c8SJeremy Fitzhardinge 2259ba470de4SRik van Riel if (mm->locked_vm) { 2260ba470de4SRik van Riel vma = mm->mmap; 2261ba470de4SRik van Riel while (vma) { 2262ba470de4SRik van Riel if (vma->vm_flags & VM_LOCKED) 2263ba470de4SRik van Riel munlock_vma_pages_all(vma); 2264ba470de4SRik van Riel vma = vma->vm_next; 2265ba470de4SRik van Riel } 2266ba470de4SRik van Riel } 22679480c53eSJeremy Fitzhardinge 22689480c53eSJeremy Fitzhardinge arch_exit_mmap(mm); 22699480c53eSJeremy Fitzhardinge 2270ba470de4SRik van Riel vma = mm->mmap; 22719480c53eSJeremy Fitzhardinge if (!vma) /* Can happen if dup_mmap() received an OOM */ 22729480c53eSJeremy Fitzhardinge return; 22739480c53eSJeremy Fitzhardinge 22741da177e4SLinus Torvalds lru_add_drain(); 22751da177e4SLinus Torvalds flush_cache_mm(mm); 2276e0da382cSHugh Dickins tlb = tlb_gather_mmu(mm, 1); 2277901608d9SOleg Nesterov /* update_hiwater_rss(mm) here? but nobody should be looking */ 2278e0da382cSHugh Dickins /* Use -1 here to ensure all VMAs in the mm are unmapped */ 2279508034a3SHugh Dickins end = unmap_vmas(&tlb, vma, 0, -1, &nr_accounted, NULL); 22801da177e4SLinus Torvalds vm_unacct_memory(nr_accounted); 22819ba69294SHugh Dickins 228242b77728SJan Beulich free_pgtables(tlb, vma, FIRST_USER_ADDRESS, 0); 2283ee39b37bSHugh Dickins tlb_finish_mmu(tlb, 0, end); 22841da177e4SLinus Torvalds 22851da177e4SLinus Torvalds /* 22868f4f8c16SHugh Dickins * Walk the list again, actually closing and freeing it, 22878f4f8c16SHugh Dickins * with preemption enabled, without holding any MM locks. 22881da177e4SLinus Torvalds */ 2289a8fb5618SHugh Dickins while (vma) 2290a8fb5618SHugh Dickins vma = remove_vma(vma); 2291e0da382cSHugh Dickins 2292e2cdef8cSHugh Dickins BUG_ON(mm->nr_ptes > (FIRST_USER_ADDRESS+PMD_SIZE-1)>>PMD_SHIFT); 22931da177e4SLinus Torvalds } 22941da177e4SLinus Torvalds 22951da177e4SLinus Torvalds /* Insert vm structure into process list sorted by address 22961da177e4SLinus Torvalds * and into the inode's i_mmap tree. If vm_file is non-NULL 22971da177e4SLinus Torvalds * then i_mmap_lock is taken here. 22981da177e4SLinus Torvalds */ 22991da177e4SLinus Torvalds int insert_vm_struct(struct mm_struct * mm, struct vm_area_struct * vma) 23001da177e4SLinus Torvalds { 23011da177e4SLinus Torvalds struct vm_area_struct * __vma, * prev; 23021da177e4SLinus Torvalds struct rb_node ** rb_link, * rb_parent; 23031da177e4SLinus Torvalds 23041da177e4SLinus Torvalds /* 23051da177e4SLinus Torvalds * The vm_pgoff of a purely anonymous vma should be irrelevant 23061da177e4SLinus Torvalds * until its first write fault, when page's anon_vma and index 23071da177e4SLinus Torvalds * are set. But now set the vm_pgoff it will almost certainly 23081da177e4SLinus Torvalds * end up with (unless mremap moves it elsewhere before that 23091da177e4SLinus Torvalds * first wfault), so /proc/pid/maps tells a consistent story. 23101da177e4SLinus Torvalds * 23111da177e4SLinus Torvalds * By setting it to reflect the virtual start address of the 23121da177e4SLinus Torvalds * vma, merges and splits can happen in a seamless way, just 23131da177e4SLinus Torvalds * using the existing file pgoff checks and manipulations. 23141da177e4SLinus Torvalds * Similarly in do_mmap_pgoff and in do_brk. 23151da177e4SLinus Torvalds */ 23161da177e4SLinus Torvalds if (!vma->vm_file) { 23171da177e4SLinus Torvalds BUG_ON(vma->anon_vma); 23181da177e4SLinus Torvalds vma->vm_pgoff = vma->vm_start >> PAGE_SHIFT; 23191da177e4SLinus Torvalds } 23201da177e4SLinus Torvalds __vma = find_vma_prepare(mm,vma->vm_start,&prev,&rb_link,&rb_parent); 23211da177e4SLinus Torvalds if (__vma && __vma->vm_start < vma->vm_end) 23221da177e4SLinus Torvalds return -ENOMEM; 23232fd4ef85SHugh Dickins if ((vma->vm_flags & VM_ACCOUNT) && 232434b4e4aaSAlan Cox security_vm_enough_memory_mm(mm, vma_pages(vma))) 23252fd4ef85SHugh Dickins return -ENOMEM; 23261da177e4SLinus Torvalds vma_link(mm, vma, prev, rb_link, rb_parent); 23271da177e4SLinus Torvalds return 0; 23281da177e4SLinus Torvalds } 23291da177e4SLinus Torvalds 23301da177e4SLinus Torvalds /* 23311da177e4SLinus Torvalds * Copy the vma structure to a new location in the same mm, 23321da177e4SLinus Torvalds * prior to moving page table entries, to effect an mremap move. 23331da177e4SLinus Torvalds */ 23341da177e4SLinus Torvalds struct vm_area_struct *copy_vma(struct vm_area_struct **vmap, 23351da177e4SLinus Torvalds unsigned long addr, unsigned long len, pgoff_t pgoff) 23361da177e4SLinus Torvalds { 23371da177e4SLinus Torvalds struct vm_area_struct *vma = *vmap; 23381da177e4SLinus Torvalds unsigned long vma_start = vma->vm_start; 23391da177e4SLinus Torvalds struct mm_struct *mm = vma->vm_mm; 23401da177e4SLinus Torvalds struct vm_area_struct *new_vma, *prev; 23411da177e4SLinus Torvalds struct rb_node **rb_link, *rb_parent; 23421da177e4SLinus Torvalds struct mempolicy *pol; 23431da177e4SLinus Torvalds 23441da177e4SLinus Torvalds /* 23451da177e4SLinus Torvalds * If anonymous vma has not yet been faulted, update new pgoff 23461da177e4SLinus Torvalds * to match new location, to increase its chance of merging. 23471da177e4SLinus Torvalds */ 23481da177e4SLinus Torvalds if (!vma->vm_file && !vma->anon_vma) 23491da177e4SLinus Torvalds pgoff = addr >> PAGE_SHIFT; 23501da177e4SLinus Torvalds 23511da177e4SLinus Torvalds find_vma_prepare(mm, addr, &prev, &rb_link, &rb_parent); 23521da177e4SLinus Torvalds new_vma = vma_merge(mm, prev, addr, addr + len, vma->vm_flags, 23531da177e4SLinus Torvalds vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma)); 23541da177e4SLinus Torvalds if (new_vma) { 23551da177e4SLinus Torvalds /* 23561da177e4SLinus Torvalds * Source vma may have been merged into new_vma 23571da177e4SLinus Torvalds */ 23581da177e4SLinus Torvalds if (vma_start >= new_vma->vm_start && 23591da177e4SLinus Torvalds vma_start < new_vma->vm_end) 23601da177e4SLinus Torvalds *vmap = new_vma; 23611da177e4SLinus Torvalds } else { 2362e94b1766SChristoph Lameter new_vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL); 23631da177e4SLinus Torvalds if (new_vma) { 23641da177e4SLinus Torvalds *new_vma = *vma; 2365846a16bfSLee Schermerhorn pol = mpol_dup(vma_policy(vma)); 23665beb4930SRik van Riel if (IS_ERR(pol)) 23675beb4930SRik van Riel goto out_free_vma; 23685beb4930SRik van Riel INIT_LIST_HEAD(&new_vma->anon_vma_chain); 23695beb4930SRik van Riel if (anon_vma_clone(new_vma, vma)) 23705beb4930SRik van Riel goto out_free_mempol; 23711da177e4SLinus Torvalds vma_set_policy(new_vma, pol); 23721da177e4SLinus Torvalds new_vma->vm_start = addr; 23731da177e4SLinus Torvalds new_vma->vm_end = addr + len; 23741da177e4SLinus Torvalds new_vma->vm_pgoff = pgoff; 2375925d1c40SMatt Helsley if (new_vma->vm_file) { 23761da177e4SLinus Torvalds get_file(new_vma->vm_file); 2377925d1c40SMatt Helsley if (vma->vm_flags & VM_EXECUTABLE) 2378925d1c40SMatt Helsley added_exe_file_vma(mm); 2379925d1c40SMatt Helsley } 23801da177e4SLinus Torvalds if (new_vma->vm_ops && new_vma->vm_ops->open) 23811da177e4SLinus Torvalds new_vma->vm_ops->open(new_vma); 23821da177e4SLinus Torvalds vma_link(mm, new_vma, prev, rb_link, rb_parent); 23831da177e4SLinus Torvalds } 23841da177e4SLinus Torvalds } 23851da177e4SLinus Torvalds return new_vma; 23865beb4930SRik van Riel 23875beb4930SRik van Riel out_free_mempol: 23885beb4930SRik van Riel mpol_put(pol); 23895beb4930SRik van Riel out_free_vma: 23905beb4930SRik van Riel kmem_cache_free(vm_area_cachep, new_vma); 23915beb4930SRik van Riel return NULL; 23921da177e4SLinus Torvalds } 2393119f657cSakpm@osdl.org 2394119f657cSakpm@osdl.org /* 2395119f657cSakpm@osdl.org * Return true if the calling process may expand its vm space by the passed 2396119f657cSakpm@osdl.org * number of pages 2397119f657cSakpm@osdl.org */ 2398119f657cSakpm@osdl.org int may_expand_vm(struct mm_struct *mm, unsigned long npages) 2399119f657cSakpm@osdl.org { 2400119f657cSakpm@osdl.org unsigned long cur = mm->total_vm; /* pages */ 2401119f657cSakpm@osdl.org unsigned long lim; 2402119f657cSakpm@osdl.org 240359e99e5bSJiri Slaby lim = rlimit(RLIMIT_AS) >> PAGE_SHIFT; 2404119f657cSakpm@osdl.org 2405119f657cSakpm@osdl.org if (cur + npages > lim) 2406119f657cSakpm@osdl.org return 0; 2407119f657cSakpm@osdl.org return 1; 2408119f657cSakpm@osdl.org } 2409fa5dc22fSRoland McGrath 2410fa5dc22fSRoland McGrath 2411b1d0e4f5SNick Piggin static int special_mapping_fault(struct vm_area_struct *vma, 2412b1d0e4f5SNick Piggin struct vm_fault *vmf) 2413fa5dc22fSRoland McGrath { 2414b1d0e4f5SNick Piggin pgoff_t pgoff; 2415fa5dc22fSRoland McGrath struct page **pages; 2416fa5dc22fSRoland McGrath 2417b1d0e4f5SNick Piggin /* 2418b1d0e4f5SNick Piggin * special mappings have no vm_file, and in that case, the mm 2419b1d0e4f5SNick Piggin * uses vm_pgoff internally. So we have to subtract it from here. 2420b1d0e4f5SNick Piggin * We are allowed to do this because we are the mm; do not copy 2421b1d0e4f5SNick Piggin * this code into drivers! 2422b1d0e4f5SNick Piggin */ 2423b1d0e4f5SNick Piggin pgoff = vmf->pgoff - vma->vm_pgoff; 2424fa5dc22fSRoland McGrath 2425b1d0e4f5SNick Piggin for (pages = vma->vm_private_data; pgoff && *pages; ++pages) 2426b1d0e4f5SNick Piggin pgoff--; 2427fa5dc22fSRoland McGrath 2428fa5dc22fSRoland McGrath if (*pages) { 2429fa5dc22fSRoland McGrath struct page *page = *pages; 2430fa5dc22fSRoland McGrath get_page(page); 2431b1d0e4f5SNick Piggin vmf->page = page; 2432b1d0e4f5SNick Piggin return 0; 2433fa5dc22fSRoland McGrath } 2434fa5dc22fSRoland McGrath 2435b1d0e4f5SNick Piggin return VM_FAULT_SIGBUS; 2436fa5dc22fSRoland McGrath } 2437fa5dc22fSRoland McGrath 2438fa5dc22fSRoland McGrath /* 2439fa5dc22fSRoland McGrath * Having a close hook prevents vma merging regardless of flags. 2440fa5dc22fSRoland McGrath */ 2441fa5dc22fSRoland McGrath static void special_mapping_close(struct vm_area_struct *vma) 2442fa5dc22fSRoland McGrath { 2443fa5dc22fSRoland McGrath } 2444fa5dc22fSRoland McGrath 2445f0f37e2fSAlexey Dobriyan static const struct vm_operations_struct special_mapping_vmops = { 2446fa5dc22fSRoland McGrath .close = special_mapping_close, 2447b1d0e4f5SNick Piggin .fault = special_mapping_fault, 2448fa5dc22fSRoland McGrath }; 2449fa5dc22fSRoland McGrath 2450fa5dc22fSRoland McGrath /* 2451fa5dc22fSRoland McGrath * Called with mm->mmap_sem held for writing. 2452fa5dc22fSRoland McGrath * Insert a new vma covering the given region, with the given flags. 2453fa5dc22fSRoland McGrath * Its pages are supplied by the given array of struct page *. 2454fa5dc22fSRoland McGrath * The array can be shorter than len >> PAGE_SHIFT if it's null-terminated. 2455fa5dc22fSRoland McGrath * The region past the last page supplied will always produce SIGBUS. 2456fa5dc22fSRoland McGrath * The array pointer and the pages it points to are assumed to stay alive 2457fa5dc22fSRoland McGrath * for as long as this mapping might exist. 2458fa5dc22fSRoland McGrath */ 2459fa5dc22fSRoland McGrath int install_special_mapping(struct mm_struct *mm, 2460fa5dc22fSRoland McGrath unsigned long addr, unsigned long len, 2461fa5dc22fSRoland McGrath unsigned long vm_flags, struct page **pages) 2462fa5dc22fSRoland McGrath { 2463fa5dc22fSRoland McGrath struct vm_area_struct *vma; 2464fa5dc22fSRoland McGrath 2465fa5dc22fSRoland McGrath vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL); 2466fa5dc22fSRoland McGrath if (unlikely(vma == NULL)) 2467fa5dc22fSRoland McGrath return -ENOMEM; 2468fa5dc22fSRoland McGrath 24695beb4930SRik van Riel INIT_LIST_HEAD(&vma->anon_vma_chain); 2470fa5dc22fSRoland McGrath vma->vm_mm = mm; 2471fa5dc22fSRoland McGrath vma->vm_start = addr; 2472fa5dc22fSRoland McGrath vma->vm_end = addr + len; 2473fa5dc22fSRoland McGrath 24742f98735cSNick Piggin vma->vm_flags = vm_flags | mm->def_flags | VM_DONTEXPAND; 24753ed75eb8SColy Li vma->vm_page_prot = vm_get_page_prot(vma->vm_flags); 2476fa5dc22fSRoland McGrath 2477fa5dc22fSRoland McGrath vma->vm_ops = &special_mapping_vmops; 2478fa5dc22fSRoland McGrath vma->vm_private_data = pages; 2479fa5dc22fSRoland McGrath 2480fa5dc22fSRoland McGrath if (unlikely(insert_vm_struct(mm, vma))) { 2481fa5dc22fSRoland McGrath kmem_cache_free(vm_area_cachep, vma); 2482fa5dc22fSRoland McGrath return -ENOMEM; 2483fa5dc22fSRoland McGrath } 2484fa5dc22fSRoland McGrath 2485fa5dc22fSRoland McGrath mm->total_vm += len >> PAGE_SHIFT; 2486fa5dc22fSRoland McGrath 2487cdd6c482SIngo Molnar perf_event_mmap(vma); 2488089dd79dSPeter Zijlstra 2489fa5dc22fSRoland McGrath return 0; 2490fa5dc22fSRoland McGrath } 24917906d00cSAndrea Arcangeli 24927906d00cSAndrea Arcangeli static DEFINE_MUTEX(mm_all_locks_mutex); 24937906d00cSAndrea Arcangeli 2494454ed842SPeter Zijlstra static void vm_lock_anon_vma(struct mm_struct *mm, struct anon_vma *anon_vma) 24957906d00cSAndrea Arcangeli { 2496012f1800SRik van Riel if (!test_bit(0, (unsigned long *) &anon_vma->root->head.next)) { 24977906d00cSAndrea Arcangeli /* 24987906d00cSAndrea Arcangeli * The LSB of head.next can't change from under us 24997906d00cSAndrea Arcangeli * because we hold the mm_all_locks_mutex. 25007906d00cSAndrea Arcangeli */ 2501012f1800SRik van Riel spin_lock_nest_lock(&anon_vma->root->lock, &mm->mmap_sem); 25027906d00cSAndrea Arcangeli /* 25037906d00cSAndrea Arcangeli * We can safely modify head.next after taking the 2504012f1800SRik van Riel * anon_vma->root->lock. If some other vma in this mm shares 25057906d00cSAndrea Arcangeli * the same anon_vma we won't take it again. 25067906d00cSAndrea Arcangeli * 25077906d00cSAndrea Arcangeli * No need of atomic instructions here, head.next 25087906d00cSAndrea Arcangeli * can't change from under us thanks to the 2509012f1800SRik van Riel * anon_vma->root->lock. 25107906d00cSAndrea Arcangeli */ 25117906d00cSAndrea Arcangeli if (__test_and_set_bit(0, (unsigned long *) 2512012f1800SRik van Riel &anon_vma->root->head.next)) 25137906d00cSAndrea Arcangeli BUG(); 25147906d00cSAndrea Arcangeli } 25157906d00cSAndrea Arcangeli } 25167906d00cSAndrea Arcangeli 2517454ed842SPeter Zijlstra static void vm_lock_mapping(struct mm_struct *mm, struct address_space *mapping) 25187906d00cSAndrea Arcangeli { 25197906d00cSAndrea Arcangeli if (!test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) { 25207906d00cSAndrea Arcangeli /* 25217906d00cSAndrea Arcangeli * AS_MM_ALL_LOCKS can't change from under us because 25227906d00cSAndrea Arcangeli * we hold the mm_all_locks_mutex. 25237906d00cSAndrea Arcangeli * 25247906d00cSAndrea Arcangeli * Operations on ->flags have to be atomic because 25257906d00cSAndrea Arcangeli * even if AS_MM_ALL_LOCKS is stable thanks to the 25267906d00cSAndrea Arcangeli * mm_all_locks_mutex, there may be other cpus 25277906d00cSAndrea Arcangeli * changing other bitflags in parallel to us. 25287906d00cSAndrea Arcangeli */ 25297906d00cSAndrea Arcangeli if (test_and_set_bit(AS_MM_ALL_LOCKS, &mapping->flags)) 25307906d00cSAndrea Arcangeli BUG(); 2531454ed842SPeter Zijlstra spin_lock_nest_lock(&mapping->i_mmap_lock, &mm->mmap_sem); 25327906d00cSAndrea Arcangeli } 25337906d00cSAndrea Arcangeli } 25347906d00cSAndrea Arcangeli 25357906d00cSAndrea Arcangeli /* 25367906d00cSAndrea Arcangeli * This operation locks against the VM for all pte/vma/mm related 25377906d00cSAndrea Arcangeli * operations that could ever happen on a certain mm. This includes 25387906d00cSAndrea Arcangeli * vmtruncate, try_to_unmap, and all page faults. 25397906d00cSAndrea Arcangeli * 25407906d00cSAndrea Arcangeli * The caller must take the mmap_sem in write mode before calling 25417906d00cSAndrea Arcangeli * mm_take_all_locks(). The caller isn't allowed to release the 25427906d00cSAndrea Arcangeli * mmap_sem until mm_drop_all_locks() returns. 25437906d00cSAndrea Arcangeli * 25447906d00cSAndrea Arcangeli * mmap_sem in write mode is required in order to block all operations 25457906d00cSAndrea Arcangeli * that could modify pagetables and free pages without need of 25467906d00cSAndrea Arcangeli * altering the vma layout (for example populate_range() with 25477906d00cSAndrea Arcangeli * nonlinear vmas). It's also needed in write mode to avoid new 25487906d00cSAndrea Arcangeli * anon_vmas to be associated with existing vmas. 25497906d00cSAndrea Arcangeli * 25507906d00cSAndrea Arcangeli * A single task can't take more than one mm_take_all_locks() in a row 25517906d00cSAndrea Arcangeli * or it would deadlock. 25527906d00cSAndrea Arcangeli * 25537906d00cSAndrea Arcangeli * The LSB in anon_vma->head.next and the AS_MM_ALL_LOCKS bitflag in 25547906d00cSAndrea Arcangeli * mapping->flags avoid to take the same lock twice, if more than one 25557906d00cSAndrea Arcangeli * vma in this mm is backed by the same anon_vma or address_space. 25567906d00cSAndrea Arcangeli * 25577906d00cSAndrea Arcangeli * We can take all the locks in random order because the VM code 25587906d00cSAndrea Arcangeli * taking i_mmap_lock or anon_vma->lock outside the mmap_sem never 25597906d00cSAndrea Arcangeli * takes more than one of them in a row. Secondly we're protected 25607906d00cSAndrea Arcangeli * against a concurrent mm_take_all_locks() by the mm_all_locks_mutex. 25617906d00cSAndrea Arcangeli * 25627906d00cSAndrea Arcangeli * mm_take_all_locks() and mm_drop_all_locks are expensive operations 25637906d00cSAndrea Arcangeli * that may have to take thousand of locks. 25647906d00cSAndrea Arcangeli * 25657906d00cSAndrea Arcangeli * mm_take_all_locks() can fail if it's interrupted by signals. 25667906d00cSAndrea Arcangeli */ 25677906d00cSAndrea Arcangeli int mm_take_all_locks(struct mm_struct *mm) 25687906d00cSAndrea Arcangeli { 25697906d00cSAndrea Arcangeli struct vm_area_struct *vma; 25705beb4930SRik van Riel struct anon_vma_chain *avc; 25717906d00cSAndrea Arcangeli int ret = -EINTR; 25727906d00cSAndrea Arcangeli 25737906d00cSAndrea Arcangeli BUG_ON(down_read_trylock(&mm->mmap_sem)); 25747906d00cSAndrea Arcangeli 25757906d00cSAndrea Arcangeli mutex_lock(&mm_all_locks_mutex); 25767906d00cSAndrea Arcangeli 25777906d00cSAndrea Arcangeli for (vma = mm->mmap; vma; vma = vma->vm_next) { 25787906d00cSAndrea Arcangeli if (signal_pending(current)) 25797906d00cSAndrea Arcangeli goto out_unlock; 25807906d00cSAndrea Arcangeli if (vma->vm_file && vma->vm_file->f_mapping) 2581454ed842SPeter Zijlstra vm_lock_mapping(mm, vma->vm_file->f_mapping); 25827906d00cSAndrea Arcangeli } 25837cd5a02fSPeter Zijlstra 25847cd5a02fSPeter Zijlstra for (vma = mm->mmap; vma; vma = vma->vm_next) { 25857cd5a02fSPeter Zijlstra if (signal_pending(current)) 25867cd5a02fSPeter Zijlstra goto out_unlock; 25877cd5a02fSPeter Zijlstra if (vma->anon_vma) 25885beb4930SRik van Riel list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) 25895beb4930SRik van Riel vm_lock_anon_vma(mm, avc->anon_vma); 25907cd5a02fSPeter Zijlstra } 25917cd5a02fSPeter Zijlstra 25927906d00cSAndrea Arcangeli ret = 0; 25937906d00cSAndrea Arcangeli 25947906d00cSAndrea Arcangeli out_unlock: 25957906d00cSAndrea Arcangeli if (ret) 25967906d00cSAndrea Arcangeli mm_drop_all_locks(mm); 25977906d00cSAndrea Arcangeli 25987906d00cSAndrea Arcangeli return ret; 25997906d00cSAndrea Arcangeli } 26007906d00cSAndrea Arcangeli 26017906d00cSAndrea Arcangeli static void vm_unlock_anon_vma(struct anon_vma *anon_vma) 26027906d00cSAndrea Arcangeli { 2603012f1800SRik van Riel if (test_bit(0, (unsigned long *) &anon_vma->root->head.next)) { 26047906d00cSAndrea Arcangeli /* 26057906d00cSAndrea Arcangeli * The LSB of head.next can't change to 0 from under 26067906d00cSAndrea Arcangeli * us because we hold the mm_all_locks_mutex. 26077906d00cSAndrea Arcangeli * 26087906d00cSAndrea Arcangeli * We must however clear the bitflag before unlocking 26097906d00cSAndrea Arcangeli * the vma so the users using the anon_vma->head will 26107906d00cSAndrea Arcangeli * never see our bitflag. 26117906d00cSAndrea Arcangeli * 26127906d00cSAndrea Arcangeli * No need of atomic instructions here, head.next 26137906d00cSAndrea Arcangeli * can't change from under us until we release the 2614012f1800SRik van Riel * anon_vma->root->lock. 26157906d00cSAndrea Arcangeli */ 26167906d00cSAndrea Arcangeli if (!__test_and_clear_bit(0, (unsigned long *) 2617012f1800SRik van Riel &anon_vma->root->head.next)) 26187906d00cSAndrea Arcangeli BUG(); 2619cba48b98SRik van Riel anon_vma_unlock(anon_vma); 26207906d00cSAndrea Arcangeli } 26217906d00cSAndrea Arcangeli } 26227906d00cSAndrea Arcangeli 26237906d00cSAndrea Arcangeli static void vm_unlock_mapping(struct address_space *mapping) 26247906d00cSAndrea Arcangeli { 26257906d00cSAndrea Arcangeli if (test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) { 26267906d00cSAndrea Arcangeli /* 26277906d00cSAndrea Arcangeli * AS_MM_ALL_LOCKS can't change to 0 from under us 26287906d00cSAndrea Arcangeli * because we hold the mm_all_locks_mutex. 26297906d00cSAndrea Arcangeli */ 26307906d00cSAndrea Arcangeli spin_unlock(&mapping->i_mmap_lock); 26317906d00cSAndrea Arcangeli if (!test_and_clear_bit(AS_MM_ALL_LOCKS, 26327906d00cSAndrea Arcangeli &mapping->flags)) 26337906d00cSAndrea Arcangeli BUG(); 26347906d00cSAndrea Arcangeli } 26357906d00cSAndrea Arcangeli } 26367906d00cSAndrea Arcangeli 26377906d00cSAndrea Arcangeli /* 26387906d00cSAndrea Arcangeli * The mmap_sem cannot be released by the caller until 26397906d00cSAndrea Arcangeli * mm_drop_all_locks() returns. 26407906d00cSAndrea Arcangeli */ 26417906d00cSAndrea Arcangeli void mm_drop_all_locks(struct mm_struct *mm) 26427906d00cSAndrea Arcangeli { 26437906d00cSAndrea Arcangeli struct vm_area_struct *vma; 26445beb4930SRik van Riel struct anon_vma_chain *avc; 26457906d00cSAndrea Arcangeli 26467906d00cSAndrea Arcangeli BUG_ON(down_read_trylock(&mm->mmap_sem)); 26477906d00cSAndrea Arcangeli BUG_ON(!mutex_is_locked(&mm_all_locks_mutex)); 26487906d00cSAndrea Arcangeli 26497906d00cSAndrea Arcangeli for (vma = mm->mmap; vma; vma = vma->vm_next) { 26507906d00cSAndrea Arcangeli if (vma->anon_vma) 26515beb4930SRik van Riel list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) 26525beb4930SRik van Riel vm_unlock_anon_vma(avc->anon_vma); 26537906d00cSAndrea Arcangeli if (vma->vm_file && vma->vm_file->f_mapping) 26547906d00cSAndrea Arcangeli vm_unlock_mapping(vma->vm_file->f_mapping); 26557906d00cSAndrea Arcangeli } 26567906d00cSAndrea Arcangeli 26577906d00cSAndrea Arcangeli mutex_unlock(&mm_all_locks_mutex); 26587906d00cSAndrea Arcangeli } 26598feae131SDavid Howells 26608feae131SDavid Howells /* 26618feae131SDavid Howells * initialise the VMA slab 26628feae131SDavid Howells */ 26638feae131SDavid Howells void __init mmap_init(void) 26648feae131SDavid Howells { 266500a62ce9SKOSAKI Motohiro int ret; 266600a62ce9SKOSAKI Motohiro 266700a62ce9SKOSAKI Motohiro ret = percpu_counter_init(&vm_committed_as, 0); 266800a62ce9SKOSAKI Motohiro VM_BUG_ON(ret); 26698feae131SDavid Howells } 2670