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> 25b95f1b31SPaul Gortmaker #include <linux/export.h> 261da177e4SLinus Torvalds #include <linux/mount.h> 271da177e4SLinus Torvalds #include <linux/mempolicy.h> 281da177e4SLinus Torvalds #include <linux/rmap.h> 29cddb8a5cSAndrea Arcangeli #include <linux/mmu_notifier.h> 30cdd6c482SIngo Molnar #include <linux/perf_event.h> 31120a795dSAl Viro #include <linux/audit.h> 32b15d00b6SAndrea Arcangeli #include <linux/khugepaged.h> 332b144498SSrikar Dronamraju #include <linux/uprobes.h> 34d3737187SMichel Lespinasse #include <linux/rbtree_augmented.h> 35cf4aebc2SClark Williams #include <linux/sched/sysctl.h> 361da177e4SLinus Torvalds 371da177e4SLinus Torvalds #include <asm/uaccess.h> 381da177e4SLinus Torvalds #include <asm/cacheflush.h> 391da177e4SLinus Torvalds #include <asm/tlb.h> 40d6dd61c8SJeremy Fitzhardinge #include <asm/mmu_context.h> 411da177e4SLinus Torvalds 4242b77728SJan Beulich #include "internal.h" 4342b77728SJan Beulich 443a459756SKirill Korotaev #ifndef arch_mmap_check 453a459756SKirill Korotaev #define arch_mmap_check(addr, len, flags) (0) 463a459756SKirill Korotaev #endif 473a459756SKirill Korotaev 4808e7d9b5SMartin Schwidefsky #ifndef arch_rebalance_pgtables 4908e7d9b5SMartin Schwidefsky #define arch_rebalance_pgtables(addr, len) (addr) 5008e7d9b5SMartin Schwidefsky #endif 5108e7d9b5SMartin Schwidefsky 52e0da382cSHugh Dickins static void unmap_region(struct mm_struct *mm, 53e0da382cSHugh Dickins struct vm_area_struct *vma, struct vm_area_struct *prev, 54e0da382cSHugh Dickins unsigned long start, unsigned long end); 55e0da382cSHugh Dickins 561da177e4SLinus Torvalds /* description of effects of mapping type and prot in current implementation. 571da177e4SLinus Torvalds * this is due to the limited x86 page protection hardware. The expected 581da177e4SLinus Torvalds * behavior is in parens: 591da177e4SLinus Torvalds * 601da177e4SLinus Torvalds * map_type prot 611da177e4SLinus Torvalds * PROT_NONE PROT_READ PROT_WRITE PROT_EXEC 621da177e4SLinus Torvalds * MAP_SHARED r: (no) no r: (yes) yes r: (no) yes r: (no) yes 631da177e4SLinus Torvalds * w: (no) no w: (no) no w: (yes) yes w: (no) no 641da177e4SLinus Torvalds * x: (no) no x: (no) yes x: (no) yes x: (yes) yes 651da177e4SLinus Torvalds * 661da177e4SLinus Torvalds * MAP_PRIVATE r: (no) no r: (yes) yes r: (no) yes r: (no) yes 671da177e4SLinus Torvalds * w: (no) no w: (no) no w: (copy) copy w: (no) no 681da177e4SLinus Torvalds * x: (no) no x: (no) yes x: (no) yes x: (yes) yes 691da177e4SLinus Torvalds * 701da177e4SLinus Torvalds */ 711da177e4SLinus Torvalds pgprot_t protection_map[16] = { 721da177e4SLinus Torvalds __P000, __P001, __P010, __P011, __P100, __P101, __P110, __P111, 731da177e4SLinus Torvalds __S000, __S001, __S010, __S011, __S100, __S101, __S110, __S111 741da177e4SLinus Torvalds }; 751da177e4SLinus Torvalds 76804af2cfSHugh Dickins pgprot_t vm_get_page_prot(unsigned long vm_flags) 77804af2cfSHugh Dickins { 78b845f313SDave Kleikamp return __pgprot(pgprot_val(protection_map[vm_flags & 79b845f313SDave Kleikamp (VM_READ|VM_WRITE|VM_EXEC|VM_SHARED)]) | 80b845f313SDave Kleikamp pgprot_val(arch_vm_get_page_prot(vm_flags))); 81804af2cfSHugh Dickins } 82804af2cfSHugh Dickins EXPORT_SYMBOL(vm_get_page_prot); 83804af2cfSHugh Dickins 8434679d7eSShaohua Li int sysctl_overcommit_memory __read_mostly = OVERCOMMIT_GUESS; /* heuristic overcommit */ 8534679d7eSShaohua Li int sysctl_overcommit_ratio __read_mostly = 50; /* default is 50% */ 86c3d8c141SChristoph Lameter int sysctl_max_map_count __read_mostly = DEFAULT_MAX_MAP_COUNT; 87c9b1d098SAndrew Shewmaker unsigned long sysctl_user_reserve_kbytes __read_mostly = 1UL << 17; /* 128MB */ 88*4eeab4f5SAndrew Shewmaker unsigned long sysctl_admin_reserve_kbytes __read_mostly = 1UL << 13; /* 8MB */ 8934679d7eSShaohua Li /* 9034679d7eSShaohua Li * Make sure vm_committed_as in one cacheline and not cacheline shared with 9134679d7eSShaohua Li * other variables. It can be updated by several CPUs frequently. 9234679d7eSShaohua Li */ 9334679d7eSShaohua Li struct percpu_counter vm_committed_as ____cacheline_aligned_in_smp; 941da177e4SLinus Torvalds 951da177e4SLinus Torvalds /* 96997071bcSK. Y. Srinivasan * The global memory commitment made in the system can be a metric 97997071bcSK. Y. Srinivasan * that can be used to drive ballooning decisions when Linux is hosted 98997071bcSK. Y. Srinivasan * as a guest. On Hyper-V, the host implements a policy engine for dynamically 99997071bcSK. Y. Srinivasan * balancing memory across competing virtual machines that are hosted. 100997071bcSK. Y. Srinivasan * Several metrics drive this policy engine including the guest reported 101997071bcSK. Y. Srinivasan * memory commitment. 102997071bcSK. Y. Srinivasan */ 103997071bcSK. Y. Srinivasan unsigned long vm_memory_committed(void) 104997071bcSK. Y. Srinivasan { 105997071bcSK. Y. Srinivasan return percpu_counter_read_positive(&vm_committed_as); 106997071bcSK. Y. Srinivasan } 107997071bcSK. Y. Srinivasan EXPORT_SYMBOL_GPL(vm_memory_committed); 108997071bcSK. Y. Srinivasan 109997071bcSK. Y. Srinivasan /* 1101da177e4SLinus Torvalds * Check that a process has enough memory to allocate a new virtual 1111da177e4SLinus Torvalds * mapping. 0 means there is enough memory for the allocation to 1121da177e4SLinus Torvalds * succeed and -ENOMEM implies there is not. 1131da177e4SLinus Torvalds * 1141da177e4SLinus Torvalds * We currently support three overcommit policies, which are set via the 1151da177e4SLinus Torvalds * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting 1161da177e4SLinus Torvalds * 1171da177e4SLinus Torvalds * Strict overcommit modes added 2002 Feb 26 by Alan Cox. 1181da177e4SLinus Torvalds * Additional code 2002 Jul 20 by Robert Love. 1191da177e4SLinus Torvalds * 1201da177e4SLinus Torvalds * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise. 1211da177e4SLinus Torvalds * 1221da177e4SLinus Torvalds * Note this is a helper function intended to be used by LSMs which 1231da177e4SLinus Torvalds * wish to use this logic. 1241da177e4SLinus Torvalds */ 12534b4e4aaSAlan Cox int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin) 1261da177e4SLinus Torvalds { 127c9b1d098SAndrew Shewmaker unsigned long free, allowed, reserve; 1281da177e4SLinus Torvalds 1291da177e4SLinus Torvalds vm_acct_memory(pages); 1301da177e4SLinus Torvalds 1311da177e4SLinus Torvalds /* 1321da177e4SLinus Torvalds * Sometimes we want to use more memory than we have 1331da177e4SLinus Torvalds */ 1341da177e4SLinus Torvalds if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS) 1351da177e4SLinus Torvalds return 0; 1361da177e4SLinus Torvalds 1371da177e4SLinus Torvalds if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) { 138c15bef30SDmitry Fink free = global_page_state(NR_FREE_PAGES); 139c15bef30SDmitry Fink free += global_page_state(NR_FILE_PAGES); 1401da177e4SLinus Torvalds 141c15bef30SDmitry Fink /* 142c15bef30SDmitry Fink * shmem pages shouldn't be counted as free in this 143c15bef30SDmitry Fink * case, they can't be purged, only swapped out, and 144c15bef30SDmitry Fink * that won't affect the overall amount of available 145c15bef30SDmitry Fink * memory in the system. 146c15bef30SDmitry Fink */ 147c15bef30SDmitry Fink free -= global_page_state(NR_SHMEM); 148c15bef30SDmitry Fink 149ec8acf20SShaohua Li free += get_nr_swap_pages(); 1501da177e4SLinus Torvalds 1511da177e4SLinus Torvalds /* 1521da177e4SLinus Torvalds * Any slabs which are created with the 1531da177e4SLinus Torvalds * SLAB_RECLAIM_ACCOUNT flag claim to have contents 1541da177e4SLinus Torvalds * which are reclaimable, under pressure. The dentry 1551da177e4SLinus Torvalds * cache and most inode caches should fall into this 1561da177e4SLinus Torvalds */ 157972d1a7bSChristoph Lameter free += global_page_state(NR_SLAB_RECLAIMABLE); 1581da177e4SLinus Torvalds 1591da177e4SLinus Torvalds /* 160c15bef30SDmitry Fink * Leave reserved pages. The pages are not for anonymous pages. 161c15bef30SDmitry Fink */ 162c15bef30SDmitry Fink if (free <= totalreserve_pages) 163c15bef30SDmitry Fink goto error; 164c15bef30SDmitry Fink else 165c15bef30SDmitry Fink free -= totalreserve_pages; 166c15bef30SDmitry Fink 167c15bef30SDmitry Fink /* 168*4eeab4f5SAndrew Shewmaker * Reserve some for root 1691da177e4SLinus Torvalds */ 1701da177e4SLinus Torvalds if (!cap_sys_admin) 171*4eeab4f5SAndrew Shewmaker free -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10); 1721da177e4SLinus Torvalds 1731da177e4SLinus Torvalds if (free > pages) 1741da177e4SLinus Torvalds return 0; 1751da177e4SLinus Torvalds 1766d9f7839SHideo AOKI goto error; 1771da177e4SLinus Torvalds } 1781da177e4SLinus Torvalds 1791da177e4SLinus Torvalds allowed = (totalram_pages - hugetlb_total_pages()) 1801da177e4SLinus Torvalds * sysctl_overcommit_ratio / 100; 1811da177e4SLinus Torvalds /* 182*4eeab4f5SAndrew Shewmaker * Reserve some for root 1831da177e4SLinus Torvalds */ 1841da177e4SLinus Torvalds if (!cap_sys_admin) 185*4eeab4f5SAndrew Shewmaker allowed -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10); 1861da177e4SLinus Torvalds allowed += total_swap_pages; 1871da177e4SLinus Torvalds 188c9b1d098SAndrew Shewmaker /* 189c9b1d098SAndrew Shewmaker * Don't let a single process grow so big a user can't recover 190c9b1d098SAndrew Shewmaker */ 191c9b1d098SAndrew Shewmaker if (mm) { 192c9b1d098SAndrew Shewmaker reserve = sysctl_user_reserve_kbytes >> (PAGE_SHIFT - 10); 193c9b1d098SAndrew Shewmaker allowed -= min(mm->total_vm / 32, reserve); 194c9b1d098SAndrew Shewmaker } 1951da177e4SLinus Torvalds 19600a62ce9SKOSAKI Motohiro if (percpu_counter_read_positive(&vm_committed_as) < allowed) 1971da177e4SLinus Torvalds return 0; 1986d9f7839SHideo AOKI error: 1991da177e4SLinus Torvalds vm_unacct_memory(pages); 2001da177e4SLinus Torvalds 2011da177e4SLinus Torvalds return -ENOMEM; 2021da177e4SLinus Torvalds } 2031da177e4SLinus Torvalds 2041da177e4SLinus Torvalds /* 2053d48ae45SPeter Zijlstra * Requires inode->i_mapping->i_mmap_mutex 2061da177e4SLinus Torvalds */ 2071da177e4SLinus Torvalds static void __remove_shared_vm_struct(struct vm_area_struct *vma, 2081da177e4SLinus Torvalds struct file *file, struct address_space *mapping) 2091da177e4SLinus Torvalds { 2101da177e4SLinus Torvalds if (vma->vm_flags & VM_DENYWRITE) 211496ad9aaSAl Viro atomic_inc(&file_inode(file)->i_writecount); 2121da177e4SLinus Torvalds if (vma->vm_flags & VM_SHARED) 2131da177e4SLinus Torvalds mapping->i_mmap_writable--; 2141da177e4SLinus Torvalds 2151da177e4SLinus Torvalds flush_dcache_mmap_lock(mapping); 2161da177e4SLinus Torvalds if (unlikely(vma->vm_flags & VM_NONLINEAR)) 2176b2dbba8SMichel Lespinasse list_del_init(&vma->shared.nonlinear); 2181da177e4SLinus Torvalds else 2196b2dbba8SMichel Lespinasse vma_interval_tree_remove(vma, &mapping->i_mmap); 2201da177e4SLinus Torvalds flush_dcache_mmap_unlock(mapping); 2211da177e4SLinus Torvalds } 2221da177e4SLinus Torvalds 2231da177e4SLinus Torvalds /* 2246b2dbba8SMichel Lespinasse * Unlink a file-based vm structure from its interval tree, to hide 225a8fb5618SHugh Dickins * vma from rmap and vmtruncate before freeing its page tables. 2261da177e4SLinus Torvalds */ 227a8fb5618SHugh Dickins void unlink_file_vma(struct vm_area_struct *vma) 2281da177e4SLinus Torvalds { 2291da177e4SLinus Torvalds struct file *file = vma->vm_file; 2301da177e4SLinus Torvalds 2311da177e4SLinus Torvalds if (file) { 2321da177e4SLinus Torvalds struct address_space *mapping = file->f_mapping; 2333d48ae45SPeter Zijlstra mutex_lock(&mapping->i_mmap_mutex); 2341da177e4SLinus Torvalds __remove_shared_vm_struct(vma, file, mapping); 2353d48ae45SPeter Zijlstra mutex_unlock(&mapping->i_mmap_mutex); 2361da177e4SLinus Torvalds } 237a8fb5618SHugh Dickins } 238a8fb5618SHugh Dickins 239a8fb5618SHugh Dickins /* 240a8fb5618SHugh Dickins * Close a vm structure and free it, returning the next. 241a8fb5618SHugh Dickins */ 242a8fb5618SHugh Dickins static struct vm_area_struct *remove_vma(struct vm_area_struct *vma) 243a8fb5618SHugh Dickins { 244a8fb5618SHugh Dickins struct vm_area_struct *next = vma->vm_next; 245a8fb5618SHugh Dickins 246a8fb5618SHugh Dickins might_sleep(); 2471da177e4SLinus Torvalds if (vma->vm_ops && vma->vm_ops->close) 2481da177e4SLinus Torvalds vma->vm_ops->close(vma); 249e9714acfSKonstantin Khlebnikov if (vma->vm_file) 250a8fb5618SHugh Dickins fput(vma->vm_file); 251f0be3d32SLee Schermerhorn mpol_put(vma_policy(vma)); 2521da177e4SLinus Torvalds kmem_cache_free(vm_area_cachep, vma); 253a8fb5618SHugh Dickins return next; 2541da177e4SLinus Torvalds } 2551da177e4SLinus Torvalds 256e4eb1ff6SLinus Torvalds static unsigned long do_brk(unsigned long addr, unsigned long len); 257e4eb1ff6SLinus Torvalds 2586a6160a7SHeiko Carstens SYSCALL_DEFINE1(brk, unsigned long, brk) 2591da177e4SLinus Torvalds { 2601da177e4SLinus Torvalds unsigned long rlim, retval; 2611da177e4SLinus Torvalds unsigned long newbrk, oldbrk; 2621da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 263a5b4592cSJiri Kosina unsigned long min_brk; 264128557ffSMichel Lespinasse bool populate; 2651da177e4SLinus Torvalds 2661da177e4SLinus Torvalds down_write(&mm->mmap_sem); 2671da177e4SLinus Torvalds 268a5b4592cSJiri Kosina #ifdef CONFIG_COMPAT_BRK 2695520e894SJiri Kosina /* 2705520e894SJiri Kosina * CONFIG_COMPAT_BRK can still be overridden by setting 2715520e894SJiri Kosina * randomize_va_space to 2, which will still cause mm->start_brk 2725520e894SJiri Kosina * to be arbitrarily shifted 2735520e894SJiri Kosina */ 2744471a675SJiri Kosina if (current->brk_randomized) 2755520e894SJiri Kosina min_brk = mm->start_brk; 2765520e894SJiri Kosina else 2775520e894SJiri Kosina min_brk = mm->end_data; 278a5b4592cSJiri Kosina #else 279a5b4592cSJiri Kosina min_brk = mm->start_brk; 280a5b4592cSJiri Kosina #endif 281a5b4592cSJiri Kosina if (brk < min_brk) 2821da177e4SLinus Torvalds goto out; 2831e624196SRam Gupta 2841e624196SRam Gupta /* 2851e624196SRam Gupta * Check against rlimit here. If this check is done later after the test 2861e624196SRam Gupta * of oldbrk with newbrk then it can escape the test and let the data 2871e624196SRam Gupta * segment grow beyond its set limit the in case where the limit is 2881e624196SRam Gupta * not page aligned -Ram Gupta 2891e624196SRam Gupta */ 29059e99e5bSJiri Slaby rlim = rlimit(RLIMIT_DATA); 291c1d171a0SJiri Kosina if (rlim < RLIM_INFINITY && (brk - mm->start_brk) + 292c1d171a0SJiri Kosina (mm->end_data - mm->start_data) > rlim) 2931e624196SRam Gupta goto out; 2941e624196SRam Gupta 2951da177e4SLinus Torvalds newbrk = PAGE_ALIGN(brk); 2961da177e4SLinus Torvalds oldbrk = PAGE_ALIGN(mm->brk); 2971da177e4SLinus Torvalds if (oldbrk == newbrk) 2981da177e4SLinus Torvalds goto set_brk; 2991da177e4SLinus Torvalds 3001da177e4SLinus Torvalds /* Always allow shrinking brk. */ 3011da177e4SLinus Torvalds if (brk <= mm->brk) { 3021da177e4SLinus Torvalds if (!do_munmap(mm, newbrk, oldbrk-newbrk)) 3031da177e4SLinus Torvalds goto set_brk; 3041da177e4SLinus Torvalds goto out; 3051da177e4SLinus Torvalds } 3061da177e4SLinus Torvalds 3071da177e4SLinus Torvalds /* Check against existing mmap mappings. */ 3081da177e4SLinus Torvalds if (find_vma_intersection(mm, oldbrk, newbrk+PAGE_SIZE)) 3091da177e4SLinus Torvalds goto out; 3101da177e4SLinus Torvalds 3111da177e4SLinus Torvalds /* Ok, looks good - let it rip. */ 3121da177e4SLinus Torvalds if (do_brk(oldbrk, newbrk-oldbrk) != oldbrk) 3131da177e4SLinus Torvalds goto out; 314128557ffSMichel Lespinasse 3151da177e4SLinus Torvalds set_brk: 3161da177e4SLinus Torvalds mm->brk = brk; 317128557ffSMichel Lespinasse populate = newbrk > oldbrk && (mm->def_flags & VM_LOCKED) != 0; 318128557ffSMichel Lespinasse up_write(&mm->mmap_sem); 319128557ffSMichel Lespinasse if (populate) 320128557ffSMichel Lespinasse mm_populate(oldbrk, newbrk - oldbrk); 321128557ffSMichel Lespinasse return brk; 322128557ffSMichel Lespinasse 3231da177e4SLinus Torvalds out: 3241da177e4SLinus Torvalds retval = mm->brk; 3251da177e4SLinus Torvalds up_write(&mm->mmap_sem); 3261da177e4SLinus Torvalds return retval; 3271da177e4SLinus Torvalds } 3281da177e4SLinus Torvalds 329d3737187SMichel Lespinasse static long vma_compute_subtree_gap(struct vm_area_struct *vma) 330d3737187SMichel Lespinasse { 331d3737187SMichel Lespinasse unsigned long max, subtree_gap; 332d3737187SMichel Lespinasse max = vma->vm_start; 333d3737187SMichel Lespinasse if (vma->vm_prev) 334d3737187SMichel Lespinasse max -= vma->vm_prev->vm_end; 335d3737187SMichel Lespinasse if (vma->vm_rb.rb_left) { 336d3737187SMichel Lespinasse subtree_gap = rb_entry(vma->vm_rb.rb_left, 337d3737187SMichel Lespinasse struct vm_area_struct, vm_rb)->rb_subtree_gap; 338d3737187SMichel Lespinasse if (subtree_gap > max) 339d3737187SMichel Lespinasse max = subtree_gap; 340d3737187SMichel Lespinasse } 341d3737187SMichel Lespinasse if (vma->vm_rb.rb_right) { 342d3737187SMichel Lespinasse subtree_gap = rb_entry(vma->vm_rb.rb_right, 343d3737187SMichel Lespinasse struct vm_area_struct, vm_rb)->rb_subtree_gap; 344d3737187SMichel Lespinasse if (subtree_gap > max) 345d3737187SMichel Lespinasse max = subtree_gap; 346d3737187SMichel Lespinasse } 347d3737187SMichel Lespinasse return max; 348d3737187SMichel Lespinasse } 349d3737187SMichel Lespinasse 350ed8ea815SMichel Lespinasse #ifdef CONFIG_DEBUG_VM_RB 3511da177e4SLinus Torvalds static int browse_rb(struct rb_root *root) 3521da177e4SLinus Torvalds { 3535a0768f6SMichel Lespinasse int i = 0, j, bug = 0; 3541da177e4SLinus Torvalds struct rb_node *nd, *pn = NULL; 3551da177e4SLinus Torvalds unsigned long prev = 0, pend = 0; 3561da177e4SLinus Torvalds 3571da177e4SLinus Torvalds for (nd = rb_first(root); nd; nd = rb_next(nd)) { 3581da177e4SLinus Torvalds struct vm_area_struct *vma; 3591da177e4SLinus Torvalds vma = rb_entry(nd, struct vm_area_struct, vm_rb); 3605a0768f6SMichel Lespinasse if (vma->vm_start < prev) { 3615a0768f6SMichel Lespinasse printk("vm_start %lx prev %lx\n", vma->vm_start, prev); 3625a0768f6SMichel Lespinasse bug = 1; 3635a0768f6SMichel Lespinasse } 3645a0768f6SMichel Lespinasse if (vma->vm_start < pend) { 3651da177e4SLinus Torvalds printk("vm_start %lx pend %lx\n", vma->vm_start, pend); 3665a0768f6SMichel Lespinasse bug = 1; 3675a0768f6SMichel Lespinasse } 3685a0768f6SMichel Lespinasse if (vma->vm_start > vma->vm_end) { 3695a0768f6SMichel Lespinasse printk("vm_end %lx < vm_start %lx\n", 3705a0768f6SMichel Lespinasse vma->vm_end, vma->vm_start); 3715a0768f6SMichel Lespinasse bug = 1; 3725a0768f6SMichel Lespinasse } 3735a0768f6SMichel Lespinasse if (vma->rb_subtree_gap != vma_compute_subtree_gap(vma)) { 3745a0768f6SMichel Lespinasse printk("free gap %lx, correct %lx\n", 3755a0768f6SMichel Lespinasse vma->rb_subtree_gap, 3765a0768f6SMichel Lespinasse vma_compute_subtree_gap(vma)); 3775a0768f6SMichel Lespinasse bug = 1; 3785a0768f6SMichel Lespinasse } 3791da177e4SLinus Torvalds i++; 3801da177e4SLinus Torvalds pn = nd; 381d1af65d1SDavid Miller prev = vma->vm_start; 382d1af65d1SDavid Miller pend = vma->vm_end; 3831da177e4SLinus Torvalds } 3841da177e4SLinus Torvalds j = 0; 3855a0768f6SMichel Lespinasse for (nd = pn; nd; nd = rb_prev(nd)) 3861da177e4SLinus Torvalds j++; 3875a0768f6SMichel Lespinasse if (i != j) { 3885a0768f6SMichel Lespinasse printk("backwards %d, forwards %d\n", j, i); 3895a0768f6SMichel Lespinasse bug = 1; 3901da177e4SLinus Torvalds } 3915a0768f6SMichel Lespinasse return bug ? -1 : i; 3921da177e4SLinus Torvalds } 3931da177e4SLinus Torvalds 394d3737187SMichel Lespinasse static void validate_mm_rb(struct rb_root *root, struct vm_area_struct *ignore) 395d3737187SMichel Lespinasse { 396d3737187SMichel Lespinasse struct rb_node *nd; 397d3737187SMichel Lespinasse 398d3737187SMichel Lespinasse for (nd = rb_first(root); nd; nd = rb_next(nd)) { 399d3737187SMichel Lespinasse struct vm_area_struct *vma; 400d3737187SMichel Lespinasse vma = rb_entry(nd, struct vm_area_struct, vm_rb); 401d3737187SMichel Lespinasse BUG_ON(vma != ignore && 402d3737187SMichel Lespinasse vma->rb_subtree_gap != vma_compute_subtree_gap(vma)); 403d3737187SMichel Lespinasse } 4041da177e4SLinus Torvalds } 4051da177e4SLinus Torvalds 4061da177e4SLinus Torvalds void validate_mm(struct mm_struct *mm) 4071da177e4SLinus Torvalds { 4081da177e4SLinus Torvalds int bug = 0; 4091da177e4SLinus Torvalds int i = 0; 4105a0768f6SMichel Lespinasse unsigned long highest_address = 0; 411ed8ea815SMichel Lespinasse struct vm_area_struct *vma = mm->mmap; 412ed8ea815SMichel Lespinasse while (vma) { 413ed8ea815SMichel Lespinasse struct anon_vma_chain *avc; 41463c3b902SMichel Lespinasse vma_lock_anon_vma(vma); 415ed8ea815SMichel Lespinasse list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) 416ed8ea815SMichel Lespinasse anon_vma_interval_tree_verify(avc); 41763c3b902SMichel Lespinasse vma_unlock_anon_vma(vma); 4185a0768f6SMichel Lespinasse highest_address = vma->vm_end; 419ed8ea815SMichel Lespinasse vma = vma->vm_next; 4201da177e4SLinus Torvalds i++; 4211da177e4SLinus Torvalds } 4225a0768f6SMichel Lespinasse if (i != mm->map_count) { 4235a0768f6SMichel Lespinasse printk("map_count %d vm_next %d\n", mm->map_count, i); 4245a0768f6SMichel Lespinasse bug = 1; 4255a0768f6SMichel Lespinasse } 4265a0768f6SMichel Lespinasse if (highest_address != mm->highest_vm_end) { 4275a0768f6SMichel Lespinasse printk("mm->highest_vm_end %lx, found %lx\n", 4285a0768f6SMichel Lespinasse mm->highest_vm_end, highest_address); 4295a0768f6SMichel Lespinasse bug = 1; 4305a0768f6SMichel Lespinasse } 4311da177e4SLinus Torvalds i = browse_rb(&mm->mm_rb); 4325a0768f6SMichel Lespinasse if (i != mm->map_count) { 4335a0768f6SMichel Lespinasse printk("map_count %d rb %d\n", mm->map_count, i); 4345a0768f6SMichel Lespinasse bug = 1; 4355a0768f6SMichel Lespinasse } 43646a350efSEric Sesterhenn BUG_ON(bug); 4371da177e4SLinus Torvalds } 4381da177e4SLinus Torvalds #else 439d3737187SMichel Lespinasse #define validate_mm_rb(root, ignore) do { } while (0) 4401da177e4SLinus Torvalds #define validate_mm(mm) do { } while (0) 4411da177e4SLinus Torvalds #endif 4421da177e4SLinus Torvalds 443d3737187SMichel Lespinasse RB_DECLARE_CALLBACKS(static, vma_gap_callbacks, struct vm_area_struct, vm_rb, 444d3737187SMichel Lespinasse unsigned long, rb_subtree_gap, vma_compute_subtree_gap) 445d3737187SMichel Lespinasse 446d3737187SMichel Lespinasse /* 447d3737187SMichel Lespinasse * Update augmented rbtree rb_subtree_gap values after vma->vm_start or 448d3737187SMichel Lespinasse * vma->vm_prev->vm_end values changed, without modifying the vma's position 449d3737187SMichel Lespinasse * in the rbtree. 450d3737187SMichel Lespinasse */ 451d3737187SMichel Lespinasse static void vma_gap_update(struct vm_area_struct *vma) 452d3737187SMichel Lespinasse { 453d3737187SMichel Lespinasse /* 454d3737187SMichel Lespinasse * As it turns out, RB_DECLARE_CALLBACKS() already created a callback 455d3737187SMichel Lespinasse * function that does exacltly what we want. 456d3737187SMichel Lespinasse */ 457d3737187SMichel Lespinasse vma_gap_callbacks_propagate(&vma->vm_rb, NULL); 458d3737187SMichel Lespinasse } 459d3737187SMichel Lespinasse 460d3737187SMichel Lespinasse static inline void vma_rb_insert(struct vm_area_struct *vma, 461d3737187SMichel Lespinasse struct rb_root *root) 462d3737187SMichel Lespinasse { 463d3737187SMichel Lespinasse /* All rb_subtree_gap values must be consistent prior to insertion */ 464d3737187SMichel Lespinasse validate_mm_rb(root, NULL); 465d3737187SMichel Lespinasse 466d3737187SMichel Lespinasse rb_insert_augmented(&vma->vm_rb, root, &vma_gap_callbacks); 467d3737187SMichel Lespinasse } 468d3737187SMichel Lespinasse 469d3737187SMichel Lespinasse static void vma_rb_erase(struct vm_area_struct *vma, struct rb_root *root) 470d3737187SMichel Lespinasse { 471d3737187SMichel Lespinasse /* 472d3737187SMichel Lespinasse * All rb_subtree_gap values must be consistent prior to erase, 473d3737187SMichel Lespinasse * with the possible exception of the vma being erased. 474d3737187SMichel Lespinasse */ 475d3737187SMichel Lespinasse validate_mm_rb(root, vma); 476d3737187SMichel Lespinasse 477d3737187SMichel Lespinasse /* 478d3737187SMichel Lespinasse * Note rb_erase_augmented is a fairly large inline function, 479d3737187SMichel Lespinasse * so make sure we instantiate it only once with our desired 480d3737187SMichel Lespinasse * augmented rbtree callbacks. 481d3737187SMichel Lespinasse */ 482d3737187SMichel Lespinasse rb_erase_augmented(&vma->vm_rb, root, &vma_gap_callbacks); 483d3737187SMichel Lespinasse } 484d3737187SMichel Lespinasse 485bf181b9fSMichel Lespinasse /* 486bf181b9fSMichel Lespinasse * vma has some anon_vma assigned, and is already inserted on that 487bf181b9fSMichel Lespinasse * anon_vma's interval trees. 488bf181b9fSMichel Lespinasse * 489bf181b9fSMichel Lespinasse * Before updating the vma's vm_start / vm_end / vm_pgoff fields, the 490bf181b9fSMichel Lespinasse * vma must be removed from the anon_vma's interval trees using 491bf181b9fSMichel Lespinasse * anon_vma_interval_tree_pre_update_vma(). 492bf181b9fSMichel Lespinasse * 493bf181b9fSMichel Lespinasse * After the update, the vma will be reinserted using 494bf181b9fSMichel Lespinasse * anon_vma_interval_tree_post_update_vma(). 495bf181b9fSMichel Lespinasse * 496bf181b9fSMichel Lespinasse * The entire update must be protected by exclusive mmap_sem and by 497bf181b9fSMichel Lespinasse * the root anon_vma's mutex. 498bf181b9fSMichel Lespinasse */ 499bf181b9fSMichel Lespinasse static inline void 500bf181b9fSMichel Lespinasse anon_vma_interval_tree_pre_update_vma(struct vm_area_struct *vma) 501bf181b9fSMichel Lespinasse { 502bf181b9fSMichel Lespinasse struct anon_vma_chain *avc; 503bf181b9fSMichel Lespinasse 504bf181b9fSMichel Lespinasse list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) 505bf181b9fSMichel Lespinasse anon_vma_interval_tree_remove(avc, &avc->anon_vma->rb_root); 506bf181b9fSMichel Lespinasse } 507bf181b9fSMichel Lespinasse 508bf181b9fSMichel Lespinasse static inline void 509bf181b9fSMichel Lespinasse anon_vma_interval_tree_post_update_vma(struct vm_area_struct *vma) 510bf181b9fSMichel Lespinasse { 511bf181b9fSMichel Lespinasse struct anon_vma_chain *avc; 512bf181b9fSMichel Lespinasse 513bf181b9fSMichel Lespinasse list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) 514bf181b9fSMichel Lespinasse anon_vma_interval_tree_insert(avc, &avc->anon_vma->rb_root); 515bf181b9fSMichel Lespinasse } 516bf181b9fSMichel Lespinasse 5176597d783SHugh Dickins static int find_vma_links(struct mm_struct *mm, unsigned long addr, 5186597d783SHugh Dickins unsigned long end, struct vm_area_struct **pprev, 5196597d783SHugh Dickins struct rb_node ***rb_link, struct rb_node **rb_parent) 5201da177e4SLinus Torvalds { 5211da177e4SLinus Torvalds struct rb_node **__rb_link, *__rb_parent, *rb_prev; 5221da177e4SLinus Torvalds 5231da177e4SLinus Torvalds __rb_link = &mm->mm_rb.rb_node; 5241da177e4SLinus Torvalds rb_prev = __rb_parent = NULL; 5251da177e4SLinus Torvalds 5261da177e4SLinus Torvalds while (*__rb_link) { 5271da177e4SLinus Torvalds struct vm_area_struct *vma_tmp; 5281da177e4SLinus Torvalds 5291da177e4SLinus Torvalds __rb_parent = *__rb_link; 5301da177e4SLinus Torvalds vma_tmp = rb_entry(__rb_parent, struct vm_area_struct, vm_rb); 5311da177e4SLinus Torvalds 5321da177e4SLinus Torvalds if (vma_tmp->vm_end > addr) { 5336597d783SHugh Dickins /* Fail if an existing vma overlaps the area */ 5346597d783SHugh Dickins if (vma_tmp->vm_start < end) 5356597d783SHugh Dickins return -ENOMEM; 5361da177e4SLinus Torvalds __rb_link = &__rb_parent->rb_left; 5371da177e4SLinus Torvalds } else { 5381da177e4SLinus Torvalds rb_prev = __rb_parent; 5391da177e4SLinus Torvalds __rb_link = &__rb_parent->rb_right; 5401da177e4SLinus Torvalds } 5411da177e4SLinus Torvalds } 5421da177e4SLinus Torvalds 5431da177e4SLinus Torvalds *pprev = NULL; 5441da177e4SLinus Torvalds if (rb_prev) 5451da177e4SLinus Torvalds *pprev = rb_entry(rb_prev, struct vm_area_struct, vm_rb); 5461da177e4SLinus Torvalds *rb_link = __rb_link; 5471da177e4SLinus Torvalds *rb_parent = __rb_parent; 5486597d783SHugh Dickins return 0; 5491da177e4SLinus Torvalds } 5501da177e4SLinus Torvalds 5511da177e4SLinus Torvalds void __vma_link_rb(struct mm_struct *mm, struct vm_area_struct *vma, 5521da177e4SLinus Torvalds struct rb_node **rb_link, struct rb_node *rb_parent) 5531da177e4SLinus Torvalds { 554d3737187SMichel Lespinasse /* Update tracking information for the gap following the new vma. */ 555d3737187SMichel Lespinasse if (vma->vm_next) 556d3737187SMichel Lespinasse vma_gap_update(vma->vm_next); 557d3737187SMichel Lespinasse else 558d3737187SMichel Lespinasse mm->highest_vm_end = vma->vm_end; 559d3737187SMichel Lespinasse 560d3737187SMichel Lespinasse /* 561d3737187SMichel Lespinasse * vma->vm_prev wasn't known when we followed the rbtree to find the 562d3737187SMichel Lespinasse * correct insertion point for that vma. As a result, we could not 563d3737187SMichel Lespinasse * update the vma vm_rb parents rb_subtree_gap values on the way down. 564d3737187SMichel Lespinasse * So, we first insert the vma with a zero rb_subtree_gap value 565d3737187SMichel Lespinasse * (to be consistent with what we did on the way down), and then 566d3737187SMichel Lespinasse * immediately update the gap to the correct value. Finally we 567d3737187SMichel Lespinasse * rebalance the rbtree after all augmented values have been set. 568d3737187SMichel Lespinasse */ 5691da177e4SLinus Torvalds rb_link_node(&vma->vm_rb, rb_parent, rb_link); 570d3737187SMichel Lespinasse vma->rb_subtree_gap = 0; 571d3737187SMichel Lespinasse vma_gap_update(vma); 572d3737187SMichel Lespinasse vma_rb_insert(vma, &mm->mm_rb); 5731da177e4SLinus Torvalds } 5741da177e4SLinus Torvalds 575cb8f488cSDenys Vlasenko static void __vma_link_file(struct vm_area_struct *vma) 5761da177e4SLinus Torvalds { 5771da177e4SLinus Torvalds struct file *file; 5781da177e4SLinus Torvalds 5791da177e4SLinus Torvalds file = vma->vm_file; 5801da177e4SLinus Torvalds if (file) { 5811da177e4SLinus Torvalds struct address_space *mapping = file->f_mapping; 5821da177e4SLinus Torvalds 5831da177e4SLinus Torvalds if (vma->vm_flags & VM_DENYWRITE) 584496ad9aaSAl Viro atomic_dec(&file_inode(file)->i_writecount); 5851da177e4SLinus Torvalds if (vma->vm_flags & VM_SHARED) 5861da177e4SLinus Torvalds mapping->i_mmap_writable++; 5871da177e4SLinus Torvalds 5881da177e4SLinus Torvalds flush_dcache_mmap_lock(mapping); 5891da177e4SLinus Torvalds if (unlikely(vma->vm_flags & VM_NONLINEAR)) 5901da177e4SLinus Torvalds vma_nonlinear_insert(vma, &mapping->i_mmap_nonlinear); 5911da177e4SLinus Torvalds else 5926b2dbba8SMichel Lespinasse vma_interval_tree_insert(vma, &mapping->i_mmap); 5931da177e4SLinus Torvalds flush_dcache_mmap_unlock(mapping); 5941da177e4SLinus Torvalds } 5951da177e4SLinus Torvalds } 5961da177e4SLinus Torvalds 5971da177e4SLinus Torvalds static void 5981da177e4SLinus Torvalds __vma_link(struct mm_struct *mm, struct vm_area_struct *vma, 5991da177e4SLinus Torvalds struct vm_area_struct *prev, struct rb_node **rb_link, 6001da177e4SLinus Torvalds struct rb_node *rb_parent) 6011da177e4SLinus Torvalds { 6021da177e4SLinus Torvalds __vma_link_list(mm, vma, prev, rb_parent); 6031da177e4SLinus Torvalds __vma_link_rb(mm, vma, rb_link, rb_parent); 6041da177e4SLinus Torvalds } 6051da177e4SLinus Torvalds 6061da177e4SLinus Torvalds static void vma_link(struct mm_struct *mm, struct vm_area_struct *vma, 6071da177e4SLinus Torvalds struct vm_area_struct *prev, struct rb_node **rb_link, 6081da177e4SLinus Torvalds struct rb_node *rb_parent) 6091da177e4SLinus Torvalds { 6101da177e4SLinus Torvalds struct address_space *mapping = NULL; 6111da177e4SLinus Torvalds 6121da177e4SLinus Torvalds if (vma->vm_file) 6131da177e4SLinus Torvalds mapping = vma->vm_file->f_mapping; 6141da177e4SLinus Torvalds 61597a89413SPeter Zijlstra if (mapping) 6163d48ae45SPeter Zijlstra mutex_lock(&mapping->i_mmap_mutex); 6171da177e4SLinus Torvalds 6181da177e4SLinus Torvalds __vma_link(mm, vma, prev, rb_link, rb_parent); 6191da177e4SLinus Torvalds __vma_link_file(vma); 6201da177e4SLinus Torvalds 6211da177e4SLinus Torvalds if (mapping) 6223d48ae45SPeter Zijlstra mutex_unlock(&mapping->i_mmap_mutex); 6231da177e4SLinus Torvalds 6241da177e4SLinus Torvalds mm->map_count++; 6251da177e4SLinus Torvalds validate_mm(mm); 6261da177e4SLinus Torvalds } 6271da177e4SLinus Torvalds 6281da177e4SLinus Torvalds /* 62988f6b4c3SKautuk Consul * Helper for vma_adjust() in the split_vma insert case: insert a vma into the 6306b2dbba8SMichel Lespinasse * mm's list and rbtree. It has already been inserted into the interval tree. 6311da177e4SLinus Torvalds */ 63248aae425SZhenwenXu static void __insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma) 6331da177e4SLinus Torvalds { 6346597d783SHugh Dickins struct vm_area_struct *prev; 6351da177e4SLinus Torvalds struct rb_node **rb_link, *rb_parent; 6361da177e4SLinus Torvalds 6376597d783SHugh Dickins if (find_vma_links(mm, vma->vm_start, vma->vm_end, 6386597d783SHugh Dickins &prev, &rb_link, &rb_parent)) 6396597d783SHugh Dickins BUG(); 6401da177e4SLinus Torvalds __vma_link(mm, vma, prev, rb_link, rb_parent); 6411da177e4SLinus Torvalds mm->map_count++; 6421da177e4SLinus Torvalds } 6431da177e4SLinus Torvalds 6441da177e4SLinus Torvalds static inline void 6451da177e4SLinus Torvalds __vma_unlink(struct mm_struct *mm, struct vm_area_struct *vma, 6461da177e4SLinus Torvalds struct vm_area_struct *prev) 6471da177e4SLinus Torvalds { 648d3737187SMichel Lespinasse struct vm_area_struct *next; 649297c5eeeSLinus Torvalds 650d3737187SMichel Lespinasse vma_rb_erase(vma, &mm->mm_rb); 651d3737187SMichel Lespinasse prev->vm_next = next = vma->vm_next; 652297c5eeeSLinus Torvalds if (next) 653297c5eeeSLinus Torvalds next->vm_prev = prev; 6541da177e4SLinus Torvalds if (mm->mmap_cache == vma) 6551da177e4SLinus Torvalds mm->mmap_cache = prev; 6561da177e4SLinus Torvalds } 6571da177e4SLinus Torvalds 6581da177e4SLinus Torvalds /* 6591da177e4SLinus Torvalds * We cannot adjust vm_start, vm_end, vm_pgoff fields of a vma that 6601da177e4SLinus Torvalds * is already present in an i_mmap tree without adjusting the tree. 6611da177e4SLinus Torvalds * The following helper function should be used when such adjustments 6621da177e4SLinus Torvalds * are necessary. The "insert" vma (if any) is to be inserted 6631da177e4SLinus Torvalds * before we drop the necessary locks. 6641da177e4SLinus Torvalds */ 6655beb4930SRik van Riel int vma_adjust(struct vm_area_struct *vma, unsigned long start, 6661da177e4SLinus Torvalds unsigned long end, pgoff_t pgoff, struct vm_area_struct *insert) 6671da177e4SLinus Torvalds { 6681da177e4SLinus Torvalds struct mm_struct *mm = vma->vm_mm; 6691da177e4SLinus Torvalds struct vm_area_struct *next = vma->vm_next; 6701da177e4SLinus Torvalds struct vm_area_struct *importer = NULL; 6711da177e4SLinus Torvalds struct address_space *mapping = NULL; 6726b2dbba8SMichel Lespinasse struct rb_root *root = NULL; 673012f1800SRik van Riel struct anon_vma *anon_vma = NULL; 6741da177e4SLinus Torvalds struct file *file = vma->vm_file; 675d3737187SMichel Lespinasse bool start_changed = false, end_changed = false; 6761da177e4SLinus Torvalds long adjust_next = 0; 6771da177e4SLinus Torvalds int remove_next = 0; 6781da177e4SLinus Torvalds 6791da177e4SLinus Torvalds if (next && !insert) { 680287d97acSLinus Torvalds struct vm_area_struct *exporter = NULL; 681287d97acSLinus Torvalds 6821da177e4SLinus Torvalds if (end >= next->vm_end) { 6831da177e4SLinus Torvalds /* 6841da177e4SLinus Torvalds * vma expands, overlapping all the next, and 6851da177e4SLinus Torvalds * perhaps the one after too (mprotect case 6). 6861da177e4SLinus Torvalds */ 6871da177e4SLinus Torvalds again: remove_next = 1 + (end > next->vm_end); 6881da177e4SLinus Torvalds end = next->vm_end; 689287d97acSLinus Torvalds exporter = next; 6901da177e4SLinus Torvalds importer = vma; 6911da177e4SLinus Torvalds } else if (end > next->vm_start) { 6921da177e4SLinus Torvalds /* 6931da177e4SLinus Torvalds * vma expands, overlapping part of the next: 6941da177e4SLinus Torvalds * mprotect case 5 shifting the boundary up. 6951da177e4SLinus Torvalds */ 6961da177e4SLinus Torvalds adjust_next = (end - next->vm_start) >> PAGE_SHIFT; 697287d97acSLinus Torvalds exporter = next; 6981da177e4SLinus Torvalds importer = vma; 6991da177e4SLinus Torvalds } else if (end < vma->vm_end) { 7001da177e4SLinus Torvalds /* 7011da177e4SLinus Torvalds * vma shrinks, and !insert tells it's not 7021da177e4SLinus Torvalds * split_vma inserting another: so it must be 7031da177e4SLinus Torvalds * mprotect case 4 shifting the boundary down. 7041da177e4SLinus Torvalds */ 7051da177e4SLinus Torvalds adjust_next = - ((vma->vm_end - end) >> PAGE_SHIFT); 706287d97acSLinus Torvalds exporter = vma; 7071da177e4SLinus Torvalds importer = next; 7081da177e4SLinus Torvalds } 7091da177e4SLinus Torvalds 7105beb4930SRik van Riel /* 7115beb4930SRik van Riel * Easily overlooked: when mprotect shifts the boundary, 7125beb4930SRik van Riel * make sure the expanding vma has anon_vma set if the 7135beb4930SRik van Riel * shrinking vma had, to cover any anon pages imported. 7145beb4930SRik van Riel */ 715287d97acSLinus Torvalds if (exporter && exporter->anon_vma && !importer->anon_vma) { 716287d97acSLinus Torvalds if (anon_vma_clone(importer, exporter)) 7175beb4930SRik van Riel return -ENOMEM; 718287d97acSLinus Torvalds importer->anon_vma = exporter->anon_vma; 7195beb4930SRik van Riel } 7205beb4930SRik van Riel } 7215beb4930SRik van Riel 7221da177e4SLinus Torvalds if (file) { 7231da177e4SLinus Torvalds mapping = file->f_mapping; 724682968e0SSrikar Dronamraju if (!(vma->vm_flags & VM_NONLINEAR)) { 7251da177e4SLinus Torvalds root = &mapping->i_mmap; 726cbc91f71SSrikar Dronamraju uprobe_munmap(vma, vma->vm_start, vma->vm_end); 727682968e0SSrikar Dronamraju 728682968e0SSrikar Dronamraju if (adjust_next) 729cbc91f71SSrikar Dronamraju uprobe_munmap(next, next->vm_start, 730cbc91f71SSrikar Dronamraju next->vm_end); 731682968e0SSrikar Dronamraju } 732682968e0SSrikar Dronamraju 7333d48ae45SPeter Zijlstra mutex_lock(&mapping->i_mmap_mutex); 7341da177e4SLinus Torvalds if (insert) { 7351da177e4SLinus Torvalds /* 7366b2dbba8SMichel Lespinasse * Put into interval tree now, so instantiated pages 7371da177e4SLinus Torvalds * are visible to arm/parisc __flush_dcache_page 7381da177e4SLinus Torvalds * throughout; but we cannot insert into address 7391da177e4SLinus Torvalds * space until vma start or end is updated. 7401da177e4SLinus Torvalds */ 7411da177e4SLinus Torvalds __vma_link_file(insert); 7421da177e4SLinus Torvalds } 7431da177e4SLinus Torvalds } 7441da177e4SLinus Torvalds 74594fcc585SAndrea Arcangeli vma_adjust_trans_huge(vma, start, end, adjust_next); 74694fcc585SAndrea Arcangeli 747012f1800SRik van Riel anon_vma = vma->anon_vma; 748bf181b9fSMichel Lespinasse if (!anon_vma && adjust_next) 749bf181b9fSMichel Lespinasse anon_vma = next->anon_vma; 750bf181b9fSMichel Lespinasse if (anon_vma) { 751ca42b26aSMichel Lespinasse VM_BUG_ON(adjust_next && next->anon_vma && 752ca42b26aSMichel Lespinasse anon_vma != next->anon_vma); 7534fc3f1d6SIngo Molnar anon_vma_lock_write(anon_vma); 754bf181b9fSMichel Lespinasse anon_vma_interval_tree_pre_update_vma(vma); 755bf181b9fSMichel Lespinasse if (adjust_next) 756bf181b9fSMichel Lespinasse anon_vma_interval_tree_pre_update_vma(next); 757bf181b9fSMichel Lespinasse } 758012f1800SRik van Riel 7591da177e4SLinus Torvalds if (root) { 7601da177e4SLinus Torvalds flush_dcache_mmap_lock(mapping); 7616b2dbba8SMichel Lespinasse vma_interval_tree_remove(vma, root); 7621da177e4SLinus Torvalds if (adjust_next) 7636b2dbba8SMichel Lespinasse vma_interval_tree_remove(next, root); 7641da177e4SLinus Torvalds } 7651da177e4SLinus Torvalds 766d3737187SMichel Lespinasse if (start != vma->vm_start) { 7671da177e4SLinus Torvalds vma->vm_start = start; 768d3737187SMichel Lespinasse start_changed = true; 769d3737187SMichel Lespinasse } 770d3737187SMichel Lespinasse if (end != vma->vm_end) { 7711da177e4SLinus Torvalds vma->vm_end = end; 772d3737187SMichel Lespinasse end_changed = true; 773d3737187SMichel Lespinasse } 7741da177e4SLinus Torvalds vma->vm_pgoff = pgoff; 7751da177e4SLinus Torvalds if (adjust_next) { 7761da177e4SLinus Torvalds next->vm_start += adjust_next << PAGE_SHIFT; 7771da177e4SLinus Torvalds next->vm_pgoff += adjust_next; 7781da177e4SLinus Torvalds } 7791da177e4SLinus Torvalds 7801da177e4SLinus Torvalds if (root) { 7811da177e4SLinus Torvalds if (adjust_next) 7826b2dbba8SMichel Lespinasse vma_interval_tree_insert(next, root); 7836b2dbba8SMichel Lespinasse vma_interval_tree_insert(vma, root); 7841da177e4SLinus Torvalds flush_dcache_mmap_unlock(mapping); 7851da177e4SLinus Torvalds } 7861da177e4SLinus Torvalds 7871da177e4SLinus Torvalds if (remove_next) { 7881da177e4SLinus Torvalds /* 7891da177e4SLinus Torvalds * vma_merge has merged next into vma, and needs 7901da177e4SLinus Torvalds * us to remove next before dropping the locks. 7911da177e4SLinus Torvalds */ 7921da177e4SLinus Torvalds __vma_unlink(mm, next, vma); 7931da177e4SLinus Torvalds if (file) 7941da177e4SLinus Torvalds __remove_shared_vm_struct(next, file, mapping); 7951da177e4SLinus Torvalds } else if (insert) { 7961da177e4SLinus Torvalds /* 7971da177e4SLinus Torvalds * split_vma has split insert from vma, and needs 7981da177e4SLinus Torvalds * us to insert it before dropping the locks 7991da177e4SLinus Torvalds * (it may either follow vma or precede it). 8001da177e4SLinus Torvalds */ 8011da177e4SLinus Torvalds __insert_vm_struct(mm, insert); 802d3737187SMichel Lespinasse } else { 803d3737187SMichel Lespinasse if (start_changed) 804d3737187SMichel Lespinasse vma_gap_update(vma); 805d3737187SMichel Lespinasse if (end_changed) { 806d3737187SMichel Lespinasse if (!next) 807d3737187SMichel Lespinasse mm->highest_vm_end = end; 808d3737187SMichel Lespinasse else if (!adjust_next) 809d3737187SMichel Lespinasse vma_gap_update(next); 810d3737187SMichel Lespinasse } 8111da177e4SLinus Torvalds } 8121da177e4SLinus Torvalds 813bf181b9fSMichel Lespinasse if (anon_vma) { 814bf181b9fSMichel Lespinasse anon_vma_interval_tree_post_update_vma(vma); 815bf181b9fSMichel Lespinasse if (adjust_next) 816bf181b9fSMichel Lespinasse anon_vma_interval_tree_post_update_vma(next); 81708b52706SKonstantin Khlebnikov anon_vma_unlock_write(anon_vma); 818bf181b9fSMichel Lespinasse } 8191da177e4SLinus Torvalds if (mapping) 8203d48ae45SPeter Zijlstra mutex_unlock(&mapping->i_mmap_mutex); 8211da177e4SLinus Torvalds 8222b144498SSrikar Dronamraju if (root) { 8237b2d81d4SIngo Molnar uprobe_mmap(vma); 8242b144498SSrikar Dronamraju 8252b144498SSrikar Dronamraju if (adjust_next) 8267b2d81d4SIngo Molnar uprobe_mmap(next); 8272b144498SSrikar Dronamraju } 8282b144498SSrikar Dronamraju 8291da177e4SLinus Torvalds if (remove_next) { 830925d1c40SMatt Helsley if (file) { 831cbc91f71SSrikar Dronamraju uprobe_munmap(next, next->vm_start, next->vm_end); 8321da177e4SLinus Torvalds fput(file); 833925d1c40SMatt Helsley } 8345beb4930SRik van Riel if (next->anon_vma) 8355beb4930SRik van Riel anon_vma_merge(vma, next); 8361da177e4SLinus Torvalds mm->map_count--; 8371444f92cSHampson, Steven T vma_set_policy(vma, vma_policy(next)); 8381da177e4SLinus Torvalds kmem_cache_free(vm_area_cachep, next); 8391da177e4SLinus Torvalds /* 8401da177e4SLinus Torvalds * In mprotect's case 6 (see comments on vma_merge), 8411da177e4SLinus Torvalds * we must remove another next too. It would clutter 8421da177e4SLinus Torvalds * up the code too much to do both in one go. 8431da177e4SLinus Torvalds */ 8441da177e4SLinus Torvalds next = vma->vm_next; 845d3737187SMichel Lespinasse if (remove_next == 2) 8461da177e4SLinus Torvalds goto again; 847d3737187SMichel Lespinasse else if (next) 848d3737187SMichel Lespinasse vma_gap_update(next); 849d3737187SMichel Lespinasse else 850d3737187SMichel Lespinasse mm->highest_vm_end = end; 8511da177e4SLinus Torvalds } 8522b144498SSrikar Dronamraju if (insert && file) 8537b2d81d4SIngo Molnar uprobe_mmap(insert); 8541da177e4SLinus Torvalds 8551da177e4SLinus Torvalds validate_mm(mm); 8565beb4930SRik van Riel 8575beb4930SRik van Riel return 0; 8581da177e4SLinus Torvalds } 8591da177e4SLinus Torvalds 8601da177e4SLinus Torvalds /* 8611da177e4SLinus Torvalds * If the vma has a ->close operation then the driver probably needs to release 8621da177e4SLinus Torvalds * per-vma resources, so we don't attempt to merge those. 8631da177e4SLinus Torvalds */ 8641da177e4SLinus Torvalds static inline int is_mergeable_vma(struct vm_area_struct *vma, 8651da177e4SLinus Torvalds struct file *file, unsigned long vm_flags) 8661da177e4SLinus Torvalds { 8670b173bc4SKonstantin Khlebnikov if (vma->vm_flags ^ vm_flags) 8681da177e4SLinus Torvalds return 0; 8691da177e4SLinus Torvalds if (vma->vm_file != file) 8701da177e4SLinus Torvalds return 0; 8711da177e4SLinus Torvalds if (vma->vm_ops && vma->vm_ops->close) 8721da177e4SLinus Torvalds return 0; 8731da177e4SLinus Torvalds return 1; 8741da177e4SLinus Torvalds } 8751da177e4SLinus Torvalds 8761da177e4SLinus Torvalds static inline int is_mergeable_anon_vma(struct anon_vma *anon_vma1, 877965f55deSShaohua Li struct anon_vma *anon_vma2, 878965f55deSShaohua Li struct vm_area_struct *vma) 8791da177e4SLinus Torvalds { 880965f55deSShaohua Li /* 881965f55deSShaohua Li * The list_is_singular() test is to avoid merging VMA cloned from 882965f55deSShaohua Li * parents. This can improve scalability caused by anon_vma lock. 883965f55deSShaohua Li */ 884965f55deSShaohua Li if ((!anon_vma1 || !anon_vma2) && (!vma || 885965f55deSShaohua Li list_is_singular(&vma->anon_vma_chain))) 886965f55deSShaohua Li return 1; 887965f55deSShaohua Li return anon_vma1 == anon_vma2; 8881da177e4SLinus Torvalds } 8891da177e4SLinus Torvalds 8901da177e4SLinus Torvalds /* 8911da177e4SLinus Torvalds * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff) 8921da177e4SLinus Torvalds * in front of (at a lower virtual address and file offset than) the vma. 8931da177e4SLinus Torvalds * 8941da177e4SLinus Torvalds * We cannot merge two vmas if they have differently assigned (non-NULL) 8951da177e4SLinus Torvalds * anon_vmas, nor if same anon_vma is assigned but offsets incompatible. 8961da177e4SLinus Torvalds * 8971da177e4SLinus Torvalds * We don't check here for the merged mmap wrapping around the end of pagecache 8981da177e4SLinus Torvalds * indices (16TB on ia32) because do_mmap_pgoff() does not permit mmap's which 8991da177e4SLinus Torvalds * wrap, nor mmaps which cover the final page at index -1UL. 9001da177e4SLinus Torvalds */ 9011da177e4SLinus Torvalds static int 9021da177e4SLinus Torvalds can_vma_merge_before(struct vm_area_struct *vma, unsigned long vm_flags, 9031da177e4SLinus Torvalds struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff) 9041da177e4SLinus Torvalds { 9051da177e4SLinus Torvalds if (is_mergeable_vma(vma, file, vm_flags) && 906965f55deSShaohua Li is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) { 9071da177e4SLinus Torvalds if (vma->vm_pgoff == vm_pgoff) 9081da177e4SLinus Torvalds return 1; 9091da177e4SLinus Torvalds } 9101da177e4SLinus Torvalds return 0; 9111da177e4SLinus Torvalds } 9121da177e4SLinus Torvalds 9131da177e4SLinus Torvalds /* 9141da177e4SLinus Torvalds * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff) 9151da177e4SLinus Torvalds * beyond (at a higher virtual address and file offset than) the vma. 9161da177e4SLinus Torvalds * 9171da177e4SLinus Torvalds * We cannot merge two vmas if they have differently assigned (non-NULL) 9181da177e4SLinus Torvalds * anon_vmas, nor if same anon_vma is assigned but offsets incompatible. 9191da177e4SLinus Torvalds */ 9201da177e4SLinus Torvalds static int 9211da177e4SLinus Torvalds can_vma_merge_after(struct vm_area_struct *vma, unsigned long vm_flags, 9221da177e4SLinus Torvalds struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff) 9231da177e4SLinus Torvalds { 9241da177e4SLinus Torvalds if (is_mergeable_vma(vma, file, vm_flags) && 925965f55deSShaohua Li is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) { 9261da177e4SLinus Torvalds pgoff_t vm_pglen; 9271da177e4SLinus Torvalds vm_pglen = (vma->vm_end - vma->vm_start) >> PAGE_SHIFT; 9281da177e4SLinus Torvalds if (vma->vm_pgoff + vm_pglen == vm_pgoff) 9291da177e4SLinus Torvalds return 1; 9301da177e4SLinus Torvalds } 9311da177e4SLinus Torvalds return 0; 9321da177e4SLinus Torvalds } 9331da177e4SLinus Torvalds 9341da177e4SLinus Torvalds /* 9351da177e4SLinus Torvalds * Given a mapping request (addr,end,vm_flags,file,pgoff), figure out 9361da177e4SLinus Torvalds * whether that can be merged with its predecessor or its successor. 9371da177e4SLinus Torvalds * Or both (it neatly fills a hole). 9381da177e4SLinus Torvalds * 9391da177e4SLinus Torvalds * In most cases - when called for mmap, brk or mremap - [addr,end) is 9401da177e4SLinus Torvalds * certain not to be mapped by the time vma_merge is called; but when 9411da177e4SLinus Torvalds * called for mprotect, it is certain to be already mapped (either at 9421da177e4SLinus Torvalds * an offset within prev, or at the start of next), and the flags of 9431da177e4SLinus Torvalds * this area are about to be changed to vm_flags - and the no-change 9441da177e4SLinus Torvalds * case has already been eliminated. 9451da177e4SLinus Torvalds * 9461da177e4SLinus Torvalds * The following mprotect cases have to be considered, where AAAA is 9471da177e4SLinus Torvalds * the area passed down from mprotect_fixup, never extending beyond one 9481da177e4SLinus Torvalds * vma, PPPPPP is the prev vma specified, and NNNNNN the next vma after: 9491da177e4SLinus Torvalds * 9501da177e4SLinus Torvalds * AAAA AAAA AAAA AAAA 9511da177e4SLinus Torvalds * PPPPPPNNNNNN PPPPPPNNNNNN PPPPPPNNNNNN PPPPNNNNXXXX 9521da177e4SLinus Torvalds * cannot merge might become might become might become 9531da177e4SLinus Torvalds * PPNNNNNNNNNN PPPPPPPPPPNN PPPPPPPPPPPP 6 or 9541da177e4SLinus Torvalds * mmap, brk or case 4 below case 5 below PPPPPPPPXXXX 7 or 9551da177e4SLinus Torvalds * mremap move: PPPPNNNNNNNN 8 9561da177e4SLinus Torvalds * AAAA 9571da177e4SLinus Torvalds * PPPP NNNN PPPPPPPPPPPP PPPPPPPPNNNN PPPPNNNNNNNN 9581da177e4SLinus Torvalds * might become case 1 below case 2 below case 3 below 9591da177e4SLinus Torvalds * 9601da177e4SLinus Torvalds * Odd one out? Case 8, because it extends NNNN but needs flags of XXXX: 9611da177e4SLinus Torvalds * mprotect_fixup updates vm_flags & vm_page_prot on successful return. 9621da177e4SLinus Torvalds */ 9631da177e4SLinus Torvalds struct vm_area_struct *vma_merge(struct mm_struct *mm, 9641da177e4SLinus Torvalds struct vm_area_struct *prev, unsigned long addr, 9651da177e4SLinus Torvalds unsigned long end, unsigned long vm_flags, 9661da177e4SLinus Torvalds struct anon_vma *anon_vma, struct file *file, 9671da177e4SLinus Torvalds pgoff_t pgoff, struct mempolicy *policy) 9681da177e4SLinus Torvalds { 9691da177e4SLinus Torvalds pgoff_t pglen = (end - addr) >> PAGE_SHIFT; 9701da177e4SLinus Torvalds struct vm_area_struct *area, *next; 9715beb4930SRik van Riel int err; 9721da177e4SLinus Torvalds 9731da177e4SLinus Torvalds /* 9741da177e4SLinus Torvalds * We later require that vma->vm_flags == vm_flags, 9751da177e4SLinus Torvalds * so this tests vma->vm_flags & VM_SPECIAL, too. 9761da177e4SLinus Torvalds */ 9771da177e4SLinus Torvalds if (vm_flags & VM_SPECIAL) 9781da177e4SLinus Torvalds return NULL; 9791da177e4SLinus Torvalds 9801da177e4SLinus Torvalds if (prev) 9811da177e4SLinus Torvalds next = prev->vm_next; 9821da177e4SLinus Torvalds else 9831da177e4SLinus Torvalds next = mm->mmap; 9841da177e4SLinus Torvalds area = next; 9851da177e4SLinus Torvalds if (next && next->vm_end == end) /* cases 6, 7, 8 */ 9861da177e4SLinus Torvalds next = next->vm_next; 9871da177e4SLinus Torvalds 9881da177e4SLinus Torvalds /* 9891da177e4SLinus Torvalds * Can it merge with the predecessor? 9901da177e4SLinus Torvalds */ 9911da177e4SLinus Torvalds if (prev && prev->vm_end == addr && 9921da177e4SLinus Torvalds mpol_equal(vma_policy(prev), policy) && 9931da177e4SLinus Torvalds can_vma_merge_after(prev, vm_flags, 9941da177e4SLinus Torvalds anon_vma, file, pgoff)) { 9951da177e4SLinus Torvalds /* 9961da177e4SLinus Torvalds * OK, it can. Can we now merge in the successor as well? 9971da177e4SLinus Torvalds */ 9981da177e4SLinus Torvalds if (next && end == next->vm_start && 9991da177e4SLinus Torvalds mpol_equal(policy, vma_policy(next)) && 10001da177e4SLinus Torvalds can_vma_merge_before(next, vm_flags, 10011da177e4SLinus Torvalds anon_vma, file, pgoff+pglen) && 10021da177e4SLinus Torvalds is_mergeable_anon_vma(prev->anon_vma, 1003965f55deSShaohua Li next->anon_vma, NULL)) { 10041da177e4SLinus Torvalds /* cases 1, 6 */ 10055beb4930SRik van Riel err = vma_adjust(prev, prev->vm_start, 10061da177e4SLinus Torvalds next->vm_end, prev->vm_pgoff, NULL); 10071da177e4SLinus Torvalds } else /* cases 2, 5, 7 */ 10085beb4930SRik van Riel err = vma_adjust(prev, prev->vm_start, 10091da177e4SLinus Torvalds end, prev->vm_pgoff, NULL); 10105beb4930SRik van Riel if (err) 10115beb4930SRik van Riel return NULL; 1012b15d00b6SAndrea Arcangeli khugepaged_enter_vma_merge(prev); 10131da177e4SLinus Torvalds return prev; 10141da177e4SLinus Torvalds } 10151da177e4SLinus Torvalds 10161da177e4SLinus Torvalds /* 10171da177e4SLinus Torvalds * Can this new request be merged in front of next? 10181da177e4SLinus Torvalds */ 10191da177e4SLinus Torvalds if (next && end == next->vm_start && 10201da177e4SLinus Torvalds mpol_equal(policy, vma_policy(next)) && 10211da177e4SLinus Torvalds can_vma_merge_before(next, vm_flags, 10221da177e4SLinus Torvalds anon_vma, file, pgoff+pglen)) { 10231da177e4SLinus Torvalds if (prev && addr < prev->vm_end) /* case 4 */ 10245beb4930SRik van Riel err = vma_adjust(prev, prev->vm_start, 10251da177e4SLinus Torvalds addr, prev->vm_pgoff, NULL); 10261da177e4SLinus Torvalds else /* cases 3, 8 */ 10275beb4930SRik van Riel err = vma_adjust(area, addr, next->vm_end, 10281da177e4SLinus Torvalds next->vm_pgoff - pglen, NULL); 10295beb4930SRik van Riel if (err) 10305beb4930SRik van Riel return NULL; 1031b15d00b6SAndrea Arcangeli khugepaged_enter_vma_merge(area); 10321da177e4SLinus Torvalds return area; 10331da177e4SLinus Torvalds } 10341da177e4SLinus Torvalds 10351da177e4SLinus Torvalds return NULL; 10361da177e4SLinus Torvalds } 10371da177e4SLinus Torvalds 10381da177e4SLinus Torvalds /* 1039d0e9fe17SLinus Torvalds * Rough compatbility check to quickly see if it's even worth looking 1040d0e9fe17SLinus Torvalds * at sharing an anon_vma. 1041d0e9fe17SLinus Torvalds * 1042d0e9fe17SLinus Torvalds * They need to have the same vm_file, and the flags can only differ 1043d0e9fe17SLinus Torvalds * in things that mprotect may change. 1044d0e9fe17SLinus Torvalds * 1045d0e9fe17SLinus Torvalds * NOTE! The fact that we share an anon_vma doesn't _have_ to mean that 1046d0e9fe17SLinus Torvalds * we can merge the two vma's. For example, we refuse to merge a vma if 1047d0e9fe17SLinus Torvalds * there is a vm_ops->close() function, because that indicates that the 1048d0e9fe17SLinus Torvalds * driver is doing some kind of reference counting. But that doesn't 1049d0e9fe17SLinus Torvalds * really matter for the anon_vma sharing case. 1050d0e9fe17SLinus Torvalds */ 1051d0e9fe17SLinus Torvalds static int anon_vma_compatible(struct vm_area_struct *a, struct vm_area_struct *b) 1052d0e9fe17SLinus Torvalds { 1053d0e9fe17SLinus Torvalds return a->vm_end == b->vm_start && 1054d0e9fe17SLinus Torvalds mpol_equal(vma_policy(a), vma_policy(b)) && 1055d0e9fe17SLinus Torvalds a->vm_file == b->vm_file && 1056d0e9fe17SLinus Torvalds !((a->vm_flags ^ b->vm_flags) & ~(VM_READ|VM_WRITE|VM_EXEC)) && 1057d0e9fe17SLinus Torvalds b->vm_pgoff == a->vm_pgoff + ((b->vm_start - a->vm_start) >> PAGE_SHIFT); 1058d0e9fe17SLinus Torvalds } 1059d0e9fe17SLinus Torvalds 1060d0e9fe17SLinus Torvalds /* 1061d0e9fe17SLinus Torvalds * Do some basic sanity checking to see if we can re-use the anon_vma 1062d0e9fe17SLinus Torvalds * from 'old'. The 'a'/'b' vma's are in VM order - one of them will be 1063d0e9fe17SLinus Torvalds * the same as 'old', the other will be the new one that is trying 1064d0e9fe17SLinus Torvalds * to share the anon_vma. 1065d0e9fe17SLinus Torvalds * 1066d0e9fe17SLinus Torvalds * NOTE! This runs with mm_sem held for reading, so it is possible that 1067d0e9fe17SLinus Torvalds * the anon_vma of 'old' is concurrently in the process of being set up 1068d0e9fe17SLinus Torvalds * by another page fault trying to merge _that_. But that's ok: if it 1069d0e9fe17SLinus Torvalds * is being set up, that automatically means that it will be a singleton 1070d0e9fe17SLinus Torvalds * acceptable for merging, so we can do all of this optimistically. But 1071d0e9fe17SLinus Torvalds * we do that ACCESS_ONCE() to make sure that we never re-load the pointer. 1072d0e9fe17SLinus Torvalds * 1073d0e9fe17SLinus Torvalds * IOW: that the "list_is_singular()" test on the anon_vma_chain only 1074d0e9fe17SLinus Torvalds * matters for the 'stable anon_vma' case (ie the thing we want to avoid 1075d0e9fe17SLinus Torvalds * is to return an anon_vma that is "complex" due to having gone through 1076d0e9fe17SLinus Torvalds * a fork). 1077d0e9fe17SLinus Torvalds * 1078d0e9fe17SLinus Torvalds * We also make sure that the two vma's are compatible (adjacent, 1079d0e9fe17SLinus Torvalds * and with the same memory policies). That's all stable, even with just 1080d0e9fe17SLinus Torvalds * a read lock on the mm_sem. 1081d0e9fe17SLinus Torvalds */ 1082d0e9fe17SLinus Torvalds static struct anon_vma *reusable_anon_vma(struct vm_area_struct *old, struct vm_area_struct *a, struct vm_area_struct *b) 1083d0e9fe17SLinus Torvalds { 1084d0e9fe17SLinus Torvalds if (anon_vma_compatible(a, b)) { 1085d0e9fe17SLinus Torvalds struct anon_vma *anon_vma = ACCESS_ONCE(old->anon_vma); 1086d0e9fe17SLinus Torvalds 1087d0e9fe17SLinus Torvalds if (anon_vma && list_is_singular(&old->anon_vma_chain)) 1088d0e9fe17SLinus Torvalds return anon_vma; 1089d0e9fe17SLinus Torvalds } 1090d0e9fe17SLinus Torvalds return NULL; 1091d0e9fe17SLinus Torvalds } 1092d0e9fe17SLinus Torvalds 1093d0e9fe17SLinus Torvalds /* 10941da177e4SLinus Torvalds * find_mergeable_anon_vma is used by anon_vma_prepare, to check 10951da177e4SLinus Torvalds * neighbouring vmas for a suitable anon_vma, before it goes off 10961da177e4SLinus Torvalds * to allocate a new anon_vma. It checks because a repetitive 10971da177e4SLinus Torvalds * sequence of mprotects and faults may otherwise lead to distinct 10981da177e4SLinus Torvalds * anon_vmas being allocated, preventing vma merge in subsequent 10991da177e4SLinus Torvalds * mprotect. 11001da177e4SLinus Torvalds */ 11011da177e4SLinus Torvalds struct anon_vma *find_mergeable_anon_vma(struct vm_area_struct *vma) 11021da177e4SLinus Torvalds { 1103d0e9fe17SLinus Torvalds struct anon_vma *anon_vma; 11041da177e4SLinus Torvalds struct vm_area_struct *near; 11051da177e4SLinus Torvalds 11061da177e4SLinus Torvalds near = vma->vm_next; 11071da177e4SLinus Torvalds if (!near) 11081da177e4SLinus Torvalds goto try_prev; 11091da177e4SLinus Torvalds 1110d0e9fe17SLinus Torvalds anon_vma = reusable_anon_vma(near, vma, near); 1111d0e9fe17SLinus Torvalds if (anon_vma) 1112d0e9fe17SLinus Torvalds return anon_vma; 11131da177e4SLinus Torvalds try_prev: 11149be34c9dSLinus Torvalds near = vma->vm_prev; 11151da177e4SLinus Torvalds if (!near) 11161da177e4SLinus Torvalds goto none; 11171da177e4SLinus Torvalds 1118d0e9fe17SLinus Torvalds anon_vma = reusable_anon_vma(near, near, vma); 1119d0e9fe17SLinus Torvalds if (anon_vma) 1120d0e9fe17SLinus Torvalds return anon_vma; 11211da177e4SLinus Torvalds none: 11221da177e4SLinus Torvalds /* 11231da177e4SLinus Torvalds * There's no absolute need to look only at touching neighbours: 11241da177e4SLinus Torvalds * we could search further afield for "compatible" anon_vmas. 11251da177e4SLinus Torvalds * But it would probably just be a waste of time searching, 11261da177e4SLinus Torvalds * or lead to too many vmas hanging off the same anon_vma. 11271da177e4SLinus Torvalds * We're trying to allow mprotect remerging later on, 11281da177e4SLinus Torvalds * not trying to minimize memory used for anon_vmas. 11291da177e4SLinus Torvalds */ 11301da177e4SLinus Torvalds return NULL; 11311da177e4SLinus Torvalds } 11321da177e4SLinus Torvalds 11331da177e4SLinus Torvalds #ifdef CONFIG_PROC_FS 1134ab50b8edSHugh Dickins void vm_stat_account(struct mm_struct *mm, unsigned long flags, 11351da177e4SLinus Torvalds struct file *file, long pages) 11361da177e4SLinus Torvalds { 11371da177e4SLinus Torvalds const unsigned long stack_flags 11381da177e4SLinus Torvalds = VM_STACK_FLAGS & (VM_GROWSUP|VM_GROWSDOWN); 11391da177e4SLinus Torvalds 114044de9d0cSHuang Shijie mm->total_vm += pages; 114144de9d0cSHuang Shijie 11421da177e4SLinus Torvalds if (file) { 11431da177e4SLinus Torvalds mm->shared_vm += pages; 11441da177e4SLinus Torvalds if ((flags & (VM_EXEC|VM_WRITE)) == VM_EXEC) 11451da177e4SLinus Torvalds mm->exec_vm += pages; 11461da177e4SLinus Torvalds } else if (flags & stack_flags) 11471da177e4SLinus Torvalds mm->stack_vm += pages; 11481da177e4SLinus Torvalds } 11491da177e4SLinus Torvalds #endif /* CONFIG_PROC_FS */ 11501da177e4SLinus Torvalds 11511da177e4SLinus Torvalds /* 115240401530SAl Viro * If a hint addr is less than mmap_min_addr change hint to be as 115340401530SAl Viro * low as possible but still greater than mmap_min_addr 115440401530SAl Viro */ 115540401530SAl Viro static inline unsigned long round_hint_to_min(unsigned long hint) 115640401530SAl Viro { 115740401530SAl Viro hint &= PAGE_MASK; 115840401530SAl Viro if (((void *)hint != NULL) && 115940401530SAl Viro (hint < mmap_min_addr)) 116040401530SAl Viro return PAGE_ALIGN(mmap_min_addr); 116140401530SAl Viro return hint; 116240401530SAl Viro } 116340401530SAl Viro 116440401530SAl Viro /* 116527f5de79SJianjun Kong * The caller must hold down_write(¤t->mm->mmap_sem). 11661da177e4SLinus Torvalds */ 11671da177e4SLinus Torvalds 1168e3fc629dSAl Viro unsigned long do_mmap_pgoff(struct file *file, unsigned long addr, 11691da177e4SLinus Torvalds unsigned long len, unsigned long prot, 1170bebeb3d6SMichel Lespinasse unsigned long flags, unsigned long pgoff, 117141badc15SMichel Lespinasse unsigned long *populate) 11721da177e4SLinus Torvalds { 11731da177e4SLinus Torvalds struct mm_struct * mm = current->mm; 11741da177e4SLinus Torvalds struct inode *inode; 1175ca16d140SKOSAKI Motohiro vm_flags_t vm_flags; 11761da177e4SLinus Torvalds 117741badc15SMichel Lespinasse *populate = 0; 1178bebeb3d6SMichel Lespinasse 11791da177e4SLinus Torvalds /* 11801da177e4SLinus Torvalds * Does the application expect PROT_READ to imply PROT_EXEC? 11811da177e4SLinus Torvalds * 11821da177e4SLinus Torvalds * (the exception is when the underlying filesystem is noexec 11831da177e4SLinus Torvalds * mounted, in which case we dont add PROT_EXEC.) 11841da177e4SLinus Torvalds */ 11851da177e4SLinus Torvalds if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC)) 1186d3ac7f89SJosef "Jeff" Sipek if (!(file && (file->f_path.mnt->mnt_flags & MNT_NOEXEC))) 11871da177e4SLinus Torvalds prot |= PROT_EXEC; 11881da177e4SLinus Torvalds 11891da177e4SLinus Torvalds if (!len) 11901da177e4SLinus Torvalds return -EINVAL; 11911da177e4SLinus Torvalds 11927cd94146SEric Paris if (!(flags & MAP_FIXED)) 11937cd94146SEric Paris addr = round_hint_to_min(addr); 11947cd94146SEric Paris 11951da177e4SLinus Torvalds /* Careful about overflows.. */ 11961da177e4SLinus Torvalds len = PAGE_ALIGN(len); 11979206de95SAl Viro if (!len) 11981da177e4SLinus Torvalds return -ENOMEM; 11991da177e4SLinus Torvalds 12001da177e4SLinus Torvalds /* offset overflow? */ 12011da177e4SLinus Torvalds if ((pgoff + (len >> PAGE_SHIFT)) < pgoff) 12021da177e4SLinus Torvalds return -EOVERFLOW; 12031da177e4SLinus Torvalds 12041da177e4SLinus Torvalds /* Too many mappings? */ 12051da177e4SLinus Torvalds if (mm->map_count > sysctl_max_map_count) 12061da177e4SLinus Torvalds return -ENOMEM; 12071da177e4SLinus Torvalds 12081da177e4SLinus Torvalds /* Obtain the address to map to. we verify (or select) it and ensure 12091da177e4SLinus Torvalds * that it represents a valid section of the address space. 12101da177e4SLinus Torvalds */ 12111da177e4SLinus Torvalds addr = get_unmapped_area(file, addr, len, pgoff, flags); 12121da177e4SLinus Torvalds if (addr & ~PAGE_MASK) 12131da177e4SLinus Torvalds return addr; 12141da177e4SLinus Torvalds 12151da177e4SLinus Torvalds /* Do simple checking here so the lower-level routines won't have 12161da177e4SLinus Torvalds * to. we assume access permissions have been handled by the open 12171da177e4SLinus Torvalds * of the memory object, so we don't do any here. 12181da177e4SLinus Torvalds */ 12191da177e4SLinus Torvalds vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags) | 12201da177e4SLinus Torvalds mm->def_flags | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC; 12211da177e4SLinus Torvalds 1222cdf7b341SHuang Shijie if (flags & MAP_LOCKED) 12231da177e4SLinus Torvalds if (!can_do_mlock()) 12241da177e4SLinus Torvalds return -EPERM; 1225ba470de4SRik van Riel 12261da177e4SLinus Torvalds /* mlock MCL_FUTURE? */ 12271da177e4SLinus Torvalds if (vm_flags & VM_LOCKED) { 12281da177e4SLinus Torvalds unsigned long locked, lock_limit; 122993ea1d0aSChris Wright locked = len >> PAGE_SHIFT; 123093ea1d0aSChris Wright locked += mm->locked_vm; 123159e99e5bSJiri Slaby lock_limit = rlimit(RLIMIT_MEMLOCK); 123293ea1d0aSChris Wright lock_limit >>= PAGE_SHIFT; 12331da177e4SLinus Torvalds if (locked > lock_limit && !capable(CAP_IPC_LOCK)) 12341da177e4SLinus Torvalds return -EAGAIN; 12351da177e4SLinus Torvalds } 12361da177e4SLinus Torvalds 1237496ad9aaSAl Viro inode = file ? file_inode(file) : NULL; 12381da177e4SLinus Torvalds 12391da177e4SLinus Torvalds if (file) { 12401da177e4SLinus Torvalds switch (flags & MAP_TYPE) { 12411da177e4SLinus Torvalds case MAP_SHARED: 12421da177e4SLinus Torvalds if ((prot&PROT_WRITE) && !(file->f_mode&FMODE_WRITE)) 12431da177e4SLinus Torvalds return -EACCES; 12441da177e4SLinus Torvalds 12451da177e4SLinus Torvalds /* 12461da177e4SLinus Torvalds * Make sure we don't allow writing to an append-only 12471da177e4SLinus Torvalds * file.. 12481da177e4SLinus Torvalds */ 12491da177e4SLinus Torvalds if (IS_APPEND(inode) && (file->f_mode & FMODE_WRITE)) 12501da177e4SLinus Torvalds return -EACCES; 12511da177e4SLinus Torvalds 12521da177e4SLinus Torvalds /* 12531da177e4SLinus Torvalds * Make sure there are no mandatory locks on the file. 12541da177e4SLinus Torvalds */ 12551da177e4SLinus Torvalds if (locks_verify_locked(inode)) 12561da177e4SLinus Torvalds return -EAGAIN; 12571da177e4SLinus Torvalds 12581da177e4SLinus Torvalds vm_flags |= VM_SHARED | VM_MAYSHARE; 12591da177e4SLinus Torvalds if (!(file->f_mode & FMODE_WRITE)) 12601da177e4SLinus Torvalds vm_flags &= ~(VM_MAYWRITE | VM_SHARED); 12611da177e4SLinus Torvalds 12621da177e4SLinus Torvalds /* fall through */ 12631da177e4SLinus Torvalds case MAP_PRIVATE: 12641da177e4SLinus Torvalds if (!(file->f_mode & FMODE_READ)) 12651da177e4SLinus Torvalds return -EACCES; 1266d3ac7f89SJosef "Jeff" Sipek if (file->f_path.mnt->mnt_flags & MNT_NOEXEC) { 126780c5606cSLinus Torvalds if (vm_flags & VM_EXEC) 126880c5606cSLinus Torvalds return -EPERM; 126980c5606cSLinus Torvalds vm_flags &= ~VM_MAYEXEC; 127080c5606cSLinus Torvalds } 127180c5606cSLinus Torvalds 127280c5606cSLinus Torvalds if (!file->f_op || !file->f_op->mmap) 127380c5606cSLinus Torvalds return -ENODEV; 12741da177e4SLinus Torvalds break; 12751da177e4SLinus Torvalds 12761da177e4SLinus Torvalds default: 12771da177e4SLinus Torvalds return -EINVAL; 12781da177e4SLinus Torvalds } 12791da177e4SLinus Torvalds } else { 12801da177e4SLinus Torvalds switch (flags & MAP_TYPE) { 12811da177e4SLinus Torvalds case MAP_SHARED: 1282ce363942STejun Heo /* 1283ce363942STejun Heo * Ignore pgoff. 1284ce363942STejun Heo */ 1285ce363942STejun Heo pgoff = 0; 12861da177e4SLinus Torvalds vm_flags |= VM_SHARED | VM_MAYSHARE; 12871da177e4SLinus Torvalds break; 12881da177e4SLinus Torvalds case MAP_PRIVATE: 12891da177e4SLinus Torvalds /* 12901da177e4SLinus Torvalds * Set pgoff according to addr for anon_vma. 12911da177e4SLinus Torvalds */ 12921da177e4SLinus Torvalds pgoff = addr >> PAGE_SHIFT; 12931da177e4SLinus Torvalds break; 12941da177e4SLinus Torvalds default: 12951da177e4SLinus Torvalds return -EINVAL; 12961da177e4SLinus Torvalds } 12971da177e4SLinus Torvalds } 12981da177e4SLinus Torvalds 1299c22c0d63SMichel Lespinasse /* 1300c22c0d63SMichel Lespinasse * Set 'VM_NORESERVE' if we should not account for the 1301c22c0d63SMichel Lespinasse * memory use of this mapping. 1302c22c0d63SMichel Lespinasse */ 1303c22c0d63SMichel Lespinasse if (flags & MAP_NORESERVE) { 1304c22c0d63SMichel Lespinasse /* We honor MAP_NORESERVE if allowed to overcommit */ 1305c22c0d63SMichel Lespinasse if (sysctl_overcommit_memory != OVERCOMMIT_NEVER) 1306c22c0d63SMichel Lespinasse vm_flags |= VM_NORESERVE; 1307c22c0d63SMichel Lespinasse 1308c22c0d63SMichel Lespinasse /* hugetlb applies strict overcommit unless MAP_NORESERVE */ 1309c22c0d63SMichel Lespinasse if (file && is_file_hugepages(file)) 1310c22c0d63SMichel Lespinasse vm_flags |= VM_NORESERVE; 1311c22c0d63SMichel Lespinasse } 1312c22c0d63SMichel Lespinasse 1313c22c0d63SMichel Lespinasse addr = mmap_region(file, addr, len, vm_flags, pgoff); 131409a9f1d2SMichel Lespinasse if (!IS_ERR_VALUE(addr) && 131509a9f1d2SMichel Lespinasse ((vm_flags & VM_LOCKED) || 131609a9f1d2SMichel Lespinasse (flags & (MAP_POPULATE | MAP_NONBLOCK)) == MAP_POPULATE)) 131741badc15SMichel Lespinasse *populate = len; 1318bebeb3d6SMichel Lespinasse return addr; 13190165ab44SMiklos Szeredi } 13206be5ceb0SLinus Torvalds 132166f0dc48SHugh Dickins SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len, 132266f0dc48SHugh Dickins unsigned long, prot, unsigned long, flags, 132366f0dc48SHugh Dickins unsigned long, fd, unsigned long, pgoff) 132466f0dc48SHugh Dickins { 132566f0dc48SHugh Dickins struct file *file = NULL; 132666f0dc48SHugh Dickins unsigned long retval = -EBADF; 132766f0dc48SHugh Dickins 132866f0dc48SHugh Dickins if (!(flags & MAP_ANONYMOUS)) { 1329120a795dSAl Viro audit_mmap_fd(fd, flags); 133066f0dc48SHugh Dickins if (unlikely(flags & MAP_HUGETLB)) 133166f0dc48SHugh Dickins return -EINVAL; 133266f0dc48SHugh Dickins file = fget(fd); 133366f0dc48SHugh Dickins if (!file) 133466f0dc48SHugh Dickins goto out; 133566f0dc48SHugh Dickins } else if (flags & MAP_HUGETLB) { 133666f0dc48SHugh Dickins struct user_struct *user = NULL; 133766f0dc48SHugh Dickins /* 133866f0dc48SHugh Dickins * VM_NORESERVE is used because the reservations will be 133966f0dc48SHugh Dickins * taken when vm_ops->mmap() is called 134066f0dc48SHugh Dickins * A dummy user value is used because we are not locking 134166f0dc48SHugh Dickins * memory so no accounting is necessary 134266f0dc48SHugh Dickins */ 134340716e29SSteven Truelove file = hugetlb_file_setup(HUGETLB_ANON_FILE, addr, len, 134442d7395fSAndi Kleen VM_NORESERVE, 134542d7395fSAndi Kleen &user, HUGETLB_ANONHUGE_INODE, 134642d7395fSAndi Kleen (flags >> MAP_HUGE_SHIFT) & MAP_HUGE_MASK); 134766f0dc48SHugh Dickins if (IS_ERR(file)) 134866f0dc48SHugh Dickins return PTR_ERR(file); 134966f0dc48SHugh Dickins } 135066f0dc48SHugh Dickins 135166f0dc48SHugh Dickins flags &= ~(MAP_EXECUTABLE | MAP_DENYWRITE); 135266f0dc48SHugh Dickins 1353eb36c587SAl Viro retval = vm_mmap_pgoff(file, addr, len, prot, flags, pgoff); 135466f0dc48SHugh Dickins if (file) 135566f0dc48SHugh Dickins fput(file); 135666f0dc48SHugh Dickins out: 135766f0dc48SHugh Dickins return retval; 135866f0dc48SHugh Dickins } 135966f0dc48SHugh Dickins 1360a4679373SChristoph Hellwig #ifdef __ARCH_WANT_SYS_OLD_MMAP 1361a4679373SChristoph Hellwig struct mmap_arg_struct { 1362a4679373SChristoph Hellwig unsigned long addr; 1363a4679373SChristoph Hellwig unsigned long len; 1364a4679373SChristoph Hellwig unsigned long prot; 1365a4679373SChristoph Hellwig unsigned long flags; 1366a4679373SChristoph Hellwig unsigned long fd; 1367a4679373SChristoph Hellwig unsigned long offset; 1368a4679373SChristoph Hellwig }; 1369a4679373SChristoph Hellwig 1370a4679373SChristoph Hellwig SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg) 1371a4679373SChristoph Hellwig { 1372a4679373SChristoph Hellwig struct mmap_arg_struct a; 1373a4679373SChristoph Hellwig 1374a4679373SChristoph Hellwig if (copy_from_user(&a, arg, sizeof(a))) 1375a4679373SChristoph Hellwig return -EFAULT; 1376a4679373SChristoph Hellwig if (a.offset & ~PAGE_MASK) 1377a4679373SChristoph Hellwig return -EINVAL; 1378a4679373SChristoph Hellwig 1379a4679373SChristoph Hellwig return sys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd, 1380a4679373SChristoph Hellwig a.offset >> PAGE_SHIFT); 1381a4679373SChristoph Hellwig } 1382a4679373SChristoph Hellwig #endif /* __ARCH_WANT_SYS_OLD_MMAP */ 1383a4679373SChristoph Hellwig 13844e950f6fSAlexey Dobriyan /* 13854e950f6fSAlexey Dobriyan * Some shared mappigns will want the pages marked read-only 13864e950f6fSAlexey Dobriyan * to track write events. If so, we'll downgrade vm_page_prot 13874e950f6fSAlexey Dobriyan * to the private version (using protection_map[] without the 13884e950f6fSAlexey Dobriyan * VM_SHARED bit). 13894e950f6fSAlexey Dobriyan */ 13904e950f6fSAlexey Dobriyan int vma_wants_writenotify(struct vm_area_struct *vma) 13914e950f6fSAlexey Dobriyan { 1392ca16d140SKOSAKI Motohiro vm_flags_t vm_flags = vma->vm_flags; 13934e950f6fSAlexey Dobriyan 13944e950f6fSAlexey Dobriyan /* If it was private or non-writable, the write bit is already clear */ 13954e950f6fSAlexey Dobriyan if ((vm_flags & (VM_WRITE|VM_SHARED)) != ((VM_WRITE|VM_SHARED))) 13964e950f6fSAlexey Dobriyan return 0; 13974e950f6fSAlexey Dobriyan 13984e950f6fSAlexey Dobriyan /* The backer wishes to know when pages are first written to? */ 13994e950f6fSAlexey Dobriyan if (vma->vm_ops && vma->vm_ops->page_mkwrite) 14004e950f6fSAlexey Dobriyan return 1; 14014e950f6fSAlexey Dobriyan 14024e950f6fSAlexey Dobriyan /* The open routine did something to the protections already? */ 14034e950f6fSAlexey Dobriyan if (pgprot_val(vma->vm_page_prot) != 14043ed75eb8SColy Li pgprot_val(vm_get_page_prot(vm_flags))) 14054e950f6fSAlexey Dobriyan return 0; 14064e950f6fSAlexey Dobriyan 14074e950f6fSAlexey Dobriyan /* Specialty mapping? */ 14084b6e1e37SKonstantin Khlebnikov if (vm_flags & VM_PFNMAP) 14094e950f6fSAlexey Dobriyan return 0; 14104e950f6fSAlexey Dobriyan 14114e950f6fSAlexey Dobriyan /* Can the mapping track the dirty pages? */ 14124e950f6fSAlexey Dobriyan return vma->vm_file && vma->vm_file->f_mapping && 14134e950f6fSAlexey Dobriyan mapping_cap_account_dirty(vma->vm_file->f_mapping); 14144e950f6fSAlexey Dobriyan } 14154e950f6fSAlexey Dobriyan 1416fc8744adSLinus Torvalds /* 1417fc8744adSLinus Torvalds * We account for memory if it's a private writeable mapping, 14185a6fe125SMel Gorman * not hugepages and VM_NORESERVE wasn't set. 1419fc8744adSLinus Torvalds */ 1420ca16d140SKOSAKI Motohiro static inline int accountable_mapping(struct file *file, vm_flags_t vm_flags) 1421fc8744adSLinus Torvalds { 14225a6fe125SMel Gorman /* 14235a6fe125SMel Gorman * hugetlb has its own accounting separate from the core VM 14245a6fe125SMel Gorman * VM_HUGETLB may not be set yet so we cannot check for that flag. 14255a6fe125SMel Gorman */ 14265a6fe125SMel Gorman if (file && is_file_hugepages(file)) 14275a6fe125SMel Gorman return 0; 14285a6fe125SMel Gorman 1429fc8744adSLinus Torvalds return (vm_flags & (VM_NORESERVE | VM_SHARED | VM_WRITE)) == VM_WRITE; 1430fc8744adSLinus Torvalds } 1431fc8744adSLinus Torvalds 14320165ab44SMiklos Szeredi unsigned long mmap_region(struct file *file, unsigned long addr, 1433c22c0d63SMichel Lespinasse unsigned long len, vm_flags_t vm_flags, unsigned long pgoff) 14340165ab44SMiklos Szeredi { 14350165ab44SMiklos Szeredi struct mm_struct *mm = current->mm; 14360165ab44SMiklos Szeredi struct vm_area_struct *vma, *prev; 14370165ab44SMiklos Szeredi int correct_wcount = 0; 14380165ab44SMiklos Szeredi int error; 14390165ab44SMiklos Szeredi struct rb_node **rb_link, *rb_parent; 14400165ab44SMiklos Szeredi unsigned long charged = 0; 1441496ad9aaSAl Viro struct inode *inode = file ? file_inode(file) : NULL; 14420165ab44SMiklos Szeredi 14431da177e4SLinus Torvalds /* Clear old maps */ 14441da177e4SLinus Torvalds error = -ENOMEM; 14451da177e4SLinus Torvalds munmap_back: 14466597d783SHugh Dickins if (find_vma_links(mm, addr, addr + len, &prev, &rb_link, &rb_parent)) { 14471da177e4SLinus Torvalds if (do_munmap(mm, addr, len)) 14481da177e4SLinus Torvalds return -ENOMEM; 14491da177e4SLinus Torvalds goto munmap_back; 14501da177e4SLinus Torvalds } 14511da177e4SLinus Torvalds 14521da177e4SLinus Torvalds /* Check against address space limit. */ 1453119f657cSakpm@osdl.org if (!may_expand_vm(mm, len >> PAGE_SHIFT)) 14541da177e4SLinus Torvalds return -ENOMEM; 14551da177e4SLinus Torvalds 1456fc8744adSLinus Torvalds /* 14571da177e4SLinus Torvalds * Private writable mapping: check memory availability 14581da177e4SLinus Torvalds */ 14595a6fe125SMel Gorman if (accountable_mapping(file, vm_flags)) { 14601da177e4SLinus Torvalds charged = len >> PAGE_SHIFT; 1461191c5424SAl Viro if (security_vm_enough_memory_mm(mm, charged)) 14621da177e4SLinus Torvalds return -ENOMEM; 14631da177e4SLinus Torvalds vm_flags |= VM_ACCOUNT; 14641da177e4SLinus Torvalds } 14651da177e4SLinus Torvalds 14661da177e4SLinus Torvalds /* 1467de33c8dbSLinus Torvalds * Can we just expand an old mapping? 14681da177e4SLinus Torvalds */ 1469de33c8dbSLinus Torvalds vma = vma_merge(mm, prev, addr, addr + len, vm_flags, NULL, file, pgoff, NULL); 1470ba470de4SRik van Riel if (vma) 14711da177e4SLinus Torvalds goto out; 14721da177e4SLinus Torvalds 14731da177e4SLinus Torvalds /* 14741da177e4SLinus Torvalds * Determine the object being mapped and call the appropriate 14751da177e4SLinus Torvalds * specific mapper. the address has already been validated, but 14761da177e4SLinus Torvalds * not unmapped, but the maps are removed from the list. 14771da177e4SLinus Torvalds */ 1478c5e3b83eSPekka Enberg vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL); 14791da177e4SLinus Torvalds if (!vma) { 14801da177e4SLinus Torvalds error = -ENOMEM; 14811da177e4SLinus Torvalds goto unacct_error; 14821da177e4SLinus Torvalds } 14831da177e4SLinus Torvalds 14841da177e4SLinus Torvalds vma->vm_mm = mm; 14851da177e4SLinus Torvalds vma->vm_start = addr; 14861da177e4SLinus Torvalds vma->vm_end = addr + len; 14871da177e4SLinus Torvalds vma->vm_flags = vm_flags; 14883ed75eb8SColy Li vma->vm_page_prot = vm_get_page_prot(vm_flags); 14891da177e4SLinus Torvalds vma->vm_pgoff = pgoff; 14905beb4930SRik van Riel INIT_LIST_HEAD(&vma->anon_vma_chain); 14911da177e4SLinus Torvalds 1492ce8fea7aSHugh Dickins error = -EINVAL; /* when rejecting VM_GROWSDOWN|VM_GROWSUP */ 1493ce8fea7aSHugh Dickins 14941da177e4SLinus Torvalds if (file) { 14951da177e4SLinus Torvalds if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP)) 14961da177e4SLinus Torvalds goto free_vma; 14971da177e4SLinus Torvalds if (vm_flags & VM_DENYWRITE) { 14981da177e4SLinus Torvalds error = deny_write_access(file); 14991da177e4SLinus Torvalds if (error) 15001da177e4SLinus Torvalds goto free_vma; 15011da177e4SLinus Torvalds correct_wcount = 1; 15021da177e4SLinus Torvalds } 1503cb0942b8SAl Viro vma->vm_file = get_file(file); 15041da177e4SLinus Torvalds error = file->f_op->mmap(file, vma); 15051da177e4SLinus Torvalds if (error) 15061da177e4SLinus Torvalds goto unmap_and_free_vma; 15071da177e4SLinus Torvalds 15081da177e4SLinus Torvalds /* Can addr have changed?? 15091da177e4SLinus Torvalds * 15101da177e4SLinus Torvalds * Answer: Yes, several device drivers can do it in their 15111da177e4SLinus Torvalds * f_op->mmap method. -DaveM 15122897b4d2SJoonsoo Kim * Bug: If addr is changed, prev, rb_link, rb_parent should 15132897b4d2SJoonsoo Kim * be updated for vma_link() 15141da177e4SLinus Torvalds */ 15152897b4d2SJoonsoo Kim WARN_ON_ONCE(addr != vma->vm_start); 15162897b4d2SJoonsoo Kim 15171da177e4SLinus Torvalds addr = vma->vm_start; 15181da177e4SLinus Torvalds pgoff = vma->vm_pgoff; 15191da177e4SLinus Torvalds vm_flags = vma->vm_flags; 1520f8dbf0a7SHuang Shijie } else if (vm_flags & VM_SHARED) { 1521835ee797SAl Viro if (unlikely(vm_flags & (VM_GROWSDOWN|VM_GROWSUP))) 1522835ee797SAl Viro goto free_vma; 1523f8dbf0a7SHuang Shijie error = shmem_zero_setup(vma); 1524f8dbf0a7SHuang Shijie if (error) 1525f8dbf0a7SHuang Shijie goto free_vma; 1526f8dbf0a7SHuang Shijie } 15271da177e4SLinus Torvalds 1528c9d0bf24SMagnus Damm if (vma_wants_writenotify(vma)) { 1529c9d0bf24SMagnus Damm pgprot_t pprot = vma->vm_page_prot; 1530c9d0bf24SMagnus Damm 1531c9d0bf24SMagnus Damm /* Can vma->vm_page_prot have changed?? 1532c9d0bf24SMagnus Damm * 1533c9d0bf24SMagnus Damm * Answer: Yes, drivers may have changed it in their 1534c9d0bf24SMagnus Damm * f_op->mmap method. 1535c9d0bf24SMagnus Damm * 1536c9d0bf24SMagnus Damm * Ensures that vmas marked as uncached stay that way. 1537c9d0bf24SMagnus Damm */ 15381ddd439eSHugh Dickins vma->vm_page_prot = vm_get_page_prot(vm_flags & ~VM_SHARED); 1539c9d0bf24SMagnus Damm if (pgprot_val(pprot) == pgprot_val(pgprot_noncached(pprot))) 1540c9d0bf24SMagnus Damm vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot); 1541c9d0bf24SMagnus Damm } 1542d08b3851SPeter Zijlstra 15434d3d5b41SOleg Nesterov vma_link(mm, vma, prev, rb_link, rb_parent); 15444d3d5b41SOleg Nesterov file = vma->vm_file; 15454d3d5b41SOleg Nesterov 15464d3d5b41SOleg Nesterov /* Once vma denies write, undo our temporary denial count */ 15474d3d5b41SOleg Nesterov if (correct_wcount) 15484d3d5b41SOleg Nesterov atomic_inc(&inode->i_writecount); 15491da177e4SLinus Torvalds out: 1550cdd6c482SIngo Molnar perf_event_mmap(vma); 15510a4a9391SPeter Zijlstra 1552ab50b8edSHugh Dickins vm_stat_account(mm, vm_flags, file, len >> PAGE_SHIFT); 15531da177e4SLinus Torvalds if (vm_flags & VM_LOCKED) { 1554bebeb3d6SMichel Lespinasse if (!((vm_flags & VM_SPECIAL) || is_vm_hugetlb_page(vma) || 1555bebeb3d6SMichel Lespinasse vma == get_gate_vma(current->mm))) 155606f9d8c2SKOSAKI Motohiro mm->locked_vm += (len >> PAGE_SHIFT); 1557bebeb3d6SMichel Lespinasse else 1558bebeb3d6SMichel Lespinasse vma->vm_flags &= ~VM_LOCKED; 1559bebeb3d6SMichel Lespinasse } 15602b144498SSrikar Dronamraju 1561c7a3a88cSOleg Nesterov if (file) 1562c7a3a88cSOleg Nesterov uprobe_mmap(vma); 15632b144498SSrikar Dronamraju 15641da177e4SLinus Torvalds return addr; 15651da177e4SLinus Torvalds 15661da177e4SLinus Torvalds unmap_and_free_vma: 15671da177e4SLinus Torvalds if (correct_wcount) 15681da177e4SLinus Torvalds atomic_inc(&inode->i_writecount); 15691da177e4SLinus Torvalds vma->vm_file = NULL; 15701da177e4SLinus Torvalds fput(file); 15711da177e4SLinus Torvalds 15721da177e4SLinus Torvalds /* Undo any partial mapping done by a device driver. */ 1573e0da382cSHugh Dickins unmap_region(mm, vma, prev, vma->vm_start, vma->vm_end); 1574e0da382cSHugh Dickins charged = 0; 15751da177e4SLinus Torvalds free_vma: 15761da177e4SLinus Torvalds kmem_cache_free(vm_area_cachep, vma); 15771da177e4SLinus Torvalds unacct_error: 15781da177e4SLinus Torvalds if (charged) 15791da177e4SLinus Torvalds vm_unacct_memory(charged); 15801da177e4SLinus Torvalds return error; 15811da177e4SLinus Torvalds } 15821da177e4SLinus Torvalds 1583db4fbfb9SMichel Lespinasse unsigned long unmapped_area(struct vm_unmapped_area_info *info) 1584db4fbfb9SMichel Lespinasse { 1585db4fbfb9SMichel Lespinasse /* 1586db4fbfb9SMichel Lespinasse * We implement the search by looking for an rbtree node that 1587db4fbfb9SMichel Lespinasse * immediately follows a suitable gap. That is, 1588db4fbfb9SMichel Lespinasse * - gap_start = vma->vm_prev->vm_end <= info->high_limit - length; 1589db4fbfb9SMichel Lespinasse * - gap_end = vma->vm_start >= info->low_limit + length; 1590db4fbfb9SMichel Lespinasse * - gap_end - gap_start >= length 1591db4fbfb9SMichel Lespinasse */ 1592db4fbfb9SMichel Lespinasse 1593db4fbfb9SMichel Lespinasse struct mm_struct *mm = current->mm; 1594db4fbfb9SMichel Lespinasse struct vm_area_struct *vma; 1595db4fbfb9SMichel Lespinasse unsigned long length, low_limit, high_limit, gap_start, gap_end; 1596db4fbfb9SMichel Lespinasse 1597db4fbfb9SMichel Lespinasse /* Adjust search length to account for worst case alignment overhead */ 1598db4fbfb9SMichel Lespinasse length = info->length + info->align_mask; 1599db4fbfb9SMichel Lespinasse if (length < info->length) 1600db4fbfb9SMichel Lespinasse return -ENOMEM; 1601db4fbfb9SMichel Lespinasse 1602db4fbfb9SMichel Lespinasse /* Adjust search limits by the desired length */ 1603db4fbfb9SMichel Lespinasse if (info->high_limit < length) 1604db4fbfb9SMichel Lespinasse return -ENOMEM; 1605db4fbfb9SMichel Lespinasse high_limit = info->high_limit - length; 1606db4fbfb9SMichel Lespinasse 1607db4fbfb9SMichel Lespinasse if (info->low_limit > high_limit) 1608db4fbfb9SMichel Lespinasse return -ENOMEM; 1609db4fbfb9SMichel Lespinasse low_limit = info->low_limit + length; 1610db4fbfb9SMichel Lespinasse 1611db4fbfb9SMichel Lespinasse /* Check if rbtree root looks promising */ 1612db4fbfb9SMichel Lespinasse if (RB_EMPTY_ROOT(&mm->mm_rb)) 1613db4fbfb9SMichel Lespinasse goto check_highest; 1614db4fbfb9SMichel Lespinasse vma = rb_entry(mm->mm_rb.rb_node, struct vm_area_struct, vm_rb); 1615db4fbfb9SMichel Lespinasse if (vma->rb_subtree_gap < length) 1616db4fbfb9SMichel Lespinasse goto check_highest; 1617db4fbfb9SMichel Lespinasse 1618db4fbfb9SMichel Lespinasse while (true) { 1619db4fbfb9SMichel Lespinasse /* Visit left subtree if it looks promising */ 1620db4fbfb9SMichel Lespinasse gap_end = vma->vm_start; 1621db4fbfb9SMichel Lespinasse if (gap_end >= low_limit && vma->vm_rb.rb_left) { 1622db4fbfb9SMichel Lespinasse struct vm_area_struct *left = 1623db4fbfb9SMichel Lespinasse rb_entry(vma->vm_rb.rb_left, 1624db4fbfb9SMichel Lespinasse struct vm_area_struct, vm_rb); 1625db4fbfb9SMichel Lespinasse if (left->rb_subtree_gap >= length) { 1626db4fbfb9SMichel Lespinasse vma = left; 1627db4fbfb9SMichel Lespinasse continue; 1628db4fbfb9SMichel Lespinasse } 1629db4fbfb9SMichel Lespinasse } 1630db4fbfb9SMichel Lespinasse 1631db4fbfb9SMichel Lespinasse gap_start = vma->vm_prev ? vma->vm_prev->vm_end : 0; 1632db4fbfb9SMichel Lespinasse check_current: 1633db4fbfb9SMichel Lespinasse /* Check if current node has a suitable gap */ 1634db4fbfb9SMichel Lespinasse if (gap_start > high_limit) 1635db4fbfb9SMichel Lespinasse return -ENOMEM; 1636db4fbfb9SMichel Lespinasse if (gap_end >= low_limit && gap_end - gap_start >= length) 1637db4fbfb9SMichel Lespinasse goto found; 1638db4fbfb9SMichel Lespinasse 1639db4fbfb9SMichel Lespinasse /* Visit right subtree if it looks promising */ 1640db4fbfb9SMichel Lespinasse if (vma->vm_rb.rb_right) { 1641db4fbfb9SMichel Lespinasse struct vm_area_struct *right = 1642db4fbfb9SMichel Lespinasse rb_entry(vma->vm_rb.rb_right, 1643db4fbfb9SMichel Lespinasse struct vm_area_struct, vm_rb); 1644db4fbfb9SMichel Lespinasse if (right->rb_subtree_gap >= length) { 1645db4fbfb9SMichel Lespinasse vma = right; 1646db4fbfb9SMichel Lespinasse continue; 1647db4fbfb9SMichel Lespinasse } 1648db4fbfb9SMichel Lespinasse } 1649db4fbfb9SMichel Lespinasse 1650db4fbfb9SMichel Lespinasse /* Go back up the rbtree to find next candidate node */ 1651db4fbfb9SMichel Lespinasse while (true) { 1652db4fbfb9SMichel Lespinasse struct rb_node *prev = &vma->vm_rb; 1653db4fbfb9SMichel Lespinasse if (!rb_parent(prev)) 1654db4fbfb9SMichel Lespinasse goto check_highest; 1655db4fbfb9SMichel Lespinasse vma = rb_entry(rb_parent(prev), 1656db4fbfb9SMichel Lespinasse struct vm_area_struct, vm_rb); 1657db4fbfb9SMichel Lespinasse if (prev == vma->vm_rb.rb_left) { 1658db4fbfb9SMichel Lespinasse gap_start = vma->vm_prev->vm_end; 1659db4fbfb9SMichel Lespinasse gap_end = vma->vm_start; 1660db4fbfb9SMichel Lespinasse goto check_current; 1661db4fbfb9SMichel Lespinasse } 1662db4fbfb9SMichel Lespinasse } 1663db4fbfb9SMichel Lespinasse } 1664db4fbfb9SMichel Lespinasse 1665db4fbfb9SMichel Lespinasse check_highest: 1666db4fbfb9SMichel Lespinasse /* Check highest gap, which does not precede any rbtree node */ 1667db4fbfb9SMichel Lespinasse gap_start = mm->highest_vm_end; 1668db4fbfb9SMichel Lespinasse gap_end = ULONG_MAX; /* Only for VM_BUG_ON below */ 1669db4fbfb9SMichel Lespinasse if (gap_start > high_limit) 1670db4fbfb9SMichel Lespinasse return -ENOMEM; 1671db4fbfb9SMichel Lespinasse 1672db4fbfb9SMichel Lespinasse found: 1673db4fbfb9SMichel Lespinasse /* We found a suitable gap. Clip it with the original low_limit. */ 1674db4fbfb9SMichel Lespinasse if (gap_start < info->low_limit) 1675db4fbfb9SMichel Lespinasse gap_start = info->low_limit; 1676db4fbfb9SMichel Lespinasse 1677db4fbfb9SMichel Lespinasse /* Adjust gap address to the desired alignment */ 1678db4fbfb9SMichel Lespinasse gap_start += (info->align_offset - gap_start) & info->align_mask; 1679db4fbfb9SMichel Lespinasse 1680db4fbfb9SMichel Lespinasse VM_BUG_ON(gap_start + info->length > info->high_limit); 1681db4fbfb9SMichel Lespinasse VM_BUG_ON(gap_start + info->length > gap_end); 1682db4fbfb9SMichel Lespinasse return gap_start; 1683db4fbfb9SMichel Lespinasse } 1684db4fbfb9SMichel Lespinasse 1685db4fbfb9SMichel Lespinasse unsigned long unmapped_area_topdown(struct vm_unmapped_area_info *info) 1686db4fbfb9SMichel Lespinasse { 1687db4fbfb9SMichel Lespinasse struct mm_struct *mm = current->mm; 1688db4fbfb9SMichel Lespinasse struct vm_area_struct *vma; 1689db4fbfb9SMichel Lespinasse unsigned long length, low_limit, high_limit, gap_start, gap_end; 1690db4fbfb9SMichel Lespinasse 1691db4fbfb9SMichel Lespinasse /* Adjust search length to account for worst case alignment overhead */ 1692db4fbfb9SMichel Lespinasse length = info->length + info->align_mask; 1693db4fbfb9SMichel Lespinasse if (length < info->length) 1694db4fbfb9SMichel Lespinasse return -ENOMEM; 1695db4fbfb9SMichel Lespinasse 1696db4fbfb9SMichel Lespinasse /* 1697db4fbfb9SMichel Lespinasse * Adjust search limits by the desired length. 1698db4fbfb9SMichel Lespinasse * See implementation comment at top of unmapped_area(). 1699db4fbfb9SMichel Lespinasse */ 1700db4fbfb9SMichel Lespinasse gap_end = info->high_limit; 1701db4fbfb9SMichel Lespinasse if (gap_end < length) 1702db4fbfb9SMichel Lespinasse return -ENOMEM; 1703db4fbfb9SMichel Lespinasse high_limit = gap_end - length; 1704db4fbfb9SMichel Lespinasse 1705db4fbfb9SMichel Lespinasse if (info->low_limit > high_limit) 1706db4fbfb9SMichel Lespinasse return -ENOMEM; 1707db4fbfb9SMichel Lespinasse low_limit = info->low_limit + length; 1708db4fbfb9SMichel Lespinasse 1709db4fbfb9SMichel Lespinasse /* Check highest gap, which does not precede any rbtree node */ 1710db4fbfb9SMichel Lespinasse gap_start = mm->highest_vm_end; 1711db4fbfb9SMichel Lespinasse if (gap_start <= high_limit) 1712db4fbfb9SMichel Lespinasse goto found_highest; 1713db4fbfb9SMichel Lespinasse 1714db4fbfb9SMichel Lespinasse /* Check if rbtree root looks promising */ 1715db4fbfb9SMichel Lespinasse if (RB_EMPTY_ROOT(&mm->mm_rb)) 1716db4fbfb9SMichel Lespinasse return -ENOMEM; 1717db4fbfb9SMichel Lespinasse vma = rb_entry(mm->mm_rb.rb_node, struct vm_area_struct, vm_rb); 1718db4fbfb9SMichel Lespinasse if (vma->rb_subtree_gap < length) 1719db4fbfb9SMichel Lespinasse return -ENOMEM; 1720db4fbfb9SMichel Lespinasse 1721db4fbfb9SMichel Lespinasse while (true) { 1722db4fbfb9SMichel Lespinasse /* Visit right subtree if it looks promising */ 1723db4fbfb9SMichel Lespinasse gap_start = vma->vm_prev ? vma->vm_prev->vm_end : 0; 1724db4fbfb9SMichel Lespinasse if (gap_start <= high_limit && vma->vm_rb.rb_right) { 1725db4fbfb9SMichel Lespinasse struct vm_area_struct *right = 1726db4fbfb9SMichel Lespinasse rb_entry(vma->vm_rb.rb_right, 1727db4fbfb9SMichel Lespinasse struct vm_area_struct, vm_rb); 1728db4fbfb9SMichel Lespinasse if (right->rb_subtree_gap >= length) { 1729db4fbfb9SMichel Lespinasse vma = right; 1730db4fbfb9SMichel Lespinasse continue; 1731db4fbfb9SMichel Lespinasse } 1732db4fbfb9SMichel Lespinasse } 1733db4fbfb9SMichel Lespinasse 1734db4fbfb9SMichel Lespinasse check_current: 1735db4fbfb9SMichel Lespinasse /* Check if current node has a suitable gap */ 1736db4fbfb9SMichel Lespinasse gap_end = vma->vm_start; 1737db4fbfb9SMichel Lespinasse if (gap_end < low_limit) 1738db4fbfb9SMichel Lespinasse return -ENOMEM; 1739db4fbfb9SMichel Lespinasse if (gap_start <= high_limit && gap_end - gap_start >= length) 1740db4fbfb9SMichel Lespinasse goto found; 1741db4fbfb9SMichel Lespinasse 1742db4fbfb9SMichel Lespinasse /* Visit left subtree if it looks promising */ 1743db4fbfb9SMichel Lespinasse if (vma->vm_rb.rb_left) { 1744db4fbfb9SMichel Lespinasse struct vm_area_struct *left = 1745db4fbfb9SMichel Lespinasse rb_entry(vma->vm_rb.rb_left, 1746db4fbfb9SMichel Lespinasse struct vm_area_struct, vm_rb); 1747db4fbfb9SMichel Lespinasse if (left->rb_subtree_gap >= length) { 1748db4fbfb9SMichel Lespinasse vma = left; 1749db4fbfb9SMichel Lespinasse continue; 1750db4fbfb9SMichel Lespinasse } 1751db4fbfb9SMichel Lespinasse } 1752db4fbfb9SMichel Lespinasse 1753db4fbfb9SMichel Lespinasse /* Go back up the rbtree to find next candidate node */ 1754db4fbfb9SMichel Lespinasse while (true) { 1755db4fbfb9SMichel Lespinasse struct rb_node *prev = &vma->vm_rb; 1756db4fbfb9SMichel Lespinasse if (!rb_parent(prev)) 1757db4fbfb9SMichel Lespinasse return -ENOMEM; 1758db4fbfb9SMichel Lespinasse vma = rb_entry(rb_parent(prev), 1759db4fbfb9SMichel Lespinasse struct vm_area_struct, vm_rb); 1760db4fbfb9SMichel Lespinasse if (prev == vma->vm_rb.rb_right) { 1761db4fbfb9SMichel Lespinasse gap_start = vma->vm_prev ? 1762db4fbfb9SMichel Lespinasse vma->vm_prev->vm_end : 0; 1763db4fbfb9SMichel Lespinasse goto check_current; 1764db4fbfb9SMichel Lespinasse } 1765db4fbfb9SMichel Lespinasse } 1766db4fbfb9SMichel Lespinasse } 1767db4fbfb9SMichel Lespinasse 1768db4fbfb9SMichel Lespinasse found: 1769db4fbfb9SMichel Lespinasse /* We found a suitable gap. Clip it with the original high_limit. */ 1770db4fbfb9SMichel Lespinasse if (gap_end > info->high_limit) 1771db4fbfb9SMichel Lespinasse gap_end = info->high_limit; 1772db4fbfb9SMichel Lespinasse 1773db4fbfb9SMichel Lespinasse found_highest: 1774db4fbfb9SMichel Lespinasse /* Compute highest gap address at the desired alignment */ 1775db4fbfb9SMichel Lespinasse gap_end -= info->length; 1776db4fbfb9SMichel Lespinasse gap_end -= (gap_end - info->align_offset) & info->align_mask; 1777db4fbfb9SMichel Lespinasse 1778db4fbfb9SMichel Lespinasse VM_BUG_ON(gap_end < info->low_limit); 1779db4fbfb9SMichel Lespinasse VM_BUG_ON(gap_end < gap_start); 1780db4fbfb9SMichel Lespinasse return gap_end; 1781db4fbfb9SMichel Lespinasse } 1782db4fbfb9SMichel Lespinasse 17831da177e4SLinus Torvalds /* Get an address range which is currently unmapped. 17841da177e4SLinus Torvalds * For shmat() with addr=0. 17851da177e4SLinus Torvalds * 17861da177e4SLinus Torvalds * Ugly calling convention alert: 17871da177e4SLinus Torvalds * Return value with the low bits set means error value, 17881da177e4SLinus Torvalds * ie 17891da177e4SLinus Torvalds * if (ret & ~PAGE_MASK) 17901da177e4SLinus Torvalds * error = ret; 17911da177e4SLinus Torvalds * 17921da177e4SLinus Torvalds * This function "knows" that -ENOMEM has the bits set. 17931da177e4SLinus Torvalds */ 17941da177e4SLinus Torvalds #ifndef HAVE_ARCH_UNMAPPED_AREA 17951da177e4SLinus Torvalds unsigned long 17961da177e4SLinus Torvalds arch_get_unmapped_area(struct file *filp, unsigned long addr, 17971da177e4SLinus Torvalds unsigned long len, unsigned long pgoff, unsigned long flags) 17981da177e4SLinus Torvalds { 17991da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 18001da177e4SLinus Torvalds struct vm_area_struct *vma; 1801db4fbfb9SMichel Lespinasse struct vm_unmapped_area_info info; 18021da177e4SLinus Torvalds 18031da177e4SLinus Torvalds if (len > TASK_SIZE) 18041da177e4SLinus Torvalds return -ENOMEM; 18051da177e4SLinus Torvalds 180606abdfb4SBenjamin Herrenschmidt if (flags & MAP_FIXED) 180706abdfb4SBenjamin Herrenschmidt return addr; 180806abdfb4SBenjamin Herrenschmidt 18091da177e4SLinus Torvalds if (addr) { 18101da177e4SLinus Torvalds addr = PAGE_ALIGN(addr); 18111da177e4SLinus Torvalds vma = find_vma(mm, addr); 18121da177e4SLinus Torvalds if (TASK_SIZE - len >= addr && 18131da177e4SLinus Torvalds (!vma || addr + len <= vma->vm_start)) 18141da177e4SLinus Torvalds return addr; 18151da177e4SLinus Torvalds } 18161da177e4SLinus Torvalds 1817db4fbfb9SMichel Lespinasse info.flags = 0; 1818db4fbfb9SMichel Lespinasse info.length = len; 1819db4fbfb9SMichel Lespinasse info.low_limit = TASK_UNMAPPED_BASE; 1820db4fbfb9SMichel Lespinasse info.high_limit = TASK_SIZE; 1821db4fbfb9SMichel Lespinasse info.align_mask = 0; 1822db4fbfb9SMichel Lespinasse return vm_unmapped_area(&info); 18231da177e4SLinus Torvalds } 18241da177e4SLinus Torvalds #endif 18251da177e4SLinus Torvalds 18261363c3cdSWolfgang Wander void arch_unmap_area(struct mm_struct *mm, unsigned long addr) 18271da177e4SLinus Torvalds { 18281da177e4SLinus Torvalds /* 18291da177e4SLinus Torvalds * Is this a new hole at the lowest possible address? 18301da177e4SLinus Torvalds */ 1831f44d2198SXiao Guangrong if (addr >= TASK_UNMAPPED_BASE && addr < mm->free_area_cache) 18321363c3cdSWolfgang Wander mm->free_area_cache = addr; 18331da177e4SLinus Torvalds } 18341da177e4SLinus Torvalds 18351da177e4SLinus Torvalds /* 18361da177e4SLinus Torvalds * This mmap-allocator allocates new areas top-down from below the 18371da177e4SLinus Torvalds * stack's low limit (the base): 18381da177e4SLinus Torvalds */ 18391da177e4SLinus Torvalds #ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN 18401da177e4SLinus Torvalds unsigned long 18411da177e4SLinus Torvalds arch_get_unmapped_area_topdown(struct file *filp, const unsigned long addr0, 18421da177e4SLinus Torvalds const unsigned long len, const unsigned long pgoff, 18431da177e4SLinus Torvalds const unsigned long flags) 18441da177e4SLinus Torvalds { 18451da177e4SLinus Torvalds struct vm_area_struct *vma; 18461da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 1847db4fbfb9SMichel Lespinasse unsigned long addr = addr0; 1848db4fbfb9SMichel Lespinasse struct vm_unmapped_area_info info; 18491da177e4SLinus Torvalds 18501da177e4SLinus Torvalds /* requested length too big for entire address space */ 18511da177e4SLinus Torvalds if (len > TASK_SIZE) 18521da177e4SLinus Torvalds return -ENOMEM; 18531da177e4SLinus Torvalds 185406abdfb4SBenjamin Herrenschmidt if (flags & MAP_FIXED) 185506abdfb4SBenjamin Herrenschmidt return addr; 185606abdfb4SBenjamin Herrenschmidt 18571da177e4SLinus Torvalds /* requesting a specific address */ 18581da177e4SLinus Torvalds if (addr) { 18591da177e4SLinus Torvalds addr = PAGE_ALIGN(addr); 18601da177e4SLinus Torvalds vma = find_vma(mm, addr); 18611da177e4SLinus Torvalds if (TASK_SIZE - len >= addr && 18621da177e4SLinus Torvalds (!vma || addr + len <= vma->vm_start)) 18631da177e4SLinus Torvalds return addr; 18641da177e4SLinus Torvalds } 18651da177e4SLinus Torvalds 1866db4fbfb9SMichel Lespinasse info.flags = VM_UNMAPPED_AREA_TOPDOWN; 1867db4fbfb9SMichel Lespinasse info.length = len; 1868db4fbfb9SMichel Lespinasse info.low_limit = PAGE_SIZE; 1869db4fbfb9SMichel Lespinasse info.high_limit = mm->mmap_base; 1870db4fbfb9SMichel Lespinasse info.align_mask = 0; 1871db4fbfb9SMichel Lespinasse addr = vm_unmapped_area(&info); 1872b716ad95SXiao Guangrong 18731da177e4SLinus Torvalds /* 18741da177e4SLinus Torvalds * A failed mmap() very likely causes application failure, 18751da177e4SLinus Torvalds * so fall back to the bottom-up function here. This scenario 18761da177e4SLinus Torvalds * can happen with large stack limits and large mmap() 18771da177e4SLinus Torvalds * allocations. 18781da177e4SLinus Torvalds */ 1879db4fbfb9SMichel Lespinasse if (addr & ~PAGE_MASK) { 1880db4fbfb9SMichel Lespinasse VM_BUG_ON(addr != -ENOMEM); 1881db4fbfb9SMichel Lespinasse info.flags = 0; 1882db4fbfb9SMichel Lespinasse info.low_limit = TASK_UNMAPPED_BASE; 1883db4fbfb9SMichel Lespinasse info.high_limit = TASK_SIZE; 1884db4fbfb9SMichel Lespinasse addr = vm_unmapped_area(&info); 1885db4fbfb9SMichel Lespinasse } 18861da177e4SLinus Torvalds 18871da177e4SLinus Torvalds return addr; 18881da177e4SLinus Torvalds } 18891da177e4SLinus Torvalds #endif 18901da177e4SLinus Torvalds 18911363c3cdSWolfgang Wander void arch_unmap_area_topdown(struct mm_struct *mm, unsigned long addr) 18921da177e4SLinus Torvalds { 18931da177e4SLinus Torvalds /* 18941da177e4SLinus Torvalds * Is this a new hole at the highest possible address? 18951da177e4SLinus Torvalds */ 18961363c3cdSWolfgang Wander if (addr > mm->free_area_cache) 18971363c3cdSWolfgang Wander mm->free_area_cache = addr; 18981da177e4SLinus Torvalds 18991da177e4SLinus Torvalds /* dont allow allocations above current base */ 19001363c3cdSWolfgang Wander if (mm->free_area_cache > mm->mmap_base) 19011363c3cdSWolfgang Wander mm->free_area_cache = mm->mmap_base; 19021da177e4SLinus Torvalds } 19031da177e4SLinus Torvalds 19041da177e4SLinus Torvalds unsigned long 19051da177e4SLinus Torvalds get_unmapped_area(struct file *file, unsigned long addr, unsigned long len, 19061da177e4SLinus Torvalds unsigned long pgoff, unsigned long flags) 19071da177e4SLinus Torvalds { 190806abdfb4SBenjamin Herrenschmidt unsigned long (*get_area)(struct file *, unsigned long, 190906abdfb4SBenjamin Herrenschmidt unsigned long, unsigned long, unsigned long); 191007ab67c8SLinus Torvalds 19119206de95SAl Viro unsigned long error = arch_mmap_check(addr, len, flags); 19129206de95SAl Viro if (error) 19139206de95SAl Viro return error; 19149206de95SAl Viro 19159206de95SAl Viro /* Careful about overflows.. */ 19169206de95SAl Viro if (len > TASK_SIZE) 19179206de95SAl Viro return -ENOMEM; 19189206de95SAl Viro 191907ab67c8SLinus Torvalds get_area = current->mm->get_unmapped_area; 192007ab67c8SLinus Torvalds if (file && file->f_op && file->f_op->get_unmapped_area) 192107ab67c8SLinus Torvalds get_area = file->f_op->get_unmapped_area; 192207ab67c8SLinus Torvalds addr = get_area(file, addr, len, pgoff, flags); 192307ab67c8SLinus Torvalds if (IS_ERR_VALUE(addr)) 192407ab67c8SLinus Torvalds return addr; 192507ab67c8SLinus Torvalds 19261da177e4SLinus Torvalds if (addr > TASK_SIZE - len) 19271da177e4SLinus Torvalds return -ENOMEM; 19281da177e4SLinus Torvalds if (addr & ~PAGE_MASK) 19291da177e4SLinus Torvalds return -EINVAL; 193006abdfb4SBenjamin Herrenschmidt 19319ac4ed4bSAl Viro addr = arch_rebalance_pgtables(addr, len); 19329ac4ed4bSAl Viro error = security_mmap_addr(addr); 19339ac4ed4bSAl Viro return error ? error : addr; 19341da177e4SLinus Torvalds } 19351da177e4SLinus Torvalds 19361da177e4SLinus Torvalds EXPORT_SYMBOL(get_unmapped_area); 19371da177e4SLinus Torvalds 19381da177e4SLinus Torvalds /* Look up the first VMA which satisfies addr < vm_end, NULL if none. */ 19391da177e4SLinus Torvalds struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr) 19401da177e4SLinus Torvalds { 19411da177e4SLinus Torvalds struct vm_area_struct *vma = NULL; 19421da177e4SLinus Torvalds 19431da177e4SLinus Torvalds /* Check the cache first. */ 19441da177e4SLinus Torvalds /* (Cache hit rate is typically around 35%.) */ 1945b6a9b7f6SJan Stancek vma = ACCESS_ONCE(mm->mmap_cache); 19461da177e4SLinus Torvalds if (!(vma && vma->vm_end > addr && vma->vm_start <= addr)) { 19471da177e4SLinus Torvalds struct rb_node *rb_node; 19481da177e4SLinus Torvalds 19491da177e4SLinus Torvalds rb_node = mm->mm_rb.rb_node; 19501da177e4SLinus Torvalds vma = NULL; 19511da177e4SLinus Torvalds 19521da177e4SLinus Torvalds while (rb_node) { 19531da177e4SLinus Torvalds struct vm_area_struct *vma_tmp; 19541da177e4SLinus Torvalds 19551da177e4SLinus Torvalds vma_tmp = rb_entry(rb_node, 19561da177e4SLinus Torvalds struct vm_area_struct, vm_rb); 19571da177e4SLinus Torvalds 19581da177e4SLinus Torvalds if (vma_tmp->vm_end > addr) { 19591da177e4SLinus Torvalds vma = vma_tmp; 19601da177e4SLinus Torvalds if (vma_tmp->vm_start <= addr) 19611da177e4SLinus Torvalds break; 19621da177e4SLinus Torvalds rb_node = rb_node->rb_left; 19631da177e4SLinus Torvalds } else 19641da177e4SLinus Torvalds rb_node = rb_node->rb_right; 19651da177e4SLinus Torvalds } 19661da177e4SLinus Torvalds if (vma) 19671da177e4SLinus Torvalds mm->mmap_cache = vma; 19681da177e4SLinus Torvalds } 19691da177e4SLinus Torvalds return vma; 19701da177e4SLinus Torvalds } 19711da177e4SLinus Torvalds 19721da177e4SLinus Torvalds EXPORT_SYMBOL(find_vma); 19731da177e4SLinus Torvalds 19746bd4837dSKOSAKI Motohiro /* 19756bd4837dSKOSAKI Motohiro * Same as find_vma, but also return a pointer to the previous VMA in *pprev. 19766bd4837dSKOSAKI Motohiro */ 19771da177e4SLinus Torvalds struct vm_area_struct * 19781da177e4SLinus Torvalds find_vma_prev(struct mm_struct *mm, unsigned long addr, 19791da177e4SLinus Torvalds struct vm_area_struct **pprev) 19801da177e4SLinus Torvalds { 19816bd4837dSKOSAKI Motohiro struct vm_area_struct *vma; 19821da177e4SLinus Torvalds 19836bd4837dSKOSAKI Motohiro vma = find_vma(mm, addr); 198483cd904dSMikulas Patocka if (vma) { 198583cd904dSMikulas Patocka *pprev = vma->vm_prev; 198683cd904dSMikulas Patocka } else { 198783cd904dSMikulas Patocka struct rb_node *rb_node = mm->mm_rb.rb_node; 198883cd904dSMikulas Patocka *pprev = NULL; 198983cd904dSMikulas Patocka while (rb_node) { 199083cd904dSMikulas Patocka *pprev = rb_entry(rb_node, struct vm_area_struct, vm_rb); 199183cd904dSMikulas Patocka rb_node = rb_node->rb_right; 199283cd904dSMikulas Patocka } 199383cd904dSMikulas Patocka } 19946bd4837dSKOSAKI Motohiro return vma; 19951da177e4SLinus Torvalds } 19961da177e4SLinus Torvalds 19971da177e4SLinus Torvalds /* 19981da177e4SLinus Torvalds * Verify that the stack growth is acceptable and 19991da177e4SLinus Torvalds * update accounting. This is shared with both the 20001da177e4SLinus Torvalds * grow-up and grow-down cases. 20011da177e4SLinus Torvalds */ 20021da177e4SLinus Torvalds static int acct_stack_growth(struct vm_area_struct *vma, unsigned long size, unsigned long grow) 20031da177e4SLinus Torvalds { 20041da177e4SLinus Torvalds struct mm_struct *mm = vma->vm_mm; 20051da177e4SLinus Torvalds struct rlimit *rlim = current->signal->rlim; 20060d59a01bSAdam Litke unsigned long new_start; 20071da177e4SLinus Torvalds 20081da177e4SLinus Torvalds /* address space limit tests */ 2009119f657cSakpm@osdl.org if (!may_expand_vm(mm, grow)) 20101da177e4SLinus Torvalds return -ENOMEM; 20111da177e4SLinus Torvalds 20121da177e4SLinus Torvalds /* Stack limit test */ 201359e99e5bSJiri Slaby if (size > ACCESS_ONCE(rlim[RLIMIT_STACK].rlim_cur)) 20141da177e4SLinus Torvalds return -ENOMEM; 20151da177e4SLinus Torvalds 20161da177e4SLinus Torvalds /* mlock limit tests */ 20171da177e4SLinus Torvalds if (vma->vm_flags & VM_LOCKED) { 20181da177e4SLinus Torvalds unsigned long locked; 20191da177e4SLinus Torvalds unsigned long limit; 20201da177e4SLinus Torvalds locked = mm->locked_vm + grow; 202159e99e5bSJiri Slaby limit = ACCESS_ONCE(rlim[RLIMIT_MEMLOCK].rlim_cur); 202259e99e5bSJiri Slaby limit >>= PAGE_SHIFT; 20231da177e4SLinus Torvalds if (locked > limit && !capable(CAP_IPC_LOCK)) 20241da177e4SLinus Torvalds return -ENOMEM; 20251da177e4SLinus Torvalds } 20261da177e4SLinus Torvalds 20270d59a01bSAdam Litke /* Check to ensure the stack will not grow into a hugetlb-only region */ 20280d59a01bSAdam Litke new_start = (vma->vm_flags & VM_GROWSUP) ? vma->vm_start : 20290d59a01bSAdam Litke vma->vm_end - size; 20300d59a01bSAdam Litke if (is_hugepage_only_range(vma->vm_mm, new_start, size)) 20310d59a01bSAdam Litke return -EFAULT; 20320d59a01bSAdam Litke 20331da177e4SLinus Torvalds /* 20341da177e4SLinus Torvalds * Overcommit.. This must be the final test, as it will 20351da177e4SLinus Torvalds * update security statistics. 20361da177e4SLinus Torvalds */ 203705fa199dSHugh Dickins if (security_vm_enough_memory_mm(mm, grow)) 20381da177e4SLinus Torvalds return -ENOMEM; 20391da177e4SLinus Torvalds 20401da177e4SLinus Torvalds /* Ok, everything looks good - let it rip */ 20411da177e4SLinus Torvalds if (vma->vm_flags & VM_LOCKED) 20421da177e4SLinus Torvalds mm->locked_vm += grow; 2043ab50b8edSHugh Dickins vm_stat_account(mm, vma->vm_flags, vma->vm_file, grow); 20441da177e4SLinus Torvalds return 0; 20451da177e4SLinus Torvalds } 20461da177e4SLinus Torvalds 204746dea3d0SHugh Dickins #if defined(CONFIG_STACK_GROWSUP) || defined(CONFIG_IA64) 20481da177e4SLinus Torvalds /* 204946dea3d0SHugh Dickins * PA-RISC uses this for its stack; IA64 for its Register Backing Store. 205046dea3d0SHugh Dickins * vma is the last one with address > vma->vm_end. Have to extend vma. 20511da177e4SLinus Torvalds */ 205246dea3d0SHugh Dickins int expand_upwards(struct vm_area_struct *vma, unsigned long address) 20531da177e4SLinus Torvalds { 20541da177e4SLinus Torvalds int error; 20551da177e4SLinus Torvalds 20561da177e4SLinus Torvalds if (!(vma->vm_flags & VM_GROWSUP)) 20571da177e4SLinus Torvalds return -EFAULT; 20581da177e4SLinus Torvalds 20591da177e4SLinus Torvalds /* 20601da177e4SLinus Torvalds * We must make sure the anon_vma is allocated 20611da177e4SLinus Torvalds * so that the anon_vma locking is not a noop. 20621da177e4SLinus Torvalds */ 20631da177e4SLinus Torvalds if (unlikely(anon_vma_prepare(vma))) 20641da177e4SLinus Torvalds return -ENOMEM; 2065bb4a340eSRik van Riel vma_lock_anon_vma(vma); 20661da177e4SLinus Torvalds 20671da177e4SLinus Torvalds /* 20681da177e4SLinus Torvalds * vma->vm_start/vm_end cannot change under us because the caller 20691da177e4SLinus Torvalds * is required to hold the mmap_sem in read mode. We need the 20701da177e4SLinus Torvalds * anon_vma lock to serialize against concurrent expand_stacks. 207106b32f3aSHelge Deller * Also guard against wrapping around to address 0. 20721da177e4SLinus Torvalds */ 207306b32f3aSHelge Deller if (address < PAGE_ALIGN(address+4)) 207406b32f3aSHelge Deller address = PAGE_ALIGN(address+4); 207506b32f3aSHelge Deller else { 2076bb4a340eSRik van Riel vma_unlock_anon_vma(vma); 207706b32f3aSHelge Deller return -ENOMEM; 207806b32f3aSHelge Deller } 20791da177e4SLinus Torvalds error = 0; 20801da177e4SLinus Torvalds 20811da177e4SLinus Torvalds /* Somebody else might have raced and expanded it already */ 20821da177e4SLinus Torvalds if (address > vma->vm_end) { 20831da177e4SLinus Torvalds unsigned long size, grow; 20841da177e4SLinus Torvalds 20851da177e4SLinus Torvalds size = address - vma->vm_start; 20861da177e4SLinus Torvalds grow = (address - vma->vm_end) >> PAGE_SHIFT; 20871da177e4SLinus Torvalds 208842c36f63SHugh Dickins error = -ENOMEM; 208942c36f63SHugh Dickins if (vma->vm_pgoff + (size >> PAGE_SHIFT) >= vma->vm_pgoff) { 20901da177e4SLinus Torvalds error = acct_stack_growth(vma, size, grow); 20913af9e859SEric B Munson if (!error) { 20924128997bSMichel Lespinasse /* 20934128997bSMichel Lespinasse * vma_gap_update() doesn't support concurrent 20944128997bSMichel Lespinasse * updates, but we only hold a shared mmap_sem 20954128997bSMichel Lespinasse * lock here, so we need to protect against 20964128997bSMichel Lespinasse * concurrent vma expansions. 20974128997bSMichel Lespinasse * vma_lock_anon_vma() doesn't help here, as 20984128997bSMichel Lespinasse * we don't guarantee that all growable vmas 20994128997bSMichel Lespinasse * in a mm share the same root anon vma. 21004128997bSMichel Lespinasse * So, we reuse mm->page_table_lock to guard 21014128997bSMichel Lespinasse * against concurrent vma expansions. 21024128997bSMichel Lespinasse */ 21034128997bSMichel Lespinasse spin_lock(&vma->vm_mm->page_table_lock); 2104bf181b9fSMichel Lespinasse anon_vma_interval_tree_pre_update_vma(vma); 21051da177e4SLinus Torvalds vma->vm_end = address; 2106bf181b9fSMichel Lespinasse anon_vma_interval_tree_post_update_vma(vma); 2107d3737187SMichel Lespinasse if (vma->vm_next) 2108d3737187SMichel Lespinasse vma_gap_update(vma->vm_next); 2109d3737187SMichel Lespinasse else 2110d3737187SMichel Lespinasse vma->vm_mm->highest_vm_end = address; 21114128997bSMichel Lespinasse spin_unlock(&vma->vm_mm->page_table_lock); 21124128997bSMichel Lespinasse 21133af9e859SEric B Munson perf_event_mmap(vma); 21143af9e859SEric B Munson } 21151da177e4SLinus Torvalds } 211642c36f63SHugh Dickins } 2117bb4a340eSRik van Riel vma_unlock_anon_vma(vma); 2118b15d00b6SAndrea Arcangeli khugepaged_enter_vma_merge(vma); 2119ed8ea815SMichel Lespinasse validate_mm(vma->vm_mm); 21201da177e4SLinus Torvalds return error; 21211da177e4SLinus Torvalds } 212246dea3d0SHugh Dickins #endif /* CONFIG_STACK_GROWSUP || CONFIG_IA64 */ 212346dea3d0SHugh Dickins 21241da177e4SLinus Torvalds /* 21251da177e4SLinus Torvalds * vma is the first one with address < vma->vm_start. Have to extend vma. 21261da177e4SLinus Torvalds */ 2127d05f3169SMichal Hocko int expand_downwards(struct vm_area_struct *vma, 2128b6a2fea3SOllie Wild unsigned long address) 21291da177e4SLinus Torvalds { 21301da177e4SLinus Torvalds int error; 21311da177e4SLinus Torvalds 21321da177e4SLinus Torvalds /* 21331da177e4SLinus Torvalds * We must make sure the anon_vma is allocated 21341da177e4SLinus Torvalds * so that the anon_vma locking is not a noop. 21351da177e4SLinus Torvalds */ 21361da177e4SLinus Torvalds if (unlikely(anon_vma_prepare(vma))) 21371da177e4SLinus Torvalds return -ENOMEM; 21388869477aSEric Paris 21398869477aSEric Paris address &= PAGE_MASK; 2140e5467859SAl Viro error = security_mmap_addr(address); 21418869477aSEric Paris if (error) 21428869477aSEric Paris return error; 21438869477aSEric Paris 2144bb4a340eSRik van Riel vma_lock_anon_vma(vma); 21451da177e4SLinus Torvalds 21461da177e4SLinus Torvalds /* 21471da177e4SLinus Torvalds * vma->vm_start/vm_end cannot change under us because the caller 21481da177e4SLinus Torvalds * is required to hold the mmap_sem in read mode. We need the 21491da177e4SLinus Torvalds * anon_vma lock to serialize against concurrent expand_stacks. 21501da177e4SLinus Torvalds */ 21511da177e4SLinus Torvalds 21521da177e4SLinus Torvalds /* Somebody else might have raced and expanded it already */ 21531da177e4SLinus Torvalds if (address < vma->vm_start) { 21541da177e4SLinus Torvalds unsigned long size, grow; 21551da177e4SLinus Torvalds 21561da177e4SLinus Torvalds size = vma->vm_end - address; 21571da177e4SLinus Torvalds grow = (vma->vm_start - address) >> PAGE_SHIFT; 21581da177e4SLinus Torvalds 2159a626ca6aSLinus Torvalds error = -ENOMEM; 2160a626ca6aSLinus Torvalds if (grow <= vma->vm_pgoff) { 21611da177e4SLinus Torvalds error = acct_stack_growth(vma, size, grow); 21621da177e4SLinus Torvalds if (!error) { 21634128997bSMichel Lespinasse /* 21644128997bSMichel Lespinasse * vma_gap_update() doesn't support concurrent 21654128997bSMichel Lespinasse * updates, but we only hold a shared mmap_sem 21664128997bSMichel Lespinasse * lock here, so we need to protect against 21674128997bSMichel Lespinasse * concurrent vma expansions. 21684128997bSMichel Lespinasse * vma_lock_anon_vma() doesn't help here, as 21694128997bSMichel Lespinasse * we don't guarantee that all growable vmas 21704128997bSMichel Lespinasse * in a mm share the same root anon vma. 21714128997bSMichel Lespinasse * So, we reuse mm->page_table_lock to guard 21724128997bSMichel Lespinasse * against concurrent vma expansions. 21734128997bSMichel Lespinasse */ 21744128997bSMichel Lespinasse spin_lock(&vma->vm_mm->page_table_lock); 2175bf181b9fSMichel Lespinasse anon_vma_interval_tree_pre_update_vma(vma); 21761da177e4SLinus Torvalds vma->vm_start = address; 21771da177e4SLinus Torvalds vma->vm_pgoff -= grow; 2178bf181b9fSMichel Lespinasse anon_vma_interval_tree_post_update_vma(vma); 2179d3737187SMichel Lespinasse vma_gap_update(vma); 21804128997bSMichel Lespinasse spin_unlock(&vma->vm_mm->page_table_lock); 21814128997bSMichel Lespinasse 21823af9e859SEric B Munson perf_event_mmap(vma); 21831da177e4SLinus Torvalds } 21841da177e4SLinus Torvalds } 2185a626ca6aSLinus Torvalds } 2186bb4a340eSRik van Riel vma_unlock_anon_vma(vma); 2187b15d00b6SAndrea Arcangeli khugepaged_enter_vma_merge(vma); 2188ed8ea815SMichel Lespinasse validate_mm(vma->vm_mm); 21891da177e4SLinus Torvalds return error; 21901da177e4SLinus Torvalds } 21911da177e4SLinus Torvalds 219209884964SLinus Torvalds /* 219309884964SLinus Torvalds * Note how expand_stack() refuses to expand the stack all the way to 219409884964SLinus Torvalds * abut the next virtual mapping, *unless* that mapping itself is also 219509884964SLinus Torvalds * a stack mapping. We want to leave room for a guard page, after all 219609884964SLinus Torvalds * (the guard page itself is not added here, that is done by the 219709884964SLinus Torvalds * actual page faulting logic) 219809884964SLinus Torvalds * 219909884964SLinus Torvalds * This matches the behavior of the guard page logic (see mm/memory.c: 220009884964SLinus Torvalds * check_stack_guard_page()), which only allows the guard page to be 220109884964SLinus Torvalds * removed under these circumstances. 220209884964SLinus Torvalds */ 2203b6a2fea3SOllie Wild #ifdef CONFIG_STACK_GROWSUP 2204b6a2fea3SOllie Wild int expand_stack(struct vm_area_struct *vma, unsigned long address) 2205b6a2fea3SOllie Wild { 220609884964SLinus Torvalds struct vm_area_struct *next; 220709884964SLinus Torvalds 220809884964SLinus Torvalds address &= PAGE_MASK; 220909884964SLinus Torvalds next = vma->vm_next; 221009884964SLinus Torvalds if (next && next->vm_start == address + PAGE_SIZE) { 221109884964SLinus Torvalds if (!(next->vm_flags & VM_GROWSUP)) 221209884964SLinus Torvalds return -ENOMEM; 221309884964SLinus Torvalds } 2214b6a2fea3SOllie Wild return expand_upwards(vma, address); 2215b6a2fea3SOllie Wild } 2216b6a2fea3SOllie Wild 2217b6a2fea3SOllie Wild struct vm_area_struct * 2218b6a2fea3SOllie Wild find_extend_vma(struct mm_struct *mm, unsigned long addr) 2219b6a2fea3SOllie Wild { 2220b6a2fea3SOllie Wild struct vm_area_struct *vma, *prev; 2221b6a2fea3SOllie Wild 2222b6a2fea3SOllie Wild addr &= PAGE_MASK; 2223b6a2fea3SOllie Wild vma = find_vma_prev(mm, addr, &prev); 2224b6a2fea3SOllie Wild if (vma && (vma->vm_start <= addr)) 2225b6a2fea3SOllie Wild return vma; 22261c127185SDenys Vlasenko if (!prev || expand_stack(prev, addr)) 2227b6a2fea3SOllie Wild return NULL; 2228cea10a19SMichel Lespinasse if (prev->vm_flags & VM_LOCKED) 2229cea10a19SMichel Lespinasse __mlock_vma_pages_range(prev, addr, prev->vm_end, NULL); 2230b6a2fea3SOllie Wild return prev; 2231b6a2fea3SOllie Wild } 2232b6a2fea3SOllie Wild #else 2233b6a2fea3SOllie Wild int expand_stack(struct vm_area_struct *vma, unsigned long address) 2234b6a2fea3SOllie Wild { 223509884964SLinus Torvalds struct vm_area_struct *prev; 223609884964SLinus Torvalds 223709884964SLinus Torvalds address &= PAGE_MASK; 223809884964SLinus Torvalds prev = vma->vm_prev; 223909884964SLinus Torvalds if (prev && prev->vm_end == address) { 224009884964SLinus Torvalds if (!(prev->vm_flags & VM_GROWSDOWN)) 224109884964SLinus Torvalds return -ENOMEM; 224209884964SLinus Torvalds } 2243b6a2fea3SOllie Wild return expand_downwards(vma, address); 2244b6a2fea3SOllie Wild } 2245b6a2fea3SOllie Wild 22461da177e4SLinus Torvalds struct vm_area_struct * 22471da177e4SLinus Torvalds find_extend_vma(struct mm_struct * mm, unsigned long addr) 22481da177e4SLinus Torvalds { 22491da177e4SLinus Torvalds struct vm_area_struct * vma; 22501da177e4SLinus Torvalds unsigned long start; 22511da177e4SLinus Torvalds 22521da177e4SLinus Torvalds addr &= PAGE_MASK; 22531da177e4SLinus Torvalds vma = find_vma(mm,addr); 22541da177e4SLinus Torvalds if (!vma) 22551da177e4SLinus Torvalds return NULL; 22561da177e4SLinus Torvalds if (vma->vm_start <= addr) 22571da177e4SLinus Torvalds return vma; 22581da177e4SLinus Torvalds if (!(vma->vm_flags & VM_GROWSDOWN)) 22591da177e4SLinus Torvalds return NULL; 22601da177e4SLinus Torvalds start = vma->vm_start; 22611da177e4SLinus Torvalds if (expand_stack(vma, addr)) 22621da177e4SLinus Torvalds return NULL; 2263cea10a19SMichel Lespinasse if (vma->vm_flags & VM_LOCKED) 2264cea10a19SMichel Lespinasse __mlock_vma_pages_range(vma, addr, start, NULL); 22651da177e4SLinus Torvalds return vma; 22661da177e4SLinus Torvalds } 22671da177e4SLinus Torvalds #endif 22681da177e4SLinus Torvalds 22692c0b3814SHugh Dickins /* 22702c0b3814SHugh Dickins * Ok - we have the memory areas we should free on the vma list, 22712c0b3814SHugh Dickins * so release them, and do the vma updates. 22721da177e4SLinus Torvalds * 22732c0b3814SHugh Dickins * Called with the mm semaphore held. 22741da177e4SLinus Torvalds */ 22752c0b3814SHugh Dickins static void remove_vma_list(struct mm_struct *mm, struct vm_area_struct *vma) 22761da177e4SLinus Torvalds { 22774f74d2c8SLinus Torvalds unsigned long nr_accounted = 0; 22784f74d2c8SLinus Torvalds 2279365e9c87SHugh Dickins /* Update high watermark before we lower total_vm */ 2280365e9c87SHugh Dickins update_hiwater_vm(mm); 22812c0b3814SHugh Dickins do { 2282ab50b8edSHugh Dickins long nrpages = vma_pages(vma); 22831da177e4SLinus Torvalds 22844f74d2c8SLinus Torvalds if (vma->vm_flags & VM_ACCOUNT) 22854f74d2c8SLinus Torvalds nr_accounted += nrpages; 2286ab50b8edSHugh Dickins vm_stat_account(mm, vma->vm_flags, vma->vm_file, -nrpages); 2287a8fb5618SHugh Dickins vma = remove_vma(vma); 2288146425a3SHugh Dickins } while (vma); 22894f74d2c8SLinus Torvalds vm_unacct_memory(nr_accounted); 22901da177e4SLinus Torvalds validate_mm(mm); 22911da177e4SLinus Torvalds } 22921da177e4SLinus Torvalds 22931da177e4SLinus Torvalds /* 22941da177e4SLinus Torvalds * Get rid of page table information in the indicated region. 22951da177e4SLinus Torvalds * 2296f10df686SPaolo 'Blaisorblade' Giarrusso * Called with the mm semaphore held. 22971da177e4SLinus Torvalds */ 22981da177e4SLinus Torvalds static void unmap_region(struct mm_struct *mm, 2299e0da382cSHugh Dickins struct vm_area_struct *vma, struct vm_area_struct *prev, 2300e0da382cSHugh Dickins unsigned long start, unsigned long end) 23011da177e4SLinus Torvalds { 2302e0da382cSHugh Dickins struct vm_area_struct *next = prev? prev->vm_next: mm->mmap; 2303d16dfc55SPeter Zijlstra struct mmu_gather tlb; 23041da177e4SLinus Torvalds 23051da177e4SLinus Torvalds lru_add_drain(); 2306d16dfc55SPeter Zijlstra tlb_gather_mmu(&tlb, mm, 0); 2307365e9c87SHugh Dickins update_hiwater_rss(mm); 23084f74d2c8SLinus Torvalds unmap_vmas(&tlb, vma, start, end); 2309d16dfc55SPeter Zijlstra free_pgtables(&tlb, vma, prev ? prev->vm_end : FIRST_USER_ADDRESS, 23106ee8630eSHugh Dickins next ? next->vm_start : USER_PGTABLES_CEILING); 2311d16dfc55SPeter Zijlstra tlb_finish_mmu(&tlb, start, end); 23121da177e4SLinus Torvalds } 23131da177e4SLinus Torvalds 23141da177e4SLinus Torvalds /* 23151da177e4SLinus Torvalds * Create a list of vma's touched by the unmap, removing them from the mm's 23161da177e4SLinus Torvalds * vma list as we go.. 23171da177e4SLinus Torvalds */ 23181da177e4SLinus Torvalds static void 23191da177e4SLinus Torvalds detach_vmas_to_be_unmapped(struct mm_struct *mm, struct vm_area_struct *vma, 23201da177e4SLinus Torvalds struct vm_area_struct *prev, unsigned long end) 23211da177e4SLinus Torvalds { 23221da177e4SLinus Torvalds struct vm_area_struct **insertion_point; 23231da177e4SLinus Torvalds struct vm_area_struct *tail_vma = NULL; 23241363c3cdSWolfgang Wander unsigned long addr; 23251da177e4SLinus Torvalds 23261da177e4SLinus Torvalds insertion_point = (prev ? &prev->vm_next : &mm->mmap); 2327297c5eeeSLinus Torvalds vma->vm_prev = NULL; 23281da177e4SLinus Torvalds do { 2329d3737187SMichel Lespinasse vma_rb_erase(vma, &mm->mm_rb); 23301da177e4SLinus Torvalds mm->map_count--; 23311da177e4SLinus Torvalds tail_vma = vma; 23321da177e4SLinus Torvalds vma = vma->vm_next; 23331da177e4SLinus Torvalds } while (vma && vma->vm_start < end); 23341da177e4SLinus Torvalds *insertion_point = vma; 2335d3737187SMichel Lespinasse if (vma) { 2336297c5eeeSLinus Torvalds vma->vm_prev = prev; 2337d3737187SMichel Lespinasse vma_gap_update(vma); 2338d3737187SMichel Lespinasse } else 2339d3737187SMichel Lespinasse mm->highest_vm_end = prev ? prev->vm_end : 0; 23401da177e4SLinus Torvalds tail_vma->vm_next = NULL; 23411363c3cdSWolfgang Wander if (mm->unmap_area == arch_unmap_area) 23421363c3cdSWolfgang Wander addr = prev ? prev->vm_end : mm->mmap_base; 23431363c3cdSWolfgang Wander else 23441363c3cdSWolfgang Wander addr = vma ? vma->vm_start : mm->mmap_base; 23451363c3cdSWolfgang Wander mm->unmap_area(mm, addr); 23461da177e4SLinus Torvalds mm->mmap_cache = NULL; /* Kill the cache. */ 23471da177e4SLinus Torvalds } 23481da177e4SLinus Torvalds 23491da177e4SLinus Torvalds /* 2350659ace58SKOSAKI Motohiro * __split_vma() bypasses sysctl_max_map_count checking. We use this on the 2351659ace58SKOSAKI Motohiro * munmap path where it doesn't make sense to fail. 23521da177e4SLinus Torvalds */ 2353659ace58SKOSAKI Motohiro static int __split_vma(struct mm_struct * mm, struct vm_area_struct * vma, 23541da177e4SLinus Torvalds unsigned long addr, int new_below) 23551da177e4SLinus Torvalds { 23561da177e4SLinus Torvalds struct mempolicy *pol; 23571da177e4SLinus Torvalds struct vm_area_struct *new; 23585beb4930SRik van Riel int err = -ENOMEM; 23591da177e4SLinus Torvalds 2360a5516438SAndi Kleen if (is_vm_hugetlb_page(vma) && (addr & 2361a5516438SAndi Kleen ~(huge_page_mask(hstate_vma(vma))))) 23621da177e4SLinus Torvalds return -EINVAL; 23631da177e4SLinus Torvalds 2364e94b1766SChristoph Lameter new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL); 23651da177e4SLinus Torvalds if (!new) 23665beb4930SRik van Riel goto out_err; 23671da177e4SLinus Torvalds 23681da177e4SLinus Torvalds /* most fields are the same, copy all, and then fixup */ 23691da177e4SLinus Torvalds *new = *vma; 23701da177e4SLinus Torvalds 23715beb4930SRik van Riel INIT_LIST_HEAD(&new->anon_vma_chain); 23725beb4930SRik van Riel 23731da177e4SLinus Torvalds if (new_below) 23741da177e4SLinus Torvalds new->vm_end = addr; 23751da177e4SLinus Torvalds else { 23761da177e4SLinus Torvalds new->vm_start = addr; 23771da177e4SLinus Torvalds new->vm_pgoff += ((addr - vma->vm_start) >> PAGE_SHIFT); 23781da177e4SLinus Torvalds } 23791da177e4SLinus Torvalds 2380846a16bfSLee Schermerhorn pol = mpol_dup(vma_policy(vma)); 23811da177e4SLinus Torvalds if (IS_ERR(pol)) { 23825beb4930SRik van Riel err = PTR_ERR(pol); 23835beb4930SRik van Riel goto out_free_vma; 23841da177e4SLinus Torvalds } 23851da177e4SLinus Torvalds vma_set_policy(new, pol); 23861da177e4SLinus Torvalds 23875beb4930SRik van Riel if (anon_vma_clone(new, vma)) 23885beb4930SRik van Riel goto out_free_mpol; 23895beb4930SRik van Riel 2390e9714acfSKonstantin Khlebnikov if (new->vm_file) 23911da177e4SLinus Torvalds get_file(new->vm_file); 23921da177e4SLinus Torvalds 23931da177e4SLinus Torvalds if (new->vm_ops && new->vm_ops->open) 23941da177e4SLinus Torvalds new->vm_ops->open(new); 23951da177e4SLinus Torvalds 23961da177e4SLinus Torvalds if (new_below) 23975beb4930SRik van Riel err = vma_adjust(vma, addr, vma->vm_end, vma->vm_pgoff + 23981da177e4SLinus Torvalds ((addr - new->vm_start) >> PAGE_SHIFT), new); 23991da177e4SLinus Torvalds else 24005beb4930SRik van Riel err = vma_adjust(vma, vma->vm_start, addr, vma->vm_pgoff, new); 24011da177e4SLinus Torvalds 24025beb4930SRik van Riel /* Success. */ 24035beb4930SRik van Riel if (!err) 24041da177e4SLinus Torvalds return 0; 24055beb4930SRik van Riel 24065beb4930SRik van Riel /* Clean everything up if vma_adjust failed. */ 240758927533SRik van Riel if (new->vm_ops && new->vm_ops->close) 24085beb4930SRik van Riel new->vm_ops->close(new); 2409e9714acfSKonstantin Khlebnikov if (new->vm_file) 24105beb4930SRik van Riel fput(new->vm_file); 24112aeadc30SAndrea Arcangeli unlink_anon_vmas(new); 24125beb4930SRik van Riel out_free_mpol: 24135beb4930SRik van Riel mpol_put(pol); 24145beb4930SRik van Riel out_free_vma: 24155beb4930SRik van Riel kmem_cache_free(vm_area_cachep, new); 24165beb4930SRik van Riel out_err: 24175beb4930SRik van Riel return err; 24181da177e4SLinus Torvalds } 24191da177e4SLinus Torvalds 2420659ace58SKOSAKI Motohiro /* 2421659ace58SKOSAKI Motohiro * Split a vma into two pieces at address 'addr', a new vma is allocated 2422659ace58SKOSAKI Motohiro * either for the first part or the tail. 2423659ace58SKOSAKI Motohiro */ 2424659ace58SKOSAKI Motohiro int split_vma(struct mm_struct *mm, struct vm_area_struct *vma, 2425659ace58SKOSAKI Motohiro unsigned long addr, int new_below) 2426659ace58SKOSAKI Motohiro { 2427659ace58SKOSAKI Motohiro if (mm->map_count >= sysctl_max_map_count) 2428659ace58SKOSAKI Motohiro return -ENOMEM; 2429659ace58SKOSAKI Motohiro 2430659ace58SKOSAKI Motohiro return __split_vma(mm, vma, addr, new_below); 2431659ace58SKOSAKI Motohiro } 2432659ace58SKOSAKI Motohiro 24331da177e4SLinus Torvalds /* Munmap is split into 2 main parts -- this part which finds 24341da177e4SLinus Torvalds * what needs doing, and the areas themselves, which do the 24351da177e4SLinus Torvalds * work. This now handles partial unmappings. 24361da177e4SLinus Torvalds * Jeremy Fitzhardinge <jeremy@goop.org> 24371da177e4SLinus Torvalds */ 24381da177e4SLinus Torvalds int do_munmap(struct mm_struct *mm, unsigned long start, size_t len) 24391da177e4SLinus Torvalds { 24401da177e4SLinus Torvalds unsigned long end; 2441146425a3SHugh Dickins struct vm_area_struct *vma, *prev, *last; 24421da177e4SLinus Torvalds 24431da177e4SLinus Torvalds if ((start & ~PAGE_MASK) || start > TASK_SIZE || len > TASK_SIZE-start) 24441da177e4SLinus Torvalds return -EINVAL; 24451da177e4SLinus Torvalds 24461da177e4SLinus Torvalds if ((len = PAGE_ALIGN(len)) == 0) 24471da177e4SLinus Torvalds return -EINVAL; 24481da177e4SLinus Torvalds 24491da177e4SLinus Torvalds /* Find the first overlapping VMA */ 24509be34c9dSLinus Torvalds vma = find_vma(mm, start); 2451146425a3SHugh Dickins if (!vma) 24521da177e4SLinus Torvalds return 0; 24539be34c9dSLinus Torvalds prev = vma->vm_prev; 2454146425a3SHugh Dickins /* we have start < vma->vm_end */ 24551da177e4SLinus Torvalds 24561da177e4SLinus Torvalds /* if it doesn't overlap, we have nothing.. */ 24571da177e4SLinus Torvalds end = start + len; 2458146425a3SHugh Dickins if (vma->vm_start >= end) 24591da177e4SLinus Torvalds return 0; 24601da177e4SLinus Torvalds 24611da177e4SLinus Torvalds /* 24621da177e4SLinus Torvalds * If we need to split any vma, do it now to save pain later. 24631da177e4SLinus Torvalds * 24641da177e4SLinus Torvalds * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially 24651da177e4SLinus Torvalds * unmapped vm_area_struct will remain in use: so lower split_vma 24661da177e4SLinus Torvalds * places tmp vma above, and higher split_vma places tmp vma below. 24671da177e4SLinus Torvalds */ 2468146425a3SHugh Dickins if (start > vma->vm_start) { 2469659ace58SKOSAKI Motohiro int error; 2470659ace58SKOSAKI Motohiro 2471659ace58SKOSAKI Motohiro /* 2472659ace58SKOSAKI Motohiro * Make sure that map_count on return from munmap() will 2473659ace58SKOSAKI Motohiro * not exceed its limit; but let map_count go just above 2474659ace58SKOSAKI Motohiro * its limit temporarily, to help free resources as expected. 2475659ace58SKOSAKI Motohiro */ 2476659ace58SKOSAKI Motohiro if (end < vma->vm_end && mm->map_count >= sysctl_max_map_count) 2477659ace58SKOSAKI Motohiro return -ENOMEM; 2478659ace58SKOSAKI Motohiro 2479659ace58SKOSAKI Motohiro error = __split_vma(mm, vma, start, 0); 24801da177e4SLinus Torvalds if (error) 24811da177e4SLinus Torvalds return error; 2482146425a3SHugh Dickins prev = vma; 24831da177e4SLinus Torvalds } 24841da177e4SLinus Torvalds 24851da177e4SLinus Torvalds /* Does it split the last one? */ 24861da177e4SLinus Torvalds last = find_vma(mm, end); 24871da177e4SLinus Torvalds if (last && end > last->vm_start) { 2488659ace58SKOSAKI Motohiro int error = __split_vma(mm, last, end, 1); 24891da177e4SLinus Torvalds if (error) 24901da177e4SLinus Torvalds return error; 24911da177e4SLinus Torvalds } 2492146425a3SHugh Dickins vma = prev? prev->vm_next: mm->mmap; 24931da177e4SLinus Torvalds 24941da177e4SLinus Torvalds /* 2495ba470de4SRik van Riel * unlock any mlock()ed ranges before detaching vmas 2496ba470de4SRik van Riel */ 2497ba470de4SRik van Riel if (mm->locked_vm) { 2498ba470de4SRik van Riel struct vm_area_struct *tmp = vma; 2499ba470de4SRik van Riel while (tmp && tmp->vm_start < end) { 2500ba470de4SRik van Riel if (tmp->vm_flags & VM_LOCKED) { 2501ba470de4SRik van Riel mm->locked_vm -= vma_pages(tmp); 2502ba470de4SRik van Riel munlock_vma_pages_all(tmp); 2503ba470de4SRik van Riel } 2504ba470de4SRik van Riel tmp = tmp->vm_next; 2505ba470de4SRik van Riel } 2506ba470de4SRik van Riel } 2507ba470de4SRik van Riel 2508ba470de4SRik van Riel /* 25091da177e4SLinus Torvalds * Remove the vma's, and unmap the actual pages 25101da177e4SLinus Torvalds */ 2511146425a3SHugh Dickins detach_vmas_to_be_unmapped(mm, vma, prev, end); 2512146425a3SHugh Dickins unmap_region(mm, vma, prev, start, end); 25131da177e4SLinus Torvalds 25141da177e4SLinus Torvalds /* Fix up all other VM information */ 25152c0b3814SHugh Dickins remove_vma_list(mm, vma); 25161da177e4SLinus Torvalds 25171da177e4SLinus Torvalds return 0; 25181da177e4SLinus Torvalds } 25191da177e4SLinus Torvalds 2520bfce281cSAl Viro int vm_munmap(unsigned long start, size_t len) 2521a46ef99dSLinus Torvalds { 2522a46ef99dSLinus Torvalds int ret; 2523bfce281cSAl Viro struct mm_struct *mm = current->mm; 2524a46ef99dSLinus Torvalds 2525a46ef99dSLinus Torvalds down_write(&mm->mmap_sem); 2526a46ef99dSLinus Torvalds ret = do_munmap(mm, start, len); 2527a46ef99dSLinus Torvalds up_write(&mm->mmap_sem); 2528a46ef99dSLinus Torvalds return ret; 2529a46ef99dSLinus Torvalds } 2530a46ef99dSLinus Torvalds EXPORT_SYMBOL(vm_munmap); 2531a46ef99dSLinus Torvalds 25326a6160a7SHeiko Carstens SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len) 25331da177e4SLinus Torvalds { 25341da177e4SLinus Torvalds profile_munmap(addr); 2535bfce281cSAl Viro return vm_munmap(addr, len); 25361da177e4SLinus Torvalds } 25371da177e4SLinus Torvalds 25381da177e4SLinus Torvalds static inline void verify_mm_writelocked(struct mm_struct *mm) 25391da177e4SLinus Torvalds { 2540a241ec65SPaul E. McKenney #ifdef CONFIG_DEBUG_VM 25411da177e4SLinus Torvalds if (unlikely(down_read_trylock(&mm->mmap_sem))) { 25421da177e4SLinus Torvalds WARN_ON(1); 25431da177e4SLinus Torvalds up_read(&mm->mmap_sem); 25441da177e4SLinus Torvalds } 25451da177e4SLinus Torvalds #endif 25461da177e4SLinus Torvalds } 25471da177e4SLinus Torvalds 25481da177e4SLinus Torvalds /* 25491da177e4SLinus Torvalds * this is really a simplified "do_mmap". it only handles 25501da177e4SLinus Torvalds * anonymous maps. eventually we may be able to do some 25511da177e4SLinus Torvalds * brk-specific accounting here. 25521da177e4SLinus Torvalds */ 2553e4eb1ff6SLinus Torvalds static unsigned long do_brk(unsigned long addr, unsigned long len) 25541da177e4SLinus Torvalds { 25551da177e4SLinus Torvalds struct mm_struct * mm = current->mm; 25561da177e4SLinus Torvalds struct vm_area_struct * vma, * prev; 25571da177e4SLinus Torvalds unsigned long flags; 25581da177e4SLinus Torvalds struct rb_node ** rb_link, * rb_parent; 25591da177e4SLinus Torvalds pgoff_t pgoff = addr >> PAGE_SHIFT; 25603a459756SKirill Korotaev int error; 25611da177e4SLinus Torvalds 25621da177e4SLinus Torvalds len = PAGE_ALIGN(len); 25631da177e4SLinus Torvalds if (!len) 25641da177e4SLinus Torvalds return addr; 25651da177e4SLinus Torvalds 25663a459756SKirill Korotaev flags = VM_DATA_DEFAULT_FLAGS | VM_ACCOUNT | mm->def_flags; 25673a459756SKirill Korotaev 25682c6a1016SAl Viro error = get_unmapped_area(NULL, addr, len, 0, MAP_FIXED); 25692c6a1016SAl Viro if (error & ~PAGE_MASK) 25703a459756SKirill Korotaev return error; 25713a459756SKirill Korotaev 25721da177e4SLinus Torvalds /* 25731da177e4SLinus Torvalds * mlock MCL_FUTURE? 25741da177e4SLinus Torvalds */ 25751da177e4SLinus Torvalds if (mm->def_flags & VM_LOCKED) { 25761da177e4SLinus Torvalds unsigned long locked, lock_limit; 257793ea1d0aSChris Wright locked = len >> PAGE_SHIFT; 257893ea1d0aSChris Wright locked += mm->locked_vm; 257959e99e5bSJiri Slaby lock_limit = rlimit(RLIMIT_MEMLOCK); 258093ea1d0aSChris Wright lock_limit >>= PAGE_SHIFT; 25811da177e4SLinus Torvalds if (locked > lock_limit && !capable(CAP_IPC_LOCK)) 25821da177e4SLinus Torvalds return -EAGAIN; 25831da177e4SLinus Torvalds } 25841da177e4SLinus Torvalds 25851da177e4SLinus Torvalds /* 25861da177e4SLinus Torvalds * mm->mmap_sem is required to protect against another thread 25871da177e4SLinus Torvalds * changing the mappings in case we sleep. 25881da177e4SLinus Torvalds */ 25891da177e4SLinus Torvalds verify_mm_writelocked(mm); 25901da177e4SLinus Torvalds 25911da177e4SLinus Torvalds /* 25921da177e4SLinus Torvalds * Clear old maps. this also does some error checking for us 25931da177e4SLinus Torvalds */ 25941da177e4SLinus Torvalds munmap_back: 25956597d783SHugh Dickins if (find_vma_links(mm, addr, addr + len, &prev, &rb_link, &rb_parent)) { 25961da177e4SLinus Torvalds if (do_munmap(mm, addr, len)) 25971da177e4SLinus Torvalds return -ENOMEM; 25981da177e4SLinus Torvalds goto munmap_back; 25991da177e4SLinus Torvalds } 26001da177e4SLinus Torvalds 26011da177e4SLinus Torvalds /* Check against address space limits *after* clearing old maps... */ 2602119f657cSakpm@osdl.org if (!may_expand_vm(mm, len >> PAGE_SHIFT)) 26031da177e4SLinus Torvalds return -ENOMEM; 26041da177e4SLinus Torvalds 26051da177e4SLinus Torvalds if (mm->map_count > sysctl_max_map_count) 26061da177e4SLinus Torvalds return -ENOMEM; 26071da177e4SLinus Torvalds 2608191c5424SAl Viro if (security_vm_enough_memory_mm(mm, len >> PAGE_SHIFT)) 26091da177e4SLinus Torvalds return -ENOMEM; 26101da177e4SLinus Torvalds 26111da177e4SLinus Torvalds /* Can we just expand an old private anonymous mapping? */ 2612ba470de4SRik van Riel vma = vma_merge(mm, prev, addr, addr + len, flags, 2613ba470de4SRik van Riel NULL, NULL, pgoff, NULL); 2614ba470de4SRik van Riel if (vma) 26151da177e4SLinus Torvalds goto out; 26161da177e4SLinus Torvalds 26171da177e4SLinus Torvalds /* 26181da177e4SLinus Torvalds * create a vma struct for an anonymous mapping 26191da177e4SLinus Torvalds */ 2620c5e3b83eSPekka Enberg vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL); 26211da177e4SLinus Torvalds if (!vma) { 26221da177e4SLinus Torvalds vm_unacct_memory(len >> PAGE_SHIFT); 26231da177e4SLinus Torvalds return -ENOMEM; 26241da177e4SLinus Torvalds } 26251da177e4SLinus Torvalds 26265beb4930SRik van Riel INIT_LIST_HEAD(&vma->anon_vma_chain); 26271da177e4SLinus Torvalds vma->vm_mm = mm; 26281da177e4SLinus Torvalds vma->vm_start = addr; 26291da177e4SLinus Torvalds vma->vm_end = addr + len; 26301da177e4SLinus Torvalds vma->vm_pgoff = pgoff; 26311da177e4SLinus Torvalds vma->vm_flags = flags; 26323ed75eb8SColy Li vma->vm_page_prot = vm_get_page_prot(flags); 26331da177e4SLinus Torvalds vma_link(mm, vma, prev, rb_link, rb_parent); 26341da177e4SLinus Torvalds out: 26353af9e859SEric B Munson perf_event_mmap(vma); 26361da177e4SLinus Torvalds mm->total_vm += len >> PAGE_SHIFT; 2637128557ffSMichel Lespinasse if (flags & VM_LOCKED) 2638ba470de4SRik van Riel mm->locked_vm += (len >> PAGE_SHIFT); 26391da177e4SLinus Torvalds return addr; 26401da177e4SLinus Torvalds } 26411da177e4SLinus Torvalds 2642e4eb1ff6SLinus Torvalds unsigned long vm_brk(unsigned long addr, unsigned long len) 2643e4eb1ff6SLinus Torvalds { 2644e4eb1ff6SLinus Torvalds struct mm_struct *mm = current->mm; 2645e4eb1ff6SLinus Torvalds unsigned long ret; 2646128557ffSMichel Lespinasse bool populate; 2647e4eb1ff6SLinus Torvalds 2648e4eb1ff6SLinus Torvalds down_write(&mm->mmap_sem); 2649e4eb1ff6SLinus Torvalds ret = do_brk(addr, len); 2650128557ffSMichel Lespinasse populate = ((mm->def_flags & VM_LOCKED) != 0); 2651e4eb1ff6SLinus Torvalds up_write(&mm->mmap_sem); 2652128557ffSMichel Lespinasse if (populate) 2653128557ffSMichel Lespinasse mm_populate(addr, len); 2654e4eb1ff6SLinus Torvalds return ret; 2655e4eb1ff6SLinus Torvalds } 2656e4eb1ff6SLinus Torvalds EXPORT_SYMBOL(vm_brk); 26571da177e4SLinus Torvalds 26581da177e4SLinus Torvalds /* Release all mmaps. */ 26591da177e4SLinus Torvalds void exit_mmap(struct mm_struct *mm) 26601da177e4SLinus Torvalds { 2661d16dfc55SPeter Zijlstra struct mmu_gather tlb; 2662ba470de4SRik van Riel struct vm_area_struct *vma; 26631da177e4SLinus Torvalds unsigned long nr_accounted = 0; 26641da177e4SLinus Torvalds 2665d6dd61c8SJeremy Fitzhardinge /* mm's last user has gone, and its about to be pulled down */ 2666cddb8a5cSAndrea Arcangeli mmu_notifier_release(mm); 2667d6dd61c8SJeremy Fitzhardinge 2668ba470de4SRik van Riel if (mm->locked_vm) { 2669ba470de4SRik van Riel vma = mm->mmap; 2670ba470de4SRik van Riel while (vma) { 2671ba470de4SRik van Riel if (vma->vm_flags & VM_LOCKED) 2672ba470de4SRik van Riel munlock_vma_pages_all(vma); 2673ba470de4SRik van Riel vma = vma->vm_next; 2674ba470de4SRik van Riel } 2675ba470de4SRik van Riel } 26769480c53eSJeremy Fitzhardinge 26779480c53eSJeremy Fitzhardinge arch_exit_mmap(mm); 26789480c53eSJeremy Fitzhardinge 2679ba470de4SRik van Riel vma = mm->mmap; 26809480c53eSJeremy Fitzhardinge if (!vma) /* Can happen if dup_mmap() received an OOM */ 26819480c53eSJeremy Fitzhardinge return; 26829480c53eSJeremy Fitzhardinge 26831da177e4SLinus Torvalds lru_add_drain(); 26841da177e4SLinus Torvalds flush_cache_mm(mm); 2685d16dfc55SPeter Zijlstra tlb_gather_mmu(&tlb, mm, 1); 2686901608d9SOleg Nesterov /* update_hiwater_rss(mm) here? but nobody should be looking */ 2687e0da382cSHugh Dickins /* Use -1 here to ensure all VMAs in the mm are unmapped */ 26884f74d2c8SLinus Torvalds unmap_vmas(&tlb, vma, 0, -1); 26899ba69294SHugh Dickins 26906ee8630eSHugh Dickins free_pgtables(&tlb, vma, FIRST_USER_ADDRESS, USER_PGTABLES_CEILING); 2691853f5e26SAl Viro tlb_finish_mmu(&tlb, 0, -1); 26921da177e4SLinus Torvalds 26931da177e4SLinus Torvalds /* 26948f4f8c16SHugh Dickins * Walk the list again, actually closing and freeing it, 26958f4f8c16SHugh Dickins * with preemption enabled, without holding any MM locks. 26961da177e4SLinus Torvalds */ 26974f74d2c8SLinus Torvalds while (vma) { 26984f74d2c8SLinus Torvalds if (vma->vm_flags & VM_ACCOUNT) 26994f74d2c8SLinus Torvalds nr_accounted += vma_pages(vma); 2700a8fb5618SHugh Dickins vma = remove_vma(vma); 27014f74d2c8SLinus Torvalds } 27024f74d2c8SLinus Torvalds vm_unacct_memory(nr_accounted); 2703e0da382cSHugh Dickins 2704f9aed62aSHugh Dickins WARN_ON(mm->nr_ptes > (FIRST_USER_ADDRESS+PMD_SIZE-1)>>PMD_SHIFT); 27051da177e4SLinus Torvalds } 27061da177e4SLinus Torvalds 27071da177e4SLinus Torvalds /* Insert vm structure into process list sorted by address 27081da177e4SLinus Torvalds * and into the inode's i_mmap tree. If vm_file is non-NULL 27093d48ae45SPeter Zijlstra * then i_mmap_mutex is taken here. 27101da177e4SLinus Torvalds */ 27111da177e4SLinus Torvalds int insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma) 27121da177e4SLinus Torvalds { 27136597d783SHugh Dickins struct vm_area_struct *prev; 27141da177e4SLinus Torvalds struct rb_node **rb_link, *rb_parent; 27151da177e4SLinus Torvalds 27161da177e4SLinus Torvalds /* 27171da177e4SLinus Torvalds * The vm_pgoff of a purely anonymous vma should be irrelevant 27181da177e4SLinus Torvalds * until its first write fault, when page's anon_vma and index 27191da177e4SLinus Torvalds * are set. But now set the vm_pgoff it will almost certainly 27201da177e4SLinus Torvalds * end up with (unless mremap moves it elsewhere before that 27211da177e4SLinus Torvalds * first wfault), so /proc/pid/maps tells a consistent story. 27221da177e4SLinus Torvalds * 27231da177e4SLinus Torvalds * By setting it to reflect the virtual start address of the 27241da177e4SLinus Torvalds * vma, merges and splits can happen in a seamless way, just 27251da177e4SLinus Torvalds * using the existing file pgoff checks and manipulations. 27261da177e4SLinus Torvalds * Similarly in do_mmap_pgoff and in do_brk. 27271da177e4SLinus Torvalds */ 27281da177e4SLinus Torvalds if (!vma->vm_file) { 27291da177e4SLinus Torvalds BUG_ON(vma->anon_vma); 27301da177e4SLinus Torvalds vma->vm_pgoff = vma->vm_start >> PAGE_SHIFT; 27311da177e4SLinus Torvalds } 27326597d783SHugh Dickins if (find_vma_links(mm, vma->vm_start, vma->vm_end, 27336597d783SHugh Dickins &prev, &rb_link, &rb_parent)) 27341da177e4SLinus Torvalds return -ENOMEM; 27352fd4ef85SHugh Dickins if ((vma->vm_flags & VM_ACCOUNT) && 273634b4e4aaSAlan Cox security_vm_enough_memory_mm(mm, vma_pages(vma))) 27372fd4ef85SHugh Dickins return -ENOMEM; 27382b144498SSrikar Dronamraju 27391da177e4SLinus Torvalds vma_link(mm, vma, prev, rb_link, rb_parent); 27401da177e4SLinus Torvalds return 0; 27411da177e4SLinus Torvalds } 27421da177e4SLinus Torvalds 27431da177e4SLinus Torvalds /* 27441da177e4SLinus Torvalds * Copy the vma structure to a new location in the same mm, 27451da177e4SLinus Torvalds * prior to moving page table entries, to effect an mremap move. 27461da177e4SLinus Torvalds */ 27471da177e4SLinus Torvalds struct vm_area_struct *copy_vma(struct vm_area_struct **vmap, 274838a76013SMichel Lespinasse unsigned long addr, unsigned long len, pgoff_t pgoff, 274938a76013SMichel Lespinasse bool *need_rmap_locks) 27501da177e4SLinus Torvalds { 27511da177e4SLinus Torvalds struct vm_area_struct *vma = *vmap; 27521da177e4SLinus Torvalds unsigned long vma_start = vma->vm_start; 27531da177e4SLinus Torvalds struct mm_struct *mm = vma->vm_mm; 27541da177e4SLinus Torvalds struct vm_area_struct *new_vma, *prev; 27551da177e4SLinus Torvalds struct rb_node **rb_link, *rb_parent; 27561da177e4SLinus Torvalds struct mempolicy *pol; 2757948f017bSAndrea Arcangeli bool faulted_in_anon_vma = true; 27581da177e4SLinus Torvalds 27591da177e4SLinus Torvalds /* 27601da177e4SLinus Torvalds * If anonymous vma has not yet been faulted, update new pgoff 27611da177e4SLinus Torvalds * to match new location, to increase its chance of merging. 27621da177e4SLinus Torvalds */ 2763948f017bSAndrea Arcangeli if (unlikely(!vma->vm_file && !vma->anon_vma)) { 27641da177e4SLinus Torvalds pgoff = addr >> PAGE_SHIFT; 2765948f017bSAndrea Arcangeli faulted_in_anon_vma = false; 2766948f017bSAndrea Arcangeli } 27671da177e4SLinus Torvalds 27686597d783SHugh Dickins if (find_vma_links(mm, addr, addr + len, &prev, &rb_link, &rb_parent)) 27696597d783SHugh Dickins return NULL; /* should never get here */ 27701da177e4SLinus Torvalds new_vma = vma_merge(mm, prev, addr, addr + len, vma->vm_flags, 27711da177e4SLinus Torvalds vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma)); 27721da177e4SLinus Torvalds if (new_vma) { 27731da177e4SLinus Torvalds /* 27741da177e4SLinus Torvalds * Source vma may have been merged into new_vma 27751da177e4SLinus Torvalds */ 2776948f017bSAndrea Arcangeli if (unlikely(vma_start >= new_vma->vm_start && 2777948f017bSAndrea Arcangeli vma_start < new_vma->vm_end)) { 2778948f017bSAndrea Arcangeli /* 2779948f017bSAndrea Arcangeli * The only way we can get a vma_merge with 2780948f017bSAndrea Arcangeli * self during an mremap is if the vma hasn't 2781948f017bSAndrea Arcangeli * been faulted in yet and we were allowed to 2782948f017bSAndrea Arcangeli * reset the dst vma->vm_pgoff to the 2783948f017bSAndrea Arcangeli * destination address of the mremap to allow 2784948f017bSAndrea Arcangeli * the merge to happen. mremap must change the 2785948f017bSAndrea Arcangeli * vm_pgoff linearity between src and dst vmas 2786948f017bSAndrea Arcangeli * (in turn preventing a vma_merge) to be 2787948f017bSAndrea Arcangeli * safe. It is only safe to keep the vm_pgoff 2788948f017bSAndrea Arcangeli * linear if there are no pages mapped yet. 2789948f017bSAndrea Arcangeli */ 2790948f017bSAndrea Arcangeli VM_BUG_ON(faulted_in_anon_vma); 279138a76013SMichel Lespinasse *vmap = vma = new_vma; 2792108d6642SMichel Lespinasse } 279338a76013SMichel Lespinasse *need_rmap_locks = (new_vma->vm_pgoff <= vma->vm_pgoff); 27941da177e4SLinus Torvalds } else { 2795e94b1766SChristoph Lameter new_vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL); 27961da177e4SLinus Torvalds if (new_vma) { 27971da177e4SLinus Torvalds *new_vma = *vma; 27981da177e4SLinus Torvalds new_vma->vm_start = addr; 27991da177e4SLinus Torvalds new_vma->vm_end = addr + len; 28001da177e4SLinus Torvalds new_vma->vm_pgoff = pgoff; 2801523d4e20SMichel Lespinasse pol = mpol_dup(vma_policy(vma)); 2802523d4e20SMichel Lespinasse if (IS_ERR(pol)) 2803523d4e20SMichel Lespinasse goto out_free_vma; 2804523d4e20SMichel Lespinasse vma_set_policy(new_vma, pol); 2805523d4e20SMichel Lespinasse INIT_LIST_HEAD(&new_vma->anon_vma_chain); 2806523d4e20SMichel Lespinasse if (anon_vma_clone(new_vma, vma)) 2807523d4e20SMichel Lespinasse goto out_free_mempol; 2808e9714acfSKonstantin Khlebnikov if (new_vma->vm_file) 28091da177e4SLinus Torvalds get_file(new_vma->vm_file); 28101da177e4SLinus Torvalds if (new_vma->vm_ops && new_vma->vm_ops->open) 28111da177e4SLinus Torvalds new_vma->vm_ops->open(new_vma); 28121da177e4SLinus Torvalds vma_link(mm, new_vma, prev, rb_link, rb_parent); 281338a76013SMichel Lespinasse *need_rmap_locks = false; 28141da177e4SLinus Torvalds } 28151da177e4SLinus Torvalds } 28161da177e4SLinus Torvalds return new_vma; 28175beb4930SRik van Riel 28185beb4930SRik van Riel out_free_mempol: 28195beb4930SRik van Riel mpol_put(pol); 28205beb4930SRik van Riel out_free_vma: 28215beb4930SRik van Riel kmem_cache_free(vm_area_cachep, new_vma); 28225beb4930SRik van Riel return NULL; 28231da177e4SLinus Torvalds } 2824119f657cSakpm@osdl.org 2825119f657cSakpm@osdl.org /* 2826119f657cSakpm@osdl.org * Return true if the calling process may expand its vm space by the passed 2827119f657cSakpm@osdl.org * number of pages 2828119f657cSakpm@osdl.org */ 2829119f657cSakpm@osdl.org int may_expand_vm(struct mm_struct *mm, unsigned long npages) 2830119f657cSakpm@osdl.org { 2831119f657cSakpm@osdl.org unsigned long cur = mm->total_vm; /* pages */ 2832119f657cSakpm@osdl.org unsigned long lim; 2833119f657cSakpm@osdl.org 283459e99e5bSJiri Slaby lim = rlimit(RLIMIT_AS) >> PAGE_SHIFT; 2835119f657cSakpm@osdl.org 2836119f657cSakpm@osdl.org if (cur + npages > lim) 2837119f657cSakpm@osdl.org return 0; 2838119f657cSakpm@osdl.org return 1; 2839119f657cSakpm@osdl.org } 2840fa5dc22fSRoland McGrath 2841fa5dc22fSRoland McGrath 2842b1d0e4f5SNick Piggin static int special_mapping_fault(struct vm_area_struct *vma, 2843b1d0e4f5SNick Piggin struct vm_fault *vmf) 2844fa5dc22fSRoland McGrath { 2845b1d0e4f5SNick Piggin pgoff_t pgoff; 2846fa5dc22fSRoland McGrath struct page **pages; 2847fa5dc22fSRoland McGrath 2848b1d0e4f5SNick Piggin /* 2849b1d0e4f5SNick Piggin * special mappings have no vm_file, and in that case, the mm 2850b1d0e4f5SNick Piggin * uses vm_pgoff internally. So we have to subtract it from here. 2851b1d0e4f5SNick Piggin * We are allowed to do this because we are the mm; do not copy 2852b1d0e4f5SNick Piggin * this code into drivers! 2853b1d0e4f5SNick Piggin */ 2854b1d0e4f5SNick Piggin pgoff = vmf->pgoff - vma->vm_pgoff; 2855fa5dc22fSRoland McGrath 2856b1d0e4f5SNick Piggin for (pages = vma->vm_private_data; pgoff && *pages; ++pages) 2857b1d0e4f5SNick Piggin pgoff--; 2858fa5dc22fSRoland McGrath 2859fa5dc22fSRoland McGrath if (*pages) { 2860fa5dc22fSRoland McGrath struct page *page = *pages; 2861fa5dc22fSRoland McGrath get_page(page); 2862b1d0e4f5SNick Piggin vmf->page = page; 2863b1d0e4f5SNick Piggin return 0; 2864fa5dc22fSRoland McGrath } 2865fa5dc22fSRoland McGrath 2866b1d0e4f5SNick Piggin return VM_FAULT_SIGBUS; 2867fa5dc22fSRoland McGrath } 2868fa5dc22fSRoland McGrath 2869fa5dc22fSRoland McGrath /* 2870fa5dc22fSRoland McGrath * Having a close hook prevents vma merging regardless of flags. 2871fa5dc22fSRoland McGrath */ 2872fa5dc22fSRoland McGrath static void special_mapping_close(struct vm_area_struct *vma) 2873fa5dc22fSRoland McGrath { 2874fa5dc22fSRoland McGrath } 2875fa5dc22fSRoland McGrath 2876f0f37e2fSAlexey Dobriyan static const struct vm_operations_struct special_mapping_vmops = { 2877fa5dc22fSRoland McGrath .close = special_mapping_close, 2878b1d0e4f5SNick Piggin .fault = special_mapping_fault, 2879fa5dc22fSRoland McGrath }; 2880fa5dc22fSRoland McGrath 2881fa5dc22fSRoland McGrath /* 2882fa5dc22fSRoland McGrath * Called with mm->mmap_sem held for writing. 2883fa5dc22fSRoland McGrath * Insert a new vma covering the given region, with the given flags. 2884fa5dc22fSRoland McGrath * Its pages are supplied by the given array of struct page *. 2885fa5dc22fSRoland McGrath * The array can be shorter than len >> PAGE_SHIFT if it's null-terminated. 2886fa5dc22fSRoland McGrath * The region past the last page supplied will always produce SIGBUS. 2887fa5dc22fSRoland McGrath * The array pointer and the pages it points to are assumed to stay alive 2888fa5dc22fSRoland McGrath * for as long as this mapping might exist. 2889fa5dc22fSRoland McGrath */ 2890fa5dc22fSRoland McGrath int install_special_mapping(struct mm_struct *mm, 2891fa5dc22fSRoland McGrath unsigned long addr, unsigned long len, 2892fa5dc22fSRoland McGrath unsigned long vm_flags, struct page **pages) 2893fa5dc22fSRoland McGrath { 2894462e635eSTavis Ormandy int ret; 2895fa5dc22fSRoland McGrath struct vm_area_struct *vma; 2896fa5dc22fSRoland McGrath 2897fa5dc22fSRoland McGrath vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL); 2898fa5dc22fSRoland McGrath if (unlikely(vma == NULL)) 2899fa5dc22fSRoland McGrath return -ENOMEM; 2900fa5dc22fSRoland McGrath 29015beb4930SRik van Riel INIT_LIST_HEAD(&vma->anon_vma_chain); 2902fa5dc22fSRoland McGrath vma->vm_mm = mm; 2903fa5dc22fSRoland McGrath vma->vm_start = addr; 2904fa5dc22fSRoland McGrath vma->vm_end = addr + len; 2905fa5dc22fSRoland McGrath 29062f98735cSNick Piggin vma->vm_flags = vm_flags | mm->def_flags | VM_DONTEXPAND; 29073ed75eb8SColy Li vma->vm_page_prot = vm_get_page_prot(vma->vm_flags); 2908fa5dc22fSRoland McGrath 2909fa5dc22fSRoland McGrath vma->vm_ops = &special_mapping_vmops; 2910fa5dc22fSRoland McGrath vma->vm_private_data = pages; 2911fa5dc22fSRoland McGrath 2912462e635eSTavis Ormandy ret = insert_vm_struct(mm, vma); 2913462e635eSTavis Ormandy if (ret) 2914462e635eSTavis Ormandy goto out; 2915fa5dc22fSRoland McGrath 2916fa5dc22fSRoland McGrath mm->total_vm += len >> PAGE_SHIFT; 2917fa5dc22fSRoland McGrath 2918cdd6c482SIngo Molnar perf_event_mmap(vma); 2919089dd79dSPeter Zijlstra 2920fa5dc22fSRoland McGrath return 0; 2921462e635eSTavis Ormandy 2922462e635eSTavis Ormandy out: 2923462e635eSTavis Ormandy kmem_cache_free(vm_area_cachep, vma); 2924462e635eSTavis Ormandy return ret; 2925fa5dc22fSRoland McGrath } 29267906d00cSAndrea Arcangeli 29277906d00cSAndrea Arcangeli static DEFINE_MUTEX(mm_all_locks_mutex); 29287906d00cSAndrea Arcangeli 2929454ed842SPeter Zijlstra static void vm_lock_anon_vma(struct mm_struct *mm, struct anon_vma *anon_vma) 29307906d00cSAndrea Arcangeli { 2931bf181b9fSMichel Lespinasse if (!test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_node)) { 29327906d00cSAndrea Arcangeli /* 29337906d00cSAndrea Arcangeli * The LSB of head.next can't change from under us 29347906d00cSAndrea Arcangeli * because we hold the mm_all_locks_mutex. 29357906d00cSAndrea Arcangeli */ 2936572043c9SJiri Kosina down_write_nest_lock(&anon_vma->root->rwsem, &mm->mmap_sem); 29377906d00cSAndrea Arcangeli /* 29387906d00cSAndrea Arcangeli * We can safely modify head.next after taking the 29395a505085SIngo Molnar * anon_vma->root->rwsem. If some other vma in this mm shares 29407906d00cSAndrea Arcangeli * the same anon_vma we won't take it again. 29417906d00cSAndrea Arcangeli * 29427906d00cSAndrea Arcangeli * No need of atomic instructions here, head.next 29437906d00cSAndrea Arcangeli * can't change from under us thanks to the 29445a505085SIngo Molnar * anon_vma->root->rwsem. 29457906d00cSAndrea Arcangeli */ 29467906d00cSAndrea Arcangeli if (__test_and_set_bit(0, (unsigned long *) 2947bf181b9fSMichel Lespinasse &anon_vma->root->rb_root.rb_node)) 29487906d00cSAndrea Arcangeli BUG(); 29497906d00cSAndrea Arcangeli } 29507906d00cSAndrea Arcangeli } 29517906d00cSAndrea Arcangeli 2952454ed842SPeter Zijlstra static void vm_lock_mapping(struct mm_struct *mm, struct address_space *mapping) 29537906d00cSAndrea Arcangeli { 29547906d00cSAndrea Arcangeli if (!test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) { 29557906d00cSAndrea Arcangeli /* 29567906d00cSAndrea Arcangeli * AS_MM_ALL_LOCKS can't change from under us because 29577906d00cSAndrea Arcangeli * we hold the mm_all_locks_mutex. 29587906d00cSAndrea Arcangeli * 29597906d00cSAndrea Arcangeli * Operations on ->flags have to be atomic because 29607906d00cSAndrea Arcangeli * even if AS_MM_ALL_LOCKS is stable thanks to the 29617906d00cSAndrea Arcangeli * mm_all_locks_mutex, there may be other cpus 29627906d00cSAndrea Arcangeli * changing other bitflags in parallel to us. 29637906d00cSAndrea Arcangeli */ 29647906d00cSAndrea Arcangeli if (test_and_set_bit(AS_MM_ALL_LOCKS, &mapping->flags)) 29657906d00cSAndrea Arcangeli BUG(); 29663d48ae45SPeter Zijlstra mutex_lock_nest_lock(&mapping->i_mmap_mutex, &mm->mmap_sem); 29677906d00cSAndrea Arcangeli } 29687906d00cSAndrea Arcangeli } 29697906d00cSAndrea Arcangeli 29707906d00cSAndrea Arcangeli /* 29717906d00cSAndrea Arcangeli * This operation locks against the VM for all pte/vma/mm related 29727906d00cSAndrea Arcangeli * operations that could ever happen on a certain mm. This includes 29737906d00cSAndrea Arcangeli * vmtruncate, try_to_unmap, and all page faults. 29747906d00cSAndrea Arcangeli * 29757906d00cSAndrea Arcangeli * The caller must take the mmap_sem in write mode before calling 29767906d00cSAndrea Arcangeli * mm_take_all_locks(). The caller isn't allowed to release the 29777906d00cSAndrea Arcangeli * mmap_sem until mm_drop_all_locks() returns. 29787906d00cSAndrea Arcangeli * 29797906d00cSAndrea Arcangeli * mmap_sem in write mode is required in order to block all operations 29807906d00cSAndrea Arcangeli * that could modify pagetables and free pages without need of 29817906d00cSAndrea Arcangeli * altering the vma layout (for example populate_range() with 29827906d00cSAndrea Arcangeli * nonlinear vmas). It's also needed in write mode to avoid new 29837906d00cSAndrea Arcangeli * anon_vmas to be associated with existing vmas. 29847906d00cSAndrea Arcangeli * 29857906d00cSAndrea Arcangeli * A single task can't take more than one mm_take_all_locks() in a row 29867906d00cSAndrea Arcangeli * or it would deadlock. 29877906d00cSAndrea Arcangeli * 2988bf181b9fSMichel Lespinasse * The LSB in anon_vma->rb_root.rb_node and the AS_MM_ALL_LOCKS bitflag in 29897906d00cSAndrea Arcangeli * mapping->flags avoid to take the same lock twice, if more than one 29907906d00cSAndrea Arcangeli * vma in this mm is backed by the same anon_vma or address_space. 29917906d00cSAndrea Arcangeli * 29927906d00cSAndrea Arcangeli * We can take all the locks in random order because the VM code 2993631b0cfdSYuanhan Liu * taking i_mmap_mutex or anon_vma->rwsem outside the mmap_sem never 29947906d00cSAndrea Arcangeli * takes more than one of them in a row. Secondly we're protected 29957906d00cSAndrea Arcangeli * against a concurrent mm_take_all_locks() by the mm_all_locks_mutex. 29967906d00cSAndrea Arcangeli * 29977906d00cSAndrea Arcangeli * mm_take_all_locks() and mm_drop_all_locks are expensive operations 29987906d00cSAndrea Arcangeli * that may have to take thousand of locks. 29997906d00cSAndrea Arcangeli * 30007906d00cSAndrea Arcangeli * mm_take_all_locks() can fail if it's interrupted by signals. 30017906d00cSAndrea Arcangeli */ 30027906d00cSAndrea Arcangeli int mm_take_all_locks(struct mm_struct *mm) 30037906d00cSAndrea Arcangeli { 30047906d00cSAndrea Arcangeli struct vm_area_struct *vma; 30055beb4930SRik van Riel struct anon_vma_chain *avc; 30067906d00cSAndrea Arcangeli 30077906d00cSAndrea Arcangeli BUG_ON(down_read_trylock(&mm->mmap_sem)); 30087906d00cSAndrea Arcangeli 30097906d00cSAndrea Arcangeli mutex_lock(&mm_all_locks_mutex); 30107906d00cSAndrea Arcangeli 30117906d00cSAndrea Arcangeli for (vma = mm->mmap; vma; vma = vma->vm_next) { 30127906d00cSAndrea Arcangeli if (signal_pending(current)) 30137906d00cSAndrea Arcangeli goto out_unlock; 30147906d00cSAndrea Arcangeli if (vma->vm_file && vma->vm_file->f_mapping) 3015454ed842SPeter Zijlstra vm_lock_mapping(mm, vma->vm_file->f_mapping); 30167906d00cSAndrea Arcangeli } 30177cd5a02fSPeter Zijlstra 30187cd5a02fSPeter Zijlstra for (vma = mm->mmap; vma; vma = vma->vm_next) { 30197cd5a02fSPeter Zijlstra if (signal_pending(current)) 30207cd5a02fSPeter Zijlstra goto out_unlock; 30217cd5a02fSPeter Zijlstra if (vma->anon_vma) 30225beb4930SRik van Riel list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) 30235beb4930SRik van Riel vm_lock_anon_vma(mm, avc->anon_vma); 30247cd5a02fSPeter Zijlstra } 30257cd5a02fSPeter Zijlstra 3026584cff54SKautuk Consul return 0; 30277906d00cSAndrea Arcangeli 30287906d00cSAndrea Arcangeli out_unlock: 30297906d00cSAndrea Arcangeli mm_drop_all_locks(mm); 3030584cff54SKautuk Consul return -EINTR; 30317906d00cSAndrea Arcangeli } 30327906d00cSAndrea Arcangeli 30337906d00cSAndrea Arcangeli static void vm_unlock_anon_vma(struct anon_vma *anon_vma) 30347906d00cSAndrea Arcangeli { 3035bf181b9fSMichel Lespinasse if (test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_node)) { 30367906d00cSAndrea Arcangeli /* 30377906d00cSAndrea Arcangeli * The LSB of head.next can't change to 0 from under 30387906d00cSAndrea Arcangeli * us because we hold the mm_all_locks_mutex. 30397906d00cSAndrea Arcangeli * 30407906d00cSAndrea Arcangeli * We must however clear the bitflag before unlocking 3041bf181b9fSMichel Lespinasse * the vma so the users using the anon_vma->rb_root will 30427906d00cSAndrea Arcangeli * never see our bitflag. 30437906d00cSAndrea Arcangeli * 30447906d00cSAndrea Arcangeli * No need of atomic instructions here, head.next 30457906d00cSAndrea Arcangeli * can't change from under us until we release the 30465a505085SIngo Molnar * anon_vma->root->rwsem. 30477906d00cSAndrea Arcangeli */ 30487906d00cSAndrea Arcangeli if (!__test_and_clear_bit(0, (unsigned long *) 3049bf181b9fSMichel Lespinasse &anon_vma->root->rb_root.rb_node)) 30507906d00cSAndrea Arcangeli BUG(); 305108b52706SKonstantin Khlebnikov anon_vma_unlock_write(anon_vma); 30527906d00cSAndrea Arcangeli } 30537906d00cSAndrea Arcangeli } 30547906d00cSAndrea Arcangeli 30557906d00cSAndrea Arcangeli static void vm_unlock_mapping(struct address_space *mapping) 30567906d00cSAndrea Arcangeli { 30577906d00cSAndrea Arcangeli if (test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) { 30587906d00cSAndrea Arcangeli /* 30597906d00cSAndrea Arcangeli * AS_MM_ALL_LOCKS can't change to 0 from under us 30607906d00cSAndrea Arcangeli * because we hold the mm_all_locks_mutex. 30617906d00cSAndrea Arcangeli */ 30623d48ae45SPeter Zijlstra mutex_unlock(&mapping->i_mmap_mutex); 30637906d00cSAndrea Arcangeli if (!test_and_clear_bit(AS_MM_ALL_LOCKS, 30647906d00cSAndrea Arcangeli &mapping->flags)) 30657906d00cSAndrea Arcangeli BUG(); 30667906d00cSAndrea Arcangeli } 30677906d00cSAndrea Arcangeli } 30687906d00cSAndrea Arcangeli 30697906d00cSAndrea Arcangeli /* 30707906d00cSAndrea Arcangeli * The mmap_sem cannot be released by the caller until 30717906d00cSAndrea Arcangeli * mm_drop_all_locks() returns. 30727906d00cSAndrea Arcangeli */ 30737906d00cSAndrea Arcangeli void mm_drop_all_locks(struct mm_struct *mm) 30747906d00cSAndrea Arcangeli { 30757906d00cSAndrea Arcangeli struct vm_area_struct *vma; 30765beb4930SRik van Riel struct anon_vma_chain *avc; 30777906d00cSAndrea Arcangeli 30787906d00cSAndrea Arcangeli BUG_ON(down_read_trylock(&mm->mmap_sem)); 30797906d00cSAndrea Arcangeli BUG_ON(!mutex_is_locked(&mm_all_locks_mutex)); 30807906d00cSAndrea Arcangeli 30817906d00cSAndrea Arcangeli for (vma = mm->mmap; vma; vma = vma->vm_next) { 30827906d00cSAndrea Arcangeli if (vma->anon_vma) 30835beb4930SRik van Riel list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) 30845beb4930SRik van Riel vm_unlock_anon_vma(avc->anon_vma); 30857906d00cSAndrea Arcangeli if (vma->vm_file && vma->vm_file->f_mapping) 30867906d00cSAndrea Arcangeli vm_unlock_mapping(vma->vm_file->f_mapping); 30877906d00cSAndrea Arcangeli } 30887906d00cSAndrea Arcangeli 30897906d00cSAndrea Arcangeli mutex_unlock(&mm_all_locks_mutex); 30907906d00cSAndrea Arcangeli } 30918feae131SDavid Howells 30928feae131SDavid Howells /* 30938feae131SDavid Howells * initialise the VMA slab 30948feae131SDavid Howells */ 30958feae131SDavid Howells void __init mmap_init(void) 30968feae131SDavid Howells { 309700a62ce9SKOSAKI Motohiro int ret; 309800a62ce9SKOSAKI Motohiro 309900a62ce9SKOSAKI Motohiro ret = percpu_counter_init(&vm_committed_as, 0); 310000a62ce9SKOSAKI Motohiro VM_BUG_ON(ret); 31018feae131SDavid Howells } 3102c9b1d098SAndrew Shewmaker 3103c9b1d098SAndrew Shewmaker /* 3104c9b1d098SAndrew Shewmaker * Initialise sysctl_user_reserve_kbytes. 3105c9b1d098SAndrew Shewmaker * 3106c9b1d098SAndrew Shewmaker * This is intended to prevent a user from starting a single memory hogging 3107c9b1d098SAndrew Shewmaker * process, such that they cannot recover (kill the hog) in OVERCOMMIT_NEVER 3108c9b1d098SAndrew Shewmaker * mode. 3109c9b1d098SAndrew Shewmaker * 3110c9b1d098SAndrew Shewmaker * The default value is min(3% of free memory, 128MB) 3111c9b1d098SAndrew Shewmaker * 128MB is enough to recover with sshd/login, bash, and top/kill. 3112c9b1d098SAndrew Shewmaker */ 3113c9b1d098SAndrew Shewmaker static int __meminit init_user_reserve(void) 3114c9b1d098SAndrew Shewmaker { 3115c9b1d098SAndrew Shewmaker unsigned long free_kbytes; 3116c9b1d098SAndrew Shewmaker 3117c9b1d098SAndrew Shewmaker free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10); 3118c9b1d098SAndrew Shewmaker 3119c9b1d098SAndrew Shewmaker sysctl_user_reserve_kbytes = min(free_kbytes / 32, 1UL << 17); 3120c9b1d098SAndrew Shewmaker return 0; 3121c9b1d098SAndrew Shewmaker } 3122c9b1d098SAndrew Shewmaker module_init(init_user_reserve) 3123*4eeab4f5SAndrew Shewmaker 3124*4eeab4f5SAndrew Shewmaker /* 3125*4eeab4f5SAndrew Shewmaker * Initialise sysctl_admin_reserve_kbytes. 3126*4eeab4f5SAndrew Shewmaker * 3127*4eeab4f5SAndrew Shewmaker * The purpose of sysctl_admin_reserve_kbytes is to allow the sys admin 3128*4eeab4f5SAndrew Shewmaker * to log in and kill a memory hogging process. 3129*4eeab4f5SAndrew Shewmaker * 3130*4eeab4f5SAndrew Shewmaker * Systems with more than 256MB will reserve 8MB, enough to recover 3131*4eeab4f5SAndrew Shewmaker * with sshd, bash, and top in OVERCOMMIT_GUESS. Smaller systems will 3132*4eeab4f5SAndrew Shewmaker * only reserve 3% of free pages by default. 3133*4eeab4f5SAndrew Shewmaker */ 3134*4eeab4f5SAndrew Shewmaker static int __meminit init_admin_reserve(void) 3135*4eeab4f5SAndrew Shewmaker { 3136*4eeab4f5SAndrew Shewmaker unsigned long free_kbytes; 3137*4eeab4f5SAndrew Shewmaker 3138*4eeab4f5SAndrew Shewmaker free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10); 3139*4eeab4f5SAndrew Shewmaker 3140*4eeab4f5SAndrew Shewmaker sysctl_admin_reserve_kbytes = min(free_kbytes / 32, 1UL << 13); 3141*4eeab4f5SAndrew Shewmaker return 0; 3142*4eeab4f5SAndrew Shewmaker } 3143*4eeab4f5SAndrew Shewmaker module_init(init_admin_reserve) 3144