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 9b1de0d13SMitchel Humpherys #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt 10b1de0d13SMitchel Humpherys 11e8420a8eSCyril Hrubis #include <linux/kernel.h> 121da177e4SLinus Torvalds #include <linux/slab.h> 134af3c9ccSAlexey Dobriyan #include <linux/backing-dev.h> 141da177e4SLinus Torvalds #include <linux/mm.h> 15615d6e87SDavidlohr Bueso #include <linux/vmacache.h> 161da177e4SLinus Torvalds #include <linux/shm.h> 171da177e4SLinus Torvalds #include <linux/mman.h> 181da177e4SLinus Torvalds #include <linux/pagemap.h> 191da177e4SLinus Torvalds #include <linux/swap.h> 201da177e4SLinus Torvalds #include <linux/syscalls.h> 21c59ede7bSRandy.Dunlap #include <linux/capability.h> 221da177e4SLinus Torvalds #include <linux/init.h> 231da177e4SLinus Torvalds #include <linux/file.h> 241da177e4SLinus Torvalds #include <linux/fs.h> 251da177e4SLinus Torvalds #include <linux/personality.h> 261da177e4SLinus Torvalds #include <linux/security.h> 271da177e4SLinus Torvalds #include <linux/hugetlb.h> 281da177e4SLinus Torvalds #include <linux/profile.h> 29b95f1b31SPaul Gortmaker #include <linux/export.h> 301da177e4SLinus Torvalds #include <linux/mount.h> 311da177e4SLinus Torvalds #include <linux/mempolicy.h> 321da177e4SLinus Torvalds #include <linux/rmap.h> 33cddb8a5cSAndrea Arcangeli #include <linux/mmu_notifier.h> 3482f71ae4SKonstantin Khlebnikov #include <linux/mmdebug.h> 35cdd6c482SIngo Molnar #include <linux/perf_event.h> 36120a795dSAl Viro #include <linux/audit.h> 37b15d00b6SAndrea Arcangeli #include <linux/khugepaged.h> 382b144498SSrikar Dronamraju #include <linux/uprobes.h> 39d3737187SMichel Lespinasse #include <linux/rbtree_augmented.h> 40cf4aebc2SClark Williams #include <linux/sched/sysctl.h> 411640879aSAndrew Shewmaker #include <linux/notifier.h> 421640879aSAndrew Shewmaker #include <linux/memory.h> 43b1de0d13SMitchel Humpherys #include <linux/printk.h> 441da177e4SLinus Torvalds 451da177e4SLinus Torvalds #include <asm/uaccess.h> 461da177e4SLinus Torvalds #include <asm/cacheflush.h> 471da177e4SLinus Torvalds #include <asm/tlb.h> 48d6dd61c8SJeremy Fitzhardinge #include <asm/mmu_context.h> 491da177e4SLinus Torvalds 5042b77728SJan Beulich #include "internal.h" 5142b77728SJan Beulich 523a459756SKirill Korotaev #ifndef arch_mmap_check 533a459756SKirill Korotaev #define arch_mmap_check(addr, len, flags) (0) 543a459756SKirill Korotaev #endif 553a459756SKirill Korotaev 5608e7d9b5SMartin Schwidefsky #ifndef arch_rebalance_pgtables 5708e7d9b5SMartin Schwidefsky #define arch_rebalance_pgtables(addr, len) (addr) 5808e7d9b5SMartin Schwidefsky #endif 5908e7d9b5SMartin Schwidefsky 60e0da382cSHugh Dickins static void unmap_region(struct mm_struct *mm, 61e0da382cSHugh Dickins struct vm_area_struct *vma, struct vm_area_struct *prev, 62e0da382cSHugh Dickins unsigned long start, unsigned long end); 63e0da382cSHugh Dickins 641da177e4SLinus Torvalds /* description of effects of mapping type and prot in current implementation. 651da177e4SLinus Torvalds * this is due to the limited x86 page protection hardware. The expected 661da177e4SLinus Torvalds * behavior is in parens: 671da177e4SLinus Torvalds * 681da177e4SLinus Torvalds * map_type prot 691da177e4SLinus Torvalds * PROT_NONE PROT_READ PROT_WRITE PROT_EXEC 701da177e4SLinus Torvalds * MAP_SHARED r: (no) no r: (yes) yes r: (no) yes r: (no) yes 711da177e4SLinus Torvalds * w: (no) no w: (no) no w: (yes) yes w: (no) no 721da177e4SLinus Torvalds * x: (no) no x: (no) yes x: (no) yes x: (yes) yes 731da177e4SLinus Torvalds * 741da177e4SLinus Torvalds * MAP_PRIVATE r: (no) no r: (yes) yes r: (no) yes r: (no) yes 751da177e4SLinus Torvalds * w: (no) no w: (no) no w: (copy) copy w: (no) no 761da177e4SLinus Torvalds * x: (no) no x: (no) yes x: (no) yes x: (yes) yes 771da177e4SLinus Torvalds * 781da177e4SLinus Torvalds */ 791da177e4SLinus Torvalds pgprot_t protection_map[16] = { 801da177e4SLinus Torvalds __P000, __P001, __P010, __P011, __P100, __P101, __P110, __P111, 811da177e4SLinus Torvalds __S000, __S001, __S010, __S011, __S100, __S101, __S110, __S111 821da177e4SLinus Torvalds }; 831da177e4SLinus Torvalds 84804af2cfSHugh Dickins pgprot_t vm_get_page_prot(unsigned long vm_flags) 85804af2cfSHugh Dickins { 86b845f313SDave Kleikamp return __pgprot(pgprot_val(protection_map[vm_flags & 87b845f313SDave Kleikamp (VM_READ|VM_WRITE|VM_EXEC|VM_SHARED)]) | 88b845f313SDave Kleikamp pgprot_val(arch_vm_get_page_prot(vm_flags))); 89804af2cfSHugh Dickins } 90804af2cfSHugh Dickins EXPORT_SYMBOL(vm_get_page_prot); 91804af2cfSHugh Dickins 9264e45507SPeter Feiner static pgprot_t vm_pgprot_modify(pgprot_t oldprot, unsigned long vm_flags) 9364e45507SPeter Feiner { 9464e45507SPeter Feiner return pgprot_modify(oldprot, vm_get_page_prot(vm_flags)); 9564e45507SPeter Feiner } 9664e45507SPeter Feiner 9764e45507SPeter Feiner /* Update vma->vm_page_prot to reflect vma->vm_flags. */ 9864e45507SPeter Feiner void vma_set_page_prot(struct vm_area_struct *vma) 9964e45507SPeter Feiner { 10064e45507SPeter Feiner unsigned long vm_flags = vma->vm_flags; 10164e45507SPeter Feiner 10264e45507SPeter Feiner vma->vm_page_prot = vm_pgprot_modify(vma->vm_page_prot, vm_flags); 10364e45507SPeter Feiner if (vma_wants_writenotify(vma)) { 10464e45507SPeter Feiner vm_flags &= ~VM_SHARED; 10564e45507SPeter Feiner vma->vm_page_prot = vm_pgprot_modify(vma->vm_page_prot, 10664e45507SPeter Feiner vm_flags); 10764e45507SPeter Feiner } 10864e45507SPeter Feiner } 10964e45507SPeter Feiner 11064e45507SPeter Feiner 11134679d7eSShaohua Li int sysctl_overcommit_memory __read_mostly = OVERCOMMIT_GUESS; /* heuristic overcommit */ 11234679d7eSShaohua Li int sysctl_overcommit_ratio __read_mostly = 50; /* default is 50% */ 11349f0ce5fSJerome Marchand unsigned long sysctl_overcommit_kbytes __read_mostly; 114c3d8c141SChristoph Lameter int sysctl_max_map_count __read_mostly = DEFAULT_MAX_MAP_COUNT; 115c9b1d098SAndrew Shewmaker unsigned long sysctl_user_reserve_kbytes __read_mostly = 1UL << 17; /* 128MB */ 1164eeab4f5SAndrew Shewmaker unsigned long sysctl_admin_reserve_kbytes __read_mostly = 1UL << 13; /* 8MB */ 11734679d7eSShaohua Li /* 11834679d7eSShaohua Li * Make sure vm_committed_as in one cacheline and not cacheline shared with 11934679d7eSShaohua Li * other variables. It can be updated by several CPUs frequently. 12034679d7eSShaohua Li */ 12134679d7eSShaohua Li struct percpu_counter vm_committed_as ____cacheline_aligned_in_smp; 1221da177e4SLinus Torvalds 1231da177e4SLinus Torvalds /* 124997071bcSK. Y. Srinivasan * The global memory commitment made in the system can be a metric 125997071bcSK. Y. Srinivasan * that can be used to drive ballooning decisions when Linux is hosted 126997071bcSK. Y. Srinivasan * as a guest. On Hyper-V, the host implements a policy engine for dynamically 127997071bcSK. Y. Srinivasan * balancing memory across competing virtual machines that are hosted. 128997071bcSK. Y. Srinivasan * Several metrics drive this policy engine including the guest reported 129997071bcSK. Y. Srinivasan * memory commitment. 130997071bcSK. Y. Srinivasan */ 131997071bcSK. Y. Srinivasan unsigned long vm_memory_committed(void) 132997071bcSK. Y. Srinivasan { 133997071bcSK. Y. Srinivasan return percpu_counter_read_positive(&vm_committed_as); 134997071bcSK. Y. Srinivasan } 135997071bcSK. Y. Srinivasan EXPORT_SYMBOL_GPL(vm_memory_committed); 136997071bcSK. Y. Srinivasan 137997071bcSK. Y. Srinivasan /* 1381da177e4SLinus Torvalds * Check that a process has enough memory to allocate a new virtual 1391da177e4SLinus Torvalds * mapping. 0 means there is enough memory for the allocation to 1401da177e4SLinus Torvalds * succeed and -ENOMEM implies there is not. 1411da177e4SLinus Torvalds * 1421da177e4SLinus Torvalds * We currently support three overcommit policies, which are set via the 1431da177e4SLinus Torvalds * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting 1441da177e4SLinus Torvalds * 1451da177e4SLinus Torvalds * Strict overcommit modes added 2002 Feb 26 by Alan Cox. 1461da177e4SLinus Torvalds * Additional code 2002 Jul 20 by Robert Love. 1471da177e4SLinus Torvalds * 1481da177e4SLinus Torvalds * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise. 1491da177e4SLinus Torvalds * 1501da177e4SLinus Torvalds * Note this is a helper function intended to be used by LSMs which 1511da177e4SLinus Torvalds * wish to use this logic. 1521da177e4SLinus Torvalds */ 15334b4e4aaSAlan Cox int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin) 1541da177e4SLinus Torvalds { 155c9b1d098SAndrew Shewmaker unsigned long free, allowed, reserve; 1561da177e4SLinus Torvalds 15782f71ae4SKonstantin Khlebnikov VM_WARN_ONCE(percpu_counter_read(&vm_committed_as) < 15882f71ae4SKonstantin Khlebnikov -(s64)vm_committed_as_batch * num_online_cpus(), 15982f71ae4SKonstantin Khlebnikov "memory commitment underflow"); 16082f71ae4SKonstantin Khlebnikov 1611da177e4SLinus Torvalds vm_acct_memory(pages); 1621da177e4SLinus Torvalds 1631da177e4SLinus Torvalds /* 1641da177e4SLinus Torvalds * Sometimes we want to use more memory than we have 1651da177e4SLinus Torvalds */ 1661da177e4SLinus Torvalds if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS) 1671da177e4SLinus Torvalds return 0; 1681da177e4SLinus Torvalds 1691da177e4SLinus Torvalds if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) { 170c15bef30SDmitry Fink free = global_page_state(NR_FREE_PAGES); 171c15bef30SDmitry Fink free += global_page_state(NR_FILE_PAGES); 1721da177e4SLinus Torvalds 173c15bef30SDmitry Fink /* 174c15bef30SDmitry Fink * shmem pages shouldn't be counted as free in this 175c15bef30SDmitry Fink * case, they can't be purged, only swapped out, and 176c15bef30SDmitry Fink * that won't affect the overall amount of available 177c15bef30SDmitry Fink * memory in the system. 178c15bef30SDmitry Fink */ 179c15bef30SDmitry Fink free -= global_page_state(NR_SHMEM); 180c15bef30SDmitry Fink 181ec8acf20SShaohua Li free += get_nr_swap_pages(); 1821da177e4SLinus Torvalds 1831da177e4SLinus Torvalds /* 1841da177e4SLinus Torvalds * Any slabs which are created with the 1851da177e4SLinus Torvalds * SLAB_RECLAIM_ACCOUNT flag claim to have contents 1861da177e4SLinus Torvalds * which are reclaimable, under pressure. The dentry 1871da177e4SLinus Torvalds * cache and most inode caches should fall into this 1881da177e4SLinus Torvalds */ 189972d1a7bSChristoph Lameter free += global_page_state(NR_SLAB_RECLAIMABLE); 1901da177e4SLinus Torvalds 1911da177e4SLinus Torvalds /* 192c15bef30SDmitry Fink * Leave reserved pages. The pages are not for anonymous pages. 193c15bef30SDmitry Fink */ 194c15bef30SDmitry Fink if (free <= totalreserve_pages) 195c15bef30SDmitry Fink goto error; 196c15bef30SDmitry Fink else 197c15bef30SDmitry Fink free -= totalreserve_pages; 198c15bef30SDmitry Fink 199c15bef30SDmitry Fink /* 2004eeab4f5SAndrew Shewmaker * Reserve some for root 2011da177e4SLinus Torvalds */ 2021da177e4SLinus Torvalds if (!cap_sys_admin) 2034eeab4f5SAndrew Shewmaker free -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10); 2041da177e4SLinus Torvalds 2051da177e4SLinus Torvalds if (free > pages) 2061da177e4SLinus Torvalds return 0; 2071da177e4SLinus Torvalds 2086d9f7839SHideo AOKI goto error; 2091da177e4SLinus Torvalds } 2101da177e4SLinus Torvalds 21100619bccSJerome Marchand allowed = vm_commit_limit(); 2121da177e4SLinus Torvalds /* 2134eeab4f5SAndrew Shewmaker * Reserve some for root 2141da177e4SLinus Torvalds */ 2151da177e4SLinus Torvalds if (!cap_sys_admin) 2164eeab4f5SAndrew Shewmaker allowed -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10); 2171da177e4SLinus Torvalds 218c9b1d098SAndrew Shewmaker /* 219c9b1d098SAndrew Shewmaker * Don't let a single process grow so big a user can't recover 220c9b1d098SAndrew Shewmaker */ 221c9b1d098SAndrew Shewmaker if (mm) { 222c9b1d098SAndrew Shewmaker reserve = sysctl_user_reserve_kbytes >> (PAGE_SHIFT - 10); 223c9b1d098SAndrew Shewmaker allowed -= min(mm->total_vm / 32, reserve); 224c9b1d098SAndrew Shewmaker } 2251da177e4SLinus Torvalds 22600a62ce9SKOSAKI Motohiro if (percpu_counter_read_positive(&vm_committed_as) < allowed) 2271da177e4SLinus Torvalds return 0; 2286d9f7839SHideo AOKI error: 2291da177e4SLinus Torvalds vm_unacct_memory(pages); 2301da177e4SLinus Torvalds 2311da177e4SLinus Torvalds return -ENOMEM; 2321da177e4SLinus Torvalds } 2331da177e4SLinus Torvalds 2341da177e4SLinus Torvalds /* 235c8c06efaSDavidlohr Bueso * Requires inode->i_mapping->i_mmap_rwsem 2361da177e4SLinus Torvalds */ 2371da177e4SLinus Torvalds static void __remove_shared_vm_struct(struct vm_area_struct *vma, 2381da177e4SLinus Torvalds struct file *file, struct address_space *mapping) 2391da177e4SLinus Torvalds { 2401da177e4SLinus Torvalds if (vma->vm_flags & VM_DENYWRITE) 241496ad9aaSAl Viro atomic_inc(&file_inode(file)->i_writecount); 2421da177e4SLinus Torvalds if (vma->vm_flags & VM_SHARED) 2434bb5f5d9SDavid Herrmann mapping_unmap_writable(mapping); 2441da177e4SLinus Torvalds 2451da177e4SLinus Torvalds flush_dcache_mmap_lock(mapping); 2461da177e4SLinus Torvalds if (unlikely(vma->vm_flags & VM_NONLINEAR)) 2476b2dbba8SMichel Lespinasse list_del_init(&vma->shared.nonlinear); 2481da177e4SLinus Torvalds else 2496b2dbba8SMichel Lespinasse vma_interval_tree_remove(vma, &mapping->i_mmap); 2501da177e4SLinus Torvalds flush_dcache_mmap_unlock(mapping); 2511da177e4SLinus Torvalds } 2521da177e4SLinus Torvalds 2531da177e4SLinus Torvalds /* 2546b2dbba8SMichel Lespinasse * Unlink a file-based vm structure from its interval tree, to hide 255a8fb5618SHugh Dickins * vma from rmap and vmtruncate before freeing its page tables. 2561da177e4SLinus Torvalds */ 257a8fb5618SHugh Dickins void unlink_file_vma(struct vm_area_struct *vma) 2581da177e4SLinus Torvalds { 2591da177e4SLinus Torvalds struct file *file = vma->vm_file; 2601da177e4SLinus Torvalds 2611da177e4SLinus Torvalds if (file) { 2621da177e4SLinus Torvalds struct address_space *mapping = file->f_mapping; 26383cde9e8SDavidlohr Bueso i_mmap_lock_write(mapping); 2641da177e4SLinus Torvalds __remove_shared_vm_struct(vma, file, mapping); 26583cde9e8SDavidlohr Bueso i_mmap_unlock_write(mapping); 2661da177e4SLinus Torvalds } 267a8fb5618SHugh Dickins } 268a8fb5618SHugh Dickins 269a8fb5618SHugh Dickins /* 270a8fb5618SHugh Dickins * Close a vm structure and free it, returning the next. 271a8fb5618SHugh Dickins */ 272a8fb5618SHugh Dickins static struct vm_area_struct *remove_vma(struct vm_area_struct *vma) 273a8fb5618SHugh Dickins { 274a8fb5618SHugh Dickins struct vm_area_struct *next = vma->vm_next; 275a8fb5618SHugh Dickins 276a8fb5618SHugh Dickins might_sleep(); 2771da177e4SLinus Torvalds if (vma->vm_ops && vma->vm_ops->close) 2781da177e4SLinus Torvalds vma->vm_ops->close(vma); 279e9714acfSKonstantin Khlebnikov if (vma->vm_file) 280a8fb5618SHugh Dickins fput(vma->vm_file); 281f0be3d32SLee Schermerhorn mpol_put(vma_policy(vma)); 2821da177e4SLinus Torvalds kmem_cache_free(vm_area_cachep, vma); 283a8fb5618SHugh Dickins return next; 2841da177e4SLinus Torvalds } 2851da177e4SLinus Torvalds 286e4eb1ff6SLinus Torvalds static unsigned long do_brk(unsigned long addr, unsigned long len); 287e4eb1ff6SLinus Torvalds 2886a6160a7SHeiko Carstens SYSCALL_DEFINE1(brk, unsigned long, brk) 2891da177e4SLinus Torvalds { 2908764b338SCyrill Gorcunov unsigned long retval; 2911da177e4SLinus Torvalds unsigned long newbrk, oldbrk; 2921da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 293a5b4592cSJiri Kosina unsigned long min_brk; 294128557ffSMichel Lespinasse bool populate; 2951da177e4SLinus Torvalds 2961da177e4SLinus Torvalds down_write(&mm->mmap_sem); 2971da177e4SLinus Torvalds 298a5b4592cSJiri Kosina #ifdef CONFIG_COMPAT_BRK 2995520e894SJiri Kosina /* 3005520e894SJiri Kosina * CONFIG_COMPAT_BRK can still be overridden by setting 3015520e894SJiri Kosina * randomize_va_space to 2, which will still cause mm->start_brk 3025520e894SJiri Kosina * to be arbitrarily shifted 3035520e894SJiri Kosina */ 3044471a675SJiri Kosina if (current->brk_randomized) 3055520e894SJiri Kosina min_brk = mm->start_brk; 3065520e894SJiri Kosina else 3075520e894SJiri Kosina min_brk = mm->end_data; 308a5b4592cSJiri Kosina #else 309a5b4592cSJiri Kosina min_brk = mm->start_brk; 310a5b4592cSJiri Kosina #endif 311a5b4592cSJiri Kosina if (brk < min_brk) 3121da177e4SLinus Torvalds goto out; 3131e624196SRam Gupta 3141e624196SRam Gupta /* 3151e624196SRam Gupta * Check against rlimit here. If this check is done later after the test 3161e624196SRam Gupta * of oldbrk with newbrk then it can escape the test and let the data 3171e624196SRam Gupta * segment grow beyond its set limit the in case where the limit is 3181e624196SRam Gupta * not page aligned -Ram Gupta 3191e624196SRam Gupta */ 3208764b338SCyrill Gorcunov if (check_data_rlimit(rlimit(RLIMIT_DATA), brk, mm->start_brk, 3218764b338SCyrill Gorcunov mm->end_data, mm->start_data)) 3221e624196SRam Gupta goto out; 3231e624196SRam Gupta 3241da177e4SLinus Torvalds newbrk = PAGE_ALIGN(brk); 3251da177e4SLinus Torvalds oldbrk = PAGE_ALIGN(mm->brk); 3261da177e4SLinus Torvalds if (oldbrk == newbrk) 3271da177e4SLinus Torvalds goto set_brk; 3281da177e4SLinus Torvalds 3291da177e4SLinus Torvalds /* Always allow shrinking brk. */ 3301da177e4SLinus Torvalds if (brk <= mm->brk) { 3311da177e4SLinus Torvalds if (!do_munmap(mm, newbrk, oldbrk-newbrk)) 3321da177e4SLinus Torvalds goto set_brk; 3331da177e4SLinus Torvalds goto out; 3341da177e4SLinus Torvalds } 3351da177e4SLinus Torvalds 3361da177e4SLinus Torvalds /* Check against existing mmap mappings. */ 3371da177e4SLinus Torvalds if (find_vma_intersection(mm, oldbrk, newbrk+PAGE_SIZE)) 3381da177e4SLinus Torvalds goto out; 3391da177e4SLinus Torvalds 3401da177e4SLinus Torvalds /* Ok, looks good - let it rip. */ 3411da177e4SLinus Torvalds if (do_brk(oldbrk, newbrk-oldbrk) != oldbrk) 3421da177e4SLinus Torvalds goto out; 343128557ffSMichel Lespinasse 3441da177e4SLinus Torvalds set_brk: 3451da177e4SLinus Torvalds mm->brk = brk; 346128557ffSMichel Lespinasse populate = newbrk > oldbrk && (mm->def_flags & VM_LOCKED) != 0; 347128557ffSMichel Lespinasse up_write(&mm->mmap_sem); 348128557ffSMichel Lespinasse if (populate) 349128557ffSMichel Lespinasse mm_populate(oldbrk, newbrk - oldbrk); 350128557ffSMichel Lespinasse return brk; 351128557ffSMichel Lespinasse 3521da177e4SLinus Torvalds out: 3531da177e4SLinus Torvalds retval = mm->brk; 3541da177e4SLinus Torvalds up_write(&mm->mmap_sem); 3551da177e4SLinus Torvalds return retval; 3561da177e4SLinus Torvalds } 3571da177e4SLinus Torvalds 358d3737187SMichel Lespinasse static long vma_compute_subtree_gap(struct vm_area_struct *vma) 359d3737187SMichel Lespinasse { 360d3737187SMichel Lespinasse unsigned long max, subtree_gap; 361d3737187SMichel Lespinasse max = vma->vm_start; 362d3737187SMichel Lespinasse if (vma->vm_prev) 363d3737187SMichel Lespinasse max -= vma->vm_prev->vm_end; 364d3737187SMichel Lespinasse if (vma->vm_rb.rb_left) { 365d3737187SMichel Lespinasse subtree_gap = rb_entry(vma->vm_rb.rb_left, 366d3737187SMichel Lespinasse struct vm_area_struct, vm_rb)->rb_subtree_gap; 367d3737187SMichel Lespinasse if (subtree_gap > max) 368d3737187SMichel Lespinasse max = subtree_gap; 369d3737187SMichel Lespinasse } 370d3737187SMichel Lespinasse if (vma->vm_rb.rb_right) { 371d3737187SMichel Lespinasse subtree_gap = rb_entry(vma->vm_rb.rb_right, 372d3737187SMichel Lespinasse struct vm_area_struct, vm_rb)->rb_subtree_gap; 373d3737187SMichel Lespinasse if (subtree_gap > max) 374d3737187SMichel Lespinasse max = subtree_gap; 375d3737187SMichel Lespinasse } 376d3737187SMichel Lespinasse return max; 377d3737187SMichel Lespinasse } 378d3737187SMichel Lespinasse 379ed8ea815SMichel Lespinasse #ifdef CONFIG_DEBUG_VM_RB 3801da177e4SLinus Torvalds static int browse_rb(struct rb_root *root) 3811da177e4SLinus Torvalds { 3825a0768f6SMichel Lespinasse int i = 0, j, bug = 0; 3831da177e4SLinus Torvalds struct rb_node *nd, *pn = NULL; 3841da177e4SLinus Torvalds unsigned long prev = 0, pend = 0; 3851da177e4SLinus Torvalds 3861da177e4SLinus Torvalds for (nd = rb_first(root); nd; nd = rb_next(nd)) { 3871da177e4SLinus Torvalds struct vm_area_struct *vma; 3881da177e4SLinus Torvalds vma = rb_entry(nd, struct vm_area_struct, vm_rb); 3895a0768f6SMichel Lespinasse if (vma->vm_start < prev) { 390ff26f70fSAndrew Morton pr_emerg("vm_start %lx < prev %lx\n", 391ff26f70fSAndrew Morton vma->vm_start, prev); 3925a0768f6SMichel Lespinasse bug = 1; 3935a0768f6SMichel Lespinasse } 3945a0768f6SMichel Lespinasse if (vma->vm_start < pend) { 395ff26f70fSAndrew Morton pr_emerg("vm_start %lx < pend %lx\n", 396ff26f70fSAndrew Morton vma->vm_start, pend); 3975a0768f6SMichel Lespinasse bug = 1; 3985a0768f6SMichel Lespinasse } 3995a0768f6SMichel Lespinasse if (vma->vm_start > vma->vm_end) { 400ff26f70fSAndrew Morton pr_emerg("vm_start %lx > vm_end %lx\n", 401ff26f70fSAndrew Morton vma->vm_start, vma->vm_end); 4025a0768f6SMichel Lespinasse bug = 1; 4035a0768f6SMichel Lespinasse } 4045a0768f6SMichel Lespinasse if (vma->rb_subtree_gap != vma_compute_subtree_gap(vma)) { 4058542bdfcSSasha Levin pr_emerg("free gap %lx, correct %lx\n", 4065a0768f6SMichel Lespinasse vma->rb_subtree_gap, 4075a0768f6SMichel Lespinasse vma_compute_subtree_gap(vma)); 4085a0768f6SMichel Lespinasse bug = 1; 4095a0768f6SMichel Lespinasse } 4101da177e4SLinus Torvalds i++; 4111da177e4SLinus Torvalds pn = nd; 412d1af65d1SDavid Miller prev = vma->vm_start; 413d1af65d1SDavid Miller pend = vma->vm_end; 4141da177e4SLinus Torvalds } 4151da177e4SLinus Torvalds j = 0; 4165a0768f6SMichel Lespinasse for (nd = pn; nd; nd = rb_prev(nd)) 4171da177e4SLinus Torvalds j++; 4185a0768f6SMichel Lespinasse if (i != j) { 4198542bdfcSSasha Levin pr_emerg("backwards %d, forwards %d\n", j, i); 4205a0768f6SMichel Lespinasse bug = 1; 4211da177e4SLinus Torvalds } 4225a0768f6SMichel Lespinasse return bug ? -1 : i; 4231da177e4SLinus Torvalds } 4241da177e4SLinus Torvalds 425d3737187SMichel Lespinasse static void validate_mm_rb(struct rb_root *root, struct vm_area_struct *ignore) 426d3737187SMichel Lespinasse { 427d3737187SMichel Lespinasse struct rb_node *nd; 428d3737187SMichel Lespinasse 429d3737187SMichel Lespinasse for (nd = rb_first(root); nd; nd = rb_next(nd)) { 430d3737187SMichel Lespinasse struct vm_area_struct *vma; 431d3737187SMichel Lespinasse vma = rb_entry(nd, struct vm_area_struct, vm_rb); 43296dad67fSSasha Levin VM_BUG_ON_VMA(vma != ignore && 43396dad67fSSasha Levin vma->rb_subtree_gap != vma_compute_subtree_gap(vma), 43496dad67fSSasha Levin vma); 435d3737187SMichel Lespinasse } 4361da177e4SLinus Torvalds } 4371da177e4SLinus Torvalds 438eafd4dc4SRashika Kheria static void validate_mm(struct mm_struct *mm) 4391da177e4SLinus Torvalds { 4401da177e4SLinus Torvalds int bug = 0; 4411da177e4SLinus Torvalds int i = 0; 4425a0768f6SMichel Lespinasse unsigned long highest_address = 0; 443ed8ea815SMichel Lespinasse struct vm_area_struct *vma = mm->mmap; 444ff26f70fSAndrew Morton 445ed8ea815SMichel Lespinasse while (vma) { 446ed8ea815SMichel Lespinasse struct anon_vma_chain *avc; 447ff26f70fSAndrew Morton 44863c3b902SMichel Lespinasse vma_lock_anon_vma(vma); 449ed8ea815SMichel Lespinasse list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) 450ed8ea815SMichel Lespinasse anon_vma_interval_tree_verify(avc); 45163c3b902SMichel Lespinasse vma_unlock_anon_vma(vma); 4525a0768f6SMichel Lespinasse highest_address = vma->vm_end; 453ed8ea815SMichel Lespinasse vma = vma->vm_next; 4541da177e4SLinus Torvalds i++; 4551da177e4SLinus Torvalds } 4565a0768f6SMichel Lespinasse if (i != mm->map_count) { 4578542bdfcSSasha Levin pr_emerg("map_count %d vm_next %d\n", mm->map_count, i); 4585a0768f6SMichel Lespinasse bug = 1; 4595a0768f6SMichel Lespinasse } 4605a0768f6SMichel Lespinasse if (highest_address != mm->highest_vm_end) { 4618542bdfcSSasha Levin pr_emerg("mm->highest_vm_end %lx, found %lx\n", 4625a0768f6SMichel Lespinasse mm->highest_vm_end, highest_address); 4635a0768f6SMichel Lespinasse bug = 1; 4645a0768f6SMichel Lespinasse } 4651da177e4SLinus Torvalds i = browse_rb(&mm->mm_rb); 4665a0768f6SMichel Lespinasse if (i != mm->map_count) { 467ff26f70fSAndrew Morton if (i != -1) 4688542bdfcSSasha Levin pr_emerg("map_count %d rb %d\n", mm->map_count, i); 4695a0768f6SMichel Lespinasse bug = 1; 4705a0768f6SMichel Lespinasse } 47196dad67fSSasha Levin VM_BUG_ON_MM(bug, mm); 4721da177e4SLinus Torvalds } 4731da177e4SLinus Torvalds #else 474d3737187SMichel Lespinasse #define validate_mm_rb(root, ignore) do { } while (0) 4751da177e4SLinus Torvalds #define validate_mm(mm) do { } while (0) 4761da177e4SLinus Torvalds #endif 4771da177e4SLinus Torvalds 478d3737187SMichel Lespinasse RB_DECLARE_CALLBACKS(static, vma_gap_callbacks, struct vm_area_struct, vm_rb, 479d3737187SMichel Lespinasse unsigned long, rb_subtree_gap, vma_compute_subtree_gap) 480d3737187SMichel Lespinasse 481d3737187SMichel Lespinasse /* 482d3737187SMichel Lespinasse * Update augmented rbtree rb_subtree_gap values after vma->vm_start or 483d3737187SMichel Lespinasse * vma->vm_prev->vm_end values changed, without modifying the vma's position 484d3737187SMichel Lespinasse * in the rbtree. 485d3737187SMichel Lespinasse */ 486d3737187SMichel Lespinasse static void vma_gap_update(struct vm_area_struct *vma) 487d3737187SMichel Lespinasse { 488d3737187SMichel Lespinasse /* 489d3737187SMichel Lespinasse * As it turns out, RB_DECLARE_CALLBACKS() already created a callback 490d3737187SMichel Lespinasse * function that does exacltly what we want. 491d3737187SMichel Lespinasse */ 492d3737187SMichel Lespinasse vma_gap_callbacks_propagate(&vma->vm_rb, NULL); 493d3737187SMichel Lespinasse } 494d3737187SMichel Lespinasse 495d3737187SMichel Lespinasse static inline void vma_rb_insert(struct vm_area_struct *vma, 496d3737187SMichel Lespinasse struct rb_root *root) 497d3737187SMichel Lespinasse { 498d3737187SMichel Lespinasse /* All rb_subtree_gap values must be consistent prior to insertion */ 499d3737187SMichel Lespinasse validate_mm_rb(root, NULL); 500d3737187SMichel Lespinasse 501d3737187SMichel Lespinasse rb_insert_augmented(&vma->vm_rb, root, &vma_gap_callbacks); 502d3737187SMichel Lespinasse } 503d3737187SMichel Lespinasse 504d3737187SMichel Lespinasse static void vma_rb_erase(struct vm_area_struct *vma, struct rb_root *root) 505d3737187SMichel Lespinasse { 506d3737187SMichel Lespinasse /* 507d3737187SMichel Lespinasse * All rb_subtree_gap values must be consistent prior to erase, 508d3737187SMichel Lespinasse * with the possible exception of the vma being erased. 509d3737187SMichel Lespinasse */ 510d3737187SMichel Lespinasse validate_mm_rb(root, vma); 511d3737187SMichel Lespinasse 512d3737187SMichel Lespinasse /* 513d3737187SMichel Lespinasse * Note rb_erase_augmented is a fairly large inline function, 514d3737187SMichel Lespinasse * so make sure we instantiate it only once with our desired 515d3737187SMichel Lespinasse * augmented rbtree callbacks. 516d3737187SMichel Lespinasse */ 517d3737187SMichel Lespinasse rb_erase_augmented(&vma->vm_rb, root, &vma_gap_callbacks); 518d3737187SMichel Lespinasse } 519d3737187SMichel Lespinasse 520bf181b9fSMichel Lespinasse /* 521bf181b9fSMichel Lespinasse * vma has some anon_vma assigned, and is already inserted on that 522bf181b9fSMichel Lespinasse * anon_vma's interval trees. 523bf181b9fSMichel Lespinasse * 524bf181b9fSMichel Lespinasse * Before updating the vma's vm_start / vm_end / vm_pgoff fields, the 525bf181b9fSMichel Lespinasse * vma must be removed from the anon_vma's interval trees using 526bf181b9fSMichel Lespinasse * anon_vma_interval_tree_pre_update_vma(). 527bf181b9fSMichel Lespinasse * 528bf181b9fSMichel Lespinasse * After the update, the vma will be reinserted using 529bf181b9fSMichel Lespinasse * anon_vma_interval_tree_post_update_vma(). 530bf181b9fSMichel Lespinasse * 531bf181b9fSMichel Lespinasse * The entire update must be protected by exclusive mmap_sem and by 532bf181b9fSMichel Lespinasse * the root anon_vma's mutex. 533bf181b9fSMichel Lespinasse */ 534bf181b9fSMichel Lespinasse static inline void 535bf181b9fSMichel Lespinasse anon_vma_interval_tree_pre_update_vma(struct vm_area_struct *vma) 536bf181b9fSMichel Lespinasse { 537bf181b9fSMichel Lespinasse struct anon_vma_chain *avc; 538bf181b9fSMichel Lespinasse 539bf181b9fSMichel Lespinasse list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) 540bf181b9fSMichel Lespinasse anon_vma_interval_tree_remove(avc, &avc->anon_vma->rb_root); 541bf181b9fSMichel Lespinasse } 542bf181b9fSMichel Lespinasse 543bf181b9fSMichel Lespinasse static inline void 544bf181b9fSMichel Lespinasse anon_vma_interval_tree_post_update_vma(struct vm_area_struct *vma) 545bf181b9fSMichel Lespinasse { 546bf181b9fSMichel Lespinasse struct anon_vma_chain *avc; 547bf181b9fSMichel Lespinasse 548bf181b9fSMichel Lespinasse list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) 549bf181b9fSMichel Lespinasse anon_vma_interval_tree_insert(avc, &avc->anon_vma->rb_root); 550bf181b9fSMichel Lespinasse } 551bf181b9fSMichel Lespinasse 5526597d783SHugh Dickins static int find_vma_links(struct mm_struct *mm, unsigned long addr, 5536597d783SHugh Dickins unsigned long end, struct vm_area_struct **pprev, 5546597d783SHugh Dickins struct rb_node ***rb_link, struct rb_node **rb_parent) 5551da177e4SLinus Torvalds { 5561da177e4SLinus Torvalds struct rb_node **__rb_link, *__rb_parent, *rb_prev; 5571da177e4SLinus Torvalds 5581da177e4SLinus Torvalds __rb_link = &mm->mm_rb.rb_node; 5591da177e4SLinus Torvalds rb_prev = __rb_parent = NULL; 5601da177e4SLinus Torvalds 5611da177e4SLinus Torvalds while (*__rb_link) { 5621da177e4SLinus Torvalds struct vm_area_struct *vma_tmp; 5631da177e4SLinus Torvalds 5641da177e4SLinus Torvalds __rb_parent = *__rb_link; 5651da177e4SLinus Torvalds vma_tmp = rb_entry(__rb_parent, struct vm_area_struct, vm_rb); 5661da177e4SLinus Torvalds 5671da177e4SLinus Torvalds if (vma_tmp->vm_end > addr) { 5686597d783SHugh Dickins /* Fail if an existing vma overlaps the area */ 5696597d783SHugh Dickins if (vma_tmp->vm_start < end) 5706597d783SHugh Dickins return -ENOMEM; 5711da177e4SLinus Torvalds __rb_link = &__rb_parent->rb_left; 5721da177e4SLinus Torvalds } else { 5731da177e4SLinus Torvalds rb_prev = __rb_parent; 5741da177e4SLinus Torvalds __rb_link = &__rb_parent->rb_right; 5751da177e4SLinus Torvalds } 5761da177e4SLinus Torvalds } 5771da177e4SLinus Torvalds 5781da177e4SLinus Torvalds *pprev = NULL; 5791da177e4SLinus Torvalds if (rb_prev) 5801da177e4SLinus Torvalds *pprev = rb_entry(rb_prev, struct vm_area_struct, vm_rb); 5811da177e4SLinus Torvalds *rb_link = __rb_link; 5821da177e4SLinus Torvalds *rb_parent = __rb_parent; 5836597d783SHugh Dickins return 0; 5841da177e4SLinus Torvalds } 5851da177e4SLinus Torvalds 586e8420a8eSCyril Hrubis static unsigned long count_vma_pages_range(struct mm_struct *mm, 587e8420a8eSCyril Hrubis unsigned long addr, unsigned long end) 588e8420a8eSCyril Hrubis { 589e8420a8eSCyril Hrubis unsigned long nr_pages = 0; 590e8420a8eSCyril Hrubis struct vm_area_struct *vma; 591e8420a8eSCyril Hrubis 592e8420a8eSCyril Hrubis /* Find first overlaping mapping */ 593e8420a8eSCyril Hrubis vma = find_vma_intersection(mm, addr, end); 594e8420a8eSCyril Hrubis if (!vma) 595e8420a8eSCyril Hrubis return 0; 596e8420a8eSCyril Hrubis 597e8420a8eSCyril Hrubis nr_pages = (min(end, vma->vm_end) - 598e8420a8eSCyril Hrubis max(addr, vma->vm_start)) >> PAGE_SHIFT; 599e8420a8eSCyril Hrubis 600e8420a8eSCyril Hrubis /* Iterate over the rest of the overlaps */ 601e8420a8eSCyril Hrubis for (vma = vma->vm_next; vma; vma = vma->vm_next) { 602e8420a8eSCyril Hrubis unsigned long overlap_len; 603e8420a8eSCyril Hrubis 604e8420a8eSCyril Hrubis if (vma->vm_start > end) 605e8420a8eSCyril Hrubis break; 606e8420a8eSCyril Hrubis 607e8420a8eSCyril Hrubis overlap_len = min(end, vma->vm_end) - vma->vm_start; 608e8420a8eSCyril Hrubis nr_pages += overlap_len >> PAGE_SHIFT; 609e8420a8eSCyril Hrubis } 610e8420a8eSCyril Hrubis 611e8420a8eSCyril Hrubis return nr_pages; 612e8420a8eSCyril Hrubis } 613e8420a8eSCyril Hrubis 6141da177e4SLinus Torvalds void __vma_link_rb(struct mm_struct *mm, struct vm_area_struct *vma, 6151da177e4SLinus Torvalds struct rb_node **rb_link, struct rb_node *rb_parent) 6161da177e4SLinus Torvalds { 617d3737187SMichel Lespinasse /* Update tracking information for the gap following the new vma. */ 618d3737187SMichel Lespinasse if (vma->vm_next) 619d3737187SMichel Lespinasse vma_gap_update(vma->vm_next); 620d3737187SMichel Lespinasse else 621d3737187SMichel Lespinasse mm->highest_vm_end = vma->vm_end; 622d3737187SMichel Lespinasse 623d3737187SMichel Lespinasse /* 624d3737187SMichel Lespinasse * vma->vm_prev wasn't known when we followed the rbtree to find the 625d3737187SMichel Lespinasse * correct insertion point for that vma. As a result, we could not 626d3737187SMichel Lespinasse * update the vma vm_rb parents rb_subtree_gap values on the way down. 627d3737187SMichel Lespinasse * So, we first insert the vma with a zero rb_subtree_gap value 628d3737187SMichel Lespinasse * (to be consistent with what we did on the way down), and then 629d3737187SMichel Lespinasse * immediately update the gap to the correct value. Finally we 630d3737187SMichel Lespinasse * rebalance the rbtree after all augmented values have been set. 631d3737187SMichel Lespinasse */ 6321da177e4SLinus Torvalds rb_link_node(&vma->vm_rb, rb_parent, rb_link); 633d3737187SMichel Lespinasse vma->rb_subtree_gap = 0; 634d3737187SMichel Lespinasse vma_gap_update(vma); 635d3737187SMichel Lespinasse vma_rb_insert(vma, &mm->mm_rb); 6361da177e4SLinus Torvalds } 6371da177e4SLinus Torvalds 638cb8f488cSDenys Vlasenko static void __vma_link_file(struct vm_area_struct *vma) 6391da177e4SLinus Torvalds { 6401da177e4SLinus Torvalds struct file *file; 6411da177e4SLinus Torvalds 6421da177e4SLinus Torvalds file = vma->vm_file; 6431da177e4SLinus Torvalds if (file) { 6441da177e4SLinus Torvalds struct address_space *mapping = file->f_mapping; 6451da177e4SLinus Torvalds 6461da177e4SLinus Torvalds if (vma->vm_flags & VM_DENYWRITE) 647496ad9aaSAl Viro atomic_dec(&file_inode(file)->i_writecount); 6481da177e4SLinus Torvalds if (vma->vm_flags & VM_SHARED) 6494bb5f5d9SDavid Herrmann atomic_inc(&mapping->i_mmap_writable); 6501da177e4SLinus Torvalds 6511da177e4SLinus Torvalds flush_dcache_mmap_lock(mapping); 6521da177e4SLinus Torvalds if (unlikely(vma->vm_flags & VM_NONLINEAR)) 6531da177e4SLinus Torvalds vma_nonlinear_insert(vma, &mapping->i_mmap_nonlinear); 6541da177e4SLinus Torvalds else 6556b2dbba8SMichel Lespinasse vma_interval_tree_insert(vma, &mapping->i_mmap); 6561da177e4SLinus Torvalds flush_dcache_mmap_unlock(mapping); 6571da177e4SLinus Torvalds } 6581da177e4SLinus Torvalds } 6591da177e4SLinus Torvalds 6601da177e4SLinus Torvalds static void 6611da177e4SLinus Torvalds __vma_link(struct mm_struct *mm, struct vm_area_struct *vma, 6621da177e4SLinus Torvalds struct vm_area_struct *prev, struct rb_node **rb_link, 6631da177e4SLinus Torvalds struct rb_node *rb_parent) 6641da177e4SLinus Torvalds { 6651da177e4SLinus Torvalds __vma_link_list(mm, vma, prev, rb_parent); 6661da177e4SLinus Torvalds __vma_link_rb(mm, vma, rb_link, rb_parent); 6671da177e4SLinus Torvalds } 6681da177e4SLinus Torvalds 6691da177e4SLinus Torvalds static void vma_link(struct mm_struct *mm, struct vm_area_struct *vma, 6701da177e4SLinus Torvalds struct vm_area_struct *prev, struct rb_node **rb_link, 6711da177e4SLinus Torvalds struct rb_node *rb_parent) 6721da177e4SLinus Torvalds { 6731da177e4SLinus Torvalds struct address_space *mapping = NULL; 6741da177e4SLinus Torvalds 67564ac4940SHuang Shijie if (vma->vm_file) { 6761da177e4SLinus Torvalds mapping = vma->vm_file->f_mapping; 67783cde9e8SDavidlohr Bueso i_mmap_lock_write(mapping); 67864ac4940SHuang Shijie } 6791da177e4SLinus Torvalds 6801da177e4SLinus Torvalds __vma_link(mm, vma, prev, rb_link, rb_parent); 6811da177e4SLinus Torvalds __vma_link_file(vma); 6821da177e4SLinus Torvalds 6831da177e4SLinus Torvalds if (mapping) 68483cde9e8SDavidlohr Bueso i_mmap_unlock_write(mapping); 6851da177e4SLinus Torvalds 6861da177e4SLinus Torvalds mm->map_count++; 6871da177e4SLinus Torvalds validate_mm(mm); 6881da177e4SLinus Torvalds } 6891da177e4SLinus Torvalds 6901da177e4SLinus Torvalds /* 69188f6b4c3SKautuk Consul * Helper for vma_adjust() in the split_vma insert case: insert a vma into the 6926b2dbba8SMichel Lespinasse * mm's list and rbtree. It has already been inserted into the interval tree. 6931da177e4SLinus Torvalds */ 69448aae425SZhenwenXu static void __insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma) 6951da177e4SLinus Torvalds { 6966597d783SHugh Dickins struct vm_area_struct *prev; 6971da177e4SLinus Torvalds struct rb_node **rb_link, *rb_parent; 6981da177e4SLinus Torvalds 6996597d783SHugh Dickins if (find_vma_links(mm, vma->vm_start, vma->vm_end, 7006597d783SHugh Dickins &prev, &rb_link, &rb_parent)) 7016597d783SHugh Dickins BUG(); 7021da177e4SLinus Torvalds __vma_link(mm, vma, prev, rb_link, rb_parent); 7031da177e4SLinus Torvalds mm->map_count++; 7041da177e4SLinus Torvalds } 7051da177e4SLinus Torvalds 7061da177e4SLinus Torvalds static inline void 7071da177e4SLinus Torvalds __vma_unlink(struct mm_struct *mm, struct vm_area_struct *vma, 7081da177e4SLinus Torvalds struct vm_area_struct *prev) 7091da177e4SLinus Torvalds { 710d3737187SMichel Lespinasse struct vm_area_struct *next; 711297c5eeeSLinus Torvalds 712d3737187SMichel Lespinasse vma_rb_erase(vma, &mm->mm_rb); 713d3737187SMichel Lespinasse prev->vm_next = next = vma->vm_next; 714297c5eeeSLinus Torvalds if (next) 715297c5eeeSLinus Torvalds next->vm_prev = prev; 716615d6e87SDavidlohr Bueso 717615d6e87SDavidlohr Bueso /* Kill the cache */ 718615d6e87SDavidlohr Bueso vmacache_invalidate(mm); 7191da177e4SLinus Torvalds } 7201da177e4SLinus Torvalds 7211da177e4SLinus Torvalds /* 7221da177e4SLinus Torvalds * We cannot adjust vm_start, vm_end, vm_pgoff fields of a vma that 7231da177e4SLinus Torvalds * is already present in an i_mmap tree without adjusting the tree. 7241da177e4SLinus Torvalds * The following helper function should be used when such adjustments 7251da177e4SLinus Torvalds * are necessary. The "insert" vma (if any) is to be inserted 7261da177e4SLinus Torvalds * before we drop the necessary locks. 7271da177e4SLinus Torvalds */ 7285beb4930SRik van Riel int vma_adjust(struct vm_area_struct *vma, unsigned long start, 7291da177e4SLinus Torvalds unsigned long end, pgoff_t pgoff, struct vm_area_struct *insert) 7301da177e4SLinus Torvalds { 7311da177e4SLinus Torvalds struct mm_struct *mm = vma->vm_mm; 7321da177e4SLinus Torvalds struct vm_area_struct *next = vma->vm_next; 7331da177e4SLinus Torvalds struct vm_area_struct *importer = NULL; 7341da177e4SLinus Torvalds struct address_space *mapping = NULL; 7356b2dbba8SMichel Lespinasse struct rb_root *root = NULL; 736012f1800SRik van Riel struct anon_vma *anon_vma = NULL; 7371da177e4SLinus Torvalds struct file *file = vma->vm_file; 738d3737187SMichel Lespinasse bool start_changed = false, end_changed = false; 7391da177e4SLinus Torvalds long adjust_next = 0; 7401da177e4SLinus Torvalds int remove_next = 0; 7411da177e4SLinus Torvalds 7421da177e4SLinus Torvalds if (next && !insert) { 743287d97acSLinus Torvalds struct vm_area_struct *exporter = NULL; 744287d97acSLinus Torvalds 7451da177e4SLinus Torvalds if (end >= next->vm_end) { 7461da177e4SLinus Torvalds /* 7471da177e4SLinus Torvalds * vma expands, overlapping all the next, and 7481da177e4SLinus Torvalds * perhaps the one after too (mprotect case 6). 7491da177e4SLinus Torvalds */ 7501da177e4SLinus Torvalds again: remove_next = 1 + (end > next->vm_end); 7511da177e4SLinus Torvalds end = next->vm_end; 752287d97acSLinus Torvalds exporter = next; 7531da177e4SLinus Torvalds importer = vma; 7541da177e4SLinus Torvalds } else if (end > next->vm_start) { 7551da177e4SLinus Torvalds /* 7561da177e4SLinus Torvalds * vma expands, overlapping part of the next: 7571da177e4SLinus Torvalds * mprotect case 5 shifting the boundary up. 7581da177e4SLinus Torvalds */ 7591da177e4SLinus Torvalds adjust_next = (end - next->vm_start) >> PAGE_SHIFT; 760287d97acSLinus Torvalds exporter = next; 7611da177e4SLinus Torvalds importer = vma; 7621da177e4SLinus Torvalds } else if (end < vma->vm_end) { 7631da177e4SLinus Torvalds /* 7641da177e4SLinus Torvalds * vma shrinks, and !insert tells it's not 7651da177e4SLinus Torvalds * split_vma inserting another: so it must be 7661da177e4SLinus Torvalds * mprotect case 4 shifting the boundary down. 7671da177e4SLinus Torvalds */ 7681da177e4SLinus Torvalds adjust_next = -((vma->vm_end - end) >> PAGE_SHIFT); 769287d97acSLinus Torvalds exporter = vma; 7701da177e4SLinus Torvalds importer = next; 7711da177e4SLinus Torvalds } 7721da177e4SLinus Torvalds 7735beb4930SRik van Riel /* 7745beb4930SRik van Riel * Easily overlooked: when mprotect shifts the boundary, 7755beb4930SRik van Riel * make sure the expanding vma has anon_vma set if the 7765beb4930SRik van Riel * shrinking vma had, to cover any anon pages imported. 7775beb4930SRik van Riel */ 778287d97acSLinus Torvalds if (exporter && exporter->anon_vma && !importer->anon_vma) { 779c4ea95d7SDaniel Forrest int error; 780c4ea95d7SDaniel Forrest 781287d97acSLinus Torvalds importer->anon_vma = exporter->anon_vma; 782*b800c91aSKonstantin Khlebnikov error = anon_vma_clone(importer, exporter); 783*b800c91aSKonstantin Khlebnikov if (error) { 784*b800c91aSKonstantin Khlebnikov importer->anon_vma = NULL; 785*b800c91aSKonstantin Khlebnikov return error; 786*b800c91aSKonstantin Khlebnikov } 7875beb4930SRik van Riel } 7885beb4930SRik van Riel } 7895beb4930SRik van Riel 7901da177e4SLinus Torvalds if (file) { 7911da177e4SLinus Torvalds mapping = file->f_mapping; 792682968e0SSrikar Dronamraju if (!(vma->vm_flags & VM_NONLINEAR)) { 7931da177e4SLinus Torvalds root = &mapping->i_mmap; 794cbc91f71SSrikar Dronamraju uprobe_munmap(vma, vma->vm_start, vma->vm_end); 795682968e0SSrikar Dronamraju 796682968e0SSrikar Dronamraju if (adjust_next) 797cbc91f71SSrikar Dronamraju uprobe_munmap(next, next->vm_start, 798cbc91f71SSrikar Dronamraju next->vm_end); 799682968e0SSrikar Dronamraju } 800682968e0SSrikar Dronamraju 80183cde9e8SDavidlohr Bueso i_mmap_lock_write(mapping); 8021da177e4SLinus Torvalds if (insert) { 8031da177e4SLinus Torvalds /* 8046b2dbba8SMichel Lespinasse * Put into interval tree now, so instantiated pages 8051da177e4SLinus Torvalds * are visible to arm/parisc __flush_dcache_page 8061da177e4SLinus Torvalds * throughout; but we cannot insert into address 8071da177e4SLinus Torvalds * space until vma start or end is updated. 8081da177e4SLinus Torvalds */ 8091da177e4SLinus Torvalds __vma_link_file(insert); 8101da177e4SLinus Torvalds } 8111da177e4SLinus Torvalds } 8121da177e4SLinus Torvalds 81394fcc585SAndrea Arcangeli vma_adjust_trans_huge(vma, start, end, adjust_next); 81494fcc585SAndrea Arcangeli 815012f1800SRik van Riel anon_vma = vma->anon_vma; 816bf181b9fSMichel Lespinasse if (!anon_vma && adjust_next) 817bf181b9fSMichel Lespinasse anon_vma = next->anon_vma; 818bf181b9fSMichel Lespinasse if (anon_vma) { 81981d1b09cSSasha Levin VM_BUG_ON_VMA(adjust_next && next->anon_vma && 82081d1b09cSSasha Levin anon_vma != next->anon_vma, next); 8214fc3f1d6SIngo Molnar anon_vma_lock_write(anon_vma); 822bf181b9fSMichel Lespinasse anon_vma_interval_tree_pre_update_vma(vma); 823bf181b9fSMichel Lespinasse if (adjust_next) 824bf181b9fSMichel Lespinasse anon_vma_interval_tree_pre_update_vma(next); 825bf181b9fSMichel Lespinasse } 826012f1800SRik van Riel 8271da177e4SLinus Torvalds if (root) { 8281da177e4SLinus Torvalds flush_dcache_mmap_lock(mapping); 8296b2dbba8SMichel Lespinasse vma_interval_tree_remove(vma, root); 8301da177e4SLinus Torvalds if (adjust_next) 8316b2dbba8SMichel Lespinasse vma_interval_tree_remove(next, root); 8321da177e4SLinus Torvalds } 8331da177e4SLinus Torvalds 834d3737187SMichel Lespinasse if (start != vma->vm_start) { 8351da177e4SLinus Torvalds vma->vm_start = start; 836d3737187SMichel Lespinasse start_changed = true; 837d3737187SMichel Lespinasse } 838d3737187SMichel Lespinasse if (end != vma->vm_end) { 8391da177e4SLinus Torvalds vma->vm_end = end; 840d3737187SMichel Lespinasse end_changed = true; 841d3737187SMichel Lespinasse } 8421da177e4SLinus Torvalds vma->vm_pgoff = pgoff; 8431da177e4SLinus Torvalds if (adjust_next) { 8441da177e4SLinus Torvalds next->vm_start += adjust_next << PAGE_SHIFT; 8451da177e4SLinus Torvalds next->vm_pgoff += adjust_next; 8461da177e4SLinus Torvalds } 8471da177e4SLinus Torvalds 8481da177e4SLinus Torvalds if (root) { 8491da177e4SLinus Torvalds if (adjust_next) 8506b2dbba8SMichel Lespinasse vma_interval_tree_insert(next, root); 8516b2dbba8SMichel Lespinasse vma_interval_tree_insert(vma, root); 8521da177e4SLinus Torvalds flush_dcache_mmap_unlock(mapping); 8531da177e4SLinus Torvalds } 8541da177e4SLinus Torvalds 8551da177e4SLinus Torvalds if (remove_next) { 8561da177e4SLinus Torvalds /* 8571da177e4SLinus Torvalds * vma_merge has merged next into vma, and needs 8581da177e4SLinus Torvalds * us to remove next before dropping the locks. 8591da177e4SLinus Torvalds */ 8601da177e4SLinus Torvalds __vma_unlink(mm, next, vma); 8611da177e4SLinus Torvalds if (file) 8621da177e4SLinus Torvalds __remove_shared_vm_struct(next, file, mapping); 8631da177e4SLinus Torvalds } else if (insert) { 8641da177e4SLinus Torvalds /* 8651da177e4SLinus Torvalds * split_vma has split insert from vma, and needs 8661da177e4SLinus Torvalds * us to insert it before dropping the locks 8671da177e4SLinus Torvalds * (it may either follow vma or precede it). 8681da177e4SLinus Torvalds */ 8691da177e4SLinus Torvalds __insert_vm_struct(mm, insert); 870d3737187SMichel Lespinasse } else { 871d3737187SMichel Lespinasse if (start_changed) 872d3737187SMichel Lespinasse vma_gap_update(vma); 873d3737187SMichel Lespinasse if (end_changed) { 874d3737187SMichel Lespinasse if (!next) 875d3737187SMichel Lespinasse mm->highest_vm_end = end; 876d3737187SMichel Lespinasse else if (!adjust_next) 877d3737187SMichel Lespinasse vma_gap_update(next); 878d3737187SMichel Lespinasse } 8791da177e4SLinus Torvalds } 8801da177e4SLinus Torvalds 881bf181b9fSMichel Lespinasse if (anon_vma) { 882bf181b9fSMichel Lespinasse anon_vma_interval_tree_post_update_vma(vma); 883bf181b9fSMichel Lespinasse if (adjust_next) 884bf181b9fSMichel Lespinasse anon_vma_interval_tree_post_update_vma(next); 88508b52706SKonstantin Khlebnikov anon_vma_unlock_write(anon_vma); 886bf181b9fSMichel Lespinasse } 8871da177e4SLinus Torvalds if (mapping) 88883cde9e8SDavidlohr Bueso i_mmap_unlock_write(mapping); 8891da177e4SLinus Torvalds 8902b144498SSrikar Dronamraju if (root) { 8917b2d81d4SIngo Molnar uprobe_mmap(vma); 8922b144498SSrikar Dronamraju 8932b144498SSrikar Dronamraju if (adjust_next) 8947b2d81d4SIngo Molnar uprobe_mmap(next); 8952b144498SSrikar Dronamraju } 8962b144498SSrikar Dronamraju 8971da177e4SLinus Torvalds if (remove_next) { 898925d1c40SMatt Helsley if (file) { 899cbc91f71SSrikar Dronamraju uprobe_munmap(next, next->vm_start, next->vm_end); 9001da177e4SLinus Torvalds fput(file); 901925d1c40SMatt Helsley } 9025beb4930SRik van Riel if (next->anon_vma) 9035beb4930SRik van Riel anon_vma_merge(vma, next); 9041da177e4SLinus Torvalds mm->map_count--; 9053964acd0SOleg Nesterov mpol_put(vma_policy(next)); 9061da177e4SLinus Torvalds kmem_cache_free(vm_area_cachep, next); 9071da177e4SLinus Torvalds /* 9081da177e4SLinus Torvalds * In mprotect's case 6 (see comments on vma_merge), 9091da177e4SLinus Torvalds * we must remove another next too. It would clutter 9101da177e4SLinus Torvalds * up the code too much to do both in one go. 9111da177e4SLinus Torvalds */ 9121da177e4SLinus Torvalds next = vma->vm_next; 913d3737187SMichel Lespinasse if (remove_next == 2) 9141da177e4SLinus Torvalds goto again; 915d3737187SMichel Lespinasse else if (next) 916d3737187SMichel Lespinasse vma_gap_update(next); 917d3737187SMichel Lespinasse else 918d3737187SMichel Lespinasse mm->highest_vm_end = end; 9191da177e4SLinus Torvalds } 9202b144498SSrikar Dronamraju if (insert && file) 9217b2d81d4SIngo Molnar uprobe_mmap(insert); 9221da177e4SLinus Torvalds 9231da177e4SLinus Torvalds validate_mm(mm); 9245beb4930SRik van Riel 9255beb4930SRik van Riel return 0; 9261da177e4SLinus Torvalds } 9271da177e4SLinus Torvalds 9281da177e4SLinus Torvalds /* 9291da177e4SLinus Torvalds * If the vma has a ->close operation then the driver probably needs to release 9301da177e4SLinus Torvalds * per-vma resources, so we don't attempt to merge those. 9311da177e4SLinus Torvalds */ 9321da177e4SLinus Torvalds static inline int is_mergeable_vma(struct vm_area_struct *vma, 9331da177e4SLinus Torvalds struct file *file, unsigned long vm_flags) 9341da177e4SLinus Torvalds { 93534228d47SCyrill Gorcunov /* 93634228d47SCyrill Gorcunov * VM_SOFTDIRTY should not prevent from VMA merging, if we 93734228d47SCyrill Gorcunov * match the flags but dirty bit -- the caller should mark 93834228d47SCyrill Gorcunov * merged VMA as dirty. If dirty bit won't be excluded from 93934228d47SCyrill Gorcunov * comparison, we increase pressue on the memory system forcing 94034228d47SCyrill Gorcunov * the kernel to generate new VMAs when old one could be 94134228d47SCyrill Gorcunov * extended instead. 94234228d47SCyrill Gorcunov */ 94334228d47SCyrill Gorcunov if ((vma->vm_flags ^ vm_flags) & ~VM_SOFTDIRTY) 9441da177e4SLinus Torvalds return 0; 9451da177e4SLinus Torvalds if (vma->vm_file != file) 9461da177e4SLinus Torvalds return 0; 9471da177e4SLinus Torvalds if (vma->vm_ops && vma->vm_ops->close) 9481da177e4SLinus Torvalds return 0; 9491da177e4SLinus Torvalds return 1; 9501da177e4SLinus Torvalds } 9511da177e4SLinus Torvalds 9521da177e4SLinus Torvalds static inline int is_mergeable_anon_vma(struct anon_vma *anon_vma1, 953965f55deSShaohua Li struct anon_vma *anon_vma2, 954965f55deSShaohua Li struct vm_area_struct *vma) 9551da177e4SLinus Torvalds { 956965f55deSShaohua Li /* 957965f55deSShaohua Li * The list_is_singular() test is to avoid merging VMA cloned from 958965f55deSShaohua Li * parents. This can improve scalability caused by anon_vma lock. 959965f55deSShaohua Li */ 960965f55deSShaohua Li if ((!anon_vma1 || !anon_vma2) && (!vma || 961965f55deSShaohua Li list_is_singular(&vma->anon_vma_chain))) 962965f55deSShaohua Li return 1; 963965f55deSShaohua Li return anon_vma1 == anon_vma2; 9641da177e4SLinus Torvalds } 9651da177e4SLinus Torvalds 9661da177e4SLinus Torvalds /* 9671da177e4SLinus Torvalds * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff) 9681da177e4SLinus Torvalds * in front of (at a lower virtual address and file offset than) the vma. 9691da177e4SLinus Torvalds * 9701da177e4SLinus Torvalds * We cannot merge two vmas if they have differently assigned (non-NULL) 9711da177e4SLinus Torvalds * anon_vmas, nor if same anon_vma is assigned but offsets incompatible. 9721da177e4SLinus Torvalds * 9731da177e4SLinus Torvalds * We don't check here for the merged mmap wrapping around the end of pagecache 9741da177e4SLinus Torvalds * indices (16TB on ia32) because do_mmap_pgoff() does not permit mmap's which 9751da177e4SLinus Torvalds * wrap, nor mmaps which cover the final page at index -1UL. 9761da177e4SLinus Torvalds */ 9771da177e4SLinus Torvalds static int 9781da177e4SLinus Torvalds can_vma_merge_before(struct vm_area_struct *vma, unsigned long vm_flags, 9791da177e4SLinus Torvalds struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff) 9801da177e4SLinus Torvalds { 9811da177e4SLinus Torvalds if (is_mergeable_vma(vma, file, vm_flags) && 982965f55deSShaohua Li is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) { 9831da177e4SLinus Torvalds if (vma->vm_pgoff == vm_pgoff) 9841da177e4SLinus Torvalds return 1; 9851da177e4SLinus Torvalds } 9861da177e4SLinus Torvalds return 0; 9871da177e4SLinus Torvalds } 9881da177e4SLinus Torvalds 9891da177e4SLinus Torvalds /* 9901da177e4SLinus Torvalds * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff) 9911da177e4SLinus Torvalds * beyond (at a higher virtual address and file offset than) the vma. 9921da177e4SLinus Torvalds * 9931da177e4SLinus Torvalds * We cannot merge two vmas if they have differently assigned (non-NULL) 9941da177e4SLinus Torvalds * anon_vmas, nor if same anon_vma is assigned but offsets incompatible. 9951da177e4SLinus Torvalds */ 9961da177e4SLinus Torvalds static int 9971da177e4SLinus Torvalds can_vma_merge_after(struct vm_area_struct *vma, unsigned long vm_flags, 9981da177e4SLinus Torvalds struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff) 9991da177e4SLinus Torvalds { 10001da177e4SLinus Torvalds if (is_mergeable_vma(vma, file, vm_flags) && 1001965f55deSShaohua Li is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) { 10021da177e4SLinus Torvalds pgoff_t vm_pglen; 1003d6e93217SLibin vm_pglen = vma_pages(vma); 10041da177e4SLinus Torvalds if (vma->vm_pgoff + vm_pglen == vm_pgoff) 10051da177e4SLinus Torvalds return 1; 10061da177e4SLinus Torvalds } 10071da177e4SLinus Torvalds return 0; 10081da177e4SLinus Torvalds } 10091da177e4SLinus Torvalds 10101da177e4SLinus Torvalds /* 10111da177e4SLinus Torvalds * Given a mapping request (addr,end,vm_flags,file,pgoff), figure out 10121da177e4SLinus Torvalds * whether that can be merged with its predecessor or its successor. 10131da177e4SLinus Torvalds * Or both (it neatly fills a hole). 10141da177e4SLinus Torvalds * 10151da177e4SLinus Torvalds * In most cases - when called for mmap, brk or mremap - [addr,end) is 10161da177e4SLinus Torvalds * certain not to be mapped by the time vma_merge is called; but when 10171da177e4SLinus Torvalds * called for mprotect, it is certain to be already mapped (either at 10181da177e4SLinus Torvalds * an offset within prev, or at the start of next), and the flags of 10191da177e4SLinus Torvalds * this area are about to be changed to vm_flags - and the no-change 10201da177e4SLinus Torvalds * case has already been eliminated. 10211da177e4SLinus Torvalds * 10221da177e4SLinus Torvalds * The following mprotect cases have to be considered, where AAAA is 10231da177e4SLinus Torvalds * the area passed down from mprotect_fixup, never extending beyond one 10241da177e4SLinus Torvalds * vma, PPPPPP is the prev vma specified, and NNNNNN the next vma after: 10251da177e4SLinus Torvalds * 10261da177e4SLinus Torvalds * AAAA AAAA AAAA AAAA 10271da177e4SLinus Torvalds * PPPPPPNNNNNN PPPPPPNNNNNN PPPPPPNNNNNN PPPPNNNNXXXX 10281da177e4SLinus Torvalds * cannot merge might become might become might become 10291da177e4SLinus Torvalds * PPNNNNNNNNNN PPPPPPPPPPNN PPPPPPPPPPPP 6 or 10301da177e4SLinus Torvalds * mmap, brk or case 4 below case 5 below PPPPPPPPXXXX 7 or 10311da177e4SLinus Torvalds * mremap move: PPPPNNNNNNNN 8 10321da177e4SLinus Torvalds * AAAA 10331da177e4SLinus Torvalds * PPPP NNNN PPPPPPPPPPPP PPPPPPPPNNNN PPPPNNNNNNNN 10341da177e4SLinus Torvalds * might become case 1 below case 2 below case 3 below 10351da177e4SLinus Torvalds * 10361da177e4SLinus Torvalds * Odd one out? Case 8, because it extends NNNN but needs flags of XXXX: 10371da177e4SLinus Torvalds * mprotect_fixup updates vm_flags & vm_page_prot on successful return. 10381da177e4SLinus Torvalds */ 10391da177e4SLinus Torvalds struct vm_area_struct *vma_merge(struct mm_struct *mm, 10401da177e4SLinus Torvalds struct vm_area_struct *prev, unsigned long addr, 10411da177e4SLinus Torvalds unsigned long end, unsigned long vm_flags, 10421da177e4SLinus Torvalds struct anon_vma *anon_vma, struct file *file, 10431da177e4SLinus Torvalds pgoff_t pgoff, struct mempolicy *policy) 10441da177e4SLinus Torvalds { 10451da177e4SLinus Torvalds pgoff_t pglen = (end - addr) >> PAGE_SHIFT; 10461da177e4SLinus Torvalds struct vm_area_struct *area, *next; 10475beb4930SRik van Riel int err; 10481da177e4SLinus Torvalds 10491da177e4SLinus Torvalds /* 10501da177e4SLinus Torvalds * We later require that vma->vm_flags == vm_flags, 10511da177e4SLinus Torvalds * so this tests vma->vm_flags & VM_SPECIAL, too. 10521da177e4SLinus Torvalds */ 10531da177e4SLinus Torvalds if (vm_flags & VM_SPECIAL) 10541da177e4SLinus Torvalds return NULL; 10551da177e4SLinus Torvalds 10561da177e4SLinus Torvalds if (prev) 10571da177e4SLinus Torvalds next = prev->vm_next; 10581da177e4SLinus Torvalds else 10591da177e4SLinus Torvalds next = mm->mmap; 10601da177e4SLinus Torvalds area = next; 10611da177e4SLinus Torvalds if (next && next->vm_end == end) /* cases 6, 7, 8 */ 10621da177e4SLinus Torvalds next = next->vm_next; 10631da177e4SLinus Torvalds 10641da177e4SLinus Torvalds /* 10651da177e4SLinus Torvalds * Can it merge with the predecessor? 10661da177e4SLinus Torvalds */ 10671da177e4SLinus Torvalds if (prev && prev->vm_end == addr && 10681da177e4SLinus Torvalds mpol_equal(vma_policy(prev), policy) && 10691da177e4SLinus Torvalds can_vma_merge_after(prev, vm_flags, 10701da177e4SLinus Torvalds anon_vma, file, pgoff)) { 10711da177e4SLinus Torvalds /* 10721da177e4SLinus Torvalds * OK, it can. Can we now merge in the successor as well? 10731da177e4SLinus Torvalds */ 10741da177e4SLinus Torvalds if (next && end == next->vm_start && 10751da177e4SLinus Torvalds mpol_equal(policy, vma_policy(next)) && 10761da177e4SLinus Torvalds can_vma_merge_before(next, vm_flags, 10771da177e4SLinus Torvalds anon_vma, file, pgoff+pglen) && 10781da177e4SLinus Torvalds is_mergeable_anon_vma(prev->anon_vma, 1079965f55deSShaohua Li next->anon_vma, NULL)) { 10801da177e4SLinus Torvalds /* cases 1, 6 */ 10815beb4930SRik van Riel err = vma_adjust(prev, prev->vm_start, 10821da177e4SLinus Torvalds next->vm_end, prev->vm_pgoff, NULL); 10831da177e4SLinus Torvalds } else /* cases 2, 5, 7 */ 10845beb4930SRik van Riel err = vma_adjust(prev, prev->vm_start, 10851da177e4SLinus Torvalds end, prev->vm_pgoff, NULL); 10865beb4930SRik van Riel if (err) 10875beb4930SRik van Riel return NULL; 10886d50e60cSDavid Rientjes khugepaged_enter_vma_merge(prev, vm_flags); 10891da177e4SLinus Torvalds return prev; 10901da177e4SLinus Torvalds } 10911da177e4SLinus Torvalds 10921da177e4SLinus Torvalds /* 10931da177e4SLinus Torvalds * Can this new request be merged in front of next? 10941da177e4SLinus Torvalds */ 10951da177e4SLinus Torvalds if (next && end == next->vm_start && 10961da177e4SLinus Torvalds mpol_equal(policy, vma_policy(next)) && 10971da177e4SLinus Torvalds can_vma_merge_before(next, vm_flags, 10981da177e4SLinus Torvalds anon_vma, file, pgoff+pglen)) { 10991da177e4SLinus Torvalds if (prev && addr < prev->vm_end) /* case 4 */ 11005beb4930SRik van Riel err = vma_adjust(prev, prev->vm_start, 11011da177e4SLinus Torvalds addr, prev->vm_pgoff, NULL); 11021da177e4SLinus Torvalds else /* cases 3, 8 */ 11035beb4930SRik van Riel err = vma_adjust(area, addr, next->vm_end, 11041da177e4SLinus Torvalds next->vm_pgoff - pglen, NULL); 11055beb4930SRik van Riel if (err) 11065beb4930SRik van Riel return NULL; 11076d50e60cSDavid Rientjes khugepaged_enter_vma_merge(area, vm_flags); 11081da177e4SLinus Torvalds return area; 11091da177e4SLinus Torvalds } 11101da177e4SLinus Torvalds 11111da177e4SLinus Torvalds return NULL; 11121da177e4SLinus Torvalds } 11131da177e4SLinus Torvalds 11141da177e4SLinus Torvalds /* 1115d0e9fe17SLinus Torvalds * Rough compatbility check to quickly see if it's even worth looking 1116d0e9fe17SLinus Torvalds * at sharing an anon_vma. 1117d0e9fe17SLinus Torvalds * 1118d0e9fe17SLinus Torvalds * They need to have the same vm_file, and the flags can only differ 1119d0e9fe17SLinus Torvalds * in things that mprotect may change. 1120d0e9fe17SLinus Torvalds * 1121d0e9fe17SLinus Torvalds * NOTE! The fact that we share an anon_vma doesn't _have_ to mean that 1122d0e9fe17SLinus Torvalds * we can merge the two vma's. For example, we refuse to merge a vma if 1123d0e9fe17SLinus Torvalds * there is a vm_ops->close() function, because that indicates that the 1124d0e9fe17SLinus Torvalds * driver is doing some kind of reference counting. But that doesn't 1125d0e9fe17SLinus Torvalds * really matter for the anon_vma sharing case. 1126d0e9fe17SLinus Torvalds */ 1127d0e9fe17SLinus Torvalds static int anon_vma_compatible(struct vm_area_struct *a, struct vm_area_struct *b) 1128d0e9fe17SLinus Torvalds { 1129d0e9fe17SLinus Torvalds return a->vm_end == b->vm_start && 1130d0e9fe17SLinus Torvalds mpol_equal(vma_policy(a), vma_policy(b)) && 1131d0e9fe17SLinus Torvalds a->vm_file == b->vm_file && 113234228d47SCyrill Gorcunov !((a->vm_flags ^ b->vm_flags) & ~(VM_READ|VM_WRITE|VM_EXEC|VM_SOFTDIRTY)) && 1133d0e9fe17SLinus Torvalds b->vm_pgoff == a->vm_pgoff + ((b->vm_start - a->vm_start) >> PAGE_SHIFT); 1134d0e9fe17SLinus Torvalds } 1135d0e9fe17SLinus Torvalds 1136d0e9fe17SLinus Torvalds /* 1137d0e9fe17SLinus Torvalds * Do some basic sanity checking to see if we can re-use the anon_vma 1138d0e9fe17SLinus Torvalds * from 'old'. The 'a'/'b' vma's are in VM order - one of them will be 1139d0e9fe17SLinus Torvalds * the same as 'old', the other will be the new one that is trying 1140d0e9fe17SLinus Torvalds * to share the anon_vma. 1141d0e9fe17SLinus Torvalds * 1142d0e9fe17SLinus Torvalds * NOTE! This runs with mm_sem held for reading, so it is possible that 1143d0e9fe17SLinus Torvalds * the anon_vma of 'old' is concurrently in the process of being set up 1144d0e9fe17SLinus Torvalds * by another page fault trying to merge _that_. But that's ok: if it 1145d0e9fe17SLinus Torvalds * is being set up, that automatically means that it will be a singleton 1146d0e9fe17SLinus Torvalds * acceptable for merging, so we can do all of this optimistically. But 1147d0e9fe17SLinus Torvalds * we do that ACCESS_ONCE() to make sure that we never re-load the pointer. 1148d0e9fe17SLinus Torvalds * 1149d0e9fe17SLinus Torvalds * IOW: that the "list_is_singular()" test on the anon_vma_chain only 1150d0e9fe17SLinus Torvalds * matters for the 'stable anon_vma' case (ie the thing we want to avoid 1151d0e9fe17SLinus Torvalds * is to return an anon_vma that is "complex" due to having gone through 1152d0e9fe17SLinus Torvalds * a fork). 1153d0e9fe17SLinus Torvalds * 1154d0e9fe17SLinus Torvalds * We also make sure that the two vma's are compatible (adjacent, 1155d0e9fe17SLinus Torvalds * and with the same memory policies). That's all stable, even with just 1156d0e9fe17SLinus Torvalds * a read lock on the mm_sem. 1157d0e9fe17SLinus Torvalds */ 1158d0e9fe17SLinus Torvalds static struct anon_vma *reusable_anon_vma(struct vm_area_struct *old, struct vm_area_struct *a, struct vm_area_struct *b) 1159d0e9fe17SLinus Torvalds { 1160d0e9fe17SLinus Torvalds if (anon_vma_compatible(a, b)) { 1161d0e9fe17SLinus Torvalds struct anon_vma *anon_vma = ACCESS_ONCE(old->anon_vma); 1162d0e9fe17SLinus Torvalds 1163d0e9fe17SLinus Torvalds if (anon_vma && list_is_singular(&old->anon_vma_chain)) 1164d0e9fe17SLinus Torvalds return anon_vma; 1165d0e9fe17SLinus Torvalds } 1166d0e9fe17SLinus Torvalds return NULL; 1167d0e9fe17SLinus Torvalds } 1168d0e9fe17SLinus Torvalds 1169d0e9fe17SLinus Torvalds /* 11701da177e4SLinus Torvalds * find_mergeable_anon_vma is used by anon_vma_prepare, to check 11711da177e4SLinus Torvalds * neighbouring vmas for a suitable anon_vma, before it goes off 11721da177e4SLinus Torvalds * to allocate a new anon_vma. It checks because a repetitive 11731da177e4SLinus Torvalds * sequence of mprotects and faults may otherwise lead to distinct 11741da177e4SLinus Torvalds * anon_vmas being allocated, preventing vma merge in subsequent 11751da177e4SLinus Torvalds * mprotect. 11761da177e4SLinus Torvalds */ 11771da177e4SLinus Torvalds struct anon_vma *find_mergeable_anon_vma(struct vm_area_struct *vma) 11781da177e4SLinus Torvalds { 1179d0e9fe17SLinus Torvalds struct anon_vma *anon_vma; 11801da177e4SLinus Torvalds struct vm_area_struct *near; 11811da177e4SLinus Torvalds 11821da177e4SLinus Torvalds near = vma->vm_next; 11831da177e4SLinus Torvalds if (!near) 11841da177e4SLinus Torvalds goto try_prev; 11851da177e4SLinus Torvalds 1186d0e9fe17SLinus Torvalds anon_vma = reusable_anon_vma(near, vma, near); 1187d0e9fe17SLinus Torvalds if (anon_vma) 1188d0e9fe17SLinus Torvalds return anon_vma; 11891da177e4SLinus Torvalds try_prev: 11909be34c9dSLinus Torvalds near = vma->vm_prev; 11911da177e4SLinus Torvalds if (!near) 11921da177e4SLinus Torvalds goto none; 11931da177e4SLinus Torvalds 1194d0e9fe17SLinus Torvalds anon_vma = reusable_anon_vma(near, near, vma); 1195d0e9fe17SLinus Torvalds if (anon_vma) 1196d0e9fe17SLinus Torvalds return anon_vma; 11971da177e4SLinus Torvalds none: 11981da177e4SLinus Torvalds /* 11991da177e4SLinus Torvalds * There's no absolute need to look only at touching neighbours: 12001da177e4SLinus Torvalds * we could search further afield for "compatible" anon_vmas. 12011da177e4SLinus Torvalds * But it would probably just be a waste of time searching, 12021da177e4SLinus Torvalds * or lead to too many vmas hanging off the same anon_vma. 12031da177e4SLinus Torvalds * We're trying to allow mprotect remerging later on, 12041da177e4SLinus Torvalds * not trying to minimize memory used for anon_vmas. 12051da177e4SLinus Torvalds */ 12061da177e4SLinus Torvalds return NULL; 12071da177e4SLinus Torvalds } 12081da177e4SLinus Torvalds 12091da177e4SLinus Torvalds #ifdef CONFIG_PROC_FS 1210ab50b8edSHugh Dickins void vm_stat_account(struct mm_struct *mm, unsigned long flags, 12111da177e4SLinus Torvalds struct file *file, long pages) 12121da177e4SLinus Torvalds { 12131da177e4SLinus Torvalds const unsigned long stack_flags 12141da177e4SLinus Torvalds = VM_STACK_FLAGS & (VM_GROWSUP|VM_GROWSDOWN); 12151da177e4SLinus Torvalds 121644de9d0cSHuang Shijie mm->total_vm += pages; 121744de9d0cSHuang Shijie 12181da177e4SLinus Torvalds if (file) { 12191da177e4SLinus Torvalds mm->shared_vm += pages; 12201da177e4SLinus Torvalds if ((flags & (VM_EXEC|VM_WRITE)) == VM_EXEC) 12211da177e4SLinus Torvalds mm->exec_vm += pages; 12221da177e4SLinus Torvalds } else if (flags & stack_flags) 12231da177e4SLinus Torvalds mm->stack_vm += pages; 12241da177e4SLinus Torvalds } 12251da177e4SLinus Torvalds #endif /* CONFIG_PROC_FS */ 12261da177e4SLinus Torvalds 12271da177e4SLinus Torvalds /* 122840401530SAl Viro * If a hint addr is less than mmap_min_addr change hint to be as 122940401530SAl Viro * low as possible but still greater than mmap_min_addr 123040401530SAl Viro */ 123140401530SAl Viro static inline unsigned long round_hint_to_min(unsigned long hint) 123240401530SAl Viro { 123340401530SAl Viro hint &= PAGE_MASK; 123440401530SAl Viro if (((void *)hint != NULL) && 123540401530SAl Viro (hint < mmap_min_addr)) 123640401530SAl Viro return PAGE_ALIGN(mmap_min_addr); 123740401530SAl Viro return hint; 123840401530SAl Viro } 123940401530SAl Viro 1240363ee17fSDavidlohr Bueso static inline int mlock_future_check(struct mm_struct *mm, 1241363ee17fSDavidlohr Bueso unsigned long flags, 1242363ee17fSDavidlohr Bueso unsigned long len) 1243363ee17fSDavidlohr Bueso { 1244363ee17fSDavidlohr Bueso unsigned long locked, lock_limit; 1245363ee17fSDavidlohr Bueso 1246363ee17fSDavidlohr Bueso /* mlock MCL_FUTURE? */ 1247363ee17fSDavidlohr Bueso if (flags & VM_LOCKED) { 1248363ee17fSDavidlohr Bueso locked = len >> PAGE_SHIFT; 1249363ee17fSDavidlohr Bueso locked += mm->locked_vm; 1250363ee17fSDavidlohr Bueso lock_limit = rlimit(RLIMIT_MEMLOCK); 1251363ee17fSDavidlohr Bueso lock_limit >>= PAGE_SHIFT; 1252363ee17fSDavidlohr Bueso if (locked > lock_limit && !capable(CAP_IPC_LOCK)) 1253363ee17fSDavidlohr Bueso return -EAGAIN; 1254363ee17fSDavidlohr Bueso } 1255363ee17fSDavidlohr Bueso return 0; 1256363ee17fSDavidlohr Bueso } 1257363ee17fSDavidlohr Bueso 125840401530SAl Viro /* 125927f5de79SJianjun Kong * The caller must hold down_write(¤t->mm->mmap_sem). 12601da177e4SLinus Torvalds */ 12611da177e4SLinus Torvalds 1262e3fc629dSAl Viro unsigned long do_mmap_pgoff(struct file *file, unsigned long addr, 12631da177e4SLinus Torvalds unsigned long len, unsigned long prot, 1264bebeb3d6SMichel Lespinasse unsigned long flags, unsigned long pgoff, 126541badc15SMichel Lespinasse unsigned long *populate) 12661da177e4SLinus Torvalds { 12671da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 1268ca16d140SKOSAKI Motohiro vm_flags_t vm_flags; 12691da177e4SLinus Torvalds 127041badc15SMichel Lespinasse *populate = 0; 1271bebeb3d6SMichel Lespinasse 12721da177e4SLinus Torvalds /* 12731da177e4SLinus Torvalds * Does the application expect PROT_READ to imply PROT_EXEC? 12741da177e4SLinus Torvalds * 12751da177e4SLinus Torvalds * (the exception is when the underlying filesystem is noexec 12761da177e4SLinus Torvalds * mounted, in which case we dont add PROT_EXEC.) 12771da177e4SLinus Torvalds */ 12781da177e4SLinus Torvalds if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC)) 1279d3ac7f89SJosef "Jeff" Sipek if (!(file && (file->f_path.mnt->mnt_flags & MNT_NOEXEC))) 12801da177e4SLinus Torvalds prot |= PROT_EXEC; 12811da177e4SLinus Torvalds 12821da177e4SLinus Torvalds if (!len) 12831da177e4SLinus Torvalds return -EINVAL; 12841da177e4SLinus Torvalds 12857cd94146SEric Paris if (!(flags & MAP_FIXED)) 12867cd94146SEric Paris addr = round_hint_to_min(addr); 12877cd94146SEric Paris 12881da177e4SLinus Torvalds /* Careful about overflows.. */ 12891da177e4SLinus Torvalds len = PAGE_ALIGN(len); 12909206de95SAl Viro if (!len) 12911da177e4SLinus Torvalds return -ENOMEM; 12921da177e4SLinus Torvalds 12931da177e4SLinus Torvalds /* offset overflow? */ 12941da177e4SLinus Torvalds if ((pgoff + (len >> PAGE_SHIFT)) < pgoff) 12951da177e4SLinus Torvalds return -EOVERFLOW; 12961da177e4SLinus Torvalds 12971da177e4SLinus Torvalds /* Too many mappings? */ 12981da177e4SLinus Torvalds if (mm->map_count > sysctl_max_map_count) 12991da177e4SLinus Torvalds return -ENOMEM; 13001da177e4SLinus Torvalds 13011da177e4SLinus Torvalds /* Obtain the address to map to. we verify (or select) it and ensure 13021da177e4SLinus Torvalds * that it represents a valid section of the address space. 13031da177e4SLinus Torvalds */ 13041da177e4SLinus Torvalds addr = get_unmapped_area(file, addr, len, pgoff, flags); 13051da177e4SLinus Torvalds if (addr & ~PAGE_MASK) 13061da177e4SLinus Torvalds return addr; 13071da177e4SLinus Torvalds 13081da177e4SLinus Torvalds /* Do simple checking here so the lower-level routines won't have 13091da177e4SLinus Torvalds * to. we assume access permissions have been handled by the open 13101da177e4SLinus Torvalds * of the memory object, so we don't do any here. 13111da177e4SLinus Torvalds */ 13121da177e4SLinus Torvalds vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags) | 13131da177e4SLinus Torvalds mm->def_flags | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC; 13141da177e4SLinus Torvalds 1315cdf7b341SHuang Shijie if (flags & MAP_LOCKED) 13161da177e4SLinus Torvalds if (!can_do_mlock()) 13171da177e4SLinus Torvalds return -EPERM; 1318ba470de4SRik van Riel 1319363ee17fSDavidlohr Bueso if (mlock_future_check(mm, vm_flags, len)) 13201da177e4SLinus Torvalds return -EAGAIN; 13211da177e4SLinus Torvalds 13221da177e4SLinus Torvalds if (file) { 1323077bf22bSOleg Nesterov struct inode *inode = file_inode(file); 1324077bf22bSOleg Nesterov 13251da177e4SLinus Torvalds switch (flags & MAP_TYPE) { 13261da177e4SLinus Torvalds case MAP_SHARED: 13271da177e4SLinus Torvalds if ((prot&PROT_WRITE) && !(file->f_mode&FMODE_WRITE)) 13281da177e4SLinus Torvalds return -EACCES; 13291da177e4SLinus Torvalds 13301da177e4SLinus Torvalds /* 13311da177e4SLinus Torvalds * Make sure we don't allow writing to an append-only 13321da177e4SLinus Torvalds * file.. 13331da177e4SLinus Torvalds */ 13341da177e4SLinus Torvalds if (IS_APPEND(inode) && (file->f_mode & FMODE_WRITE)) 13351da177e4SLinus Torvalds return -EACCES; 13361da177e4SLinus Torvalds 13371da177e4SLinus Torvalds /* 13381da177e4SLinus Torvalds * Make sure there are no mandatory locks on the file. 13391da177e4SLinus Torvalds */ 1340d7a06983SJeff Layton if (locks_verify_locked(file)) 13411da177e4SLinus Torvalds return -EAGAIN; 13421da177e4SLinus Torvalds 13431da177e4SLinus Torvalds vm_flags |= VM_SHARED | VM_MAYSHARE; 13441da177e4SLinus Torvalds if (!(file->f_mode & FMODE_WRITE)) 13451da177e4SLinus Torvalds vm_flags &= ~(VM_MAYWRITE | VM_SHARED); 13461da177e4SLinus Torvalds 13471da177e4SLinus Torvalds /* fall through */ 13481da177e4SLinus Torvalds case MAP_PRIVATE: 13491da177e4SLinus Torvalds if (!(file->f_mode & FMODE_READ)) 13501da177e4SLinus Torvalds return -EACCES; 1351d3ac7f89SJosef "Jeff" Sipek if (file->f_path.mnt->mnt_flags & MNT_NOEXEC) { 135280c5606cSLinus Torvalds if (vm_flags & VM_EXEC) 135380c5606cSLinus Torvalds return -EPERM; 135480c5606cSLinus Torvalds vm_flags &= ~VM_MAYEXEC; 135580c5606cSLinus Torvalds } 135680c5606cSLinus Torvalds 135772c2d531SAl Viro if (!file->f_op->mmap) 135880c5606cSLinus Torvalds return -ENODEV; 1359b2c56e4fSOleg Nesterov if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP)) 1360b2c56e4fSOleg Nesterov return -EINVAL; 13611da177e4SLinus Torvalds break; 13621da177e4SLinus Torvalds 13631da177e4SLinus Torvalds default: 13641da177e4SLinus Torvalds return -EINVAL; 13651da177e4SLinus Torvalds } 13661da177e4SLinus Torvalds } else { 13671da177e4SLinus Torvalds switch (flags & MAP_TYPE) { 13681da177e4SLinus Torvalds case MAP_SHARED: 1369b2c56e4fSOleg Nesterov if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP)) 1370b2c56e4fSOleg Nesterov return -EINVAL; 1371ce363942STejun Heo /* 1372ce363942STejun Heo * Ignore pgoff. 1373ce363942STejun Heo */ 1374ce363942STejun Heo pgoff = 0; 13751da177e4SLinus Torvalds vm_flags |= VM_SHARED | VM_MAYSHARE; 13761da177e4SLinus Torvalds break; 13771da177e4SLinus Torvalds case MAP_PRIVATE: 13781da177e4SLinus Torvalds /* 13791da177e4SLinus Torvalds * Set pgoff according to addr for anon_vma. 13801da177e4SLinus Torvalds */ 13811da177e4SLinus Torvalds pgoff = addr >> PAGE_SHIFT; 13821da177e4SLinus Torvalds break; 13831da177e4SLinus Torvalds default: 13841da177e4SLinus Torvalds return -EINVAL; 13851da177e4SLinus Torvalds } 13861da177e4SLinus Torvalds } 13871da177e4SLinus Torvalds 1388c22c0d63SMichel Lespinasse /* 1389c22c0d63SMichel Lespinasse * Set 'VM_NORESERVE' if we should not account for the 1390c22c0d63SMichel Lespinasse * memory use of this mapping. 1391c22c0d63SMichel Lespinasse */ 1392c22c0d63SMichel Lespinasse if (flags & MAP_NORESERVE) { 1393c22c0d63SMichel Lespinasse /* We honor MAP_NORESERVE if allowed to overcommit */ 1394c22c0d63SMichel Lespinasse if (sysctl_overcommit_memory != OVERCOMMIT_NEVER) 1395c22c0d63SMichel Lespinasse vm_flags |= VM_NORESERVE; 1396c22c0d63SMichel Lespinasse 1397c22c0d63SMichel Lespinasse /* hugetlb applies strict overcommit unless MAP_NORESERVE */ 1398c22c0d63SMichel Lespinasse if (file && is_file_hugepages(file)) 1399c22c0d63SMichel Lespinasse vm_flags |= VM_NORESERVE; 1400c22c0d63SMichel Lespinasse } 1401c22c0d63SMichel Lespinasse 1402c22c0d63SMichel Lespinasse addr = mmap_region(file, addr, len, vm_flags, pgoff); 140309a9f1d2SMichel Lespinasse if (!IS_ERR_VALUE(addr) && 140409a9f1d2SMichel Lespinasse ((vm_flags & VM_LOCKED) || 140509a9f1d2SMichel Lespinasse (flags & (MAP_POPULATE | MAP_NONBLOCK)) == MAP_POPULATE)) 140641badc15SMichel Lespinasse *populate = len; 1407bebeb3d6SMichel Lespinasse return addr; 14080165ab44SMiklos Szeredi } 14096be5ceb0SLinus Torvalds 141066f0dc48SHugh Dickins SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len, 141166f0dc48SHugh Dickins unsigned long, prot, unsigned long, flags, 141266f0dc48SHugh Dickins unsigned long, fd, unsigned long, pgoff) 141366f0dc48SHugh Dickins { 141466f0dc48SHugh Dickins struct file *file = NULL; 141566f0dc48SHugh Dickins unsigned long retval = -EBADF; 141666f0dc48SHugh Dickins 141766f0dc48SHugh Dickins if (!(flags & MAP_ANONYMOUS)) { 1418120a795dSAl Viro audit_mmap_fd(fd, flags); 141966f0dc48SHugh Dickins file = fget(fd); 142066f0dc48SHugh Dickins if (!file) 142166f0dc48SHugh Dickins goto out; 1422af73e4d9SNaoya Horiguchi if (is_file_hugepages(file)) 1423af73e4d9SNaoya Horiguchi len = ALIGN(len, huge_page_size(hstate_file(file))); 1424493af578SJörn Engel retval = -EINVAL; 1425493af578SJörn Engel if (unlikely(flags & MAP_HUGETLB && !is_file_hugepages(file))) 1426493af578SJörn Engel goto out_fput; 142766f0dc48SHugh Dickins } else if (flags & MAP_HUGETLB) { 142866f0dc48SHugh Dickins struct user_struct *user = NULL; 1429c103a4dcSAndrew Morton struct hstate *hs; 1430af73e4d9SNaoya Horiguchi 1431c103a4dcSAndrew Morton hs = hstate_sizelog((flags >> MAP_HUGE_SHIFT) & SHM_HUGE_MASK); 1432091d0d55SLi Zefan if (!hs) 1433091d0d55SLi Zefan return -EINVAL; 1434091d0d55SLi Zefan 1435091d0d55SLi Zefan len = ALIGN(len, huge_page_size(hs)); 143666f0dc48SHugh Dickins /* 143766f0dc48SHugh Dickins * VM_NORESERVE is used because the reservations will be 143866f0dc48SHugh Dickins * taken when vm_ops->mmap() is called 143966f0dc48SHugh Dickins * A dummy user value is used because we are not locking 144066f0dc48SHugh Dickins * memory so no accounting is necessary 144166f0dc48SHugh Dickins */ 1442af73e4d9SNaoya Horiguchi file = hugetlb_file_setup(HUGETLB_ANON_FILE, len, 144342d7395fSAndi Kleen VM_NORESERVE, 144442d7395fSAndi Kleen &user, HUGETLB_ANONHUGE_INODE, 144542d7395fSAndi Kleen (flags >> MAP_HUGE_SHIFT) & MAP_HUGE_MASK); 144666f0dc48SHugh Dickins if (IS_ERR(file)) 144766f0dc48SHugh Dickins return PTR_ERR(file); 144866f0dc48SHugh Dickins } 144966f0dc48SHugh Dickins 145066f0dc48SHugh Dickins flags &= ~(MAP_EXECUTABLE | MAP_DENYWRITE); 145166f0dc48SHugh Dickins 1452eb36c587SAl Viro retval = vm_mmap_pgoff(file, addr, len, prot, flags, pgoff); 1453493af578SJörn Engel out_fput: 145466f0dc48SHugh Dickins if (file) 145566f0dc48SHugh Dickins fput(file); 145666f0dc48SHugh Dickins out: 145766f0dc48SHugh Dickins return retval; 145866f0dc48SHugh Dickins } 145966f0dc48SHugh Dickins 1460a4679373SChristoph Hellwig #ifdef __ARCH_WANT_SYS_OLD_MMAP 1461a4679373SChristoph Hellwig struct mmap_arg_struct { 1462a4679373SChristoph Hellwig unsigned long addr; 1463a4679373SChristoph Hellwig unsigned long len; 1464a4679373SChristoph Hellwig unsigned long prot; 1465a4679373SChristoph Hellwig unsigned long flags; 1466a4679373SChristoph Hellwig unsigned long fd; 1467a4679373SChristoph Hellwig unsigned long offset; 1468a4679373SChristoph Hellwig }; 1469a4679373SChristoph Hellwig 1470a4679373SChristoph Hellwig SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg) 1471a4679373SChristoph Hellwig { 1472a4679373SChristoph Hellwig struct mmap_arg_struct a; 1473a4679373SChristoph Hellwig 1474a4679373SChristoph Hellwig if (copy_from_user(&a, arg, sizeof(a))) 1475a4679373SChristoph Hellwig return -EFAULT; 1476a4679373SChristoph Hellwig if (a.offset & ~PAGE_MASK) 1477a4679373SChristoph Hellwig return -EINVAL; 1478a4679373SChristoph Hellwig 1479a4679373SChristoph Hellwig return sys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd, 1480a4679373SChristoph Hellwig a.offset >> PAGE_SHIFT); 1481a4679373SChristoph Hellwig } 1482a4679373SChristoph Hellwig #endif /* __ARCH_WANT_SYS_OLD_MMAP */ 1483a4679373SChristoph Hellwig 14844e950f6fSAlexey Dobriyan /* 14854e950f6fSAlexey Dobriyan * Some shared mappigns will want the pages marked read-only 14864e950f6fSAlexey Dobriyan * to track write events. If so, we'll downgrade vm_page_prot 14874e950f6fSAlexey Dobriyan * to the private version (using protection_map[] without the 14884e950f6fSAlexey Dobriyan * VM_SHARED bit). 14894e950f6fSAlexey Dobriyan */ 14904e950f6fSAlexey Dobriyan int vma_wants_writenotify(struct vm_area_struct *vma) 14914e950f6fSAlexey Dobriyan { 1492ca16d140SKOSAKI Motohiro vm_flags_t vm_flags = vma->vm_flags; 14934e950f6fSAlexey Dobriyan 14944e950f6fSAlexey Dobriyan /* If it was private or non-writable, the write bit is already clear */ 14954e950f6fSAlexey Dobriyan if ((vm_flags & (VM_WRITE|VM_SHARED)) != ((VM_WRITE|VM_SHARED))) 14964e950f6fSAlexey Dobriyan return 0; 14974e950f6fSAlexey Dobriyan 14984e950f6fSAlexey Dobriyan /* The backer wishes to know when pages are first written to? */ 14994e950f6fSAlexey Dobriyan if (vma->vm_ops && vma->vm_ops->page_mkwrite) 15004e950f6fSAlexey Dobriyan return 1; 15014e950f6fSAlexey Dobriyan 150264e45507SPeter Feiner /* The open routine did something to the protections that pgprot_modify 150364e45507SPeter Feiner * won't preserve? */ 15044e950f6fSAlexey Dobriyan if (pgprot_val(vma->vm_page_prot) != 150564e45507SPeter Feiner pgprot_val(vm_pgprot_modify(vma->vm_page_prot, vm_flags))) 15064e950f6fSAlexey Dobriyan return 0; 15074e950f6fSAlexey Dobriyan 150864e45507SPeter Feiner /* Do we need to track softdirty? */ 150964e45507SPeter Feiner if (IS_ENABLED(CONFIG_MEM_SOFT_DIRTY) && !(vm_flags & VM_SOFTDIRTY)) 151064e45507SPeter Feiner return 1; 151164e45507SPeter Feiner 15124e950f6fSAlexey Dobriyan /* Specialty mapping? */ 15134b6e1e37SKonstantin Khlebnikov if (vm_flags & VM_PFNMAP) 15144e950f6fSAlexey Dobriyan return 0; 15154e950f6fSAlexey Dobriyan 15164e950f6fSAlexey Dobriyan /* Can the mapping track the dirty pages? */ 15174e950f6fSAlexey Dobriyan return vma->vm_file && vma->vm_file->f_mapping && 15184e950f6fSAlexey Dobriyan mapping_cap_account_dirty(vma->vm_file->f_mapping); 15194e950f6fSAlexey Dobriyan } 15204e950f6fSAlexey Dobriyan 1521fc8744adSLinus Torvalds /* 1522fc8744adSLinus Torvalds * We account for memory if it's a private writeable mapping, 15235a6fe125SMel Gorman * not hugepages and VM_NORESERVE wasn't set. 1524fc8744adSLinus Torvalds */ 1525ca16d140SKOSAKI Motohiro static inline int accountable_mapping(struct file *file, vm_flags_t vm_flags) 1526fc8744adSLinus Torvalds { 15275a6fe125SMel Gorman /* 15285a6fe125SMel Gorman * hugetlb has its own accounting separate from the core VM 15295a6fe125SMel Gorman * VM_HUGETLB may not be set yet so we cannot check for that flag. 15305a6fe125SMel Gorman */ 15315a6fe125SMel Gorman if (file && is_file_hugepages(file)) 15325a6fe125SMel Gorman return 0; 15335a6fe125SMel Gorman 1534fc8744adSLinus Torvalds return (vm_flags & (VM_NORESERVE | VM_SHARED | VM_WRITE)) == VM_WRITE; 1535fc8744adSLinus Torvalds } 1536fc8744adSLinus Torvalds 15370165ab44SMiklos Szeredi unsigned long mmap_region(struct file *file, unsigned long addr, 1538c22c0d63SMichel Lespinasse unsigned long len, vm_flags_t vm_flags, unsigned long pgoff) 15390165ab44SMiklos Szeredi { 15400165ab44SMiklos Szeredi struct mm_struct *mm = current->mm; 15410165ab44SMiklos Szeredi struct vm_area_struct *vma, *prev; 15420165ab44SMiklos Szeredi int error; 15430165ab44SMiklos Szeredi struct rb_node **rb_link, *rb_parent; 15440165ab44SMiklos Szeredi unsigned long charged = 0; 15450165ab44SMiklos Szeredi 1546e8420a8eSCyril Hrubis /* Check against address space limit. */ 1547e8420a8eSCyril Hrubis if (!may_expand_vm(mm, len >> PAGE_SHIFT)) { 1548e8420a8eSCyril Hrubis unsigned long nr_pages; 1549e8420a8eSCyril Hrubis 1550e8420a8eSCyril Hrubis /* 1551e8420a8eSCyril Hrubis * MAP_FIXED may remove pages of mappings that intersects with 1552e8420a8eSCyril Hrubis * requested mapping. Account for the pages it would unmap. 1553e8420a8eSCyril Hrubis */ 1554e8420a8eSCyril Hrubis if (!(vm_flags & MAP_FIXED)) 1555e8420a8eSCyril Hrubis return -ENOMEM; 1556e8420a8eSCyril Hrubis 1557e8420a8eSCyril Hrubis nr_pages = count_vma_pages_range(mm, addr, addr + len); 1558e8420a8eSCyril Hrubis 1559e8420a8eSCyril Hrubis if (!may_expand_vm(mm, (len >> PAGE_SHIFT) - nr_pages)) 1560e8420a8eSCyril Hrubis return -ENOMEM; 1561e8420a8eSCyril Hrubis } 1562e8420a8eSCyril Hrubis 15631da177e4SLinus Torvalds /* Clear old maps */ 15641da177e4SLinus Torvalds error = -ENOMEM; 15651da177e4SLinus Torvalds munmap_back: 15666597d783SHugh Dickins if (find_vma_links(mm, addr, addr + len, &prev, &rb_link, &rb_parent)) { 15671da177e4SLinus Torvalds if (do_munmap(mm, addr, len)) 15681da177e4SLinus Torvalds return -ENOMEM; 15691da177e4SLinus Torvalds goto munmap_back; 15701da177e4SLinus Torvalds } 15711da177e4SLinus Torvalds 1572fc8744adSLinus Torvalds /* 15731da177e4SLinus Torvalds * Private writable mapping: check memory availability 15741da177e4SLinus Torvalds */ 15755a6fe125SMel Gorman if (accountable_mapping(file, vm_flags)) { 15761da177e4SLinus Torvalds charged = len >> PAGE_SHIFT; 1577191c5424SAl Viro if (security_vm_enough_memory_mm(mm, charged)) 15781da177e4SLinus Torvalds return -ENOMEM; 15791da177e4SLinus Torvalds vm_flags |= VM_ACCOUNT; 15801da177e4SLinus Torvalds } 15811da177e4SLinus Torvalds 15821da177e4SLinus Torvalds /* 1583de33c8dbSLinus Torvalds * Can we just expand an old mapping? 15841da177e4SLinus Torvalds */ 1585de33c8dbSLinus Torvalds vma = vma_merge(mm, prev, addr, addr + len, vm_flags, NULL, file, pgoff, NULL); 1586ba470de4SRik van Riel if (vma) 15871da177e4SLinus Torvalds goto out; 15881da177e4SLinus Torvalds 15891da177e4SLinus Torvalds /* 15901da177e4SLinus Torvalds * Determine the object being mapped and call the appropriate 15911da177e4SLinus Torvalds * specific mapper. the address has already been validated, but 15921da177e4SLinus Torvalds * not unmapped, but the maps are removed from the list. 15931da177e4SLinus Torvalds */ 1594c5e3b83eSPekka Enberg vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL); 15951da177e4SLinus Torvalds if (!vma) { 15961da177e4SLinus Torvalds error = -ENOMEM; 15971da177e4SLinus Torvalds goto unacct_error; 15981da177e4SLinus Torvalds } 15991da177e4SLinus Torvalds 16001da177e4SLinus Torvalds vma->vm_mm = mm; 16011da177e4SLinus Torvalds vma->vm_start = addr; 16021da177e4SLinus Torvalds vma->vm_end = addr + len; 16031da177e4SLinus Torvalds vma->vm_flags = vm_flags; 16043ed75eb8SColy Li vma->vm_page_prot = vm_get_page_prot(vm_flags); 16051da177e4SLinus Torvalds vma->vm_pgoff = pgoff; 16065beb4930SRik van Riel INIT_LIST_HEAD(&vma->anon_vma_chain); 16071da177e4SLinus Torvalds 16081da177e4SLinus Torvalds if (file) { 16091da177e4SLinus Torvalds if (vm_flags & VM_DENYWRITE) { 16101da177e4SLinus Torvalds error = deny_write_access(file); 16111da177e4SLinus Torvalds if (error) 16121da177e4SLinus Torvalds goto free_vma; 16131da177e4SLinus Torvalds } 16144bb5f5d9SDavid Herrmann if (vm_flags & VM_SHARED) { 16154bb5f5d9SDavid Herrmann error = mapping_map_writable(file->f_mapping); 16164bb5f5d9SDavid Herrmann if (error) 16174bb5f5d9SDavid Herrmann goto allow_write_and_free_vma; 16184bb5f5d9SDavid Herrmann } 16194bb5f5d9SDavid Herrmann 16204bb5f5d9SDavid Herrmann /* ->mmap() can change vma->vm_file, but must guarantee that 16214bb5f5d9SDavid Herrmann * vma_link() below can deny write-access if VM_DENYWRITE is set 16224bb5f5d9SDavid Herrmann * and map writably if VM_SHARED is set. This usually means the 16234bb5f5d9SDavid Herrmann * new file must not have been exposed to user-space, yet. 16244bb5f5d9SDavid Herrmann */ 1625cb0942b8SAl Viro vma->vm_file = get_file(file); 16261da177e4SLinus Torvalds error = file->f_op->mmap(file, vma); 16271da177e4SLinus Torvalds if (error) 16281da177e4SLinus Torvalds goto unmap_and_free_vma; 16291da177e4SLinus Torvalds 16301da177e4SLinus Torvalds /* Can addr have changed?? 16311da177e4SLinus Torvalds * 16321da177e4SLinus Torvalds * Answer: Yes, several device drivers can do it in their 16331da177e4SLinus Torvalds * f_op->mmap method. -DaveM 16342897b4d2SJoonsoo Kim * Bug: If addr is changed, prev, rb_link, rb_parent should 16352897b4d2SJoonsoo Kim * be updated for vma_link() 16361da177e4SLinus Torvalds */ 16372897b4d2SJoonsoo Kim WARN_ON_ONCE(addr != vma->vm_start); 16382897b4d2SJoonsoo Kim 16391da177e4SLinus Torvalds addr = vma->vm_start; 16401da177e4SLinus Torvalds vm_flags = vma->vm_flags; 1641f8dbf0a7SHuang Shijie } else if (vm_flags & VM_SHARED) { 1642f8dbf0a7SHuang Shijie error = shmem_zero_setup(vma); 1643f8dbf0a7SHuang Shijie if (error) 1644f8dbf0a7SHuang Shijie goto free_vma; 1645f8dbf0a7SHuang Shijie } 16461da177e4SLinus Torvalds 16474d3d5b41SOleg Nesterov vma_link(mm, vma, prev, rb_link, rb_parent); 16484d3d5b41SOleg Nesterov /* Once vma denies write, undo our temporary denial count */ 16494bb5f5d9SDavid Herrmann if (file) { 16504bb5f5d9SDavid Herrmann if (vm_flags & VM_SHARED) 16514bb5f5d9SDavid Herrmann mapping_unmap_writable(file->f_mapping); 1652e8686772SOleg Nesterov if (vm_flags & VM_DENYWRITE) 1653e8686772SOleg Nesterov allow_write_access(file); 16544bb5f5d9SDavid Herrmann } 1655e8686772SOleg Nesterov file = vma->vm_file; 16561da177e4SLinus Torvalds out: 1657cdd6c482SIngo Molnar perf_event_mmap(vma); 16580a4a9391SPeter Zijlstra 1659ab50b8edSHugh Dickins vm_stat_account(mm, vm_flags, file, len >> PAGE_SHIFT); 16601da177e4SLinus Torvalds if (vm_flags & VM_LOCKED) { 1661bebeb3d6SMichel Lespinasse if (!((vm_flags & VM_SPECIAL) || is_vm_hugetlb_page(vma) || 1662bebeb3d6SMichel Lespinasse vma == get_gate_vma(current->mm))) 166306f9d8c2SKOSAKI Motohiro mm->locked_vm += (len >> PAGE_SHIFT); 1664bebeb3d6SMichel Lespinasse else 1665bebeb3d6SMichel Lespinasse vma->vm_flags &= ~VM_LOCKED; 1666bebeb3d6SMichel Lespinasse } 16672b144498SSrikar Dronamraju 1668c7a3a88cSOleg Nesterov if (file) 1669c7a3a88cSOleg Nesterov uprobe_mmap(vma); 16702b144498SSrikar Dronamraju 1671d9104d1cSCyrill Gorcunov /* 1672d9104d1cSCyrill Gorcunov * New (or expanded) vma always get soft dirty status. 1673d9104d1cSCyrill Gorcunov * Otherwise user-space soft-dirty page tracker won't 1674d9104d1cSCyrill Gorcunov * be able to distinguish situation when vma area unmapped, 1675d9104d1cSCyrill Gorcunov * then new mapped in-place (which must be aimed as 1676d9104d1cSCyrill Gorcunov * a completely new data area). 1677d9104d1cSCyrill Gorcunov */ 1678d9104d1cSCyrill Gorcunov vma->vm_flags |= VM_SOFTDIRTY; 1679d9104d1cSCyrill Gorcunov 168064e45507SPeter Feiner vma_set_page_prot(vma); 168164e45507SPeter Feiner 16821da177e4SLinus Torvalds return addr; 16831da177e4SLinus Torvalds 16841da177e4SLinus Torvalds unmap_and_free_vma: 16851da177e4SLinus Torvalds vma->vm_file = NULL; 16861da177e4SLinus Torvalds fput(file); 16871da177e4SLinus Torvalds 16881da177e4SLinus Torvalds /* Undo any partial mapping done by a device driver. */ 1689e0da382cSHugh Dickins unmap_region(mm, vma, prev, vma->vm_start, vma->vm_end); 1690e0da382cSHugh Dickins charged = 0; 16914bb5f5d9SDavid Herrmann if (vm_flags & VM_SHARED) 16924bb5f5d9SDavid Herrmann mapping_unmap_writable(file->f_mapping); 16934bb5f5d9SDavid Herrmann allow_write_and_free_vma: 16944bb5f5d9SDavid Herrmann if (vm_flags & VM_DENYWRITE) 16954bb5f5d9SDavid Herrmann allow_write_access(file); 16961da177e4SLinus Torvalds free_vma: 16971da177e4SLinus Torvalds kmem_cache_free(vm_area_cachep, vma); 16981da177e4SLinus Torvalds unacct_error: 16991da177e4SLinus Torvalds if (charged) 17001da177e4SLinus Torvalds vm_unacct_memory(charged); 17011da177e4SLinus Torvalds return error; 17021da177e4SLinus Torvalds } 17031da177e4SLinus Torvalds 1704db4fbfb9SMichel Lespinasse unsigned long unmapped_area(struct vm_unmapped_area_info *info) 1705db4fbfb9SMichel Lespinasse { 1706db4fbfb9SMichel Lespinasse /* 1707db4fbfb9SMichel Lespinasse * We implement the search by looking for an rbtree node that 1708db4fbfb9SMichel Lespinasse * immediately follows a suitable gap. That is, 1709db4fbfb9SMichel Lespinasse * - gap_start = vma->vm_prev->vm_end <= info->high_limit - length; 1710db4fbfb9SMichel Lespinasse * - gap_end = vma->vm_start >= info->low_limit + length; 1711db4fbfb9SMichel Lespinasse * - gap_end - gap_start >= length 1712db4fbfb9SMichel Lespinasse */ 1713db4fbfb9SMichel Lespinasse 1714db4fbfb9SMichel Lespinasse struct mm_struct *mm = current->mm; 1715db4fbfb9SMichel Lespinasse struct vm_area_struct *vma; 1716db4fbfb9SMichel Lespinasse unsigned long length, low_limit, high_limit, gap_start, gap_end; 1717db4fbfb9SMichel Lespinasse 1718db4fbfb9SMichel Lespinasse /* Adjust search length to account for worst case alignment overhead */ 1719db4fbfb9SMichel Lespinasse length = info->length + info->align_mask; 1720db4fbfb9SMichel Lespinasse if (length < info->length) 1721db4fbfb9SMichel Lespinasse return -ENOMEM; 1722db4fbfb9SMichel Lespinasse 1723db4fbfb9SMichel Lespinasse /* Adjust search limits by the desired length */ 1724db4fbfb9SMichel Lespinasse if (info->high_limit < length) 1725db4fbfb9SMichel Lespinasse return -ENOMEM; 1726db4fbfb9SMichel Lespinasse high_limit = info->high_limit - length; 1727db4fbfb9SMichel Lespinasse 1728db4fbfb9SMichel Lespinasse if (info->low_limit > high_limit) 1729db4fbfb9SMichel Lespinasse return -ENOMEM; 1730db4fbfb9SMichel Lespinasse low_limit = info->low_limit + length; 1731db4fbfb9SMichel Lespinasse 1732db4fbfb9SMichel Lespinasse /* Check if rbtree root looks promising */ 1733db4fbfb9SMichel Lespinasse if (RB_EMPTY_ROOT(&mm->mm_rb)) 1734db4fbfb9SMichel Lespinasse goto check_highest; 1735db4fbfb9SMichel Lespinasse vma = rb_entry(mm->mm_rb.rb_node, struct vm_area_struct, vm_rb); 1736db4fbfb9SMichel Lespinasse if (vma->rb_subtree_gap < length) 1737db4fbfb9SMichel Lespinasse goto check_highest; 1738db4fbfb9SMichel Lespinasse 1739db4fbfb9SMichel Lespinasse while (true) { 1740db4fbfb9SMichel Lespinasse /* Visit left subtree if it looks promising */ 1741db4fbfb9SMichel Lespinasse gap_end = vma->vm_start; 1742db4fbfb9SMichel Lespinasse if (gap_end >= low_limit && vma->vm_rb.rb_left) { 1743db4fbfb9SMichel Lespinasse struct vm_area_struct *left = 1744db4fbfb9SMichel Lespinasse rb_entry(vma->vm_rb.rb_left, 1745db4fbfb9SMichel Lespinasse struct vm_area_struct, vm_rb); 1746db4fbfb9SMichel Lespinasse if (left->rb_subtree_gap >= length) { 1747db4fbfb9SMichel Lespinasse vma = left; 1748db4fbfb9SMichel Lespinasse continue; 1749db4fbfb9SMichel Lespinasse } 1750db4fbfb9SMichel Lespinasse } 1751db4fbfb9SMichel Lespinasse 1752db4fbfb9SMichel Lespinasse gap_start = vma->vm_prev ? vma->vm_prev->vm_end : 0; 1753db4fbfb9SMichel Lespinasse check_current: 1754db4fbfb9SMichel Lespinasse /* Check if current node has a suitable gap */ 1755db4fbfb9SMichel Lespinasse if (gap_start > high_limit) 1756db4fbfb9SMichel Lespinasse return -ENOMEM; 1757db4fbfb9SMichel Lespinasse if (gap_end >= low_limit && gap_end - gap_start >= length) 1758db4fbfb9SMichel Lespinasse goto found; 1759db4fbfb9SMichel Lespinasse 1760db4fbfb9SMichel Lespinasse /* Visit right subtree if it looks promising */ 1761db4fbfb9SMichel Lespinasse if (vma->vm_rb.rb_right) { 1762db4fbfb9SMichel Lespinasse struct vm_area_struct *right = 1763db4fbfb9SMichel Lespinasse rb_entry(vma->vm_rb.rb_right, 1764db4fbfb9SMichel Lespinasse struct vm_area_struct, vm_rb); 1765db4fbfb9SMichel Lespinasse if (right->rb_subtree_gap >= length) { 1766db4fbfb9SMichel Lespinasse vma = right; 1767db4fbfb9SMichel Lespinasse continue; 1768db4fbfb9SMichel Lespinasse } 1769db4fbfb9SMichel Lespinasse } 1770db4fbfb9SMichel Lespinasse 1771db4fbfb9SMichel Lespinasse /* Go back up the rbtree to find next candidate node */ 1772db4fbfb9SMichel Lespinasse while (true) { 1773db4fbfb9SMichel Lespinasse struct rb_node *prev = &vma->vm_rb; 1774db4fbfb9SMichel Lespinasse if (!rb_parent(prev)) 1775db4fbfb9SMichel Lespinasse goto check_highest; 1776db4fbfb9SMichel Lespinasse vma = rb_entry(rb_parent(prev), 1777db4fbfb9SMichel Lespinasse struct vm_area_struct, vm_rb); 1778db4fbfb9SMichel Lespinasse if (prev == vma->vm_rb.rb_left) { 1779db4fbfb9SMichel Lespinasse gap_start = vma->vm_prev->vm_end; 1780db4fbfb9SMichel Lespinasse gap_end = vma->vm_start; 1781db4fbfb9SMichel Lespinasse goto check_current; 1782db4fbfb9SMichel Lespinasse } 1783db4fbfb9SMichel Lespinasse } 1784db4fbfb9SMichel Lespinasse } 1785db4fbfb9SMichel Lespinasse 1786db4fbfb9SMichel Lespinasse check_highest: 1787db4fbfb9SMichel Lespinasse /* Check highest gap, which does not precede any rbtree node */ 1788db4fbfb9SMichel Lespinasse gap_start = mm->highest_vm_end; 1789db4fbfb9SMichel Lespinasse gap_end = ULONG_MAX; /* Only for VM_BUG_ON below */ 1790db4fbfb9SMichel Lespinasse if (gap_start > high_limit) 1791db4fbfb9SMichel Lespinasse return -ENOMEM; 1792db4fbfb9SMichel Lespinasse 1793db4fbfb9SMichel Lespinasse found: 1794db4fbfb9SMichel Lespinasse /* We found a suitable gap. Clip it with the original low_limit. */ 1795db4fbfb9SMichel Lespinasse if (gap_start < info->low_limit) 1796db4fbfb9SMichel Lespinasse gap_start = info->low_limit; 1797db4fbfb9SMichel Lespinasse 1798db4fbfb9SMichel Lespinasse /* Adjust gap address to the desired alignment */ 1799db4fbfb9SMichel Lespinasse gap_start += (info->align_offset - gap_start) & info->align_mask; 1800db4fbfb9SMichel Lespinasse 1801db4fbfb9SMichel Lespinasse VM_BUG_ON(gap_start + info->length > info->high_limit); 1802db4fbfb9SMichel Lespinasse VM_BUG_ON(gap_start + info->length > gap_end); 1803db4fbfb9SMichel Lespinasse return gap_start; 1804db4fbfb9SMichel Lespinasse } 1805db4fbfb9SMichel Lespinasse 1806db4fbfb9SMichel Lespinasse unsigned long unmapped_area_topdown(struct vm_unmapped_area_info *info) 1807db4fbfb9SMichel Lespinasse { 1808db4fbfb9SMichel Lespinasse struct mm_struct *mm = current->mm; 1809db4fbfb9SMichel Lespinasse struct vm_area_struct *vma; 1810db4fbfb9SMichel Lespinasse unsigned long length, low_limit, high_limit, gap_start, gap_end; 1811db4fbfb9SMichel Lespinasse 1812db4fbfb9SMichel Lespinasse /* Adjust search length to account for worst case alignment overhead */ 1813db4fbfb9SMichel Lespinasse length = info->length + info->align_mask; 1814db4fbfb9SMichel Lespinasse if (length < info->length) 1815db4fbfb9SMichel Lespinasse return -ENOMEM; 1816db4fbfb9SMichel Lespinasse 1817db4fbfb9SMichel Lespinasse /* 1818db4fbfb9SMichel Lespinasse * Adjust search limits by the desired length. 1819db4fbfb9SMichel Lespinasse * See implementation comment at top of unmapped_area(). 1820db4fbfb9SMichel Lespinasse */ 1821db4fbfb9SMichel Lespinasse gap_end = info->high_limit; 1822db4fbfb9SMichel Lespinasse if (gap_end < length) 1823db4fbfb9SMichel Lespinasse return -ENOMEM; 1824db4fbfb9SMichel Lespinasse high_limit = gap_end - length; 1825db4fbfb9SMichel Lespinasse 1826db4fbfb9SMichel Lespinasse if (info->low_limit > high_limit) 1827db4fbfb9SMichel Lespinasse return -ENOMEM; 1828db4fbfb9SMichel Lespinasse low_limit = info->low_limit + length; 1829db4fbfb9SMichel Lespinasse 1830db4fbfb9SMichel Lespinasse /* Check highest gap, which does not precede any rbtree node */ 1831db4fbfb9SMichel Lespinasse gap_start = mm->highest_vm_end; 1832db4fbfb9SMichel Lespinasse if (gap_start <= high_limit) 1833db4fbfb9SMichel Lespinasse goto found_highest; 1834db4fbfb9SMichel Lespinasse 1835db4fbfb9SMichel Lespinasse /* Check if rbtree root looks promising */ 1836db4fbfb9SMichel Lespinasse if (RB_EMPTY_ROOT(&mm->mm_rb)) 1837db4fbfb9SMichel Lespinasse return -ENOMEM; 1838db4fbfb9SMichel Lespinasse vma = rb_entry(mm->mm_rb.rb_node, struct vm_area_struct, vm_rb); 1839db4fbfb9SMichel Lespinasse if (vma->rb_subtree_gap < length) 1840db4fbfb9SMichel Lespinasse return -ENOMEM; 1841db4fbfb9SMichel Lespinasse 1842db4fbfb9SMichel Lespinasse while (true) { 1843db4fbfb9SMichel Lespinasse /* Visit right subtree if it looks promising */ 1844db4fbfb9SMichel Lespinasse gap_start = vma->vm_prev ? vma->vm_prev->vm_end : 0; 1845db4fbfb9SMichel Lespinasse if (gap_start <= high_limit && vma->vm_rb.rb_right) { 1846db4fbfb9SMichel Lespinasse struct vm_area_struct *right = 1847db4fbfb9SMichel Lespinasse rb_entry(vma->vm_rb.rb_right, 1848db4fbfb9SMichel Lespinasse struct vm_area_struct, vm_rb); 1849db4fbfb9SMichel Lespinasse if (right->rb_subtree_gap >= length) { 1850db4fbfb9SMichel Lespinasse vma = right; 1851db4fbfb9SMichel Lespinasse continue; 1852db4fbfb9SMichel Lespinasse } 1853db4fbfb9SMichel Lespinasse } 1854db4fbfb9SMichel Lespinasse 1855db4fbfb9SMichel Lespinasse check_current: 1856db4fbfb9SMichel Lespinasse /* Check if current node has a suitable gap */ 1857db4fbfb9SMichel Lespinasse gap_end = vma->vm_start; 1858db4fbfb9SMichel Lespinasse if (gap_end < low_limit) 1859db4fbfb9SMichel Lespinasse return -ENOMEM; 1860db4fbfb9SMichel Lespinasse if (gap_start <= high_limit && gap_end - gap_start >= length) 1861db4fbfb9SMichel Lespinasse goto found; 1862db4fbfb9SMichel Lespinasse 1863db4fbfb9SMichel Lespinasse /* Visit left subtree if it looks promising */ 1864db4fbfb9SMichel Lespinasse if (vma->vm_rb.rb_left) { 1865db4fbfb9SMichel Lespinasse struct vm_area_struct *left = 1866db4fbfb9SMichel Lespinasse rb_entry(vma->vm_rb.rb_left, 1867db4fbfb9SMichel Lespinasse struct vm_area_struct, vm_rb); 1868db4fbfb9SMichel Lespinasse if (left->rb_subtree_gap >= length) { 1869db4fbfb9SMichel Lespinasse vma = left; 1870db4fbfb9SMichel Lespinasse continue; 1871db4fbfb9SMichel Lespinasse } 1872db4fbfb9SMichel Lespinasse } 1873db4fbfb9SMichel Lespinasse 1874db4fbfb9SMichel Lespinasse /* Go back up the rbtree to find next candidate node */ 1875db4fbfb9SMichel Lespinasse while (true) { 1876db4fbfb9SMichel Lespinasse struct rb_node *prev = &vma->vm_rb; 1877db4fbfb9SMichel Lespinasse if (!rb_parent(prev)) 1878db4fbfb9SMichel Lespinasse return -ENOMEM; 1879db4fbfb9SMichel Lespinasse vma = rb_entry(rb_parent(prev), 1880db4fbfb9SMichel Lespinasse struct vm_area_struct, vm_rb); 1881db4fbfb9SMichel Lespinasse if (prev == vma->vm_rb.rb_right) { 1882db4fbfb9SMichel Lespinasse gap_start = vma->vm_prev ? 1883db4fbfb9SMichel Lespinasse vma->vm_prev->vm_end : 0; 1884db4fbfb9SMichel Lespinasse goto check_current; 1885db4fbfb9SMichel Lespinasse } 1886db4fbfb9SMichel Lespinasse } 1887db4fbfb9SMichel Lespinasse } 1888db4fbfb9SMichel Lespinasse 1889db4fbfb9SMichel Lespinasse found: 1890db4fbfb9SMichel Lespinasse /* We found a suitable gap. Clip it with the original high_limit. */ 1891db4fbfb9SMichel Lespinasse if (gap_end > info->high_limit) 1892db4fbfb9SMichel Lespinasse gap_end = info->high_limit; 1893db4fbfb9SMichel Lespinasse 1894db4fbfb9SMichel Lespinasse found_highest: 1895db4fbfb9SMichel Lespinasse /* Compute highest gap address at the desired alignment */ 1896db4fbfb9SMichel Lespinasse gap_end -= info->length; 1897db4fbfb9SMichel Lespinasse gap_end -= (gap_end - info->align_offset) & info->align_mask; 1898db4fbfb9SMichel Lespinasse 1899db4fbfb9SMichel Lespinasse VM_BUG_ON(gap_end < info->low_limit); 1900db4fbfb9SMichel Lespinasse VM_BUG_ON(gap_end < gap_start); 1901db4fbfb9SMichel Lespinasse return gap_end; 1902db4fbfb9SMichel Lespinasse } 1903db4fbfb9SMichel Lespinasse 19041da177e4SLinus Torvalds /* Get an address range which is currently unmapped. 19051da177e4SLinus Torvalds * For shmat() with addr=0. 19061da177e4SLinus Torvalds * 19071da177e4SLinus Torvalds * Ugly calling convention alert: 19081da177e4SLinus Torvalds * Return value with the low bits set means error value, 19091da177e4SLinus Torvalds * ie 19101da177e4SLinus Torvalds * if (ret & ~PAGE_MASK) 19111da177e4SLinus Torvalds * error = ret; 19121da177e4SLinus Torvalds * 19131da177e4SLinus Torvalds * This function "knows" that -ENOMEM has the bits set. 19141da177e4SLinus Torvalds */ 19151da177e4SLinus Torvalds #ifndef HAVE_ARCH_UNMAPPED_AREA 19161da177e4SLinus Torvalds unsigned long 19171da177e4SLinus Torvalds arch_get_unmapped_area(struct file *filp, unsigned long addr, 19181da177e4SLinus Torvalds unsigned long len, unsigned long pgoff, unsigned long flags) 19191da177e4SLinus Torvalds { 19201da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 19211da177e4SLinus Torvalds struct vm_area_struct *vma; 1922db4fbfb9SMichel Lespinasse struct vm_unmapped_area_info info; 19231da177e4SLinus Torvalds 19242afc745fSAkira Takeuchi if (len > TASK_SIZE - mmap_min_addr) 19251da177e4SLinus Torvalds return -ENOMEM; 19261da177e4SLinus Torvalds 192706abdfb4SBenjamin Herrenschmidt if (flags & MAP_FIXED) 192806abdfb4SBenjamin Herrenschmidt return addr; 192906abdfb4SBenjamin Herrenschmidt 19301da177e4SLinus Torvalds if (addr) { 19311da177e4SLinus Torvalds addr = PAGE_ALIGN(addr); 19321da177e4SLinus Torvalds vma = find_vma(mm, addr); 19332afc745fSAkira Takeuchi if (TASK_SIZE - len >= addr && addr >= mmap_min_addr && 19341da177e4SLinus Torvalds (!vma || addr + len <= vma->vm_start)) 19351da177e4SLinus Torvalds return addr; 19361da177e4SLinus Torvalds } 19371da177e4SLinus Torvalds 1938db4fbfb9SMichel Lespinasse info.flags = 0; 1939db4fbfb9SMichel Lespinasse info.length = len; 19404e99b021SHeiko Carstens info.low_limit = mm->mmap_base; 1941db4fbfb9SMichel Lespinasse info.high_limit = TASK_SIZE; 1942db4fbfb9SMichel Lespinasse info.align_mask = 0; 1943db4fbfb9SMichel Lespinasse return vm_unmapped_area(&info); 19441da177e4SLinus Torvalds } 19451da177e4SLinus Torvalds #endif 19461da177e4SLinus Torvalds 19471da177e4SLinus Torvalds /* 19481da177e4SLinus Torvalds * This mmap-allocator allocates new areas top-down from below the 19491da177e4SLinus Torvalds * stack's low limit (the base): 19501da177e4SLinus Torvalds */ 19511da177e4SLinus Torvalds #ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN 19521da177e4SLinus Torvalds unsigned long 19531da177e4SLinus Torvalds arch_get_unmapped_area_topdown(struct file *filp, const unsigned long addr0, 19541da177e4SLinus Torvalds const unsigned long len, const unsigned long pgoff, 19551da177e4SLinus Torvalds const unsigned long flags) 19561da177e4SLinus Torvalds { 19571da177e4SLinus Torvalds struct vm_area_struct *vma; 19581da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 1959db4fbfb9SMichel Lespinasse unsigned long addr = addr0; 1960db4fbfb9SMichel Lespinasse struct vm_unmapped_area_info info; 19611da177e4SLinus Torvalds 19621da177e4SLinus Torvalds /* requested length too big for entire address space */ 19632afc745fSAkira Takeuchi if (len > TASK_SIZE - mmap_min_addr) 19641da177e4SLinus Torvalds return -ENOMEM; 19651da177e4SLinus Torvalds 196606abdfb4SBenjamin Herrenschmidt if (flags & MAP_FIXED) 196706abdfb4SBenjamin Herrenschmidt return addr; 196806abdfb4SBenjamin Herrenschmidt 19691da177e4SLinus Torvalds /* requesting a specific address */ 19701da177e4SLinus Torvalds if (addr) { 19711da177e4SLinus Torvalds addr = PAGE_ALIGN(addr); 19721da177e4SLinus Torvalds vma = find_vma(mm, addr); 19732afc745fSAkira Takeuchi if (TASK_SIZE - len >= addr && addr >= mmap_min_addr && 19741da177e4SLinus Torvalds (!vma || addr + len <= vma->vm_start)) 19751da177e4SLinus Torvalds return addr; 19761da177e4SLinus Torvalds } 19771da177e4SLinus Torvalds 1978db4fbfb9SMichel Lespinasse info.flags = VM_UNMAPPED_AREA_TOPDOWN; 1979db4fbfb9SMichel Lespinasse info.length = len; 19802afc745fSAkira Takeuchi info.low_limit = max(PAGE_SIZE, mmap_min_addr); 1981db4fbfb9SMichel Lespinasse info.high_limit = mm->mmap_base; 1982db4fbfb9SMichel Lespinasse info.align_mask = 0; 1983db4fbfb9SMichel Lespinasse addr = vm_unmapped_area(&info); 1984b716ad95SXiao Guangrong 19851da177e4SLinus Torvalds /* 19861da177e4SLinus Torvalds * A failed mmap() very likely causes application failure, 19871da177e4SLinus Torvalds * so fall back to the bottom-up function here. This scenario 19881da177e4SLinus Torvalds * can happen with large stack limits and large mmap() 19891da177e4SLinus Torvalds * allocations. 19901da177e4SLinus Torvalds */ 1991db4fbfb9SMichel Lespinasse if (addr & ~PAGE_MASK) { 1992db4fbfb9SMichel Lespinasse VM_BUG_ON(addr != -ENOMEM); 1993db4fbfb9SMichel Lespinasse info.flags = 0; 1994db4fbfb9SMichel Lespinasse info.low_limit = TASK_UNMAPPED_BASE; 1995db4fbfb9SMichel Lespinasse info.high_limit = TASK_SIZE; 1996db4fbfb9SMichel Lespinasse addr = vm_unmapped_area(&info); 1997db4fbfb9SMichel Lespinasse } 19981da177e4SLinus Torvalds 19991da177e4SLinus Torvalds return addr; 20001da177e4SLinus Torvalds } 20011da177e4SLinus Torvalds #endif 20021da177e4SLinus Torvalds 20031da177e4SLinus Torvalds unsigned long 20041da177e4SLinus Torvalds get_unmapped_area(struct file *file, unsigned long addr, unsigned long len, 20051da177e4SLinus Torvalds unsigned long pgoff, unsigned long flags) 20061da177e4SLinus Torvalds { 200706abdfb4SBenjamin Herrenschmidt unsigned long (*get_area)(struct file *, unsigned long, 200806abdfb4SBenjamin Herrenschmidt unsigned long, unsigned long, unsigned long); 200907ab67c8SLinus Torvalds 20109206de95SAl Viro unsigned long error = arch_mmap_check(addr, len, flags); 20119206de95SAl Viro if (error) 20129206de95SAl Viro return error; 20139206de95SAl Viro 20149206de95SAl Viro /* Careful about overflows.. */ 20159206de95SAl Viro if (len > TASK_SIZE) 20169206de95SAl Viro return -ENOMEM; 20179206de95SAl Viro 201807ab67c8SLinus Torvalds get_area = current->mm->get_unmapped_area; 201972c2d531SAl Viro if (file && file->f_op->get_unmapped_area) 202007ab67c8SLinus Torvalds get_area = file->f_op->get_unmapped_area; 202107ab67c8SLinus Torvalds addr = get_area(file, addr, len, pgoff, flags); 202207ab67c8SLinus Torvalds if (IS_ERR_VALUE(addr)) 202307ab67c8SLinus Torvalds return addr; 202407ab67c8SLinus Torvalds 20251da177e4SLinus Torvalds if (addr > TASK_SIZE - len) 20261da177e4SLinus Torvalds return -ENOMEM; 20271da177e4SLinus Torvalds if (addr & ~PAGE_MASK) 20281da177e4SLinus Torvalds return -EINVAL; 202906abdfb4SBenjamin Herrenschmidt 20309ac4ed4bSAl Viro addr = arch_rebalance_pgtables(addr, len); 20319ac4ed4bSAl Viro error = security_mmap_addr(addr); 20329ac4ed4bSAl Viro return error ? error : addr; 20331da177e4SLinus Torvalds } 20341da177e4SLinus Torvalds 20351da177e4SLinus Torvalds EXPORT_SYMBOL(get_unmapped_area); 20361da177e4SLinus Torvalds 20371da177e4SLinus Torvalds /* Look up the first VMA which satisfies addr < vm_end, NULL if none. */ 20381da177e4SLinus Torvalds struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr) 20391da177e4SLinus Torvalds { 2040615d6e87SDavidlohr Bueso struct rb_node *rb_node; 2041615d6e87SDavidlohr Bueso struct vm_area_struct *vma; 20421da177e4SLinus Torvalds 20431da177e4SLinus Torvalds /* Check the cache first. */ 2044615d6e87SDavidlohr Bueso vma = vmacache_find(mm, addr); 2045615d6e87SDavidlohr Bueso if (likely(vma)) 2046615d6e87SDavidlohr Bueso return vma; 20471da177e4SLinus Torvalds 20481da177e4SLinus Torvalds rb_node = mm->mm_rb.rb_node; 20491da177e4SLinus Torvalds vma = NULL; 20501da177e4SLinus Torvalds 20511da177e4SLinus Torvalds while (rb_node) { 2052615d6e87SDavidlohr Bueso struct vm_area_struct *tmp; 20531da177e4SLinus Torvalds 2054615d6e87SDavidlohr Bueso tmp = rb_entry(rb_node, struct vm_area_struct, vm_rb); 20551da177e4SLinus Torvalds 2056615d6e87SDavidlohr Bueso if (tmp->vm_end > addr) { 2057615d6e87SDavidlohr Bueso vma = tmp; 2058615d6e87SDavidlohr Bueso if (tmp->vm_start <= addr) 20591da177e4SLinus Torvalds break; 20601da177e4SLinus Torvalds rb_node = rb_node->rb_left; 20611da177e4SLinus Torvalds } else 20621da177e4SLinus Torvalds rb_node = rb_node->rb_right; 20631da177e4SLinus Torvalds } 2064615d6e87SDavidlohr Bueso 20651da177e4SLinus Torvalds if (vma) 2066615d6e87SDavidlohr Bueso vmacache_update(addr, vma); 20671da177e4SLinus Torvalds return vma; 20681da177e4SLinus Torvalds } 20691da177e4SLinus Torvalds 20701da177e4SLinus Torvalds EXPORT_SYMBOL(find_vma); 20711da177e4SLinus Torvalds 20726bd4837dSKOSAKI Motohiro /* 20736bd4837dSKOSAKI Motohiro * Same as find_vma, but also return a pointer to the previous VMA in *pprev. 20746bd4837dSKOSAKI Motohiro */ 20751da177e4SLinus Torvalds struct vm_area_struct * 20761da177e4SLinus Torvalds find_vma_prev(struct mm_struct *mm, unsigned long addr, 20771da177e4SLinus Torvalds struct vm_area_struct **pprev) 20781da177e4SLinus Torvalds { 20796bd4837dSKOSAKI Motohiro struct vm_area_struct *vma; 20801da177e4SLinus Torvalds 20816bd4837dSKOSAKI Motohiro vma = find_vma(mm, addr); 208283cd904dSMikulas Patocka if (vma) { 208383cd904dSMikulas Patocka *pprev = vma->vm_prev; 208483cd904dSMikulas Patocka } else { 208583cd904dSMikulas Patocka struct rb_node *rb_node = mm->mm_rb.rb_node; 208683cd904dSMikulas Patocka *pprev = NULL; 208783cd904dSMikulas Patocka while (rb_node) { 208883cd904dSMikulas Patocka *pprev = rb_entry(rb_node, struct vm_area_struct, vm_rb); 208983cd904dSMikulas Patocka rb_node = rb_node->rb_right; 209083cd904dSMikulas Patocka } 209183cd904dSMikulas Patocka } 20926bd4837dSKOSAKI Motohiro return vma; 20931da177e4SLinus Torvalds } 20941da177e4SLinus Torvalds 20951da177e4SLinus Torvalds /* 20961da177e4SLinus Torvalds * Verify that the stack growth is acceptable and 20971da177e4SLinus Torvalds * update accounting. This is shared with both the 20981da177e4SLinus Torvalds * grow-up and grow-down cases. 20991da177e4SLinus Torvalds */ 21001da177e4SLinus Torvalds static int acct_stack_growth(struct vm_area_struct *vma, unsigned long size, unsigned long grow) 21011da177e4SLinus Torvalds { 21021da177e4SLinus Torvalds struct mm_struct *mm = vma->vm_mm; 21031da177e4SLinus Torvalds struct rlimit *rlim = current->signal->rlim; 2104690eac53SLinus Torvalds unsigned long new_start, actual_size; 21051da177e4SLinus Torvalds 21061da177e4SLinus Torvalds /* address space limit tests */ 2107119f657cSakpm@osdl.org if (!may_expand_vm(mm, grow)) 21081da177e4SLinus Torvalds return -ENOMEM; 21091da177e4SLinus Torvalds 21101da177e4SLinus Torvalds /* Stack limit test */ 2111690eac53SLinus Torvalds actual_size = size; 2112690eac53SLinus Torvalds if (size && (vma->vm_flags & (VM_GROWSUP | VM_GROWSDOWN))) 2113690eac53SLinus Torvalds actual_size -= PAGE_SIZE; 2114690eac53SLinus Torvalds if (actual_size > ACCESS_ONCE(rlim[RLIMIT_STACK].rlim_cur)) 21151da177e4SLinus Torvalds return -ENOMEM; 21161da177e4SLinus Torvalds 21171da177e4SLinus Torvalds /* mlock limit tests */ 21181da177e4SLinus Torvalds if (vma->vm_flags & VM_LOCKED) { 21191da177e4SLinus Torvalds unsigned long locked; 21201da177e4SLinus Torvalds unsigned long limit; 21211da177e4SLinus Torvalds locked = mm->locked_vm + grow; 212259e99e5bSJiri Slaby limit = ACCESS_ONCE(rlim[RLIMIT_MEMLOCK].rlim_cur); 212359e99e5bSJiri Slaby limit >>= PAGE_SHIFT; 21241da177e4SLinus Torvalds if (locked > limit && !capable(CAP_IPC_LOCK)) 21251da177e4SLinus Torvalds return -ENOMEM; 21261da177e4SLinus Torvalds } 21271da177e4SLinus Torvalds 21280d59a01bSAdam Litke /* Check to ensure the stack will not grow into a hugetlb-only region */ 21290d59a01bSAdam Litke new_start = (vma->vm_flags & VM_GROWSUP) ? vma->vm_start : 21300d59a01bSAdam Litke vma->vm_end - size; 21310d59a01bSAdam Litke if (is_hugepage_only_range(vma->vm_mm, new_start, size)) 21320d59a01bSAdam Litke return -EFAULT; 21330d59a01bSAdam Litke 21341da177e4SLinus Torvalds /* 21351da177e4SLinus Torvalds * Overcommit.. This must be the final test, as it will 21361da177e4SLinus Torvalds * update security statistics. 21371da177e4SLinus Torvalds */ 213805fa199dSHugh Dickins if (security_vm_enough_memory_mm(mm, grow)) 21391da177e4SLinus Torvalds return -ENOMEM; 21401da177e4SLinus Torvalds 21411da177e4SLinus Torvalds /* Ok, everything looks good - let it rip */ 21421da177e4SLinus Torvalds if (vma->vm_flags & VM_LOCKED) 21431da177e4SLinus Torvalds mm->locked_vm += grow; 2144ab50b8edSHugh Dickins vm_stat_account(mm, vma->vm_flags, vma->vm_file, grow); 21451da177e4SLinus Torvalds return 0; 21461da177e4SLinus Torvalds } 21471da177e4SLinus Torvalds 214846dea3d0SHugh Dickins #if defined(CONFIG_STACK_GROWSUP) || defined(CONFIG_IA64) 21491da177e4SLinus Torvalds /* 215046dea3d0SHugh Dickins * PA-RISC uses this for its stack; IA64 for its Register Backing Store. 215146dea3d0SHugh Dickins * vma is the last one with address > vma->vm_end. Have to extend vma. 21521da177e4SLinus Torvalds */ 215346dea3d0SHugh Dickins int expand_upwards(struct vm_area_struct *vma, unsigned long address) 21541da177e4SLinus Torvalds { 21551da177e4SLinus Torvalds int error; 21561da177e4SLinus Torvalds 21571da177e4SLinus Torvalds if (!(vma->vm_flags & VM_GROWSUP)) 21581da177e4SLinus Torvalds return -EFAULT; 21591da177e4SLinus Torvalds 21601da177e4SLinus Torvalds /* 21611da177e4SLinus Torvalds * We must make sure the anon_vma is allocated 21621da177e4SLinus Torvalds * so that the anon_vma locking is not a noop. 21631da177e4SLinus Torvalds */ 21641da177e4SLinus Torvalds if (unlikely(anon_vma_prepare(vma))) 21651da177e4SLinus Torvalds return -ENOMEM; 2166bb4a340eSRik van Riel vma_lock_anon_vma(vma); 21671da177e4SLinus Torvalds 21681da177e4SLinus Torvalds /* 21691da177e4SLinus Torvalds * vma->vm_start/vm_end cannot change under us because the caller 21701da177e4SLinus Torvalds * is required to hold the mmap_sem in read mode. We need the 21711da177e4SLinus Torvalds * anon_vma lock to serialize against concurrent expand_stacks. 217206b32f3aSHelge Deller * Also guard against wrapping around to address 0. 21731da177e4SLinus Torvalds */ 217406b32f3aSHelge Deller if (address < PAGE_ALIGN(address+4)) 217506b32f3aSHelge Deller address = PAGE_ALIGN(address+4); 217606b32f3aSHelge Deller else { 2177bb4a340eSRik van Riel vma_unlock_anon_vma(vma); 217806b32f3aSHelge Deller return -ENOMEM; 217906b32f3aSHelge Deller } 21801da177e4SLinus Torvalds error = 0; 21811da177e4SLinus Torvalds 21821da177e4SLinus Torvalds /* Somebody else might have raced and expanded it already */ 21831da177e4SLinus Torvalds if (address > vma->vm_end) { 21841da177e4SLinus Torvalds unsigned long size, grow; 21851da177e4SLinus Torvalds 21861da177e4SLinus Torvalds size = address - vma->vm_start; 21871da177e4SLinus Torvalds grow = (address - vma->vm_end) >> PAGE_SHIFT; 21881da177e4SLinus Torvalds 218942c36f63SHugh Dickins error = -ENOMEM; 219042c36f63SHugh Dickins if (vma->vm_pgoff + (size >> PAGE_SHIFT) >= vma->vm_pgoff) { 21911da177e4SLinus Torvalds error = acct_stack_growth(vma, size, grow); 21923af9e859SEric B Munson if (!error) { 21934128997bSMichel Lespinasse /* 21944128997bSMichel Lespinasse * vma_gap_update() doesn't support concurrent 21954128997bSMichel Lespinasse * updates, but we only hold a shared mmap_sem 21964128997bSMichel Lespinasse * lock here, so we need to protect against 21974128997bSMichel Lespinasse * concurrent vma expansions. 21984128997bSMichel Lespinasse * vma_lock_anon_vma() doesn't help here, as 21994128997bSMichel Lespinasse * we don't guarantee that all growable vmas 22004128997bSMichel Lespinasse * in a mm share the same root anon vma. 22014128997bSMichel Lespinasse * So, we reuse mm->page_table_lock to guard 22024128997bSMichel Lespinasse * against concurrent vma expansions. 22034128997bSMichel Lespinasse */ 22044128997bSMichel Lespinasse spin_lock(&vma->vm_mm->page_table_lock); 2205bf181b9fSMichel Lespinasse anon_vma_interval_tree_pre_update_vma(vma); 22061da177e4SLinus Torvalds vma->vm_end = address; 2207bf181b9fSMichel Lespinasse anon_vma_interval_tree_post_update_vma(vma); 2208d3737187SMichel Lespinasse if (vma->vm_next) 2209d3737187SMichel Lespinasse vma_gap_update(vma->vm_next); 2210d3737187SMichel Lespinasse else 2211d3737187SMichel Lespinasse vma->vm_mm->highest_vm_end = address; 22124128997bSMichel Lespinasse spin_unlock(&vma->vm_mm->page_table_lock); 22134128997bSMichel Lespinasse 22143af9e859SEric B Munson perf_event_mmap(vma); 22153af9e859SEric B Munson } 22161da177e4SLinus Torvalds } 221742c36f63SHugh Dickins } 2218bb4a340eSRik van Riel vma_unlock_anon_vma(vma); 22196d50e60cSDavid Rientjes khugepaged_enter_vma_merge(vma, vma->vm_flags); 2220ed8ea815SMichel Lespinasse validate_mm(vma->vm_mm); 22211da177e4SLinus Torvalds return error; 22221da177e4SLinus Torvalds } 222346dea3d0SHugh Dickins #endif /* CONFIG_STACK_GROWSUP || CONFIG_IA64 */ 222446dea3d0SHugh Dickins 22251da177e4SLinus Torvalds /* 22261da177e4SLinus Torvalds * vma is the first one with address < vma->vm_start. Have to extend vma. 22271da177e4SLinus Torvalds */ 2228d05f3169SMichal Hocko int expand_downwards(struct vm_area_struct *vma, 2229b6a2fea3SOllie Wild unsigned long address) 22301da177e4SLinus Torvalds { 22311da177e4SLinus Torvalds int error; 22321da177e4SLinus Torvalds 22331da177e4SLinus Torvalds /* 22341da177e4SLinus Torvalds * We must make sure the anon_vma is allocated 22351da177e4SLinus Torvalds * so that the anon_vma locking is not a noop. 22361da177e4SLinus Torvalds */ 22371da177e4SLinus Torvalds if (unlikely(anon_vma_prepare(vma))) 22381da177e4SLinus Torvalds return -ENOMEM; 22398869477aSEric Paris 22408869477aSEric Paris address &= PAGE_MASK; 2241e5467859SAl Viro error = security_mmap_addr(address); 22428869477aSEric Paris if (error) 22438869477aSEric Paris return error; 22448869477aSEric Paris 2245bb4a340eSRik van Riel vma_lock_anon_vma(vma); 22461da177e4SLinus Torvalds 22471da177e4SLinus Torvalds /* 22481da177e4SLinus Torvalds * vma->vm_start/vm_end cannot change under us because the caller 22491da177e4SLinus Torvalds * is required to hold the mmap_sem in read mode. We need the 22501da177e4SLinus Torvalds * anon_vma lock to serialize against concurrent expand_stacks. 22511da177e4SLinus Torvalds */ 22521da177e4SLinus Torvalds 22531da177e4SLinus Torvalds /* Somebody else might have raced and expanded it already */ 22541da177e4SLinus Torvalds if (address < vma->vm_start) { 22551da177e4SLinus Torvalds unsigned long size, grow; 22561da177e4SLinus Torvalds 22571da177e4SLinus Torvalds size = vma->vm_end - address; 22581da177e4SLinus Torvalds grow = (vma->vm_start - address) >> PAGE_SHIFT; 22591da177e4SLinus Torvalds 2260a626ca6aSLinus Torvalds error = -ENOMEM; 2261a626ca6aSLinus Torvalds if (grow <= vma->vm_pgoff) { 22621da177e4SLinus Torvalds error = acct_stack_growth(vma, size, grow); 22631da177e4SLinus Torvalds if (!error) { 22644128997bSMichel Lespinasse /* 22654128997bSMichel Lespinasse * vma_gap_update() doesn't support concurrent 22664128997bSMichel Lespinasse * updates, but we only hold a shared mmap_sem 22674128997bSMichel Lespinasse * lock here, so we need to protect against 22684128997bSMichel Lespinasse * concurrent vma expansions. 22694128997bSMichel Lespinasse * vma_lock_anon_vma() doesn't help here, as 22704128997bSMichel Lespinasse * we don't guarantee that all growable vmas 22714128997bSMichel Lespinasse * in a mm share the same root anon vma. 22724128997bSMichel Lespinasse * So, we reuse mm->page_table_lock to guard 22734128997bSMichel Lespinasse * against concurrent vma expansions. 22744128997bSMichel Lespinasse */ 22754128997bSMichel Lespinasse spin_lock(&vma->vm_mm->page_table_lock); 2276bf181b9fSMichel Lespinasse anon_vma_interval_tree_pre_update_vma(vma); 22771da177e4SLinus Torvalds vma->vm_start = address; 22781da177e4SLinus Torvalds vma->vm_pgoff -= grow; 2279bf181b9fSMichel Lespinasse anon_vma_interval_tree_post_update_vma(vma); 2280d3737187SMichel Lespinasse vma_gap_update(vma); 22814128997bSMichel Lespinasse spin_unlock(&vma->vm_mm->page_table_lock); 22824128997bSMichel Lespinasse 22833af9e859SEric B Munson perf_event_mmap(vma); 22841da177e4SLinus Torvalds } 22851da177e4SLinus Torvalds } 2286a626ca6aSLinus Torvalds } 2287bb4a340eSRik van Riel vma_unlock_anon_vma(vma); 22886d50e60cSDavid Rientjes khugepaged_enter_vma_merge(vma, vma->vm_flags); 2289ed8ea815SMichel Lespinasse validate_mm(vma->vm_mm); 22901da177e4SLinus Torvalds return error; 22911da177e4SLinus Torvalds } 22921da177e4SLinus Torvalds 229309884964SLinus Torvalds /* 229409884964SLinus Torvalds * Note how expand_stack() refuses to expand the stack all the way to 229509884964SLinus Torvalds * abut the next virtual mapping, *unless* that mapping itself is also 229609884964SLinus Torvalds * a stack mapping. We want to leave room for a guard page, after all 229709884964SLinus Torvalds * (the guard page itself is not added here, that is done by the 229809884964SLinus Torvalds * actual page faulting logic) 229909884964SLinus Torvalds * 230009884964SLinus Torvalds * This matches the behavior of the guard page logic (see mm/memory.c: 230109884964SLinus Torvalds * check_stack_guard_page()), which only allows the guard page to be 230209884964SLinus Torvalds * removed under these circumstances. 230309884964SLinus Torvalds */ 2304b6a2fea3SOllie Wild #ifdef CONFIG_STACK_GROWSUP 2305b6a2fea3SOllie Wild int expand_stack(struct vm_area_struct *vma, unsigned long address) 2306b6a2fea3SOllie Wild { 230709884964SLinus Torvalds struct vm_area_struct *next; 230809884964SLinus Torvalds 230909884964SLinus Torvalds address &= PAGE_MASK; 231009884964SLinus Torvalds next = vma->vm_next; 231109884964SLinus Torvalds if (next && next->vm_start == address + PAGE_SIZE) { 231209884964SLinus Torvalds if (!(next->vm_flags & VM_GROWSUP)) 231309884964SLinus Torvalds return -ENOMEM; 231409884964SLinus Torvalds } 2315b6a2fea3SOllie Wild return expand_upwards(vma, address); 2316b6a2fea3SOllie Wild } 2317b6a2fea3SOllie Wild 2318b6a2fea3SOllie Wild struct vm_area_struct * 2319b6a2fea3SOllie Wild find_extend_vma(struct mm_struct *mm, unsigned long addr) 2320b6a2fea3SOllie Wild { 2321b6a2fea3SOllie Wild struct vm_area_struct *vma, *prev; 2322b6a2fea3SOllie Wild 2323b6a2fea3SOllie Wild addr &= PAGE_MASK; 2324b6a2fea3SOllie Wild vma = find_vma_prev(mm, addr, &prev); 2325b6a2fea3SOllie Wild if (vma && (vma->vm_start <= addr)) 2326b6a2fea3SOllie Wild return vma; 23271c127185SDenys Vlasenko if (!prev || expand_stack(prev, addr)) 2328b6a2fea3SOllie Wild return NULL; 2329cea10a19SMichel Lespinasse if (prev->vm_flags & VM_LOCKED) 2330cea10a19SMichel Lespinasse __mlock_vma_pages_range(prev, addr, prev->vm_end, NULL); 2331b6a2fea3SOllie Wild return prev; 2332b6a2fea3SOllie Wild } 2333b6a2fea3SOllie Wild #else 2334b6a2fea3SOllie Wild int expand_stack(struct vm_area_struct *vma, unsigned long address) 2335b6a2fea3SOllie Wild { 233609884964SLinus Torvalds struct vm_area_struct *prev; 233709884964SLinus Torvalds 233809884964SLinus Torvalds address &= PAGE_MASK; 233909884964SLinus Torvalds prev = vma->vm_prev; 234009884964SLinus Torvalds if (prev && prev->vm_end == address) { 234109884964SLinus Torvalds if (!(prev->vm_flags & VM_GROWSDOWN)) 234209884964SLinus Torvalds return -ENOMEM; 234309884964SLinus Torvalds } 2344b6a2fea3SOllie Wild return expand_downwards(vma, address); 2345b6a2fea3SOllie Wild } 2346b6a2fea3SOllie Wild 23471da177e4SLinus Torvalds struct vm_area_struct * 23481da177e4SLinus Torvalds find_extend_vma(struct mm_struct *mm, unsigned long addr) 23491da177e4SLinus Torvalds { 23501da177e4SLinus Torvalds struct vm_area_struct *vma; 23511da177e4SLinus Torvalds unsigned long start; 23521da177e4SLinus Torvalds 23531da177e4SLinus Torvalds addr &= PAGE_MASK; 23541da177e4SLinus Torvalds vma = find_vma(mm, addr); 23551da177e4SLinus Torvalds if (!vma) 23561da177e4SLinus Torvalds return NULL; 23571da177e4SLinus Torvalds if (vma->vm_start <= addr) 23581da177e4SLinus Torvalds return vma; 23591da177e4SLinus Torvalds if (!(vma->vm_flags & VM_GROWSDOWN)) 23601da177e4SLinus Torvalds return NULL; 23611da177e4SLinus Torvalds start = vma->vm_start; 23621da177e4SLinus Torvalds if (expand_stack(vma, addr)) 23631da177e4SLinus Torvalds return NULL; 2364cea10a19SMichel Lespinasse if (vma->vm_flags & VM_LOCKED) 2365cea10a19SMichel Lespinasse __mlock_vma_pages_range(vma, addr, start, NULL); 23661da177e4SLinus Torvalds return vma; 23671da177e4SLinus Torvalds } 23681da177e4SLinus Torvalds #endif 23691da177e4SLinus Torvalds 2370e1d6d01aSJesse Barnes EXPORT_SYMBOL_GPL(find_extend_vma); 2371e1d6d01aSJesse Barnes 23722c0b3814SHugh Dickins /* 23732c0b3814SHugh Dickins * Ok - we have the memory areas we should free on the vma list, 23742c0b3814SHugh Dickins * so release them, and do the vma updates. 23751da177e4SLinus Torvalds * 23762c0b3814SHugh Dickins * Called with the mm semaphore held. 23771da177e4SLinus Torvalds */ 23782c0b3814SHugh Dickins static void remove_vma_list(struct mm_struct *mm, struct vm_area_struct *vma) 23791da177e4SLinus Torvalds { 23804f74d2c8SLinus Torvalds unsigned long nr_accounted = 0; 23814f74d2c8SLinus Torvalds 2382365e9c87SHugh Dickins /* Update high watermark before we lower total_vm */ 2383365e9c87SHugh Dickins update_hiwater_vm(mm); 23842c0b3814SHugh Dickins do { 2385ab50b8edSHugh Dickins long nrpages = vma_pages(vma); 23861da177e4SLinus Torvalds 23874f74d2c8SLinus Torvalds if (vma->vm_flags & VM_ACCOUNT) 23884f74d2c8SLinus Torvalds nr_accounted += nrpages; 2389ab50b8edSHugh Dickins vm_stat_account(mm, vma->vm_flags, vma->vm_file, -nrpages); 2390a8fb5618SHugh Dickins vma = remove_vma(vma); 2391146425a3SHugh Dickins } while (vma); 23924f74d2c8SLinus Torvalds vm_unacct_memory(nr_accounted); 23931da177e4SLinus Torvalds validate_mm(mm); 23941da177e4SLinus Torvalds } 23951da177e4SLinus Torvalds 23961da177e4SLinus Torvalds /* 23971da177e4SLinus Torvalds * Get rid of page table information in the indicated region. 23981da177e4SLinus Torvalds * 2399f10df686SPaolo 'Blaisorblade' Giarrusso * Called with the mm semaphore held. 24001da177e4SLinus Torvalds */ 24011da177e4SLinus Torvalds static void unmap_region(struct mm_struct *mm, 2402e0da382cSHugh Dickins struct vm_area_struct *vma, struct vm_area_struct *prev, 2403e0da382cSHugh Dickins unsigned long start, unsigned long end) 24041da177e4SLinus Torvalds { 2405e0da382cSHugh Dickins struct vm_area_struct *next = prev ? prev->vm_next : mm->mmap; 2406d16dfc55SPeter Zijlstra struct mmu_gather tlb; 24071da177e4SLinus Torvalds 24081da177e4SLinus Torvalds lru_add_drain(); 24092b047252SLinus Torvalds tlb_gather_mmu(&tlb, mm, start, end); 2410365e9c87SHugh Dickins update_hiwater_rss(mm); 24114f74d2c8SLinus Torvalds unmap_vmas(&tlb, vma, start, end); 2412d16dfc55SPeter Zijlstra free_pgtables(&tlb, vma, prev ? prev->vm_end : FIRST_USER_ADDRESS, 24136ee8630eSHugh Dickins next ? next->vm_start : USER_PGTABLES_CEILING); 2414d16dfc55SPeter Zijlstra tlb_finish_mmu(&tlb, start, end); 24151da177e4SLinus Torvalds } 24161da177e4SLinus Torvalds 24171da177e4SLinus Torvalds /* 24181da177e4SLinus Torvalds * Create a list of vma's touched by the unmap, removing them from the mm's 24191da177e4SLinus Torvalds * vma list as we go.. 24201da177e4SLinus Torvalds */ 24211da177e4SLinus Torvalds static void 24221da177e4SLinus Torvalds detach_vmas_to_be_unmapped(struct mm_struct *mm, struct vm_area_struct *vma, 24231da177e4SLinus Torvalds struct vm_area_struct *prev, unsigned long end) 24241da177e4SLinus Torvalds { 24251da177e4SLinus Torvalds struct vm_area_struct **insertion_point; 24261da177e4SLinus Torvalds struct vm_area_struct *tail_vma = NULL; 24271da177e4SLinus Torvalds 24281da177e4SLinus Torvalds insertion_point = (prev ? &prev->vm_next : &mm->mmap); 2429297c5eeeSLinus Torvalds vma->vm_prev = NULL; 24301da177e4SLinus Torvalds do { 2431d3737187SMichel Lespinasse vma_rb_erase(vma, &mm->mm_rb); 24321da177e4SLinus Torvalds mm->map_count--; 24331da177e4SLinus Torvalds tail_vma = vma; 24341da177e4SLinus Torvalds vma = vma->vm_next; 24351da177e4SLinus Torvalds } while (vma && vma->vm_start < end); 24361da177e4SLinus Torvalds *insertion_point = vma; 2437d3737187SMichel Lespinasse if (vma) { 2438297c5eeeSLinus Torvalds vma->vm_prev = prev; 2439d3737187SMichel Lespinasse vma_gap_update(vma); 2440d3737187SMichel Lespinasse } else 2441d3737187SMichel Lespinasse mm->highest_vm_end = prev ? prev->vm_end : 0; 24421da177e4SLinus Torvalds tail_vma->vm_next = NULL; 2443615d6e87SDavidlohr Bueso 2444615d6e87SDavidlohr Bueso /* Kill the cache */ 2445615d6e87SDavidlohr Bueso vmacache_invalidate(mm); 24461da177e4SLinus Torvalds } 24471da177e4SLinus Torvalds 24481da177e4SLinus Torvalds /* 2449659ace58SKOSAKI Motohiro * __split_vma() bypasses sysctl_max_map_count checking. We use this on the 2450659ace58SKOSAKI Motohiro * munmap path where it doesn't make sense to fail. 24511da177e4SLinus Torvalds */ 2452659ace58SKOSAKI Motohiro static int __split_vma(struct mm_struct *mm, struct vm_area_struct *vma, 24531da177e4SLinus Torvalds unsigned long addr, int new_below) 24541da177e4SLinus Torvalds { 24551da177e4SLinus Torvalds struct vm_area_struct *new; 24565beb4930SRik van Riel int err = -ENOMEM; 24571da177e4SLinus Torvalds 2458a5516438SAndi Kleen if (is_vm_hugetlb_page(vma) && (addr & 2459a5516438SAndi Kleen ~(huge_page_mask(hstate_vma(vma))))) 24601da177e4SLinus Torvalds return -EINVAL; 24611da177e4SLinus Torvalds 2462e94b1766SChristoph Lameter new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL); 24631da177e4SLinus Torvalds if (!new) 24645beb4930SRik van Riel goto out_err; 24651da177e4SLinus Torvalds 24661da177e4SLinus Torvalds /* most fields are the same, copy all, and then fixup */ 24671da177e4SLinus Torvalds *new = *vma; 24681da177e4SLinus Torvalds 24695beb4930SRik van Riel INIT_LIST_HEAD(&new->anon_vma_chain); 24705beb4930SRik van Riel 24711da177e4SLinus Torvalds if (new_below) 24721da177e4SLinus Torvalds new->vm_end = addr; 24731da177e4SLinus Torvalds else { 24741da177e4SLinus Torvalds new->vm_start = addr; 24751da177e4SLinus Torvalds new->vm_pgoff += ((addr - vma->vm_start) >> PAGE_SHIFT); 24761da177e4SLinus Torvalds } 24771da177e4SLinus Torvalds 2478ef0855d3SOleg Nesterov err = vma_dup_policy(vma, new); 2479ef0855d3SOleg Nesterov if (err) 24805beb4930SRik van Riel goto out_free_vma; 24811da177e4SLinus Torvalds 2482c4ea95d7SDaniel Forrest err = anon_vma_clone(new, vma); 2483c4ea95d7SDaniel Forrest if (err) 24845beb4930SRik van Riel goto out_free_mpol; 24855beb4930SRik van Riel 2486e9714acfSKonstantin Khlebnikov if (new->vm_file) 24871da177e4SLinus Torvalds get_file(new->vm_file); 24881da177e4SLinus Torvalds 24891da177e4SLinus Torvalds if (new->vm_ops && new->vm_ops->open) 24901da177e4SLinus Torvalds new->vm_ops->open(new); 24911da177e4SLinus Torvalds 24921da177e4SLinus Torvalds if (new_below) 24935beb4930SRik van Riel err = vma_adjust(vma, addr, vma->vm_end, vma->vm_pgoff + 24941da177e4SLinus Torvalds ((addr - new->vm_start) >> PAGE_SHIFT), new); 24951da177e4SLinus Torvalds else 24965beb4930SRik van Riel err = vma_adjust(vma, vma->vm_start, addr, vma->vm_pgoff, new); 24971da177e4SLinus Torvalds 24985beb4930SRik van Riel /* Success. */ 24995beb4930SRik van Riel if (!err) 25001da177e4SLinus Torvalds return 0; 25015beb4930SRik van Riel 25025beb4930SRik van Riel /* Clean everything up if vma_adjust failed. */ 250358927533SRik van Riel if (new->vm_ops && new->vm_ops->close) 25045beb4930SRik van Riel new->vm_ops->close(new); 2505e9714acfSKonstantin Khlebnikov if (new->vm_file) 25065beb4930SRik van Riel fput(new->vm_file); 25072aeadc30SAndrea Arcangeli unlink_anon_vmas(new); 25085beb4930SRik van Riel out_free_mpol: 2509ef0855d3SOleg Nesterov mpol_put(vma_policy(new)); 25105beb4930SRik van Riel out_free_vma: 25115beb4930SRik van Riel kmem_cache_free(vm_area_cachep, new); 25125beb4930SRik van Riel out_err: 25135beb4930SRik van Riel return err; 25141da177e4SLinus Torvalds } 25151da177e4SLinus Torvalds 2516659ace58SKOSAKI Motohiro /* 2517659ace58SKOSAKI Motohiro * Split a vma into two pieces at address 'addr', a new vma is allocated 2518659ace58SKOSAKI Motohiro * either for the first part or the tail. 2519659ace58SKOSAKI Motohiro */ 2520659ace58SKOSAKI Motohiro int split_vma(struct mm_struct *mm, struct vm_area_struct *vma, 2521659ace58SKOSAKI Motohiro unsigned long addr, int new_below) 2522659ace58SKOSAKI Motohiro { 2523659ace58SKOSAKI Motohiro if (mm->map_count >= sysctl_max_map_count) 2524659ace58SKOSAKI Motohiro return -ENOMEM; 2525659ace58SKOSAKI Motohiro 2526659ace58SKOSAKI Motohiro return __split_vma(mm, vma, addr, new_below); 2527659ace58SKOSAKI Motohiro } 2528659ace58SKOSAKI Motohiro 25291da177e4SLinus Torvalds /* Munmap is split into 2 main parts -- this part which finds 25301da177e4SLinus Torvalds * what needs doing, and the areas themselves, which do the 25311da177e4SLinus Torvalds * work. This now handles partial unmappings. 25321da177e4SLinus Torvalds * Jeremy Fitzhardinge <jeremy@goop.org> 25331da177e4SLinus Torvalds */ 25341da177e4SLinus Torvalds int do_munmap(struct mm_struct *mm, unsigned long start, size_t len) 25351da177e4SLinus Torvalds { 25361da177e4SLinus Torvalds unsigned long end; 2537146425a3SHugh Dickins struct vm_area_struct *vma, *prev, *last; 25381da177e4SLinus Torvalds 25391da177e4SLinus Torvalds if ((start & ~PAGE_MASK) || start > TASK_SIZE || len > TASK_SIZE-start) 25401da177e4SLinus Torvalds return -EINVAL; 25411da177e4SLinus Torvalds 2542cc71aba3Svishnu.ps len = PAGE_ALIGN(len); 2543cc71aba3Svishnu.ps if (len == 0) 25441da177e4SLinus Torvalds return -EINVAL; 25451da177e4SLinus Torvalds 25461da177e4SLinus Torvalds /* Find the first overlapping VMA */ 25479be34c9dSLinus Torvalds vma = find_vma(mm, start); 2548146425a3SHugh Dickins if (!vma) 25491da177e4SLinus Torvalds return 0; 25509be34c9dSLinus Torvalds prev = vma->vm_prev; 2551146425a3SHugh Dickins /* we have start < vma->vm_end */ 25521da177e4SLinus Torvalds 25531da177e4SLinus Torvalds /* if it doesn't overlap, we have nothing.. */ 25541da177e4SLinus Torvalds end = start + len; 2555146425a3SHugh Dickins if (vma->vm_start >= end) 25561da177e4SLinus Torvalds return 0; 25571da177e4SLinus Torvalds 25581da177e4SLinus Torvalds /* 25591da177e4SLinus Torvalds * If we need to split any vma, do it now to save pain later. 25601da177e4SLinus Torvalds * 25611da177e4SLinus Torvalds * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially 25621da177e4SLinus Torvalds * unmapped vm_area_struct will remain in use: so lower split_vma 25631da177e4SLinus Torvalds * places tmp vma above, and higher split_vma places tmp vma below. 25641da177e4SLinus Torvalds */ 2565146425a3SHugh Dickins if (start > vma->vm_start) { 2566659ace58SKOSAKI Motohiro int error; 2567659ace58SKOSAKI Motohiro 2568659ace58SKOSAKI Motohiro /* 2569659ace58SKOSAKI Motohiro * Make sure that map_count on return from munmap() will 2570659ace58SKOSAKI Motohiro * not exceed its limit; but let map_count go just above 2571659ace58SKOSAKI Motohiro * its limit temporarily, to help free resources as expected. 2572659ace58SKOSAKI Motohiro */ 2573659ace58SKOSAKI Motohiro if (end < vma->vm_end && mm->map_count >= sysctl_max_map_count) 2574659ace58SKOSAKI Motohiro return -ENOMEM; 2575659ace58SKOSAKI Motohiro 2576659ace58SKOSAKI Motohiro error = __split_vma(mm, vma, start, 0); 25771da177e4SLinus Torvalds if (error) 25781da177e4SLinus Torvalds return error; 2579146425a3SHugh Dickins prev = vma; 25801da177e4SLinus Torvalds } 25811da177e4SLinus Torvalds 25821da177e4SLinus Torvalds /* Does it split the last one? */ 25831da177e4SLinus Torvalds last = find_vma(mm, end); 25841da177e4SLinus Torvalds if (last && end > last->vm_start) { 2585659ace58SKOSAKI Motohiro int error = __split_vma(mm, last, end, 1); 25861da177e4SLinus Torvalds if (error) 25871da177e4SLinus Torvalds return error; 25881da177e4SLinus Torvalds } 2589146425a3SHugh Dickins vma = prev ? prev->vm_next : mm->mmap; 25901da177e4SLinus Torvalds 25911da177e4SLinus Torvalds /* 2592ba470de4SRik van Riel * unlock any mlock()ed ranges before detaching vmas 2593ba470de4SRik van Riel */ 2594ba470de4SRik van Riel if (mm->locked_vm) { 2595ba470de4SRik van Riel struct vm_area_struct *tmp = vma; 2596ba470de4SRik van Riel while (tmp && tmp->vm_start < end) { 2597ba470de4SRik van Riel if (tmp->vm_flags & VM_LOCKED) { 2598ba470de4SRik van Riel mm->locked_vm -= vma_pages(tmp); 2599ba470de4SRik van Riel munlock_vma_pages_all(tmp); 2600ba470de4SRik van Riel } 2601ba470de4SRik van Riel tmp = tmp->vm_next; 2602ba470de4SRik van Riel } 2603ba470de4SRik van Riel } 2604ba470de4SRik van Riel 2605ba470de4SRik van Riel /* 26061da177e4SLinus Torvalds * Remove the vma's, and unmap the actual pages 26071da177e4SLinus Torvalds */ 2608146425a3SHugh Dickins detach_vmas_to_be_unmapped(mm, vma, prev, end); 2609146425a3SHugh Dickins unmap_region(mm, vma, prev, start, end); 26101da177e4SLinus Torvalds 26111de4fa14SDave Hansen arch_unmap(mm, vma, start, end); 26121de4fa14SDave Hansen 26131da177e4SLinus Torvalds /* Fix up all other VM information */ 26142c0b3814SHugh Dickins remove_vma_list(mm, vma); 26151da177e4SLinus Torvalds 26161da177e4SLinus Torvalds return 0; 26171da177e4SLinus Torvalds } 26181da177e4SLinus Torvalds 2619bfce281cSAl Viro int vm_munmap(unsigned long start, size_t len) 2620a46ef99dSLinus Torvalds { 2621a46ef99dSLinus Torvalds int ret; 2622bfce281cSAl Viro struct mm_struct *mm = current->mm; 2623a46ef99dSLinus Torvalds 2624a46ef99dSLinus Torvalds down_write(&mm->mmap_sem); 2625a46ef99dSLinus Torvalds ret = do_munmap(mm, start, len); 2626a46ef99dSLinus Torvalds up_write(&mm->mmap_sem); 2627a46ef99dSLinus Torvalds return ret; 2628a46ef99dSLinus Torvalds } 2629a46ef99dSLinus Torvalds EXPORT_SYMBOL(vm_munmap); 2630a46ef99dSLinus Torvalds 26316a6160a7SHeiko Carstens SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len) 26321da177e4SLinus Torvalds { 26331da177e4SLinus Torvalds profile_munmap(addr); 2634bfce281cSAl Viro return vm_munmap(addr, len); 26351da177e4SLinus Torvalds } 26361da177e4SLinus Torvalds 26371da177e4SLinus Torvalds static inline void verify_mm_writelocked(struct mm_struct *mm) 26381da177e4SLinus Torvalds { 2639a241ec65SPaul E. McKenney #ifdef CONFIG_DEBUG_VM 26401da177e4SLinus Torvalds if (unlikely(down_read_trylock(&mm->mmap_sem))) { 26411da177e4SLinus Torvalds WARN_ON(1); 26421da177e4SLinus Torvalds up_read(&mm->mmap_sem); 26431da177e4SLinus Torvalds } 26441da177e4SLinus Torvalds #endif 26451da177e4SLinus Torvalds } 26461da177e4SLinus Torvalds 26471da177e4SLinus Torvalds /* 26481da177e4SLinus Torvalds * this is really a simplified "do_mmap". it only handles 26491da177e4SLinus Torvalds * anonymous maps. eventually we may be able to do some 26501da177e4SLinus Torvalds * brk-specific accounting here. 26511da177e4SLinus Torvalds */ 2652e4eb1ff6SLinus Torvalds static unsigned long do_brk(unsigned long addr, unsigned long len) 26531da177e4SLinus Torvalds { 26541da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 26551da177e4SLinus Torvalds struct vm_area_struct *vma, *prev; 26561da177e4SLinus Torvalds unsigned long flags; 26571da177e4SLinus Torvalds struct rb_node **rb_link, *rb_parent; 26581da177e4SLinus Torvalds pgoff_t pgoff = addr >> PAGE_SHIFT; 26593a459756SKirill Korotaev int error; 26601da177e4SLinus Torvalds 26611da177e4SLinus Torvalds len = PAGE_ALIGN(len); 26621da177e4SLinus Torvalds if (!len) 26631da177e4SLinus Torvalds return addr; 26641da177e4SLinus Torvalds 26653a459756SKirill Korotaev flags = VM_DATA_DEFAULT_FLAGS | VM_ACCOUNT | mm->def_flags; 26663a459756SKirill Korotaev 26672c6a1016SAl Viro error = get_unmapped_area(NULL, addr, len, 0, MAP_FIXED); 26682c6a1016SAl Viro if (error & ~PAGE_MASK) 26693a459756SKirill Korotaev return error; 26703a459756SKirill Korotaev 2671363ee17fSDavidlohr Bueso error = mlock_future_check(mm, mm->def_flags, len); 2672363ee17fSDavidlohr Bueso if (error) 2673363ee17fSDavidlohr Bueso return error; 26741da177e4SLinus Torvalds 26751da177e4SLinus Torvalds /* 26761da177e4SLinus Torvalds * mm->mmap_sem is required to protect against another thread 26771da177e4SLinus Torvalds * changing the mappings in case we sleep. 26781da177e4SLinus Torvalds */ 26791da177e4SLinus Torvalds verify_mm_writelocked(mm); 26801da177e4SLinus Torvalds 26811da177e4SLinus Torvalds /* 26821da177e4SLinus Torvalds * Clear old maps. this also does some error checking for us 26831da177e4SLinus Torvalds */ 26841da177e4SLinus Torvalds munmap_back: 26856597d783SHugh Dickins if (find_vma_links(mm, addr, addr + len, &prev, &rb_link, &rb_parent)) { 26861da177e4SLinus Torvalds if (do_munmap(mm, addr, len)) 26871da177e4SLinus Torvalds return -ENOMEM; 26881da177e4SLinus Torvalds goto munmap_back; 26891da177e4SLinus Torvalds } 26901da177e4SLinus Torvalds 26911da177e4SLinus Torvalds /* Check against address space limits *after* clearing old maps... */ 2692119f657cSakpm@osdl.org if (!may_expand_vm(mm, len >> PAGE_SHIFT)) 26931da177e4SLinus Torvalds return -ENOMEM; 26941da177e4SLinus Torvalds 26951da177e4SLinus Torvalds if (mm->map_count > sysctl_max_map_count) 26961da177e4SLinus Torvalds return -ENOMEM; 26971da177e4SLinus Torvalds 2698191c5424SAl Viro if (security_vm_enough_memory_mm(mm, len >> PAGE_SHIFT)) 26991da177e4SLinus Torvalds return -ENOMEM; 27001da177e4SLinus Torvalds 27011da177e4SLinus Torvalds /* Can we just expand an old private anonymous mapping? */ 2702ba470de4SRik van Riel vma = vma_merge(mm, prev, addr, addr + len, flags, 2703ba470de4SRik van Riel NULL, NULL, pgoff, NULL); 2704ba470de4SRik van Riel if (vma) 27051da177e4SLinus Torvalds goto out; 27061da177e4SLinus Torvalds 27071da177e4SLinus Torvalds /* 27081da177e4SLinus Torvalds * create a vma struct for an anonymous mapping 27091da177e4SLinus Torvalds */ 2710c5e3b83eSPekka Enberg vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL); 27111da177e4SLinus Torvalds if (!vma) { 27121da177e4SLinus Torvalds vm_unacct_memory(len >> PAGE_SHIFT); 27131da177e4SLinus Torvalds return -ENOMEM; 27141da177e4SLinus Torvalds } 27151da177e4SLinus Torvalds 27165beb4930SRik van Riel INIT_LIST_HEAD(&vma->anon_vma_chain); 27171da177e4SLinus Torvalds vma->vm_mm = mm; 27181da177e4SLinus Torvalds vma->vm_start = addr; 27191da177e4SLinus Torvalds vma->vm_end = addr + len; 27201da177e4SLinus Torvalds vma->vm_pgoff = pgoff; 27211da177e4SLinus Torvalds vma->vm_flags = flags; 27223ed75eb8SColy Li vma->vm_page_prot = vm_get_page_prot(flags); 27231da177e4SLinus Torvalds vma_link(mm, vma, prev, rb_link, rb_parent); 27241da177e4SLinus Torvalds out: 27253af9e859SEric B Munson perf_event_mmap(vma); 27261da177e4SLinus Torvalds mm->total_vm += len >> PAGE_SHIFT; 2727128557ffSMichel Lespinasse if (flags & VM_LOCKED) 2728ba470de4SRik van Riel mm->locked_vm += (len >> PAGE_SHIFT); 2729d9104d1cSCyrill Gorcunov vma->vm_flags |= VM_SOFTDIRTY; 27301da177e4SLinus Torvalds return addr; 27311da177e4SLinus Torvalds } 27321da177e4SLinus Torvalds 2733e4eb1ff6SLinus Torvalds unsigned long vm_brk(unsigned long addr, unsigned long len) 2734e4eb1ff6SLinus Torvalds { 2735e4eb1ff6SLinus Torvalds struct mm_struct *mm = current->mm; 2736e4eb1ff6SLinus Torvalds unsigned long ret; 2737128557ffSMichel Lespinasse bool populate; 2738e4eb1ff6SLinus Torvalds 2739e4eb1ff6SLinus Torvalds down_write(&mm->mmap_sem); 2740e4eb1ff6SLinus Torvalds ret = do_brk(addr, len); 2741128557ffSMichel Lespinasse populate = ((mm->def_flags & VM_LOCKED) != 0); 2742e4eb1ff6SLinus Torvalds up_write(&mm->mmap_sem); 2743128557ffSMichel Lespinasse if (populate) 2744128557ffSMichel Lespinasse mm_populate(addr, len); 2745e4eb1ff6SLinus Torvalds return ret; 2746e4eb1ff6SLinus Torvalds } 2747e4eb1ff6SLinus Torvalds EXPORT_SYMBOL(vm_brk); 27481da177e4SLinus Torvalds 27491da177e4SLinus Torvalds /* Release all mmaps. */ 27501da177e4SLinus Torvalds void exit_mmap(struct mm_struct *mm) 27511da177e4SLinus Torvalds { 2752d16dfc55SPeter Zijlstra struct mmu_gather tlb; 2753ba470de4SRik van Riel struct vm_area_struct *vma; 27541da177e4SLinus Torvalds unsigned long nr_accounted = 0; 27551da177e4SLinus Torvalds 2756d6dd61c8SJeremy Fitzhardinge /* mm's last user has gone, and its about to be pulled down */ 2757cddb8a5cSAndrea Arcangeli mmu_notifier_release(mm); 2758d6dd61c8SJeremy Fitzhardinge 2759ba470de4SRik van Riel if (mm->locked_vm) { 2760ba470de4SRik van Riel vma = mm->mmap; 2761ba470de4SRik van Riel while (vma) { 2762ba470de4SRik van Riel if (vma->vm_flags & VM_LOCKED) 2763ba470de4SRik van Riel munlock_vma_pages_all(vma); 2764ba470de4SRik van Riel vma = vma->vm_next; 2765ba470de4SRik van Riel } 2766ba470de4SRik van Riel } 27679480c53eSJeremy Fitzhardinge 27689480c53eSJeremy Fitzhardinge arch_exit_mmap(mm); 27699480c53eSJeremy Fitzhardinge 2770ba470de4SRik van Riel vma = mm->mmap; 27719480c53eSJeremy Fitzhardinge if (!vma) /* Can happen if dup_mmap() received an OOM */ 27729480c53eSJeremy Fitzhardinge return; 27739480c53eSJeremy Fitzhardinge 27741da177e4SLinus Torvalds lru_add_drain(); 27751da177e4SLinus Torvalds flush_cache_mm(mm); 27762b047252SLinus Torvalds tlb_gather_mmu(&tlb, mm, 0, -1); 2777901608d9SOleg Nesterov /* update_hiwater_rss(mm) here? but nobody should be looking */ 2778e0da382cSHugh Dickins /* Use -1 here to ensure all VMAs in the mm are unmapped */ 27794f74d2c8SLinus Torvalds unmap_vmas(&tlb, vma, 0, -1); 27809ba69294SHugh Dickins 27816ee8630eSHugh Dickins free_pgtables(&tlb, vma, FIRST_USER_ADDRESS, USER_PGTABLES_CEILING); 2782853f5e26SAl Viro tlb_finish_mmu(&tlb, 0, -1); 27831da177e4SLinus Torvalds 27841da177e4SLinus Torvalds /* 27858f4f8c16SHugh Dickins * Walk the list again, actually closing and freeing it, 27868f4f8c16SHugh Dickins * with preemption enabled, without holding any MM locks. 27871da177e4SLinus Torvalds */ 27884f74d2c8SLinus Torvalds while (vma) { 27894f74d2c8SLinus Torvalds if (vma->vm_flags & VM_ACCOUNT) 27904f74d2c8SLinus Torvalds nr_accounted += vma_pages(vma); 2791a8fb5618SHugh Dickins vma = remove_vma(vma); 27924f74d2c8SLinus Torvalds } 27934f74d2c8SLinus Torvalds vm_unacct_memory(nr_accounted); 2794e0da382cSHugh Dickins 2795e1f56c89SKirill A. Shutemov WARN_ON(atomic_long_read(&mm->nr_ptes) > 2796e1f56c89SKirill A. Shutemov (FIRST_USER_ADDRESS+PMD_SIZE-1)>>PMD_SHIFT); 27971da177e4SLinus Torvalds } 27981da177e4SLinus Torvalds 27991da177e4SLinus Torvalds /* Insert vm structure into process list sorted by address 28001da177e4SLinus Torvalds * and into the inode's i_mmap tree. If vm_file is non-NULL 2801c8c06efaSDavidlohr Bueso * then i_mmap_rwsem is taken here. 28021da177e4SLinus Torvalds */ 28031da177e4SLinus Torvalds int insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma) 28041da177e4SLinus Torvalds { 28056597d783SHugh Dickins struct vm_area_struct *prev; 28061da177e4SLinus Torvalds struct rb_node **rb_link, *rb_parent; 28071da177e4SLinus Torvalds 28081da177e4SLinus Torvalds /* 28091da177e4SLinus Torvalds * The vm_pgoff of a purely anonymous vma should be irrelevant 28101da177e4SLinus Torvalds * until its first write fault, when page's anon_vma and index 28111da177e4SLinus Torvalds * are set. But now set the vm_pgoff it will almost certainly 28121da177e4SLinus Torvalds * end up with (unless mremap moves it elsewhere before that 28131da177e4SLinus Torvalds * first wfault), so /proc/pid/maps tells a consistent story. 28141da177e4SLinus Torvalds * 28151da177e4SLinus Torvalds * By setting it to reflect the virtual start address of the 28161da177e4SLinus Torvalds * vma, merges and splits can happen in a seamless way, just 28171da177e4SLinus Torvalds * using the existing file pgoff checks and manipulations. 28181da177e4SLinus Torvalds * Similarly in do_mmap_pgoff and in do_brk. 28191da177e4SLinus Torvalds */ 28201da177e4SLinus Torvalds if (!vma->vm_file) { 28211da177e4SLinus Torvalds BUG_ON(vma->anon_vma); 28221da177e4SLinus Torvalds vma->vm_pgoff = vma->vm_start >> PAGE_SHIFT; 28231da177e4SLinus Torvalds } 28246597d783SHugh Dickins if (find_vma_links(mm, vma->vm_start, vma->vm_end, 28256597d783SHugh Dickins &prev, &rb_link, &rb_parent)) 28261da177e4SLinus Torvalds return -ENOMEM; 28272fd4ef85SHugh Dickins if ((vma->vm_flags & VM_ACCOUNT) && 282834b4e4aaSAlan Cox security_vm_enough_memory_mm(mm, vma_pages(vma))) 28292fd4ef85SHugh Dickins return -ENOMEM; 28302b144498SSrikar Dronamraju 28311da177e4SLinus Torvalds vma_link(mm, vma, prev, rb_link, rb_parent); 28321da177e4SLinus Torvalds return 0; 28331da177e4SLinus Torvalds } 28341da177e4SLinus Torvalds 28351da177e4SLinus Torvalds /* 28361da177e4SLinus Torvalds * Copy the vma structure to a new location in the same mm, 28371da177e4SLinus Torvalds * prior to moving page table entries, to effect an mremap move. 28381da177e4SLinus Torvalds */ 28391da177e4SLinus Torvalds struct vm_area_struct *copy_vma(struct vm_area_struct **vmap, 284038a76013SMichel Lespinasse unsigned long addr, unsigned long len, pgoff_t pgoff, 284138a76013SMichel Lespinasse bool *need_rmap_locks) 28421da177e4SLinus Torvalds { 28431da177e4SLinus Torvalds struct vm_area_struct *vma = *vmap; 28441da177e4SLinus Torvalds unsigned long vma_start = vma->vm_start; 28451da177e4SLinus Torvalds struct mm_struct *mm = vma->vm_mm; 28461da177e4SLinus Torvalds struct vm_area_struct *new_vma, *prev; 28471da177e4SLinus Torvalds struct rb_node **rb_link, *rb_parent; 2848948f017bSAndrea Arcangeli bool faulted_in_anon_vma = true; 28491da177e4SLinus Torvalds 28501da177e4SLinus Torvalds /* 28511da177e4SLinus Torvalds * If anonymous vma has not yet been faulted, update new pgoff 28521da177e4SLinus Torvalds * to match new location, to increase its chance of merging. 28531da177e4SLinus Torvalds */ 2854948f017bSAndrea Arcangeli if (unlikely(!vma->vm_file && !vma->anon_vma)) { 28551da177e4SLinus Torvalds pgoff = addr >> PAGE_SHIFT; 2856948f017bSAndrea Arcangeli faulted_in_anon_vma = false; 2857948f017bSAndrea Arcangeli } 28581da177e4SLinus Torvalds 28596597d783SHugh Dickins if (find_vma_links(mm, addr, addr + len, &prev, &rb_link, &rb_parent)) 28606597d783SHugh Dickins return NULL; /* should never get here */ 28611da177e4SLinus Torvalds new_vma = vma_merge(mm, prev, addr, addr + len, vma->vm_flags, 28621da177e4SLinus Torvalds vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma)); 28631da177e4SLinus Torvalds if (new_vma) { 28641da177e4SLinus Torvalds /* 28651da177e4SLinus Torvalds * Source vma may have been merged into new_vma 28661da177e4SLinus Torvalds */ 2867948f017bSAndrea Arcangeli if (unlikely(vma_start >= new_vma->vm_start && 2868948f017bSAndrea Arcangeli vma_start < new_vma->vm_end)) { 2869948f017bSAndrea Arcangeli /* 2870948f017bSAndrea Arcangeli * The only way we can get a vma_merge with 2871948f017bSAndrea Arcangeli * self during an mremap is if the vma hasn't 2872948f017bSAndrea Arcangeli * been faulted in yet and we were allowed to 2873948f017bSAndrea Arcangeli * reset the dst vma->vm_pgoff to the 2874948f017bSAndrea Arcangeli * destination address of the mremap to allow 2875948f017bSAndrea Arcangeli * the merge to happen. mremap must change the 2876948f017bSAndrea Arcangeli * vm_pgoff linearity between src and dst vmas 2877948f017bSAndrea Arcangeli * (in turn preventing a vma_merge) to be 2878948f017bSAndrea Arcangeli * safe. It is only safe to keep the vm_pgoff 2879948f017bSAndrea Arcangeli * linear if there are no pages mapped yet. 2880948f017bSAndrea Arcangeli */ 288181d1b09cSSasha Levin VM_BUG_ON_VMA(faulted_in_anon_vma, new_vma); 288238a76013SMichel Lespinasse *vmap = vma = new_vma; 2883108d6642SMichel Lespinasse } 288438a76013SMichel Lespinasse *need_rmap_locks = (new_vma->vm_pgoff <= vma->vm_pgoff); 28851da177e4SLinus Torvalds } else { 2886e94b1766SChristoph Lameter new_vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL); 28871da177e4SLinus Torvalds if (new_vma) { 28881da177e4SLinus Torvalds *new_vma = *vma; 28891da177e4SLinus Torvalds new_vma->vm_start = addr; 28901da177e4SLinus Torvalds new_vma->vm_end = addr + len; 28911da177e4SLinus Torvalds new_vma->vm_pgoff = pgoff; 2892ef0855d3SOleg Nesterov if (vma_dup_policy(vma, new_vma)) 2893523d4e20SMichel Lespinasse goto out_free_vma; 2894523d4e20SMichel Lespinasse INIT_LIST_HEAD(&new_vma->anon_vma_chain); 2895523d4e20SMichel Lespinasse if (anon_vma_clone(new_vma, vma)) 2896523d4e20SMichel Lespinasse goto out_free_mempol; 2897e9714acfSKonstantin Khlebnikov if (new_vma->vm_file) 28981da177e4SLinus Torvalds get_file(new_vma->vm_file); 28991da177e4SLinus Torvalds if (new_vma->vm_ops && new_vma->vm_ops->open) 29001da177e4SLinus Torvalds new_vma->vm_ops->open(new_vma); 29011da177e4SLinus Torvalds vma_link(mm, new_vma, prev, rb_link, rb_parent); 290238a76013SMichel Lespinasse *need_rmap_locks = false; 29031da177e4SLinus Torvalds } 29041da177e4SLinus Torvalds } 29051da177e4SLinus Torvalds return new_vma; 29065beb4930SRik van Riel 29075beb4930SRik van Riel out_free_mempol: 2908ef0855d3SOleg Nesterov mpol_put(vma_policy(new_vma)); 29095beb4930SRik van Riel out_free_vma: 29105beb4930SRik van Riel kmem_cache_free(vm_area_cachep, new_vma); 29115beb4930SRik van Riel return NULL; 29121da177e4SLinus Torvalds } 2913119f657cSakpm@osdl.org 2914119f657cSakpm@osdl.org /* 2915119f657cSakpm@osdl.org * Return true if the calling process may expand its vm space by the passed 2916119f657cSakpm@osdl.org * number of pages 2917119f657cSakpm@osdl.org */ 2918119f657cSakpm@osdl.org int may_expand_vm(struct mm_struct *mm, unsigned long npages) 2919119f657cSakpm@osdl.org { 2920119f657cSakpm@osdl.org unsigned long cur = mm->total_vm; /* pages */ 2921119f657cSakpm@osdl.org unsigned long lim; 2922119f657cSakpm@osdl.org 292359e99e5bSJiri Slaby lim = rlimit(RLIMIT_AS) >> PAGE_SHIFT; 2924119f657cSakpm@osdl.org 2925119f657cSakpm@osdl.org if (cur + npages > lim) 2926119f657cSakpm@osdl.org return 0; 2927119f657cSakpm@osdl.org return 1; 2928119f657cSakpm@osdl.org } 2929fa5dc22fSRoland McGrath 2930a62c34bdSAndy Lutomirski static int special_mapping_fault(struct vm_area_struct *vma, 2931a62c34bdSAndy Lutomirski struct vm_fault *vmf); 2932a62c34bdSAndy Lutomirski 2933a62c34bdSAndy Lutomirski /* 2934a62c34bdSAndy Lutomirski * Having a close hook prevents vma merging regardless of flags. 2935a62c34bdSAndy Lutomirski */ 2936a62c34bdSAndy Lutomirski static void special_mapping_close(struct vm_area_struct *vma) 2937a62c34bdSAndy Lutomirski { 2938a62c34bdSAndy Lutomirski } 2939a62c34bdSAndy Lutomirski 2940a62c34bdSAndy Lutomirski static const char *special_mapping_name(struct vm_area_struct *vma) 2941a62c34bdSAndy Lutomirski { 2942a62c34bdSAndy Lutomirski return ((struct vm_special_mapping *)vma->vm_private_data)->name; 2943a62c34bdSAndy Lutomirski } 2944a62c34bdSAndy Lutomirski 2945a62c34bdSAndy Lutomirski static const struct vm_operations_struct special_mapping_vmops = { 2946a62c34bdSAndy Lutomirski .close = special_mapping_close, 2947a62c34bdSAndy Lutomirski .fault = special_mapping_fault, 2948a62c34bdSAndy Lutomirski .name = special_mapping_name, 2949a62c34bdSAndy Lutomirski }; 2950a62c34bdSAndy Lutomirski 2951a62c34bdSAndy Lutomirski static const struct vm_operations_struct legacy_special_mapping_vmops = { 2952a62c34bdSAndy Lutomirski .close = special_mapping_close, 2953a62c34bdSAndy Lutomirski .fault = special_mapping_fault, 2954a62c34bdSAndy Lutomirski }; 2955fa5dc22fSRoland McGrath 2956b1d0e4f5SNick Piggin static int special_mapping_fault(struct vm_area_struct *vma, 2957b1d0e4f5SNick Piggin struct vm_fault *vmf) 2958fa5dc22fSRoland McGrath { 2959b1d0e4f5SNick Piggin pgoff_t pgoff; 2960fa5dc22fSRoland McGrath struct page **pages; 2961fa5dc22fSRoland McGrath 2962b1d0e4f5SNick Piggin /* 2963b1d0e4f5SNick Piggin * special mappings have no vm_file, and in that case, the mm 2964b1d0e4f5SNick Piggin * uses vm_pgoff internally. So we have to subtract it from here. 2965b1d0e4f5SNick Piggin * We are allowed to do this because we are the mm; do not copy 2966b1d0e4f5SNick Piggin * this code into drivers! 2967b1d0e4f5SNick Piggin */ 2968b1d0e4f5SNick Piggin pgoff = vmf->pgoff - vma->vm_pgoff; 2969fa5dc22fSRoland McGrath 2970a62c34bdSAndy Lutomirski if (vma->vm_ops == &legacy_special_mapping_vmops) 2971a62c34bdSAndy Lutomirski pages = vma->vm_private_data; 2972a62c34bdSAndy Lutomirski else 2973a62c34bdSAndy Lutomirski pages = ((struct vm_special_mapping *)vma->vm_private_data)-> 2974a62c34bdSAndy Lutomirski pages; 2975a62c34bdSAndy Lutomirski 2976a62c34bdSAndy Lutomirski for (; pgoff && *pages; ++pages) 2977b1d0e4f5SNick Piggin pgoff--; 2978fa5dc22fSRoland McGrath 2979fa5dc22fSRoland McGrath if (*pages) { 2980fa5dc22fSRoland McGrath struct page *page = *pages; 2981fa5dc22fSRoland McGrath get_page(page); 2982b1d0e4f5SNick Piggin vmf->page = page; 2983b1d0e4f5SNick Piggin return 0; 2984fa5dc22fSRoland McGrath } 2985fa5dc22fSRoland McGrath 2986b1d0e4f5SNick Piggin return VM_FAULT_SIGBUS; 2987fa5dc22fSRoland McGrath } 2988fa5dc22fSRoland McGrath 2989a62c34bdSAndy Lutomirski static struct vm_area_struct *__install_special_mapping( 2990a62c34bdSAndy Lutomirski struct mm_struct *mm, 2991fa5dc22fSRoland McGrath unsigned long addr, unsigned long len, 2992a62c34bdSAndy Lutomirski unsigned long vm_flags, const struct vm_operations_struct *ops, 2993a62c34bdSAndy Lutomirski void *priv) 2994fa5dc22fSRoland McGrath { 2995462e635eSTavis Ormandy int ret; 2996fa5dc22fSRoland McGrath struct vm_area_struct *vma; 2997fa5dc22fSRoland McGrath 2998fa5dc22fSRoland McGrath vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL); 2999fa5dc22fSRoland McGrath if (unlikely(vma == NULL)) 30003935ed6aSStefani Seibold return ERR_PTR(-ENOMEM); 3001fa5dc22fSRoland McGrath 30025beb4930SRik van Riel INIT_LIST_HEAD(&vma->anon_vma_chain); 3003fa5dc22fSRoland McGrath vma->vm_mm = mm; 3004fa5dc22fSRoland McGrath vma->vm_start = addr; 3005fa5dc22fSRoland McGrath vma->vm_end = addr + len; 3006fa5dc22fSRoland McGrath 3007d9104d1cSCyrill Gorcunov vma->vm_flags = vm_flags | mm->def_flags | VM_DONTEXPAND | VM_SOFTDIRTY; 30083ed75eb8SColy Li vma->vm_page_prot = vm_get_page_prot(vma->vm_flags); 3009fa5dc22fSRoland McGrath 3010a62c34bdSAndy Lutomirski vma->vm_ops = ops; 3011a62c34bdSAndy Lutomirski vma->vm_private_data = priv; 3012fa5dc22fSRoland McGrath 3013462e635eSTavis Ormandy ret = insert_vm_struct(mm, vma); 3014462e635eSTavis Ormandy if (ret) 3015462e635eSTavis Ormandy goto out; 3016fa5dc22fSRoland McGrath 3017fa5dc22fSRoland McGrath mm->total_vm += len >> PAGE_SHIFT; 3018fa5dc22fSRoland McGrath 3019cdd6c482SIngo Molnar perf_event_mmap(vma); 3020089dd79dSPeter Zijlstra 30213935ed6aSStefani Seibold return vma; 3022462e635eSTavis Ormandy 3023462e635eSTavis Ormandy out: 3024462e635eSTavis Ormandy kmem_cache_free(vm_area_cachep, vma); 30253935ed6aSStefani Seibold return ERR_PTR(ret); 30263935ed6aSStefani Seibold } 30273935ed6aSStefani Seibold 3028a62c34bdSAndy Lutomirski /* 3029a62c34bdSAndy Lutomirski * Called with mm->mmap_sem held for writing. 3030a62c34bdSAndy Lutomirski * Insert a new vma covering the given region, with the given flags. 3031a62c34bdSAndy Lutomirski * Its pages are supplied by the given array of struct page *. 3032a62c34bdSAndy Lutomirski * The array can be shorter than len >> PAGE_SHIFT if it's null-terminated. 3033a62c34bdSAndy Lutomirski * The region past the last page supplied will always produce SIGBUS. 3034a62c34bdSAndy Lutomirski * The array pointer and the pages it points to are assumed to stay alive 3035a62c34bdSAndy Lutomirski * for as long as this mapping might exist. 3036a62c34bdSAndy Lutomirski */ 3037a62c34bdSAndy Lutomirski struct vm_area_struct *_install_special_mapping( 3038a62c34bdSAndy Lutomirski struct mm_struct *mm, 3039a62c34bdSAndy Lutomirski unsigned long addr, unsigned long len, 3040a62c34bdSAndy Lutomirski unsigned long vm_flags, const struct vm_special_mapping *spec) 3041a62c34bdSAndy Lutomirski { 3042a62c34bdSAndy Lutomirski return __install_special_mapping(mm, addr, len, vm_flags, 3043a62c34bdSAndy Lutomirski &special_mapping_vmops, (void *)spec); 3044a62c34bdSAndy Lutomirski } 3045a62c34bdSAndy Lutomirski 30463935ed6aSStefani Seibold int install_special_mapping(struct mm_struct *mm, 30473935ed6aSStefani Seibold unsigned long addr, unsigned long len, 30483935ed6aSStefani Seibold unsigned long vm_flags, struct page **pages) 30493935ed6aSStefani Seibold { 3050a62c34bdSAndy Lutomirski struct vm_area_struct *vma = __install_special_mapping( 3051a62c34bdSAndy Lutomirski mm, addr, len, vm_flags, &legacy_special_mapping_vmops, 3052a62c34bdSAndy Lutomirski (void *)pages); 30533935ed6aSStefani Seibold 305414bd5b45SDuan Jiong return PTR_ERR_OR_ZERO(vma); 3055fa5dc22fSRoland McGrath } 30567906d00cSAndrea Arcangeli 30577906d00cSAndrea Arcangeli static DEFINE_MUTEX(mm_all_locks_mutex); 30587906d00cSAndrea Arcangeli 3059454ed842SPeter Zijlstra static void vm_lock_anon_vma(struct mm_struct *mm, struct anon_vma *anon_vma) 30607906d00cSAndrea Arcangeli { 3061bf181b9fSMichel Lespinasse if (!test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_node)) { 30627906d00cSAndrea Arcangeli /* 30637906d00cSAndrea Arcangeli * The LSB of head.next can't change from under us 30647906d00cSAndrea Arcangeli * because we hold the mm_all_locks_mutex. 30657906d00cSAndrea Arcangeli */ 3066572043c9SJiri Kosina down_write_nest_lock(&anon_vma->root->rwsem, &mm->mmap_sem); 30677906d00cSAndrea Arcangeli /* 30687906d00cSAndrea Arcangeli * We can safely modify head.next after taking the 30695a505085SIngo Molnar * anon_vma->root->rwsem. If some other vma in this mm shares 30707906d00cSAndrea Arcangeli * the same anon_vma we won't take it again. 30717906d00cSAndrea Arcangeli * 30727906d00cSAndrea Arcangeli * No need of atomic instructions here, head.next 30737906d00cSAndrea Arcangeli * can't change from under us thanks to the 30745a505085SIngo Molnar * anon_vma->root->rwsem. 30757906d00cSAndrea Arcangeli */ 30767906d00cSAndrea Arcangeli if (__test_and_set_bit(0, (unsigned long *) 3077bf181b9fSMichel Lespinasse &anon_vma->root->rb_root.rb_node)) 30787906d00cSAndrea Arcangeli BUG(); 30797906d00cSAndrea Arcangeli } 30807906d00cSAndrea Arcangeli } 30817906d00cSAndrea Arcangeli 3082454ed842SPeter Zijlstra static void vm_lock_mapping(struct mm_struct *mm, struct address_space *mapping) 30837906d00cSAndrea Arcangeli { 30847906d00cSAndrea Arcangeli if (!test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) { 30857906d00cSAndrea Arcangeli /* 30867906d00cSAndrea Arcangeli * AS_MM_ALL_LOCKS can't change from under us because 30877906d00cSAndrea Arcangeli * we hold the mm_all_locks_mutex. 30887906d00cSAndrea Arcangeli * 30897906d00cSAndrea Arcangeli * Operations on ->flags have to be atomic because 30907906d00cSAndrea Arcangeli * even if AS_MM_ALL_LOCKS is stable thanks to the 30917906d00cSAndrea Arcangeli * mm_all_locks_mutex, there may be other cpus 30927906d00cSAndrea Arcangeli * changing other bitflags in parallel to us. 30937906d00cSAndrea Arcangeli */ 30947906d00cSAndrea Arcangeli if (test_and_set_bit(AS_MM_ALL_LOCKS, &mapping->flags)) 30957906d00cSAndrea Arcangeli BUG(); 3096c8c06efaSDavidlohr Bueso down_write_nest_lock(&mapping->i_mmap_rwsem, &mm->mmap_sem); 30977906d00cSAndrea Arcangeli } 30987906d00cSAndrea Arcangeli } 30997906d00cSAndrea Arcangeli 31007906d00cSAndrea Arcangeli /* 31017906d00cSAndrea Arcangeli * This operation locks against the VM for all pte/vma/mm related 31027906d00cSAndrea Arcangeli * operations that could ever happen on a certain mm. This includes 31037906d00cSAndrea Arcangeli * vmtruncate, try_to_unmap, and all page faults. 31047906d00cSAndrea Arcangeli * 31057906d00cSAndrea Arcangeli * The caller must take the mmap_sem in write mode before calling 31067906d00cSAndrea Arcangeli * mm_take_all_locks(). The caller isn't allowed to release the 31077906d00cSAndrea Arcangeli * mmap_sem until mm_drop_all_locks() returns. 31087906d00cSAndrea Arcangeli * 31097906d00cSAndrea Arcangeli * mmap_sem in write mode is required in order to block all operations 31107906d00cSAndrea Arcangeli * that could modify pagetables and free pages without need of 31117906d00cSAndrea Arcangeli * altering the vma layout (for example populate_range() with 31127906d00cSAndrea Arcangeli * nonlinear vmas). It's also needed in write mode to avoid new 31137906d00cSAndrea Arcangeli * anon_vmas to be associated with existing vmas. 31147906d00cSAndrea Arcangeli * 31157906d00cSAndrea Arcangeli * A single task can't take more than one mm_take_all_locks() in a row 31167906d00cSAndrea Arcangeli * or it would deadlock. 31177906d00cSAndrea Arcangeli * 3118bf181b9fSMichel Lespinasse * The LSB in anon_vma->rb_root.rb_node and the AS_MM_ALL_LOCKS bitflag in 31197906d00cSAndrea Arcangeli * mapping->flags avoid to take the same lock twice, if more than one 31207906d00cSAndrea Arcangeli * vma in this mm is backed by the same anon_vma or address_space. 31217906d00cSAndrea Arcangeli * 31227906d00cSAndrea Arcangeli * We can take all the locks in random order because the VM code 3123c8c06efaSDavidlohr Bueso * taking i_mmap_rwsem or anon_vma->rwsem outside the mmap_sem never 31247906d00cSAndrea Arcangeli * takes more than one of them in a row. Secondly we're protected 31257906d00cSAndrea Arcangeli * against a concurrent mm_take_all_locks() by the mm_all_locks_mutex. 31267906d00cSAndrea Arcangeli * 31277906d00cSAndrea Arcangeli * mm_take_all_locks() and mm_drop_all_locks are expensive operations 31287906d00cSAndrea Arcangeli * that may have to take thousand of locks. 31297906d00cSAndrea Arcangeli * 31307906d00cSAndrea Arcangeli * mm_take_all_locks() can fail if it's interrupted by signals. 31317906d00cSAndrea Arcangeli */ 31327906d00cSAndrea Arcangeli int mm_take_all_locks(struct mm_struct *mm) 31337906d00cSAndrea Arcangeli { 31347906d00cSAndrea Arcangeli struct vm_area_struct *vma; 31355beb4930SRik van Riel struct anon_vma_chain *avc; 31367906d00cSAndrea Arcangeli 31377906d00cSAndrea Arcangeli BUG_ON(down_read_trylock(&mm->mmap_sem)); 31387906d00cSAndrea Arcangeli 31397906d00cSAndrea Arcangeli mutex_lock(&mm_all_locks_mutex); 31407906d00cSAndrea Arcangeli 31417906d00cSAndrea Arcangeli for (vma = mm->mmap; vma; vma = vma->vm_next) { 31427906d00cSAndrea Arcangeli if (signal_pending(current)) 31437906d00cSAndrea Arcangeli goto out_unlock; 31447906d00cSAndrea Arcangeli if (vma->vm_file && vma->vm_file->f_mapping) 3145454ed842SPeter Zijlstra vm_lock_mapping(mm, vma->vm_file->f_mapping); 31467906d00cSAndrea Arcangeli } 31477cd5a02fSPeter Zijlstra 31487cd5a02fSPeter Zijlstra for (vma = mm->mmap; vma; vma = vma->vm_next) { 31497cd5a02fSPeter Zijlstra if (signal_pending(current)) 31507cd5a02fSPeter Zijlstra goto out_unlock; 31517cd5a02fSPeter Zijlstra if (vma->anon_vma) 31525beb4930SRik van Riel list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) 31535beb4930SRik van Riel vm_lock_anon_vma(mm, avc->anon_vma); 31547cd5a02fSPeter Zijlstra } 31557cd5a02fSPeter Zijlstra 3156584cff54SKautuk Consul return 0; 31577906d00cSAndrea Arcangeli 31587906d00cSAndrea Arcangeli out_unlock: 31597906d00cSAndrea Arcangeli mm_drop_all_locks(mm); 3160584cff54SKautuk Consul return -EINTR; 31617906d00cSAndrea Arcangeli } 31627906d00cSAndrea Arcangeli 31637906d00cSAndrea Arcangeli static void vm_unlock_anon_vma(struct anon_vma *anon_vma) 31647906d00cSAndrea Arcangeli { 3165bf181b9fSMichel Lespinasse if (test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_node)) { 31667906d00cSAndrea Arcangeli /* 31677906d00cSAndrea Arcangeli * The LSB of head.next can't change to 0 from under 31687906d00cSAndrea Arcangeli * us because we hold the mm_all_locks_mutex. 31697906d00cSAndrea Arcangeli * 31707906d00cSAndrea Arcangeli * We must however clear the bitflag before unlocking 3171bf181b9fSMichel Lespinasse * the vma so the users using the anon_vma->rb_root will 31727906d00cSAndrea Arcangeli * never see our bitflag. 31737906d00cSAndrea Arcangeli * 31747906d00cSAndrea Arcangeli * No need of atomic instructions here, head.next 31757906d00cSAndrea Arcangeli * can't change from under us until we release the 31765a505085SIngo Molnar * anon_vma->root->rwsem. 31777906d00cSAndrea Arcangeli */ 31787906d00cSAndrea Arcangeli if (!__test_and_clear_bit(0, (unsigned long *) 3179bf181b9fSMichel Lespinasse &anon_vma->root->rb_root.rb_node)) 31807906d00cSAndrea Arcangeli BUG(); 318108b52706SKonstantin Khlebnikov anon_vma_unlock_write(anon_vma); 31827906d00cSAndrea Arcangeli } 31837906d00cSAndrea Arcangeli } 31847906d00cSAndrea Arcangeli 31857906d00cSAndrea Arcangeli static void vm_unlock_mapping(struct address_space *mapping) 31867906d00cSAndrea Arcangeli { 31877906d00cSAndrea Arcangeli if (test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) { 31887906d00cSAndrea Arcangeli /* 31897906d00cSAndrea Arcangeli * AS_MM_ALL_LOCKS can't change to 0 from under us 31907906d00cSAndrea Arcangeli * because we hold the mm_all_locks_mutex. 31917906d00cSAndrea Arcangeli */ 319283cde9e8SDavidlohr Bueso i_mmap_unlock_write(mapping); 31937906d00cSAndrea Arcangeli if (!test_and_clear_bit(AS_MM_ALL_LOCKS, 31947906d00cSAndrea Arcangeli &mapping->flags)) 31957906d00cSAndrea Arcangeli BUG(); 31967906d00cSAndrea Arcangeli } 31977906d00cSAndrea Arcangeli } 31987906d00cSAndrea Arcangeli 31997906d00cSAndrea Arcangeli /* 32007906d00cSAndrea Arcangeli * The mmap_sem cannot be released by the caller until 32017906d00cSAndrea Arcangeli * mm_drop_all_locks() returns. 32027906d00cSAndrea Arcangeli */ 32037906d00cSAndrea Arcangeli void mm_drop_all_locks(struct mm_struct *mm) 32047906d00cSAndrea Arcangeli { 32057906d00cSAndrea Arcangeli struct vm_area_struct *vma; 32065beb4930SRik van Riel struct anon_vma_chain *avc; 32077906d00cSAndrea Arcangeli 32087906d00cSAndrea Arcangeli BUG_ON(down_read_trylock(&mm->mmap_sem)); 32097906d00cSAndrea Arcangeli BUG_ON(!mutex_is_locked(&mm_all_locks_mutex)); 32107906d00cSAndrea Arcangeli 32117906d00cSAndrea Arcangeli for (vma = mm->mmap; vma; vma = vma->vm_next) { 32127906d00cSAndrea Arcangeli if (vma->anon_vma) 32135beb4930SRik van Riel list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) 32145beb4930SRik van Riel vm_unlock_anon_vma(avc->anon_vma); 32157906d00cSAndrea Arcangeli if (vma->vm_file && vma->vm_file->f_mapping) 32167906d00cSAndrea Arcangeli vm_unlock_mapping(vma->vm_file->f_mapping); 32177906d00cSAndrea Arcangeli } 32187906d00cSAndrea Arcangeli 32197906d00cSAndrea Arcangeli mutex_unlock(&mm_all_locks_mutex); 32207906d00cSAndrea Arcangeli } 32218feae131SDavid Howells 32228feae131SDavid Howells /* 32238feae131SDavid Howells * initialise the VMA slab 32248feae131SDavid Howells */ 32258feae131SDavid Howells void __init mmap_init(void) 32268feae131SDavid Howells { 322700a62ce9SKOSAKI Motohiro int ret; 322800a62ce9SKOSAKI Motohiro 3229908c7f19STejun Heo ret = percpu_counter_init(&vm_committed_as, 0, GFP_KERNEL); 323000a62ce9SKOSAKI Motohiro VM_BUG_ON(ret); 32318feae131SDavid Howells } 3232c9b1d098SAndrew Shewmaker 3233c9b1d098SAndrew Shewmaker /* 3234c9b1d098SAndrew Shewmaker * Initialise sysctl_user_reserve_kbytes. 3235c9b1d098SAndrew Shewmaker * 3236c9b1d098SAndrew Shewmaker * This is intended to prevent a user from starting a single memory hogging 3237c9b1d098SAndrew Shewmaker * process, such that they cannot recover (kill the hog) in OVERCOMMIT_NEVER 3238c9b1d098SAndrew Shewmaker * mode. 3239c9b1d098SAndrew Shewmaker * 3240c9b1d098SAndrew Shewmaker * The default value is min(3% of free memory, 128MB) 3241c9b1d098SAndrew Shewmaker * 128MB is enough to recover with sshd/login, bash, and top/kill. 3242c9b1d098SAndrew Shewmaker */ 32431640879aSAndrew Shewmaker static int init_user_reserve(void) 3244c9b1d098SAndrew Shewmaker { 3245c9b1d098SAndrew Shewmaker unsigned long free_kbytes; 3246c9b1d098SAndrew Shewmaker 3247c9b1d098SAndrew Shewmaker free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10); 3248c9b1d098SAndrew Shewmaker 3249c9b1d098SAndrew Shewmaker sysctl_user_reserve_kbytes = min(free_kbytes / 32, 1UL << 17); 3250c9b1d098SAndrew Shewmaker return 0; 3251c9b1d098SAndrew Shewmaker } 3252a64fb3cdSPaul Gortmaker subsys_initcall(init_user_reserve); 32534eeab4f5SAndrew Shewmaker 32544eeab4f5SAndrew Shewmaker /* 32554eeab4f5SAndrew Shewmaker * Initialise sysctl_admin_reserve_kbytes. 32564eeab4f5SAndrew Shewmaker * 32574eeab4f5SAndrew Shewmaker * The purpose of sysctl_admin_reserve_kbytes is to allow the sys admin 32584eeab4f5SAndrew Shewmaker * to log in and kill a memory hogging process. 32594eeab4f5SAndrew Shewmaker * 32604eeab4f5SAndrew Shewmaker * Systems with more than 256MB will reserve 8MB, enough to recover 32614eeab4f5SAndrew Shewmaker * with sshd, bash, and top in OVERCOMMIT_GUESS. Smaller systems will 32624eeab4f5SAndrew Shewmaker * only reserve 3% of free pages by default. 32634eeab4f5SAndrew Shewmaker */ 32641640879aSAndrew Shewmaker static int init_admin_reserve(void) 32654eeab4f5SAndrew Shewmaker { 32664eeab4f5SAndrew Shewmaker unsigned long free_kbytes; 32674eeab4f5SAndrew Shewmaker 32684eeab4f5SAndrew Shewmaker free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10); 32694eeab4f5SAndrew Shewmaker 32704eeab4f5SAndrew Shewmaker sysctl_admin_reserve_kbytes = min(free_kbytes / 32, 1UL << 13); 32714eeab4f5SAndrew Shewmaker return 0; 32724eeab4f5SAndrew Shewmaker } 3273a64fb3cdSPaul Gortmaker subsys_initcall(init_admin_reserve); 32741640879aSAndrew Shewmaker 32751640879aSAndrew Shewmaker /* 32761640879aSAndrew Shewmaker * Reinititalise user and admin reserves if memory is added or removed. 32771640879aSAndrew Shewmaker * 32781640879aSAndrew Shewmaker * The default user reserve max is 128MB, and the default max for the 32791640879aSAndrew Shewmaker * admin reserve is 8MB. These are usually, but not always, enough to 32801640879aSAndrew Shewmaker * enable recovery from a memory hogging process using login/sshd, a shell, 32811640879aSAndrew Shewmaker * and tools like top. It may make sense to increase or even disable the 32821640879aSAndrew Shewmaker * reserve depending on the existence of swap or variations in the recovery 32831640879aSAndrew Shewmaker * tools. So, the admin may have changed them. 32841640879aSAndrew Shewmaker * 32851640879aSAndrew Shewmaker * If memory is added and the reserves have been eliminated or increased above 32861640879aSAndrew Shewmaker * the default max, then we'll trust the admin. 32871640879aSAndrew Shewmaker * 32881640879aSAndrew Shewmaker * If memory is removed and there isn't enough free memory, then we 32891640879aSAndrew Shewmaker * need to reset the reserves. 32901640879aSAndrew Shewmaker * 32911640879aSAndrew Shewmaker * Otherwise keep the reserve set by the admin. 32921640879aSAndrew Shewmaker */ 32931640879aSAndrew Shewmaker static int reserve_mem_notifier(struct notifier_block *nb, 32941640879aSAndrew Shewmaker unsigned long action, void *data) 32951640879aSAndrew Shewmaker { 32961640879aSAndrew Shewmaker unsigned long tmp, free_kbytes; 32971640879aSAndrew Shewmaker 32981640879aSAndrew Shewmaker switch (action) { 32991640879aSAndrew Shewmaker case MEM_ONLINE: 33001640879aSAndrew Shewmaker /* Default max is 128MB. Leave alone if modified by operator. */ 33011640879aSAndrew Shewmaker tmp = sysctl_user_reserve_kbytes; 33021640879aSAndrew Shewmaker if (0 < tmp && tmp < (1UL << 17)) 33031640879aSAndrew Shewmaker init_user_reserve(); 33041640879aSAndrew Shewmaker 33051640879aSAndrew Shewmaker /* Default max is 8MB. Leave alone if modified by operator. */ 33061640879aSAndrew Shewmaker tmp = sysctl_admin_reserve_kbytes; 33071640879aSAndrew Shewmaker if (0 < tmp && tmp < (1UL << 13)) 33081640879aSAndrew Shewmaker init_admin_reserve(); 33091640879aSAndrew Shewmaker 33101640879aSAndrew Shewmaker break; 33111640879aSAndrew Shewmaker case MEM_OFFLINE: 33121640879aSAndrew Shewmaker free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10); 33131640879aSAndrew Shewmaker 33141640879aSAndrew Shewmaker if (sysctl_user_reserve_kbytes > free_kbytes) { 33151640879aSAndrew Shewmaker init_user_reserve(); 33161640879aSAndrew Shewmaker pr_info("vm.user_reserve_kbytes reset to %lu\n", 33171640879aSAndrew Shewmaker sysctl_user_reserve_kbytes); 33181640879aSAndrew Shewmaker } 33191640879aSAndrew Shewmaker 33201640879aSAndrew Shewmaker if (sysctl_admin_reserve_kbytes > free_kbytes) { 33211640879aSAndrew Shewmaker init_admin_reserve(); 33221640879aSAndrew Shewmaker pr_info("vm.admin_reserve_kbytes reset to %lu\n", 33231640879aSAndrew Shewmaker sysctl_admin_reserve_kbytes); 33241640879aSAndrew Shewmaker } 33251640879aSAndrew Shewmaker break; 33261640879aSAndrew Shewmaker default: 33271640879aSAndrew Shewmaker break; 33281640879aSAndrew Shewmaker } 33291640879aSAndrew Shewmaker return NOTIFY_OK; 33301640879aSAndrew Shewmaker } 33311640879aSAndrew Shewmaker 33321640879aSAndrew Shewmaker static struct notifier_block reserve_mem_nb = { 33331640879aSAndrew Shewmaker .notifier_call = reserve_mem_notifier, 33341640879aSAndrew Shewmaker }; 33351640879aSAndrew Shewmaker 33361640879aSAndrew Shewmaker static int __meminit init_reserve_notifier(void) 33371640879aSAndrew Shewmaker { 33381640879aSAndrew Shewmaker if (register_hotmemory_notifier(&reserve_mem_nb)) 3339b1de0d13SMitchel Humpherys pr_err("Failed registering memory add/remove notifier for admin reserve\n"); 33401640879aSAndrew Shewmaker 33411640879aSAndrew Shewmaker return 0; 33421640879aSAndrew Shewmaker } 3343a64fb3cdSPaul Gortmaker subsys_initcall(init_reserve_notifier); 3344