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> 4419a809afSAndrea Arcangeli #include <linux/userfaultfd_k.h> 45d977d56cSKonstantin Khlebnikov #include <linux/moduleparam.h> 46*62b5f7d0SDave Hansen #include <linux/pkeys.h> 471da177e4SLinus Torvalds 481da177e4SLinus Torvalds #include <asm/uaccess.h> 491da177e4SLinus Torvalds #include <asm/cacheflush.h> 501da177e4SLinus Torvalds #include <asm/tlb.h> 51d6dd61c8SJeremy Fitzhardinge #include <asm/mmu_context.h> 521da177e4SLinus Torvalds 5342b77728SJan Beulich #include "internal.h" 5442b77728SJan Beulich 553a459756SKirill Korotaev #ifndef arch_mmap_check 563a459756SKirill Korotaev #define arch_mmap_check(addr, len, flags) (0) 573a459756SKirill Korotaev #endif 583a459756SKirill Korotaev 5908e7d9b5SMartin Schwidefsky #ifndef arch_rebalance_pgtables 6008e7d9b5SMartin Schwidefsky #define arch_rebalance_pgtables(addr, len) (addr) 6108e7d9b5SMartin Schwidefsky #endif 6208e7d9b5SMartin Schwidefsky 63d07e2259SDaniel Cashman #ifdef CONFIG_HAVE_ARCH_MMAP_RND_BITS 64d07e2259SDaniel Cashman const int mmap_rnd_bits_min = CONFIG_ARCH_MMAP_RND_BITS_MIN; 65d07e2259SDaniel Cashman const int mmap_rnd_bits_max = CONFIG_ARCH_MMAP_RND_BITS_MAX; 66d07e2259SDaniel Cashman int mmap_rnd_bits __read_mostly = CONFIG_ARCH_MMAP_RND_BITS; 67d07e2259SDaniel Cashman #endif 68d07e2259SDaniel Cashman #ifdef CONFIG_HAVE_ARCH_MMAP_RND_COMPAT_BITS 69d07e2259SDaniel Cashman const int mmap_rnd_compat_bits_min = CONFIG_ARCH_MMAP_RND_COMPAT_BITS_MIN; 70d07e2259SDaniel Cashman const int mmap_rnd_compat_bits_max = CONFIG_ARCH_MMAP_RND_COMPAT_BITS_MAX; 71d07e2259SDaniel Cashman int mmap_rnd_compat_bits __read_mostly = CONFIG_ARCH_MMAP_RND_COMPAT_BITS; 72d07e2259SDaniel Cashman #endif 73d07e2259SDaniel Cashman 74d977d56cSKonstantin Khlebnikov static bool ignore_rlimit_data = true; 75d977d56cSKonstantin Khlebnikov core_param(ignore_rlimit_data, ignore_rlimit_data, bool, 0644); 76d07e2259SDaniel Cashman 77e0da382cSHugh Dickins static void unmap_region(struct mm_struct *mm, 78e0da382cSHugh Dickins struct vm_area_struct *vma, struct vm_area_struct *prev, 79e0da382cSHugh Dickins unsigned long start, unsigned long end); 80e0da382cSHugh Dickins 811da177e4SLinus Torvalds /* description of effects of mapping type and prot in current implementation. 821da177e4SLinus Torvalds * this is due to the limited x86 page protection hardware. The expected 831da177e4SLinus Torvalds * behavior is in parens: 841da177e4SLinus Torvalds * 851da177e4SLinus Torvalds * map_type prot 861da177e4SLinus Torvalds * PROT_NONE PROT_READ PROT_WRITE PROT_EXEC 871da177e4SLinus Torvalds * MAP_SHARED r: (no) no r: (yes) yes r: (no) yes r: (no) yes 881da177e4SLinus Torvalds * w: (no) no w: (no) no w: (yes) yes w: (no) no 891da177e4SLinus Torvalds * x: (no) no x: (no) yes x: (no) yes x: (yes) yes 901da177e4SLinus Torvalds * 911da177e4SLinus Torvalds * MAP_PRIVATE r: (no) no r: (yes) yes r: (no) yes r: (no) yes 921da177e4SLinus Torvalds * w: (no) no w: (no) no w: (copy) copy w: (no) no 931da177e4SLinus Torvalds * x: (no) no x: (no) yes x: (no) yes x: (yes) yes 941da177e4SLinus Torvalds * 951da177e4SLinus Torvalds */ 961da177e4SLinus Torvalds pgprot_t protection_map[16] = { 971da177e4SLinus Torvalds __P000, __P001, __P010, __P011, __P100, __P101, __P110, __P111, 981da177e4SLinus Torvalds __S000, __S001, __S010, __S011, __S100, __S101, __S110, __S111 991da177e4SLinus Torvalds }; 1001da177e4SLinus Torvalds 101804af2cfSHugh Dickins pgprot_t vm_get_page_prot(unsigned long vm_flags) 102804af2cfSHugh Dickins { 103b845f313SDave Kleikamp return __pgprot(pgprot_val(protection_map[vm_flags & 104b845f313SDave Kleikamp (VM_READ|VM_WRITE|VM_EXEC|VM_SHARED)]) | 105b845f313SDave Kleikamp pgprot_val(arch_vm_get_page_prot(vm_flags))); 106804af2cfSHugh Dickins } 107804af2cfSHugh Dickins EXPORT_SYMBOL(vm_get_page_prot); 108804af2cfSHugh Dickins 10964e45507SPeter Feiner static pgprot_t vm_pgprot_modify(pgprot_t oldprot, unsigned long vm_flags) 11064e45507SPeter Feiner { 11164e45507SPeter Feiner return pgprot_modify(oldprot, vm_get_page_prot(vm_flags)); 11264e45507SPeter Feiner } 11364e45507SPeter Feiner 11464e45507SPeter Feiner /* Update vma->vm_page_prot to reflect vma->vm_flags. */ 11564e45507SPeter Feiner void vma_set_page_prot(struct vm_area_struct *vma) 11664e45507SPeter Feiner { 11764e45507SPeter Feiner unsigned long vm_flags = vma->vm_flags; 11864e45507SPeter Feiner 11964e45507SPeter Feiner vma->vm_page_prot = vm_pgprot_modify(vma->vm_page_prot, vm_flags); 12064e45507SPeter Feiner if (vma_wants_writenotify(vma)) { 12164e45507SPeter Feiner vm_flags &= ~VM_SHARED; 12264e45507SPeter Feiner vma->vm_page_prot = vm_pgprot_modify(vma->vm_page_prot, 12364e45507SPeter Feiner vm_flags); 12464e45507SPeter Feiner } 12564e45507SPeter Feiner } 12664e45507SPeter Feiner 12764e45507SPeter Feiner 12834679d7eSShaohua Li int sysctl_overcommit_memory __read_mostly = OVERCOMMIT_GUESS; /* heuristic overcommit */ 12934679d7eSShaohua Li int sysctl_overcommit_ratio __read_mostly = 50; /* default is 50% */ 13049f0ce5fSJerome Marchand unsigned long sysctl_overcommit_kbytes __read_mostly; 131c3d8c141SChristoph Lameter int sysctl_max_map_count __read_mostly = DEFAULT_MAX_MAP_COUNT; 132c9b1d098SAndrew Shewmaker unsigned long sysctl_user_reserve_kbytes __read_mostly = 1UL << 17; /* 128MB */ 1334eeab4f5SAndrew Shewmaker unsigned long sysctl_admin_reserve_kbytes __read_mostly = 1UL << 13; /* 8MB */ 13434679d7eSShaohua Li /* 13534679d7eSShaohua Li * Make sure vm_committed_as in one cacheline and not cacheline shared with 13634679d7eSShaohua Li * other variables. It can be updated by several CPUs frequently. 13734679d7eSShaohua Li */ 13834679d7eSShaohua Li struct percpu_counter vm_committed_as ____cacheline_aligned_in_smp; 1391da177e4SLinus Torvalds 1401da177e4SLinus Torvalds /* 141997071bcSK. Y. Srinivasan * The global memory commitment made in the system can be a metric 142997071bcSK. Y. Srinivasan * that can be used to drive ballooning decisions when Linux is hosted 143997071bcSK. Y. Srinivasan * as a guest. On Hyper-V, the host implements a policy engine for dynamically 144997071bcSK. Y. Srinivasan * balancing memory across competing virtual machines that are hosted. 145997071bcSK. Y. Srinivasan * Several metrics drive this policy engine including the guest reported 146997071bcSK. Y. Srinivasan * memory commitment. 147997071bcSK. Y. Srinivasan */ 148997071bcSK. Y. Srinivasan unsigned long vm_memory_committed(void) 149997071bcSK. Y. Srinivasan { 150997071bcSK. Y. Srinivasan return percpu_counter_read_positive(&vm_committed_as); 151997071bcSK. Y. Srinivasan } 152997071bcSK. Y. Srinivasan EXPORT_SYMBOL_GPL(vm_memory_committed); 153997071bcSK. Y. Srinivasan 154997071bcSK. Y. Srinivasan /* 1551da177e4SLinus Torvalds * Check that a process has enough memory to allocate a new virtual 1561da177e4SLinus Torvalds * mapping. 0 means there is enough memory for the allocation to 1571da177e4SLinus Torvalds * succeed and -ENOMEM implies there is not. 1581da177e4SLinus Torvalds * 1591da177e4SLinus Torvalds * We currently support three overcommit policies, which are set via the 1601da177e4SLinus Torvalds * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting 1611da177e4SLinus Torvalds * 1621da177e4SLinus Torvalds * Strict overcommit modes added 2002 Feb 26 by Alan Cox. 1631da177e4SLinus Torvalds * Additional code 2002 Jul 20 by Robert Love. 1641da177e4SLinus Torvalds * 1651da177e4SLinus Torvalds * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise. 1661da177e4SLinus Torvalds * 1671da177e4SLinus Torvalds * Note this is a helper function intended to be used by LSMs which 1681da177e4SLinus Torvalds * wish to use this logic. 1691da177e4SLinus Torvalds */ 17034b4e4aaSAlan Cox int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin) 1711da177e4SLinus Torvalds { 1725703b087SRoman Gushchin long free, allowed, reserve; 1731da177e4SLinus Torvalds 17482f71ae4SKonstantin Khlebnikov VM_WARN_ONCE(percpu_counter_read(&vm_committed_as) < 17582f71ae4SKonstantin Khlebnikov -(s64)vm_committed_as_batch * num_online_cpus(), 17682f71ae4SKonstantin Khlebnikov "memory commitment underflow"); 17782f71ae4SKonstantin Khlebnikov 1781da177e4SLinus Torvalds vm_acct_memory(pages); 1791da177e4SLinus Torvalds 1801da177e4SLinus Torvalds /* 1811da177e4SLinus Torvalds * Sometimes we want to use more memory than we have 1821da177e4SLinus Torvalds */ 1831da177e4SLinus Torvalds if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS) 1841da177e4SLinus Torvalds return 0; 1851da177e4SLinus Torvalds 1861da177e4SLinus Torvalds if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) { 187c15bef30SDmitry Fink free = global_page_state(NR_FREE_PAGES); 188c15bef30SDmitry Fink free += global_page_state(NR_FILE_PAGES); 1891da177e4SLinus Torvalds 190c15bef30SDmitry Fink /* 191c15bef30SDmitry Fink * shmem pages shouldn't be counted as free in this 192c15bef30SDmitry Fink * case, they can't be purged, only swapped out, and 193c15bef30SDmitry Fink * that won't affect the overall amount of available 194c15bef30SDmitry Fink * memory in the system. 195c15bef30SDmitry Fink */ 196c15bef30SDmitry Fink free -= global_page_state(NR_SHMEM); 197c15bef30SDmitry Fink 198ec8acf20SShaohua Li free += get_nr_swap_pages(); 1991da177e4SLinus Torvalds 2001da177e4SLinus Torvalds /* 2011da177e4SLinus Torvalds * Any slabs which are created with the 2021da177e4SLinus Torvalds * SLAB_RECLAIM_ACCOUNT flag claim to have contents 2031da177e4SLinus Torvalds * which are reclaimable, under pressure. The dentry 2041da177e4SLinus Torvalds * cache and most inode caches should fall into this 2051da177e4SLinus Torvalds */ 206972d1a7bSChristoph Lameter free += global_page_state(NR_SLAB_RECLAIMABLE); 2071da177e4SLinus Torvalds 2081da177e4SLinus Torvalds /* 209c15bef30SDmitry Fink * Leave reserved pages. The pages are not for anonymous pages. 210c15bef30SDmitry Fink */ 211c15bef30SDmitry Fink if (free <= totalreserve_pages) 212c15bef30SDmitry Fink goto error; 213c15bef30SDmitry Fink else 214c15bef30SDmitry Fink free -= totalreserve_pages; 215c15bef30SDmitry Fink 216c15bef30SDmitry Fink /* 2174eeab4f5SAndrew Shewmaker * Reserve some for root 2181da177e4SLinus Torvalds */ 2191da177e4SLinus Torvalds if (!cap_sys_admin) 2204eeab4f5SAndrew Shewmaker free -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10); 2211da177e4SLinus Torvalds 2221da177e4SLinus Torvalds if (free > pages) 2231da177e4SLinus Torvalds return 0; 2241da177e4SLinus Torvalds 2256d9f7839SHideo AOKI goto error; 2261da177e4SLinus Torvalds } 2271da177e4SLinus Torvalds 22800619bccSJerome Marchand allowed = vm_commit_limit(); 2291da177e4SLinus Torvalds /* 2304eeab4f5SAndrew Shewmaker * Reserve some for root 2311da177e4SLinus Torvalds */ 2321da177e4SLinus Torvalds if (!cap_sys_admin) 2334eeab4f5SAndrew Shewmaker allowed -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10); 2341da177e4SLinus Torvalds 235c9b1d098SAndrew Shewmaker /* 236c9b1d098SAndrew Shewmaker * Don't let a single process grow so big a user can't recover 237c9b1d098SAndrew Shewmaker */ 238c9b1d098SAndrew Shewmaker if (mm) { 239c9b1d098SAndrew Shewmaker reserve = sysctl_user_reserve_kbytes >> (PAGE_SHIFT - 10); 2405703b087SRoman Gushchin allowed -= min_t(long, mm->total_vm / 32, reserve); 241c9b1d098SAndrew Shewmaker } 2421da177e4SLinus Torvalds 24300a62ce9SKOSAKI Motohiro if (percpu_counter_read_positive(&vm_committed_as) < allowed) 2441da177e4SLinus Torvalds return 0; 2456d9f7839SHideo AOKI error: 2461da177e4SLinus Torvalds vm_unacct_memory(pages); 2471da177e4SLinus Torvalds 2481da177e4SLinus Torvalds return -ENOMEM; 2491da177e4SLinus Torvalds } 2501da177e4SLinus Torvalds 2511da177e4SLinus Torvalds /* 252c8c06efaSDavidlohr Bueso * Requires inode->i_mapping->i_mmap_rwsem 2531da177e4SLinus Torvalds */ 2541da177e4SLinus Torvalds static void __remove_shared_vm_struct(struct vm_area_struct *vma, 2551da177e4SLinus Torvalds struct file *file, struct address_space *mapping) 2561da177e4SLinus Torvalds { 2571da177e4SLinus Torvalds if (vma->vm_flags & VM_DENYWRITE) 258496ad9aaSAl Viro atomic_inc(&file_inode(file)->i_writecount); 2591da177e4SLinus Torvalds if (vma->vm_flags & VM_SHARED) 2604bb5f5d9SDavid Herrmann mapping_unmap_writable(mapping); 2611da177e4SLinus Torvalds 2621da177e4SLinus Torvalds flush_dcache_mmap_lock(mapping); 2636b2dbba8SMichel Lespinasse vma_interval_tree_remove(vma, &mapping->i_mmap); 2641da177e4SLinus Torvalds flush_dcache_mmap_unlock(mapping); 2651da177e4SLinus Torvalds } 2661da177e4SLinus Torvalds 2671da177e4SLinus Torvalds /* 2686b2dbba8SMichel Lespinasse * Unlink a file-based vm structure from its interval tree, to hide 269a8fb5618SHugh Dickins * vma from rmap and vmtruncate before freeing its page tables. 2701da177e4SLinus Torvalds */ 271a8fb5618SHugh Dickins void unlink_file_vma(struct vm_area_struct *vma) 2721da177e4SLinus Torvalds { 2731da177e4SLinus Torvalds struct file *file = vma->vm_file; 2741da177e4SLinus Torvalds 2751da177e4SLinus Torvalds if (file) { 2761da177e4SLinus Torvalds struct address_space *mapping = file->f_mapping; 27783cde9e8SDavidlohr Bueso i_mmap_lock_write(mapping); 2781da177e4SLinus Torvalds __remove_shared_vm_struct(vma, file, mapping); 27983cde9e8SDavidlohr Bueso i_mmap_unlock_write(mapping); 2801da177e4SLinus Torvalds } 281a8fb5618SHugh Dickins } 282a8fb5618SHugh Dickins 283a8fb5618SHugh Dickins /* 284a8fb5618SHugh Dickins * Close a vm structure and free it, returning the next. 285a8fb5618SHugh Dickins */ 286a8fb5618SHugh Dickins static struct vm_area_struct *remove_vma(struct vm_area_struct *vma) 287a8fb5618SHugh Dickins { 288a8fb5618SHugh Dickins struct vm_area_struct *next = vma->vm_next; 289a8fb5618SHugh Dickins 290a8fb5618SHugh Dickins might_sleep(); 2911da177e4SLinus Torvalds if (vma->vm_ops && vma->vm_ops->close) 2921da177e4SLinus Torvalds vma->vm_ops->close(vma); 293e9714acfSKonstantin Khlebnikov if (vma->vm_file) 294a8fb5618SHugh Dickins fput(vma->vm_file); 295f0be3d32SLee Schermerhorn mpol_put(vma_policy(vma)); 2961da177e4SLinus Torvalds kmem_cache_free(vm_area_cachep, vma); 297a8fb5618SHugh Dickins return next; 2981da177e4SLinus Torvalds } 2991da177e4SLinus Torvalds 300e4eb1ff6SLinus Torvalds static unsigned long do_brk(unsigned long addr, unsigned long len); 301e4eb1ff6SLinus Torvalds 3026a6160a7SHeiko Carstens SYSCALL_DEFINE1(brk, unsigned long, brk) 3031da177e4SLinus Torvalds { 3048764b338SCyrill Gorcunov unsigned long retval; 3051da177e4SLinus Torvalds unsigned long newbrk, oldbrk; 3061da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 307a5b4592cSJiri Kosina unsigned long min_brk; 308128557ffSMichel Lespinasse bool populate; 3091da177e4SLinus Torvalds 3101da177e4SLinus Torvalds down_write(&mm->mmap_sem); 3111da177e4SLinus Torvalds 312a5b4592cSJiri Kosina #ifdef CONFIG_COMPAT_BRK 3135520e894SJiri Kosina /* 3145520e894SJiri Kosina * CONFIG_COMPAT_BRK can still be overridden by setting 3155520e894SJiri Kosina * randomize_va_space to 2, which will still cause mm->start_brk 3165520e894SJiri Kosina * to be arbitrarily shifted 3175520e894SJiri Kosina */ 3184471a675SJiri Kosina if (current->brk_randomized) 3195520e894SJiri Kosina min_brk = mm->start_brk; 3205520e894SJiri Kosina else 3215520e894SJiri Kosina min_brk = mm->end_data; 322a5b4592cSJiri Kosina #else 323a5b4592cSJiri Kosina min_brk = mm->start_brk; 324a5b4592cSJiri Kosina #endif 325a5b4592cSJiri Kosina if (brk < min_brk) 3261da177e4SLinus Torvalds goto out; 3271e624196SRam Gupta 3281e624196SRam Gupta /* 3291e624196SRam Gupta * Check against rlimit here. If this check is done later after the test 3301e624196SRam Gupta * of oldbrk with newbrk then it can escape the test and let the data 3311e624196SRam Gupta * segment grow beyond its set limit the in case where the limit is 3321e624196SRam Gupta * not page aligned -Ram Gupta 3331e624196SRam Gupta */ 3348764b338SCyrill Gorcunov if (check_data_rlimit(rlimit(RLIMIT_DATA), brk, mm->start_brk, 3358764b338SCyrill Gorcunov mm->end_data, mm->start_data)) 3361e624196SRam Gupta goto out; 3371e624196SRam Gupta 3381da177e4SLinus Torvalds newbrk = PAGE_ALIGN(brk); 3391da177e4SLinus Torvalds oldbrk = PAGE_ALIGN(mm->brk); 3401da177e4SLinus Torvalds if (oldbrk == newbrk) 3411da177e4SLinus Torvalds goto set_brk; 3421da177e4SLinus Torvalds 3431da177e4SLinus Torvalds /* Always allow shrinking brk. */ 3441da177e4SLinus Torvalds if (brk <= mm->brk) { 3451da177e4SLinus Torvalds if (!do_munmap(mm, newbrk, oldbrk-newbrk)) 3461da177e4SLinus Torvalds goto set_brk; 3471da177e4SLinus Torvalds goto out; 3481da177e4SLinus Torvalds } 3491da177e4SLinus Torvalds 3501da177e4SLinus Torvalds /* Check against existing mmap mappings. */ 3511da177e4SLinus Torvalds if (find_vma_intersection(mm, oldbrk, newbrk+PAGE_SIZE)) 3521da177e4SLinus Torvalds goto out; 3531da177e4SLinus Torvalds 3541da177e4SLinus Torvalds /* Ok, looks good - let it rip. */ 3551da177e4SLinus Torvalds if (do_brk(oldbrk, newbrk-oldbrk) != oldbrk) 3561da177e4SLinus Torvalds goto out; 357128557ffSMichel Lespinasse 3581da177e4SLinus Torvalds set_brk: 3591da177e4SLinus Torvalds mm->brk = brk; 360128557ffSMichel Lespinasse populate = newbrk > oldbrk && (mm->def_flags & VM_LOCKED) != 0; 361128557ffSMichel Lespinasse up_write(&mm->mmap_sem); 362128557ffSMichel Lespinasse if (populate) 363128557ffSMichel Lespinasse mm_populate(oldbrk, newbrk - oldbrk); 364128557ffSMichel Lespinasse return brk; 365128557ffSMichel Lespinasse 3661da177e4SLinus Torvalds out: 3671da177e4SLinus Torvalds retval = mm->brk; 3681da177e4SLinus Torvalds up_write(&mm->mmap_sem); 3691da177e4SLinus Torvalds return retval; 3701da177e4SLinus Torvalds } 3711da177e4SLinus Torvalds 372d3737187SMichel Lespinasse static long vma_compute_subtree_gap(struct vm_area_struct *vma) 373d3737187SMichel Lespinasse { 374d3737187SMichel Lespinasse unsigned long max, subtree_gap; 375d3737187SMichel Lespinasse max = vma->vm_start; 376d3737187SMichel Lespinasse if (vma->vm_prev) 377d3737187SMichel Lespinasse max -= vma->vm_prev->vm_end; 378d3737187SMichel Lespinasse if (vma->vm_rb.rb_left) { 379d3737187SMichel Lespinasse subtree_gap = rb_entry(vma->vm_rb.rb_left, 380d3737187SMichel Lespinasse struct vm_area_struct, vm_rb)->rb_subtree_gap; 381d3737187SMichel Lespinasse if (subtree_gap > max) 382d3737187SMichel Lespinasse max = subtree_gap; 383d3737187SMichel Lespinasse } 384d3737187SMichel Lespinasse if (vma->vm_rb.rb_right) { 385d3737187SMichel Lespinasse subtree_gap = rb_entry(vma->vm_rb.rb_right, 386d3737187SMichel Lespinasse struct vm_area_struct, vm_rb)->rb_subtree_gap; 387d3737187SMichel Lespinasse if (subtree_gap > max) 388d3737187SMichel Lespinasse max = subtree_gap; 389d3737187SMichel Lespinasse } 390d3737187SMichel Lespinasse return max; 391d3737187SMichel Lespinasse } 392d3737187SMichel Lespinasse 393ed8ea815SMichel Lespinasse #ifdef CONFIG_DEBUG_VM_RB 394acf128d0SAndrea Arcangeli static int browse_rb(struct mm_struct *mm) 3951da177e4SLinus Torvalds { 396acf128d0SAndrea Arcangeli struct rb_root *root = &mm->mm_rb; 3975a0768f6SMichel Lespinasse int i = 0, j, bug = 0; 3981da177e4SLinus Torvalds struct rb_node *nd, *pn = NULL; 3991da177e4SLinus Torvalds unsigned long prev = 0, pend = 0; 4001da177e4SLinus Torvalds 4011da177e4SLinus Torvalds for (nd = rb_first(root); nd; nd = rb_next(nd)) { 4021da177e4SLinus Torvalds struct vm_area_struct *vma; 4031da177e4SLinus Torvalds vma = rb_entry(nd, struct vm_area_struct, vm_rb); 4045a0768f6SMichel Lespinasse if (vma->vm_start < prev) { 405ff26f70fSAndrew Morton pr_emerg("vm_start %lx < prev %lx\n", 406ff26f70fSAndrew Morton vma->vm_start, prev); 4075a0768f6SMichel Lespinasse bug = 1; 4085a0768f6SMichel Lespinasse } 4095a0768f6SMichel Lespinasse if (vma->vm_start < pend) { 410ff26f70fSAndrew Morton pr_emerg("vm_start %lx < pend %lx\n", 411ff26f70fSAndrew Morton vma->vm_start, pend); 4125a0768f6SMichel Lespinasse bug = 1; 4135a0768f6SMichel Lespinasse } 4145a0768f6SMichel Lespinasse if (vma->vm_start > vma->vm_end) { 415ff26f70fSAndrew Morton pr_emerg("vm_start %lx > vm_end %lx\n", 416ff26f70fSAndrew Morton vma->vm_start, vma->vm_end); 4175a0768f6SMichel Lespinasse bug = 1; 4185a0768f6SMichel Lespinasse } 419acf128d0SAndrea Arcangeli spin_lock(&mm->page_table_lock); 4205a0768f6SMichel Lespinasse if (vma->rb_subtree_gap != vma_compute_subtree_gap(vma)) { 4218542bdfcSSasha Levin pr_emerg("free gap %lx, correct %lx\n", 4225a0768f6SMichel Lespinasse vma->rb_subtree_gap, 4235a0768f6SMichel Lespinasse vma_compute_subtree_gap(vma)); 4245a0768f6SMichel Lespinasse bug = 1; 4255a0768f6SMichel Lespinasse } 426acf128d0SAndrea Arcangeli spin_unlock(&mm->page_table_lock); 4271da177e4SLinus Torvalds i++; 4281da177e4SLinus Torvalds pn = nd; 429d1af65d1SDavid Miller prev = vma->vm_start; 430d1af65d1SDavid Miller pend = vma->vm_end; 4311da177e4SLinus Torvalds } 4321da177e4SLinus Torvalds j = 0; 4335a0768f6SMichel Lespinasse for (nd = pn; nd; nd = rb_prev(nd)) 4341da177e4SLinus Torvalds j++; 4355a0768f6SMichel Lespinasse if (i != j) { 4368542bdfcSSasha Levin pr_emerg("backwards %d, forwards %d\n", j, i); 4375a0768f6SMichel Lespinasse bug = 1; 4381da177e4SLinus Torvalds } 4395a0768f6SMichel Lespinasse return bug ? -1 : i; 4401da177e4SLinus Torvalds } 4411da177e4SLinus Torvalds 442d3737187SMichel Lespinasse static void validate_mm_rb(struct rb_root *root, struct vm_area_struct *ignore) 443d3737187SMichel Lespinasse { 444d3737187SMichel Lespinasse struct rb_node *nd; 445d3737187SMichel Lespinasse 446d3737187SMichel Lespinasse for (nd = rb_first(root); nd; nd = rb_next(nd)) { 447d3737187SMichel Lespinasse struct vm_area_struct *vma; 448d3737187SMichel Lespinasse vma = rb_entry(nd, struct vm_area_struct, vm_rb); 44996dad67fSSasha Levin VM_BUG_ON_VMA(vma != ignore && 45096dad67fSSasha Levin vma->rb_subtree_gap != vma_compute_subtree_gap(vma), 45196dad67fSSasha Levin vma); 452d3737187SMichel Lespinasse } 4531da177e4SLinus Torvalds } 4541da177e4SLinus Torvalds 455eafd4dc4SRashika Kheria static void validate_mm(struct mm_struct *mm) 4561da177e4SLinus Torvalds { 4571da177e4SLinus Torvalds int bug = 0; 4581da177e4SLinus Torvalds int i = 0; 4595a0768f6SMichel Lespinasse unsigned long highest_address = 0; 460ed8ea815SMichel Lespinasse struct vm_area_struct *vma = mm->mmap; 461ff26f70fSAndrew Morton 462ed8ea815SMichel Lespinasse while (vma) { 46312352d3cSKonstantin Khlebnikov struct anon_vma *anon_vma = vma->anon_vma; 464ed8ea815SMichel Lespinasse struct anon_vma_chain *avc; 465ff26f70fSAndrew Morton 46612352d3cSKonstantin Khlebnikov if (anon_vma) { 46712352d3cSKonstantin Khlebnikov anon_vma_lock_read(anon_vma); 468ed8ea815SMichel Lespinasse list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) 469ed8ea815SMichel Lespinasse anon_vma_interval_tree_verify(avc); 47012352d3cSKonstantin Khlebnikov anon_vma_unlock_read(anon_vma); 47112352d3cSKonstantin Khlebnikov } 47212352d3cSKonstantin Khlebnikov 4735a0768f6SMichel Lespinasse highest_address = vma->vm_end; 474ed8ea815SMichel Lespinasse vma = vma->vm_next; 4751da177e4SLinus Torvalds i++; 4761da177e4SLinus Torvalds } 4775a0768f6SMichel Lespinasse if (i != mm->map_count) { 4788542bdfcSSasha Levin pr_emerg("map_count %d vm_next %d\n", mm->map_count, i); 4795a0768f6SMichel Lespinasse bug = 1; 4805a0768f6SMichel Lespinasse } 4815a0768f6SMichel Lespinasse if (highest_address != mm->highest_vm_end) { 4828542bdfcSSasha Levin pr_emerg("mm->highest_vm_end %lx, found %lx\n", 4835a0768f6SMichel Lespinasse mm->highest_vm_end, highest_address); 4845a0768f6SMichel Lespinasse bug = 1; 4855a0768f6SMichel Lespinasse } 486acf128d0SAndrea Arcangeli i = browse_rb(mm); 4875a0768f6SMichel Lespinasse if (i != mm->map_count) { 488ff26f70fSAndrew Morton if (i != -1) 4898542bdfcSSasha Levin pr_emerg("map_count %d rb %d\n", mm->map_count, i); 4905a0768f6SMichel Lespinasse bug = 1; 4915a0768f6SMichel Lespinasse } 49296dad67fSSasha Levin VM_BUG_ON_MM(bug, mm); 4931da177e4SLinus Torvalds } 4941da177e4SLinus Torvalds #else 495d3737187SMichel Lespinasse #define validate_mm_rb(root, ignore) do { } while (0) 4961da177e4SLinus Torvalds #define validate_mm(mm) do { } while (0) 4971da177e4SLinus Torvalds #endif 4981da177e4SLinus Torvalds 499d3737187SMichel Lespinasse RB_DECLARE_CALLBACKS(static, vma_gap_callbacks, struct vm_area_struct, vm_rb, 500d3737187SMichel Lespinasse unsigned long, rb_subtree_gap, vma_compute_subtree_gap) 501d3737187SMichel Lespinasse 502d3737187SMichel Lespinasse /* 503d3737187SMichel Lespinasse * Update augmented rbtree rb_subtree_gap values after vma->vm_start or 504d3737187SMichel Lespinasse * vma->vm_prev->vm_end values changed, without modifying the vma's position 505d3737187SMichel Lespinasse * in the rbtree. 506d3737187SMichel Lespinasse */ 507d3737187SMichel Lespinasse static void vma_gap_update(struct vm_area_struct *vma) 508d3737187SMichel Lespinasse { 509d3737187SMichel Lespinasse /* 510d3737187SMichel Lespinasse * As it turns out, RB_DECLARE_CALLBACKS() already created a callback 511d3737187SMichel Lespinasse * function that does exacltly what we want. 512d3737187SMichel Lespinasse */ 513d3737187SMichel Lespinasse vma_gap_callbacks_propagate(&vma->vm_rb, NULL); 514d3737187SMichel Lespinasse } 515d3737187SMichel Lespinasse 516d3737187SMichel Lespinasse static inline void vma_rb_insert(struct vm_area_struct *vma, 517d3737187SMichel Lespinasse struct rb_root *root) 518d3737187SMichel Lespinasse { 519d3737187SMichel Lespinasse /* All rb_subtree_gap values must be consistent prior to insertion */ 520d3737187SMichel Lespinasse validate_mm_rb(root, NULL); 521d3737187SMichel Lespinasse 522d3737187SMichel Lespinasse rb_insert_augmented(&vma->vm_rb, root, &vma_gap_callbacks); 523d3737187SMichel Lespinasse } 524d3737187SMichel Lespinasse 525d3737187SMichel Lespinasse static void vma_rb_erase(struct vm_area_struct *vma, struct rb_root *root) 526d3737187SMichel Lespinasse { 527d3737187SMichel Lespinasse /* 528d3737187SMichel Lespinasse * All rb_subtree_gap values must be consistent prior to erase, 529d3737187SMichel Lespinasse * with the possible exception of the vma being erased. 530d3737187SMichel Lespinasse */ 531d3737187SMichel Lespinasse validate_mm_rb(root, vma); 532d3737187SMichel Lespinasse 533d3737187SMichel Lespinasse /* 534d3737187SMichel Lespinasse * Note rb_erase_augmented is a fairly large inline function, 535d3737187SMichel Lespinasse * so make sure we instantiate it only once with our desired 536d3737187SMichel Lespinasse * augmented rbtree callbacks. 537d3737187SMichel Lespinasse */ 538d3737187SMichel Lespinasse rb_erase_augmented(&vma->vm_rb, root, &vma_gap_callbacks); 539d3737187SMichel Lespinasse } 540d3737187SMichel Lespinasse 541bf181b9fSMichel Lespinasse /* 542bf181b9fSMichel Lespinasse * vma has some anon_vma assigned, and is already inserted on that 543bf181b9fSMichel Lespinasse * anon_vma's interval trees. 544bf181b9fSMichel Lespinasse * 545bf181b9fSMichel Lespinasse * Before updating the vma's vm_start / vm_end / vm_pgoff fields, the 546bf181b9fSMichel Lespinasse * vma must be removed from the anon_vma's interval trees using 547bf181b9fSMichel Lespinasse * anon_vma_interval_tree_pre_update_vma(). 548bf181b9fSMichel Lespinasse * 549bf181b9fSMichel Lespinasse * After the update, the vma will be reinserted using 550bf181b9fSMichel Lespinasse * anon_vma_interval_tree_post_update_vma(). 551bf181b9fSMichel Lespinasse * 552bf181b9fSMichel Lespinasse * The entire update must be protected by exclusive mmap_sem and by 553bf181b9fSMichel Lespinasse * the root anon_vma's mutex. 554bf181b9fSMichel Lespinasse */ 555bf181b9fSMichel Lespinasse static inline void 556bf181b9fSMichel Lespinasse anon_vma_interval_tree_pre_update_vma(struct vm_area_struct *vma) 557bf181b9fSMichel Lespinasse { 558bf181b9fSMichel Lespinasse struct anon_vma_chain *avc; 559bf181b9fSMichel Lespinasse 560bf181b9fSMichel Lespinasse list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) 561bf181b9fSMichel Lespinasse anon_vma_interval_tree_remove(avc, &avc->anon_vma->rb_root); 562bf181b9fSMichel Lespinasse } 563bf181b9fSMichel Lespinasse 564bf181b9fSMichel Lespinasse static inline void 565bf181b9fSMichel Lespinasse anon_vma_interval_tree_post_update_vma(struct vm_area_struct *vma) 566bf181b9fSMichel Lespinasse { 567bf181b9fSMichel Lespinasse struct anon_vma_chain *avc; 568bf181b9fSMichel Lespinasse 569bf181b9fSMichel Lespinasse list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) 570bf181b9fSMichel Lespinasse anon_vma_interval_tree_insert(avc, &avc->anon_vma->rb_root); 571bf181b9fSMichel Lespinasse } 572bf181b9fSMichel Lespinasse 5736597d783SHugh Dickins static int find_vma_links(struct mm_struct *mm, unsigned long addr, 5746597d783SHugh Dickins unsigned long end, struct vm_area_struct **pprev, 5756597d783SHugh Dickins struct rb_node ***rb_link, struct rb_node **rb_parent) 5761da177e4SLinus Torvalds { 5771da177e4SLinus Torvalds struct rb_node **__rb_link, *__rb_parent, *rb_prev; 5781da177e4SLinus Torvalds 5791da177e4SLinus Torvalds __rb_link = &mm->mm_rb.rb_node; 5801da177e4SLinus Torvalds rb_prev = __rb_parent = NULL; 5811da177e4SLinus Torvalds 5821da177e4SLinus Torvalds while (*__rb_link) { 5831da177e4SLinus Torvalds struct vm_area_struct *vma_tmp; 5841da177e4SLinus Torvalds 5851da177e4SLinus Torvalds __rb_parent = *__rb_link; 5861da177e4SLinus Torvalds vma_tmp = rb_entry(__rb_parent, struct vm_area_struct, vm_rb); 5871da177e4SLinus Torvalds 5881da177e4SLinus Torvalds if (vma_tmp->vm_end > addr) { 5896597d783SHugh Dickins /* Fail if an existing vma overlaps the area */ 5906597d783SHugh Dickins if (vma_tmp->vm_start < end) 5916597d783SHugh Dickins return -ENOMEM; 5921da177e4SLinus Torvalds __rb_link = &__rb_parent->rb_left; 5931da177e4SLinus Torvalds } else { 5941da177e4SLinus Torvalds rb_prev = __rb_parent; 5951da177e4SLinus Torvalds __rb_link = &__rb_parent->rb_right; 5961da177e4SLinus Torvalds } 5971da177e4SLinus Torvalds } 5981da177e4SLinus Torvalds 5991da177e4SLinus Torvalds *pprev = NULL; 6001da177e4SLinus Torvalds if (rb_prev) 6011da177e4SLinus Torvalds *pprev = rb_entry(rb_prev, struct vm_area_struct, vm_rb); 6021da177e4SLinus Torvalds *rb_link = __rb_link; 6031da177e4SLinus Torvalds *rb_parent = __rb_parent; 6046597d783SHugh Dickins return 0; 6051da177e4SLinus Torvalds } 6061da177e4SLinus Torvalds 607e8420a8eSCyril Hrubis static unsigned long count_vma_pages_range(struct mm_struct *mm, 608e8420a8eSCyril Hrubis unsigned long addr, unsigned long end) 609e8420a8eSCyril Hrubis { 610e8420a8eSCyril Hrubis unsigned long nr_pages = 0; 611e8420a8eSCyril Hrubis struct vm_area_struct *vma; 612e8420a8eSCyril Hrubis 613e8420a8eSCyril Hrubis /* Find first overlaping mapping */ 614e8420a8eSCyril Hrubis vma = find_vma_intersection(mm, addr, end); 615e8420a8eSCyril Hrubis if (!vma) 616e8420a8eSCyril Hrubis return 0; 617e8420a8eSCyril Hrubis 618e8420a8eSCyril Hrubis nr_pages = (min(end, vma->vm_end) - 619e8420a8eSCyril Hrubis max(addr, vma->vm_start)) >> PAGE_SHIFT; 620e8420a8eSCyril Hrubis 621e8420a8eSCyril Hrubis /* Iterate over the rest of the overlaps */ 622e8420a8eSCyril Hrubis for (vma = vma->vm_next; vma; vma = vma->vm_next) { 623e8420a8eSCyril Hrubis unsigned long overlap_len; 624e8420a8eSCyril Hrubis 625e8420a8eSCyril Hrubis if (vma->vm_start > end) 626e8420a8eSCyril Hrubis break; 627e8420a8eSCyril Hrubis 628e8420a8eSCyril Hrubis overlap_len = min(end, vma->vm_end) - vma->vm_start; 629e8420a8eSCyril Hrubis nr_pages += overlap_len >> PAGE_SHIFT; 630e8420a8eSCyril Hrubis } 631e8420a8eSCyril Hrubis 632e8420a8eSCyril Hrubis return nr_pages; 633e8420a8eSCyril Hrubis } 634e8420a8eSCyril Hrubis 6351da177e4SLinus Torvalds void __vma_link_rb(struct mm_struct *mm, struct vm_area_struct *vma, 6361da177e4SLinus Torvalds struct rb_node **rb_link, struct rb_node *rb_parent) 6371da177e4SLinus Torvalds { 638d3737187SMichel Lespinasse /* Update tracking information for the gap following the new vma. */ 639d3737187SMichel Lespinasse if (vma->vm_next) 640d3737187SMichel Lespinasse vma_gap_update(vma->vm_next); 641d3737187SMichel Lespinasse else 642d3737187SMichel Lespinasse mm->highest_vm_end = vma->vm_end; 643d3737187SMichel Lespinasse 644d3737187SMichel Lespinasse /* 645d3737187SMichel Lespinasse * vma->vm_prev wasn't known when we followed the rbtree to find the 646d3737187SMichel Lespinasse * correct insertion point for that vma. As a result, we could not 647d3737187SMichel Lespinasse * update the vma vm_rb parents rb_subtree_gap values on the way down. 648d3737187SMichel Lespinasse * So, we first insert the vma with a zero rb_subtree_gap value 649d3737187SMichel Lespinasse * (to be consistent with what we did on the way down), and then 650d3737187SMichel Lespinasse * immediately update the gap to the correct value. Finally we 651d3737187SMichel Lespinasse * rebalance the rbtree after all augmented values have been set. 652d3737187SMichel Lespinasse */ 6531da177e4SLinus Torvalds rb_link_node(&vma->vm_rb, rb_parent, rb_link); 654d3737187SMichel Lespinasse vma->rb_subtree_gap = 0; 655d3737187SMichel Lespinasse vma_gap_update(vma); 656d3737187SMichel Lespinasse vma_rb_insert(vma, &mm->mm_rb); 6571da177e4SLinus Torvalds } 6581da177e4SLinus Torvalds 659cb8f488cSDenys Vlasenko static void __vma_link_file(struct vm_area_struct *vma) 6601da177e4SLinus Torvalds { 6611da177e4SLinus Torvalds struct file *file; 6621da177e4SLinus Torvalds 6631da177e4SLinus Torvalds file = vma->vm_file; 6641da177e4SLinus Torvalds if (file) { 6651da177e4SLinus Torvalds struct address_space *mapping = file->f_mapping; 6661da177e4SLinus Torvalds 6671da177e4SLinus Torvalds if (vma->vm_flags & VM_DENYWRITE) 668496ad9aaSAl Viro atomic_dec(&file_inode(file)->i_writecount); 6691da177e4SLinus Torvalds if (vma->vm_flags & VM_SHARED) 6704bb5f5d9SDavid Herrmann atomic_inc(&mapping->i_mmap_writable); 6711da177e4SLinus Torvalds 6721da177e4SLinus Torvalds flush_dcache_mmap_lock(mapping); 6736b2dbba8SMichel Lespinasse vma_interval_tree_insert(vma, &mapping->i_mmap); 6741da177e4SLinus Torvalds flush_dcache_mmap_unlock(mapping); 6751da177e4SLinus Torvalds } 6761da177e4SLinus Torvalds } 6771da177e4SLinus Torvalds 6781da177e4SLinus Torvalds static void 6791da177e4SLinus Torvalds __vma_link(struct mm_struct *mm, struct vm_area_struct *vma, 6801da177e4SLinus Torvalds struct vm_area_struct *prev, struct rb_node **rb_link, 6811da177e4SLinus Torvalds struct rb_node *rb_parent) 6821da177e4SLinus Torvalds { 6831da177e4SLinus Torvalds __vma_link_list(mm, vma, prev, rb_parent); 6841da177e4SLinus Torvalds __vma_link_rb(mm, vma, rb_link, rb_parent); 6851da177e4SLinus Torvalds } 6861da177e4SLinus Torvalds 6871da177e4SLinus Torvalds static void vma_link(struct mm_struct *mm, struct vm_area_struct *vma, 6881da177e4SLinus Torvalds struct vm_area_struct *prev, struct rb_node **rb_link, 6891da177e4SLinus Torvalds struct rb_node *rb_parent) 6901da177e4SLinus Torvalds { 6911da177e4SLinus Torvalds struct address_space *mapping = NULL; 6921da177e4SLinus Torvalds 69364ac4940SHuang Shijie if (vma->vm_file) { 6941da177e4SLinus Torvalds mapping = vma->vm_file->f_mapping; 69583cde9e8SDavidlohr Bueso i_mmap_lock_write(mapping); 69664ac4940SHuang Shijie } 6971da177e4SLinus Torvalds 6981da177e4SLinus Torvalds __vma_link(mm, vma, prev, rb_link, rb_parent); 6991da177e4SLinus Torvalds __vma_link_file(vma); 7001da177e4SLinus Torvalds 7011da177e4SLinus Torvalds if (mapping) 70283cde9e8SDavidlohr Bueso i_mmap_unlock_write(mapping); 7031da177e4SLinus Torvalds 7041da177e4SLinus Torvalds mm->map_count++; 7051da177e4SLinus Torvalds validate_mm(mm); 7061da177e4SLinus Torvalds } 7071da177e4SLinus Torvalds 7081da177e4SLinus Torvalds /* 70988f6b4c3SKautuk Consul * Helper for vma_adjust() in the split_vma insert case: insert a vma into the 7106b2dbba8SMichel Lespinasse * mm's list and rbtree. It has already been inserted into the interval tree. 7111da177e4SLinus Torvalds */ 71248aae425SZhenwenXu static void __insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma) 7131da177e4SLinus Torvalds { 7146597d783SHugh Dickins struct vm_area_struct *prev; 7151da177e4SLinus Torvalds struct rb_node **rb_link, *rb_parent; 7161da177e4SLinus Torvalds 7176597d783SHugh Dickins if (find_vma_links(mm, vma->vm_start, vma->vm_end, 7186597d783SHugh Dickins &prev, &rb_link, &rb_parent)) 7196597d783SHugh Dickins BUG(); 7201da177e4SLinus Torvalds __vma_link(mm, vma, prev, rb_link, rb_parent); 7211da177e4SLinus Torvalds mm->map_count++; 7221da177e4SLinus Torvalds } 7231da177e4SLinus Torvalds 7241da177e4SLinus Torvalds static inline void 7251da177e4SLinus Torvalds __vma_unlink(struct mm_struct *mm, struct vm_area_struct *vma, 7261da177e4SLinus Torvalds struct vm_area_struct *prev) 7271da177e4SLinus Torvalds { 728d3737187SMichel Lespinasse struct vm_area_struct *next; 729297c5eeeSLinus Torvalds 730d3737187SMichel Lespinasse vma_rb_erase(vma, &mm->mm_rb); 731d3737187SMichel Lespinasse prev->vm_next = next = vma->vm_next; 732297c5eeeSLinus Torvalds if (next) 733297c5eeeSLinus Torvalds next->vm_prev = prev; 734615d6e87SDavidlohr Bueso 735615d6e87SDavidlohr Bueso /* Kill the cache */ 736615d6e87SDavidlohr Bueso vmacache_invalidate(mm); 7371da177e4SLinus Torvalds } 7381da177e4SLinus Torvalds 7391da177e4SLinus Torvalds /* 7401da177e4SLinus Torvalds * We cannot adjust vm_start, vm_end, vm_pgoff fields of a vma that 7411da177e4SLinus Torvalds * is already present in an i_mmap tree without adjusting the tree. 7421da177e4SLinus Torvalds * The following helper function should be used when such adjustments 7431da177e4SLinus Torvalds * are necessary. The "insert" vma (if any) is to be inserted 7441da177e4SLinus Torvalds * before we drop the necessary locks. 7451da177e4SLinus Torvalds */ 7465beb4930SRik van Riel int vma_adjust(struct vm_area_struct *vma, unsigned long start, 7471da177e4SLinus Torvalds unsigned long end, pgoff_t pgoff, struct vm_area_struct *insert) 7481da177e4SLinus Torvalds { 7491da177e4SLinus Torvalds struct mm_struct *mm = vma->vm_mm; 7501da177e4SLinus Torvalds struct vm_area_struct *next = vma->vm_next; 7511da177e4SLinus Torvalds struct vm_area_struct *importer = NULL; 7521da177e4SLinus Torvalds struct address_space *mapping = NULL; 7536b2dbba8SMichel Lespinasse struct rb_root *root = NULL; 754012f1800SRik van Riel struct anon_vma *anon_vma = NULL; 7551da177e4SLinus Torvalds struct file *file = vma->vm_file; 756d3737187SMichel Lespinasse bool start_changed = false, end_changed = false; 7571da177e4SLinus Torvalds long adjust_next = 0; 7581da177e4SLinus Torvalds int remove_next = 0; 7591da177e4SLinus Torvalds 7601da177e4SLinus Torvalds if (next && !insert) { 761287d97acSLinus Torvalds struct vm_area_struct *exporter = NULL; 762287d97acSLinus Torvalds 7631da177e4SLinus Torvalds if (end >= next->vm_end) { 7641da177e4SLinus Torvalds /* 7651da177e4SLinus Torvalds * vma expands, overlapping all the next, and 7661da177e4SLinus Torvalds * perhaps the one after too (mprotect case 6). 7671da177e4SLinus Torvalds */ 7681da177e4SLinus Torvalds again: remove_next = 1 + (end > next->vm_end); 7691da177e4SLinus Torvalds end = next->vm_end; 770287d97acSLinus Torvalds exporter = next; 7711da177e4SLinus Torvalds importer = vma; 7721da177e4SLinus Torvalds } else if (end > next->vm_start) { 7731da177e4SLinus Torvalds /* 7741da177e4SLinus Torvalds * vma expands, overlapping part of the next: 7751da177e4SLinus Torvalds * mprotect case 5 shifting the boundary up. 7761da177e4SLinus Torvalds */ 7771da177e4SLinus Torvalds adjust_next = (end - next->vm_start) >> PAGE_SHIFT; 778287d97acSLinus Torvalds exporter = next; 7791da177e4SLinus Torvalds importer = vma; 7801da177e4SLinus Torvalds } else if (end < vma->vm_end) { 7811da177e4SLinus Torvalds /* 7821da177e4SLinus Torvalds * vma shrinks, and !insert tells it's not 7831da177e4SLinus Torvalds * split_vma inserting another: so it must be 7841da177e4SLinus Torvalds * mprotect case 4 shifting the boundary down. 7851da177e4SLinus Torvalds */ 7861da177e4SLinus Torvalds adjust_next = -((vma->vm_end - end) >> PAGE_SHIFT); 787287d97acSLinus Torvalds exporter = vma; 7881da177e4SLinus Torvalds importer = next; 7891da177e4SLinus Torvalds } 7901da177e4SLinus Torvalds 7915beb4930SRik van Riel /* 7925beb4930SRik van Riel * Easily overlooked: when mprotect shifts the boundary, 7935beb4930SRik van Riel * make sure the expanding vma has anon_vma set if the 7945beb4930SRik van Riel * shrinking vma had, to cover any anon pages imported. 7955beb4930SRik van Riel */ 796287d97acSLinus Torvalds if (exporter && exporter->anon_vma && !importer->anon_vma) { 797c4ea95d7SDaniel Forrest int error; 798c4ea95d7SDaniel Forrest 799287d97acSLinus Torvalds importer->anon_vma = exporter->anon_vma; 800b800c91aSKonstantin Khlebnikov error = anon_vma_clone(importer, exporter); 8013fe89b3eSLeon Yu if (error) 802b800c91aSKonstantin Khlebnikov return error; 803b800c91aSKonstantin Khlebnikov } 8045beb4930SRik van Riel } 8055beb4930SRik van Riel 8061da177e4SLinus Torvalds if (file) { 8071da177e4SLinus Torvalds mapping = file->f_mapping; 8081da177e4SLinus Torvalds root = &mapping->i_mmap; 809cbc91f71SSrikar Dronamraju uprobe_munmap(vma, vma->vm_start, vma->vm_end); 810682968e0SSrikar Dronamraju 811682968e0SSrikar Dronamraju if (adjust_next) 81227ba0644SKirill A. Shutemov uprobe_munmap(next, next->vm_start, next->vm_end); 813682968e0SSrikar Dronamraju 81483cde9e8SDavidlohr Bueso i_mmap_lock_write(mapping); 8151da177e4SLinus Torvalds if (insert) { 8161da177e4SLinus Torvalds /* 8176b2dbba8SMichel Lespinasse * Put into interval tree now, so instantiated pages 8181da177e4SLinus Torvalds * are visible to arm/parisc __flush_dcache_page 8191da177e4SLinus Torvalds * throughout; but we cannot insert into address 8201da177e4SLinus Torvalds * space until vma start or end is updated. 8211da177e4SLinus Torvalds */ 8221da177e4SLinus Torvalds __vma_link_file(insert); 8231da177e4SLinus Torvalds } 8241da177e4SLinus Torvalds } 8251da177e4SLinus Torvalds 82694fcc585SAndrea Arcangeli vma_adjust_trans_huge(vma, start, end, adjust_next); 82794fcc585SAndrea Arcangeli 828012f1800SRik van Riel anon_vma = vma->anon_vma; 829bf181b9fSMichel Lespinasse if (!anon_vma && adjust_next) 830bf181b9fSMichel Lespinasse anon_vma = next->anon_vma; 831bf181b9fSMichel Lespinasse if (anon_vma) { 83281d1b09cSSasha Levin VM_BUG_ON_VMA(adjust_next && next->anon_vma && 83381d1b09cSSasha Levin anon_vma != next->anon_vma, next); 8344fc3f1d6SIngo Molnar anon_vma_lock_write(anon_vma); 835bf181b9fSMichel Lespinasse anon_vma_interval_tree_pre_update_vma(vma); 836bf181b9fSMichel Lespinasse if (adjust_next) 837bf181b9fSMichel Lespinasse anon_vma_interval_tree_pre_update_vma(next); 838bf181b9fSMichel Lespinasse } 839012f1800SRik van Riel 8401da177e4SLinus Torvalds if (root) { 8411da177e4SLinus Torvalds flush_dcache_mmap_lock(mapping); 8426b2dbba8SMichel Lespinasse vma_interval_tree_remove(vma, root); 8431da177e4SLinus Torvalds if (adjust_next) 8446b2dbba8SMichel Lespinasse vma_interval_tree_remove(next, root); 8451da177e4SLinus Torvalds } 8461da177e4SLinus Torvalds 847d3737187SMichel Lespinasse if (start != vma->vm_start) { 8481da177e4SLinus Torvalds vma->vm_start = start; 849d3737187SMichel Lespinasse start_changed = true; 850d3737187SMichel Lespinasse } 851d3737187SMichel Lespinasse if (end != vma->vm_end) { 8521da177e4SLinus Torvalds vma->vm_end = end; 853d3737187SMichel Lespinasse end_changed = true; 854d3737187SMichel Lespinasse } 8551da177e4SLinus Torvalds vma->vm_pgoff = pgoff; 8561da177e4SLinus Torvalds if (adjust_next) { 8571da177e4SLinus Torvalds next->vm_start += adjust_next << PAGE_SHIFT; 8581da177e4SLinus Torvalds next->vm_pgoff += adjust_next; 8591da177e4SLinus Torvalds } 8601da177e4SLinus Torvalds 8611da177e4SLinus Torvalds if (root) { 8621da177e4SLinus Torvalds if (adjust_next) 8636b2dbba8SMichel Lespinasse vma_interval_tree_insert(next, root); 8646b2dbba8SMichel Lespinasse vma_interval_tree_insert(vma, root); 8651da177e4SLinus Torvalds flush_dcache_mmap_unlock(mapping); 8661da177e4SLinus Torvalds } 8671da177e4SLinus Torvalds 8681da177e4SLinus Torvalds if (remove_next) { 8691da177e4SLinus Torvalds /* 8701da177e4SLinus Torvalds * vma_merge has merged next into vma, and needs 8711da177e4SLinus Torvalds * us to remove next before dropping the locks. 8721da177e4SLinus Torvalds */ 8731da177e4SLinus Torvalds __vma_unlink(mm, next, vma); 8741da177e4SLinus Torvalds if (file) 8751da177e4SLinus Torvalds __remove_shared_vm_struct(next, file, mapping); 8761da177e4SLinus Torvalds } else if (insert) { 8771da177e4SLinus Torvalds /* 8781da177e4SLinus Torvalds * split_vma has split insert from vma, and needs 8791da177e4SLinus Torvalds * us to insert it before dropping the locks 8801da177e4SLinus Torvalds * (it may either follow vma or precede it). 8811da177e4SLinus Torvalds */ 8821da177e4SLinus Torvalds __insert_vm_struct(mm, insert); 883d3737187SMichel Lespinasse } else { 884d3737187SMichel Lespinasse if (start_changed) 885d3737187SMichel Lespinasse vma_gap_update(vma); 886d3737187SMichel Lespinasse if (end_changed) { 887d3737187SMichel Lespinasse if (!next) 888d3737187SMichel Lespinasse mm->highest_vm_end = end; 889d3737187SMichel Lespinasse else if (!adjust_next) 890d3737187SMichel Lespinasse vma_gap_update(next); 891d3737187SMichel Lespinasse } 8921da177e4SLinus Torvalds } 8931da177e4SLinus Torvalds 894bf181b9fSMichel Lespinasse if (anon_vma) { 895bf181b9fSMichel Lespinasse anon_vma_interval_tree_post_update_vma(vma); 896bf181b9fSMichel Lespinasse if (adjust_next) 897bf181b9fSMichel Lespinasse anon_vma_interval_tree_post_update_vma(next); 89808b52706SKonstantin Khlebnikov anon_vma_unlock_write(anon_vma); 899bf181b9fSMichel Lespinasse } 9001da177e4SLinus Torvalds if (mapping) 90183cde9e8SDavidlohr Bueso i_mmap_unlock_write(mapping); 9021da177e4SLinus Torvalds 9032b144498SSrikar Dronamraju if (root) { 9047b2d81d4SIngo Molnar uprobe_mmap(vma); 9052b144498SSrikar Dronamraju 9062b144498SSrikar Dronamraju if (adjust_next) 9077b2d81d4SIngo Molnar uprobe_mmap(next); 9082b144498SSrikar Dronamraju } 9092b144498SSrikar Dronamraju 9101da177e4SLinus Torvalds if (remove_next) { 911925d1c40SMatt Helsley if (file) { 912cbc91f71SSrikar Dronamraju uprobe_munmap(next, next->vm_start, next->vm_end); 9131da177e4SLinus Torvalds fput(file); 914925d1c40SMatt Helsley } 9155beb4930SRik van Riel if (next->anon_vma) 9165beb4930SRik van Riel anon_vma_merge(vma, next); 9171da177e4SLinus Torvalds mm->map_count--; 9183964acd0SOleg Nesterov mpol_put(vma_policy(next)); 9191da177e4SLinus Torvalds kmem_cache_free(vm_area_cachep, next); 9201da177e4SLinus Torvalds /* 9211da177e4SLinus Torvalds * In mprotect's case 6 (see comments on vma_merge), 9221da177e4SLinus Torvalds * we must remove another next too. It would clutter 9231da177e4SLinus Torvalds * up the code too much to do both in one go. 9241da177e4SLinus Torvalds */ 9251da177e4SLinus Torvalds next = vma->vm_next; 926d3737187SMichel Lespinasse if (remove_next == 2) 9271da177e4SLinus Torvalds goto again; 928d3737187SMichel Lespinasse else if (next) 929d3737187SMichel Lespinasse vma_gap_update(next); 930d3737187SMichel Lespinasse else 931d3737187SMichel Lespinasse mm->highest_vm_end = end; 9321da177e4SLinus Torvalds } 9332b144498SSrikar Dronamraju if (insert && file) 9347b2d81d4SIngo Molnar uprobe_mmap(insert); 9351da177e4SLinus Torvalds 9361da177e4SLinus Torvalds validate_mm(mm); 9375beb4930SRik van Riel 9385beb4930SRik van Riel return 0; 9391da177e4SLinus Torvalds } 9401da177e4SLinus Torvalds 9411da177e4SLinus Torvalds /* 9421da177e4SLinus Torvalds * If the vma has a ->close operation then the driver probably needs to release 9431da177e4SLinus Torvalds * per-vma resources, so we don't attempt to merge those. 9441da177e4SLinus Torvalds */ 9451da177e4SLinus Torvalds static inline int is_mergeable_vma(struct vm_area_struct *vma, 94619a809afSAndrea Arcangeli struct file *file, unsigned long vm_flags, 94719a809afSAndrea Arcangeli struct vm_userfaultfd_ctx vm_userfaultfd_ctx) 9481da177e4SLinus Torvalds { 94934228d47SCyrill Gorcunov /* 95034228d47SCyrill Gorcunov * VM_SOFTDIRTY should not prevent from VMA merging, if we 95134228d47SCyrill Gorcunov * match the flags but dirty bit -- the caller should mark 95234228d47SCyrill Gorcunov * merged VMA as dirty. If dirty bit won't be excluded from 95334228d47SCyrill Gorcunov * comparison, we increase pressue on the memory system forcing 95434228d47SCyrill Gorcunov * the kernel to generate new VMAs when old one could be 95534228d47SCyrill Gorcunov * extended instead. 95634228d47SCyrill Gorcunov */ 95734228d47SCyrill Gorcunov if ((vma->vm_flags ^ vm_flags) & ~VM_SOFTDIRTY) 9581da177e4SLinus Torvalds return 0; 9591da177e4SLinus Torvalds if (vma->vm_file != file) 9601da177e4SLinus Torvalds return 0; 9611da177e4SLinus Torvalds if (vma->vm_ops && vma->vm_ops->close) 9621da177e4SLinus Torvalds return 0; 96319a809afSAndrea Arcangeli if (!is_mergeable_vm_userfaultfd_ctx(vma, vm_userfaultfd_ctx)) 96419a809afSAndrea Arcangeli return 0; 9651da177e4SLinus Torvalds return 1; 9661da177e4SLinus Torvalds } 9671da177e4SLinus Torvalds 9681da177e4SLinus Torvalds static inline int is_mergeable_anon_vma(struct anon_vma *anon_vma1, 969965f55deSShaohua Li struct anon_vma *anon_vma2, 970965f55deSShaohua Li struct vm_area_struct *vma) 9711da177e4SLinus Torvalds { 972965f55deSShaohua Li /* 973965f55deSShaohua Li * The list_is_singular() test is to avoid merging VMA cloned from 974965f55deSShaohua Li * parents. This can improve scalability caused by anon_vma lock. 975965f55deSShaohua Li */ 976965f55deSShaohua Li if ((!anon_vma1 || !anon_vma2) && (!vma || 977965f55deSShaohua Li list_is_singular(&vma->anon_vma_chain))) 978965f55deSShaohua Li return 1; 979965f55deSShaohua Li return anon_vma1 == anon_vma2; 9801da177e4SLinus Torvalds } 9811da177e4SLinus Torvalds 9821da177e4SLinus Torvalds /* 9831da177e4SLinus Torvalds * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff) 9841da177e4SLinus Torvalds * in front of (at a lower virtual address and file offset than) the vma. 9851da177e4SLinus Torvalds * 9861da177e4SLinus Torvalds * We cannot merge two vmas if they have differently assigned (non-NULL) 9871da177e4SLinus Torvalds * anon_vmas, nor if same anon_vma is assigned but offsets incompatible. 9881da177e4SLinus Torvalds * 9891da177e4SLinus Torvalds * We don't check here for the merged mmap wrapping around the end of pagecache 9901da177e4SLinus Torvalds * indices (16TB on ia32) because do_mmap_pgoff() does not permit mmap's which 9911da177e4SLinus Torvalds * wrap, nor mmaps which cover the final page at index -1UL. 9921da177e4SLinus Torvalds */ 9931da177e4SLinus Torvalds static int 9941da177e4SLinus Torvalds can_vma_merge_before(struct vm_area_struct *vma, unsigned long vm_flags, 99519a809afSAndrea Arcangeli struct anon_vma *anon_vma, struct file *file, 99619a809afSAndrea Arcangeli pgoff_t vm_pgoff, 99719a809afSAndrea Arcangeli struct vm_userfaultfd_ctx vm_userfaultfd_ctx) 9981da177e4SLinus Torvalds { 99919a809afSAndrea Arcangeli if (is_mergeable_vma(vma, file, vm_flags, vm_userfaultfd_ctx) && 1000965f55deSShaohua Li is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) { 10011da177e4SLinus Torvalds if (vma->vm_pgoff == vm_pgoff) 10021da177e4SLinus Torvalds return 1; 10031da177e4SLinus Torvalds } 10041da177e4SLinus Torvalds return 0; 10051da177e4SLinus Torvalds } 10061da177e4SLinus Torvalds 10071da177e4SLinus Torvalds /* 10081da177e4SLinus Torvalds * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff) 10091da177e4SLinus Torvalds * beyond (at a higher virtual address and file offset than) the vma. 10101da177e4SLinus Torvalds * 10111da177e4SLinus Torvalds * We cannot merge two vmas if they have differently assigned (non-NULL) 10121da177e4SLinus Torvalds * anon_vmas, nor if same anon_vma is assigned but offsets incompatible. 10131da177e4SLinus Torvalds */ 10141da177e4SLinus Torvalds static int 10151da177e4SLinus Torvalds can_vma_merge_after(struct vm_area_struct *vma, unsigned long vm_flags, 101619a809afSAndrea Arcangeli struct anon_vma *anon_vma, struct file *file, 101719a809afSAndrea Arcangeli pgoff_t vm_pgoff, 101819a809afSAndrea Arcangeli struct vm_userfaultfd_ctx vm_userfaultfd_ctx) 10191da177e4SLinus Torvalds { 102019a809afSAndrea Arcangeli if (is_mergeable_vma(vma, file, vm_flags, vm_userfaultfd_ctx) && 1021965f55deSShaohua Li is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) { 10221da177e4SLinus Torvalds pgoff_t vm_pglen; 1023d6e93217SLibin vm_pglen = vma_pages(vma); 10241da177e4SLinus Torvalds if (vma->vm_pgoff + vm_pglen == vm_pgoff) 10251da177e4SLinus Torvalds return 1; 10261da177e4SLinus Torvalds } 10271da177e4SLinus Torvalds return 0; 10281da177e4SLinus Torvalds } 10291da177e4SLinus Torvalds 10301da177e4SLinus Torvalds /* 10311da177e4SLinus Torvalds * Given a mapping request (addr,end,vm_flags,file,pgoff), figure out 10321da177e4SLinus Torvalds * whether that can be merged with its predecessor or its successor. 10331da177e4SLinus Torvalds * Or both (it neatly fills a hole). 10341da177e4SLinus Torvalds * 10351da177e4SLinus Torvalds * In most cases - when called for mmap, brk or mremap - [addr,end) is 10361da177e4SLinus Torvalds * certain not to be mapped by the time vma_merge is called; but when 10371da177e4SLinus Torvalds * called for mprotect, it is certain to be already mapped (either at 10381da177e4SLinus Torvalds * an offset within prev, or at the start of next), and the flags of 10391da177e4SLinus Torvalds * this area are about to be changed to vm_flags - and the no-change 10401da177e4SLinus Torvalds * case has already been eliminated. 10411da177e4SLinus Torvalds * 10421da177e4SLinus Torvalds * The following mprotect cases have to be considered, where AAAA is 10431da177e4SLinus Torvalds * the area passed down from mprotect_fixup, never extending beyond one 10441da177e4SLinus Torvalds * vma, PPPPPP is the prev vma specified, and NNNNNN the next vma after: 10451da177e4SLinus Torvalds * 10461da177e4SLinus Torvalds * AAAA AAAA AAAA AAAA 10471da177e4SLinus Torvalds * PPPPPPNNNNNN PPPPPPNNNNNN PPPPPPNNNNNN PPPPNNNNXXXX 10481da177e4SLinus Torvalds * cannot merge might become might become might become 10491da177e4SLinus Torvalds * PPNNNNNNNNNN PPPPPPPPPPNN PPPPPPPPPPPP 6 or 10501da177e4SLinus Torvalds * mmap, brk or case 4 below case 5 below PPPPPPPPXXXX 7 or 10511da177e4SLinus Torvalds * mremap move: PPPPNNNNNNNN 8 10521da177e4SLinus Torvalds * AAAA 10531da177e4SLinus Torvalds * PPPP NNNN PPPPPPPPPPPP PPPPPPPPNNNN PPPPNNNNNNNN 10541da177e4SLinus Torvalds * might become case 1 below case 2 below case 3 below 10551da177e4SLinus Torvalds * 10561da177e4SLinus Torvalds * Odd one out? Case 8, because it extends NNNN but needs flags of XXXX: 10571da177e4SLinus Torvalds * mprotect_fixup updates vm_flags & vm_page_prot on successful return. 10581da177e4SLinus Torvalds */ 10591da177e4SLinus Torvalds struct vm_area_struct *vma_merge(struct mm_struct *mm, 10601da177e4SLinus Torvalds struct vm_area_struct *prev, unsigned long addr, 10611da177e4SLinus Torvalds unsigned long end, unsigned long vm_flags, 10621da177e4SLinus Torvalds struct anon_vma *anon_vma, struct file *file, 106319a809afSAndrea Arcangeli pgoff_t pgoff, struct mempolicy *policy, 106419a809afSAndrea Arcangeli struct vm_userfaultfd_ctx vm_userfaultfd_ctx) 10651da177e4SLinus Torvalds { 10661da177e4SLinus Torvalds pgoff_t pglen = (end - addr) >> PAGE_SHIFT; 10671da177e4SLinus Torvalds struct vm_area_struct *area, *next; 10685beb4930SRik van Riel int err; 10691da177e4SLinus Torvalds 10701da177e4SLinus Torvalds /* 10711da177e4SLinus Torvalds * We later require that vma->vm_flags == vm_flags, 10721da177e4SLinus Torvalds * so this tests vma->vm_flags & VM_SPECIAL, too. 10731da177e4SLinus Torvalds */ 10741da177e4SLinus Torvalds if (vm_flags & VM_SPECIAL) 10751da177e4SLinus Torvalds return NULL; 10761da177e4SLinus Torvalds 10771da177e4SLinus Torvalds if (prev) 10781da177e4SLinus Torvalds next = prev->vm_next; 10791da177e4SLinus Torvalds else 10801da177e4SLinus Torvalds next = mm->mmap; 10811da177e4SLinus Torvalds area = next; 10821da177e4SLinus Torvalds if (next && next->vm_end == end) /* cases 6, 7, 8 */ 10831da177e4SLinus Torvalds next = next->vm_next; 10841da177e4SLinus Torvalds 10851da177e4SLinus Torvalds /* 10861da177e4SLinus Torvalds * Can it merge with the predecessor? 10871da177e4SLinus Torvalds */ 10881da177e4SLinus Torvalds if (prev && prev->vm_end == addr && 10891da177e4SLinus Torvalds mpol_equal(vma_policy(prev), policy) && 10901da177e4SLinus Torvalds can_vma_merge_after(prev, vm_flags, 109119a809afSAndrea Arcangeli anon_vma, file, pgoff, 109219a809afSAndrea Arcangeli vm_userfaultfd_ctx)) { 10931da177e4SLinus Torvalds /* 10941da177e4SLinus Torvalds * OK, it can. Can we now merge in the successor as well? 10951da177e4SLinus Torvalds */ 10961da177e4SLinus Torvalds if (next && end == next->vm_start && 10971da177e4SLinus Torvalds mpol_equal(policy, vma_policy(next)) && 10981da177e4SLinus Torvalds can_vma_merge_before(next, vm_flags, 109919a809afSAndrea Arcangeli anon_vma, file, 110019a809afSAndrea Arcangeli pgoff+pglen, 110119a809afSAndrea Arcangeli vm_userfaultfd_ctx) && 11021da177e4SLinus Torvalds is_mergeable_anon_vma(prev->anon_vma, 1103965f55deSShaohua Li next->anon_vma, NULL)) { 11041da177e4SLinus Torvalds /* cases 1, 6 */ 11055beb4930SRik van Riel err = vma_adjust(prev, prev->vm_start, 11061da177e4SLinus Torvalds next->vm_end, prev->vm_pgoff, NULL); 11071da177e4SLinus Torvalds } else /* cases 2, 5, 7 */ 11085beb4930SRik van Riel err = vma_adjust(prev, prev->vm_start, 11091da177e4SLinus Torvalds end, prev->vm_pgoff, NULL); 11105beb4930SRik van Riel if (err) 11115beb4930SRik van Riel return NULL; 11126d50e60cSDavid Rientjes khugepaged_enter_vma_merge(prev, vm_flags); 11131da177e4SLinus Torvalds return prev; 11141da177e4SLinus Torvalds } 11151da177e4SLinus Torvalds 11161da177e4SLinus Torvalds /* 11171da177e4SLinus Torvalds * Can this new request be merged in front of next? 11181da177e4SLinus Torvalds */ 11191da177e4SLinus Torvalds if (next && end == next->vm_start && 11201da177e4SLinus Torvalds mpol_equal(policy, vma_policy(next)) && 11211da177e4SLinus Torvalds can_vma_merge_before(next, vm_flags, 112219a809afSAndrea Arcangeli anon_vma, file, pgoff+pglen, 112319a809afSAndrea Arcangeli vm_userfaultfd_ctx)) { 11241da177e4SLinus Torvalds if (prev && addr < prev->vm_end) /* case 4 */ 11255beb4930SRik van Riel err = vma_adjust(prev, prev->vm_start, 11261da177e4SLinus Torvalds addr, prev->vm_pgoff, NULL); 11271da177e4SLinus Torvalds else /* cases 3, 8 */ 11285beb4930SRik van Riel err = vma_adjust(area, addr, next->vm_end, 11291da177e4SLinus Torvalds next->vm_pgoff - pglen, NULL); 11305beb4930SRik van Riel if (err) 11315beb4930SRik van Riel return NULL; 11326d50e60cSDavid Rientjes khugepaged_enter_vma_merge(area, vm_flags); 11331da177e4SLinus Torvalds return area; 11341da177e4SLinus Torvalds } 11351da177e4SLinus Torvalds 11361da177e4SLinus Torvalds return NULL; 11371da177e4SLinus Torvalds } 11381da177e4SLinus Torvalds 11391da177e4SLinus Torvalds /* 1140d0e9fe17SLinus Torvalds * Rough compatbility check to quickly see if it's even worth looking 1141d0e9fe17SLinus Torvalds * at sharing an anon_vma. 1142d0e9fe17SLinus Torvalds * 1143d0e9fe17SLinus Torvalds * They need to have the same vm_file, and the flags can only differ 1144d0e9fe17SLinus Torvalds * in things that mprotect may change. 1145d0e9fe17SLinus Torvalds * 1146d0e9fe17SLinus Torvalds * NOTE! The fact that we share an anon_vma doesn't _have_ to mean that 1147d0e9fe17SLinus Torvalds * we can merge the two vma's. For example, we refuse to merge a vma if 1148d0e9fe17SLinus Torvalds * there is a vm_ops->close() function, because that indicates that the 1149d0e9fe17SLinus Torvalds * driver is doing some kind of reference counting. But that doesn't 1150d0e9fe17SLinus Torvalds * really matter for the anon_vma sharing case. 1151d0e9fe17SLinus Torvalds */ 1152d0e9fe17SLinus Torvalds static int anon_vma_compatible(struct vm_area_struct *a, struct vm_area_struct *b) 1153d0e9fe17SLinus Torvalds { 1154d0e9fe17SLinus Torvalds return a->vm_end == b->vm_start && 1155d0e9fe17SLinus Torvalds mpol_equal(vma_policy(a), vma_policy(b)) && 1156d0e9fe17SLinus Torvalds a->vm_file == b->vm_file && 115734228d47SCyrill Gorcunov !((a->vm_flags ^ b->vm_flags) & ~(VM_READ|VM_WRITE|VM_EXEC|VM_SOFTDIRTY)) && 1158d0e9fe17SLinus Torvalds b->vm_pgoff == a->vm_pgoff + ((b->vm_start - a->vm_start) >> PAGE_SHIFT); 1159d0e9fe17SLinus Torvalds } 1160d0e9fe17SLinus Torvalds 1161d0e9fe17SLinus Torvalds /* 1162d0e9fe17SLinus Torvalds * Do some basic sanity checking to see if we can re-use the anon_vma 1163d0e9fe17SLinus Torvalds * from 'old'. The 'a'/'b' vma's are in VM order - one of them will be 1164d0e9fe17SLinus Torvalds * the same as 'old', the other will be the new one that is trying 1165d0e9fe17SLinus Torvalds * to share the anon_vma. 1166d0e9fe17SLinus Torvalds * 1167d0e9fe17SLinus Torvalds * NOTE! This runs with mm_sem held for reading, so it is possible that 1168d0e9fe17SLinus Torvalds * the anon_vma of 'old' is concurrently in the process of being set up 1169d0e9fe17SLinus Torvalds * by another page fault trying to merge _that_. But that's ok: if it 1170d0e9fe17SLinus Torvalds * is being set up, that automatically means that it will be a singleton 1171d0e9fe17SLinus Torvalds * acceptable for merging, so we can do all of this optimistically. But 11724db0c3c2SJason Low * we do that READ_ONCE() to make sure that we never re-load the pointer. 1173d0e9fe17SLinus Torvalds * 1174d0e9fe17SLinus Torvalds * IOW: that the "list_is_singular()" test on the anon_vma_chain only 1175d0e9fe17SLinus Torvalds * matters for the 'stable anon_vma' case (ie the thing we want to avoid 1176d0e9fe17SLinus Torvalds * is to return an anon_vma that is "complex" due to having gone through 1177d0e9fe17SLinus Torvalds * a fork). 1178d0e9fe17SLinus Torvalds * 1179d0e9fe17SLinus Torvalds * We also make sure that the two vma's are compatible (adjacent, 1180d0e9fe17SLinus Torvalds * and with the same memory policies). That's all stable, even with just 1181d0e9fe17SLinus Torvalds * a read lock on the mm_sem. 1182d0e9fe17SLinus Torvalds */ 1183d0e9fe17SLinus Torvalds static struct anon_vma *reusable_anon_vma(struct vm_area_struct *old, struct vm_area_struct *a, struct vm_area_struct *b) 1184d0e9fe17SLinus Torvalds { 1185d0e9fe17SLinus Torvalds if (anon_vma_compatible(a, b)) { 11864db0c3c2SJason Low struct anon_vma *anon_vma = READ_ONCE(old->anon_vma); 1187d0e9fe17SLinus Torvalds 1188d0e9fe17SLinus Torvalds if (anon_vma && list_is_singular(&old->anon_vma_chain)) 1189d0e9fe17SLinus Torvalds return anon_vma; 1190d0e9fe17SLinus Torvalds } 1191d0e9fe17SLinus Torvalds return NULL; 1192d0e9fe17SLinus Torvalds } 1193d0e9fe17SLinus Torvalds 1194d0e9fe17SLinus Torvalds /* 11951da177e4SLinus Torvalds * find_mergeable_anon_vma is used by anon_vma_prepare, to check 11961da177e4SLinus Torvalds * neighbouring vmas for a suitable anon_vma, before it goes off 11971da177e4SLinus Torvalds * to allocate a new anon_vma. It checks because a repetitive 11981da177e4SLinus Torvalds * sequence of mprotects and faults may otherwise lead to distinct 11991da177e4SLinus Torvalds * anon_vmas being allocated, preventing vma merge in subsequent 12001da177e4SLinus Torvalds * mprotect. 12011da177e4SLinus Torvalds */ 12021da177e4SLinus Torvalds struct anon_vma *find_mergeable_anon_vma(struct vm_area_struct *vma) 12031da177e4SLinus Torvalds { 1204d0e9fe17SLinus Torvalds struct anon_vma *anon_vma; 12051da177e4SLinus Torvalds struct vm_area_struct *near; 12061da177e4SLinus Torvalds 12071da177e4SLinus Torvalds near = vma->vm_next; 12081da177e4SLinus Torvalds if (!near) 12091da177e4SLinus Torvalds goto try_prev; 12101da177e4SLinus Torvalds 1211d0e9fe17SLinus Torvalds anon_vma = reusable_anon_vma(near, vma, near); 1212d0e9fe17SLinus Torvalds if (anon_vma) 1213d0e9fe17SLinus Torvalds return anon_vma; 12141da177e4SLinus Torvalds try_prev: 12159be34c9dSLinus Torvalds near = vma->vm_prev; 12161da177e4SLinus Torvalds if (!near) 12171da177e4SLinus Torvalds goto none; 12181da177e4SLinus Torvalds 1219d0e9fe17SLinus Torvalds anon_vma = reusable_anon_vma(near, near, vma); 1220d0e9fe17SLinus Torvalds if (anon_vma) 1221d0e9fe17SLinus Torvalds return anon_vma; 12221da177e4SLinus Torvalds none: 12231da177e4SLinus Torvalds /* 12241da177e4SLinus Torvalds * There's no absolute need to look only at touching neighbours: 12251da177e4SLinus Torvalds * we could search further afield for "compatible" anon_vmas. 12261da177e4SLinus Torvalds * But it would probably just be a waste of time searching, 12271da177e4SLinus Torvalds * or lead to too many vmas hanging off the same anon_vma. 12281da177e4SLinus Torvalds * We're trying to allow mprotect remerging later on, 12291da177e4SLinus Torvalds * not trying to minimize memory used for anon_vmas. 12301da177e4SLinus Torvalds */ 12311da177e4SLinus Torvalds return NULL; 12321da177e4SLinus Torvalds } 12331da177e4SLinus Torvalds 12341da177e4SLinus Torvalds /* 123540401530SAl Viro * If a hint addr is less than mmap_min_addr change hint to be as 123640401530SAl Viro * low as possible but still greater than mmap_min_addr 123740401530SAl Viro */ 123840401530SAl Viro static inline unsigned long round_hint_to_min(unsigned long hint) 123940401530SAl Viro { 124040401530SAl Viro hint &= PAGE_MASK; 124140401530SAl Viro if (((void *)hint != NULL) && 124240401530SAl Viro (hint < mmap_min_addr)) 124340401530SAl Viro return PAGE_ALIGN(mmap_min_addr); 124440401530SAl Viro return hint; 124540401530SAl Viro } 124640401530SAl Viro 1247363ee17fSDavidlohr Bueso static inline int mlock_future_check(struct mm_struct *mm, 1248363ee17fSDavidlohr Bueso unsigned long flags, 1249363ee17fSDavidlohr Bueso unsigned long len) 1250363ee17fSDavidlohr Bueso { 1251363ee17fSDavidlohr Bueso unsigned long locked, lock_limit; 1252363ee17fSDavidlohr Bueso 1253363ee17fSDavidlohr Bueso /* mlock MCL_FUTURE? */ 1254363ee17fSDavidlohr Bueso if (flags & VM_LOCKED) { 1255363ee17fSDavidlohr Bueso locked = len >> PAGE_SHIFT; 1256363ee17fSDavidlohr Bueso locked += mm->locked_vm; 1257363ee17fSDavidlohr Bueso lock_limit = rlimit(RLIMIT_MEMLOCK); 1258363ee17fSDavidlohr Bueso lock_limit >>= PAGE_SHIFT; 1259363ee17fSDavidlohr Bueso if (locked > lock_limit && !capable(CAP_IPC_LOCK)) 1260363ee17fSDavidlohr Bueso return -EAGAIN; 1261363ee17fSDavidlohr Bueso } 1262363ee17fSDavidlohr Bueso return 0; 1263363ee17fSDavidlohr Bueso } 1264363ee17fSDavidlohr Bueso 126540401530SAl Viro /* 126627f5de79SJianjun Kong * The caller must hold down_write(¤t->mm->mmap_sem). 12671da177e4SLinus Torvalds */ 12681fcfd8dbSOleg Nesterov unsigned long do_mmap(struct file *file, unsigned long addr, 12691da177e4SLinus Torvalds unsigned long len, unsigned long prot, 12701fcfd8dbSOleg Nesterov unsigned long flags, vm_flags_t vm_flags, 12711fcfd8dbSOleg Nesterov unsigned long pgoff, unsigned long *populate) 12721da177e4SLinus Torvalds { 12731da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 1274*62b5f7d0SDave Hansen int pkey = 0; 12751da177e4SLinus Torvalds 127641badc15SMichel Lespinasse *populate = 0; 1277bebeb3d6SMichel Lespinasse 1278e37609bbSPiotr Kwapulinski if (!len) 1279e37609bbSPiotr Kwapulinski return -EINVAL; 1280e37609bbSPiotr Kwapulinski 12811da177e4SLinus Torvalds /* 12821da177e4SLinus Torvalds * Does the application expect PROT_READ to imply PROT_EXEC? 12831da177e4SLinus Torvalds * 12841da177e4SLinus Torvalds * (the exception is when the underlying filesystem is noexec 12851da177e4SLinus Torvalds * mounted, in which case we dont add PROT_EXEC.) 12861da177e4SLinus Torvalds */ 12871da177e4SLinus Torvalds if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC)) 128890f8572bSEric W. Biederman if (!(file && path_noexec(&file->f_path))) 12891da177e4SLinus Torvalds prot |= PROT_EXEC; 12901da177e4SLinus Torvalds 12917cd94146SEric Paris if (!(flags & MAP_FIXED)) 12927cd94146SEric Paris addr = round_hint_to_min(addr); 12937cd94146SEric Paris 12941da177e4SLinus Torvalds /* Careful about overflows.. */ 12951da177e4SLinus Torvalds len = PAGE_ALIGN(len); 12969206de95SAl Viro if (!len) 12971da177e4SLinus Torvalds return -ENOMEM; 12981da177e4SLinus Torvalds 12991da177e4SLinus Torvalds /* offset overflow? */ 13001da177e4SLinus Torvalds if ((pgoff + (len >> PAGE_SHIFT)) < pgoff) 13011da177e4SLinus Torvalds return -EOVERFLOW; 13021da177e4SLinus Torvalds 13031da177e4SLinus Torvalds /* Too many mappings? */ 13041da177e4SLinus Torvalds if (mm->map_count > sysctl_max_map_count) 13051da177e4SLinus Torvalds return -ENOMEM; 13061da177e4SLinus Torvalds 13071da177e4SLinus Torvalds /* Obtain the address to map to. we verify (or select) it and ensure 13081da177e4SLinus Torvalds * that it represents a valid section of the address space. 13091da177e4SLinus Torvalds */ 13101da177e4SLinus Torvalds addr = get_unmapped_area(file, addr, len, pgoff, flags); 1311de1741a1SAlexander Kuleshov if (offset_in_page(addr)) 13121da177e4SLinus Torvalds return addr; 13131da177e4SLinus Torvalds 1314*62b5f7d0SDave Hansen if (prot == PROT_EXEC) { 1315*62b5f7d0SDave Hansen pkey = execute_only_pkey(mm); 1316*62b5f7d0SDave Hansen if (pkey < 0) 1317*62b5f7d0SDave Hansen pkey = 0; 1318*62b5f7d0SDave Hansen } 1319*62b5f7d0SDave Hansen 13201da177e4SLinus Torvalds /* Do simple checking here so the lower-level routines won't have 13211da177e4SLinus Torvalds * to. we assume access permissions have been handled by the open 13221da177e4SLinus Torvalds * of the memory object, so we don't do any here. 13231da177e4SLinus Torvalds */ 1324*62b5f7d0SDave Hansen vm_flags |= calc_vm_prot_bits(prot, pkey) | calc_vm_flag_bits(flags) | 13251da177e4SLinus Torvalds mm->def_flags | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC; 13261da177e4SLinus Torvalds 1327cdf7b341SHuang Shijie if (flags & MAP_LOCKED) 13281da177e4SLinus Torvalds if (!can_do_mlock()) 13291da177e4SLinus Torvalds return -EPERM; 1330ba470de4SRik van Riel 1331363ee17fSDavidlohr Bueso if (mlock_future_check(mm, vm_flags, len)) 13321da177e4SLinus Torvalds return -EAGAIN; 13331da177e4SLinus Torvalds 13341da177e4SLinus Torvalds if (file) { 1335077bf22bSOleg Nesterov struct inode *inode = file_inode(file); 1336077bf22bSOleg Nesterov 13371da177e4SLinus Torvalds switch (flags & MAP_TYPE) { 13381da177e4SLinus Torvalds case MAP_SHARED: 13391da177e4SLinus Torvalds if ((prot&PROT_WRITE) && !(file->f_mode&FMODE_WRITE)) 13401da177e4SLinus Torvalds return -EACCES; 13411da177e4SLinus Torvalds 13421da177e4SLinus Torvalds /* 13431da177e4SLinus Torvalds * Make sure we don't allow writing to an append-only 13441da177e4SLinus Torvalds * file.. 13451da177e4SLinus Torvalds */ 13461da177e4SLinus Torvalds if (IS_APPEND(inode) && (file->f_mode & FMODE_WRITE)) 13471da177e4SLinus Torvalds return -EACCES; 13481da177e4SLinus Torvalds 13491da177e4SLinus Torvalds /* 13501da177e4SLinus Torvalds * Make sure there are no mandatory locks on the file. 13511da177e4SLinus Torvalds */ 1352d7a06983SJeff Layton if (locks_verify_locked(file)) 13531da177e4SLinus Torvalds return -EAGAIN; 13541da177e4SLinus Torvalds 13551da177e4SLinus Torvalds vm_flags |= VM_SHARED | VM_MAYSHARE; 13561da177e4SLinus Torvalds if (!(file->f_mode & FMODE_WRITE)) 13571da177e4SLinus Torvalds vm_flags &= ~(VM_MAYWRITE | VM_SHARED); 13581da177e4SLinus Torvalds 13591da177e4SLinus Torvalds /* fall through */ 13601da177e4SLinus Torvalds case MAP_PRIVATE: 13611da177e4SLinus Torvalds if (!(file->f_mode & FMODE_READ)) 13621da177e4SLinus Torvalds return -EACCES; 136390f8572bSEric W. Biederman if (path_noexec(&file->f_path)) { 136480c5606cSLinus Torvalds if (vm_flags & VM_EXEC) 136580c5606cSLinus Torvalds return -EPERM; 136680c5606cSLinus Torvalds vm_flags &= ~VM_MAYEXEC; 136780c5606cSLinus Torvalds } 136880c5606cSLinus Torvalds 136972c2d531SAl Viro if (!file->f_op->mmap) 137080c5606cSLinus Torvalds return -ENODEV; 1371b2c56e4fSOleg Nesterov if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP)) 1372b2c56e4fSOleg Nesterov return -EINVAL; 13731da177e4SLinus Torvalds break; 13741da177e4SLinus Torvalds 13751da177e4SLinus Torvalds default: 13761da177e4SLinus Torvalds return -EINVAL; 13771da177e4SLinus Torvalds } 13781da177e4SLinus Torvalds } else { 13791da177e4SLinus Torvalds switch (flags & MAP_TYPE) { 13801da177e4SLinus Torvalds case MAP_SHARED: 1381b2c56e4fSOleg Nesterov if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP)) 1382b2c56e4fSOleg Nesterov return -EINVAL; 1383ce363942STejun Heo /* 1384ce363942STejun Heo * Ignore pgoff. 1385ce363942STejun Heo */ 1386ce363942STejun Heo pgoff = 0; 13871da177e4SLinus Torvalds vm_flags |= VM_SHARED | VM_MAYSHARE; 13881da177e4SLinus Torvalds break; 13891da177e4SLinus Torvalds case MAP_PRIVATE: 13901da177e4SLinus Torvalds /* 13911da177e4SLinus Torvalds * Set pgoff according to addr for anon_vma. 13921da177e4SLinus Torvalds */ 13931da177e4SLinus Torvalds pgoff = addr >> PAGE_SHIFT; 13941da177e4SLinus Torvalds break; 13951da177e4SLinus Torvalds default: 13961da177e4SLinus Torvalds return -EINVAL; 13971da177e4SLinus Torvalds } 13981da177e4SLinus Torvalds } 13991da177e4SLinus Torvalds 1400c22c0d63SMichel Lespinasse /* 1401c22c0d63SMichel Lespinasse * Set 'VM_NORESERVE' if we should not account for the 1402c22c0d63SMichel Lespinasse * memory use of this mapping. 1403c22c0d63SMichel Lespinasse */ 1404c22c0d63SMichel Lespinasse if (flags & MAP_NORESERVE) { 1405c22c0d63SMichel Lespinasse /* We honor MAP_NORESERVE if allowed to overcommit */ 1406c22c0d63SMichel Lespinasse if (sysctl_overcommit_memory != OVERCOMMIT_NEVER) 1407c22c0d63SMichel Lespinasse vm_flags |= VM_NORESERVE; 1408c22c0d63SMichel Lespinasse 1409c22c0d63SMichel Lespinasse /* hugetlb applies strict overcommit unless MAP_NORESERVE */ 1410c22c0d63SMichel Lespinasse if (file && is_file_hugepages(file)) 1411c22c0d63SMichel Lespinasse vm_flags |= VM_NORESERVE; 1412c22c0d63SMichel Lespinasse } 1413c22c0d63SMichel Lespinasse 1414c22c0d63SMichel Lespinasse addr = mmap_region(file, addr, len, vm_flags, pgoff); 141509a9f1d2SMichel Lespinasse if (!IS_ERR_VALUE(addr) && 141609a9f1d2SMichel Lespinasse ((vm_flags & VM_LOCKED) || 141709a9f1d2SMichel Lespinasse (flags & (MAP_POPULATE | MAP_NONBLOCK)) == MAP_POPULATE)) 141841badc15SMichel Lespinasse *populate = len; 1419bebeb3d6SMichel Lespinasse return addr; 14200165ab44SMiklos Szeredi } 14216be5ceb0SLinus Torvalds 142266f0dc48SHugh Dickins SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len, 142366f0dc48SHugh Dickins unsigned long, prot, unsigned long, flags, 142466f0dc48SHugh Dickins unsigned long, fd, unsigned long, pgoff) 142566f0dc48SHugh Dickins { 142666f0dc48SHugh Dickins struct file *file = NULL; 14271e3ee14bSChen Gang unsigned long retval; 142866f0dc48SHugh Dickins 142966f0dc48SHugh Dickins if (!(flags & MAP_ANONYMOUS)) { 1430120a795dSAl Viro audit_mmap_fd(fd, flags); 143166f0dc48SHugh Dickins file = fget(fd); 143266f0dc48SHugh Dickins if (!file) 14331e3ee14bSChen Gang return -EBADF; 1434af73e4d9SNaoya Horiguchi if (is_file_hugepages(file)) 1435af73e4d9SNaoya Horiguchi len = ALIGN(len, huge_page_size(hstate_file(file))); 1436493af578SJörn Engel retval = -EINVAL; 1437493af578SJörn Engel if (unlikely(flags & MAP_HUGETLB && !is_file_hugepages(file))) 1438493af578SJörn Engel goto out_fput; 143966f0dc48SHugh Dickins } else if (flags & MAP_HUGETLB) { 144066f0dc48SHugh Dickins struct user_struct *user = NULL; 1441c103a4dcSAndrew Morton struct hstate *hs; 1442af73e4d9SNaoya Horiguchi 1443c103a4dcSAndrew Morton hs = hstate_sizelog((flags >> MAP_HUGE_SHIFT) & SHM_HUGE_MASK); 1444091d0d55SLi Zefan if (!hs) 1445091d0d55SLi Zefan return -EINVAL; 1446091d0d55SLi Zefan 1447091d0d55SLi Zefan len = ALIGN(len, huge_page_size(hs)); 144866f0dc48SHugh Dickins /* 144966f0dc48SHugh Dickins * VM_NORESERVE is used because the reservations will be 145066f0dc48SHugh Dickins * taken when vm_ops->mmap() is called 145166f0dc48SHugh Dickins * A dummy user value is used because we are not locking 145266f0dc48SHugh Dickins * memory so no accounting is necessary 145366f0dc48SHugh Dickins */ 1454af73e4d9SNaoya Horiguchi file = hugetlb_file_setup(HUGETLB_ANON_FILE, len, 145542d7395fSAndi Kleen VM_NORESERVE, 145642d7395fSAndi Kleen &user, HUGETLB_ANONHUGE_INODE, 145742d7395fSAndi Kleen (flags >> MAP_HUGE_SHIFT) & MAP_HUGE_MASK); 145866f0dc48SHugh Dickins if (IS_ERR(file)) 145966f0dc48SHugh Dickins return PTR_ERR(file); 146066f0dc48SHugh Dickins } 146166f0dc48SHugh Dickins 146266f0dc48SHugh Dickins flags &= ~(MAP_EXECUTABLE | MAP_DENYWRITE); 146366f0dc48SHugh Dickins 1464eb36c587SAl Viro retval = vm_mmap_pgoff(file, addr, len, prot, flags, pgoff); 1465493af578SJörn Engel out_fput: 146666f0dc48SHugh Dickins if (file) 146766f0dc48SHugh Dickins fput(file); 146866f0dc48SHugh Dickins return retval; 146966f0dc48SHugh Dickins } 147066f0dc48SHugh Dickins 1471a4679373SChristoph Hellwig #ifdef __ARCH_WANT_SYS_OLD_MMAP 1472a4679373SChristoph Hellwig struct mmap_arg_struct { 1473a4679373SChristoph Hellwig unsigned long addr; 1474a4679373SChristoph Hellwig unsigned long len; 1475a4679373SChristoph Hellwig unsigned long prot; 1476a4679373SChristoph Hellwig unsigned long flags; 1477a4679373SChristoph Hellwig unsigned long fd; 1478a4679373SChristoph Hellwig unsigned long offset; 1479a4679373SChristoph Hellwig }; 1480a4679373SChristoph Hellwig 1481a4679373SChristoph Hellwig SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg) 1482a4679373SChristoph Hellwig { 1483a4679373SChristoph Hellwig struct mmap_arg_struct a; 1484a4679373SChristoph Hellwig 1485a4679373SChristoph Hellwig if (copy_from_user(&a, arg, sizeof(a))) 1486a4679373SChristoph Hellwig return -EFAULT; 1487de1741a1SAlexander Kuleshov if (offset_in_page(a.offset)) 1488a4679373SChristoph Hellwig return -EINVAL; 1489a4679373SChristoph Hellwig 1490a4679373SChristoph Hellwig return sys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd, 1491a4679373SChristoph Hellwig a.offset >> PAGE_SHIFT); 1492a4679373SChristoph Hellwig } 1493a4679373SChristoph Hellwig #endif /* __ARCH_WANT_SYS_OLD_MMAP */ 1494a4679373SChristoph Hellwig 14954e950f6fSAlexey Dobriyan /* 14964e950f6fSAlexey Dobriyan * Some shared mappigns will want the pages marked read-only 14974e950f6fSAlexey Dobriyan * to track write events. If so, we'll downgrade vm_page_prot 14984e950f6fSAlexey Dobriyan * to the private version (using protection_map[] without the 14994e950f6fSAlexey Dobriyan * VM_SHARED bit). 15004e950f6fSAlexey Dobriyan */ 15014e950f6fSAlexey Dobriyan int vma_wants_writenotify(struct vm_area_struct *vma) 15024e950f6fSAlexey Dobriyan { 1503ca16d140SKOSAKI Motohiro vm_flags_t vm_flags = vma->vm_flags; 15048a04446aSKirill A. Shutemov const struct vm_operations_struct *vm_ops = vma->vm_ops; 15054e950f6fSAlexey Dobriyan 15064e950f6fSAlexey Dobriyan /* If it was private or non-writable, the write bit is already clear */ 15074e950f6fSAlexey Dobriyan if ((vm_flags & (VM_WRITE|VM_SHARED)) != ((VM_WRITE|VM_SHARED))) 15084e950f6fSAlexey Dobriyan return 0; 15094e950f6fSAlexey Dobriyan 15104e950f6fSAlexey Dobriyan /* The backer wishes to know when pages are first written to? */ 15118a04446aSKirill A. Shutemov if (vm_ops && (vm_ops->page_mkwrite || vm_ops->pfn_mkwrite)) 15124e950f6fSAlexey Dobriyan return 1; 15134e950f6fSAlexey Dobriyan 151464e45507SPeter Feiner /* The open routine did something to the protections that pgprot_modify 151564e45507SPeter Feiner * won't preserve? */ 15164e950f6fSAlexey Dobriyan if (pgprot_val(vma->vm_page_prot) != 151764e45507SPeter Feiner pgprot_val(vm_pgprot_modify(vma->vm_page_prot, vm_flags))) 15184e950f6fSAlexey Dobriyan return 0; 15194e950f6fSAlexey Dobriyan 152064e45507SPeter Feiner /* Do we need to track softdirty? */ 152164e45507SPeter Feiner if (IS_ENABLED(CONFIG_MEM_SOFT_DIRTY) && !(vm_flags & VM_SOFTDIRTY)) 152264e45507SPeter Feiner return 1; 152364e45507SPeter Feiner 15244e950f6fSAlexey Dobriyan /* Specialty mapping? */ 15254b6e1e37SKonstantin Khlebnikov if (vm_flags & VM_PFNMAP) 15264e950f6fSAlexey Dobriyan return 0; 15274e950f6fSAlexey Dobriyan 15284e950f6fSAlexey Dobriyan /* Can the mapping track the dirty pages? */ 15294e950f6fSAlexey Dobriyan return vma->vm_file && vma->vm_file->f_mapping && 15304e950f6fSAlexey Dobriyan mapping_cap_account_dirty(vma->vm_file->f_mapping); 15314e950f6fSAlexey Dobriyan } 15324e950f6fSAlexey Dobriyan 1533fc8744adSLinus Torvalds /* 1534fc8744adSLinus Torvalds * We account for memory if it's a private writeable mapping, 15355a6fe125SMel Gorman * not hugepages and VM_NORESERVE wasn't set. 1536fc8744adSLinus Torvalds */ 1537ca16d140SKOSAKI Motohiro static inline int accountable_mapping(struct file *file, vm_flags_t vm_flags) 1538fc8744adSLinus Torvalds { 15395a6fe125SMel Gorman /* 15405a6fe125SMel Gorman * hugetlb has its own accounting separate from the core VM 15415a6fe125SMel Gorman * VM_HUGETLB may not be set yet so we cannot check for that flag. 15425a6fe125SMel Gorman */ 15435a6fe125SMel Gorman if (file && is_file_hugepages(file)) 15445a6fe125SMel Gorman return 0; 15455a6fe125SMel Gorman 1546fc8744adSLinus Torvalds return (vm_flags & (VM_NORESERVE | VM_SHARED | VM_WRITE)) == VM_WRITE; 1547fc8744adSLinus Torvalds } 1548fc8744adSLinus Torvalds 15490165ab44SMiklos Szeredi unsigned long mmap_region(struct file *file, unsigned long addr, 1550c22c0d63SMichel Lespinasse unsigned long len, vm_flags_t vm_flags, unsigned long pgoff) 15510165ab44SMiklos Szeredi { 15520165ab44SMiklos Szeredi struct mm_struct *mm = current->mm; 15530165ab44SMiklos Szeredi struct vm_area_struct *vma, *prev; 15540165ab44SMiklos Szeredi int error; 15550165ab44SMiklos Szeredi struct rb_node **rb_link, *rb_parent; 15560165ab44SMiklos Szeredi unsigned long charged = 0; 15570165ab44SMiklos Szeredi 1558e8420a8eSCyril Hrubis /* Check against address space limit. */ 155984638335SKonstantin Khlebnikov if (!may_expand_vm(mm, vm_flags, len >> PAGE_SHIFT)) { 1560e8420a8eSCyril Hrubis unsigned long nr_pages; 1561e8420a8eSCyril Hrubis 1562e8420a8eSCyril Hrubis /* 1563e8420a8eSCyril Hrubis * MAP_FIXED may remove pages of mappings that intersects with 1564e8420a8eSCyril Hrubis * requested mapping. Account for the pages it would unmap. 1565e8420a8eSCyril Hrubis */ 1566e8420a8eSCyril Hrubis nr_pages = count_vma_pages_range(mm, addr, addr + len); 1567e8420a8eSCyril Hrubis 156884638335SKonstantin Khlebnikov if (!may_expand_vm(mm, vm_flags, 156984638335SKonstantin Khlebnikov (len >> PAGE_SHIFT) - nr_pages)) 1570e8420a8eSCyril Hrubis return -ENOMEM; 1571e8420a8eSCyril Hrubis } 1572e8420a8eSCyril Hrubis 15731da177e4SLinus Torvalds /* Clear old maps */ 15749fcd1457SRasmus Villemoes while (find_vma_links(mm, addr, addr + len, &prev, &rb_link, 15759fcd1457SRasmus Villemoes &rb_parent)) { 15761da177e4SLinus Torvalds if (do_munmap(mm, addr, len)) 15771da177e4SLinus Torvalds return -ENOMEM; 15781da177e4SLinus Torvalds } 15791da177e4SLinus Torvalds 1580fc8744adSLinus Torvalds /* 15811da177e4SLinus Torvalds * Private writable mapping: check memory availability 15821da177e4SLinus Torvalds */ 15835a6fe125SMel Gorman if (accountable_mapping(file, vm_flags)) { 15841da177e4SLinus Torvalds charged = len >> PAGE_SHIFT; 1585191c5424SAl Viro if (security_vm_enough_memory_mm(mm, charged)) 15861da177e4SLinus Torvalds return -ENOMEM; 15871da177e4SLinus Torvalds vm_flags |= VM_ACCOUNT; 15881da177e4SLinus Torvalds } 15891da177e4SLinus Torvalds 15901da177e4SLinus Torvalds /* 1591de33c8dbSLinus Torvalds * Can we just expand an old mapping? 15921da177e4SLinus Torvalds */ 159319a809afSAndrea Arcangeli vma = vma_merge(mm, prev, addr, addr + len, vm_flags, 159419a809afSAndrea Arcangeli NULL, file, pgoff, NULL, NULL_VM_UFFD_CTX); 1595ba470de4SRik van Riel if (vma) 15961da177e4SLinus Torvalds goto out; 15971da177e4SLinus Torvalds 15981da177e4SLinus Torvalds /* 15991da177e4SLinus Torvalds * Determine the object being mapped and call the appropriate 16001da177e4SLinus Torvalds * specific mapper. the address has already been validated, but 16011da177e4SLinus Torvalds * not unmapped, but the maps are removed from the list. 16021da177e4SLinus Torvalds */ 1603c5e3b83eSPekka Enberg vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL); 16041da177e4SLinus Torvalds if (!vma) { 16051da177e4SLinus Torvalds error = -ENOMEM; 16061da177e4SLinus Torvalds goto unacct_error; 16071da177e4SLinus Torvalds } 16081da177e4SLinus Torvalds 16091da177e4SLinus Torvalds vma->vm_mm = mm; 16101da177e4SLinus Torvalds vma->vm_start = addr; 16111da177e4SLinus Torvalds vma->vm_end = addr + len; 16121da177e4SLinus Torvalds vma->vm_flags = vm_flags; 16133ed75eb8SColy Li vma->vm_page_prot = vm_get_page_prot(vm_flags); 16141da177e4SLinus Torvalds vma->vm_pgoff = pgoff; 16155beb4930SRik van Riel INIT_LIST_HEAD(&vma->anon_vma_chain); 16161da177e4SLinus Torvalds 16171da177e4SLinus Torvalds if (file) { 16181da177e4SLinus Torvalds if (vm_flags & VM_DENYWRITE) { 16191da177e4SLinus Torvalds error = deny_write_access(file); 16201da177e4SLinus Torvalds if (error) 16211da177e4SLinus Torvalds goto free_vma; 16221da177e4SLinus Torvalds } 16234bb5f5d9SDavid Herrmann if (vm_flags & VM_SHARED) { 16244bb5f5d9SDavid Herrmann error = mapping_map_writable(file->f_mapping); 16254bb5f5d9SDavid Herrmann if (error) 16264bb5f5d9SDavid Herrmann goto allow_write_and_free_vma; 16274bb5f5d9SDavid Herrmann } 16284bb5f5d9SDavid Herrmann 16294bb5f5d9SDavid Herrmann /* ->mmap() can change vma->vm_file, but must guarantee that 16304bb5f5d9SDavid Herrmann * vma_link() below can deny write-access if VM_DENYWRITE is set 16314bb5f5d9SDavid Herrmann * and map writably if VM_SHARED is set. This usually means the 16324bb5f5d9SDavid Herrmann * new file must not have been exposed to user-space, yet. 16334bb5f5d9SDavid Herrmann */ 1634cb0942b8SAl Viro vma->vm_file = get_file(file); 16351da177e4SLinus Torvalds error = file->f_op->mmap(file, vma); 16361da177e4SLinus Torvalds if (error) 16371da177e4SLinus Torvalds goto unmap_and_free_vma; 16381da177e4SLinus Torvalds 16391da177e4SLinus Torvalds /* Can addr have changed?? 16401da177e4SLinus Torvalds * 16411da177e4SLinus Torvalds * Answer: Yes, several device drivers can do it in their 16421da177e4SLinus Torvalds * f_op->mmap method. -DaveM 16432897b4d2SJoonsoo Kim * Bug: If addr is changed, prev, rb_link, rb_parent should 16442897b4d2SJoonsoo Kim * be updated for vma_link() 16451da177e4SLinus Torvalds */ 16462897b4d2SJoonsoo Kim WARN_ON_ONCE(addr != vma->vm_start); 16472897b4d2SJoonsoo Kim 16481da177e4SLinus Torvalds addr = vma->vm_start; 16491da177e4SLinus Torvalds vm_flags = vma->vm_flags; 1650f8dbf0a7SHuang Shijie } else if (vm_flags & VM_SHARED) { 1651f8dbf0a7SHuang Shijie error = shmem_zero_setup(vma); 1652f8dbf0a7SHuang Shijie if (error) 1653f8dbf0a7SHuang Shijie goto free_vma; 1654f8dbf0a7SHuang Shijie } 16551da177e4SLinus Torvalds 16564d3d5b41SOleg Nesterov vma_link(mm, vma, prev, rb_link, rb_parent); 16574d3d5b41SOleg Nesterov /* Once vma denies write, undo our temporary denial count */ 16584bb5f5d9SDavid Herrmann if (file) { 16594bb5f5d9SDavid Herrmann if (vm_flags & VM_SHARED) 16604bb5f5d9SDavid Herrmann mapping_unmap_writable(file->f_mapping); 1661e8686772SOleg Nesterov if (vm_flags & VM_DENYWRITE) 1662e8686772SOleg Nesterov allow_write_access(file); 16634bb5f5d9SDavid Herrmann } 1664e8686772SOleg Nesterov file = vma->vm_file; 16651da177e4SLinus Torvalds out: 1666cdd6c482SIngo Molnar perf_event_mmap(vma); 16670a4a9391SPeter Zijlstra 166884638335SKonstantin Khlebnikov vm_stat_account(mm, vm_flags, len >> PAGE_SHIFT); 16691da177e4SLinus Torvalds if (vm_flags & VM_LOCKED) { 1670bebeb3d6SMichel Lespinasse if (!((vm_flags & VM_SPECIAL) || is_vm_hugetlb_page(vma) || 1671bebeb3d6SMichel Lespinasse vma == get_gate_vma(current->mm))) 167206f9d8c2SKOSAKI Motohiro mm->locked_vm += (len >> PAGE_SHIFT); 1673bebeb3d6SMichel Lespinasse else 1674de60f5f1SEric B Munson vma->vm_flags &= VM_LOCKED_CLEAR_MASK; 1675bebeb3d6SMichel Lespinasse } 16762b144498SSrikar Dronamraju 1677c7a3a88cSOleg Nesterov if (file) 1678c7a3a88cSOleg Nesterov uprobe_mmap(vma); 16792b144498SSrikar Dronamraju 1680d9104d1cSCyrill Gorcunov /* 1681d9104d1cSCyrill Gorcunov * New (or expanded) vma always get soft dirty status. 1682d9104d1cSCyrill Gorcunov * Otherwise user-space soft-dirty page tracker won't 1683d9104d1cSCyrill Gorcunov * be able to distinguish situation when vma area unmapped, 1684d9104d1cSCyrill Gorcunov * then new mapped in-place (which must be aimed as 1685d9104d1cSCyrill Gorcunov * a completely new data area). 1686d9104d1cSCyrill Gorcunov */ 1687d9104d1cSCyrill Gorcunov vma->vm_flags |= VM_SOFTDIRTY; 1688d9104d1cSCyrill Gorcunov 168964e45507SPeter Feiner vma_set_page_prot(vma); 169064e45507SPeter Feiner 16911da177e4SLinus Torvalds return addr; 16921da177e4SLinus Torvalds 16931da177e4SLinus Torvalds unmap_and_free_vma: 16941da177e4SLinus Torvalds vma->vm_file = NULL; 16951da177e4SLinus Torvalds fput(file); 16961da177e4SLinus Torvalds 16971da177e4SLinus Torvalds /* Undo any partial mapping done by a device driver. */ 1698e0da382cSHugh Dickins unmap_region(mm, vma, prev, vma->vm_start, vma->vm_end); 1699e0da382cSHugh Dickins charged = 0; 17004bb5f5d9SDavid Herrmann if (vm_flags & VM_SHARED) 17014bb5f5d9SDavid Herrmann mapping_unmap_writable(file->f_mapping); 17024bb5f5d9SDavid Herrmann allow_write_and_free_vma: 17034bb5f5d9SDavid Herrmann if (vm_flags & VM_DENYWRITE) 17044bb5f5d9SDavid Herrmann allow_write_access(file); 17051da177e4SLinus Torvalds free_vma: 17061da177e4SLinus Torvalds kmem_cache_free(vm_area_cachep, vma); 17071da177e4SLinus Torvalds unacct_error: 17081da177e4SLinus Torvalds if (charged) 17091da177e4SLinus Torvalds vm_unacct_memory(charged); 17101da177e4SLinus Torvalds return error; 17111da177e4SLinus Torvalds } 17121da177e4SLinus Torvalds 1713db4fbfb9SMichel Lespinasse unsigned long unmapped_area(struct vm_unmapped_area_info *info) 1714db4fbfb9SMichel Lespinasse { 1715db4fbfb9SMichel Lespinasse /* 1716db4fbfb9SMichel Lespinasse * We implement the search by looking for an rbtree node that 1717db4fbfb9SMichel Lespinasse * immediately follows a suitable gap. That is, 1718db4fbfb9SMichel Lespinasse * - gap_start = vma->vm_prev->vm_end <= info->high_limit - length; 1719db4fbfb9SMichel Lespinasse * - gap_end = vma->vm_start >= info->low_limit + length; 1720db4fbfb9SMichel Lespinasse * - gap_end - gap_start >= length 1721db4fbfb9SMichel Lespinasse */ 1722db4fbfb9SMichel Lespinasse 1723db4fbfb9SMichel Lespinasse struct mm_struct *mm = current->mm; 1724db4fbfb9SMichel Lespinasse struct vm_area_struct *vma; 1725db4fbfb9SMichel Lespinasse unsigned long length, low_limit, high_limit, gap_start, gap_end; 1726db4fbfb9SMichel Lespinasse 1727db4fbfb9SMichel Lespinasse /* Adjust search length to account for worst case alignment overhead */ 1728db4fbfb9SMichel Lespinasse length = info->length + info->align_mask; 1729db4fbfb9SMichel Lespinasse if (length < info->length) 1730db4fbfb9SMichel Lespinasse return -ENOMEM; 1731db4fbfb9SMichel Lespinasse 1732db4fbfb9SMichel Lespinasse /* Adjust search limits by the desired length */ 1733db4fbfb9SMichel Lespinasse if (info->high_limit < length) 1734db4fbfb9SMichel Lespinasse return -ENOMEM; 1735db4fbfb9SMichel Lespinasse high_limit = info->high_limit - length; 1736db4fbfb9SMichel Lespinasse 1737db4fbfb9SMichel Lespinasse if (info->low_limit > high_limit) 1738db4fbfb9SMichel Lespinasse return -ENOMEM; 1739db4fbfb9SMichel Lespinasse low_limit = info->low_limit + length; 1740db4fbfb9SMichel Lespinasse 1741db4fbfb9SMichel Lespinasse /* Check if rbtree root looks promising */ 1742db4fbfb9SMichel Lespinasse if (RB_EMPTY_ROOT(&mm->mm_rb)) 1743db4fbfb9SMichel Lespinasse goto check_highest; 1744db4fbfb9SMichel Lespinasse vma = rb_entry(mm->mm_rb.rb_node, struct vm_area_struct, vm_rb); 1745db4fbfb9SMichel Lespinasse if (vma->rb_subtree_gap < length) 1746db4fbfb9SMichel Lespinasse goto check_highest; 1747db4fbfb9SMichel Lespinasse 1748db4fbfb9SMichel Lespinasse while (true) { 1749db4fbfb9SMichel Lespinasse /* Visit left subtree if it looks promising */ 1750db4fbfb9SMichel Lespinasse gap_end = vma->vm_start; 1751db4fbfb9SMichel Lespinasse if (gap_end >= low_limit && vma->vm_rb.rb_left) { 1752db4fbfb9SMichel Lespinasse struct vm_area_struct *left = 1753db4fbfb9SMichel Lespinasse rb_entry(vma->vm_rb.rb_left, 1754db4fbfb9SMichel Lespinasse struct vm_area_struct, vm_rb); 1755db4fbfb9SMichel Lespinasse if (left->rb_subtree_gap >= length) { 1756db4fbfb9SMichel Lespinasse vma = left; 1757db4fbfb9SMichel Lespinasse continue; 1758db4fbfb9SMichel Lespinasse } 1759db4fbfb9SMichel Lespinasse } 1760db4fbfb9SMichel Lespinasse 1761db4fbfb9SMichel Lespinasse gap_start = vma->vm_prev ? vma->vm_prev->vm_end : 0; 1762db4fbfb9SMichel Lespinasse check_current: 1763db4fbfb9SMichel Lespinasse /* Check if current node has a suitable gap */ 1764db4fbfb9SMichel Lespinasse if (gap_start > high_limit) 1765db4fbfb9SMichel Lespinasse return -ENOMEM; 1766db4fbfb9SMichel Lespinasse if (gap_end >= low_limit && gap_end - gap_start >= length) 1767db4fbfb9SMichel Lespinasse goto found; 1768db4fbfb9SMichel Lespinasse 1769db4fbfb9SMichel Lespinasse /* Visit right subtree if it looks promising */ 1770db4fbfb9SMichel Lespinasse if (vma->vm_rb.rb_right) { 1771db4fbfb9SMichel Lespinasse struct vm_area_struct *right = 1772db4fbfb9SMichel Lespinasse rb_entry(vma->vm_rb.rb_right, 1773db4fbfb9SMichel Lespinasse struct vm_area_struct, vm_rb); 1774db4fbfb9SMichel Lespinasse if (right->rb_subtree_gap >= length) { 1775db4fbfb9SMichel Lespinasse vma = right; 1776db4fbfb9SMichel Lespinasse continue; 1777db4fbfb9SMichel Lespinasse } 1778db4fbfb9SMichel Lespinasse } 1779db4fbfb9SMichel Lespinasse 1780db4fbfb9SMichel Lespinasse /* Go back up the rbtree to find next candidate node */ 1781db4fbfb9SMichel Lespinasse while (true) { 1782db4fbfb9SMichel Lespinasse struct rb_node *prev = &vma->vm_rb; 1783db4fbfb9SMichel Lespinasse if (!rb_parent(prev)) 1784db4fbfb9SMichel Lespinasse goto check_highest; 1785db4fbfb9SMichel Lespinasse vma = rb_entry(rb_parent(prev), 1786db4fbfb9SMichel Lespinasse struct vm_area_struct, vm_rb); 1787db4fbfb9SMichel Lespinasse if (prev == vma->vm_rb.rb_left) { 1788db4fbfb9SMichel Lespinasse gap_start = vma->vm_prev->vm_end; 1789db4fbfb9SMichel Lespinasse gap_end = vma->vm_start; 1790db4fbfb9SMichel Lespinasse goto check_current; 1791db4fbfb9SMichel Lespinasse } 1792db4fbfb9SMichel Lespinasse } 1793db4fbfb9SMichel Lespinasse } 1794db4fbfb9SMichel Lespinasse 1795db4fbfb9SMichel Lespinasse check_highest: 1796db4fbfb9SMichel Lespinasse /* Check highest gap, which does not precede any rbtree node */ 1797db4fbfb9SMichel Lespinasse gap_start = mm->highest_vm_end; 1798db4fbfb9SMichel Lespinasse gap_end = ULONG_MAX; /* Only for VM_BUG_ON below */ 1799db4fbfb9SMichel Lespinasse if (gap_start > high_limit) 1800db4fbfb9SMichel Lespinasse return -ENOMEM; 1801db4fbfb9SMichel Lespinasse 1802db4fbfb9SMichel Lespinasse found: 1803db4fbfb9SMichel Lespinasse /* We found a suitable gap. Clip it with the original low_limit. */ 1804db4fbfb9SMichel Lespinasse if (gap_start < info->low_limit) 1805db4fbfb9SMichel Lespinasse gap_start = info->low_limit; 1806db4fbfb9SMichel Lespinasse 1807db4fbfb9SMichel Lespinasse /* Adjust gap address to the desired alignment */ 1808db4fbfb9SMichel Lespinasse gap_start += (info->align_offset - gap_start) & info->align_mask; 1809db4fbfb9SMichel Lespinasse 1810db4fbfb9SMichel Lespinasse VM_BUG_ON(gap_start + info->length > info->high_limit); 1811db4fbfb9SMichel Lespinasse VM_BUG_ON(gap_start + info->length > gap_end); 1812db4fbfb9SMichel Lespinasse return gap_start; 1813db4fbfb9SMichel Lespinasse } 1814db4fbfb9SMichel Lespinasse 1815db4fbfb9SMichel Lespinasse unsigned long unmapped_area_topdown(struct vm_unmapped_area_info *info) 1816db4fbfb9SMichel Lespinasse { 1817db4fbfb9SMichel Lespinasse struct mm_struct *mm = current->mm; 1818db4fbfb9SMichel Lespinasse struct vm_area_struct *vma; 1819db4fbfb9SMichel Lespinasse unsigned long length, low_limit, high_limit, gap_start, gap_end; 1820db4fbfb9SMichel Lespinasse 1821db4fbfb9SMichel Lespinasse /* Adjust search length to account for worst case alignment overhead */ 1822db4fbfb9SMichel Lespinasse length = info->length + info->align_mask; 1823db4fbfb9SMichel Lespinasse if (length < info->length) 1824db4fbfb9SMichel Lespinasse return -ENOMEM; 1825db4fbfb9SMichel Lespinasse 1826db4fbfb9SMichel Lespinasse /* 1827db4fbfb9SMichel Lespinasse * Adjust search limits by the desired length. 1828db4fbfb9SMichel Lespinasse * See implementation comment at top of unmapped_area(). 1829db4fbfb9SMichel Lespinasse */ 1830db4fbfb9SMichel Lespinasse gap_end = info->high_limit; 1831db4fbfb9SMichel Lespinasse if (gap_end < length) 1832db4fbfb9SMichel Lespinasse return -ENOMEM; 1833db4fbfb9SMichel Lespinasse high_limit = gap_end - length; 1834db4fbfb9SMichel Lespinasse 1835db4fbfb9SMichel Lespinasse if (info->low_limit > high_limit) 1836db4fbfb9SMichel Lespinasse return -ENOMEM; 1837db4fbfb9SMichel Lespinasse low_limit = info->low_limit + length; 1838db4fbfb9SMichel Lespinasse 1839db4fbfb9SMichel Lespinasse /* Check highest gap, which does not precede any rbtree node */ 1840db4fbfb9SMichel Lespinasse gap_start = mm->highest_vm_end; 1841db4fbfb9SMichel Lespinasse if (gap_start <= high_limit) 1842db4fbfb9SMichel Lespinasse goto found_highest; 1843db4fbfb9SMichel Lespinasse 1844db4fbfb9SMichel Lespinasse /* Check if rbtree root looks promising */ 1845db4fbfb9SMichel Lespinasse if (RB_EMPTY_ROOT(&mm->mm_rb)) 1846db4fbfb9SMichel Lespinasse return -ENOMEM; 1847db4fbfb9SMichel Lespinasse vma = rb_entry(mm->mm_rb.rb_node, struct vm_area_struct, vm_rb); 1848db4fbfb9SMichel Lespinasse if (vma->rb_subtree_gap < length) 1849db4fbfb9SMichel Lespinasse return -ENOMEM; 1850db4fbfb9SMichel Lespinasse 1851db4fbfb9SMichel Lespinasse while (true) { 1852db4fbfb9SMichel Lespinasse /* Visit right subtree if it looks promising */ 1853db4fbfb9SMichel Lespinasse gap_start = vma->vm_prev ? vma->vm_prev->vm_end : 0; 1854db4fbfb9SMichel Lespinasse if (gap_start <= high_limit && vma->vm_rb.rb_right) { 1855db4fbfb9SMichel Lespinasse struct vm_area_struct *right = 1856db4fbfb9SMichel Lespinasse rb_entry(vma->vm_rb.rb_right, 1857db4fbfb9SMichel Lespinasse struct vm_area_struct, vm_rb); 1858db4fbfb9SMichel Lespinasse if (right->rb_subtree_gap >= length) { 1859db4fbfb9SMichel Lespinasse vma = right; 1860db4fbfb9SMichel Lespinasse continue; 1861db4fbfb9SMichel Lespinasse } 1862db4fbfb9SMichel Lespinasse } 1863db4fbfb9SMichel Lespinasse 1864db4fbfb9SMichel Lespinasse check_current: 1865db4fbfb9SMichel Lespinasse /* Check if current node has a suitable gap */ 1866db4fbfb9SMichel Lespinasse gap_end = vma->vm_start; 1867db4fbfb9SMichel Lespinasse if (gap_end < low_limit) 1868db4fbfb9SMichel Lespinasse return -ENOMEM; 1869db4fbfb9SMichel Lespinasse if (gap_start <= high_limit && gap_end - gap_start >= length) 1870db4fbfb9SMichel Lespinasse goto found; 1871db4fbfb9SMichel Lespinasse 1872db4fbfb9SMichel Lespinasse /* Visit left subtree if it looks promising */ 1873db4fbfb9SMichel Lespinasse if (vma->vm_rb.rb_left) { 1874db4fbfb9SMichel Lespinasse struct vm_area_struct *left = 1875db4fbfb9SMichel Lespinasse rb_entry(vma->vm_rb.rb_left, 1876db4fbfb9SMichel Lespinasse struct vm_area_struct, vm_rb); 1877db4fbfb9SMichel Lespinasse if (left->rb_subtree_gap >= length) { 1878db4fbfb9SMichel Lespinasse vma = left; 1879db4fbfb9SMichel Lespinasse continue; 1880db4fbfb9SMichel Lespinasse } 1881db4fbfb9SMichel Lespinasse } 1882db4fbfb9SMichel Lespinasse 1883db4fbfb9SMichel Lespinasse /* Go back up the rbtree to find next candidate node */ 1884db4fbfb9SMichel Lespinasse while (true) { 1885db4fbfb9SMichel Lespinasse struct rb_node *prev = &vma->vm_rb; 1886db4fbfb9SMichel Lespinasse if (!rb_parent(prev)) 1887db4fbfb9SMichel Lespinasse return -ENOMEM; 1888db4fbfb9SMichel Lespinasse vma = rb_entry(rb_parent(prev), 1889db4fbfb9SMichel Lespinasse struct vm_area_struct, vm_rb); 1890db4fbfb9SMichel Lespinasse if (prev == vma->vm_rb.rb_right) { 1891db4fbfb9SMichel Lespinasse gap_start = vma->vm_prev ? 1892db4fbfb9SMichel Lespinasse vma->vm_prev->vm_end : 0; 1893db4fbfb9SMichel Lespinasse goto check_current; 1894db4fbfb9SMichel Lespinasse } 1895db4fbfb9SMichel Lespinasse } 1896db4fbfb9SMichel Lespinasse } 1897db4fbfb9SMichel Lespinasse 1898db4fbfb9SMichel Lespinasse found: 1899db4fbfb9SMichel Lespinasse /* We found a suitable gap. Clip it with the original high_limit. */ 1900db4fbfb9SMichel Lespinasse if (gap_end > info->high_limit) 1901db4fbfb9SMichel Lespinasse gap_end = info->high_limit; 1902db4fbfb9SMichel Lespinasse 1903db4fbfb9SMichel Lespinasse found_highest: 1904db4fbfb9SMichel Lespinasse /* Compute highest gap address at the desired alignment */ 1905db4fbfb9SMichel Lespinasse gap_end -= info->length; 1906db4fbfb9SMichel Lespinasse gap_end -= (gap_end - info->align_offset) & info->align_mask; 1907db4fbfb9SMichel Lespinasse 1908db4fbfb9SMichel Lespinasse VM_BUG_ON(gap_end < info->low_limit); 1909db4fbfb9SMichel Lespinasse VM_BUG_ON(gap_end < gap_start); 1910db4fbfb9SMichel Lespinasse return gap_end; 1911db4fbfb9SMichel Lespinasse } 1912db4fbfb9SMichel Lespinasse 19131da177e4SLinus Torvalds /* Get an address range which is currently unmapped. 19141da177e4SLinus Torvalds * For shmat() with addr=0. 19151da177e4SLinus Torvalds * 19161da177e4SLinus Torvalds * Ugly calling convention alert: 19171da177e4SLinus Torvalds * Return value with the low bits set means error value, 19181da177e4SLinus Torvalds * ie 19191da177e4SLinus Torvalds * if (ret & ~PAGE_MASK) 19201da177e4SLinus Torvalds * error = ret; 19211da177e4SLinus Torvalds * 19221da177e4SLinus Torvalds * This function "knows" that -ENOMEM has the bits set. 19231da177e4SLinus Torvalds */ 19241da177e4SLinus Torvalds #ifndef HAVE_ARCH_UNMAPPED_AREA 19251da177e4SLinus Torvalds unsigned long 19261da177e4SLinus Torvalds arch_get_unmapped_area(struct file *filp, unsigned long addr, 19271da177e4SLinus Torvalds unsigned long len, unsigned long pgoff, unsigned long flags) 19281da177e4SLinus Torvalds { 19291da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 19301da177e4SLinus Torvalds struct vm_area_struct *vma; 1931db4fbfb9SMichel Lespinasse struct vm_unmapped_area_info info; 19321da177e4SLinus Torvalds 19332afc745fSAkira Takeuchi if (len > TASK_SIZE - mmap_min_addr) 19341da177e4SLinus Torvalds return -ENOMEM; 19351da177e4SLinus Torvalds 193606abdfb4SBenjamin Herrenschmidt if (flags & MAP_FIXED) 193706abdfb4SBenjamin Herrenschmidt return addr; 193806abdfb4SBenjamin Herrenschmidt 19391da177e4SLinus Torvalds if (addr) { 19401da177e4SLinus Torvalds addr = PAGE_ALIGN(addr); 19411da177e4SLinus Torvalds vma = find_vma(mm, addr); 19422afc745fSAkira Takeuchi if (TASK_SIZE - len >= addr && addr >= mmap_min_addr && 19431da177e4SLinus Torvalds (!vma || addr + len <= vma->vm_start)) 19441da177e4SLinus Torvalds return addr; 19451da177e4SLinus Torvalds } 19461da177e4SLinus Torvalds 1947db4fbfb9SMichel Lespinasse info.flags = 0; 1948db4fbfb9SMichel Lespinasse info.length = len; 19494e99b021SHeiko Carstens info.low_limit = mm->mmap_base; 1950db4fbfb9SMichel Lespinasse info.high_limit = TASK_SIZE; 1951db4fbfb9SMichel Lespinasse info.align_mask = 0; 1952db4fbfb9SMichel Lespinasse return vm_unmapped_area(&info); 19531da177e4SLinus Torvalds } 19541da177e4SLinus Torvalds #endif 19551da177e4SLinus Torvalds 19561da177e4SLinus Torvalds /* 19571da177e4SLinus Torvalds * This mmap-allocator allocates new areas top-down from below the 19581da177e4SLinus Torvalds * stack's low limit (the base): 19591da177e4SLinus Torvalds */ 19601da177e4SLinus Torvalds #ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN 19611da177e4SLinus Torvalds unsigned long 19621da177e4SLinus Torvalds arch_get_unmapped_area_topdown(struct file *filp, const unsigned long addr0, 19631da177e4SLinus Torvalds const unsigned long len, const unsigned long pgoff, 19641da177e4SLinus Torvalds const unsigned long flags) 19651da177e4SLinus Torvalds { 19661da177e4SLinus Torvalds struct vm_area_struct *vma; 19671da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 1968db4fbfb9SMichel Lespinasse unsigned long addr = addr0; 1969db4fbfb9SMichel Lespinasse struct vm_unmapped_area_info info; 19701da177e4SLinus Torvalds 19711da177e4SLinus Torvalds /* requested length too big for entire address space */ 19722afc745fSAkira Takeuchi if (len > TASK_SIZE - mmap_min_addr) 19731da177e4SLinus Torvalds return -ENOMEM; 19741da177e4SLinus Torvalds 197506abdfb4SBenjamin Herrenschmidt if (flags & MAP_FIXED) 197606abdfb4SBenjamin Herrenschmidt return addr; 197706abdfb4SBenjamin Herrenschmidt 19781da177e4SLinus Torvalds /* requesting a specific address */ 19791da177e4SLinus Torvalds if (addr) { 19801da177e4SLinus Torvalds addr = PAGE_ALIGN(addr); 19811da177e4SLinus Torvalds vma = find_vma(mm, addr); 19822afc745fSAkira Takeuchi if (TASK_SIZE - len >= addr && addr >= mmap_min_addr && 19831da177e4SLinus Torvalds (!vma || addr + len <= vma->vm_start)) 19841da177e4SLinus Torvalds return addr; 19851da177e4SLinus Torvalds } 19861da177e4SLinus Torvalds 1987db4fbfb9SMichel Lespinasse info.flags = VM_UNMAPPED_AREA_TOPDOWN; 1988db4fbfb9SMichel Lespinasse info.length = len; 19892afc745fSAkira Takeuchi info.low_limit = max(PAGE_SIZE, mmap_min_addr); 1990db4fbfb9SMichel Lespinasse info.high_limit = mm->mmap_base; 1991db4fbfb9SMichel Lespinasse info.align_mask = 0; 1992db4fbfb9SMichel Lespinasse addr = vm_unmapped_area(&info); 1993b716ad95SXiao Guangrong 19941da177e4SLinus Torvalds /* 19951da177e4SLinus Torvalds * A failed mmap() very likely causes application failure, 19961da177e4SLinus Torvalds * so fall back to the bottom-up function here. This scenario 19971da177e4SLinus Torvalds * can happen with large stack limits and large mmap() 19981da177e4SLinus Torvalds * allocations. 19991da177e4SLinus Torvalds */ 2000de1741a1SAlexander Kuleshov if (offset_in_page(addr)) { 2001db4fbfb9SMichel Lespinasse VM_BUG_ON(addr != -ENOMEM); 2002db4fbfb9SMichel Lespinasse info.flags = 0; 2003db4fbfb9SMichel Lespinasse info.low_limit = TASK_UNMAPPED_BASE; 2004db4fbfb9SMichel Lespinasse info.high_limit = TASK_SIZE; 2005db4fbfb9SMichel Lespinasse addr = vm_unmapped_area(&info); 2006db4fbfb9SMichel Lespinasse } 20071da177e4SLinus Torvalds 20081da177e4SLinus Torvalds return addr; 20091da177e4SLinus Torvalds } 20101da177e4SLinus Torvalds #endif 20111da177e4SLinus Torvalds 20121da177e4SLinus Torvalds unsigned long 20131da177e4SLinus Torvalds get_unmapped_area(struct file *file, unsigned long addr, unsigned long len, 20141da177e4SLinus Torvalds unsigned long pgoff, unsigned long flags) 20151da177e4SLinus Torvalds { 201606abdfb4SBenjamin Herrenschmidt unsigned long (*get_area)(struct file *, unsigned long, 201706abdfb4SBenjamin Herrenschmidt unsigned long, unsigned long, unsigned long); 201807ab67c8SLinus Torvalds 20199206de95SAl Viro unsigned long error = arch_mmap_check(addr, len, flags); 20209206de95SAl Viro if (error) 20219206de95SAl Viro return error; 20229206de95SAl Viro 20239206de95SAl Viro /* Careful about overflows.. */ 20249206de95SAl Viro if (len > TASK_SIZE) 20259206de95SAl Viro return -ENOMEM; 20269206de95SAl Viro 202707ab67c8SLinus Torvalds get_area = current->mm->get_unmapped_area; 202872c2d531SAl Viro if (file && file->f_op->get_unmapped_area) 202907ab67c8SLinus Torvalds get_area = file->f_op->get_unmapped_area; 203007ab67c8SLinus Torvalds addr = get_area(file, addr, len, pgoff, flags); 203107ab67c8SLinus Torvalds if (IS_ERR_VALUE(addr)) 203207ab67c8SLinus Torvalds return addr; 203307ab67c8SLinus Torvalds 20341da177e4SLinus Torvalds if (addr > TASK_SIZE - len) 20351da177e4SLinus Torvalds return -ENOMEM; 2036de1741a1SAlexander Kuleshov if (offset_in_page(addr)) 20371da177e4SLinus Torvalds return -EINVAL; 203806abdfb4SBenjamin Herrenschmidt 20399ac4ed4bSAl Viro addr = arch_rebalance_pgtables(addr, len); 20409ac4ed4bSAl Viro error = security_mmap_addr(addr); 20419ac4ed4bSAl Viro return error ? error : addr; 20421da177e4SLinus Torvalds } 20431da177e4SLinus Torvalds 20441da177e4SLinus Torvalds EXPORT_SYMBOL(get_unmapped_area); 20451da177e4SLinus Torvalds 20461da177e4SLinus Torvalds /* Look up the first VMA which satisfies addr < vm_end, NULL if none. */ 20471da177e4SLinus Torvalds struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr) 20481da177e4SLinus Torvalds { 2049615d6e87SDavidlohr Bueso struct rb_node *rb_node; 2050615d6e87SDavidlohr Bueso struct vm_area_struct *vma; 20511da177e4SLinus Torvalds 20521da177e4SLinus Torvalds /* Check the cache first. */ 2053615d6e87SDavidlohr Bueso vma = vmacache_find(mm, addr); 2054615d6e87SDavidlohr Bueso if (likely(vma)) 2055615d6e87SDavidlohr Bueso return vma; 20561da177e4SLinus Torvalds 20571da177e4SLinus Torvalds rb_node = mm->mm_rb.rb_node; 20581da177e4SLinus Torvalds 20591da177e4SLinus Torvalds while (rb_node) { 2060615d6e87SDavidlohr Bueso struct vm_area_struct *tmp; 20611da177e4SLinus Torvalds 2062615d6e87SDavidlohr Bueso tmp = rb_entry(rb_node, struct vm_area_struct, vm_rb); 20631da177e4SLinus Torvalds 2064615d6e87SDavidlohr Bueso if (tmp->vm_end > addr) { 2065615d6e87SDavidlohr Bueso vma = tmp; 2066615d6e87SDavidlohr Bueso if (tmp->vm_start <= addr) 20671da177e4SLinus Torvalds break; 20681da177e4SLinus Torvalds rb_node = rb_node->rb_left; 20691da177e4SLinus Torvalds } else 20701da177e4SLinus Torvalds rb_node = rb_node->rb_right; 20711da177e4SLinus Torvalds } 2072615d6e87SDavidlohr Bueso 20731da177e4SLinus Torvalds if (vma) 2074615d6e87SDavidlohr Bueso vmacache_update(addr, vma); 20751da177e4SLinus Torvalds return vma; 20761da177e4SLinus Torvalds } 20771da177e4SLinus Torvalds 20781da177e4SLinus Torvalds EXPORT_SYMBOL(find_vma); 20791da177e4SLinus Torvalds 20806bd4837dSKOSAKI Motohiro /* 20816bd4837dSKOSAKI Motohiro * Same as find_vma, but also return a pointer to the previous VMA in *pprev. 20826bd4837dSKOSAKI Motohiro */ 20831da177e4SLinus Torvalds struct vm_area_struct * 20841da177e4SLinus Torvalds find_vma_prev(struct mm_struct *mm, unsigned long addr, 20851da177e4SLinus Torvalds struct vm_area_struct **pprev) 20861da177e4SLinus Torvalds { 20876bd4837dSKOSAKI Motohiro struct vm_area_struct *vma; 20881da177e4SLinus Torvalds 20896bd4837dSKOSAKI Motohiro vma = find_vma(mm, addr); 209083cd904dSMikulas Patocka if (vma) { 209183cd904dSMikulas Patocka *pprev = vma->vm_prev; 209283cd904dSMikulas Patocka } else { 209383cd904dSMikulas Patocka struct rb_node *rb_node = mm->mm_rb.rb_node; 209483cd904dSMikulas Patocka *pprev = NULL; 209583cd904dSMikulas Patocka while (rb_node) { 209683cd904dSMikulas Patocka *pprev = rb_entry(rb_node, struct vm_area_struct, vm_rb); 209783cd904dSMikulas Patocka rb_node = rb_node->rb_right; 209883cd904dSMikulas Patocka } 209983cd904dSMikulas Patocka } 21006bd4837dSKOSAKI Motohiro return vma; 21011da177e4SLinus Torvalds } 21021da177e4SLinus Torvalds 21031da177e4SLinus Torvalds /* 21041da177e4SLinus Torvalds * Verify that the stack growth is acceptable and 21051da177e4SLinus Torvalds * update accounting. This is shared with both the 21061da177e4SLinus Torvalds * grow-up and grow-down cases. 21071da177e4SLinus Torvalds */ 21081da177e4SLinus Torvalds static int acct_stack_growth(struct vm_area_struct *vma, unsigned long size, unsigned long grow) 21091da177e4SLinus Torvalds { 21101da177e4SLinus Torvalds struct mm_struct *mm = vma->vm_mm; 21111da177e4SLinus Torvalds struct rlimit *rlim = current->signal->rlim; 2112690eac53SLinus Torvalds unsigned long new_start, actual_size; 21131da177e4SLinus Torvalds 21141da177e4SLinus Torvalds /* address space limit tests */ 211584638335SKonstantin Khlebnikov if (!may_expand_vm(mm, vma->vm_flags, grow)) 21161da177e4SLinus Torvalds return -ENOMEM; 21171da177e4SLinus Torvalds 21181da177e4SLinus Torvalds /* Stack limit test */ 2119690eac53SLinus Torvalds actual_size = size; 2120690eac53SLinus Torvalds if (size && (vma->vm_flags & (VM_GROWSUP | VM_GROWSDOWN))) 2121690eac53SLinus Torvalds actual_size -= PAGE_SIZE; 21224db0c3c2SJason Low if (actual_size > READ_ONCE(rlim[RLIMIT_STACK].rlim_cur)) 21231da177e4SLinus Torvalds return -ENOMEM; 21241da177e4SLinus Torvalds 21251da177e4SLinus Torvalds /* mlock limit tests */ 21261da177e4SLinus Torvalds if (vma->vm_flags & VM_LOCKED) { 21271da177e4SLinus Torvalds unsigned long locked; 21281da177e4SLinus Torvalds unsigned long limit; 21291da177e4SLinus Torvalds locked = mm->locked_vm + grow; 21304db0c3c2SJason Low limit = READ_ONCE(rlim[RLIMIT_MEMLOCK].rlim_cur); 213159e99e5bSJiri Slaby limit >>= PAGE_SHIFT; 21321da177e4SLinus Torvalds if (locked > limit && !capable(CAP_IPC_LOCK)) 21331da177e4SLinus Torvalds return -ENOMEM; 21341da177e4SLinus Torvalds } 21351da177e4SLinus Torvalds 21360d59a01bSAdam Litke /* Check to ensure the stack will not grow into a hugetlb-only region */ 21370d59a01bSAdam Litke new_start = (vma->vm_flags & VM_GROWSUP) ? vma->vm_start : 21380d59a01bSAdam Litke vma->vm_end - size; 21390d59a01bSAdam Litke if (is_hugepage_only_range(vma->vm_mm, new_start, size)) 21400d59a01bSAdam Litke return -EFAULT; 21410d59a01bSAdam Litke 21421da177e4SLinus Torvalds /* 21431da177e4SLinus Torvalds * Overcommit.. This must be the final test, as it will 21441da177e4SLinus Torvalds * update security statistics. 21451da177e4SLinus Torvalds */ 214605fa199dSHugh Dickins if (security_vm_enough_memory_mm(mm, grow)) 21471da177e4SLinus Torvalds return -ENOMEM; 21481da177e4SLinus Torvalds 21491da177e4SLinus Torvalds return 0; 21501da177e4SLinus Torvalds } 21511da177e4SLinus Torvalds 215246dea3d0SHugh Dickins #if defined(CONFIG_STACK_GROWSUP) || defined(CONFIG_IA64) 21531da177e4SLinus Torvalds /* 215446dea3d0SHugh Dickins * PA-RISC uses this for its stack; IA64 for its Register Backing Store. 215546dea3d0SHugh Dickins * vma is the last one with address > vma->vm_end. Have to extend vma. 21561da177e4SLinus Torvalds */ 215746dea3d0SHugh Dickins int expand_upwards(struct vm_area_struct *vma, unsigned long address) 21581da177e4SLinus Torvalds { 215909357814SOleg Nesterov struct mm_struct *mm = vma->vm_mm; 216012352d3cSKonstantin Khlebnikov int error = 0; 21611da177e4SLinus Torvalds 21621da177e4SLinus Torvalds if (!(vma->vm_flags & VM_GROWSUP)) 21631da177e4SLinus Torvalds return -EFAULT; 21641da177e4SLinus Torvalds 216512352d3cSKonstantin Khlebnikov /* Guard against wrapping around to address 0. */ 216612352d3cSKonstantin Khlebnikov if (address < PAGE_ALIGN(address+4)) 216712352d3cSKonstantin Khlebnikov address = PAGE_ALIGN(address+4); 216812352d3cSKonstantin Khlebnikov else 216912352d3cSKonstantin Khlebnikov return -ENOMEM; 217012352d3cSKonstantin Khlebnikov 217112352d3cSKonstantin Khlebnikov /* We must make sure the anon_vma is allocated. */ 21721da177e4SLinus Torvalds if (unlikely(anon_vma_prepare(vma))) 21731da177e4SLinus Torvalds return -ENOMEM; 21741da177e4SLinus Torvalds 21751da177e4SLinus Torvalds /* 21761da177e4SLinus Torvalds * vma->vm_start/vm_end cannot change under us because the caller 21771da177e4SLinus Torvalds * is required to hold the mmap_sem in read mode. We need the 21781da177e4SLinus Torvalds * anon_vma lock to serialize against concurrent expand_stacks. 21791da177e4SLinus Torvalds */ 218012352d3cSKonstantin Khlebnikov anon_vma_lock_write(vma->anon_vma); 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. 219812352d3cSKonstantin Khlebnikov * anon_vma_lock_write() 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 */ 220409357814SOleg Nesterov spin_lock(&mm->page_table_lock); 220587e8827bSOleg Nesterov if (vma->vm_flags & VM_LOCKED) 220609357814SOleg Nesterov mm->locked_vm += grow; 220784638335SKonstantin Khlebnikov vm_stat_account(mm, vma->vm_flags, grow); 2208bf181b9fSMichel Lespinasse anon_vma_interval_tree_pre_update_vma(vma); 22091da177e4SLinus Torvalds vma->vm_end = address; 2210bf181b9fSMichel Lespinasse anon_vma_interval_tree_post_update_vma(vma); 2211d3737187SMichel Lespinasse if (vma->vm_next) 2212d3737187SMichel Lespinasse vma_gap_update(vma->vm_next); 2213d3737187SMichel Lespinasse else 221409357814SOleg Nesterov mm->highest_vm_end = address; 221509357814SOleg Nesterov spin_unlock(&mm->page_table_lock); 22164128997bSMichel Lespinasse 22173af9e859SEric B Munson perf_event_mmap(vma); 22183af9e859SEric B Munson } 22191da177e4SLinus Torvalds } 222042c36f63SHugh Dickins } 222112352d3cSKonstantin Khlebnikov anon_vma_unlock_write(vma->anon_vma); 22226d50e60cSDavid Rientjes khugepaged_enter_vma_merge(vma, vma->vm_flags); 222309357814SOleg Nesterov validate_mm(mm); 22241da177e4SLinus Torvalds return error; 22251da177e4SLinus Torvalds } 222646dea3d0SHugh Dickins #endif /* CONFIG_STACK_GROWSUP || CONFIG_IA64 */ 222746dea3d0SHugh Dickins 22281da177e4SLinus Torvalds /* 22291da177e4SLinus Torvalds * vma is the first one with address < vma->vm_start. Have to extend vma. 22301da177e4SLinus Torvalds */ 2231d05f3169SMichal Hocko int expand_downwards(struct vm_area_struct *vma, 2232b6a2fea3SOllie Wild unsigned long address) 22331da177e4SLinus Torvalds { 223409357814SOleg Nesterov struct mm_struct *mm = vma->vm_mm; 22351da177e4SLinus Torvalds int error; 22361da177e4SLinus Torvalds 22378869477aSEric Paris address &= PAGE_MASK; 2238e5467859SAl Viro error = security_mmap_addr(address); 22398869477aSEric Paris if (error) 22408869477aSEric Paris return error; 22418869477aSEric Paris 224212352d3cSKonstantin Khlebnikov /* We must make sure the anon_vma is allocated. */ 224312352d3cSKonstantin Khlebnikov if (unlikely(anon_vma_prepare(vma))) 224412352d3cSKonstantin Khlebnikov return -ENOMEM; 22451da177e4SLinus Torvalds 22461da177e4SLinus Torvalds /* 22471da177e4SLinus Torvalds * vma->vm_start/vm_end cannot change under us because the caller 22481da177e4SLinus Torvalds * is required to hold the mmap_sem in read mode. We need the 22491da177e4SLinus Torvalds * anon_vma lock to serialize against concurrent expand_stacks. 22501da177e4SLinus Torvalds */ 225112352d3cSKonstantin Khlebnikov anon_vma_lock_write(vma->anon_vma); 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. 226912352d3cSKonstantin Khlebnikov * anon_vma_lock_write() 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 */ 227509357814SOleg Nesterov spin_lock(&mm->page_table_lock); 227687e8827bSOleg Nesterov if (vma->vm_flags & VM_LOCKED) 227709357814SOleg Nesterov mm->locked_vm += grow; 227884638335SKonstantin Khlebnikov vm_stat_account(mm, vma->vm_flags, grow); 2279bf181b9fSMichel Lespinasse anon_vma_interval_tree_pre_update_vma(vma); 22801da177e4SLinus Torvalds vma->vm_start = address; 22811da177e4SLinus Torvalds vma->vm_pgoff -= grow; 2282bf181b9fSMichel Lespinasse anon_vma_interval_tree_post_update_vma(vma); 2283d3737187SMichel Lespinasse vma_gap_update(vma); 228409357814SOleg Nesterov spin_unlock(&mm->page_table_lock); 22854128997bSMichel Lespinasse 22863af9e859SEric B Munson perf_event_mmap(vma); 22871da177e4SLinus Torvalds } 22881da177e4SLinus Torvalds } 2289a626ca6aSLinus Torvalds } 229012352d3cSKonstantin Khlebnikov anon_vma_unlock_write(vma->anon_vma); 22916d50e60cSDavid Rientjes khugepaged_enter_vma_merge(vma, vma->vm_flags); 229209357814SOleg Nesterov validate_mm(mm); 22931da177e4SLinus Torvalds return error; 22941da177e4SLinus Torvalds } 22951da177e4SLinus Torvalds 229609884964SLinus Torvalds /* 229709884964SLinus Torvalds * Note how expand_stack() refuses to expand the stack all the way to 229809884964SLinus Torvalds * abut the next virtual mapping, *unless* that mapping itself is also 229909884964SLinus Torvalds * a stack mapping. We want to leave room for a guard page, after all 230009884964SLinus Torvalds * (the guard page itself is not added here, that is done by the 230109884964SLinus Torvalds * actual page faulting logic) 230209884964SLinus Torvalds * 230309884964SLinus Torvalds * This matches the behavior of the guard page logic (see mm/memory.c: 230409884964SLinus Torvalds * check_stack_guard_page()), which only allows the guard page to be 230509884964SLinus Torvalds * removed under these circumstances. 230609884964SLinus Torvalds */ 2307b6a2fea3SOllie Wild #ifdef CONFIG_STACK_GROWSUP 2308b6a2fea3SOllie Wild int expand_stack(struct vm_area_struct *vma, unsigned long address) 2309b6a2fea3SOllie Wild { 231009884964SLinus Torvalds struct vm_area_struct *next; 231109884964SLinus Torvalds 231209884964SLinus Torvalds address &= PAGE_MASK; 231309884964SLinus Torvalds next = vma->vm_next; 231409884964SLinus Torvalds if (next && next->vm_start == address + PAGE_SIZE) { 231509884964SLinus Torvalds if (!(next->vm_flags & VM_GROWSUP)) 231609884964SLinus Torvalds return -ENOMEM; 231709884964SLinus Torvalds } 2318b6a2fea3SOllie Wild return expand_upwards(vma, address); 2319b6a2fea3SOllie Wild } 2320b6a2fea3SOllie Wild 2321b6a2fea3SOllie Wild struct vm_area_struct * 2322b6a2fea3SOllie Wild find_extend_vma(struct mm_struct *mm, unsigned long addr) 2323b6a2fea3SOllie Wild { 2324b6a2fea3SOllie Wild struct vm_area_struct *vma, *prev; 2325b6a2fea3SOllie Wild 2326b6a2fea3SOllie Wild addr &= PAGE_MASK; 2327b6a2fea3SOllie Wild vma = find_vma_prev(mm, addr, &prev); 2328b6a2fea3SOllie Wild if (vma && (vma->vm_start <= addr)) 2329b6a2fea3SOllie Wild return vma; 23301c127185SDenys Vlasenko if (!prev || expand_stack(prev, addr)) 2331b6a2fea3SOllie Wild return NULL; 2332cea10a19SMichel Lespinasse if (prev->vm_flags & VM_LOCKED) 2333fc05f566SKirill A. Shutemov populate_vma_page_range(prev, addr, prev->vm_end, NULL); 2334b6a2fea3SOllie Wild return prev; 2335b6a2fea3SOllie Wild } 2336b6a2fea3SOllie Wild #else 2337b6a2fea3SOllie Wild int expand_stack(struct vm_area_struct *vma, unsigned long address) 2338b6a2fea3SOllie Wild { 233909884964SLinus Torvalds struct vm_area_struct *prev; 234009884964SLinus Torvalds 234109884964SLinus Torvalds address &= PAGE_MASK; 234209884964SLinus Torvalds prev = vma->vm_prev; 234309884964SLinus Torvalds if (prev && prev->vm_end == address) { 234409884964SLinus Torvalds if (!(prev->vm_flags & VM_GROWSDOWN)) 234509884964SLinus Torvalds return -ENOMEM; 234609884964SLinus Torvalds } 2347b6a2fea3SOllie Wild return expand_downwards(vma, address); 2348b6a2fea3SOllie Wild } 2349b6a2fea3SOllie Wild 23501da177e4SLinus Torvalds struct vm_area_struct * 23511da177e4SLinus Torvalds find_extend_vma(struct mm_struct *mm, unsigned long addr) 23521da177e4SLinus Torvalds { 23531da177e4SLinus Torvalds struct vm_area_struct *vma; 23541da177e4SLinus Torvalds unsigned long start; 23551da177e4SLinus Torvalds 23561da177e4SLinus Torvalds addr &= PAGE_MASK; 23571da177e4SLinus Torvalds vma = find_vma(mm, addr); 23581da177e4SLinus Torvalds if (!vma) 23591da177e4SLinus Torvalds return NULL; 23601da177e4SLinus Torvalds if (vma->vm_start <= addr) 23611da177e4SLinus Torvalds return vma; 23621da177e4SLinus Torvalds if (!(vma->vm_flags & VM_GROWSDOWN)) 23631da177e4SLinus Torvalds return NULL; 23641da177e4SLinus Torvalds start = vma->vm_start; 23651da177e4SLinus Torvalds if (expand_stack(vma, addr)) 23661da177e4SLinus Torvalds return NULL; 2367cea10a19SMichel Lespinasse if (vma->vm_flags & VM_LOCKED) 2368fc05f566SKirill A. Shutemov populate_vma_page_range(vma, addr, start, NULL); 23691da177e4SLinus Torvalds return vma; 23701da177e4SLinus Torvalds } 23711da177e4SLinus Torvalds #endif 23721da177e4SLinus Torvalds 2373e1d6d01aSJesse Barnes EXPORT_SYMBOL_GPL(find_extend_vma); 2374e1d6d01aSJesse Barnes 23752c0b3814SHugh Dickins /* 23762c0b3814SHugh Dickins * Ok - we have the memory areas we should free on the vma list, 23772c0b3814SHugh Dickins * so release them, and do the vma updates. 23781da177e4SLinus Torvalds * 23792c0b3814SHugh Dickins * Called with the mm semaphore held. 23801da177e4SLinus Torvalds */ 23812c0b3814SHugh Dickins static void remove_vma_list(struct mm_struct *mm, struct vm_area_struct *vma) 23821da177e4SLinus Torvalds { 23834f74d2c8SLinus Torvalds unsigned long nr_accounted = 0; 23844f74d2c8SLinus Torvalds 2385365e9c87SHugh Dickins /* Update high watermark before we lower total_vm */ 2386365e9c87SHugh Dickins update_hiwater_vm(mm); 23872c0b3814SHugh Dickins do { 2388ab50b8edSHugh Dickins long nrpages = vma_pages(vma); 23891da177e4SLinus Torvalds 23904f74d2c8SLinus Torvalds if (vma->vm_flags & VM_ACCOUNT) 23914f74d2c8SLinus Torvalds nr_accounted += nrpages; 239284638335SKonstantin Khlebnikov vm_stat_account(mm, vma->vm_flags, -nrpages); 2393a8fb5618SHugh Dickins vma = remove_vma(vma); 2394146425a3SHugh Dickins } while (vma); 23954f74d2c8SLinus Torvalds vm_unacct_memory(nr_accounted); 23961da177e4SLinus Torvalds validate_mm(mm); 23971da177e4SLinus Torvalds } 23981da177e4SLinus Torvalds 23991da177e4SLinus Torvalds /* 24001da177e4SLinus Torvalds * Get rid of page table information in the indicated region. 24011da177e4SLinus Torvalds * 2402f10df686SPaolo 'Blaisorblade' Giarrusso * Called with the mm semaphore held. 24031da177e4SLinus Torvalds */ 24041da177e4SLinus Torvalds static void unmap_region(struct mm_struct *mm, 2405e0da382cSHugh Dickins struct vm_area_struct *vma, struct vm_area_struct *prev, 2406e0da382cSHugh Dickins unsigned long start, unsigned long end) 24071da177e4SLinus Torvalds { 2408e0da382cSHugh Dickins struct vm_area_struct *next = prev ? prev->vm_next : mm->mmap; 2409d16dfc55SPeter Zijlstra struct mmu_gather tlb; 24101da177e4SLinus Torvalds 24111da177e4SLinus Torvalds lru_add_drain(); 24122b047252SLinus Torvalds tlb_gather_mmu(&tlb, mm, start, end); 2413365e9c87SHugh Dickins update_hiwater_rss(mm); 24144f74d2c8SLinus Torvalds unmap_vmas(&tlb, vma, start, end); 2415d16dfc55SPeter Zijlstra free_pgtables(&tlb, vma, prev ? prev->vm_end : FIRST_USER_ADDRESS, 24166ee8630eSHugh Dickins next ? next->vm_start : USER_PGTABLES_CEILING); 2417d16dfc55SPeter Zijlstra tlb_finish_mmu(&tlb, start, end); 24181da177e4SLinus Torvalds } 24191da177e4SLinus Torvalds 24201da177e4SLinus Torvalds /* 24211da177e4SLinus Torvalds * Create a list of vma's touched by the unmap, removing them from the mm's 24221da177e4SLinus Torvalds * vma list as we go.. 24231da177e4SLinus Torvalds */ 24241da177e4SLinus Torvalds static void 24251da177e4SLinus Torvalds detach_vmas_to_be_unmapped(struct mm_struct *mm, struct vm_area_struct *vma, 24261da177e4SLinus Torvalds struct vm_area_struct *prev, unsigned long end) 24271da177e4SLinus Torvalds { 24281da177e4SLinus Torvalds struct vm_area_struct **insertion_point; 24291da177e4SLinus Torvalds struct vm_area_struct *tail_vma = NULL; 24301da177e4SLinus Torvalds 24311da177e4SLinus Torvalds insertion_point = (prev ? &prev->vm_next : &mm->mmap); 2432297c5eeeSLinus Torvalds vma->vm_prev = NULL; 24331da177e4SLinus Torvalds do { 2434d3737187SMichel Lespinasse vma_rb_erase(vma, &mm->mm_rb); 24351da177e4SLinus Torvalds mm->map_count--; 24361da177e4SLinus Torvalds tail_vma = vma; 24371da177e4SLinus Torvalds vma = vma->vm_next; 24381da177e4SLinus Torvalds } while (vma && vma->vm_start < end); 24391da177e4SLinus Torvalds *insertion_point = vma; 2440d3737187SMichel Lespinasse if (vma) { 2441297c5eeeSLinus Torvalds vma->vm_prev = prev; 2442d3737187SMichel Lespinasse vma_gap_update(vma); 2443d3737187SMichel Lespinasse } else 2444d3737187SMichel Lespinasse mm->highest_vm_end = prev ? prev->vm_end : 0; 24451da177e4SLinus Torvalds tail_vma->vm_next = NULL; 2446615d6e87SDavidlohr Bueso 2447615d6e87SDavidlohr Bueso /* Kill the cache */ 2448615d6e87SDavidlohr Bueso vmacache_invalidate(mm); 24491da177e4SLinus Torvalds } 24501da177e4SLinus Torvalds 24511da177e4SLinus Torvalds /* 2452659ace58SKOSAKI Motohiro * __split_vma() bypasses sysctl_max_map_count checking. We use this on the 2453659ace58SKOSAKI Motohiro * munmap path where it doesn't make sense to fail. 24541da177e4SLinus Torvalds */ 2455659ace58SKOSAKI Motohiro static int __split_vma(struct mm_struct *mm, struct vm_area_struct *vma, 24561da177e4SLinus Torvalds unsigned long addr, int new_below) 24571da177e4SLinus Torvalds { 24581da177e4SLinus Torvalds struct vm_area_struct *new; 2459e3975891SChen Gang int err; 24601da177e4SLinus Torvalds 2461a5516438SAndi Kleen if (is_vm_hugetlb_page(vma) && (addr & 2462a5516438SAndi Kleen ~(huge_page_mask(hstate_vma(vma))))) 24631da177e4SLinus Torvalds return -EINVAL; 24641da177e4SLinus Torvalds 2465e94b1766SChristoph Lameter new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL); 24661da177e4SLinus Torvalds if (!new) 2467e3975891SChen Gang return -ENOMEM; 24681da177e4SLinus Torvalds 24691da177e4SLinus Torvalds /* most fields are the same, copy all, and then fixup */ 24701da177e4SLinus Torvalds *new = *vma; 24711da177e4SLinus Torvalds 24725beb4930SRik van Riel INIT_LIST_HEAD(&new->anon_vma_chain); 24735beb4930SRik van Riel 24741da177e4SLinus Torvalds if (new_below) 24751da177e4SLinus Torvalds new->vm_end = addr; 24761da177e4SLinus Torvalds else { 24771da177e4SLinus Torvalds new->vm_start = addr; 24781da177e4SLinus Torvalds new->vm_pgoff += ((addr - vma->vm_start) >> PAGE_SHIFT); 24791da177e4SLinus Torvalds } 24801da177e4SLinus Torvalds 2481ef0855d3SOleg Nesterov err = vma_dup_policy(vma, new); 2482ef0855d3SOleg Nesterov if (err) 24835beb4930SRik van Riel goto out_free_vma; 24841da177e4SLinus Torvalds 2485c4ea95d7SDaniel Forrest err = anon_vma_clone(new, vma); 2486c4ea95d7SDaniel Forrest if (err) 24875beb4930SRik van Riel goto out_free_mpol; 24885beb4930SRik van Riel 2489e9714acfSKonstantin Khlebnikov if (new->vm_file) 24901da177e4SLinus Torvalds get_file(new->vm_file); 24911da177e4SLinus Torvalds 24921da177e4SLinus Torvalds if (new->vm_ops && new->vm_ops->open) 24931da177e4SLinus Torvalds new->vm_ops->open(new); 24941da177e4SLinus Torvalds 24951da177e4SLinus Torvalds if (new_below) 24965beb4930SRik van Riel err = vma_adjust(vma, addr, vma->vm_end, vma->vm_pgoff + 24971da177e4SLinus Torvalds ((addr - new->vm_start) >> PAGE_SHIFT), new); 24981da177e4SLinus Torvalds else 24995beb4930SRik van Riel err = vma_adjust(vma, vma->vm_start, addr, vma->vm_pgoff, new); 25001da177e4SLinus Torvalds 25015beb4930SRik van Riel /* Success. */ 25025beb4930SRik van Riel if (!err) 25031da177e4SLinus Torvalds return 0; 25045beb4930SRik van Riel 25055beb4930SRik van Riel /* Clean everything up if vma_adjust failed. */ 250658927533SRik van Riel if (new->vm_ops && new->vm_ops->close) 25075beb4930SRik van Riel new->vm_ops->close(new); 2508e9714acfSKonstantin Khlebnikov if (new->vm_file) 25095beb4930SRik van Riel fput(new->vm_file); 25102aeadc30SAndrea Arcangeli unlink_anon_vmas(new); 25115beb4930SRik van Riel out_free_mpol: 2512ef0855d3SOleg Nesterov mpol_put(vma_policy(new)); 25135beb4930SRik van Riel out_free_vma: 25145beb4930SRik van Riel kmem_cache_free(vm_area_cachep, new); 25155beb4930SRik van Riel return err; 25161da177e4SLinus Torvalds } 25171da177e4SLinus Torvalds 2518659ace58SKOSAKI Motohiro /* 2519659ace58SKOSAKI Motohiro * Split a vma into two pieces at address 'addr', a new vma is allocated 2520659ace58SKOSAKI Motohiro * either for the first part or the tail. 2521659ace58SKOSAKI Motohiro */ 2522659ace58SKOSAKI Motohiro int split_vma(struct mm_struct *mm, struct vm_area_struct *vma, 2523659ace58SKOSAKI Motohiro unsigned long addr, int new_below) 2524659ace58SKOSAKI Motohiro { 2525659ace58SKOSAKI Motohiro if (mm->map_count >= sysctl_max_map_count) 2526659ace58SKOSAKI Motohiro return -ENOMEM; 2527659ace58SKOSAKI Motohiro 2528659ace58SKOSAKI Motohiro return __split_vma(mm, vma, addr, new_below); 2529659ace58SKOSAKI Motohiro } 2530659ace58SKOSAKI Motohiro 25311da177e4SLinus Torvalds /* Munmap is split into 2 main parts -- this part which finds 25321da177e4SLinus Torvalds * what needs doing, and the areas themselves, which do the 25331da177e4SLinus Torvalds * work. This now handles partial unmappings. 25341da177e4SLinus Torvalds * Jeremy Fitzhardinge <jeremy@goop.org> 25351da177e4SLinus Torvalds */ 25361da177e4SLinus Torvalds int do_munmap(struct mm_struct *mm, unsigned long start, size_t len) 25371da177e4SLinus Torvalds { 25381da177e4SLinus Torvalds unsigned long end; 2539146425a3SHugh Dickins struct vm_area_struct *vma, *prev, *last; 25401da177e4SLinus Torvalds 2541de1741a1SAlexander Kuleshov if ((offset_in_page(start)) || start > TASK_SIZE || len > TASK_SIZE-start) 25421da177e4SLinus Torvalds return -EINVAL; 25431da177e4SLinus Torvalds 2544cc71aba3Svishnu.ps len = PAGE_ALIGN(len); 2545cc71aba3Svishnu.ps if (len == 0) 25461da177e4SLinus Torvalds return -EINVAL; 25471da177e4SLinus Torvalds 25481da177e4SLinus Torvalds /* Find the first overlapping VMA */ 25499be34c9dSLinus Torvalds vma = find_vma(mm, start); 2550146425a3SHugh Dickins if (!vma) 25511da177e4SLinus Torvalds return 0; 25529be34c9dSLinus Torvalds prev = vma->vm_prev; 2553146425a3SHugh Dickins /* we have start < vma->vm_end */ 25541da177e4SLinus Torvalds 25551da177e4SLinus Torvalds /* if it doesn't overlap, we have nothing.. */ 25561da177e4SLinus Torvalds end = start + len; 2557146425a3SHugh Dickins if (vma->vm_start >= end) 25581da177e4SLinus Torvalds return 0; 25591da177e4SLinus Torvalds 25601da177e4SLinus Torvalds /* 25611da177e4SLinus Torvalds * If we need to split any vma, do it now to save pain later. 25621da177e4SLinus Torvalds * 25631da177e4SLinus Torvalds * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially 25641da177e4SLinus Torvalds * unmapped vm_area_struct will remain in use: so lower split_vma 25651da177e4SLinus Torvalds * places tmp vma above, and higher split_vma places tmp vma below. 25661da177e4SLinus Torvalds */ 2567146425a3SHugh Dickins if (start > vma->vm_start) { 2568659ace58SKOSAKI Motohiro int error; 2569659ace58SKOSAKI Motohiro 2570659ace58SKOSAKI Motohiro /* 2571659ace58SKOSAKI Motohiro * Make sure that map_count on return from munmap() will 2572659ace58SKOSAKI Motohiro * not exceed its limit; but let map_count go just above 2573659ace58SKOSAKI Motohiro * its limit temporarily, to help free resources as expected. 2574659ace58SKOSAKI Motohiro */ 2575659ace58SKOSAKI Motohiro if (end < vma->vm_end && mm->map_count >= sysctl_max_map_count) 2576659ace58SKOSAKI Motohiro return -ENOMEM; 2577659ace58SKOSAKI Motohiro 2578659ace58SKOSAKI Motohiro error = __split_vma(mm, vma, start, 0); 25791da177e4SLinus Torvalds if (error) 25801da177e4SLinus Torvalds return error; 2581146425a3SHugh Dickins prev = vma; 25821da177e4SLinus Torvalds } 25831da177e4SLinus Torvalds 25841da177e4SLinus Torvalds /* Does it split the last one? */ 25851da177e4SLinus Torvalds last = find_vma(mm, end); 25861da177e4SLinus Torvalds if (last && end > last->vm_start) { 2587659ace58SKOSAKI Motohiro int error = __split_vma(mm, last, end, 1); 25881da177e4SLinus Torvalds if (error) 25891da177e4SLinus Torvalds return error; 25901da177e4SLinus Torvalds } 2591146425a3SHugh Dickins vma = prev ? prev->vm_next : mm->mmap; 25921da177e4SLinus Torvalds 25931da177e4SLinus Torvalds /* 2594ba470de4SRik van Riel * unlock any mlock()ed ranges before detaching vmas 2595ba470de4SRik van Riel */ 2596ba470de4SRik van Riel if (mm->locked_vm) { 2597ba470de4SRik van Riel struct vm_area_struct *tmp = vma; 2598ba470de4SRik van Riel while (tmp && tmp->vm_start < end) { 2599ba470de4SRik van Riel if (tmp->vm_flags & VM_LOCKED) { 2600ba470de4SRik van Riel mm->locked_vm -= vma_pages(tmp); 2601ba470de4SRik van Riel munlock_vma_pages_all(tmp); 2602ba470de4SRik van Riel } 2603ba470de4SRik van Riel tmp = tmp->vm_next; 2604ba470de4SRik van Riel } 2605ba470de4SRik van Riel } 2606ba470de4SRik van Riel 2607ba470de4SRik van Riel /* 26081da177e4SLinus Torvalds * Remove the vma's, and unmap the actual pages 26091da177e4SLinus Torvalds */ 2610146425a3SHugh Dickins detach_vmas_to_be_unmapped(mm, vma, prev, end); 2611146425a3SHugh Dickins unmap_region(mm, vma, prev, start, end); 26121da177e4SLinus Torvalds 26131de4fa14SDave Hansen arch_unmap(mm, vma, start, end); 26141de4fa14SDave Hansen 26151da177e4SLinus Torvalds /* Fix up all other VM information */ 26162c0b3814SHugh Dickins remove_vma_list(mm, vma); 26171da177e4SLinus Torvalds 26181da177e4SLinus Torvalds return 0; 26191da177e4SLinus Torvalds } 26201da177e4SLinus Torvalds 2621bfce281cSAl Viro int vm_munmap(unsigned long start, size_t len) 2622a46ef99dSLinus Torvalds { 2623a46ef99dSLinus Torvalds int ret; 2624bfce281cSAl Viro struct mm_struct *mm = current->mm; 2625a46ef99dSLinus Torvalds 2626a46ef99dSLinus Torvalds down_write(&mm->mmap_sem); 2627a46ef99dSLinus Torvalds ret = do_munmap(mm, start, len); 2628a46ef99dSLinus Torvalds up_write(&mm->mmap_sem); 2629a46ef99dSLinus Torvalds return ret; 2630a46ef99dSLinus Torvalds } 2631a46ef99dSLinus Torvalds EXPORT_SYMBOL(vm_munmap); 2632a46ef99dSLinus Torvalds 26336a6160a7SHeiko Carstens SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len) 26341da177e4SLinus Torvalds { 26351da177e4SLinus Torvalds profile_munmap(addr); 2636bfce281cSAl Viro return vm_munmap(addr, len); 26371da177e4SLinus Torvalds } 26381da177e4SLinus Torvalds 2639c8d78c18SKirill A. Shutemov 2640c8d78c18SKirill A. Shutemov /* 2641c8d78c18SKirill A. Shutemov * Emulation of deprecated remap_file_pages() syscall. 2642c8d78c18SKirill A. Shutemov */ 2643c8d78c18SKirill A. Shutemov SYSCALL_DEFINE5(remap_file_pages, unsigned long, start, unsigned long, size, 2644c8d78c18SKirill A. Shutemov unsigned long, prot, unsigned long, pgoff, unsigned long, flags) 2645c8d78c18SKirill A. Shutemov { 2646c8d78c18SKirill A. Shutemov 2647c8d78c18SKirill A. Shutemov struct mm_struct *mm = current->mm; 2648c8d78c18SKirill A. Shutemov struct vm_area_struct *vma; 2649c8d78c18SKirill A. Shutemov unsigned long populate = 0; 2650c8d78c18SKirill A. Shutemov unsigned long ret = -EINVAL; 2651c8d78c18SKirill A. Shutemov struct file *file; 2652c8d78c18SKirill A. Shutemov 2653c8d78c18SKirill A. Shutemov pr_warn_once("%s (%d) uses deprecated remap_file_pages() syscall. " 2654c8d78c18SKirill A. Shutemov "See Documentation/vm/remap_file_pages.txt.\n", 2655c8d78c18SKirill A. Shutemov current->comm, current->pid); 2656c8d78c18SKirill A. Shutemov 2657c8d78c18SKirill A. Shutemov if (prot) 2658c8d78c18SKirill A. Shutemov return ret; 2659c8d78c18SKirill A. Shutemov start = start & PAGE_MASK; 2660c8d78c18SKirill A. Shutemov size = size & PAGE_MASK; 2661c8d78c18SKirill A. Shutemov 2662c8d78c18SKirill A. Shutemov if (start + size <= start) 2663c8d78c18SKirill A. Shutemov return ret; 2664c8d78c18SKirill A. Shutemov 2665c8d78c18SKirill A. Shutemov /* Does pgoff wrap? */ 2666c8d78c18SKirill A. Shutemov if (pgoff + (size >> PAGE_SHIFT) < pgoff) 2667c8d78c18SKirill A. Shutemov return ret; 2668c8d78c18SKirill A. Shutemov 2669c8d78c18SKirill A. Shutemov down_write(&mm->mmap_sem); 2670c8d78c18SKirill A. Shutemov vma = find_vma(mm, start); 2671c8d78c18SKirill A. Shutemov 2672c8d78c18SKirill A. Shutemov if (!vma || !(vma->vm_flags & VM_SHARED)) 2673c8d78c18SKirill A. Shutemov goto out; 2674c8d78c18SKirill A. Shutemov 2675c8d78c18SKirill A. Shutemov if (start < vma->vm_start || start + size > vma->vm_end) 2676c8d78c18SKirill A. Shutemov goto out; 2677c8d78c18SKirill A. Shutemov 2678c8d78c18SKirill A. Shutemov if (pgoff == linear_page_index(vma, start)) { 2679c8d78c18SKirill A. Shutemov ret = 0; 2680c8d78c18SKirill A. Shutemov goto out; 2681c8d78c18SKirill A. Shutemov } 2682c8d78c18SKirill A. Shutemov 2683c8d78c18SKirill A. Shutemov prot |= vma->vm_flags & VM_READ ? PROT_READ : 0; 2684c8d78c18SKirill A. Shutemov prot |= vma->vm_flags & VM_WRITE ? PROT_WRITE : 0; 2685c8d78c18SKirill A. Shutemov prot |= vma->vm_flags & VM_EXEC ? PROT_EXEC : 0; 2686c8d78c18SKirill A. Shutemov 2687c8d78c18SKirill A. Shutemov flags &= MAP_NONBLOCK; 2688c8d78c18SKirill A. Shutemov flags |= MAP_SHARED | MAP_FIXED | MAP_POPULATE; 2689c8d78c18SKirill A. Shutemov if (vma->vm_flags & VM_LOCKED) { 2690c8d78c18SKirill A. Shutemov flags |= MAP_LOCKED; 2691c8d78c18SKirill A. Shutemov /* drop PG_Mlocked flag for over-mapped range */ 2692c8d78c18SKirill A. Shutemov munlock_vma_pages_range(vma, start, start + size); 2693c8d78c18SKirill A. Shutemov } 2694c8d78c18SKirill A. Shutemov 2695c8d78c18SKirill A. Shutemov file = get_file(vma->vm_file); 2696c8d78c18SKirill A. Shutemov ret = do_mmap_pgoff(vma->vm_file, start, size, 2697c8d78c18SKirill A. Shutemov prot, flags, pgoff, &populate); 2698c8d78c18SKirill A. Shutemov fput(file); 2699c8d78c18SKirill A. Shutemov out: 2700c8d78c18SKirill A. Shutemov up_write(&mm->mmap_sem); 2701c8d78c18SKirill A. Shutemov if (populate) 2702c8d78c18SKirill A. Shutemov mm_populate(ret, populate); 2703c8d78c18SKirill A. Shutemov if (!IS_ERR_VALUE(ret)) 2704c8d78c18SKirill A. Shutemov ret = 0; 2705c8d78c18SKirill A. Shutemov return ret; 2706c8d78c18SKirill A. Shutemov } 2707c8d78c18SKirill A. Shutemov 27081da177e4SLinus Torvalds static inline void verify_mm_writelocked(struct mm_struct *mm) 27091da177e4SLinus Torvalds { 2710a241ec65SPaul E. McKenney #ifdef CONFIG_DEBUG_VM 27111da177e4SLinus Torvalds if (unlikely(down_read_trylock(&mm->mmap_sem))) { 27121da177e4SLinus Torvalds WARN_ON(1); 27131da177e4SLinus Torvalds up_read(&mm->mmap_sem); 27141da177e4SLinus Torvalds } 27151da177e4SLinus Torvalds #endif 27161da177e4SLinus Torvalds } 27171da177e4SLinus Torvalds 27181da177e4SLinus Torvalds /* 27191da177e4SLinus Torvalds * this is really a simplified "do_mmap". it only handles 27201da177e4SLinus Torvalds * anonymous maps. eventually we may be able to do some 27211da177e4SLinus Torvalds * brk-specific accounting here. 27221da177e4SLinus Torvalds */ 2723e4eb1ff6SLinus Torvalds static unsigned long do_brk(unsigned long addr, unsigned long len) 27241da177e4SLinus Torvalds { 27251da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 27261da177e4SLinus Torvalds struct vm_area_struct *vma, *prev; 27271da177e4SLinus Torvalds unsigned long flags; 27281da177e4SLinus Torvalds struct rb_node **rb_link, *rb_parent; 27291da177e4SLinus Torvalds pgoff_t pgoff = addr >> PAGE_SHIFT; 27303a459756SKirill Korotaev int error; 27311da177e4SLinus Torvalds 27321da177e4SLinus Torvalds len = PAGE_ALIGN(len); 27331da177e4SLinus Torvalds if (!len) 27341da177e4SLinus Torvalds return addr; 27351da177e4SLinus Torvalds 27363a459756SKirill Korotaev flags = VM_DATA_DEFAULT_FLAGS | VM_ACCOUNT | mm->def_flags; 27373a459756SKirill Korotaev 27382c6a1016SAl Viro error = get_unmapped_area(NULL, addr, len, 0, MAP_FIXED); 2739de1741a1SAlexander Kuleshov if (offset_in_page(error)) 27403a459756SKirill Korotaev return error; 27413a459756SKirill Korotaev 2742363ee17fSDavidlohr Bueso error = mlock_future_check(mm, mm->def_flags, len); 2743363ee17fSDavidlohr Bueso if (error) 2744363ee17fSDavidlohr Bueso return error; 27451da177e4SLinus Torvalds 27461da177e4SLinus Torvalds /* 27471da177e4SLinus Torvalds * mm->mmap_sem is required to protect against another thread 27481da177e4SLinus Torvalds * changing the mappings in case we sleep. 27491da177e4SLinus Torvalds */ 27501da177e4SLinus Torvalds verify_mm_writelocked(mm); 27511da177e4SLinus Torvalds 27521da177e4SLinus Torvalds /* 27531da177e4SLinus Torvalds * Clear old maps. this also does some error checking for us 27541da177e4SLinus Torvalds */ 27559fcd1457SRasmus Villemoes while (find_vma_links(mm, addr, addr + len, &prev, &rb_link, 27569fcd1457SRasmus Villemoes &rb_parent)) { 27571da177e4SLinus Torvalds if (do_munmap(mm, addr, len)) 27581da177e4SLinus Torvalds return -ENOMEM; 27591da177e4SLinus Torvalds } 27601da177e4SLinus Torvalds 27611da177e4SLinus Torvalds /* Check against address space limits *after* clearing old maps... */ 276284638335SKonstantin Khlebnikov if (!may_expand_vm(mm, flags, len >> PAGE_SHIFT)) 27631da177e4SLinus Torvalds return -ENOMEM; 27641da177e4SLinus Torvalds 27651da177e4SLinus Torvalds if (mm->map_count > sysctl_max_map_count) 27661da177e4SLinus Torvalds return -ENOMEM; 27671da177e4SLinus Torvalds 2768191c5424SAl Viro if (security_vm_enough_memory_mm(mm, len >> PAGE_SHIFT)) 27691da177e4SLinus Torvalds return -ENOMEM; 27701da177e4SLinus Torvalds 27711da177e4SLinus Torvalds /* Can we just expand an old private anonymous mapping? */ 2772ba470de4SRik van Riel vma = vma_merge(mm, prev, addr, addr + len, flags, 277319a809afSAndrea Arcangeli NULL, NULL, pgoff, NULL, NULL_VM_UFFD_CTX); 2774ba470de4SRik van Riel if (vma) 27751da177e4SLinus Torvalds goto out; 27761da177e4SLinus Torvalds 27771da177e4SLinus Torvalds /* 27781da177e4SLinus Torvalds * create a vma struct for an anonymous mapping 27791da177e4SLinus Torvalds */ 2780c5e3b83eSPekka Enberg vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL); 27811da177e4SLinus Torvalds if (!vma) { 27821da177e4SLinus Torvalds vm_unacct_memory(len >> PAGE_SHIFT); 27831da177e4SLinus Torvalds return -ENOMEM; 27841da177e4SLinus Torvalds } 27851da177e4SLinus Torvalds 27865beb4930SRik van Riel INIT_LIST_HEAD(&vma->anon_vma_chain); 27871da177e4SLinus Torvalds vma->vm_mm = mm; 27881da177e4SLinus Torvalds vma->vm_start = addr; 27891da177e4SLinus Torvalds vma->vm_end = addr + len; 27901da177e4SLinus Torvalds vma->vm_pgoff = pgoff; 27911da177e4SLinus Torvalds vma->vm_flags = flags; 27923ed75eb8SColy Li vma->vm_page_prot = vm_get_page_prot(flags); 27931da177e4SLinus Torvalds vma_link(mm, vma, prev, rb_link, rb_parent); 27941da177e4SLinus Torvalds out: 27953af9e859SEric B Munson perf_event_mmap(vma); 27961da177e4SLinus Torvalds mm->total_vm += len >> PAGE_SHIFT; 279784638335SKonstantin Khlebnikov mm->data_vm += len >> PAGE_SHIFT; 2798128557ffSMichel Lespinasse if (flags & VM_LOCKED) 2799ba470de4SRik van Riel mm->locked_vm += (len >> PAGE_SHIFT); 2800d9104d1cSCyrill Gorcunov vma->vm_flags |= VM_SOFTDIRTY; 28011da177e4SLinus Torvalds return addr; 28021da177e4SLinus Torvalds } 28031da177e4SLinus Torvalds 2804e4eb1ff6SLinus Torvalds unsigned long vm_brk(unsigned long addr, unsigned long len) 2805e4eb1ff6SLinus Torvalds { 2806e4eb1ff6SLinus Torvalds struct mm_struct *mm = current->mm; 2807e4eb1ff6SLinus Torvalds unsigned long ret; 2808128557ffSMichel Lespinasse bool populate; 2809e4eb1ff6SLinus Torvalds 2810e4eb1ff6SLinus Torvalds down_write(&mm->mmap_sem); 2811e4eb1ff6SLinus Torvalds ret = do_brk(addr, len); 2812128557ffSMichel Lespinasse populate = ((mm->def_flags & VM_LOCKED) != 0); 2813e4eb1ff6SLinus Torvalds up_write(&mm->mmap_sem); 2814128557ffSMichel Lespinasse if (populate) 2815128557ffSMichel Lespinasse mm_populate(addr, len); 2816e4eb1ff6SLinus Torvalds return ret; 2817e4eb1ff6SLinus Torvalds } 2818e4eb1ff6SLinus Torvalds EXPORT_SYMBOL(vm_brk); 28191da177e4SLinus Torvalds 28201da177e4SLinus Torvalds /* Release all mmaps. */ 28211da177e4SLinus Torvalds void exit_mmap(struct mm_struct *mm) 28221da177e4SLinus Torvalds { 2823d16dfc55SPeter Zijlstra struct mmu_gather tlb; 2824ba470de4SRik van Riel struct vm_area_struct *vma; 28251da177e4SLinus Torvalds unsigned long nr_accounted = 0; 28261da177e4SLinus Torvalds 2827d6dd61c8SJeremy Fitzhardinge /* mm's last user has gone, and its about to be pulled down */ 2828cddb8a5cSAndrea Arcangeli mmu_notifier_release(mm); 2829d6dd61c8SJeremy Fitzhardinge 2830ba470de4SRik van Riel if (mm->locked_vm) { 2831ba470de4SRik van Riel vma = mm->mmap; 2832ba470de4SRik van Riel while (vma) { 2833ba470de4SRik van Riel if (vma->vm_flags & VM_LOCKED) 2834ba470de4SRik van Riel munlock_vma_pages_all(vma); 2835ba470de4SRik van Riel vma = vma->vm_next; 2836ba470de4SRik van Riel } 2837ba470de4SRik van Riel } 28389480c53eSJeremy Fitzhardinge 28399480c53eSJeremy Fitzhardinge arch_exit_mmap(mm); 28409480c53eSJeremy Fitzhardinge 2841ba470de4SRik van Riel vma = mm->mmap; 28429480c53eSJeremy Fitzhardinge if (!vma) /* Can happen if dup_mmap() received an OOM */ 28439480c53eSJeremy Fitzhardinge return; 28449480c53eSJeremy Fitzhardinge 28451da177e4SLinus Torvalds lru_add_drain(); 28461da177e4SLinus Torvalds flush_cache_mm(mm); 28472b047252SLinus Torvalds tlb_gather_mmu(&tlb, mm, 0, -1); 2848901608d9SOleg Nesterov /* update_hiwater_rss(mm) here? but nobody should be looking */ 2849e0da382cSHugh Dickins /* Use -1 here to ensure all VMAs in the mm are unmapped */ 28504f74d2c8SLinus Torvalds unmap_vmas(&tlb, vma, 0, -1); 28519ba69294SHugh Dickins 28526ee8630eSHugh Dickins free_pgtables(&tlb, vma, FIRST_USER_ADDRESS, USER_PGTABLES_CEILING); 2853853f5e26SAl Viro tlb_finish_mmu(&tlb, 0, -1); 28541da177e4SLinus Torvalds 28551da177e4SLinus Torvalds /* 28568f4f8c16SHugh Dickins * Walk the list again, actually closing and freeing it, 28578f4f8c16SHugh Dickins * with preemption enabled, without holding any MM locks. 28581da177e4SLinus Torvalds */ 28594f74d2c8SLinus Torvalds while (vma) { 28604f74d2c8SLinus Torvalds if (vma->vm_flags & VM_ACCOUNT) 28614f74d2c8SLinus Torvalds nr_accounted += vma_pages(vma); 2862a8fb5618SHugh Dickins vma = remove_vma(vma); 28634f74d2c8SLinus Torvalds } 28644f74d2c8SLinus Torvalds vm_unacct_memory(nr_accounted); 28651da177e4SLinus Torvalds } 28661da177e4SLinus Torvalds 28671da177e4SLinus Torvalds /* Insert vm structure into process list sorted by address 28681da177e4SLinus Torvalds * and into the inode's i_mmap tree. If vm_file is non-NULL 2869c8c06efaSDavidlohr Bueso * then i_mmap_rwsem is taken here. 28701da177e4SLinus Torvalds */ 28711da177e4SLinus Torvalds int insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma) 28721da177e4SLinus Torvalds { 28736597d783SHugh Dickins struct vm_area_struct *prev; 28741da177e4SLinus Torvalds struct rb_node **rb_link, *rb_parent; 28751da177e4SLinus Torvalds 2876c9d13f5fSChen Gang if (find_vma_links(mm, vma->vm_start, vma->vm_end, 2877c9d13f5fSChen Gang &prev, &rb_link, &rb_parent)) 2878c9d13f5fSChen Gang return -ENOMEM; 2879c9d13f5fSChen Gang if ((vma->vm_flags & VM_ACCOUNT) && 2880c9d13f5fSChen Gang security_vm_enough_memory_mm(mm, vma_pages(vma))) 2881c9d13f5fSChen Gang return -ENOMEM; 2882c9d13f5fSChen Gang 28831da177e4SLinus Torvalds /* 28841da177e4SLinus Torvalds * The vm_pgoff of a purely anonymous vma should be irrelevant 28851da177e4SLinus Torvalds * until its first write fault, when page's anon_vma and index 28861da177e4SLinus Torvalds * are set. But now set the vm_pgoff it will almost certainly 28871da177e4SLinus Torvalds * end up with (unless mremap moves it elsewhere before that 28881da177e4SLinus Torvalds * first wfault), so /proc/pid/maps tells a consistent story. 28891da177e4SLinus Torvalds * 28901da177e4SLinus Torvalds * By setting it to reflect the virtual start address of the 28911da177e4SLinus Torvalds * vma, merges and splits can happen in a seamless way, just 28921da177e4SLinus Torvalds * using the existing file pgoff checks and manipulations. 28931da177e4SLinus Torvalds * Similarly in do_mmap_pgoff and in do_brk. 28941da177e4SLinus Torvalds */ 28958a9cc3b5SOleg Nesterov if (vma_is_anonymous(vma)) { 28961da177e4SLinus Torvalds BUG_ON(vma->anon_vma); 28971da177e4SLinus Torvalds vma->vm_pgoff = vma->vm_start >> PAGE_SHIFT; 28981da177e4SLinus Torvalds } 28992b144498SSrikar Dronamraju 29001da177e4SLinus Torvalds vma_link(mm, vma, prev, rb_link, rb_parent); 29011da177e4SLinus Torvalds return 0; 29021da177e4SLinus Torvalds } 29031da177e4SLinus Torvalds 29041da177e4SLinus Torvalds /* 29051da177e4SLinus Torvalds * Copy the vma structure to a new location in the same mm, 29061da177e4SLinus Torvalds * prior to moving page table entries, to effect an mremap move. 29071da177e4SLinus Torvalds */ 29081da177e4SLinus Torvalds struct vm_area_struct *copy_vma(struct vm_area_struct **vmap, 290938a76013SMichel Lespinasse unsigned long addr, unsigned long len, pgoff_t pgoff, 291038a76013SMichel Lespinasse bool *need_rmap_locks) 29111da177e4SLinus Torvalds { 29121da177e4SLinus Torvalds struct vm_area_struct *vma = *vmap; 29131da177e4SLinus Torvalds unsigned long vma_start = vma->vm_start; 29141da177e4SLinus Torvalds struct mm_struct *mm = vma->vm_mm; 29151da177e4SLinus Torvalds struct vm_area_struct *new_vma, *prev; 29161da177e4SLinus Torvalds struct rb_node **rb_link, *rb_parent; 2917948f017bSAndrea Arcangeli bool faulted_in_anon_vma = true; 29181da177e4SLinus Torvalds 29191da177e4SLinus Torvalds /* 29201da177e4SLinus Torvalds * If anonymous vma has not yet been faulted, update new pgoff 29211da177e4SLinus Torvalds * to match new location, to increase its chance of merging. 29221da177e4SLinus Torvalds */ 2923ce75799bSOleg Nesterov if (unlikely(vma_is_anonymous(vma) && !vma->anon_vma)) { 29241da177e4SLinus Torvalds pgoff = addr >> PAGE_SHIFT; 2925948f017bSAndrea Arcangeli faulted_in_anon_vma = false; 2926948f017bSAndrea Arcangeli } 29271da177e4SLinus Torvalds 29286597d783SHugh Dickins if (find_vma_links(mm, addr, addr + len, &prev, &rb_link, &rb_parent)) 29296597d783SHugh Dickins return NULL; /* should never get here */ 29301da177e4SLinus Torvalds new_vma = vma_merge(mm, prev, addr, addr + len, vma->vm_flags, 293119a809afSAndrea Arcangeli vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma), 293219a809afSAndrea Arcangeli vma->vm_userfaultfd_ctx); 29331da177e4SLinus Torvalds if (new_vma) { 29341da177e4SLinus Torvalds /* 29351da177e4SLinus Torvalds * Source vma may have been merged into new_vma 29361da177e4SLinus Torvalds */ 2937948f017bSAndrea Arcangeli if (unlikely(vma_start >= new_vma->vm_start && 2938948f017bSAndrea Arcangeli vma_start < new_vma->vm_end)) { 2939948f017bSAndrea Arcangeli /* 2940948f017bSAndrea Arcangeli * The only way we can get a vma_merge with 2941948f017bSAndrea Arcangeli * self during an mremap is if the vma hasn't 2942948f017bSAndrea Arcangeli * been faulted in yet and we were allowed to 2943948f017bSAndrea Arcangeli * reset the dst vma->vm_pgoff to the 2944948f017bSAndrea Arcangeli * destination address of the mremap to allow 2945948f017bSAndrea Arcangeli * the merge to happen. mremap must change the 2946948f017bSAndrea Arcangeli * vm_pgoff linearity between src and dst vmas 2947948f017bSAndrea Arcangeli * (in turn preventing a vma_merge) to be 2948948f017bSAndrea Arcangeli * safe. It is only safe to keep the vm_pgoff 2949948f017bSAndrea Arcangeli * linear if there are no pages mapped yet. 2950948f017bSAndrea Arcangeli */ 295181d1b09cSSasha Levin VM_BUG_ON_VMA(faulted_in_anon_vma, new_vma); 295238a76013SMichel Lespinasse *vmap = vma = new_vma; 2953108d6642SMichel Lespinasse } 295438a76013SMichel Lespinasse *need_rmap_locks = (new_vma->vm_pgoff <= vma->vm_pgoff); 29551da177e4SLinus Torvalds } else { 2956e94b1766SChristoph Lameter new_vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL); 2957e3975891SChen Gang if (!new_vma) 2958e3975891SChen Gang goto out; 29591da177e4SLinus Torvalds *new_vma = *vma; 29601da177e4SLinus Torvalds new_vma->vm_start = addr; 29611da177e4SLinus Torvalds new_vma->vm_end = addr + len; 29621da177e4SLinus Torvalds new_vma->vm_pgoff = pgoff; 2963ef0855d3SOleg Nesterov if (vma_dup_policy(vma, new_vma)) 2964523d4e20SMichel Lespinasse goto out_free_vma; 2965523d4e20SMichel Lespinasse INIT_LIST_HEAD(&new_vma->anon_vma_chain); 2966523d4e20SMichel Lespinasse if (anon_vma_clone(new_vma, vma)) 2967523d4e20SMichel Lespinasse goto out_free_mempol; 2968e9714acfSKonstantin Khlebnikov if (new_vma->vm_file) 29691da177e4SLinus Torvalds get_file(new_vma->vm_file); 29701da177e4SLinus Torvalds if (new_vma->vm_ops && new_vma->vm_ops->open) 29711da177e4SLinus Torvalds new_vma->vm_ops->open(new_vma); 29721da177e4SLinus Torvalds vma_link(mm, new_vma, prev, rb_link, rb_parent); 297338a76013SMichel Lespinasse *need_rmap_locks = false; 29741da177e4SLinus Torvalds } 29751da177e4SLinus Torvalds return new_vma; 29765beb4930SRik van Riel 29775beb4930SRik van Riel out_free_mempol: 2978ef0855d3SOleg Nesterov mpol_put(vma_policy(new_vma)); 29795beb4930SRik van Riel out_free_vma: 29805beb4930SRik van Riel kmem_cache_free(vm_area_cachep, new_vma); 2981e3975891SChen Gang out: 29825beb4930SRik van Riel return NULL; 29831da177e4SLinus Torvalds } 2984119f657cSakpm@osdl.org 2985119f657cSakpm@osdl.org /* 2986119f657cSakpm@osdl.org * Return true if the calling process may expand its vm space by the passed 2987119f657cSakpm@osdl.org * number of pages 2988119f657cSakpm@osdl.org */ 298984638335SKonstantin Khlebnikov bool may_expand_vm(struct mm_struct *mm, vm_flags_t flags, unsigned long npages) 2990119f657cSakpm@osdl.org { 299184638335SKonstantin Khlebnikov if (mm->total_vm + npages > rlimit(RLIMIT_AS) >> PAGE_SHIFT) 299284638335SKonstantin Khlebnikov return false; 299384638335SKonstantin Khlebnikov 2994d977d56cSKonstantin Khlebnikov if (is_data_mapping(flags) && 2995d977d56cSKonstantin Khlebnikov mm->data_vm + npages > rlimit(RLIMIT_DATA) >> PAGE_SHIFT) { 2996d977d56cSKonstantin Khlebnikov if (ignore_rlimit_data) 2997d977d56cSKonstantin Khlebnikov pr_warn_once("%s (%d): VmData %lu exceed data ulimit " 2998d977d56cSKonstantin Khlebnikov "%lu. Will be forbidden soon.\n", 2999d977d56cSKonstantin Khlebnikov current->comm, current->pid, 3000d977d56cSKonstantin Khlebnikov (mm->data_vm + npages) << PAGE_SHIFT, 3001d977d56cSKonstantin Khlebnikov rlimit(RLIMIT_DATA)); 3002d977d56cSKonstantin Khlebnikov else 3003d977d56cSKonstantin Khlebnikov return false; 3004d977d56cSKonstantin Khlebnikov } 300584638335SKonstantin Khlebnikov 300684638335SKonstantin Khlebnikov return true; 300784638335SKonstantin Khlebnikov } 300884638335SKonstantin Khlebnikov 300984638335SKonstantin Khlebnikov void vm_stat_account(struct mm_struct *mm, vm_flags_t flags, long npages) 301084638335SKonstantin Khlebnikov { 301184638335SKonstantin Khlebnikov mm->total_vm += npages; 301284638335SKonstantin Khlebnikov 3013d977d56cSKonstantin Khlebnikov if (is_exec_mapping(flags)) 301484638335SKonstantin Khlebnikov mm->exec_vm += npages; 3015d977d56cSKonstantin Khlebnikov else if (is_stack_mapping(flags)) 301684638335SKonstantin Khlebnikov mm->stack_vm += npages; 3017d977d56cSKonstantin Khlebnikov else if (is_data_mapping(flags)) 301884638335SKonstantin Khlebnikov mm->data_vm += npages; 3019119f657cSakpm@osdl.org } 3020fa5dc22fSRoland McGrath 3021a62c34bdSAndy Lutomirski static int special_mapping_fault(struct vm_area_struct *vma, 3022a62c34bdSAndy Lutomirski struct vm_fault *vmf); 3023a62c34bdSAndy Lutomirski 3024a62c34bdSAndy Lutomirski /* 3025a62c34bdSAndy Lutomirski * Having a close hook prevents vma merging regardless of flags. 3026a62c34bdSAndy Lutomirski */ 3027a62c34bdSAndy Lutomirski static void special_mapping_close(struct vm_area_struct *vma) 3028a62c34bdSAndy Lutomirski { 3029a62c34bdSAndy Lutomirski } 3030a62c34bdSAndy Lutomirski 3031a62c34bdSAndy Lutomirski static const char *special_mapping_name(struct vm_area_struct *vma) 3032a62c34bdSAndy Lutomirski { 3033a62c34bdSAndy Lutomirski return ((struct vm_special_mapping *)vma->vm_private_data)->name; 3034a62c34bdSAndy Lutomirski } 3035a62c34bdSAndy Lutomirski 3036a62c34bdSAndy Lutomirski static const struct vm_operations_struct special_mapping_vmops = { 3037a62c34bdSAndy Lutomirski .close = special_mapping_close, 3038a62c34bdSAndy Lutomirski .fault = special_mapping_fault, 3039a62c34bdSAndy Lutomirski .name = special_mapping_name, 3040a62c34bdSAndy Lutomirski }; 3041a62c34bdSAndy Lutomirski 3042a62c34bdSAndy Lutomirski static const struct vm_operations_struct legacy_special_mapping_vmops = { 3043a62c34bdSAndy Lutomirski .close = special_mapping_close, 3044a62c34bdSAndy Lutomirski .fault = special_mapping_fault, 3045a62c34bdSAndy Lutomirski }; 3046fa5dc22fSRoland McGrath 3047b1d0e4f5SNick Piggin static int special_mapping_fault(struct vm_area_struct *vma, 3048b1d0e4f5SNick Piggin struct vm_fault *vmf) 3049fa5dc22fSRoland McGrath { 3050b1d0e4f5SNick Piggin pgoff_t pgoff; 3051fa5dc22fSRoland McGrath struct page **pages; 3052fa5dc22fSRoland McGrath 3053f872f540SAndy Lutomirski if (vma->vm_ops == &legacy_special_mapping_vmops) { 3054a62c34bdSAndy Lutomirski pages = vma->vm_private_data; 3055f872f540SAndy Lutomirski } else { 3056f872f540SAndy Lutomirski struct vm_special_mapping *sm = vma->vm_private_data; 3057f872f540SAndy Lutomirski 3058f872f540SAndy Lutomirski if (sm->fault) 3059f872f540SAndy Lutomirski return sm->fault(sm, vma, vmf); 3060f872f540SAndy Lutomirski 3061f872f540SAndy Lutomirski pages = sm->pages; 3062f872f540SAndy Lutomirski } 3063a62c34bdSAndy Lutomirski 30648a9cc3b5SOleg Nesterov for (pgoff = vmf->pgoff; pgoff && *pages; ++pages) 3065b1d0e4f5SNick Piggin pgoff--; 3066fa5dc22fSRoland McGrath 3067fa5dc22fSRoland McGrath if (*pages) { 3068fa5dc22fSRoland McGrath struct page *page = *pages; 3069fa5dc22fSRoland McGrath get_page(page); 3070b1d0e4f5SNick Piggin vmf->page = page; 3071b1d0e4f5SNick Piggin return 0; 3072fa5dc22fSRoland McGrath } 3073fa5dc22fSRoland McGrath 3074b1d0e4f5SNick Piggin return VM_FAULT_SIGBUS; 3075fa5dc22fSRoland McGrath } 3076fa5dc22fSRoland McGrath 3077a62c34bdSAndy Lutomirski static struct vm_area_struct *__install_special_mapping( 3078a62c34bdSAndy Lutomirski struct mm_struct *mm, 3079fa5dc22fSRoland McGrath unsigned long addr, unsigned long len, 308027f28b97SChen Gang unsigned long vm_flags, void *priv, 308127f28b97SChen Gang const struct vm_operations_struct *ops) 3082fa5dc22fSRoland McGrath { 3083462e635eSTavis Ormandy int ret; 3084fa5dc22fSRoland McGrath struct vm_area_struct *vma; 3085fa5dc22fSRoland McGrath 3086fa5dc22fSRoland McGrath vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL); 3087fa5dc22fSRoland McGrath if (unlikely(vma == NULL)) 30883935ed6aSStefani Seibold return ERR_PTR(-ENOMEM); 3089fa5dc22fSRoland McGrath 30905beb4930SRik van Riel INIT_LIST_HEAD(&vma->anon_vma_chain); 3091fa5dc22fSRoland McGrath vma->vm_mm = mm; 3092fa5dc22fSRoland McGrath vma->vm_start = addr; 3093fa5dc22fSRoland McGrath vma->vm_end = addr + len; 3094fa5dc22fSRoland McGrath 3095d9104d1cSCyrill Gorcunov vma->vm_flags = vm_flags | mm->def_flags | VM_DONTEXPAND | VM_SOFTDIRTY; 30963ed75eb8SColy Li vma->vm_page_prot = vm_get_page_prot(vma->vm_flags); 3097fa5dc22fSRoland McGrath 3098a62c34bdSAndy Lutomirski vma->vm_ops = ops; 3099a62c34bdSAndy Lutomirski vma->vm_private_data = priv; 3100fa5dc22fSRoland McGrath 3101462e635eSTavis Ormandy ret = insert_vm_struct(mm, vma); 3102462e635eSTavis Ormandy if (ret) 3103462e635eSTavis Ormandy goto out; 3104fa5dc22fSRoland McGrath 310584638335SKonstantin Khlebnikov vm_stat_account(mm, vma->vm_flags, len >> PAGE_SHIFT); 3106fa5dc22fSRoland McGrath 3107cdd6c482SIngo Molnar perf_event_mmap(vma); 3108089dd79dSPeter Zijlstra 31093935ed6aSStefani Seibold return vma; 3110462e635eSTavis Ormandy 3111462e635eSTavis Ormandy out: 3112462e635eSTavis Ormandy kmem_cache_free(vm_area_cachep, vma); 31133935ed6aSStefani Seibold return ERR_PTR(ret); 31143935ed6aSStefani Seibold } 31153935ed6aSStefani Seibold 3116a62c34bdSAndy Lutomirski /* 3117a62c34bdSAndy Lutomirski * Called with mm->mmap_sem held for writing. 3118a62c34bdSAndy Lutomirski * Insert a new vma covering the given region, with the given flags. 3119a62c34bdSAndy Lutomirski * Its pages are supplied by the given array of struct page *. 3120a62c34bdSAndy Lutomirski * The array can be shorter than len >> PAGE_SHIFT if it's null-terminated. 3121a62c34bdSAndy Lutomirski * The region past the last page supplied will always produce SIGBUS. 3122a62c34bdSAndy Lutomirski * The array pointer and the pages it points to are assumed to stay alive 3123a62c34bdSAndy Lutomirski * for as long as this mapping might exist. 3124a62c34bdSAndy Lutomirski */ 3125a62c34bdSAndy Lutomirski struct vm_area_struct *_install_special_mapping( 3126a62c34bdSAndy Lutomirski struct mm_struct *mm, 3127a62c34bdSAndy Lutomirski unsigned long addr, unsigned long len, 3128a62c34bdSAndy Lutomirski unsigned long vm_flags, const struct vm_special_mapping *spec) 3129a62c34bdSAndy Lutomirski { 313027f28b97SChen Gang return __install_special_mapping(mm, addr, len, vm_flags, (void *)spec, 313127f28b97SChen Gang &special_mapping_vmops); 3132a62c34bdSAndy Lutomirski } 3133a62c34bdSAndy Lutomirski 31343935ed6aSStefani Seibold int install_special_mapping(struct mm_struct *mm, 31353935ed6aSStefani Seibold unsigned long addr, unsigned long len, 31363935ed6aSStefani Seibold unsigned long vm_flags, struct page **pages) 31373935ed6aSStefani Seibold { 3138a62c34bdSAndy Lutomirski struct vm_area_struct *vma = __install_special_mapping( 313927f28b97SChen Gang mm, addr, len, vm_flags, (void *)pages, 314027f28b97SChen Gang &legacy_special_mapping_vmops); 31413935ed6aSStefani Seibold 314214bd5b45SDuan Jiong return PTR_ERR_OR_ZERO(vma); 3143fa5dc22fSRoland McGrath } 31447906d00cSAndrea Arcangeli 31457906d00cSAndrea Arcangeli static DEFINE_MUTEX(mm_all_locks_mutex); 31467906d00cSAndrea Arcangeli 3147454ed842SPeter Zijlstra static void vm_lock_anon_vma(struct mm_struct *mm, struct anon_vma *anon_vma) 31487906d00cSAndrea Arcangeli { 3149bf181b9fSMichel Lespinasse if (!test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_node)) { 31507906d00cSAndrea Arcangeli /* 31517906d00cSAndrea Arcangeli * The LSB of head.next can't change from under us 31527906d00cSAndrea Arcangeli * because we hold the mm_all_locks_mutex. 31537906d00cSAndrea Arcangeli */ 3154572043c9SJiri Kosina down_write_nest_lock(&anon_vma->root->rwsem, &mm->mmap_sem); 31557906d00cSAndrea Arcangeli /* 31567906d00cSAndrea Arcangeli * We can safely modify head.next after taking the 31575a505085SIngo Molnar * anon_vma->root->rwsem. If some other vma in this mm shares 31587906d00cSAndrea Arcangeli * the same anon_vma we won't take it again. 31597906d00cSAndrea Arcangeli * 31607906d00cSAndrea Arcangeli * No need of atomic instructions here, head.next 31617906d00cSAndrea Arcangeli * can't change from under us thanks to the 31625a505085SIngo Molnar * anon_vma->root->rwsem. 31637906d00cSAndrea Arcangeli */ 31647906d00cSAndrea Arcangeli if (__test_and_set_bit(0, (unsigned long *) 3165bf181b9fSMichel Lespinasse &anon_vma->root->rb_root.rb_node)) 31667906d00cSAndrea Arcangeli BUG(); 31677906d00cSAndrea Arcangeli } 31687906d00cSAndrea Arcangeli } 31697906d00cSAndrea Arcangeli 3170454ed842SPeter Zijlstra static void vm_lock_mapping(struct mm_struct *mm, struct address_space *mapping) 31717906d00cSAndrea Arcangeli { 31727906d00cSAndrea Arcangeli if (!test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) { 31737906d00cSAndrea Arcangeli /* 31747906d00cSAndrea Arcangeli * AS_MM_ALL_LOCKS can't change from under us because 31757906d00cSAndrea Arcangeli * we hold the mm_all_locks_mutex. 31767906d00cSAndrea Arcangeli * 31777906d00cSAndrea Arcangeli * Operations on ->flags have to be atomic because 31787906d00cSAndrea Arcangeli * even if AS_MM_ALL_LOCKS is stable thanks to the 31797906d00cSAndrea Arcangeli * mm_all_locks_mutex, there may be other cpus 31807906d00cSAndrea Arcangeli * changing other bitflags in parallel to us. 31817906d00cSAndrea Arcangeli */ 31827906d00cSAndrea Arcangeli if (test_and_set_bit(AS_MM_ALL_LOCKS, &mapping->flags)) 31837906d00cSAndrea Arcangeli BUG(); 3184c8c06efaSDavidlohr Bueso down_write_nest_lock(&mapping->i_mmap_rwsem, &mm->mmap_sem); 31857906d00cSAndrea Arcangeli } 31867906d00cSAndrea Arcangeli } 31877906d00cSAndrea Arcangeli 31887906d00cSAndrea Arcangeli /* 31897906d00cSAndrea Arcangeli * This operation locks against the VM for all pte/vma/mm related 31907906d00cSAndrea Arcangeli * operations that could ever happen on a certain mm. This includes 31917906d00cSAndrea Arcangeli * vmtruncate, try_to_unmap, and all page faults. 31927906d00cSAndrea Arcangeli * 31937906d00cSAndrea Arcangeli * The caller must take the mmap_sem in write mode before calling 31947906d00cSAndrea Arcangeli * mm_take_all_locks(). The caller isn't allowed to release the 31957906d00cSAndrea Arcangeli * mmap_sem until mm_drop_all_locks() returns. 31967906d00cSAndrea Arcangeli * 31977906d00cSAndrea Arcangeli * mmap_sem in write mode is required in order to block all operations 31987906d00cSAndrea Arcangeli * that could modify pagetables and free pages without need of 319927ba0644SKirill A. Shutemov * altering the vma layout. It's also needed in write mode to avoid new 32007906d00cSAndrea Arcangeli * anon_vmas to be associated with existing vmas. 32017906d00cSAndrea Arcangeli * 32027906d00cSAndrea Arcangeli * A single task can't take more than one mm_take_all_locks() in a row 32037906d00cSAndrea Arcangeli * or it would deadlock. 32047906d00cSAndrea Arcangeli * 3205bf181b9fSMichel Lespinasse * The LSB in anon_vma->rb_root.rb_node and the AS_MM_ALL_LOCKS bitflag in 32067906d00cSAndrea Arcangeli * mapping->flags avoid to take the same lock twice, if more than one 32077906d00cSAndrea Arcangeli * vma in this mm is backed by the same anon_vma or address_space. 32087906d00cSAndrea Arcangeli * 320988f306b6SKirill A. Shutemov * We take locks in following order, accordingly to comment at beginning 321088f306b6SKirill A. Shutemov * of mm/rmap.c: 321188f306b6SKirill A. Shutemov * - all hugetlbfs_i_mmap_rwsem_key locks (aka mapping->i_mmap_rwsem for 321288f306b6SKirill A. Shutemov * hugetlb mapping); 321388f306b6SKirill A. Shutemov * - all i_mmap_rwsem locks; 321488f306b6SKirill A. Shutemov * - all anon_vma->rwseml 321588f306b6SKirill A. Shutemov * 321688f306b6SKirill A. Shutemov * We can take all locks within these types randomly because the VM code 321788f306b6SKirill A. Shutemov * doesn't nest them and we protected from parallel mm_take_all_locks() by 321888f306b6SKirill A. Shutemov * mm_all_locks_mutex. 32197906d00cSAndrea Arcangeli * 32207906d00cSAndrea Arcangeli * mm_take_all_locks() and mm_drop_all_locks are expensive operations 32217906d00cSAndrea Arcangeli * that may have to take thousand of locks. 32227906d00cSAndrea Arcangeli * 32237906d00cSAndrea Arcangeli * mm_take_all_locks() can fail if it's interrupted by signals. 32247906d00cSAndrea Arcangeli */ 32257906d00cSAndrea Arcangeli int mm_take_all_locks(struct mm_struct *mm) 32267906d00cSAndrea Arcangeli { 32277906d00cSAndrea Arcangeli struct vm_area_struct *vma; 32285beb4930SRik van Riel struct anon_vma_chain *avc; 32297906d00cSAndrea Arcangeli 32307906d00cSAndrea Arcangeli BUG_ON(down_read_trylock(&mm->mmap_sem)); 32317906d00cSAndrea Arcangeli 32327906d00cSAndrea Arcangeli mutex_lock(&mm_all_locks_mutex); 32337906d00cSAndrea Arcangeli 32347906d00cSAndrea Arcangeli for (vma = mm->mmap; vma; vma = vma->vm_next) { 32357906d00cSAndrea Arcangeli if (signal_pending(current)) 32367906d00cSAndrea Arcangeli goto out_unlock; 323788f306b6SKirill A. Shutemov if (vma->vm_file && vma->vm_file->f_mapping && 323888f306b6SKirill A. Shutemov is_vm_hugetlb_page(vma)) 323988f306b6SKirill A. Shutemov vm_lock_mapping(mm, vma->vm_file->f_mapping); 324088f306b6SKirill A. Shutemov } 324188f306b6SKirill A. Shutemov 324288f306b6SKirill A. Shutemov for (vma = mm->mmap; vma; vma = vma->vm_next) { 324388f306b6SKirill A. Shutemov if (signal_pending(current)) 324488f306b6SKirill A. Shutemov goto out_unlock; 324588f306b6SKirill A. Shutemov if (vma->vm_file && vma->vm_file->f_mapping && 324688f306b6SKirill A. Shutemov !is_vm_hugetlb_page(vma)) 3247454ed842SPeter Zijlstra vm_lock_mapping(mm, vma->vm_file->f_mapping); 32487906d00cSAndrea Arcangeli } 32497cd5a02fSPeter Zijlstra 32507cd5a02fSPeter Zijlstra for (vma = mm->mmap; vma; vma = vma->vm_next) { 32517cd5a02fSPeter Zijlstra if (signal_pending(current)) 32527cd5a02fSPeter Zijlstra goto out_unlock; 32537cd5a02fSPeter Zijlstra if (vma->anon_vma) 32545beb4930SRik van Riel list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) 32555beb4930SRik van Riel vm_lock_anon_vma(mm, avc->anon_vma); 32567cd5a02fSPeter Zijlstra } 32577cd5a02fSPeter Zijlstra 3258584cff54SKautuk Consul return 0; 32597906d00cSAndrea Arcangeli 32607906d00cSAndrea Arcangeli out_unlock: 32617906d00cSAndrea Arcangeli mm_drop_all_locks(mm); 3262584cff54SKautuk Consul return -EINTR; 32637906d00cSAndrea Arcangeli } 32647906d00cSAndrea Arcangeli 32657906d00cSAndrea Arcangeli static void vm_unlock_anon_vma(struct anon_vma *anon_vma) 32667906d00cSAndrea Arcangeli { 3267bf181b9fSMichel Lespinasse if (test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_node)) { 32687906d00cSAndrea Arcangeli /* 32697906d00cSAndrea Arcangeli * The LSB of head.next can't change to 0 from under 32707906d00cSAndrea Arcangeli * us because we hold the mm_all_locks_mutex. 32717906d00cSAndrea Arcangeli * 32727906d00cSAndrea Arcangeli * We must however clear the bitflag before unlocking 3273bf181b9fSMichel Lespinasse * the vma so the users using the anon_vma->rb_root will 32747906d00cSAndrea Arcangeli * never see our bitflag. 32757906d00cSAndrea Arcangeli * 32767906d00cSAndrea Arcangeli * No need of atomic instructions here, head.next 32777906d00cSAndrea Arcangeli * can't change from under us until we release the 32785a505085SIngo Molnar * anon_vma->root->rwsem. 32797906d00cSAndrea Arcangeli */ 32807906d00cSAndrea Arcangeli if (!__test_and_clear_bit(0, (unsigned long *) 3281bf181b9fSMichel Lespinasse &anon_vma->root->rb_root.rb_node)) 32827906d00cSAndrea Arcangeli BUG(); 328308b52706SKonstantin Khlebnikov anon_vma_unlock_write(anon_vma); 32847906d00cSAndrea Arcangeli } 32857906d00cSAndrea Arcangeli } 32867906d00cSAndrea Arcangeli 32877906d00cSAndrea Arcangeli static void vm_unlock_mapping(struct address_space *mapping) 32887906d00cSAndrea Arcangeli { 32897906d00cSAndrea Arcangeli if (test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) { 32907906d00cSAndrea Arcangeli /* 32917906d00cSAndrea Arcangeli * AS_MM_ALL_LOCKS can't change to 0 from under us 32927906d00cSAndrea Arcangeli * because we hold the mm_all_locks_mutex. 32937906d00cSAndrea Arcangeli */ 329483cde9e8SDavidlohr Bueso i_mmap_unlock_write(mapping); 32957906d00cSAndrea Arcangeli if (!test_and_clear_bit(AS_MM_ALL_LOCKS, 32967906d00cSAndrea Arcangeli &mapping->flags)) 32977906d00cSAndrea Arcangeli BUG(); 32987906d00cSAndrea Arcangeli } 32997906d00cSAndrea Arcangeli } 33007906d00cSAndrea Arcangeli 33017906d00cSAndrea Arcangeli /* 33027906d00cSAndrea Arcangeli * The mmap_sem cannot be released by the caller until 33037906d00cSAndrea Arcangeli * mm_drop_all_locks() returns. 33047906d00cSAndrea Arcangeli */ 33057906d00cSAndrea Arcangeli void mm_drop_all_locks(struct mm_struct *mm) 33067906d00cSAndrea Arcangeli { 33077906d00cSAndrea Arcangeli struct vm_area_struct *vma; 33085beb4930SRik van Riel struct anon_vma_chain *avc; 33097906d00cSAndrea Arcangeli 33107906d00cSAndrea Arcangeli BUG_ON(down_read_trylock(&mm->mmap_sem)); 33117906d00cSAndrea Arcangeli BUG_ON(!mutex_is_locked(&mm_all_locks_mutex)); 33127906d00cSAndrea Arcangeli 33137906d00cSAndrea Arcangeli for (vma = mm->mmap; vma; vma = vma->vm_next) { 33147906d00cSAndrea Arcangeli if (vma->anon_vma) 33155beb4930SRik van Riel list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) 33165beb4930SRik van Riel vm_unlock_anon_vma(avc->anon_vma); 33177906d00cSAndrea Arcangeli if (vma->vm_file && vma->vm_file->f_mapping) 33187906d00cSAndrea Arcangeli vm_unlock_mapping(vma->vm_file->f_mapping); 33197906d00cSAndrea Arcangeli } 33207906d00cSAndrea Arcangeli 33217906d00cSAndrea Arcangeli mutex_unlock(&mm_all_locks_mutex); 33227906d00cSAndrea Arcangeli } 33238feae131SDavid Howells 33248feae131SDavid Howells /* 33258feae131SDavid Howells * initialise the VMA slab 33268feae131SDavid Howells */ 33278feae131SDavid Howells void __init mmap_init(void) 33288feae131SDavid Howells { 332900a62ce9SKOSAKI Motohiro int ret; 333000a62ce9SKOSAKI Motohiro 3331908c7f19STejun Heo ret = percpu_counter_init(&vm_committed_as, 0, GFP_KERNEL); 333200a62ce9SKOSAKI Motohiro VM_BUG_ON(ret); 33338feae131SDavid Howells } 3334c9b1d098SAndrew Shewmaker 3335c9b1d098SAndrew Shewmaker /* 3336c9b1d098SAndrew Shewmaker * Initialise sysctl_user_reserve_kbytes. 3337c9b1d098SAndrew Shewmaker * 3338c9b1d098SAndrew Shewmaker * This is intended to prevent a user from starting a single memory hogging 3339c9b1d098SAndrew Shewmaker * process, such that they cannot recover (kill the hog) in OVERCOMMIT_NEVER 3340c9b1d098SAndrew Shewmaker * mode. 3341c9b1d098SAndrew Shewmaker * 3342c9b1d098SAndrew Shewmaker * The default value is min(3% of free memory, 128MB) 3343c9b1d098SAndrew Shewmaker * 128MB is enough to recover with sshd/login, bash, and top/kill. 3344c9b1d098SAndrew Shewmaker */ 33451640879aSAndrew Shewmaker static int init_user_reserve(void) 3346c9b1d098SAndrew Shewmaker { 3347c9b1d098SAndrew Shewmaker unsigned long free_kbytes; 3348c9b1d098SAndrew Shewmaker 3349c9b1d098SAndrew Shewmaker free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10); 3350c9b1d098SAndrew Shewmaker 3351c9b1d098SAndrew Shewmaker sysctl_user_reserve_kbytes = min(free_kbytes / 32, 1UL << 17); 3352c9b1d098SAndrew Shewmaker return 0; 3353c9b1d098SAndrew Shewmaker } 3354a64fb3cdSPaul Gortmaker subsys_initcall(init_user_reserve); 33554eeab4f5SAndrew Shewmaker 33564eeab4f5SAndrew Shewmaker /* 33574eeab4f5SAndrew Shewmaker * Initialise sysctl_admin_reserve_kbytes. 33584eeab4f5SAndrew Shewmaker * 33594eeab4f5SAndrew Shewmaker * The purpose of sysctl_admin_reserve_kbytes is to allow the sys admin 33604eeab4f5SAndrew Shewmaker * to log in and kill a memory hogging process. 33614eeab4f5SAndrew Shewmaker * 33624eeab4f5SAndrew Shewmaker * Systems with more than 256MB will reserve 8MB, enough to recover 33634eeab4f5SAndrew Shewmaker * with sshd, bash, and top in OVERCOMMIT_GUESS. Smaller systems will 33644eeab4f5SAndrew Shewmaker * only reserve 3% of free pages by default. 33654eeab4f5SAndrew Shewmaker */ 33661640879aSAndrew Shewmaker static int init_admin_reserve(void) 33674eeab4f5SAndrew Shewmaker { 33684eeab4f5SAndrew Shewmaker unsigned long free_kbytes; 33694eeab4f5SAndrew Shewmaker 33704eeab4f5SAndrew Shewmaker free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10); 33714eeab4f5SAndrew Shewmaker 33724eeab4f5SAndrew Shewmaker sysctl_admin_reserve_kbytes = min(free_kbytes / 32, 1UL << 13); 33734eeab4f5SAndrew Shewmaker return 0; 33744eeab4f5SAndrew Shewmaker } 3375a64fb3cdSPaul Gortmaker subsys_initcall(init_admin_reserve); 33761640879aSAndrew Shewmaker 33771640879aSAndrew Shewmaker /* 33781640879aSAndrew Shewmaker * Reinititalise user and admin reserves if memory is added or removed. 33791640879aSAndrew Shewmaker * 33801640879aSAndrew Shewmaker * The default user reserve max is 128MB, and the default max for the 33811640879aSAndrew Shewmaker * admin reserve is 8MB. These are usually, but not always, enough to 33821640879aSAndrew Shewmaker * enable recovery from a memory hogging process using login/sshd, a shell, 33831640879aSAndrew Shewmaker * and tools like top. It may make sense to increase or even disable the 33841640879aSAndrew Shewmaker * reserve depending on the existence of swap or variations in the recovery 33851640879aSAndrew Shewmaker * tools. So, the admin may have changed them. 33861640879aSAndrew Shewmaker * 33871640879aSAndrew Shewmaker * If memory is added and the reserves have been eliminated or increased above 33881640879aSAndrew Shewmaker * the default max, then we'll trust the admin. 33891640879aSAndrew Shewmaker * 33901640879aSAndrew Shewmaker * If memory is removed and there isn't enough free memory, then we 33911640879aSAndrew Shewmaker * need to reset the reserves. 33921640879aSAndrew Shewmaker * 33931640879aSAndrew Shewmaker * Otherwise keep the reserve set by the admin. 33941640879aSAndrew Shewmaker */ 33951640879aSAndrew Shewmaker static int reserve_mem_notifier(struct notifier_block *nb, 33961640879aSAndrew Shewmaker unsigned long action, void *data) 33971640879aSAndrew Shewmaker { 33981640879aSAndrew Shewmaker unsigned long tmp, free_kbytes; 33991640879aSAndrew Shewmaker 34001640879aSAndrew Shewmaker switch (action) { 34011640879aSAndrew Shewmaker case MEM_ONLINE: 34021640879aSAndrew Shewmaker /* Default max is 128MB. Leave alone if modified by operator. */ 34031640879aSAndrew Shewmaker tmp = sysctl_user_reserve_kbytes; 34041640879aSAndrew Shewmaker if (0 < tmp && tmp < (1UL << 17)) 34051640879aSAndrew Shewmaker init_user_reserve(); 34061640879aSAndrew Shewmaker 34071640879aSAndrew Shewmaker /* Default max is 8MB. Leave alone if modified by operator. */ 34081640879aSAndrew Shewmaker tmp = sysctl_admin_reserve_kbytes; 34091640879aSAndrew Shewmaker if (0 < tmp && tmp < (1UL << 13)) 34101640879aSAndrew Shewmaker init_admin_reserve(); 34111640879aSAndrew Shewmaker 34121640879aSAndrew Shewmaker break; 34131640879aSAndrew Shewmaker case MEM_OFFLINE: 34141640879aSAndrew Shewmaker free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10); 34151640879aSAndrew Shewmaker 34161640879aSAndrew Shewmaker if (sysctl_user_reserve_kbytes > free_kbytes) { 34171640879aSAndrew Shewmaker init_user_reserve(); 34181640879aSAndrew Shewmaker pr_info("vm.user_reserve_kbytes reset to %lu\n", 34191640879aSAndrew Shewmaker sysctl_user_reserve_kbytes); 34201640879aSAndrew Shewmaker } 34211640879aSAndrew Shewmaker 34221640879aSAndrew Shewmaker if (sysctl_admin_reserve_kbytes > free_kbytes) { 34231640879aSAndrew Shewmaker init_admin_reserve(); 34241640879aSAndrew Shewmaker pr_info("vm.admin_reserve_kbytes reset to %lu\n", 34251640879aSAndrew Shewmaker sysctl_admin_reserve_kbytes); 34261640879aSAndrew Shewmaker } 34271640879aSAndrew Shewmaker break; 34281640879aSAndrew Shewmaker default: 34291640879aSAndrew Shewmaker break; 34301640879aSAndrew Shewmaker } 34311640879aSAndrew Shewmaker return NOTIFY_OK; 34321640879aSAndrew Shewmaker } 34331640879aSAndrew Shewmaker 34341640879aSAndrew Shewmaker static struct notifier_block reserve_mem_nb = { 34351640879aSAndrew Shewmaker .notifier_call = reserve_mem_notifier, 34361640879aSAndrew Shewmaker }; 34371640879aSAndrew Shewmaker 34381640879aSAndrew Shewmaker static int __meminit init_reserve_notifier(void) 34391640879aSAndrew Shewmaker { 34401640879aSAndrew Shewmaker if (register_hotmemory_notifier(&reserve_mem_nb)) 3441b1de0d13SMitchel Humpherys pr_err("Failed registering memory add/remove notifier for admin reserve\n"); 34421640879aSAndrew Shewmaker 34431640879aSAndrew Shewmaker return 0; 34441640879aSAndrew Shewmaker } 3445a64fb3cdSPaul Gortmaker subsys_initcall(init_reserve_notifier); 3446