11da177e4SLinus Torvalds /* 21da177e4SLinus Torvalds * mm/mmap.c 31da177e4SLinus Torvalds * 41da177e4SLinus Torvalds * Written by obz. 51da177e4SLinus Torvalds * 6046c6884SAlan Cox * Address space accounting code <alan@lxorguk.ukuu.org.uk> 71da177e4SLinus Torvalds */ 81da177e4SLinus Torvalds 9b1de0d13SMitchel Humpherys #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt 10b1de0d13SMitchel Humpherys 11e8420a8eSCyril Hrubis #include <linux/kernel.h> 121da177e4SLinus Torvalds #include <linux/slab.h> 134af3c9ccSAlexey Dobriyan #include <linux/backing-dev.h> 141da177e4SLinus Torvalds #include <linux/mm.h> 15615d6e87SDavidlohr Bueso #include <linux/vmacache.h> 161da177e4SLinus Torvalds #include <linux/shm.h> 171da177e4SLinus Torvalds #include <linux/mman.h> 181da177e4SLinus Torvalds #include <linux/pagemap.h> 191da177e4SLinus Torvalds #include <linux/swap.h> 201da177e4SLinus Torvalds #include <linux/syscalls.h> 21c59ede7bSRandy.Dunlap #include <linux/capability.h> 221da177e4SLinus Torvalds #include <linux/init.h> 231da177e4SLinus Torvalds #include <linux/file.h> 241da177e4SLinus Torvalds #include <linux/fs.h> 251da177e4SLinus Torvalds #include <linux/personality.h> 261da177e4SLinus Torvalds #include <linux/security.h> 271da177e4SLinus Torvalds #include <linux/hugetlb.h> 281da177e4SLinus Torvalds #include <linux/profile.h> 29b95f1b31SPaul Gortmaker #include <linux/export.h> 301da177e4SLinus Torvalds #include <linux/mount.h> 311da177e4SLinus Torvalds #include <linux/mempolicy.h> 321da177e4SLinus Torvalds #include <linux/rmap.h> 33cddb8a5cSAndrea Arcangeli #include <linux/mmu_notifier.h> 3482f71ae4SKonstantin Khlebnikov #include <linux/mmdebug.h> 35cdd6c482SIngo Molnar #include <linux/perf_event.h> 36120a795dSAl Viro #include <linux/audit.h> 37b15d00b6SAndrea Arcangeli #include <linux/khugepaged.h> 382b144498SSrikar Dronamraju #include <linux/uprobes.h> 39d3737187SMichel Lespinasse #include <linux/rbtree_augmented.h> 40cf4aebc2SClark Williams #include <linux/sched/sysctl.h> 411640879aSAndrew Shewmaker #include <linux/notifier.h> 421640879aSAndrew Shewmaker #include <linux/memory.h> 43b1de0d13SMitchel Humpherys #include <linux/printk.h> 441da177e4SLinus Torvalds 451da177e4SLinus Torvalds #include <asm/uaccess.h> 461da177e4SLinus Torvalds #include <asm/cacheflush.h> 471da177e4SLinus Torvalds #include <asm/tlb.h> 48d6dd61c8SJeremy Fitzhardinge #include <asm/mmu_context.h> 491da177e4SLinus Torvalds 5042b77728SJan Beulich #include "internal.h" 5142b77728SJan Beulich 523a459756SKirill Korotaev #ifndef arch_mmap_check 533a459756SKirill Korotaev #define arch_mmap_check(addr, len, flags) (0) 543a459756SKirill Korotaev #endif 553a459756SKirill Korotaev 5608e7d9b5SMartin Schwidefsky #ifndef arch_rebalance_pgtables 5708e7d9b5SMartin Schwidefsky #define arch_rebalance_pgtables(addr, len) (addr) 5808e7d9b5SMartin Schwidefsky #endif 5908e7d9b5SMartin Schwidefsky 60e0da382cSHugh Dickins static void unmap_region(struct mm_struct *mm, 61e0da382cSHugh Dickins struct vm_area_struct *vma, struct vm_area_struct *prev, 62e0da382cSHugh Dickins unsigned long start, unsigned long end); 63e0da382cSHugh Dickins 641da177e4SLinus Torvalds /* description of effects of mapping type and prot in current implementation. 651da177e4SLinus Torvalds * this is due to the limited x86 page protection hardware. The expected 661da177e4SLinus Torvalds * behavior is in parens: 671da177e4SLinus Torvalds * 681da177e4SLinus Torvalds * map_type prot 691da177e4SLinus Torvalds * PROT_NONE PROT_READ PROT_WRITE PROT_EXEC 701da177e4SLinus Torvalds * MAP_SHARED r: (no) no r: (yes) yes r: (no) yes r: (no) yes 711da177e4SLinus Torvalds * w: (no) no w: (no) no w: (yes) yes w: (no) no 721da177e4SLinus Torvalds * x: (no) no x: (no) yes x: (no) yes x: (yes) yes 731da177e4SLinus Torvalds * 741da177e4SLinus Torvalds * MAP_PRIVATE r: (no) no r: (yes) yes r: (no) yes r: (no) yes 751da177e4SLinus Torvalds * w: (no) no w: (no) no w: (copy) copy w: (no) no 761da177e4SLinus Torvalds * x: (no) no x: (no) yes x: (no) yes x: (yes) yes 771da177e4SLinus Torvalds * 781da177e4SLinus Torvalds */ 791da177e4SLinus Torvalds pgprot_t protection_map[16] = { 801da177e4SLinus Torvalds __P000, __P001, __P010, __P011, __P100, __P101, __P110, __P111, 811da177e4SLinus Torvalds __S000, __S001, __S010, __S011, __S100, __S101, __S110, __S111 821da177e4SLinus Torvalds }; 831da177e4SLinus Torvalds 84804af2cfSHugh Dickins pgprot_t vm_get_page_prot(unsigned long vm_flags) 85804af2cfSHugh Dickins { 86b845f313SDave Kleikamp return __pgprot(pgprot_val(protection_map[vm_flags & 87b845f313SDave Kleikamp (VM_READ|VM_WRITE|VM_EXEC|VM_SHARED)]) | 88b845f313SDave Kleikamp pgprot_val(arch_vm_get_page_prot(vm_flags))); 89804af2cfSHugh Dickins } 90804af2cfSHugh Dickins EXPORT_SYMBOL(vm_get_page_prot); 91804af2cfSHugh Dickins 9264e45507SPeter Feiner static pgprot_t vm_pgprot_modify(pgprot_t oldprot, unsigned long vm_flags) 9364e45507SPeter Feiner { 9464e45507SPeter Feiner return pgprot_modify(oldprot, vm_get_page_prot(vm_flags)); 9564e45507SPeter Feiner } 9664e45507SPeter Feiner 9764e45507SPeter Feiner /* Update vma->vm_page_prot to reflect vma->vm_flags. */ 9864e45507SPeter Feiner void vma_set_page_prot(struct vm_area_struct *vma) 9964e45507SPeter Feiner { 10064e45507SPeter Feiner unsigned long vm_flags = vma->vm_flags; 10164e45507SPeter Feiner 10264e45507SPeter Feiner vma->vm_page_prot = vm_pgprot_modify(vma->vm_page_prot, vm_flags); 10364e45507SPeter Feiner if (vma_wants_writenotify(vma)) { 10464e45507SPeter Feiner vm_flags &= ~VM_SHARED; 10564e45507SPeter Feiner vma->vm_page_prot = vm_pgprot_modify(vma->vm_page_prot, 10664e45507SPeter Feiner vm_flags); 10764e45507SPeter Feiner } 10864e45507SPeter Feiner } 10964e45507SPeter Feiner 11064e45507SPeter Feiner 11134679d7eSShaohua Li int sysctl_overcommit_memory __read_mostly = OVERCOMMIT_GUESS; /* heuristic overcommit */ 11234679d7eSShaohua Li int sysctl_overcommit_ratio __read_mostly = 50; /* default is 50% */ 11349f0ce5fSJerome Marchand unsigned long sysctl_overcommit_kbytes __read_mostly; 114c3d8c141SChristoph Lameter int sysctl_max_map_count __read_mostly = DEFAULT_MAX_MAP_COUNT; 115c9b1d098SAndrew Shewmaker unsigned long sysctl_user_reserve_kbytes __read_mostly = 1UL << 17; /* 128MB */ 1164eeab4f5SAndrew Shewmaker unsigned long sysctl_admin_reserve_kbytes __read_mostly = 1UL << 13; /* 8MB */ 11734679d7eSShaohua Li /* 11834679d7eSShaohua Li * Make sure vm_committed_as in one cacheline and not cacheline shared with 11934679d7eSShaohua Li * other variables. It can be updated by several CPUs frequently. 12034679d7eSShaohua Li */ 12134679d7eSShaohua Li struct percpu_counter vm_committed_as ____cacheline_aligned_in_smp; 1221da177e4SLinus Torvalds 1231da177e4SLinus Torvalds /* 124997071bcSK. Y. Srinivasan * The global memory commitment made in the system can be a metric 125997071bcSK. Y. Srinivasan * that can be used to drive ballooning decisions when Linux is hosted 126997071bcSK. Y. Srinivasan * as a guest. On Hyper-V, the host implements a policy engine for dynamically 127997071bcSK. Y. Srinivasan * balancing memory across competing virtual machines that are hosted. 128997071bcSK. Y. Srinivasan * Several metrics drive this policy engine including the guest reported 129997071bcSK. Y. Srinivasan * memory commitment. 130997071bcSK. Y. Srinivasan */ 131997071bcSK. Y. Srinivasan unsigned long vm_memory_committed(void) 132997071bcSK. Y. Srinivasan { 133997071bcSK. Y. Srinivasan return percpu_counter_read_positive(&vm_committed_as); 134997071bcSK. Y. Srinivasan } 135997071bcSK. Y. Srinivasan EXPORT_SYMBOL_GPL(vm_memory_committed); 136997071bcSK. Y. Srinivasan 137997071bcSK. Y. Srinivasan /* 1381da177e4SLinus Torvalds * Check that a process has enough memory to allocate a new virtual 1391da177e4SLinus Torvalds * mapping. 0 means there is enough memory for the allocation to 1401da177e4SLinus Torvalds * succeed and -ENOMEM implies there is not. 1411da177e4SLinus Torvalds * 1421da177e4SLinus Torvalds * We currently support three overcommit policies, which are set via the 1431da177e4SLinus Torvalds * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting 1441da177e4SLinus Torvalds * 1451da177e4SLinus Torvalds * Strict overcommit modes added 2002 Feb 26 by Alan Cox. 1461da177e4SLinus Torvalds * Additional code 2002 Jul 20 by Robert Love. 1471da177e4SLinus Torvalds * 1481da177e4SLinus Torvalds * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise. 1491da177e4SLinus Torvalds * 1501da177e4SLinus Torvalds * Note this is a helper function intended to be used by LSMs which 1511da177e4SLinus Torvalds * wish to use this logic. 1521da177e4SLinus Torvalds */ 15334b4e4aaSAlan Cox int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin) 1541da177e4SLinus Torvalds { 155c9b1d098SAndrew Shewmaker unsigned long free, allowed, reserve; 1561da177e4SLinus Torvalds 15782f71ae4SKonstantin Khlebnikov VM_WARN_ONCE(percpu_counter_read(&vm_committed_as) < 15882f71ae4SKonstantin Khlebnikov -(s64)vm_committed_as_batch * num_online_cpus(), 15982f71ae4SKonstantin Khlebnikov "memory commitment underflow"); 16082f71ae4SKonstantin Khlebnikov 1611da177e4SLinus Torvalds vm_acct_memory(pages); 1621da177e4SLinus Torvalds 1631da177e4SLinus Torvalds /* 1641da177e4SLinus Torvalds * Sometimes we want to use more memory than we have 1651da177e4SLinus Torvalds */ 1661da177e4SLinus Torvalds if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS) 1671da177e4SLinus Torvalds return 0; 1681da177e4SLinus Torvalds 1691da177e4SLinus Torvalds if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) { 170c15bef30SDmitry Fink free = global_page_state(NR_FREE_PAGES); 171c15bef30SDmitry Fink free += global_page_state(NR_FILE_PAGES); 1721da177e4SLinus Torvalds 173c15bef30SDmitry Fink /* 174c15bef30SDmitry Fink * shmem pages shouldn't be counted as free in this 175c15bef30SDmitry Fink * case, they can't be purged, only swapped out, and 176c15bef30SDmitry Fink * that won't affect the overall amount of available 177c15bef30SDmitry Fink * memory in the system. 178c15bef30SDmitry Fink */ 179c15bef30SDmitry Fink free -= global_page_state(NR_SHMEM); 180c15bef30SDmitry Fink 181ec8acf20SShaohua Li free += get_nr_swap_pages(); 1821da177e4SLinus Torvalds 1831da177e4SLinus Torvalds /* 1841da177e4SLinus Torvalds * Any slabs which are created with the 1851da177e4SLinus Torvalds * SLAB_RECLAIM_ACCOUNT flag claim to have contents 1861da177e4SLinus Torvalds * which are reclaimable, under pressure. The dentry 1871da177e4SLinus Torvalds * cache and most inode caches should fall into this 1881da177e4SLinus Torvalds */ 189972d1a7bSChristoph Lameter free += global_page_state(NR_SLAB_RECLAIMABLE); 1901da177e4SLinus Torvalds 1911da177e4SLinus Torvalds /* 192c15bef30SDmitry Fink * Leave reserved pages. The pages are not for anonymous pages. 193c15bef30SDmitry Fink */ 194c15bef30SDmitry Fink if (free <= totalreserve_pages) 195c15bef30SDmitry Fink goto error; 196c15bef30SDmitry Fink else 197c15bef30SDmitry Fink free -= totalreserve_pages; 198c15bef30SDmitry Fink 199c15bef30SDmitry Fink /* 2004eeab4f5SAndrew Shewmaker * Reserve some for root 2011da177e4SLinus Torvalds */ 2021da177e4SLinus Torvalds if (!cap_sys_admin) 2034eeab4f5SAndrew Shewmaker free -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10); 2041da177e4SLinus Torvalds 2051da177e4SLinus Torvalds if (free > pages) 2061da177e4SLinus Torvalds return 0; 2071da177e4SLinus Torvalds 2086d9f7839SHideo AOKI goto error; 2091da177e4SLinus Torvalds } 2101da177e4SLinus Torvalds 21100619bccSJerome Marchand allowed = vm_commit_limit(); 2121da177e4SLinus Torvalds /* 2134eeab4f5SAndrew Shewmaker * Reserve some for root 2141da177e4SLinus Torvalds */ 2151da177e4SLinus Torvalds if (!cap_sys_admin) 2164eeab4f5SAndrew Shewmaker allowed -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10); 2171da177e4SLinus Torvalds 218c9b1d098SAndrew Shewmaker /* 219c9b1d098SAndrew Shewmaker * Don't let a single process grow so big a user can't recover 220c9b1d098SAndrew Shewmaker */ 221c9b1d098SAndrew Shewmaker if (mm) { 222c9b1d098SAndrew Shewmaker reserve = sysctl_user_reserve_kbytes >> (PAGE_SHIFT - 10); 223c9b1d098SAndrew Shewmaker allowed -= min(mm->total_vm / 32, reserve); 224c9b1d098SAndrew Shewmaker } 2251da177e4SLinus Torvalds 22600a62ce9SKOSAKI Motohiro if (percpu_counter_read_positive(&vm_committed_as) < allowed) 2271da177e4SLinus Torvalds return 0; 2286d9f7839SHideo AOKI error: 2291da177e4SLinus Torvalds vm_unacct_memory(pages); 2301da177e4SLinus Torvalds 2311da177e4SLinus Torvalds return -ENOMEM; 2321da177e4SLinus Torvalds } 2331da177e4SLinus Torvalds 2341da177e4SLinus Torvalds /* 235c8c06efaSDavidlohr Bueso * Requires inode->i_mapping->i_mmap_rwsem 2361da177e4SLinus Torvalds */ 2371da177e4SLinus Torvalds static void __remove_shared_vm_struct(struct vm_area_struct *vma, 2381da177e4SLinus Torvalds struct file *file, struct address_space *mapping) 2391da177e4SLinus Torvalds { 2401da177e4SLinus Torvalds if (vma->vm_flags & VM_DENYWRITE) 241496ad9aaSAl Viro atomic_inc(&file_inode(file)->i_writecount); 2421da177e4SLinus Torvalds if (vma->vm_flags & VM_SHARED) 2434bb5f5d9SDavid Herrmann mapping_unmap_writable(mapping); 2441da177e4SLinus Torvalds 2451da177e4SLinus Torvalds flush_dcache_mmap_lock(mapping); 2466b2dbba8SMichel Lespinasse vma_interval_tree_remove(vma, &mapping->i_mmap); 2471da177e4SLinus Torvalds flush_dcache_mmap_unlock(mapping); 2481da177e4SLinus Torvalds } 2491da177e4SLinus Torvalds 2501da177e4SLinus Torvalds /* 2516b2dbba8SMichel Lespinasse * Unlink a file-based vm structure from its interval tree, to hide 252a8fb5618SHugh Dickins * vma from rmap and vmtruncate before freeing its page tables. 2531da177e4SLinus Torvalds */ 254a8fb5618SHugh Dickins void unlink_file_vma(struct vm_area_struct *vma) 2551da177e4SLinus Torvalds { 2561da177e4SLinus Torvalds struct file *file = vma->vm_file; 2571da177e4SLinus Torvalds 2581da177e4SLinus Torvalds if (file) { 2591da177e4SLinus Torvalds struct address_space *mapping = file->f_mapping; 26083cde9e8SDavidlohr Bueso i_mmap_lock_write(mapping); 2611da177e4SLinus Torvalds __remove_shared_vm_struct(vma, file, mapping); 26283cde9e8SDavidlohr Bueso i_mmap_unlock_write(mapping); 2631da177e4SLinus Torvalds } 264a8fb5618SHugh Dickins } 265a8fb5618SHugh Dickins 266a8fb5618SHugh Dickins /* 267a8fb5618SHugh Dickins * Close a vm structure and free it, returning the next. 268a8fb5618SHugh Dickins */ 269a8fb5618SHugh Dickins static struct vm_area_struct *remove_vma(struct vm_area_struct *vma) 270a8fb5618SHugh Dickins { 271a8fb5618SHugh Dickins struct vm_area_struct *next = vma->vm_next; 272a8fb5618SHugh Dickins 273a8fb5618SHugh Dickins might_sleep(); 2741da177e4SLinus Torvalds if (vma->vm_ops && vma->vm_ops->close) 2751da177e4SLinus Torvalds vma->vm_ops->close(vma); 276e9714acfSKonstantin Khlebnikov if (vma->vm_file) 277a8fb5618SHugh Dickins fput(vma->vm_file); 278f0be3d32SLee Schermerhorn mpol_put(vma_policy(vma)); 2791da177e4SLinus Torvalds kmem_cache_free(vm_area_cachep, vma); 280a8fb5618SHugh Dickins return next; 2811da177e4SLinus Torvalds } 2821da177e4SLinus Torvalds 283e4eb1ff6SLinus Torvalds static unsigned long do_brk(unsigned long addr, unsigned long len); 284e4eb1ff6SLinus Torvalds 2856a6160a7SHeiko Carstens SYSCALL_DEFINE1(brk, unsigned long, brk) 2861da177e4SLinus Torvalds { 2878764b338SCyrill Gorcunov unsigned long retval; 2881da177e4SLinus Torvalds unsigned long newbrk, oldbrk; 2891da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 290a5b4592cSJiri Kosina unsigned long min_brk; 291128557ffSMichel Lespinasse bool populate; 2921da177e4SLinus Torvalds 2931da177e4SLinus Torvalds down_write(&mm->mmap_sem); 2941da177e4SLinus Torvalds 295a5b4592cSJiri Kosina #ifdef CONFIG_COMPAT_BRK 2965520e894SJiri Kosina /* 2975520e894SJiri Kosina * CONFIG_COMPAT_BRK can still be overridden by setting 2985520e894SJiri Kosina * randomize_va_space to 2, which will still cause mm->start_brk 2995520e894SJiri Kosina * to be arbitrarily shifted 3005520e894SJiri Kosina */ 3014471a675SJiri Kosina if (current->brk_randomized) 3025520e894SJiri Kosina min_brk = mm->start_brk; 3035520e894SJiri Kosina else 3045520e894SJiri Kosina min_brk = mm->end_data; 305a5b4592cSJiri Kosina #else 306a5b4592cSJiri Kosina min_brk = mm->start_brk; 307a5b4592cSJiri Kosina #endif 308a5b4592cSJiri Kosina if (brk < min_brk) 3091da177e4SLinus Torvalds goto out; 3101e624196SRam Gupta 3111e624196SRam Gupta /* 3121e624196SRam Gupta * Check against rlimit here. If this check is done later after the test 3131e624196SRam Gupta * of oldbrk with newbrk then it can escape the test and let the data 3141e624196SRam Gupta * segment grow beyond its set limit the in case where the limit is 3151e624196SRam Gupta * not page aligned -Ram Gupta 3161e624196SRam Gupta */ 3178764b338SCyrill Gorcunov if (check_data_rlimit(rlimit(RLIMIT_DATA), brk, mm->start_brk, 3188764b338SCyrill Gorcunov mm->end_data, mm->start_data)) 3191e624196SRam Gupta goto out; 3201e624196SRam Gupta 3211da177e4SLinus Torvalds newbrk = PAGE_ALIGN(brk); 3221da177e4SLinus Torvalds oldbrk = PAGE_ALIGN(mm->brk); 3231da177e4SLinus Torvalds if (oldbrk == newbrk) 3241da177e4SLinus Torvalds goto set_brk; 3251da177e4SLinus Torvalds 3261da177e4SLinus Torvalds /* Always allow shrinking brk. */ 3271da177e4SLinus Torvalds if (brk <= mm->brk) { 3281da177e4SLinus Torvalds if (!do_munmap(mm, newbrk, oldbrk-newbrk)) 3291da177e4SLinus Torvalds goto set_brk; 3301da177e4SLinus Torvalds goto out; 3311da177e4SLinus Torvalds } 3321da177e4SLinus Torvalds 3331da177e4SLinus Torvalds /* Check against existing mmap mappings. */ 3341da177e4SLinus Torvalds if (find_vma_intersection(mm, oldbrk, newbrk+PAGE_SIZE)) 3351da177e4SLinus Torvalds goto out; 3361da177e4SLinus Torvalds 3371da177e4SLinus Torvalds /* Ok, looks good - let it rip. */ 3381da177e4SLinus Torvalds if (do_brk(oldbrk, newbrk-oldbrk) != oldbrk) 3391da177e4SLinus Torvalds goto out; 340128557ffSMichel Lespinasse 3411da177e4SLinus Torvalds set_brk: 3421da177e4SLinus Torvalds mm->brk = brk; 343128557ffSMichel Lespinasse populate = newbrk > oldbrk && (mm->def_flags & VM_LOCKED) != 0; 344128557ffSMichel Lespinasse up_write(&mm->mmap_sem); 345128557ffSMichel Lespinasse if (populate) 346128557ffSMichel Lespinasse mm_populate(oldbrk, newbrk - oldbrk); 347128557ffSMichel Lespinasse return brk; 348128557ffSMichel Lespinasse 3491da177e4SLinus Torvalds out: 3501da177e4SLinus Torvalds retval = mm->brk; 3511da177e4SLinus Torvalds up_write(&mm->mmap_sem); 3521da177e4SLinus Torvalds return retval; 3531da177e4SLinus Torvalds } 3541da177e4SLinus Torvalds 355d3737187SMichel Lespinasse static long vma_compute_subtree_gap(struct vm_area_struct *vma) 356d3737187SMichel Lespinasse { 357d3737187SMichel Lespinasse unsigned long max, subtree_gap; 358d3737187SMichel Lespinasse max = vma->vm_start; 359d3737187SMichel Lespinasse if (vma->vm_prev) 360d3737187SMichel Lespinasse max -= vma->vm_prev->vm_end; 361d3737187SMichel Lespinasse if (vma->vm_rb.rb_left) { 362d3737187SMichel Lespinasse subtree_gap = rb_entry(vma->vm_rb.rb_left, 363d3737187SMichel Lespinasse struct vm_area_struct, vm_rb)->rb_subtree_gap; 364d3737187SMichel Lespinasse if (subtree_gap > max) 365d3737187SMichel Lespinasse max = subtree_gap; 366d3737187SMichel Lespinasse } 367d3737187SMichel Lespinasse if (vma->vm_rb.rb_right) { 368d3737187SMichel Lespinasse subtree_gap = rb_entry(vma->vm_rb.rb_right, 369d3737187SMichel Lespinasse struct vm_area_struct, vm_rb)->rb_subtree_gap; 370d3737187SMichel Lespinasse if (subtree_gap > max) 371d3737187SMichel Lespinasse max = subtree_gap; 372d3737187SMichel Lespinasse } 373d3737187SMichel Lespinasse return max; 374d3737187SMichel Lespinasse } 375d3737187SMichel Lespinasse 376ed8ea815SMichel Lespinasse #ifdef CONFIG_DEBUG_VM_RB 3771da177e4SLinus Torvalds static int browse_rb(struct rb_root *root) 3781da177e4SLinus Torvalds { 3795a0768f6SMichel Lespinasse int i = 0, j, bug = 0; 3801da177e4SLinus Torvalds struct rb_node *nd, *pn = NULL; 3811da177e4SLinus Torvalds unsigned long prev = 0, pend = 0; 3821da177e4SLinus Torvalds 3831da177e4SLinus Torvalds for (nd = rb_first(root); nd; nd = rb_next(nd)) { 3841da177e4SLinus Torvalds struct vm_area_struct *vma; 3851da177e4SLinus Torvalds vma = rb_entry(nd, struct vm_area_struct, vm_rb); 3865a0768f6SMichel Lespinasse if (vma->vm_start < prev) { 387ff26f70fSAndrew Morton pr_emerg("vm_start %lx < prev %lx\n", 388ff26f70fSAndrew Morton vma->vm_start, prev); 3895a0768f6SMichel Lespinasse bug = 1; 3905a0768f6SMichel Lespinasse } 3915a0768f6SMichel Lespinasse if (vma->vm_start < pend) { 392ff26f70fSAndrew Morton pr_emerg("vm_start %lx < pend %lx\n", 393ff26f70fSAndrew Morton vma->vm_start, pend); 3945a0768f6SMichel Lespinasse bug = 1; 3955a0768f6SMichel Lespinasse } 3965a0768f6SMichel Lespinasse if (vma->vm_start > vma->vm_end) { 397ff26f70fSAndrew Morton pr_emerg("vm_start %lx > vm_end %lx\n", 398ff26f70fSAndrew Morton vma->vm_start, vma->vm_end); 3995a0768f6SMichel Lespinasse bug = 1; 4005a0768f6SMichel Lespinasse } 4015a0768f6SMichel Lespinasse if (vma->rb_subtree_gap != vma_compute_subtree_gap(vma)) { 4028542bdfcSSasha Levin pr_emerg("free gap %lx, correct %lx\n", 4035a0768f6SMichel Lespinasse vma->rb_subtree_gap, 4045a0768f6SMichel Lespinasse vma_compute_subtree_gap(vma)); 4055a0768f6SMichel Lespinasse bug = 1; 4065a0768f6SMichel Lespinasse } 4071da177e4SLinus Torvalds i++; 4081da177e4SLinus Torvalds pn = nd; 409d1af65d1SDavid Miller prev = vma->vm_start; 410d1af65d1SDavid Miller pend = vma->vm_end; 4111da177e4SLinus Torvalds } 4121da177e4SLinus Torvalds j = 0; 4135a0768f6SMichel Lespinasse for (nd = pn; nd; nd = rb_prev(nd)) 4141da177e4SLinus Torvalds j++; 4155a0768f6SMichel Lespinasse if (i != j) { 4168542bdfcSSasha Levin pr_emerg("backwards %d, forwards %d\n", j, i); 4175a0768f6SMichel Lespinasse bug = 1; 4181da177e4SLinus Torvalds } 4195a0768f6SMichel Lespinasse return bug ? -1 : i; 4201da177e4SLinus Torvalds } 4211da177e4SLinus Torvalds 422d3737187SMichel Lespinasse static void validate_mm_rb(struct rb_root *root, struct vm_area_struct *ignore) 423d3737187SMichel Lespinasse { 424d3737187SMichel Lespinasse struct rb_node *nd; 425d3737187SMichel Lespinasse 426d3737187SMichel Lespinasse for (nd = rb_first(root); nd; nd = rb_next(nd)) { 427d3737187SMichel Lespinasse struct vm_area_struct *vma; 428d3737187SMichel Lespinasse vma = rb_entry(nd, struct vm_area_struct, vm_rb); 42996dad67fSSasha Levin VM_BUG_ON_VMA(vma != ignore && 43096dad67fSSasha Levin vma->rb_subtree_gap != vma_compute_subtree_gap(vma), 43196dad67fSSasha Levin vma); 432d3737187SMichel Lespinasse } 4331da177e4SLinus Torvalds } 4341da177e4SLinus Torvalds 435eafd4dc4SRashika Kheria static void validate_mm(struct mm_struct *mm) 4361da177e4SLinus Torvalds { 4371da177e4SLinus Torvalds int bug = 0; 4381da177e4SLinus Torvalds int i = 0; 4395a0768f6SMichel Lespinasse unsigned long highest_address = 0; 440ed8ea815SMichel Lespinasse struct vm_area_struct *vma = mm->mmap; 441ff26f70fSAndrew Morton 442ed8ea815SMichel Lespinasse while (vma) { 443ed8ea815SMichel Lespinasse struct anon_vma_chain *avc; 444ff26f70fSAndrew Morton 44563c3b902SMichel Lespinasse vma_lock_anon_vma(vma); 446ed8ea815SMichel Lespinasse list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) 447ed8ea815SMichel Lespinasse anon_vma_interval_tree_verify(avc); 44863c3b902SMichel Lespinasse vma_unlock_anon_vma(vma); 4495a0768f6SMichel Lespinasse highest_address = vma->vm_end; 450ed8ea815SMichel Lespinasse vma = vma->vm_next; 4511da177e4SLinus Torvalds i++; 4521da177e4SLinus Torvalds } 4535a0768f6SMichel Lespinasse if (i != mm->map_count) { 4548542bdfcSSasha Levin pr_emerg("map_count %d vm_next %d\n", mm->map_count, i); 4555a0768f6SMichel Lespinasse bug = 1; 4565a0768f6SMichel Lespinasse } 4575a0768f6SMichel Lespinasse if (highest_address != mm->highest_vm_end) { 4588542bdfcSSasha Levin pr_emerg("mm->highest_vm_end %lx, found %lx\n", 4595a0768f6SMichel Lespinasse mm->highest_vm_end, highest_address); 4605a0768f6SMichel Lespinasse bug = 1; 4615a0768f6SMichel Lespinasse } 4621da177e4SLinus Torvalds i = browse_rb(&mm->mm_rb); 4635a0768f6SMichel Lespinasse if (i != mm->map_count) { 464ff26f70fSAndrew Morton if (i != -1) 4658542bdfcSSasha Levin pr_emerg("map_count %d rb %d\n", mm->map_count, i); 4665a0768f6SMichel Lespinasse bug = 1; 4675a0768f6SMichel Lespinasse } 46896dad67fSSasha Levin VM_BUG_ON_MM(bug, mm); 4691da177e4SLinus Torvalds } 4701da177e4SLinus Torvalds #else 471d3737187SMichel Lespinasse #define validate_mm_rb(root, ignore) do { } while (0) 4721da177e4SLinus Torvalds #define validate_mm(mm) do { } while (0) 4731da177e4SLinus Torvalds #endif 4741da177e4SLinus Torvalds 475d3737187SMichel Lespinasse RB_DECLARE_CALLBACKS(static, vma_gap_callbacks, struct vm_area_struct, vm_rb, 476d3737187SMichel Lespinasse unsigned long, rb_subtree_gap, vma_compute_subtree_gap) 477d3737187SMichel Lespinasse 478d3737187SMichel Lespinasse /* 479d3737187SMichel Lespinasse * Update augmented rbtree rb_subtree_gap values after vma->vm_start or 480d3737187SMichel Lespinasse * vma->vm_prev->vm_end values changed, without modifying the vma's position 481d3737187SMichel Lespinasse * in the rbtree. 482d3737187SMichel Lespinasse */ 483d3737187SMichel Lespinasse static void vma_gap_update(struct vm_area_struct *vma) 484d3737187SMichel Lespinasse { 485d3737187SMichel Lespinasse /* 486d3737187SMichel Lespinasse * As it turns out, RB_DECLARE_CALLBACKS() already created a callback 487d3737187SMichel Lespinasse * function that does exacltly what we want. 488d3737187SMichel Lespinasse */ 489d3737187SMichel Lespinasse vma_gap_callbacks_propagate(&vma->vm_rb, NULL); 490d3737187SMichel Lespinasse } 491d3737187SMichel Lespinasse 492d3737187SMichel Lespinasse static inline void vma_rb_insert(struct vm_area_struct *vma, 493d3737187SMichel Lespinasse struct rb_root *root) 494d3737187SMichel Lespinasse { 495d3737187SMichel Lespinasse /* All rb_subtree_gap values must be consistent prior to insertion */ 496d3737187SMichel Lespinasse validate_mm_rb(root, NULL); 497d3737187SMichel Lespinasse 498d3737187SMichel Lespinasse rb_insert_augmented(&vma->vm_rb, root, &vma_gap_callbacks); 499d3737187SMichel Lespinasse } 500d3737187SMichel Lespinasse 501d3737187SMichel Lespinasse static void vma_rb_erase(struct vm_area_struct *vma, struct rb_root *root) 502d3737187SMichel Lespinasse { 503d3737187SMichel Lespinasse /* 504d3737187SMichel Lespinasse * All rb_subtree_gap values must be consistent prior to erase, 505d3737187SMichel Lespinasse * with the possible exception of the vma being erased. 506d3737187SMichel Lespinasse */ 507d3737187SMichel Lespinasse validate_mm_rb(root, vma); 508d3737187SMichel Lespinasse 509d3737187SMichel Lespinasse /* 510d3737187SMichel Lespinasse * Note rb_erase_augmented is a fairly large inline function, 511d3737187SMichel Lespinasse * so make sure we instantiate it only once with our desired 512d3737187SMichel Lespinasse * augmented rbtree callbacks. 513d3737187SMichel Lespinasse */ 514d3737187SMichel Lespinasse rb_erase_augmented(&vma->vm_rb, root, &vma_gap_callbacks); 515d3737187SMichel Lespinasse } 516d3737187SMichel Lespinasse 517bf181b9fSMichel Lespinasse /* 518bf181b9fSMichel Lespinasse * vma has some anon_vma assigned, and is already inserted on that 519bf181b9fSMichel Lespinasse * anon_vma's interval trees. 520bf181b9fSMichel Lespinasse * 521bf181b9fSMichel Lespinasse * Before updating the vma's vm_start / vm_end / vm_pgoff fields, the 522bf181b9fSMichel Lespinasse * vma must be removed from the anon_vma's interval trees using 523bf181b9fSMichel Lespinasse * anon_vma_interval_tree_pre_update_vma(). 524bf181b9fSMichel Lespinasse * 525bf181b9fSMichel Lespinasse * After the update, the vma will be reinserted using 526bf181b9fSMichel Lespinasse * anon_vma_interval_tree_post_update_vma(). 527bf181b9fSMichel Lespinasse * 528bf181b9fSMichel Lespinasse * The entire update must be protected by exclusive mmap_sem and by 529bf181b9fSMichel Lespinasse * the root anon_vma's mutex. 530bf181b9fSMichel Lespinasse */ 531bf181b9fSMichel Lespinasse static inline void 532bf181b9fSMichel Lespinasse anon_vma_interval_tree_pre_update_vma(struct vm_area_struct *vma) 533bf181b9fSMichel Lespinasse { 534bf181b9fSMichel Lespinasse struct anon_vma_chain *avc; 535bf181b9fSMichel Lespinasse 536bf181b9fSMichel Lespinasse list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) 537bf181b9fSMichel Lespinasse anon_vma_interval_tree_remove(avc, &avc->anon_vma->rb_root); 538bf181b9fSMichel Lespinasse } 539bf181b9fSMichel Lespinasse 540bf181b9fSMichel Lespinasse static inline void 541bf181b9fSMichel Lespinasse anon_vma_interval_tree_post_update_vma(struct vm_area_struct *vma) 542bf181b9fSMichel Lespinasse { 543bf181b9fSMichel Lespinasse struct anon_vma_chain *avc; 544bf181b9fSMichel Lespinasse 545bf181b9fSMichel Lespinasse list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) 546bf181b9fSMichel Lespinasse anon_vma_interval_tree_insert(avc, &avc->anon_vma->rb_root); 547bf181b9fSMichel Lespinasse } 548bf181b9fSMichel Lespinasse 5496597d783SHugh Dickins static int find_vma_links(struct mm_struct *mm, unsigned long addr, 5506597d783SHugh Dickins unsigned long end, struct vm_area_struct **pprev, 5516597d783SHugh Dickins struct rb_node ***rb_link, struct rb_node **rb_parent) 5521da177e4SLinus Torvalds { 5531da177e4SLinus Torvalds struct rb_node **__rb_link, *__rb_parent, *rb_prev; 5541da177e4SLinus Torvalds 5551da177e4SLinus Torvalds __rb_link = &mm->mm_rb.rb_node; 5561da177e4SLinus Torvalds rb_prev = __rb_parent = NULL; 5571da177e4SLinus Torvalds 5581da177e4SLinus Torvalds while (*__rb_link) { 5591da177e4SLinus Torvalds struct vm_area_struct *vma_tmp; 5601da177e4SLinus Torvalds 5611da177e4SLinus Torvalds __rb_parent = *__rb_link; 5621da177e4SLinus Torvalds vma_tmp = rb_entry(__rb_parent, struct vm_area_struct, vm_rb); 5631da177e4SLinus Torvalds 5641da177e4SLinus Torvalds if (vma_tmp->vm_end > addr) { 5656597d783SHugh Dickins /* Fail if an existing vma overlaps the area */ 5666597d783SHugh Dickins if (vma_tmp->vm_start < end) 5676597d783SHugh Dickins return -ENOMEM; 5681da177e4SLinus Torvalds __rb_link = &__rb_parent->rb_left; 5691da177e4SLinus Torvalds } else { 5701da177e4SLinus Torvalds rb_prev = __rb_parent; 5711da177e4SLinus Torvalds __rb_link = &__rb_parent->rb_right; 5721da177e4SLinus Torvalds } 5731da177e4SLinus Torvalds } 5741da177e4SLinus Torvalds 5751da177e4SLinus Torvalds *pprev = NULL; 5761da177e4SLinus Torvalds if (rb_prev) 5771da177e4SLinus Torvalds *pprev = rb_entry(rb_prev, struct vm_area_struct, vm_rb); 5781da177e4SLinus Torvalds *rb_link = __rb_link; 5791da177e4SLinus Torvalds *rb_parent = __rb_parent; 5806597d783SHugh Dickins return 0; 5811da177e4SLinus Torvalds } 5821da177e4SLinus Torvalds 583e8420a8eSCyril Hrubis static unsigned long count_vma_pages_range(struct mm_struct *mm, 584e8420a8eSCyril Hrubis unsigned long addr, unsigned long end) 585e8420a8eSCyril Hrubis { 586e8420a8eSCyril Hrubis unsigned long nr_pages = 0; 587e8420a8eSCyril Hrubis struct vm_area_struct *vma; 588e8420a8eSCyril Hrubis 589e8420a8eSCyril Hrubis /* Find first overlaping mapping */ 590e8420a8eSCyril Hrubis vma = find_vma_intersection(mm, addr, end); 591e8420a8eSCyril Hrubis if (!vma) 592e8420a8eSCyril Hrubis return 0; 593e8420a8eSCyril Hrubis 594e8420a8eSCyril Hrubis nr_pages = (min(end, vma->vm_end) - 595e8420a8eSCyril Hrubis max(addr, vma->vm_start)) >> PAGE_SHIFT; 596e8420a8eSCyril Hrubis 597e8420a8eSCyril Hrubis /* Iterate over the rest of the overlaps */ 598e8420a8eSCyril Hrubis for (vma = vma->vm_next; vma; vma = vma->vm_next) { 599e8420a8eSCyril Hrubis unsigned long overlap_len; 600e8420a8eSCyril Hrubis 601e8420a8eSCyril Hrubis if (vma->vm_start > end) 602e8420a8eSCyril Hrubis break; 603e8420a8eSCyril Hrubis 604e8420a8eSCyril Hrubis overlap_len = min(end, vma->vm_end) - vma->vm_start; 605e8420a8eSCyril Hrubis nr_pages += overlap_len >> PAGE_SHIFT; 606e8420a8eSCyril Hrubis } 607e8420a8eSCyril Hrubis 608e8420a8eSCyril Hrubis return nr_pages; 609e8420a8eSCyril Hrubis } 610e8420a8eSCyril Hrubis 6111da177e4SLinus Torvalds void __vma_link_rb(struct mm_struct *mm, struct vm_area_struct *vma, 6121da177e4SLinus Torvalds struct rb_node **rb_link, struct rb_node *rb_parent) 6131da177e4SLinus Torvalds { 614d3737187SMichel Lespinasse /* Update tracking information for the gap following the new vma. */ 615d3737187SMichel Lespinasse if (vma->vm_next) 616d3737187SMichel Lespinasse vma_gap_update(vma->vm_next); 617d3737187SMichel Lespinasse else 618d3737187SMichel Lespinasse mm->highest_vm_end = vma->vm_end; 619d3737187SMichel Lespinasse 620d3737187SMichel Lespinasse /* 621d3737187SMichel Lespinasse * vma->vm_prev wasn't known when we followed the rbtree to find the 622d3737187SMichel Lespinasse * correct insertion point for that vma. As a result, we could not 623d3737187SMichel Lespinasse * update the vma vm_rb parents rb_subtree_gap values on the way down. 624d3737187SMichel Lespinasse * So, we first insert the vma with a zero rb_subtree_gap value 625d3737187SMichel Lespinasse * (to be consistent with what we did on the way down), and then 626d3737187SMichel Lespinasse * immediately update the gap to the correct value. Finally we 627d3737187SMichel Lespinasse * rebalance the rbtree after all augmented values have been set. 628d3737187SMichel Lespinasse */ 6291da177e4SLinus Torvalds rb_link_node(&vma->vm_rb, rb_parent, rb_link); 630d3737187SMichel Lespinasse vma->rb_subtree_gap = 0; 631d3737187SMichel Lespinasse vma_gap_update(vma); 632d3737187SMichel Lespinasse vma_rb_insert(vma, &mm->mm_rb); 6331da177e4SLinus Torvalds } 6341da177e4SLinus Torvalds 635cb8f488cSDenys Vlasenko static void __vma_link_file(struct vm_area_struct *vma) 6361da177e4SLinus Torvalds { 6371da177e4SLinus Torvalds struct file *file; 6381da177e4SLinus Torvalds 6391da177e4SLinus Torvalds file = vma->vm_file; 6401da177e4SLinus Torvalds if (file) { 6411da177e4SLinus Torvalds struct address_space *mapping = file->f_mapping; 6421da177e4SLinus Torvalds 6431da177e4SLinus Torvalds if (vma->vm_flags & VM_DENYWRITE) 644496ad9aaSAl Viro atomic_dec(&file_inode(file)->i_writecount); 6451da177e4SLinus Torvalds if (vma->vm_flags & VM_SHARED) 6464bb5f5d9SDavid Herrmann atomic_inc(&mapping->i_mmap_writable); 6471da177e4SLinus Torvalds 6481da177e4SLinus Torvalds flush_dcache_mmap_lock(mapping); 6496b2dbba8SMichel Lespinasse vma_interval_tree_insert(vma, &mapping->i_mmap); 6501da177e4SLinus Torvalds flush_dcache_mmap_unlock(mapping); 6511da177e4SLinus Torvalds } 6521da177e4SLinus Torvalds } 6531da177e4SLinus Torvalds 6541da177e4SLinus Torvalds static void 6551da177e4SLinus Torvalds __vma_link(struct mm_struct *mm, struct vm_area_struct *vma, 6561da177e4SLinus Torvalds struct vm_area_struct *prev, struct rb_node **rb_link, 6571da177e4SLinus Torvalds struct rb_node *rb_parent) 6581da177e4SLinus Torvalds { 6591da177e4SLinus Torvalds __vma_link_list(mm, vma, prev, rb_parent); 6601da177e4SLinus Torvalds __vma_link_rb(mm, vma, rb_link, rb_parent); 6611da177e4SLinus Torvalds } 6621da177e4SLinus Torvalds 6631da177e4SLinus Torvalds static void vma_link(struct mm_struct *mm, struct vm_area_struct *vma, 6641da177e4SLinus Torvalds struct vm_area_struct *prev, struct rb_node **rb_link, 6651da177e4SLinus Torvalds struct rb_node *rb_parent) 6661da177e4SLinus Torvalds { 6671da177e4SLinus Torvalds struct address_space *mapping = NULL; 6681da177e4SLinus Torvalds 66964ac4940SHuang Shijie if (vma->vm_file) { 6701da177e4SLinus Torvalds mapping = vma->vm_file->f_mapping; 67183cde9e8SDavidlohr Bueso i_mmap_lock_write(mapping); 67264ac4940SHuang Shijie } 6731da177e4SLinus Torvalds 6741da177e4SLinus Torvalds __vma_link(mm, vma, prev, rb_link, rb_parent); 6751da177e4SLinus Torvalds __vma_link_file(vma); 6761da177e4SLinus Torvalds 6771da177e4SLinus Torvalds if (mapping) 67883cde9e8SDavidlohr Bueso i_mmap_unlock_write(mapping); 6791da177e4SLinus Torvalds 6801da177e4SLinus Torvalds mm->map_count++; 6811da177e4SLinus Torvalds validate_mm(mm); 6821da177e4SLinus Torvalds } 6831da177e4SLinus Torvalds 6841da177e4SLinus Torvalds /* 68588f6b4c3SKautuk Consul * Helper for vma_adjust() in the split_vma insert case: insert a vma into the 6866b2dbba8SMichel Lespinasse * mm's list and rbtree. It has already been inserted into the interval tree. 6871da177e4SLinus Torvalds */ 68848aae425SZhenwenXu static void __insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma) 6891da177e4SLinus Torvalds { 6906597d783SHugh Dickins struct vm_area_struct *prev; 6911da177e4SLinus Torvalds struct rb_node **rb_link, *rb_parent; 6921da177e4SLinus Torvalds 6936597d783SHugh Dickins if (find_vma_links(mm, vma->vm_start, vma->vm_end, 6946597d783SHugh Dickins &prev, &rb_link, &rb_parent)) 6956597d783SHugh Dickins BUG(); 6961da177e4SLinus Torvalds __vma_link(mm, vma, prev, rb_link, rb_parent); 6971da177e4SLinus Torvalds mm->map_count++; 6981da177e4SLinus Torvalds } 6991da177e4SLinus Torvalds 7001da177e4SLinus Torvalds static inline void 7011da177e4SLinus Torvalds __vma_unlink(struct mm_struct *mm, struct vm_area_struct *vma, 7021da177e4SLinus Torvalds struct vm_area_struct *prev) 7031da177e4SLinus Torvalds { 704d3737187SMichel Lespinasse struct vm_area_struct *next; 705297c5eeeSLinus Torvalds 706d3737187SMichel Lespinasse vma_rb_erase(vma, &mm->mm_rb); 707d3737187SMichel Lespinasse prev->vm_next = next = vma->vm_next; 708297c5eeeSLinus Torvalds if (next) 709297c5eeeSLinus Torvalds next->vm_prev = prev; 710615d6e87SDavidlohr Bueso 711615d6e87SDavidlohr Bueso /* Kill the cache */ 712615d6e87SDavidlohr Bueso vmacache_invalidate(mm); 7131da177e4SLinus Torvalds } 7141da177e4SLinus Torvalds 7151da177e4SLinus Torvalds /* 7161da177e4SLinus Torvalds * We cannot adjust vm_start, vm_end, vm_pgoff fields of a vma that 7171da177e4SLinus Torvalds * is already present in an i_mmap tree without adjusting the tree. 7181da177e4SLinus Torvalds * The following helper function should be used when such adjustments 7191da177e4SLinus Torvalds * are necessary. The "insert" vma (if any) is to be inserted 7201da177e4SLinus Torvalds * before we drop the necessary locks. 7211da177e4SLinus Torvalds */ 7225beb4930SRik van Riel int vma_adjust(struct vm_area_struct *vma, unsigned long start, 7231da177e4SLinus Torvalds unsigned long end, pgoff_t pgoff, struct vm_area_struct *insert) 7241da177e4SLinus Torvalds { 7251da177e4SLinus Torvalds struct mm_struct *mm = vma->vm_mm; 7261da177e4SLinus Torvalds struct vm_area_struct *next = vma->vm_next; 7271da177e4SLinus Torvalds struct vm_area_struct *importer = NULL; 7281da177e4SLinus Torvalds struct address_space *mapping = NULL; 7296b2dbba8SMichel Lespinasse struct rb_root *root = NULL; 730012f1800SRik van Riel struct anon_vma *anon_vma = NULL; 7311da177e4SLinus Torvalds struct file *file = vma->vm_file; 732d3737187SMichel Lespinasse bool start_changed = false, end_changed = false; 7331da177e4SLinus Torvalds long adjust_next = 0; 7341da177e4SLinus Torvalds int remove_next = 0; 7351da177e4SLinus Torvalds 7361da177e4SLinus Torvalds if (next && !insert) { 737287d97acSLinus Torvalds struct vm_area_struct *exporter = NULL; 738287d97acSLinus Torvalds 7391da177e4SLinus Torvalds if (end >= next->vm_end) { 7401da177e4SLinus Torvalds /* 7411da177e4SLinus Torvalds * vma expands, overlapping all the next, and 7421da177e4SLinus Torvalds * perhaps the one after too (mprotect case 6). 7431da177e4SLinus Torvalds */ 7441da177e4SLinus Torvalds again: remove_next = 1 + (end > next->vm_end); 7451da177e4SLinus Torvalds end = next->vm_end; 746287d97acSLinus Torvalds exporter = next; 7471da177e4SLinus Torvalds importer = vma; 7481da177e4SLinus Torvalds } else if (end > next->vm_start) { 7491da177e4SLinus Torvalds /* 7501da177e4SLinus Torvalds * vma expands, overlapping part of the next: 7511da177e4SLinus Torvalds * mprotect case 5 shifting the boundary up. 7521da177e4SLinus Torvalds */ 7531da177e4SLinus Torvalds adjust_next = (end - next->vm_start) >> PAGE_SHIFT; 754287d97acSLinus Torvalds exporter = next; 7551da177e4SLinus Torvalds importer = vma; 7561da177e4SLinus Torvalds } else if (end < vma->vm_end) { 7571da177e4SLinus Torvalds /* 7581da177e4SLinus Torvalds * vma shrinks, and !insert tells it's not 7591da177e4SLinus Torvalds * split_vma inserting another: so it must be 7601da177e4SLinus Torvalds * mprotect case 4 shifting the boundary down. 7611da177e4SLinus Torvalds */ 7621da177e4SLinus Torvalds adjust_next = -((vma->vm_end - end) >> PAGE_SHIFT); 763287d97acSLinus Torvalds exporter = vma; 7641da177e4SLinus Torvalds importer = next; 7651da177e4SLinus Torvalds } 7661da177e4SLinus Torvalds 7675beb4930SRik van Riel /* 7685beb4930SRik van Riel * Easily overlooked: when mprotect shifts the boundary, 7695beb4930SRik van Riel * make sure the expanding vma has anon_vma set if the 7705beb4930SRik van Riel * shrinking vma had, to cover any anon pages imported. 7715beb4930SRik van Riel */ 772287d97acSLinus Torvalds if (exporter && exporter->anon_vma && !importer->anon_vma) { 773c4ea95d7SDaniel Forrest int error; 774c4ea95d7SDaniel Forrest 775287d97acSLinus Torvalds importer->anon_vma = exporter->anon_vma; 776b800c91aSKonstantin Khlebnikov error = anon_vma_clone(importer, exporter); 777b800c91aSKonstantin Khlebnikov if (error) { 778b800c91aSKonstantin Khlebnikov importer->anon_vma = NULL; 779b800c91aSKonstantin Khlebnikov return error; 780b800c91aSKonstantin Khlebnikov } 7815beb4930SRik van Riel } 7825beb4930SRik van Riel } 7835beb4930SRik van Riel 7841da177e4SLinus Torvalds if (file) { 7851da177e4SLinus Torvalds mapping = file->f_mapping; 7861da177e4SLinus Torvalds root = &mapping->i_mmap; 787cbc91f71SSrikar Dronamraju uprobe_munmap(vma, vma->vm_start, vma->vm_end); 788682968e0SSrikar Dronamraju 789682968e0SSrikar Dronamraju if (adjust_next) 790*27ba0644SKirill A. Shutemov uprobe_munmap(next, next->vm_start, next->vm_end); 791682968e0SSrikar Dronamraju 79283cde9e8SDavidlohr Bueso i_mmap_lock_write(mapping); 7931da177e4SLinus Torvalds if (insert) { 7941da177e4SLinus Torvalds /* 7956b2dbba8SMichel Lespinasse * Put into interval tree now, so instantiated pages 7961da177e4SLinus Torvalds * are visible to arm/parisc __flush_dcache_page 7971da177e4SLinus Torvalds * throughout; but we cannot insert into address 7981da177e4SLinus Torvalds * space until vma start or end is updated. 7991da177e4SLinus Torvalds */ 8001da177e4SLinus Torvalds __vma_link_file(insert); 8011da177e4SLinus Torvalds } 8021da177e4SLinus Torvalds } 8031da177e4SLinus Torvalds 80494fcc585SAndrea Arcangeli vma_adjust_trans_huge(vma, start, end, adjust_next); 80594fcc585SAndrea Arcangeli 806012f1800SRik van Riel anon_vma = vma->anon_vma; 807bf181b9fSMichel Lespinasse if (!anon_vma && adjust_next) 808bf181b9fSMichel Lespinasse anon_vma = next->anon_vma; 809bf181b9fSMichel Lespinasse if (anon_vma) { 81081d1b09cSSasha Levin VM_BUG_ON_VMA(adjust_next && next->anon_vma && 81181d1b09cSSasha Levin anon_vma != next->anon_vma, next); 8124fc3f1d6SIngo Molnar anon_vma_lock_write(anon_vma); 813bf181b9fSMichel Lespinasse anon_vma_interval_tree_pre_update_vma(vma); 814bf181b9fSMichel Lespinasse if (adjust_next) 815bf181b9fSMichel Lespinasse anon_vma_interval_tree_pre_update_vma(next); 816bf181b9fSMichel Lespinasse } 817012f1800SRik van Riel 8181da177e4SLinus Torvalds if (root) { 8191da177e4SLinus Torvalds flush_dcache_mmap_lock(mapping); 8206b2dbba8SMichel Lespinasse vma_interval_tree_remove(vma, root); 8211da177e4SLinus Torvalds if (adjust_next) 8226b2dbba8SMichel Lespinasse vma_interval_tree_remove(next, root); 8231da177e4SLinus Torvalds } 8241da177e4SLinus Torvalds 825d3737187SMichel Lespinasse if (start != vma->vm_start) { 8261da177e4SLinus Torvalds vma->vm_start = start; 827d3737187SMichel Lespinasse start_changed = true; 828d3737187SMichel Lespinasse } 829d3737187SMichel Lespinasse if (end != vma->vm_end) { 8301da177e4SLinus Torvalds vma->vm_end = end; 831d3737187SMichel Lespinasse end_changed = true; 832d3737187SMichel Lespinasse } 8331da177e4SLinus Torvalds vma->vm_pgoff = pgoff; 8341da177e4SLinus Torvalds if (adjust_next) { 8351da177e4SLinus Torvalds next->vm_start += adjust_next << PAGE_SHIFT; 8361da177e4SLinus Torvalds next->vm_pgoff += adjust_next; 8371da177e4SLinus Torvalds } 8381da177e4SLinus Torvalds 8391da177e4SLinus Torvalds if (root) { 8401da177e4SLinus Torvalds if (adjust_next) 8416b2dbba8SMichel Lespinasse vma_interval_tree_insert(next, root); 8426b2dbba8SMichel Lespinasse vma_interval_tree_insert(vma, root); 8431da177e4SLinus Torvalds flush_dcache_mmap_unlock(mapping); 8441da177e4SLinus Torvalds } 8451da177e4SLinus Torvalds 8461da177e4SLinus Torvalds if (remove_next) { 8471da177e4SLinus Torvalds /* 8481da177e4SLinus Torvalds * vma_merge has merged next into vma, and needs 8491da177e4SLinus Torvalds * us to remove next before dropping the locks. 8501da177e4SLinus Torvalds */ 8511da177e4SLinus Torvalds __vma_unlink(mm, next, vma); 8521da177e4SLinus Torvalds if (file) 8531da177e4SLinus Torvalds __remove_shared_vm_struct(next, file, mapping); 8541da177e4SLinus Torvalds } else if (insert) { 8551da177e4SLinus Torvalds /* 8561da177e4SLinus Torvalds * split_vma has split insert from vma, and needs 8571da177e4SLinus Torvalds * us to insert it before dropping the locks 8581da177e4SLinus Torvalds * (it may either follow vma or precede it). 8591da177e4SLinus Torvalds */ 8601da177e4SLinus Torvalds __insert_vm_struct(mm, insert); 861d3737187SMichel Lespinasse } else { 862d3737187SMichel Lespinasse if (start_changed) 863d3737187SMichel Lespinasse vma_gap_update(vma); 864d3737187SMichel Lespinasse if (end_changed) { 865d3737187SMichel Lespinasse if (!next) 866d3737187SMichel Lespinasse mm->highest_vm_end = end; 867d3737187SMichel Lespinasse else if (!adjust_next) 868d3737187SMichel Lespinasse vma_gap_update(next); 869d3737187SMichel Lespinasse } 8701da177e4SLinus Torvalds } 8711da177e4SLinus Torvalds 872bf181b9fSMichel Lespinasse if (anon_vma) { 873bf181b9fSMichel Lespinasse anon_vma_interval_tree_post_update_vma(vma); 874bf181b9fSMichel Lespinasse if (adjust_next) 875bf181b9fSMichel Lespinasse anon_vma_interval_tree_post_update_vma(next); 87608b52706SKonstantin Khlebnikov anon_vma_unlock_write(anon_vma); 877bf181b9fSMichel Lespinasse } 8781da177e4SLinus Torvalds if (mapping) 87983cde9e8SDavidlohr Bueso i_mmap_unlock_write(mapping); 8801da177e4SLinus Torvalds 8812b144498SSrikar Dronamraju if (root) { 8827b2d81d4SIngo Molnar uprobe_mmap(vma); 8832b144498SSrikar Dronamraju 8842b144498SSrikar Dronamraju if (adjust_next) 8857b2d81d4SIngo Molnar uprobe_mmap(next); 8862b144498SSrikar Dronamraju } 8872b144498SSrikar Dronamraju 8881da177e4SLinus Torvalds if (remove_next) { 889925d1c40SMatt Helsley if (file) { 890cbc91f71SSrikar Dronamraju uprobe_munmap(next, next->vm_start, next->vm_end); 8911da177e4SLinus Torvalds fput(file); 892925d1c40SMatt Helsley } 8935beb4930SRik van Riel if (next->anon_vma) 8945beb4930SRik van Riel anon_vma_merge(vma, next); 8951da177e4SLinus Torvalds mm->map_count--; 8963964acd0SOleg Nesterov mpol_put(vma_policy(next)); 8971da177e4SLinus Torvalds kmem_cache_free(vm_area_cachep, next); 8981da177e4SLinus Torvalds /* 8991da177e4SLinus Torvalds * In mprotect's case 6 (see comments on vma_merge), 9001da177e4SLinus Torvalds * we must remove another next too. It would clutter 9011da177e4SLinus Torvalds * up the code too much to do both in one go. 9021da177e4SLinus Torvalds */ 9031da177e4SLinus Torvalds next = vma->vm_next; 904d3737187SMichel Lespinasse if (remove_next == 2) 9051da177e4SLinus Torvalds goto again; 906d3737187SMichel Lespinasse else if (next) 907d3737187SMichel Lespinasse vma_gap_update(next); 908d3737187SMichel Lespinasse else 909d3737187SMichel Lespinasse mm->highest_vm_end = end; 9101da177e4SLinus Torvalds } 9112b144498SSrikar Dronamraju if (insert && file) 9127b2d81d4SIngo Molnar uprobe_mmap(insert); 9131da177e4SLinus Torvalds 9141da177e4SLinus Torvalds validate_mm(mm); 9155beb4930SRik van Riel 9165beb4930SRik van Riel return 0; 9171da177e4SLinus Torvalds } 9181da177e4SLinus Torvalds 9191da177e4SLinus Torvalds /* 9201da177e4SLinus Torvalds * If the vma has a ->close operation then the driver probably needs to release 9211da177e4SLinus Torvalds * per-vma resources, so we don't attempt to merge those. 9221da177e4SLinus Torvalds */ 9231da177e4SLinus Torvalds static inline int is_mergeable_vma(struct vm_area_struct *vma, 9241da177e4SLinus Torvalds struct file *file, unsigned long vm_flags) 9251da177e4SLinus Torvalds { 92634228d47SCyrill Gorcunov /* 92734228d47SCyrill Gorcunov * VM_SOFTDIRTY should not prevent from VMA merging, if we 92834228d47SCyrill Gorcunov * match the flags but dirty bit -- the caller should mark 92934228d47SCyrill Gorcunov * merged VMA as dirty. If dirty bit won't be excluded from 93034228d47SCyrill Gorcunov * comparison, we increase pressue on the memory system forcing 93134228d47SCyrill Gorcunov * the kernel to generate new VMAs when old one could be 93234228d47SCyrill Gorcunov * extended instead. 93334228d47SCyrill Gorcunov */ 93434228d47SCyrill Gorcunov if ((vma->vm_flags ^ vm_flags) & ~VM_SOFTDIRTY) 9351da177e4SLinus Torvalds return 0; 9361da177e4SLinus Torvalds if (vma->vm_file != file) 9371da177e4SLinus Torvalds return 0; 9381da177e4SLinus Torvalds if (vma->vm_ops && vma->vm_ops->close) 9391da177e4SLinus Torvalds return 0; 9401da177e4SLinus Torvalds return 1; 9411da177e4SLinus Torvalds } 9421da177e4SLinus Torvalds 9431da177e4SLinus Torvalds static inline int is_mergeable_anon_vma(struct anon_vma *anon_vma1, 944965f55deSShaohua Li struct anon_vma *anon_vma2, 945965f55deSShaohua Li struct vm_area_struct *vma) 9461da177e4SLinus Torvalds { 947965f55deSShaohua Li /* 948965f55deSShaohua Li * The list_is_singular() test is to avoid merging VMA cloned from 949965f55deSShaohua Li * parents. This can improve scalability caused by anon_vma lock. 950965f55deSShaohua Li */ 951965f55deSShaohua Li if ((!anon_vma1 || !anon_vma2) && (!vma || 952965f55deSShaohua Li list_is_singular(&vma->anon_vma_chain))) 953965f55deSShaohua Li return 1; 954965f55deSShaohua Li return anon_vma1 == anon_vma2; 9551da177e4SLinus Torvalds } 9561da177e4SLinus Torvalds 9571da177e4SLinus Torvalds /* 9581da177e4SLinus Torvalds * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff) 9591da177e4SLinus Torvalds * in front of (at a lower virtual address and file offset than) the vma. 9601da177e4SLinus Torvalds * 9611da177e4SLinus Torvalds * We cannot merge two vmas if they have differently assigned (non-NULL) 9621da177e4SLinus Torvalds * anon_vmas, nor if same anon_vma is assigned but offsets incompatible. 9631da177e4SLinus Torvalds * 9641da177e4SLinus Torvalds * We don't check here for the merged mmap wrapping around the end of pagecache 9651da177e4SLinus Torvalds * indices (16TB on ia32) because do_mmap_pgoff() does not permit mmap's which 9661da177e4SLinus Torvalds * wrap, nor mmaps which cover the final page at index -1UL. 9671da177e4SLinus Torvalds */ 9681da177e4SLinus Torvalds static int 9691da177e4SLinus Torvalds can_vma_merge_before(struct vm_area_struct *vma, unsigned long vm_flags, 9701da177e4SLinus Torvalds struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff) 9711da177e4SLinus Torvalds { 9721da177e4SLinus Torvalds if (is_mergeable_vma(vma, file, vm_flags) && 973965f55deSShaohua Li is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) { 9741da177e4SLinus Torvalds if (vma->vm_pgoff == vm_pgoff) 9751da177e4SLinus Torvalds return 1; 9761da177e4SLinus Torvalds } 9771da177e4SLinus Torvalds return 0; 9781da177e4SLinus Torvalds } 9791da177e4SLinus Torvalds 9801da177e4SLinus Torvalds /* 9811da177e4SLinus Torvalds * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff) 9821da177e4SLinus Torvalds * beyond (at a higher virtual address and file offset than) the vma. 9831da177e4SLinus Torvalds * 9841da177e4SLinus Torvalds * We cannot merge two vmas if they have differently assigned (non-NULL) 9851da177e4SLinus Torvalds * anon_vmas, nor if same anon_vma is assigned but offsets incompatible. 9861da177e4SLinus Torvalds */ 9871da177e4SLinus Torvalds static int 9881da177e4SLinus Torvalds can_vma_merge_after(struct vm_area_struct *vma, unsigned long vm_flags, 9891da177e4SLinus Torvalds struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff) 9901da177e4SLinus Torvalds { 9911da177e4SLinus Torvalds if (is_mergeable_vma(vma, file, vm_flags) && 992965f55deSShaohua Li is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) { 9931da177e4SLinus Torvalds pgoff_t vm_pglen; 994d6e93217SLibin vm_pglen = vma_pages(vma); 9951da177e4SLinus Torvalds if (vma->vm_pgoff + vm_pglen == vm_pgoff) 9961da177e4SLinus Torvalds return 1; 9971da177e4SLinus Torvalds } 9981da177e4SLinus Torvalds return 0; 9991da177e4SLinus Torvalds } 10001da177e4SLinus Torvalds 10011da177e4SLinus Torvalds /* 10021da177e4SLinus Torvalds * Given a mapping request (addr,end,vm_flags,file,pgoff), figure out 10031da177e4SLinus Torvalds * whether that can be merged with its predecessor or its successor. 10041da177e4SLinus Torvalds * Or both (it neatly fills a hole). 10051da177e4SLinus Torvalds * 10061da177e4SLinus Torvalds * In most cases - when called for mmap, brk or mremap - [addr,end) is 10071da177e4SLinus Torvalds * certain not to be mapped by the time vma_merge is called; but when 10081da177e4SLinus Torvalds * called for mprotect, it is certain to be already mapped (either at 10091da177e4SLinus Torvalds * an offset within prev, or at the start of next), and the flags of 10101da177e4SLinus Torvalds * this area are about to be changed to vm_flags - and the no-change 10111da177e4SLinus Torvalds * case has already been eliminated. 10121da177e4SLinus Torvalds * 10131da177e4SLinus Torvalds * The following mprotect cases have to be considered, where AAAA is 10141da177e4SLinus Torvalds * the area passed down from mprotect_fixup, never extending beyond one 10151da177e4SLinus Torvalds * vma, PPPPPP is the prev vma specified, and NNNNNN the next vma after: 10161da177e4SLinus Torvalds * 10171da177e4SLinus Torvalds * AAAA AAAA AAAA AAAA 10181da177e4SLinus Torvalds * PPPPPPNNNNNN PPPPPPNNNNNN PPPPPPNNNNNN PPPPNNNNXXXX 10191da177e4SLinus Torvalds * cannot merge might become might become might become 10201da177e4SLinus Torvalds * PPNNNNNNNNNN PPPPPPPPPPNN PPPPPPPPPPPP 6 or 10211da177e4SLinus Torvalds * mmap, brk or case 4 below case 5 below PPPPPPPPXXXX 7 or 10221da177e4SLinus Torvalds * mremap move: PPPPNNNNNNNN 8 10231da177e4SLinus Torvalds * AAAA 10241da177e4SLinus Torvalds * PPPP NNNN PPPPPPPPPPPP PPPPPPPPNNNN PPPPNNNNNNNN 10251da177e4SLinus Torvalds * might become case 1 below case 2 below case 3 below 10261da177e4SLinus Torvalds * 10271da177e4SLinus Torvalds * Odd one out? Case 8, because it extends NNNN but needs flags of XXXX: 10281da177e4SLinus Torvalds * mprotect_fixup updates vm_flags & vm_page_prot on successful return. 10291da177e4SLinus Torvalds */ 10301da177e4SLinus Torvalds struct vm_area_struct *vma_merge(struct mm_struct *mm, 10311da177e4SLinus Torvalds struct vm_area_struct *prev, unsigned long addr, 10321da177e4SLinus Torvalds unsigned long end, unsigned long vm_flags, 10331da177e4SLinus Torvalds struct anon_vma *anon_vma, struct file *file, 10341da177e4SLinus Torvalds pgoff_t pgoff, struct mempolicy *policy) 10351da177e4SLinus Torvalds { 10361da177e4SLinus Torvalds pgoff_t pglen = (end - addr) >> PAGE_SHIFT; 10371da177e4SLinus Torvalds struct vm_area_struct *area, *next; 10385beb4930SRik van Riel int err; 10391da177e4SLinus Torvalds 10401da177e4SLinus Torvalds /* 10411da177e4SLinus Torvalds * We later require that vma->vm_flags == vm_flags, 10421da177e4SLinus Torvalds * so this tests vma->vm_flags & VM_SPECIAL, too. 10431da177e4SLinus Torvalds */ 10441da177e4SLinus Torvalds if (vm_flags & VM_SPECIAL) 10451da177e4SLinus Torvalds return NULL; 10461da177e4SLinus Torvalds 10471da177e4SLinus Torvalds if (prev) 10481da177e4SLinus Torvalds next = prev->vm_next; 10491da177e4SLinus Torvalds else 10501da177e4SLinus Torvalds next = mm->mmap; 10511da177e4SLinus Torvalds area = next; 10521da177e4SLinus Torvalds if (next && next->vm_end == end) /* cases 6, 7, 8 */ 10531da177e4SLinus Torvalds next = next->vm_next; 10541da177e4SLinus Torvalds 10551da177e4SLinus Torvalds /* 10561da177e4SLinus Torvalds * Can it merge with the predecessor? 10571da177e4SLinus Torvalds */ 10581da177e4SLinus Torvalds if (prev && prev->vm_end == addr && 10591da177e4SLinus Torvalds mpol_equal(vma_policy(prev), policy) && 10601da177e4SLinus Torvalds can_vma_merge_after(prev, vm_flags, 10611da177e4SLinus Torvalds anon_vma, file, pgoff)) { 10621da177e4SLinus Torvalds /* 10631da177e4SLinus Torvalds * OK, it can. Can we now merge in the successor as well? 10641da177e4SLinus Torvalds */ 10651da177e4SLinus Torvalds if (next && end == next->vm_start && 10661da177e4SLinus Torvalds mpol_equal(policy, vma_policy(next)) && 10671da177e4SLinus Torvalds can_vma_merge_before(next, vm_flags, 10681da177e4SLinus Torvalds anon_vma, file, pgoff+pglen) && 10691da177e4SLinus Torvalds is_mergeable_anon_vma(prev->anon_vma, 1070965f55deSShaohua Li next->anon_vma, NULL)) { 10711da177e4SLinus Torvalds /* cases 1, 6 */ 10725beb4930SRik van Riel err = vma_adjust(prev, prev->vm_start, 10731da177e4SLinus Torvalds next->vm_end, prev->vm_pgoff, NULL); 10741da177e4SLinus Torvalds } else /* cases 2, 5, 7 */ 10755beb4930SRik van Riel err = vma_adjust(prev, prev->vm_start, 10761da177e4SLinus Torvalds end, prev->vm_pgoff, NULL); 10775beb4930SRik van Riel if (err) 10785beb4930SRik van Riel return NULL; 10796d50e60cSDavid Rientjes khugepaged_enter_vma_merge(prev, vm_flags); 10801da177e4SLinus Torvalds return prev; 10811da177e4SLinus Torvalds } 10821da177e4SLinus Torvalds 10831da177e4SLinus Torvalds /* 10841da177e4SLinus Torvalds * Can this new request be merged in front of next? 10851da177e4SLinus Torvalds */ 10861da177e4SLinus Torvalds if (next && end == next->vm_start && 10871da177e4SLinus Torvalds mpol_equal(policy, vma_policy(next)) && 10881da177e4SLinus Torvalds can_vma_merge_before(next, vm_flags, 10891da177e4SLinus Torvalds anon_vma, file, pgoff+pglen)) { 10901da177e4SLinus Torvalds if (prev && addr < prev->vm_end) /* case 4 */ 10915beb4930SRik van Riel err = vma_adjust(prev, prev->vm_start, 10921da177e4SLinus Torvalds addr, prev->vm_pgoff, NULL); 10931da177e4SLinus Torvalds else /* cases 3, 8 */ 10945beb4930SRik van Riel err = vma_adjust(area, addr, next->vm_end, 10951da177e4SLinus Torvalds next->vm_pgoff - pglen, NULL); 10965beb4930SRik van Riel if (err) 10975beb4930SRik van Riel return NULL; 10986d50e60cSDavid Rientjes khugepaged_enter_vma_merge(area, vm_flags); 10991da177e4SLinus Torvalds return area; 11001da177e4SLinus Torvalds } 11011da177e4SLinus Torvalds 11021da177e4SLinus Torvalds return NULL; 11031da177e4SLinus Torvalds } 11041da177e4SLinus Torvalds 11051da177e4SLinus Torvalds /* 1106d0e9fe17SLinus Torvalds * Rough compatbility check to quickly see if it's even worth looking 1107d0e9fe17SLinus Torvalds * at sharing an anon_vma. 1108d0e9fe17SLinus Torvalds * 1109d0e9fe17SLinus Torvalds * They need to have the same vm_file, and the flags can only differ 1110d0e9fe17SLinus Torvalds * in things that mprotect may change. 1111d0e9fe17SLinus Torvalds * 1112d0e9fe17SLinus Torvalds * NOTE! The fact that we share an anon_vma doesn't _have_ to mean that 1113d0e9fe17SLinus Torvalds * we can merge the two vma's. For example, we refuse to merge a vma if 1114d0e9fe17SLinus Torvalds * there is a vm_ops->close() function, because that indicates that the 1115d0e9fe17SLinus Torvalds * driver is doing some kind of reference counting. But that doesn't 1116d0e9fe17SLinus Torvalds * really matter for the anon_vma sharing case. 1117d0e9fe17SLinus Torvalds */ 1118d0e9fe17SLinus Torvalds static int anon_vma_compatible(struct vm_area_struct *a, struct vm_area_struct *b) 1119d0e9fe17SLinus Torvalds { 1120d0e9fe17SLinus Torvalds return a->vm_end == b->vm_start && 1121d0e9fe17SLinus Torvalds mpol_equal(vma_policy(a), vma_policy(b)) && 1122d0e9fe17SLinus Torvalds a->vm_file == b->vm_file && 112334228d47SCyrill Gorcunov !((a->vm_flags ^ b->vm_flags) & ~(VM_READ|VM_WRITE|VM_EXEC|VM_SOFTDIRTY)) && 1124d0e9fe17SLinus Torvalds b->vm_pgoff == a->vm_pgoff + ((b->vm_start - a->vm_start) >> PAGE_SHIFT); 1125d0e9fe17SLinus Torvalds } 1126d0e9fe17SLinus Torvalds 1127d0e9fe17SLinus Torvalds /* 1128d0e9fe17SLinus Torvalds * Do some basic sanity checking to see if we can re-use the anon_vma 1129d0e9fe17SLinus Torvalds * from 'old'. The 'a'/'b' vma's are in VM order - one of them will be 1130d0e9fe17SLinus Torvalds * the same as 'old', the other will be the new one that is trying 1131d0e9fe17SLinus Torvalds * to share the anon_vma. 1132d0e9fe17SLinus Torvalds * 1133d0e9fe17SLinus Torvalds * NOTE! This runs with mm_sem held for reading, so it is possible that 1134d0e9fe17SLinus Torvalds * the anon_vma of 'old' is concurrently in the process of being set up 1135d0e9fe17SLinus Torvalds * by another page fault trying to merge _that_. But that's ok: if it 1136d0e9fe17SLinus Torvalds * is being set up, that automatically means that it will be a singleton 1137d0e9fe17SLinus Torvalds * acceptable for merging, so we can do all of this optimistically. But 1138d0e9fe17SLinus Torvalds * we do that ACCESS_ONCE() to make sure that we never re-load the pointer. 1139d0e9fe17SLinus Torvalds * 1140d0e9fe17SLinus Torvalds * IOW: that the "list_is_singular()" test on the anon_vma_chain only 1141d0e9fe17SLinus Torvalds * matters for the 'stable anon_vma' case (ie the thing we want to avoid 1142d0e9fe17SLinus Torvalds * is to return an anon_vma that is "complex" due to having gone through 1143d0e9fe17SLinus Torvalds * a fork). 1144d0e9fe17SLinus Torvalds * 1145d0e9fe17SLinus Torvalds * We also make sure that the two vma's are compatible (adjacent, 1146d0e9fe17SLinus Torvalds * and with the same memory policies). That's all stable, even with just 1147d0e9fe17SLinus Torvalds * a read lock on the mm_sem. 1148d0e9fe17SLinus Torvalds */ 1149d0e9fe17SLinus Torvalds static struct anon_vma *reusable_anon_vma(struct vm_area_struct *old, struct vm_area_struct *a, struct vm_area_struct *b) 1150d0e9fe17SLinus Torvalds { 1151d0e9fe17SLinus Torvalds if (anon_vma_compatible(a, b)) { 1152d0e9fe17SLinus Torvalds struct anon_vma *anon_vma = ACCESS_ONCE(old->anon_vma); 1153d0e9fe17SLinus Torvalds 1154d0e9fe17SLinus Torvalds if (anon_vma && list_is_singular(&old->anon_vma_chain)) 1155d0e9fe17SLinus Torvalds return anon_vma; 1156d0e9fe17SLinus Torvalds } 1157d0e9fe17SLinus Torvalds return NULL; 1158d0e9fe17SLinus Torvalds } 1159d0e9fe17SLinus Torvalds 1160d0e9fe17SLinus Torvalds /* 11611da177e4SLinus Torvalds * find_mergeable_anon_vma is used by anon_vma_prepare, to check 11621da177e4SLinus Torvalds * neighbouring vmas for a suitable anon_vma, before it goes off 11631da177e4SLinus Torvalds * to allocate a new anon_vma. It checks because a repetitive 11641da177e4SLinus Torvalds * sequence of mprotects and faults may otherwise lead to distinct 11651da177e4SLinus Torvalds * anon_vmas being allocated, preventing vma merge in subsequent 11661da177e4SLinus Torvalds * mprotect. 11671da177e4SLinus Torvalds */ 11681da177e4SLinus Torvalds struct anon_vma *find_mergeable_anon_vma(struct vm_area_struct *vma) 11691da177e4SLinus Torvalds { 1170d0e9fe17SLinus Torvalds struct anon_vma *anon_vma; 11711da177e4SLinus Torvalds struct vm_area_struct *near; 11721da177e4SLinus Torvalds 11731da177e4SLinus Torvalds near = vma->vm_next; 11741da177e4SLinus Torvalds if (!near) 11751da177e4SLinus Torvalds goto try_prev; 11761da177e4SLinus Torvalds 1177d0e9fe17SLinus Torvalds anon_vma = reusable_anon_vma(near, vma, near); 1178d0e9fe17SLinus Torvalds if (anon_vma) 1179d0e9fe17SLinus Torvalds return anon_vma; 11801da177e4SLinus Torvalds try_prev: 11819be34c9dSLinus Torvalds near = vma->vm_prev; 11821da177e4SLinus Torvalds if (!near) 11831da177e4SLinus Torvalds goto none; 11841da177e4SLinus Torvalds 1185d0e9fe17SLinus Torvalds anon_vma = reusable_anon_vma(near, near, vma); 1186d0e9fe17SLinus Torvalds if (anon_vma) 1187d0e9fe17SLinus Torvalds return anon_vma; 11881da177e4SLinus Torvalds none: 11891da177e4SLinus Torvalds /* 11901da177e4SLinus Torvalds * There's no absolute need to look only at touching neighbours: 11911da177e4SLinus Torvalds * we could search further afield for "compatible" anon_vmas. 11921da177e4SLinus Torvalds * But it would probably just be a waste of time searching, 11931da177e4SLinus Torvalds * or lead to too many vmas hanging off the same anon_vma. 11941da177e4SLinus Torvalds * We're trying to allow mprotect remerging later on, 11951da177e4SLinus Torvalds * not trying to minimize memory used for anon_vmas. 11961da177e4SLinus Torvalds */ 11971da177e4SLinus Torvalds return NULL; 11981da177e4SLinus Torvalds } 11991da177e4SLinus Torvalds 12001da177e4SLinus Torvalds #ifdef CONFIG_PROC_FS 1201ab50b8edSHugh Dickins void vm_stat_account(struct mm_struct *mm, unsigned long flags, 12021da177e4SLinus Torvalds struct file *file, long pages) 12031da177e4SLinus Torvalds { 12041da177e4SLinus Torvalds const unsigned long stack_flags 12051da177e4SLinus Torvalds = VM_STACK_FLAGS & (VM_GROWSUP|VM_GROWSDOWN); 12061da177e4SLinus Torvalds 120744de9d0cSHuang Shijie mm->total_vm += pages; 120844de9d0cSHuang Shijie 12091da177e4SLinus Torvalds if (file) { 12101da177e4SLinus Torvalds mm->shared_vm += pages; 12111da177e4SLinus Torvalds if ((flags & (VM_EXEC|VM_WRITE)) == VM_EXEC) 12121da177e4SLinus Torvalds mm->exec_vm += pages; 12131da177e4SLinus Torvalds } else if (flags & stack_flags) 12141da177e4SLinus Torvalds mm->stack_vm += pages; 12151da177e4SLinus Torvalds } 12161da177e4SLinus Torvalds #endif /* CONFIG_PROC_FS */ 12171da177e4SLinus Torvalds 12181da177e4SLinus Torvalds /* 121940401530SAl Viro * If a hint addr is less than mmap_min_addr change hint to be as 122040401530SAl Viro * low as possible but still greater than mmap_min_addr 122140401530SAl Viro */ 122240401530SAl Viro static inline unsigned long round_hint_to_min(unsigned long hint) 122340401530SAl Viro { 122440401530SAl Viro hint &= PAGE_MASK; 122540401530SAl Viro if (((void *)hint != NULL) && 122640401530SAl Viro (hint < mmap_min_addr)) 122740401530SAl Viro return PAGE_ALIGN(mmap_min_addr); 122840401530SAl Viro return hint; 122940401530SAl Viro } 123040401530SAl Viro 1231363ee17fSDavidlohr Bueso static inline int mlock_future_check(struct mm_struct *mm, 1232363ee17fSDavidlohr Bueso unsigned long flags, 1233363ee17fSDavidlohr Bueso unsigned long len) 1234363ee17fSDavidlohr Bueso { 1235363ee17fSDavidlohr Bueso unsigned long locked, lock_limit; 1236363ee17fSDavidlohr Bueso 1237363ee17fSDavidlohr Bueso /* mlock MCL_FUTURE? */ 1238363ee17fSDavidlohr Bueso if (flags & VM_LOCKED) { 1239363ee17fSDavidlohr Bueso locked = len >> PAGE_SHIFT; 1240363ee17fSDavidlohr Bueso locked += mm->locked_vm; 1241363ee17fSDavidlohr Bueso lock_limit = rlimit(RLIMIT_MEMLOCK); 1242363ee17fSDavidlohr Bueso lock_limit >>= PAGE_SHIFT; 1243363ee17fSDavidlohr Bueso if (locked > lock_limit && !capable(CAP_IPC_LOCK)) 1244363ee17fSDavidlohr Bueso return -EAGAIN; 1245363ee17fSDavidlohr Bueso } 1246363ee17fSDavidlohr Bueso return 0; 1247363ee17fSDavidlohr Bueso } 1248363ee17fSDavidlohr Bueso 124940401530SAl Viro /* 125027f5de79SJianjun Kong * The caller must hold down_write(¤t->mm->mmap_sem). 12511da177e4SLinus Torvalds */ 12521da177e4SLinus Torvalds 1253e3fc629dSAl Viro unsigned long do_mmap_pgoff(struct file *file, unsigned long addr, 12541da177e4SLinus Torvalds unsigned long len, unsigned long prot, 1255bebeb3d6SMichel Lespinasse unsigned long flags, unsigned long pgoff, 125641badc15SMichel Lespinasse unsigned long *populate) 12571da177e4SLinus Torvalds { 12581da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 1259ca16d140SKOSAKI Motohiro vm_flags_t vm_flags; 12601da177e4SLinus Torvalds 126141badc15SMichel Lespinasse *populate = 0; 1262bebeb3d6SMichel Lespinasse 12631da177e4SLinus Torvalds /* 12641da177e4SLinus Torvalds * Does the application expect PROT_READ to imply PROT_EXEC? 12651da177e4SLinus Torvalds * 12661da177e4SLinus Torvalds * (the exception is when the underlying filesystem is noexec 12671da177e4SLinus Torvalds * mounted, in which case we dont add PROT_EXEC.) 12681da177e4SLinus Torvalds */ 12691da177e4SLinus Torvalds if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC)) 1270d3ac7f89SJosef "Jeff" Sipek if (!(file && (file->f_path.mnt->mnt_flags & MNT_NOEXEC))) 12711da177e4SLinus Torvalds prot |= PROT_EXEC; 12721da177e4SLinus Torvalds 12731da177e4SLinus Torvalds if (!len) 12741da177e4SLinus Torvalds return -EINVAL; 12751da177e4SLinus Torvalds 12767cd94146SEric Paris if (!(flags & MAP_FIXED)) 12777cd94146SEric Paris addr = round_hint_to_min(addr); 12787cd94146SEric Paris 12791da177e4SLinus Torvalds /* Careful about overflows.. */ 12801da177e4SLinus Torvalds len = PAGE_ALIGN(len); 12819206de95SAl Viro if (!len) 12821da177e4SLinus Torvalds return -ENOMEM; 12831da177e4SLinus Torvalds 12841da177e4SLinus Torvalds /* offset overflow? */ 12851da177e4SLinus Torvalds if ((pgoff + (len >> PAGE_SHIFT)) < pgoff) 12861da177e4SLinus Torvalds return -EOVERFLOW; 12871da177e4SLinus Torvalds 12881da177e4SLinus Torvalds /* Too many mappings? */ 12891da177e4SLinus Torvalds if (mm->map_count > sysctl_max_map_count) 12901da177e4SLinus Torvalds return -ENOMEM; 12911da177e4SLinus Torvalds 12921da177e4SLinus Torvalds /* Obtain the address to map to. we verify (or select) it and ensure 12931da177e4SLinus Torvalds * that it represents a valid section of the address space. 12941da177e4SLinus Torvalds */ 12951da177e4SLinus Torvalds addr = get_unmapped_area(file, addr, len, pgoff, flags); 12961da177e4SLinus Torvalds if (addr & ~PAGE_MASK) 12971da177e4SLinus Torvalds return addr; 12981da177e4SLinus Torvalds 12991da177e4SLinus Torvalds /* Do simple checking here so the lower-level routines won't have 13001da177e4SLinus Torvalds * to. we assume access permissions have been handled by the open 13011da177e4SLinus Torvalds * of the memory object, so we don't do any here. 13021da177e4SLinus Torvalds */ 13031da177e4SLinus Torvalds vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags) | 13041da177e4SLinus Torvalds mm->def_flags | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC; 13051da177e4SLinus Torvalds 1306cdf7b341SHuang Shijie if (flags & MAP_LOCKED) 13071da177e4SLinus Torvalds if (!can_do_mlock()) 13081da177e4SLinus Torvalds return -EPERM; 1309ba470de4SRik van Riel 1310363ee17fSDavidlohr Bueso if (mlock_future_check(mm, vm_flags, len)) 13111da177e4SLinus Torvalds return -EAGAIN; 13121da177e4SLinus Torvalds 13131da177e4SLinus Torvalds if (file) { 1314077bf22bSOleg Nesterov struct inode *inode = file_inode(file); 1315077bf22bSOleg Nesterov 13161da177e4SLinus Torvalds switch (flags & MAP_TYPE) { 13171da177e4SLinus Torvalds case MAP_SHARED: 13181da177e4SLinus Torvalds if ((prot&PROT_WRITE) && !(file->f_mode&FMODE_WRITE)) 13191da177e4SLinus Torvalds return -EACCES; 13201da177e4SLinus Torvalds 13211da177e4SLinus Torvalds /* 13221da177e4SLinus Torvalds * Make sure we don't allow writing to an append-only 13231da177e4SLinus Torvalds * file.. 13241da177e4SLinus Torvalds */ 13251da177e4SLinus Torvalds if (IS_APPEND(inode) && (file->f_mode & FMODE_WRITE)) 13261da177e4SLinus Torvalds return -EACCES; 13271da177e4SLinus Torvalds 13281da177e4SLinus Torvalds /* 13291da177e4SLinus Torvalds * Make sure there are no mandatory locks on the file. 13301da177e4SLinus Torvalds */ 1331d7a06983SJeff Layton if (locks_verify_locked(file)) 13321da177e4SLinus Torvalds return -EAGAIN; 13331da177e4SLinus Torvalds 13341da177e4SLinus Torvalds vm_flags |= VM_SHARED | VM_MAYSHARE; 13351da177e4SLinus Torvalds if (!(file->f_mode & FMODE_WRITE)) 13361da177e4SLinus Torvalds vm_flags &= ~(VM_MAYWRITE | VM_SHARED); 13371da177e4SLinus Torvalds 13381da177e4SLinus Torvalds /* fall through */ 13391da177e4SLinus Torvalds case MAP_PRIVATE: 13401da177e4SLinus Torvalds if (!(file->f_mode & FMODE_READ)) 13411da177e4SLinus Torvalds return -EACCES; 1342d3ac7f89SJosef "Jeff" Sipek if (file->f_path.mnt->mnt_flags & MNT_NOEXEC) { 134380c5606cSLinus Torvalds if (vm_flags & VM_EXEC) 134480c5606cSLinus Torvalds return -EPERM; 134580c5606cSLinus Torvalds vm_flags &= ~VM_MAYEXEC; 134680c5606cSLinus Torvalds } 134780c5606cSLinus Torvalds 134872c2d531SAl Viro if (!file->f_op->mmap) 134980c5606cSLinus Torvalds return -ENODEV; 1350b2c56e4fSOleg Nesterov if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP)) 1351b2c56e4fSOleg Nesterov return -EINVAL; 13521da177e4SLinus Torvalds break; 13531da177e4SLinus Torvalds 13541da177e4SLinus Torvalds default: 13551da177e4SLinus Torvalds return -EINVAL; 13561da177e4SLinus Torvalds } 13571da177e4SLinus Torvalds } else { 13581da177e4SLinus Torvalds switch (flags & MAP_TYPE) { 13591da177e4SLinus Torvalds case MAP_SHARED: 1360b2c56e4fSOleg Nesterov if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP)) 1361b2c56e4fSOleg Nesterov return -EINVAL; 1362ce363942STejun Heo /* 1363ce363942STejun Heo * Ignore pgoff. 1364ce363942STejun Heo */ 1365ce363942STejun Heo pgoff = 0; 13661da177e4SLinus Torvalds vm_flags |= VM_SHARED | VM_MAYSHARE; 13671da177e4SLinus Torvalds break; 13681da177e4SLinus Torvalds case MAP_PRIVATE: 13691da177e4SLinus Torvalds /* 13701da177e4SLinus Torvalds * Set pgoff according to addr for anon_vma. 13711da177e4SLinus Torvalds */ 13721da177e4SLinus Torvalds pgoff = addr >> PAGE_SHIFT; 13731da177e4SLinus Torvalds break; 13741da177e4SLinus Torvalds default: 13751da177e4SLinus Torvalds return -EINVAL; 13761da177e4SLinus Torvalds } 13771da177e4SLinus Torvalds } 13781da177e4SLinus Torvalds 1379c22c0d63SMichel Lespinasse /* 1380c22c0d63SMichel Lespinasse * Set 'VM_NORESERVE' if we should not account for the 1381c22c0d63SMichel Lespinasse * memory use of this mapping. 1382c22c0d63SMichel Lespinasse */ 1383c22c0d63SMichel Lespinasse if (flags & MAP_NORESERVE) { 1384c22c0d63SMichel Lespinasse /* We honor MAP_NORESERVE if allowed to overcommit */ 1385c22c0d63SMichel Lespinasse if (sysctl_overcommit_memory != OVERCOMMIT_NEVER) 1386c22c0d63SMichel Lespinasse vm_flags |= VM_NORESERVE; 1387c22c0d63SMichel Lespinasse 1388c22c0d63SMichel Lespinasse /* hugetlb applies strict overcommit unless MAP_NORESERVE */ 1389c22c0d63SMichel Lespinasse if (file && is_file_hugepages(file)) 1390c22c0d63SMichel Lespinasse vm_flags |= VM_NORESERVE; 1391c22c0d63SMichel Lespinasse } 1392c22c0d63SMichel Lespinasse 1393c22c0d63SMichel Lespinasse addr = mmap_region(file, addr, len, vm_flags, pgoff); 139409a9f1d2SMichel Lespinasse if (!IS_ERR_VALUE(addr) && 139509a9f1d2SMichel Lespinasse ((vm_flags & VM_LOCKED) || 139609a9f1d2SMichel Lespinasse (flags & (MAP_POPULATE | MAP_NONBLOCK)) == MAP_POPULATE)) 139741badc15SMichel Lespinasse *populate = len; 1398bebeb3d6SMichel Lespinasse return addr; 13990165ab44SMiklos Szeredi } 14006be5ceb0SLinus Torvalds 140166f0dc48SHugh Dickins SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len, 140266f0dc48SHugh Dickins unsigned long, prot, unsigned long, flags, 140366f0dc48SHugh Dickins unsigned long, fd, unsigned long, pgoff) 140466f0dc48SHugh Dickins { 140566f0dc48SHugh Dickins struct file *file = NULL; 140666f0dc48SHugh Dickins unsigned long retval = -EBADF; 140766f0dc48SHugh Dickins 140866f0dc48SHugh Dickins if (!(flags & MAP_ANONYMOUS)) { 1409120a795dSAl Viro audit_mmap_fd(fd, flags); 141066f0dc48SHugh Dickins file = fget(fd); 141166f0dc48SHugh Dickins if (!file) 141266f0dc48SHugh Dickins goto out; 1413af73e4d9SNaoya Horiguchi if (is_file_hugepages(file)) 1414af73e4d9SNaoya Horiguchi len = ALIGN(len, huge_page_size(hstate_file(file))); 1415493af578SJörn Engel retval = -EINVAL; 1416493af578SJörn Engel if (unlikely(flags & MAP_HUGETLB && !is_file_hugepages(file))) 1417493af578SJörn Engel goto out_fput; 141866f0dc48SHugh Dickins } else if (flags & MAP_HUGETLB) { 141966f0dc48SHugh Dickins struct user_struct *user = NULL; 1420c103a4dcSAndrew Morton struct hstate *hs; 1421af73e4d9SNaoya Horiguchi 1422c103a4dcSAndrew Morton hs = hstate_sizelog((flags >> MAP_HUGE_SHIFT) & SHM_HUGE_MASK); 1423091d0d55SLi Zefan if (!hs) 1424091d0d55SLi Zefan return -EINVAL; 1425091d0d55SLi Zefan 1426091d0d55SLi Zefan len = ALIGN(len, huge_page_size(hs)); 142766f0dc48SHugh Dickins /* 142866f0dc48SHugh Dickins * VM_NORESERVE is used because the reservations will be 142966f0dc48SHugh Dickins * taken when vm_ops->mmap() is called 143066f0dc48SHugh Dickins * A dummy user value is used because we are not locking 143166f0dc48SHugh Dickins * memory so no accounting is necessary 143266f0dc48SHugh Dickins */ 1433af73e4d9SNaoya Horiguchi file = hugetlb_file_setup(HUGETLB_ANON_FILE, len, 143442d7395fSAndi Kleen VM_NORESERVE, 143542d7395fSAndi Kleen &user, HUGETLB_ANONHUGE_INODE, 143642d7395fSAndi Kleen (flags >> MAP_HUGE_SHIFT) & MAP_HUGE_MASK); 143766f0dc48SHugh Dickins if (IS_ERR(file)) 143866f0dc48SHugh Dickins return PTR_ERR(file); 143966f0dc48SHugh Dickins } 144066f0dc48SHugh Dickins 144166f0dc48SHugh Dickins flags &= ~(MAP_EXECUTABLE | MAP_DENYWRITE); 144266f0dc48SHugh Dickins 1443eb36c587SAl Viro retval = vm_mmap_pgoff(file, addr, len, prot, flags, pgoff); 1444493af578SJörn Engel out_fput: 144566f0dc48SHugh Dickins if (file) 144666f0dc48SHugh Dickins fput(file); 144766f0dc48SHugh Dickins out: 144866f0dc48SHugh Dickins return retval; 144966f0dc48SHugh Dickins } 145066f0dc48SHugh Dickins 1451a4679373SChristoph Hellwig #ifdef __ARCH_WANT_SYS_OLD_MMAP 1452a4679373SChristoph Hellwig struct mmap_arg_struct { 1453a4679373SChristoph Hellwig unsigned long addr; 1454a4679373SChristoph Hellwig unsigned long len; 1455a4679373SChristoph Hellwig unsigned long prot; 1456a4679373SChristoph Hellwig unsigned long flags; 1457a4679373SChristoph Hellwig unsigned long fd; 1458a4679373SChristoph Hellwig unsigned long offset; 1459a4679373SChristoph Hellwig }; 1460a4679373SChristoph Hellwig 1461a4679373SChristoph Hellwig SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg) 1462a4679373SChristoph Hellwig { 1463a4679373SChristoph Hellwig struct mmap_arg_struct a; 1464a4679373SChristoph Hellwig 1465a4679373SChristoph Hellwig if (copy_from_user(&a, arg, sizeof(a))) 1466a4679373SChristoph Hellwig return -EFAULT; 1467a4679373SChristoph Hellwig if (a.offset & ~PAGE_MASK) 1468a4679373SChristoph Hellwig return -EINVAL; 1469a4679373SChristoph Hellwig 1470a4679373SChristoph Hellwig return sys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd, 1471a4679373SChristoph Hellwig a.offset >> PAGE_SHIFT); 1472a4679373SChristoph Hellwig } 1473a4679373SChristoph Hellwig #endif /* __ARCH_WANT_SYS_OLD_MMAP */ 1474a4679373SChristoph Hellwig 14754e950f6fSAlexey Dobriyan /* 14764e950f6fSAlexey Dobriyan * Some shared mappigns will want the pages marked read-only 14774e950f6fSAlexey Dobriyan * to track write events. If so, we'll downgrade vm_page_prot 14784e950f6fSAlexey Dobriyan * to the private version (using protection_map[] without the 14794e950f6fSAlexey Dobriyan * VM_SHARED bit). 14804e950f6fSAlexey Dobriyan */ 14814e950f6fSAlexey Dobriyan int vma_wants_writenotify(struct vm_area_struct *vma) 14824e950f6fSAlexey Dobriyan { 1483ca16d140SKOSAKI Motohiro vm_flags_t vm_flags = vma->vm_flags; 14844e950f6fSAlexey Dobriyan 14854e950f6fSAlexey Dobriyan /* If it was private or non-writable, the write bit is already clear */ 14864e950f6fSAlexey Dobriyan if ((vm_flags & (VM_WRITE|VM_SHARED)) != ((VM_WRITE|VM_SHARED))) 14874e950f6fSAlexey Dobriyan return 0; 14884e950f6fSAlexey Dobriyan 14894e950f6fSAlexey Dobriyan /* The backer wishes to know when pages are first written to? */ 14904e950f6fSAlexey Dobriyan if (vma->vm_ops && vma->vm_ops->page_mkwrite) 14914e950f6fSAlexey Dobriyan return 1; 14924e950f6fSAlexey Dobriyan 149364e45507SPeter Feiner /* The open routine did something to the protections that pgprot_modify 149464e45507SPeter Feiner * won't preserve? */ 14954e950f6fSAlexey Dobriyan if (pgprot_val(vma->vm_page_prot) != 149664e45507SPeter Feiner pgprot_val(vm_pgprot_modify(vma->vm_page_prot, vm_flags))) 14974e950f6fSAlexey Dobriyan return 0; 14984e950f6fSAlexey Dobriyan 149964e45507SPeter Feiner /* Do we need to track softdirty? */ 150064e45507SPeter Feiner if (IS_ENABLED(CONFIG_MEM_SOFT_DIRTY) && !(vm_flags & VM_SOFTDIRTY)) 150164e45507SPeter Feiner return 1; 150264e45507SPeter Feiner 15034e950f6fSAlexey Dobriyan /* Specialty mapping? */ 15044b6e1e37SKonstantin Khlebnikov if (vm_flags & VM_PFNMAP) 15054e950f6fSAlexey Dobriyan return 0; 15064e950f6fSAlexey Dobriyan 15074e950f6fSAlexey Dobriyan /* Can the mapping track the dirty pages? */ 15084e950f6fSAlexey Dobriyan return vma->vm_file && vma->vm_file->f_mapping && 15094e950f6fSAlexey Dobriyan mapping_cap_account_dirty(vma->vm_file->f_mapping); 15104e950f6fSAlexey Dobriyan } 15114e950f6fSAlexey Dobriyan 1512fc8744adSLinus Torvalds /* 1513fc8744adSLinus Torvalds * We account for memory if it's a private writeable mapping, 15145a6fe125SMel Gorman * not hugepages and VM_NORESERVE wasn't set. 1515fc8744adSLinus Torvalds */ 1516ca16d140SKOSAKI Motohiro static inline int accountable_mapping(struct file *file, vm_flags_t vm_flags) 1517fc8744adSLinus Torvalds { 15185a6fe125SMel Gorman /* 15195a6fe125SMel Gorman * hugetlb has its own accounting separate from the core VM 15205a6fe125SMel Gorman * VM_HUGETLB may not be set yet so we cannot check for that flag. 15215a6fe125SMel Gorman */ 15225a6fe125SMel Gorman if (file && is_file_hugepages(file)) 15235a6fe125SMel Gorman return 0; 15245a6fe125SMel Gorman 1525fc8744adSLinus Torvalds return (vm_flags & (VM_NORESERVE | VM_SHARED | VM_WRITE)) == VM_WRITE; 1526fc8744adSLinus Torvalds } 1527fc8744adSLinus Torvalds 15280165ab44SMiklos Szeredi unsigned long mmap_region(struct file *file, unsigned long addr, 1529c22c0d63SMichel Lespinasse unsigned long len, vm_flags_t vm_flags, unsigned long pgoff) 15300165ab44SMiklos Szeredi { 15310165ab44SMiklos Szeredi struct mm_struct *mm = current->mm; 15320165ab44SMiklos Szeredi struct vm_area_struct *vma, *prev; 15330165ab44SMiklos Szeredi int error; 15340165ab44SMiklos Szeredi struct rb_node **rb_link, *rb_parent; 15350165ab44SMiklos Szeredi unsigned long charged = 0; 15360165ab44SMiklos Szeredi 1537e8420a8eSCyril Hrubis /* Check against address space limit. */ 1538e8420a8eSCyril Hrubis if (!may_expand_vm(mm, len >> PAGE_SHIFT)) { 1539e8420a8eSCyril Hrubis unsigned long nr_pages; 1540e8420a8eSCyril Hrubis 1541e8420a8eSCyril Hrubis /* 1542e8420a8eSCyril Hrubis * MAP_FIXED may remove pages of mappings that intersects with 1543e8420a8eSCyril Hrubis * requested mapping. Account for the pages it would unmap. 1544e8420a8eSCyril Hrubis */ 1545e8420a8eSCyril Hrubis if (!(vm_flags & MAP_FIXED)) 1546e8420a8eSCyril Hrubis return -ENOMEM; 1547e8420a8eSCyril Hrubis 1548e8420a8eSCyril Hrubis nr_pages = count_vma_pages_range(mm, addr, addr + len); 1549e8420a8eSCyril Hrubis 1550e8420a8eSCyril Hrubis if (!may_expand_vm(mm, (len >> PAGE_SHIFT) - nr_pages)) 1551e8420a8eSCyril Hrubis return -ENOMEM; 1552e8420a8eSCyril Hrubis } 1553e8420a8eSCyril Hrubis 15541da177e4SLinus Torvalds /* Clear old maps */ 15551da177e4SLinus Torvalds error = -ENOMEM; 15561da177e4SLinus Torvalds munmap_back: 15576597d783SHugh Dickins if (find_vma_links(mm, addr, addr + len, &prev, &rb_link, &rb_parent)) { 15581da177e4SLinus Torvalds if (do_munmap(mm, addr, len)) 15591da177e4SLinus Torvalds return -ENOMEM; 15601da177e4SLinus Torvalds goto munmap_back; 15611da177e4SLinus Torvalds } 15621da177e4SLinus Torvalds 1563fc8744adSLinus Torvalds /* 15641da177e4SLinus Torvalds * Private writable mapping: check memory availability 15651da177e4SLinus Torvalds */ 15665a6fe125SMel Gorman if (accountable_mapping(file, vm_flags)) { 15671da177e4SLinus Torvalds charged = len >> PAGE_SHIFT; 1568191c5424SAl Viro if (security_vm_enough_memory_mm(mm, charged)) 15691da177e4SLinus Torvalds return -ENOMEM; 15701da177e4SLinus Torvalds vm_flags |= VM_ACCOUNT; 15711da177e4SLinus Torvalds } 15721da177e4SLinus Torvalds 15731da177e4SLinus Torvalds /* 1574de33c8dbSLinus Torvalds * Can we just expand an old mapping? 15751da177e4SLinus Torvalds */ 1576de33c8dbSLinus Torvalds vma = vma_merge(mm, prev, addr, addr + len, vm_flags, NULL, file, pgoff, NULL); 1577ba470de4SRik van Riel if (vma) 15781da177e4SLinus Torvalds goto out; 15791da177e4SLinus Torvalds 15801da177e4SLinus Torvalds /* 15811da177e4SLinus Torvalds * Determine the object being mapped and call the appropriate 15821da177e4SLinus Torvalds * specific mapper. the address has already been validated, but 15831da177e4SLinus Torvalds * not unmapped, but the maps are removed from the list. 15841da177e4SLinus Torvalds */ 1585c5e3b83eSPekka Enberg vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL); 15861da177e4SLinus Torvalds if (!vma) { 15871da177e4SLinus Torvalds error = -ENOMEM; 15881da177e4SLinus Torvalds goto unacct_error; 15891da177e4SLinus Torvalds } 15901da177e4SLinus Torvalds 15911da177e4SLinus Torvalds vma->vm_mm = mm; 15921da177e4SLinus Torvalds vma->vm_start = addr; 15931da177e4SLinus Torvalds vma->vm_end = addr + len; 15941da177e4SLinus Torvalds vma->vm_flags = vm_flags; 15953ed75eb8SColy Li vma->vm_page_prot = vm_get_page_prot(vm_flags); 15961da177e4SLinus Torvalds vma->vm_pgoff = pgoff; 15975beb4930SRik van Riel INIT_LIST_HEAD(&vma->anon_vma_chain); 15981da177e4SLinus Torvalds 15991da177e4SLinus Torvalds if (file) { 16001da177e4SLinus Torvalds if (vm_flags & VM_DENYWRITE) { 16011da177e4SLinus Torvalds error = deny_write_access(file); 16021da177e4SLinus Torvalds if (error) 16031da177e4SLinus Torvalds goto free_vma; 16041da177e4SLinus Torvalds } 16054bb5f5d9SDavid Herrmann if (vm_flags & VM_SHARED) { 16064bb5f5d9SDavid Herrmann error = mapping_map_writable(file->f_mapping); 16074bb5f5d9SDavid Herrmann if (error) 16084bb5f5d9SDavid Herrmann goto allow_write_and_free_vma; 16094bb5f5d9SDavid Herrmann } 16104bb5f5d9SDavid Herrmann 16114bb5f5d9SDavid Herrmann /* ->mmap() can change vma->vm_file, but must guarantee that 16124bb5f5d9SDavid Herrmann * vma_link() below can deny write-access if VM_DENYWRITE is set 16134bb5f5d9SDavid Herrmann * and map writably if VM_SHARED is set. This usually means the 16144bb5f5d9SDavid Herrmann * new file must not have been exposed to user-space, yet. 16154bb5f5d9SDavid Herrmann */ 1616cb0942b8SAl Viro vma->vm_file = get_file(file); 16171da177e4SLinus Torvalds error = file->f_op->mmap(file, vma); 16181da177e4SLinus Torvalds if (error) 16191da177e4SLinus Torvalds goto unmap_and_free_vma; 16201da177e4SLinus Torvalds 16211da177e4SLinus Torvalds /* Can addr have changed?? 16221da177e4SLinus Torvalds * 16231da177e4SLinus Torvalds * Answer: Yes, several device drivers can do it in their 16241da177e4SLinus Torvalds * f_op->mmap method. -DaveM 16252897b4d2SJoonsoo Kim * Bug: If addr is changed, prev, rb_link, rb_parent should 16262897b4d2SJoonsoo Kim * be updated for vma_link() 16271da177e4SLinus Torvalds */ 16282897b4d2SJoonsoo Kim WARN_ON_ONCE(addr != vma->vm_start); 16292897b4d2SJoonsoo Kim 16301da177e4SLinus Torvalds addr = vma->vm_start; 16311da177e4SLinus Torvalds vm_flags = vma->vm_flags; 1632f8dbf0a7SHuang Shijie } else if (vm_flags & VM_SHARED) { 1633f8dbf0a7SHuang Shijie error = shmem_zero_setup(vma); 1634f8dbf0a7SHuang Shijie if (error) 1635f8dbf0a7SHuang Shijie goto free_vma; 1636f8dbf0a7SHuang Shijie } 16371da177e4SLinus Torvalds 16384d3d5b41SOleg Nesterov vma_link(mm, vma, prev, rb_link, rb_parent); 16394d3d5b41SOleg Nesterov /* Once vma denies write, undo our temporary denial count */ 16404bb5f5d9SDavid Herrmann if (file) { 16414bb5f5d9SDavid Herrmann if (vm_flags & VM_SHARED) 16424bb5f5d9SDavid Herrmann mapping_unmap_writable(file->f_mapping); 1643e8686772SOleg Nesterov if (vm_flags & VM_DENYWRITE) 1644e8686772SOleg Nesterov allow_write_access(file); 16454bb5f5d9SDavid Herrmann } 1646e8686772SOleg Nesterov file = vma->vm_file; 16471da177e4SLinus Torvalds out: 1648cdd6c482SIngo Molnar perf_event_mmap(vma); 16490a4a9391SPeter Zijlstra 1650ab50b8edSHugh Dickins vm_stat_account(mm, vm_flags, file, len >> PAGE_SHIFT); 16511da177e4SLinus Torvalds if (vm_flags & VM_LOCKED) { 1652bebeb3d6SMichel Lespinasse if (!((vm_flags & VM_SPECIAL) || is_vm_hugetlb_page(vma) || 1653bebeb3d6SMichel Lespinasse vma == get_gate_vma(current->mm))) 165406f9d8c2SKOSAKI Motohiro mm->locked_vm += (len >> PAGE_SHIFT); 1655bebeb3d6SMichel Lespinasse else 1656bebeb3d6SMichel Lespinasse vma->vm_flags &= ~VM_LOCKED; 1657bebeb3d6SMichel Lespinasse } 16582b144498SSrikar Dronamraju 1659c7a3a88cSOleg Nesterov if (file) 1660c7a3a88cSOleg Nesterov uprobe_mmap(vma); 16612b144498SSrikar Dronamraju 1662d9104d1cSCyrill Gorcunov /* 1663d9104d1cSCyrill Gorcunov * New (or expanded) vma always get soft dirty status. 1664d9104d1cSCyrill Gorcunov * Otherwise user-space soft-dirty page tracker won't 1665d9104d1cSCyrill Gorcunov * be able to distinguish situation when vma area unmapped, 1666d9104d1cSCyrill Gorcunov * then new mapped in-place (which must be aimed as 1667d9104d1cSCyrill Gorcunov * a completely new data area). 1668d9104d1cSCyrill Gorcunov */ 1669d9104d1cSCyrill Gorcunov vma->vm_flags |= VM_SOFTDIRTY; 1670d9104d1cSCyrill Gorcunov 167164e45507SPeter Feiner vma_set_page_prot(vma); 167264e45507SPeter Feiner 16731da177e4SLinus Torvalds return addr; 16741da177e4SLinus Torvalds 16751da177e4SLinus Torvalds unmap_and_free_vma: 16761da177e4SLinus Torvalds vma->vm_file = NULL; 16771da177e4SLinus Torvalds fput(file); 16781da177e4SLinus Torvalds 16791da177e4SLinus Torvalds /* Undo any partial mapping done by a device driver. */ 1680e0da382cSHugh Dickins unmap_region(mm, vma, prev, vma->vm_start, vma->vm_end); 1681e0da382cSHugh Dickins charged = 0; 16824bb5f5d9SDavid Herrmann if (vm_flags & VM_SHARED) 16834bb5f5d9SDavid Herrmann mapping_unmap_writable(file->f_mapping); 16844bb5f5d9SDavid Herrmann allow_write_and_free_vma: 16854bb5f5d9SDavid Herrmann if (vm_flags & VM_DENYWRITE) 16864bb5f5d9SDavid Herrmann allow_write_access(file); 16871da177e4SLinus Torvalds free_vma: 16881da177e4SLinus Torvalds kmem_cache_free(vm_area_cachep, vma); 16891da177e4SLinus Torvalds unacct_error: 16901da177e4SLinus Torvalds if (charged) 16911da177e4SLinus Torvalds vm_unacct_memory(charged); 16921da177e4SLinus Torvalds return error; 16931da177e4SLinus Torvalds } 16941da177e4SLinus Torvalds 1695db4fbfb9SMichel Lespinasse unsigned long unmapped_area(struct vm_unmapped_area_info *info) 1696db4fbfb9SMichel Lespinasse { 1697db4fbfb9SMichel Lespinasse /* 1698db4fbfb9SMichel Lespinasse * We implement the search by looking for an rbtree node that 1699db4fbfb9SMichel Lespinasse * immediately follows a suitable gap. That is, 1700db4fbfb9SMichel Lespinasse * - gap_start = vma->vm_prev->vm_end <= info->high_limit - length; 1701db4fbfb9SMichel Lespinasse * - gap_end = vma->vm_start >= info->low_limit + length; 1702db4fbfb9SMichel Lespinasse * - gap_end - gap_start >= length 1703db4fbfb9SMichel Lespinasse */ 1704db4fbfb9SMichel Lespinasse 1705db4fbfb9SMichel Lespinasse struct mm_struct *mm = current->mm; 1706db4fbfb9SMichel Lespinasse struct vm_area_struct *vma; 1707db4fbfb9SMichel Lespinasse unsigned long length, low_limit, high_limit, gap_start, gap_end; 1708db4fbfb9SMichel Lespinasse 1709db4fbfb9SMichel Lespinasse /* Adjust search length to account for worst case alignment overhead */ 1710db4fbfb9SMichel Lespinasse length = info->length + info->align_mask; 1711db4fbfb9SMichel Lespinasse if (length < info->length) 1712db4fbfb9SMichel Lespinasse return -ENOMEM; 1713db4fbfb9SMichel Lespinasse 1714db4fbfb9SMichel Lespinasse /* Adjust search limits by the desired length */ 1715db4fbfb9SMichel Lespinasse if (info->high_limit < length) 1716db4fbfb9SMichel Lespinasse return -ENOMEM; 1717db4fbfb9SMichel Lespinasse high_limit = info->high_limit - length; 1718db4fbfb9SMichel Lespinasse 1719db4fbfb9SMichel Lespinasse if (info->low_limit > high_limit) 1720db4fbfb9SMichel Lespinasse return -ENOMEM; 1721db4fbfb9SMichel Lespinasse low_limit = info->low_limit + length; 1722db4fbfb9SMichel Lespinasse 1723db4fbfb9SMichel Lespinasse /* Check if rbtree root looks promising */ 1724db4fbfb9SMichel Lespinasse if (RB_EMPTY_ROOT(&mm->mm_rb)) 1725db4fbfb9SMichel Lespinasse goto check_highest; 1726db4fbfb9SMichel Lespinasse vma = rb_entry(mm->mm_rb.rb_node, struct vm_area_struct, vm_rb); 1727db4fbfb9SMichel Lespinasse if (vma->rb_subtree_gap < length) 1728db4fbfb9SMichel Lespinasse goto check_highest; 1729db4fbfb9SMichel Lespinasse 1730db4fbfb9SMichel Lespinasse while (true) { 1731db4fbfb9SMichel Lespinasse /* Visit left subtree if it looks promising */ 1732db4fbfb9SMichel Lespinasse gap_end = vma->vm_start; 1733db4fbfb9SMichel Lespinasse if (gap_end >= low_limit && vma->vm_rb.rb_left) { 1734db4fbfb9SMichel Lespinasse struct vm_area_struct *left = 1735db4fbfb9SMichel Lespinasse rb_entry(vma->vm_rb.rb_left, 1736db4fbfb9SMichel Lespinasse struct vm_area_struct, vm_rb); 1737db4fbfb9SMichel Lespinasse if (left->rb_subtree_gap >= length) { 1738db4fbfb9SMichel Lespinasse vma = left; 1739db4fbfb9SMichel Lespinasse continue; 1740db4fbfb9SMichel Lespinasse } 1741db4fbfb9SMichel Lespinasse } 1742db4fbfb9SMichel Lespinasse 1743db4fbfb9SMichel Lespinasse gap_start = vma->vm_prev ? vma->vm_prev->vm_end : 0; 1744db4fbfb9SMichel Lespinasse check_current: 1745db4fbfb9SMichel Lespinasse /* Check if current node has a suitable gap */ 1746db4fbfb9SMichel Lespinasse if (gap_start > high_limit) 1747db4fbfb9SMichel Lespinasse return -ENOMEM; 1748db4fbfb9SMichel Lespinasse if (gap_end >= low_limit && gap_end - gap_start >= length) 1749db4fbfb9SMichel Lespinasse goto found; 1750db4fbfb9SMichel Lespinasse 1751db4fbfb9SMichel Lespinasse /* Visit right subtree if it looks promising */ 1752db4fbfb9SMichel Lespinasse if (vma->vm_rb.rb_right) { 1753db4fbfb9SMichel Lespinasse struct vm_area_struct *right = 1754db4fbfb9SMichel Lespinasse rb_entry(vma->vm_rb.rb_right, 1755db4fbfb9SMichel Lespinasse struct vm_area_struct, vm_rb); 1756db4fbfb9SMichel Lespinasse if (right->rb_subtree_gap >= length) { 1757db4fbfb9SMichel Lespinasse vma = right; 1758db4fbfb9SMichel Lespinasse continue; 1759db4fbfb9SMichel Lespinasse } 1760db4fbfb9SMichel Lespinasse } 1761db4fbfb9SMichel Lespinasse 1762db4fbfb9SMichel Lespinasse /* Go back up the rbtree to find next candidate node */ 1763db4fbfb9SMichel Lespinasse while (true) { 1764db4fbfb9SMichel Lespinasse struct rb_node *prev = &vma->vm_rb; 1765db4fbfb9SMichel Lespinasse if (!rb_parent(prev)) 1766db4fbfb9SMichel Lespinasse goto check_highest; 1767db4fbfb9SMichel Lespinasse vma = rb_entry(rb_parent(prev), 1768db4fbfb9SMichel Lespinasse struct vm_area_struct, vm_rb); 1769db4fbfb9SMichel Lespinasse if (prev == vma->vm_rb.rb_left) { 1770db4fbfb9SMichel Lespinasse gap_start = vma->vm_prev->vm_end; 1771db4fbfb9SMichel Lespinasse gap_end = vma->vm_start; 1772db4fbfb9SMichel Lespinasse goto check_current; 1773db4fbfb9SMichel Lespinasse } 1774db4fbfb9SMichel Lespinasse } 1775db4fbfb9SMichel Lespinasse } 1776db4fbfb9SMichel Lespinasse 1777db4fbfb9SMichel Lespinasse check_highest: 1778db4fbfb9SMichel Lespinasse /* Check highest gap, which does not precede any rbtree node */ 1779db4fbfb9SMichel Lespinasse gap_start = mm->highest_vm_end; 1780db4fbfb9SMichel Lespinasse gap_end = ULONG_MAX; /* Only for VM_BUG_ON below */ 1781db4fbfb9SMichel Lespinasse if (gap_start > high_limit) 1782db4fbfb9SMichel Lespinasse return -ENOMEM; 1783db4fbfb9SMichel Lespinasse 1784db4fbfb9SMichel Lespinasse found: 1785db4fbfb9SMichel Lespinasse /* We found a suitable gap. Clip it with the original low_limit. */ 1786db4fbfb9SMichel Lespinasse if (gap_start < info->low_limit) 1787db4fbfb9SMichel Lespinasse gap_start = info->low_limit; 1788db4fbfb9SMichel Lespinasse 1789db4fbfb9SMichel Lespinasse /* Adjust gap address to the desired alignment */ 1790db4fbfb9SMichel Lespinasse gap_start += (info->align_offset - gap_start) & info->align_mask; 1791db4fbfb9SMichel Lespinasse 1792db4fbfb9SMichel Lespinasse VM_BUG_ON(gap_start + info->length > info->high_limit); 1793db4fbfb9SMichel Lespinasse VM_BUG_ON(gap_start + info->length > gap_end); 1794db4fbfb9SMichel Lespinasse return gap_start; 1795db4fbfb9SMichel Lespinasse } 1796db4fbfb9SMichel Lespinasse 1797db4fbfb9SMichel Lespinasse unsigned long unmapped_area_topdown(struct vm_unmapped_area_info *info) 1798db4fbfb9SMichel Lespinasse { 1799db4fbfb9SMichel Lespinasse struct mm_struct *mm = current->mm; 1800db4fbfb9SMichel Lespinasse struct vm_area_struct *vma; 1801db4fbfb9SMichel Lespinasse unsigned long length, low_limit, high_limit, gap_start, gap_end; 1802db4fbfb9SMichel Lespinasse 1803db4fbfb9SMichel Lespinasse /* Adjust search length to account for worst case alignment overhead */ 1804db4fbfb9SMichel Lespinasse length = info->length + info->align_mask; 1805db4fbfb9SMichel Lespinasse if (length < info->length) 1806db4fbfb9SMichel Lespinasse return -ENOMEM; 1807db4fbfb9SMichel Lespinasse 1808db4fbfb9SMichel Lespinasse /* 1809db4fbfb9SMichel Lespinasse * Adjust search limits by the desired length. 1810db4fbfb9SMichel Lespinasse * See implementation comment at top of unmapped_area(). 1811db4fbfb9SMichel Lespinasse */ 1812db4fbfb9SMichel Lespinasse gap_end = info->high_limit; 1813db4fbfb9SMichel Lespinasse if (gap_end < length) 1814db4fbfb9SMichel Lespinasse return -ENOMEM; 1815db4fbfb9SMichel Lespinasse high_limit = gap_end - length; 1816db4fbfb9SMichel Lespinasse 1817db4fbfb9SMichel Lespinasse if (info->low_limit > high_limit) 1818db4fbfb9SMichel Lespinasse return -ENOMEM; 1819db4fbfb9SMichel Lespinasse low_limit = info->low_limit + length; 1820db4fbfb9SMichel Lespinasse 1821db4fbfb9SMichel Lespinasse /* Check highest gap, which does not precede any rbtree node */ 1822db4fbfb9SMichel Lespinasse gap_start = mm->highest_vm_end; 1823db4fbfb9SMichel Lespinasse if (gap_start <= high_limit) 1824db4fbfb9SMichel Lespinasse goto found_highest; 1825db4fbfb9SMichel Lespinasse 1826db4fbfb9SMichel Lespinasse /* Check if rbtree root looks promising */ 1827db4fbfb9SMichel Lespinasse if (RB_EMPTY_ROOT(&mm->mm_rb)) 1828db4fbfb9SMichel Lespinasse return -ENOMEM; 1829db4fbfb9SMichel Lespinasse vma = rb_entry(mm->mm_rb.rb_node, struct vm_area_struct, vm_rb); 1830db4fbfb9SMichel Lespinasse if (vma->rb_subtree_gap < length) 1831db4fbfb9SMichel Lespinasse return -ENOMEM; 1832db4fbfb9SMichel Lespinasse 1833db4fbfb9SMichel Lespinasse while (true) { 1834db4fbfb9SMichel Lespinasse /* Visit right subtree if it looks promising */ 1835db4fbfb9SMichel Lespinasse gap_start = vma->vm_prev ? vma->vm_prev->vm_end : 0; 1836db4fbfb9SMichel Lespinasse if (gap_start <= high_limit && vma->vm_rb.rb_right) { 1837db4fbfb9SMichel Lespinasse struct vm_area_struct *right = 1838db4fbfb9SMichel Lespinasse rb_entry(vma->vm_rb.rb_right, 1839db4fbfb9SMichel Lespinasse struct vm_area_struct, vm_rb); 1840db4fbfb9SMichel Lespinasse if (right->rb_subtree_gap >= length) { 1841db4fbfb9SMichel Lespinasse vma = right; 1842db4fbfb9SMichel Lespinasse continue; 1843db4fbfb9SMichel Lespinasse } 1844db4fbfb9SMichel Lespinasse } 1845db4fbfb9SMichel Lespinasse 1846db4fbfb9SMichel Lespinasse check_current: 1847db4fbfb9SMichel Lespinasse /* Check if current node has a suitable gap */ 1848db4fbfb9SMichel Lespinasse gap_end = vma->vm_start; 1849db4fbfb9SMichel Lespinasse if (gap_end < low_limit) 1850db4fbfb9SMichel Lespinasse return -ENOMEM; 1851db4fbfb9SMichel Lespinasse if (gap_start <= high_limit && gap_end - gap_start >= length) 1852db4fbfb9SMichel Lespinasse goto found; 1853db4fbfb9SMichel Lespinasse 1854db4fbfb9SMichel Lespinasse /* Visit left subtree if it looks promising */ 1855db4fbfb9SMichel Lespinasse if (vma->vm_rb.rb_left) { 1856db4fbfb9SMichel Lespinasse struct vm_area_struct *left = 1857db4fbfb9SMichel Lespinasse rb_entry(vma->vm_rb.rb_left, 1858db4fbfb9SMichel Lespinasse struct vm_area_struct, vm_rb); 1859db4fbfb9SMichel Lespinasse if (left->rb_subtree_gap >= length) { 1860db4fbfb9SMichel Lespinasse vma = left; 1861db4fbfb9SMichel Lespinasse continue; 1862db4fbfb9SMichel Lespinasse } 1863db4fbfb9SMichel Lespinasse } 1864db4fbfb9SMichel Lespinasse 1865db4fbfb9SMichel Lespinasse /* Go back up the rbtree to find next candidate node */ 1866db4fbfb9SMichel Lespinasse while (true) { 1867db4fbfb9SMichel Lespinasse struct rb_node *prev = &vma->vm_rb; 1868db4fbfb9SMichel Lespinasse if (!rb_parent(prev)) 1869db4fbfb9SMichel Lespinasse return -ENOMEM; 1870db4fbfb9SMichel Lespinasse vma = rb_entry(rb_parent(prev), 1871db4fbfb9SMichel Lespinasse struct vm_area_struct, vm_rb); 1872db4fbfb9SMichel Lespinasse if (prev == vma->vm_rb.rb_right) { 1873db4fbfb9SMichel Lespinasse gap_start = vma->vm_prev ? 1874db4fbfb9SMichel Lespinasse vma->vm_prev->vm_end : 0; 1875db4fbfb9SMichel Lespinasse goto check_current; 1876db4fbfb9SMichel Lespinasse } 1877db4fbfb9SMichel Lespinasse } 1878db4fbfb9SMichel Lespinasse } 1879db4fbfb9SMichel Lespinasse 1880db4fbfb9SMichel Lespinasse found: 1881db4fbfb9SMichel Lespinasse /* We found a suitable gap. Clip it with the original high_limit. */ 1882db4fbfb9SMichel Lespinasse if (gap_end > info->high_limit) 1883db4fbfb9SMichel Lespinasse gap_end = info->high_limit; 1884db4fbfb9SMichel Lespinasse 1885db4fbfb9SMichel Lespinasse found_highest: 1886db4fbfb9SMichel Lespinasse /* Compute highest gap address at the desired alignment */ 1887db4fbfb9SMichel Lespinasse gap_end -= info->length; 1888db4fbfb9SMichel Lespinasse gap_end -= (gap_end - info->align_offset) & info->align_mask; 1889db4fbfb9SMichel Lespinasse 1890db4fbfb9SMichel Lespinasse VM_BUG_ON(gap_end < info->low_limit); 1891db4fbfb9SMichel Lespinasse VM_BUG_ON(gap_end < gap_start); 1892db4fbfb9SMichel Lespinasse return gap_end; 1893db4fbfb9SMichel Lespinasse } 1894db4fbfb9SMichel Lespinasse 18951da177e4SLinus Torvalds /* Get an address range which is currently unmapped. 18961da177e4SLinus Torvalds * For shmat() with addr=0. 18971da177e4SLinus Torvalds * 18981da177e4SLinus Torvalds * Ugly calling convention alert: 18991da177e4SLinus Torvalds * Return value with the low bits set means error value, 19001da177e4SLinus Torvalds * ie 19011da177e4SLinus Torvalds * if (ret & ~PAGE_MASK) 19021da177e4SLinus Torvalds * error = ret; 19031da177e4SLinus Torvalds * 19041da177e4SLinus Torvalds * This function "knows" that -ENOMEM has the bits set. 19051da177e4SLinus Torvalds */ 19061da177e4SLinus Torvalds #ifndef HAVE_ARCH_UNMAPPED_AREA 19071da177e4SLinus Torvalds unsigned long 19081da177e4SLinus Torvalds arch_get_unmapped_area(struct file *filp, unsigned long addr, 19091da177e4SLinus Torvalds unsigned long len, unsigned long pgoff, unsigned long flags) 19101da177e4SLinus Torvalds { 19111da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 19121da177e4SLinus Torvalds struct vm_area_struct *vma; 1913db4fbfb9SMichel Lespinasse struct vm_unmapped_area_info info; 19141da177e4SLinus Torvalds 19152afc745fSAkira Takeuchi if (len > TASK_SIZE - mmap_min_addr) 19161da177e4SLinus Torvalds return -ENOMEM; 19171da177e4SLinus Torvalds 191806abdfb4SBenjamin Herrenschmidt if (flags & MAP_FIXED) 191906abdfb4SBenjamin Herrenschmidt return addr; 192006abdfb4SBenjamin Herrenschmidt 19211da177e4SLinus Torvalds if (addr) { 19221da177e4SLinus Torvalds addr = PAGE_ALIGN(addr); 19231da177e4SLinus Torvalds vma = find_vma(mm, addr); 19242afc745fSAkira Takeuchi if (TASK_SIZE - len >= addr && addr >= mmap_min_addr && 19251da177e4SLinus Torvalds (!vma || addr + len <= vma->vm_start)) 19261da177e4SLinus Torvalds return addr; 19271da177e4SLinus Torvalds } 19281da177e4SLinus Torvalds 1929db4fbfb9SMichel Lespinasse info.flags = 0; 1930db4fbfb9SMichel Lespinasse info.length = len; 19314e99b021SHeiko Carstens info.low_limit = mm->mmap_base; 1932db4fbfb9SMichel Lespinasse info.high_limit = TASK_SIZE; 1933db4fbfb9SMichel Lespinasse info.align_mask = 0; 1934db4fbfb9SMichel Lespinasse return vm_unmapped_area(&info); 19351da177e4SLinus Torvalds } 19361da177e4SLinus Torvalds #endif 19371da177e4SLinus Torvalds 19381da177e4SLinus Torvalds /* 19391da177e4SLinus Torvalds * This mmap-allocator allocates new areas top-down from below the 19401da177e4SLinus Torvalds * stack's low limit (the base): 19411da177e4SLinus Torvalds */ 19421da177e4SLinus Torvalds #ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN 19431da177e4SLinus Torvalds unsigned long 19441da177e4SLinus Torvalds arch_get_unmapped_area_topdown(struct file *filp, const unsigned long addr0, 19451da177e4SLinus Torvalds const unsigned long len, const unsigned long pgoff, 19461da177e4SLinus Torvalds const unsigned long flags) 19471da177e4SLinus Torvalds { 19481da177e4SLinus Torvalds struct vm_area_struct *vma; 19491da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 1950db4fbfb9SMichel Lespinasse unsigned long addr = addr0; 1951db4fbfb9SMichel Lespinasse struct vm_unmapped_area_info info; 19521da177e4SLinus Torvalds 19531da177e4SLinus Torvalds /* requested length too big for entire address space */ 19542afc745fSAkira Takeuchi if (len > TASK_SIZE - mmap_min_addr) 19551da177e4SLinus Torvalds return -ENOMEM; 19561da177e4SLinus Torvalds 195706abdfb4SBenjamin Herrenschmidt if (flags & MAP_FIXED) 195806abdfb4SBenjamin Herrenschmidt return addr; 195906abdfb4SBenjamin Herrenschmidt 19601da177e4SLinus Torvalds /* requesting a specific address */ 19611da177e4SLinus Torvalds if (addr) { 19621da177e4SLinus Torvalds addr = PAGE_ALIGN(addr); 19631da177e4SLinus Torvalds vma = find_vma(mm, addr); 19642afc745fSAkira Takeuchi if (TASK_SIZE - len >= addr && addr >= mmap_min_addr && 19651da177e4SLinus Torvalds (!vma || addr + len <= vma->vm_start)) 19661da177e4SLinus Torvalds return addr; 19671da177e4SLinus Torvalds } 19681da177e4SLinus Torvalds 1969db4fbfb9SMichel Lespinasse info.flags = VM_UNMAPPED_AREA_TOPDOWN; 1970db4fbfb9SMichel Lespinasse info.length = len; 19712afc745fSAkira Takeuchi info.low_limit = max(PAGE_SIZE, mmap_min_addr); 1972db4fbfb9SMichel Lespinasse info.high_limit = mm->mmap_base; 1973db4fbfb9SMichel Lespinasse info.align_mask = 0; 1974db4fbfb9SMichel Lespinasse addr = vm_unmapped_area(&info); 1975b716ad95SXiao Guangrong 19761da177e4SLinus Torvalds /* 19771da177e4SLinus Torvalds * A failed mmap() very likely causes application failure, 19781da177e4SLinus Torvalds * so fall back to the bottom-up function here. This scenario 19791da177e4SLinus Torvalds * can happen with large stack limits and large mmap() 19801da177e4SLinus Torvalds * allocations. 19811da177e4SLinus Torvalds */ 1982db4fbfb9SMichel Lespinasse if (addr & ~PAGE_MASK) { 1983db4fbfb9SMichel Lespinasse VM_BUG_ON(addr != -ENOMEM); 1984db4fbfb9SMichel Lespinasse info.flags = 0; 1985db4fbfb9SMichel Lespinasse info.low_limit = TASK_UNMAPPED_BASE; 1986db4fbfb9SMichel Lespinasse info.high_limit = TASK_SIZE; 1987db4fbfb9SMichel Lespinasse addr = vm_unmapped_area(&info); 1988db4fbfb9SMichel Lespinasse } 19891da177e4SLinus Torvalds 19901da177e4SLinus Torvalds return addr; 19911da177e4SLinus Torvalds } 19921da177e4SLinus Torvalds #endif 19931da177e4SLinus Torvalds 19941da177e4SLinus Torvalds unsigned long 19951da177e4SLinus Torvalds get_unmapped_area(struct file *file, unsigned long addr, unsigned long len, 19961da177e4SLinus Torvalds unsigned long pgoff, unsigned long flags) 19971da177e4SLinus Torvalds { 199806abdfb4SBenjamin Herrenschmidt unsigned long (*get_area)(struct file *, unsigned long, 199906abdfb4SBenjamin Herrenschmidt unsigned long, unsigned long, unsigned long); 200007ab67c8SLinus Torvalds 20019206de95SAl Viro unsigned long error = arch_mmap_check(addr, len, flags); 20029206de95SAl Viro if (error) 20039206de95SAl Viro return error; 20049206de95SAl Viro 20059206de95SAl Viro /* Careful about overflows.. */ 20069206de95SAl Viro if (len > TASK_SIZE) 20079206de95SAl Viro return -ENOMEM; 20089206de95SAl Viro 200907ab67c8SLinus Torvalds get_area = current->mm->get_unmapped_area; 201072c2d531SAl Viro if (file && file->f_op->get_unmapped_area) 201107ab67c8SLinus Torvalds get_area = file->f_op->get_unmapped_area; 201207ab67c8SLinus Torvalds addr = get_area(file, addr, len, pgoff, flags); 201307ab67c8SLinus Torvalds if (IS_ERR_VALUE(addr)) 201407ab67c8SLinus Torvalds return addr; 201507ab67c8SLinus Torvalds 20161da177e4SLinus Torvalds if (addr > TASK_SIZE - len) 20171da177e4SLinus Torvalds return -ENOMEM; 20181da177e4SLinus Torvalds if (addr & ~PAGE_MASK) 20191da177e4SLinus Torvalds return -EINVAL; 202006abdfb4SBenjamin Herrenschmidt 20219ac4ed4bSAl Viro addr = arch_rebalance_pgtables(addr, len); 20229ac4ed4bSAl Viro error = security_mmap_addr(addr); 20239ac4ed4bSAl Viro return error ? error : addr; 20241da177e4SLinus Torvalds } 20251da177e4SLinus Torvalds 20261da177e4SLinus Torvalds EXPORT_SYMBOL(get_unmapped_area); 20271da177e4SLinus Torvalds 20281da177e4SLinus Torvalds /* Look up the first VMA which satisfies addr < vm_end, NULL if none. */ 20291da177e4SLinus Torvalds struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr) 20301da177e4SLinus Torvalds { 2031615d6e87SDavidlohr Bueso struct rb_node *rb_node; 2032615d6e87SDavidlohr Bueso struct vm_area_struct *vma; 20331da177e4SLinus Torvalds 20341da177e4SLinus Torvalds /* Check the cache first. */ 2035615d6e87SDavidlohr Bueso vma = vmacache_find(mm, addr); 2036615d6e87SDavidlohr Bueso if (likely(vma)) 2037615d6e87SDavidlohr Bueso return vma; 20381da177e4SLinus Torvalds 20391da177e4SLinus Torvalds rb_node = mm->mm_rb.rb_node; 20401da177e4SLinus Torvalds vma = NULL; 20411da177e4SLinus Torvalds 20421da177e4SLinus Torvalds while (rb_node) { 2043615d6e87SDavidlohr Bueso struct vm_area_struct *tmp; 20441da177e4SLinus Torvalds 2045615d6e87SDavidlohr Bueso tmp = rb_entry(rb_node, struct vm_area_struct, vm_rb); 20461da177e4SLinus Torvalds 2047615d6e87SDavidlohr Bueso if (tmp->vm_end > addr) { 2048615d6e87SDavidlohr Bueso vma = tmp; 2049615d6e87SDavidlohr Bueso if (tmp->vm_start <= addr) 20501da177e4SLinus Torvalds break; 20511da177e4SLinus Torvalds rb_node = rb_node->rb_left; 20521da177e4SLinus Torvalds } else 20531da177e4SLinus Torvalds rb_node = rb_node->rb_right; 20541da177e4SLinus Torvalds } 2055615d6e87SDavidlohr Bueso 20561da177e4SLinus Torvalds if (vma) 2057615d6e87SDavidlohr Bueso vmacache_update(addr, vma); 20581da177e4SLinus Torvalds return vma; 20591da177e4SLinus Torvalds } 20601da177e4SLinus Torvalds 20611da177e4SLinus Torvalds EXPORT_SYMBOL(find_vma); 20621da177e4SLinus Torvalds 20636bd4837dSKOSAKI Motohiro /* 20646bd4837dSKOSAKI Motohiro * Same as find_vma, but also return a pointer to the previous VMA in *pprev. 20656bd4837dSKOSAKI Motohiro */ 20661da177e4SLinus Torvalds struct vm_area_struct * 20671da177e4SLinus Torvalds find_vma_prev(struct mm_struct *mm, unsigned long addr, 20681da177e4SLinus Torvalds struct vm_area_struct **pprev) 20691da177e4SLinus Torvalds { 20706bd4837dSKOSAKI Motohiro struct vm_area_struct *vma; 20711da177e4SLinus Torvalds 20726bd4837dSKOSAKI Motohiro vma = find_vma(mm, addr); 207383cd904dSMikulas Patocka if (vma) { 207483cd904dSMikulas Patocka *pprev = vma->vm_prev; 207583cd904dSMikulas Patocka } else { 207683cd904dSMikulas Patocka struct rb_node *rb_node = mm->mm_rb.rb_node; 207783cd904dSMikulas Patocka *pprev = NULL; 207883cd904dSMikulas Patocka while (rb_node) { 207983cd904dSMikulas Patocka *pprev = rb_entry(rb_node, struct vm_area_struct, vm_rb); 208083cd904dSMikulas Patocka rb_node = rb_node->rb_right; 208183cd904dSMikulas Patocka } 208283cd904dSMikulas Patocka } 20836bd4837dSKOSAKI Motohiro return vma; 20841da177e4SLinus Torvalds } 20851da177e4SLinus Torvalds 20861da177e4SLinus Torvalds /* 20871da177e4SLinus Torvalds * Verify that the stack growth is acceptable and 20881da177e4SLinus Torvalds * update accounting. This is shared with both the 20891da177e4SLinus Torvalds * grow-up and grow-down cases. 20901da177e4SLinus Torvalds */ 20911da177e4SLinus Torvalds static int acct_stack_growth(struct vm_area_struct *vma, unsigned long size, unsigned long grow) 20921da177e4SLinus Torvalds { 20931da177e4SLinus Torvalds struct mm_struct *mm = vma->vm_mm; 20941da177e4SLinus Torvalds struct rlimit *rlim = current->signal->rlim; 2095690eac53SLinus Torvalds unsigned long new_start, actual_size; 20961da177e4SLinus Torvalds 20971da177e4SLinus Torvalds /* address space limit tests */ 2098119f657cSakpm@osdl.org if (!may_expand_vm(mm, grow)) 20991da177e4SLinus Torvalds return -ENOMEM; 21001da177e4SLinus Torvalds 21011da177e4SLinus Torvalds /* Stack limit test */ 2102690eac53SLinus Torvalds actual_size = size; 2103690eac53SLinus Torvalds if (size && (vma->vm_flags & (VM_GROWSUP | VM_GROWSDOWN))) 2104690eac53SLinus Torvalds actual_size -= PAGE_SIZE; 2105690eac53SLinus Torvalds if (actual_size > ACCESS_ONCE(rlim[RLIMIT_STACK].rlim_cur)) 21061da177e4SLinus Torvalds return -ENOMEM; 21071da177e4SLinus Torvalds 21081da177e4SLinus Torvalds /* mlock limit tests */ 21091da177e4SLinus Torvalds if (vma->vm_flags & VM_LOCKED) { 21101da177e4SLinus Torvalds unsigned long locked; 21111da177e4SLinus Torvalds unsigned long limit; 21121da177e4SLinus Torvalds locked = mm->locked_vm + grow; 211359e99e5bSJiri Slaby limit = ACCESS_ONCE(rlim[RLIMIT_MEMLOCK].rlim_cur); 211459e99e5bSJiri Slaby limit >>= PAGE_SHIFT; 21151da177e4SLinus Torvalds if (locked > limit && !capable(CAP_IPC_LOCK)) 21161da177e4SLinus Torvalds return -ENOMEM; 21171da177e4SLinus Torvalds } 21181da177e4SLinus Torvalds 21190d59a01bSAdam Litke /* Check to ensure the stack will not grow into a hugetlb-only region */ 21200d59a01bSAdam Litke new_start = (vma->vm_flags & VM_GROWSUP) ? vma->vm_start : 21210d59a01bSAdam Litke vma->vm_end - size; 21220d59a01bSAdam Litke if (is_hugepage_only_range(vma->vm_mm, new_start, size)) 21230d59a01bSAdam Litke return -EFAULT; 21240d59a01bSAdam Litke 21251da177e4SLinus Torvalds /* 21261da177e4SLinus Torvalds * Overcommit.. This must be the final test, as it will 21271da177e4SLinus Torvalds * update security statistics. 21281da177e4SLinus Torvalds */ 212905fa199dSHugh Dickins if (security_vm_enough_memory_mm(mm, grow)) 21301da177e4SLinus Torvalds return -ENOMEM; 21311da177e4SLinus Torvalds 21321da177e4SLinus Torvalds /* Ok, everything looks good - let it rip */ 21331da177e4SLinus Torvalds if (vma->vm_flags & VM_LOCKED) 21341da177e4SLinus Torvalds mm->locked_vm += grow; 2135ab50b8edSHugh Dickins vm_stat_account(mm, vma->vm_flags, vma->vm_file, grow); 21361da177e4SLinus Torvalds return 0; 21371da177e4SLinus Torvalds } 21381da177e4SLinus Torvalds 213946dea3d0SHugh Dickins #if defined(CONFIG_STACK_GROWSUP) || defined(CONFIG_IA64) 21401da177e4SLinus Torvalds /* 214146dea3d0SHugh Dickins * PA-RISC uses this for its stack; IA64 for its Register Backing Store. 214246dea3d0SHugh Dickins * vma is the last one with address > vma->vm_end. Have to extend vma. 21431da177e4SLinus Torvalds */ 214446dea3d0SHugh Dickins int expand_upwards(struct vm_area_struct *vma, unsigned long address) 21451da177e4SLinus Torvalds { 21461da177e4SLinus Torvalds int error; 21471da177e4SLinus Torvalds 21481da177e4SLinus Torvalds if (!(vma->vm_flags & VM_GROWSUP)) 21491da177e4SLinus Torvalds return -EFAULT; 21501da177e4SLinus Torvalds 21511da177e4SLinus Torvalds /* 21521da177e4SLinus Torvalds * We must make sure the anon_vma is allocated 21531da177e4SLinus Torvalds * so that the anon_vma locking is not a noop. 21541da177e4SLinus Torvalds */ 21551da177e4SLinus Torvalds if (unlikely(anon_vma_prepare(vma))) 21561da177e4SLinus Torvalds return -ENOMEM; 2157bb4a340eSRik van Riel vma_lock_anon_vma(vma); 21581da177e4SLinus Torvalds 21591da177e4SLinus Torvalds /* 21601da177e4SLinus Torvalds * vma->vm_start/vm_end cannot change under us because the caller 21611da177e4SLinus Torvalds * is required to hold the mmap_sem in read mode. We need the 21621da177e4SLinus Torvalds * anon_vma lock to serialize against concurrent expand_stacks. 216306b32f3aSHelge Deller * Also guard against wrapping around to address 0. 21641da177e4SLinus Torvalds */ 216506b32f3aSHelge Deller if (address < PAGE_ALIGN(address+4)) 216606b32f3aSHelge Deller address = PAGE_ALIGN(address+4); 216706b32f3aSHelge Deller else { 2168bb4a340eSRik van Riel vma_unlock_anon_vma(vma); 216906b32f3aSHelge Deller return -ENOMEM; 217006b32f3aSHelge Deller } 21711da177e4SLinus Torvalds error = 0; 21721da177e4SLinus Torvalds 21731da177e4SLinus Torvalds /* Somebody else might have raced and expanded it already */ 21741da177e4SLinus Torvalds if (address > vma->vm_end) { 21751da177e4SLinus Torvalds unsigned long size, grow; 21761da177e4SLinus Torvalds 21771da177e4SLinus Torvalds size = address - vma->vm_start; 21781da177e4SLinus Torvalds grow = (address - vma->vm_end) >> PAGE_SHIFT; 21791da177e4SLinus Torvalds 218042c36f63SHugh Dickins error = -ENOMEM; 218142c36f63SHugh Dickins if (vma->vm_pgoff + (size >> PAGE_SHIFT) >= vma->vm_pgoff) { 21821da177e4SLinus Torvalds error = acct_stack_growth(vma, size, grow); 21833af9e859SEric B Munson if (!error) { 21844128997bSMichel Lespinasse /* 21854128997bSMichel Lespinasse * vma_gap_update() doesn't support concurrent 21864128997bSMichel Lespinasse * updates, but we only hold a shared mmap_sem 21874128997bSMichel Lespinasse * lock here, so we need to protect against 21884128997bSMichel Lespinasse * concurrent vma expansions. 21894128997bSMichel Lespinasse * vma_lock_anon_vma() doesn't help here, as 21904128997bSMichel Lespinasse * we don't guarantee that all growable vmas 21914128997bSMichel Lespinasse * in a mm share the same root anon vma. 21924128997bSMichel Lespinasse * So, we reuse mm->page_table_lock to guard 21934128997bSMichel Lespinasse * against concurrent vma expansions. 21944128997bSMichel Lespinasse */ 21954128997bSMichel Lespinasse spin_lock(&vma->vm_mm->page_table_lock); 2196bf181b9fSMichel Lespinasse anon_vma_interval_tree_pre_update_vma(vma); 21971da177e4SLinus Torvalds vma->vm_end = address; 2198bf181b9fSMichel Lespinasse anon_vma_interval_tree_post_update_vma(vma); 2199d3737187SMichel Lespinasse if (vma->vm_next) 2200d3737187SMichel Lespinasse vma_gap_update(vma->vm_next); 2201d3737187SMichel Lespinasse else 2202d3737187SMichel Lespinasse vma->vm_mm->highest_vm_end = address; 22034128997bSMichel Lespinasse spin_unlock(&vma->vm_mm->page_table_lock); 22044128997bSMichel Lespinasse 22053af9e859SEric B Munson perf_event_mmap(vma); 22063af9e859SEric B Munson } 22071da177e4SLinus Torvalds } 220842c36f63SHugh Dickins } 2209bb4a340eSRik van Riel vma_unlock_anon_vma(vma); 22106d50e60cSDavid Rientjes khugepaged_enter_vma_merge(vma, vma->vm_flags); 2211ed8ea815SMichel Lespinasse validate_mm(vma->vm_mm); 22121da177e4SLinus Torvalds return error; 22131da177e4SLinus Torvalds } 221446dea3d0SHugh Dickins #endif /* CONFIG_STACK_GROWSUP || CONFIG_IA64 */ 221546dea3d0SHugh Dickins 22161da177e4SLinus Torvalds /* 22171da177e4SLinus Torvalds * vma is the first one with address < vma->vm_start. Have to extend vma. 22181da177e4SLinus Torvalds */ 2219d05f3169SMichal Hocko int expand_downwards(struct vm_area_struct *vma, 2220b6a2fea3SOllie Wild unsigned long address) 22211da177e4SLinus Torvalds { 22221da177e4SLinus Torvalds int error; 22231da177e4SLinus Torvalds 22241da177e4SLinus Torvalds /* 22251da177e4SLinus Torvalds * We must make sure the anon_vma is allocated 22261da177e4SLinus Torvalds * so that the anon_vma locking is not a noop. 22271da177e4SLinus Torvalds */ 22281da177e4SLinus Torvalds if (unlikely(anon_vma_prepare(vma))) 22291da177e4SLinus Torvalds return -ENOMEM; 22308869477aSEric Paris 22318869477aSEric Paris address &= PAGE_MASK; 2232e5467859SAl Viro error = security_mmap_addr(address); 22338869477aSEric Paris if (error) 22348869477aSEric Paris return error; 22358869477aSEric Paris 2236bb4a340eSRik van Riel vma_lock_anon_vma(vma); 22371da177e4SLinus Torvalds 22381da177e4SLinus Torvalds /* 22391da177e4SLinus Torvalds * vma->vm_start/vm_end cannot change under us because the caller 22401da177e4SLinus Torvalds * is required to hold the mmap_sem in read mode. We need the 22411da177e4SLinus Torvalds * anon_vma lock to serialize against concurrent expand_stacks. 22421da177e4SLinus Torvalds */ 22431da177e4SLinus Torvalds 22441da177e4SLinus Torvalds /* Somebody else might have raced and expanded it already */ 22451da177e4SLinus Torvalds if (address < vma->vm_start) { 22461da177e4SLinus Torvalds unsigned long size, grow; 22471da177e4SLinus Torvalds 22481da177e4SLinus Torvalds size = vma->vm_end - address; 22491da177e4SLinus Torvalds grow = (vma->vm_start - address) >> PAGE_SHIFT; 22501da177e4SLinus Torvalds 2251a626ca6aSLinus Torvalds error = -ENOMEM; 2252a626ca6aSLinus Torvalds if (grow <= vma->vm_pgoff) { 22531da177e4SLinus Torvalds error = acct_stack_growth(vma, size, grow); 22541da177e4SLinus Torvalds if (!error) { 22554128997bSMichel Lespinasse /* 22564128997bSMichel Lespinasse * vma_gap_update() doesn't support concurrent 22574128997bSMichel Lespinasse * updates, but we only hold a shared mmap_sem 22584128997bSMichel Lespinasse * lock here, so we need to protect against 22594128997bSMichel Lespinasse * concurrent vma expansions. 22604128997bSMichel Lespinasse * vma_lock_anon_vma() doesn't help here, as 22614128997bSMichel Lespinasse * we don't guarantee that all growable vmas 22624128997bSMichel Lespinasse * in a mm share the same root anon vma. 22634128997bSMichel Lespinasse * So, we reuse mm->page_table_lock to guard 22644128997bSMichel Lespinasse * against concurrent vma expansions. 22654128997bSMichel Lespinasse */ 22664128997bSMichel Lespinasse spin_lock(&vma->vm_mm->page_table_lock); 2267bf181b9fSMichel Lespinasse anon_vma_interval_tree_pre_update_vma(vma); 22681da177e4SLinus Torvalds vma->vm_start = address; 22691da177e4SLinus Torvalds vma->vm_pgoff -= grow; 2270bf181b9fSMichel Lespinasse anon_vma_interval_tree_post_update_vma(vma); 2271d3737187SMichel Lespinasse vma_gap_update(vma); 22724128997bSMichel Lespinasse spin_unlock(&vma->vm_mm->page_table_lock); 22734128997bSMichel Lespinasse 22743af9e859SEric B Munson perf_event_mmap(vma); 22751da177e4SLinus Torvalds } 22761da177e4SLinus Torvalds } 2277a626ca6aSLinus Torvalds } 2278bb4a340eSRik van Riel vma_unlock_anon_vma(vma); 22796d50e60cSDavid Rientjes khugepaged_enter_vma_merge(vma, vma->vm_flags); 2280ed8ea815SMichel Lespinasse validate_mm(vma->vm_mm); 22811da177e4SLinus Torvalds return error; 22821da177e4SLinus Torvalds } 22831da177e4SLinus Torvalds 228409884964SLinus Torvalds /* 228509884964SLinus Torvalds * Note how expand_stack() refuses to expand the stack all the way to 228609884964SLinus Torvalds * abut the next virtual mapping, *unless* that mapping itself is also 228709884964SLinus Torvalds * a stack mapping. We want to leave room for a guard page, after all 228809884964SLinus Torvalds * (the guard page itself is not added here, that is done by the 228909884964SLinus Torvalds * actual page faulting logic) 229009884964SLinus Torvalds * 229109884964SLinus Torvalds * This matches the behavior of the guard page logic (see mm/memory.c: 229209884964SLinus Torvalds * check_stack_guard_page()), which only allows the guard page to be 229309884964SLinus Torvalds * removed under these circumstances. 229409884964SLinus Torvalds */ 2295b6a2fea3SOllie Wild #ifdef CONFIG_STACK_GROWSUP 2296b6a2fea3SOllie Wild int expand_stack(struct vm_area_struct *vma, unsigned long address) 2297b6a2fea3SOllie Wild { 229809884964SLinus Torvalds struct vm_area_struct *next; 229909884964SLinus Torvalds 230009884964SLinus Torvalds address &= PAGE_MASK; 230109884964SLinus Torvalds next = vma->vm_next; 230209884964SLinus Torvalds if (next && next->vm_start == address + PAGE_SIZE) { 230309884964SLinus Torvalds if (!(next->vm_flags & VM_GROWSUP)) 230409884964SLinus Torvalds return -ENOMEM; 230509884964SLinus Torvalds } 2306b6a2fea3SOllie Wild return expand_upwards(vma, address); 2307b6a2fea3SOllie Wild } 2308b6a2fea3SOllie Wild 2309b6a2fea3SOllie Wild struct vm_area_struct * 2310b6a2fea3SOllie Wild find_extend_vma(struct mm_struct *mm, unsigned long addr) 2311b6a2fea3SOllie Wild { 2312b6a2fea3SOllie Wild struct vm_area_struct *vma, *prev; 2313b6a2fea3SOllie Wild 2314b6a2fea3SOllie Wild addr &= PAGE_MASK; 2315b6a2fea3SOllie Wild vma = find_vma_prev(mm, addr, &prev); 2316b6a2fea3SOllie Wild if (vma && (vma->vm_start <= addr)) 2317b6a2fea3SOllie Wild return vma; 23181c127185SDenys Vlasenko if (!prev || expand_stack(prev, addr)) 2319b6a2fea3SOllie Wild return NULL; 2320cea10a19SMichel Lespinasse if (prev->vm_flags & VM_LOCKED) 2321cea10a19SMichel Lespinasse __mlock_vma_pages_range(prev, addr, prev->vm_end, NULL); 2322b6a2fea3SOllie Wild return prev; 2323b6a2fea3SOllie Wild } 2324b6a2fea3SOllie Wild #else 2325b6a2fea3SOllie Wild int expand_stack(struct vm_area_struct *vma, unsigned long address) 2326b6a2fea3SOllie Wild { 232709884964SLinus Torvalds struct vm_area_struct *prev; 232809884964SLinus Torvalds 232909884964SLinus Torvalds address &= PAGE_MASK; 233009884964SLinus Torvalds prev = vma->vm_prev; 233109884964SLinus Torvalds if (prev && prev->vm_end == address) { 233209884964SLinus Torvalds if (!(prev->vm_flags & VM_GROWSDOWN)) 233309884964SLinus Torvalds return -ENOMEM; 233409884964SLinus Torvalds } 2335b6a2fea3SOllie Wild return expand_downwards(vma, address); 2336b6a2fea3SOllie Wild } 2337b6a2fea3SOllie Wild 23381da177e4SLinus Torvalds struct vm_area_struct * 23391da177e4SLinus Torvalds find_extend_vma(struct mm_struct *mm, unsigned long addr) 23401da177e4SLinus Torvalds { 23411da177e4SLinus Torvalds struct vm_area_struct *vma; 23421da177e4SLinus Torvalds unsigned long start; 23431da177e4SLinus Torvalds 23441da177e4SLinus Torvalds addr &= PAGE_MASK; 23451da177e4SLinus Torvalds vma = find_vma(mm, addr); 23461da177e4SLinus Torvalds if (!vma) 23471da177e4SLinus Torvalds return NULL; 23481da177e4SLinus Torvalds if (vma->vm_start <= addr) 23491da177e4SLinus Torvalds return vma; 23501da177e4SLinus Torvalds if (!(vma->vm_flags & VM_GROWSDOWN)) 23511da177e4SLinus Torvalds return NULL; 23521da177e4SLinus Torvalds start = vma->vm_start; 23531da177e4SLinus Torvalds if (expand_stack(vma, addr)) 23541da177e4SLinus Torvalds return NULL; 2355cea10a19SMichel Lespinasse if (vma->vm_flags & VM_LOCKED) 2356cea10a19SMichel Lespinasse __mlock_vma_pages_range(vma, addr, start, NULL); 23571da177e4SLinus Torvalds return vma; 23581da177e4SLinus Torvalds } 23591da177e4SLinus Torvalds #endif 23601da177e4SLinus Torvalds 2361e1d6d01aSJesse Barnes EXPORT_SYMBOL_GPL(find_extend_vma); 2362e1d6d01aSJesse Barnes 23632c0b3814SHugh Dickins /* 23642c0b3814SHugh Dickins * Ok - we have the memory areas we should free on the vma list, 23652c0b3814SHugh Dickins * so release them, and do the vma updates. 23661da177e4SLinus Torvalds * 23672c0b3814SHugh Dickins * Called with the mm semaphore held. 23681da177e4SLinus Torvalds */ 23692c0b3814SHugh Dickins static void remove_vma_list(struct mm_struct *mm, struct vm_area_struct *vma) 23701da177e4SLinus Torvalds { 23714f74d2c8SLinus Torvalds unsigned long nr_accounted = 0; 23724f74d2c8SLinus Torvalds 2373365e9c87SHugh Dickins /* Update high watermark before we lower total_vm */ 2374365e9c87SHugh Dickins update_hiwater_vm(mm); 23752c0b3814SHugh Dickins do { 2376ab50b8edSHugh Dickins long nrpages = vma_pages(vma); 23771da177e4SLinus Torvalds 23784f74d2c8SLinus Torvalds if (vma->vm_flags & VM_ACCOUNT) 23794f74d2c8SLinus Torvalds nr_accounted += nrpages; 2380ab50b8edSHugh Dickins vm_stat_account(mm, vma->vm_flags, vma->vm_file, -nrpages); 2381a8fb5618SHugh Dickins vma = remove_vma(vma); 2382146425a3SHugh Dickins } while (vma); 23834f74d2c8SLinus Torvalds vm_unacct_memory(nr_accounted); 23841da177e4SLinus Torvalds validate_mm(mm); 23851da177e4SLinus Torvalds } 23861da177e4SLinus Torvalds 23871da177e4SLinus Torvalds /* 23881da177e4SLinus Torvalds * Get rid of page table information in the indicated region. 23891da177e4SLinus Torvalds * 2390f10df686SPaolo 'Blaisorblade' Giarrusso * Called with the mm semaphore held. 23911da177e4SLinus Torvalds */ 23921da177e4SLinus Torvalds static void unmap_region(struct mm_struct *mm, 2393e0da382cSHugh Dickins struct vm_area_struct *vma, struct vm_area_struct *prev, 2394e0da382cSHugh Dickins unsigned long start, unsigned long end) 23951da177e4SLinus Torvalds { 2396e0da382cSHugh Dickins struct vm_area_struct *next = prev ? prev->vm_next : mm->mmap; 2397d16dfc55SPeter Zijlstra struct mmu_gather tlb; 23981da177e4SLinus Torvalds 23991da177e4SLinus Torvalds lru_add_drain(); 24002b047252SLinus Torvalds tlb_gather_mmu(&tlb, mm, start, end); 2401365e9c87SHugh Dickins update_hiwater_rss(mm); 24024f74d2c8SLinus Torvalds unmap_vmas(&tlb, vma, start, end); 2403d16dfc55SPeter Zijlstra free_pgtables(&tlb, vma, prev ? prev->vm_end : FIRST_USER_ADDRESS, 24046ee8630eSHugh Dickins next ? next->vm_start : USER_PGTABLES_CEILING); 2405d16dfc55SPeter Zijlstra tlb_finish_mmu(&tlb, start, end); 24061da177e4SLinus Torvalds } 24071da177e4SLinus Torvalds 24081da177e4SLinus Torvalds /* 24091da177e4SLinus Torvalds * Create a list of vma's touched by the unmap, removing them from the mm's 24101da177e4SLinus Torvalds * vma list as we go.. 24111da177e4SLinus Torvalds */ 24121da177e4SLinus Torvalds static void 24131da177e4SLinus Torvalds detach_vmas_to_be_unmapped(struct mm_struct *mm, struct vm_area_struct *vma, 24141da177e4SLinus Torvalds struct vm_area_struct *prev, unsigned long end) 24151da177e4SLinus Torvalds { 24161da177e4SLinus Torvalds struct vm_area_struct **insertion_point; 24171da177e4SLinus Torvalds struct vm_area_struct *tail_vma = NULL; 24181da177e4SLinus Torvalds 24191da177e4SLinus Torvalds insertion_point = (prev ? &prev->vm_next : &mm->mmap); 2420297c5eeeSLinus Torvalds vma->vm_prev = NULL; 24211da177e4SLinus Torvalds do { 2422d3737187SMichel Lespinasse vma_rb_erase(vma, &mm->mm_rb); 24231da177e4SLinus Torvalds mm->map_count--; 24241da177e4SLinus Torvalds tail_vma = vma; 24251da177e4SLinus Torvalds vma = vma->vm_next; 24261da177e4SLinus Torvalds } while (vma && vma->vm_start < end); 24271da177e4SLinus Torvalds *insertion_point = vma; 2428d3737187SMichel Lespinasse if (vma) { 2429297c5eeeSLinus Torvalds vma->vm_prev = prev; 2430d3737187SMichel Lespinasse vma_gap_update(vma); 2431d3737187SMichel Lespinasse } else 2432d3737187SMichel Lespinasse mm->highest_vm_end = prev ? prev->vm_end : 0; 24331da177e4SLinus Torvalds tail_vma->vm_next = NULL; 2434615d6e87SDavidlohr Bueso 2435615d6e87SDavidlohr Bueso /* Kill the cache */ 2436615d6e87SDavidlohr Bueso vmacache_invalidate(mm); 24371da177e4SLinus Torvalds } 24381da177e4SLinus Torvalds 24391da177e4SLinus Torvalds /* 2440659ace58SKOSAKI Motohiro * __split_vma() bypasses sysctl_max_map_count checking. We use this on the 2441659ace58SKOSAKI Motohiro * munmap path where it doesn't make sense to fail. 24421da177e4SLinus Torvalds */ 2443659ace58SKOSAKI Motohiro static int __split_vma(struct mm_struct *mm, struct vm_area_struct *vma, 24441da177e4SLinus Torvalds unsigned long addr, int new_below) 24451da177e4SLinus Torvalds { 24461da177e4SLinus Torvalds struct vm_area_struct *new; 24475beb4930SRik van Riel int err = -ENOMEM; 24481da177e4SLinus Torvalds 2449a5516438SAndi Kleen if (is_vm_hugetlb_page(vma) && (addr & 2450a5516438SAndi Kleen ~(huge_page_mask(hstate_vma(vma))))) 24511da177e4SLinus Torvalds return -EINVAL; 24521da177e4SLinus Torvalds 2453e94b1766SChristoph Lameter new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL); 24541da177e4SLinus Torvalds if (!new) 24555beb4930SRik van Riel goto out_err; 24561da177e4SLinus Torvalds 24571da177e4SLinus Torvalds /* most fields are the same, copy all, and then fixup */ 24581da177e4SLinus Torvalds *new = *vma; 24591da177e4SLinus Torvalds 24605beb4930SRik van Riel INIT_LIST_HEAD(&new->anon_vma_chain); 24615beb4930SRik van Riel 24621da177e4SLinus Torvalds if (new_below) 24631da177e4SLinus Torvalds new->vm_end = addr; 24641da177e4SLinus Torvalds else { 24651da177e4SLinus Torvalds new->vm_start = addr; 24661da177e4SLinus Torvalds new->vm_pgoff += ((addr - vma->vm_start) >> PAGE_SHIFT); 24671da177e4SLinus Torvalds } 24681da177e4SLinus Torvalds 2469ef0855d3SOleg Nesterov err = vma_dup_policy(vma, new); 2470ef0855d3SOleg Nesterov if (err) 24715beb4930SRik van Riel goto out_free_vma; 24721da177e4SLinus Torvalds 2473c4ea95d7SDaniel Forrest err = anon_vma_clone(new, vma); 2474c4ea95d7SDaniel Forrest if (err) 24755beb4930SRik van Riel goto out_free_mpol; 24765beb4930SRik van Riel 2477e9714acfSKonstantin Khlebnikov if (new->vm_file) 24781da177e4SLinus Torvalds get_file(new->vm_file); 24791da177e4SLinus Torvalds 24801da177e4SLinus Torvalds if (new->vm_ops && new->vm_ops->open) 24811da177e4SLinus Torvalds new->vm_ops->open(new); 24821da177e4SLinus Torvalds 24831da177e4SLinus Torvalds if (new_below) 24845beb4930SRik van Riel err = vma_adjust(vma, addr, vma->vm_end, vma->vm_pgoff + 24851da177e4SLinus Torvalds ((addr - new->vm_start) >> PAGE_SHIFT), new); 24861da177e4SLinus Torvalds else 24875beb4930SRik van Riel err = vma_adjust(vma, vma->vm_start, addr, vma->vm_pgoff, new); 24881da177e4SLinus Torvalds 24895beb4930SRik van Riel /* Success. */ 24905beb4930SRik van Riel if (!err) 24911da177e4SLinus Torvalds return 0; 24925beb4930SRik van Riel 24935beb4930SRik van Riel /* Clean everything up if vma_adjust failed. */ 249458927533SRik van Riel if (new->vm_ops && new->vm_ops->close) 24955beb4930SRik van Riel new->vm_ops->close(new); 2496e9714acfSKonstantin Khlebnikov if (new->vm_file) 24975beb4930SRik van Riel fput(new->vm_file); 24982aeadc30SAndrea Arcangeli unlink_anon_vmas(new); 24995beb4930SRik van Riel out_free_mpol: 2500ef0855d3SOleg Nesterov mpol_put(vma_policy(new)); 25015beb4930SRik van Riel out_free_vma: 25025beb4930SRik van Riel kmem_cache_free(vm_area_cachep, new); 25035beb4930SRik van Riel out_err: 25045beb4930SRik van Riel return err; 25051da177e4SLinus Torvalds } 25061da177e4SLinus Torvalds 2507659ace58SKOSAKI Motohiro /* 2508659ace58SKOSAKI Motohiro * Split a vma into two pieces at address 'addr', a new vma is allocated 2509659ace58SKOSAKI Motohiro * either for the first part or the tail. 2510659ace58SKOSAKI Motohiro */ 2511659ace58SKOSAKI Motohiro int split_vma(struct mm_struct *mm, struct vm_area_struct *vma, 2512659ace58SKOSAKI Motohiro unsigned long addr, int new_below) 2513659ace58SKOSAKI Motohiro { 2514659ace58SKOSAKI Motohiro if (mm->map_count >= sysctl_max_map_count) 2515659ace58SKOSAKI Motohiro return -ENOMEM; 2516659ace58SKOSAKI Motohiro 2517659ace58SKOSAKI Motohiro return __split_vma(mm, vma, addr, new_below); 2518659ace58SKOSAKI Motohiro } 2519659ace58SKOSAKI Motohiro 25201da177e4SLinus Torvalds /* Munmap is split into 2 main parts -- this part which finds 25211da177e4SLinus Torvalds * what needs doing, and the areas themselves, which do the 25221da177e4SLinus Torvalds * work. This now handles partial unmappings. 25231da177e4SLinus Torvalds * Jeremy Fitzhardinge <jeremy@goop.org> 25241da177e4SLinus Torvalds */ 25251da177e4SLinus Torvalds int do_munmap(struct mm_struct *mm, unsigned long start, size_t len) 25261da177e4SLinus Torvalds { 25271da177e4SLinus Torvalds unsigned long end; 2528146425a3SHugh Dickins struct vm_area_struct *vma, *prev, *last; 25291da177e4SLinus Torvalds 25301da177e4SLinus Torvalds if ((start & ~PAGE_MASK) || start > TASK_SIZE || len > TASK_SIZE-start) 25311da177e4SLinus Torvalds return -EINVAL; 25321da177e4SLinus Torvalds 2533cc71aba3Svishnu.ps len = PAGE_ALIGN(len); 2534cc71aba3Svishnu.ps if (len == 0) 25351da177e4SLinus Torvalds return -EINVAL; 25361da177e4SLinus Torvalds 25371da177e4SLinus Torvalds /* Find the first overlapping VMA */ 25389be34c9dSLinus Torvalds vma = find_vma(mm, start); 2539146425a3SHugh Dickins if (!vma) 25401da177e4SLinus Torvalds return 0; 25419be34c9dSLinus Torvalds prev = vma->vm_prev; 2542146425a3SHugh Dickins /* we have start < vma->vm_end */ 25431da177e4SLinus Torvalds 25441da177e4SLinus Torvalds /* if it doesn't overlap, we have nothing.. */ 25451da177e4SLinus Torvalds end = start + len; 2546146425a3SHugh Dickins if (vma->vm_start >= end) 25471da177e4SLinus Torvalds return 0; 25481da177e4SLinus Torvalds 25491da177e4SLinus Torvalds /* 25501da177e4SLinus Torvalds * If we need to split any vma, do it now to save pain later. 25511da177e4SLinus Torvalds * 25521da177e4SLinus Torvalds * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially 25531da177e4SLinus Torvalds * unmapped vm_area_struct will remain in use: so lower split_vma 25541da177e4SLinus Torvalds * places tmp vma above, and higher split_vma places tmp vma below. 25551da177e4SLinus Torvalds */ 2556146425a3SHugh Dickins if (start > vma->vm_start) { 2557659ace58SKOSAKI Motohiro int error; 2558659ace58SKOSAKI Motohiro 2559659ace58SKOSAKI Motohiro /* 2560659ace58SKOSAKI Motohiro * Make sure that map_count on return from munmap() will 2561659ace58SKOSAKI Motohiro * not exceed its limit; but let map_count go just above 2562659ace58SKOSAKI Motohiro * its limit temporarily, to help free resources as expected. 2563659ace58SKOSAKI Motohiro */ 2564659ace58SKOSAKI Motohiro if (end < vma->vm_end && mm->map_count >= sysctl_max_map_count) 2565659ace58SKOSAKI Motohiro return -ENOMEM; 2566659ace58SKOSAKI Motohiro 2567659ace58SKOSAKI Motohiro error = __split_vma(mm, vma, start, 0); 25681da177e4SLinus Torvalds if (error) 25691da177e4SLinus Torvalds return error; 2570146425a3SHugh Dickins prev = vma; 25711da177e4SLinus Torvalds } 25721da177e4SLinus Torvalds 25731da177e4SLinus Torvalds /* Does it split the last one? */ 25741da177e4SLinus Torvalds last = find_vma(mm, end); 25751da177e4SLinus Torvalds if (last && end > last->vm_start) { 2576659ace58SKOSAKI Motohiro int error = __split_vma(mm, last, end, 1); 25771da177e4SLinus Torvalds if (error) 25781da177e4SLinus Torvalds return error; 25791da177e4SLinus Torvalds } 2580146425a3SHugh Dickins vma = prev ? prev->vm_next : mm->mmap; 25811da177e4SLinus Torvalds 25821da177e4SLinus Torvalds /* 2583ba470de4SRik van Riel * unlock any mlock()ed ranges before detaching vmas 2584ba470de4SRik van Riel */ 2585ba470de4SRik van Riel if (mm->locked_vm) { 2586ba470de4SRik van Riel struct vm_area_struct *tmp = vma; 2587ba470de4SRik van Riel while (tmp && tmp->vm_start < end) { 2588ba470de4SRik van Riel if (tmp->vm_flags & VM_LOCKED) { 2589ba470de4SRik van Riel mm->locked_vm -= vma_pages(tmp); 2590ba470de4SRik van Riel munlock_vma_pages_all(tmp); 2591ba470de4SRik van Riel } 2592ba470de4SRik van Riel tmp = tmp->vm_next; 2593ba470de4SRik van Riel } 2594ba470de4SRik van Riel } 2595ba470de4SRik van Riel 2596ba470de4SRik van Riel /* 25971da177e4SLinus Torvalds * Remove the vma's, and unmap the actual pages 25981da177e4SLinus Torvalds */ 2599146425a3SHugh Dickins detach_vmas_to_be_unmapped(mm, vma, prev, end); 2600146425a3SHugh Dickins unmap_region(mm, vma, prev, start, end); 26011da177e4SLinus Torvalds 26021de4fa14SDave Hansen arch_unmap(mm, vma, start, end); 26031de4fa14SDave Hansen 26041da177e4SLinus Torvalds /* Fix up all other VM information */ 26052c0b3814SHugh Dickins remove_vma_list(mm, vma); 26061da177e4SLinus Torvalds 26071da177e4SLinus Torvalds return 0; 26081da177e4SLinus Torvalds } 26091da177e4SLinus Torvalds 2610bfce281cSAl Viro int vm_munmap(unsigned long start, size_t len) 2611a46ef99dSLinus Torvalds { 2612a46ef99dSLinus Torvalds int ret; 2613bfce281cSAl Viro struct mm_struct *mm = current->mm; 2614a46ef99dSLinus Torvalds 2615a46ef99dSLinus Torvalds down_write(&mm->mmap_sem); 2616a46ef99dSLinus Torvalds ret = do_munmap(mm, start, len); 2617a46ef99dSLinus Torvalds up_write(&mm->mmap_sem); 2618a46ef99dSLinus Torvalds return ret; 2619a46ef99dSLinus Torvalds } 2620a46ef99dSLinus Torvalds EXPORT_SYMBOL(vm_munmap); 2621a46ef99dSLinus Torvalds 26226a6160a7SHeiko Carstens SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len) 26231da177e4SLinus Torvalds { 26241da177e4SLinus Torvalds profile_munmap(addr); 2625bfce281cSAl Viro return vm_munmap(addr, len); 26261da177e4SLinus Torvalds } 26271da177e4SLinus Torvalds 2628c8d78c18SKirill A. Shutemov 2629c8d78c18SKirill A. Shutemov /* 2630c8d78c18SKirill A. Shutemov * Emulation of deprecated remap_file_pages() syscall. 2631c8d78c18SKirill A. Shutemov */ 2632c8d78c18SKirill A. Shutemov SYSCALL_DEFINE5(remap_file_pages, unsigned long, start, unsigned long, size, 2633c8d78c18SKirill A. Shutemov unsigned long, prot, unsigned long, pgoff, unsigned long, flags) 2634c8d78c18SKirill A. Shutemov { 2635c8d78c18SKirill A. Shutemov 2636c8d78c18SKirill A. Shutemov struct mm_struct *mm = current->mm; 2637c8d78c18SKirill A. Shutemov struct vm_area_struct *vma; 2638c8d78c18SKirill A. Shutemov unsigned long populate = 0; 2639c8d78c18SKirill A. Shutemov unsigned long ret = -EINVAL; 2640c8d78c18SKirill A. Shutemov struct file *file; 2641c8d78c18SKirill A. Shutemov 2642c8d78c18SKirill A. Shutemov pr_warn_once("%s (%d) uses deprecated remap_file_pages() syscall. " 2643c8d78c18SKirill A. Shutemov "See Documentation/vm/remap_file_pages.txt.\n", 2644c8d78c18SKirill A. Shutemov current->comm, current->pid); 2645c8d78c18SKirill A. Shutemov 2646c8d78c18SKirill A. Shutemov if (prot) 2647c8d78c18SKirill A. Shutemov return ret; 2648c8d78c18SKirill A. Shutemov start = start & PAGE_MASK; 2649c8d78c18SKirill A. Shutemov size = size & PAGE_MASK; 2650c8d78c18SKirill A. Shutemov 2651c8d78c18SKirill A. Shutemov if (start + size <= start) 2652c8d78c18SKirill A. Shutemov return ret; 2653c8d78c18SKirill A. Shutemov 2654c8d78c18SKirill A. Shutemov /* Does pgoff wrap? */ 2655c8d78c18SKirill A. Shutemov if (pgoff + (size >> PAGE_SHIFT) < pgoff) 2656c8d78c18SKirill A. Shutemov return ret; 2657c8d78c18SKirill A. Shutemov 2658c8d78c18SKirill A. Shutemov down_write(&mm->mmap_sem); 2659c8d78c18SKirill A. Shutemov vma = find_vma(mm, start); 2660c8d78c18SKirill A. Shutemov 2661c8d78c18SKirill A. Shutemov if (!vma || !(vma->vm_flags & VM_SHARED)) 2662c8d78c18SKirill A. Shutemov goto out; 2663c8d78c18SKirill A. Shutemov 2664c8d78c18SKirill A. Shutemov if (start < vma->vm_start || start + size > vma->vm_end) 2665c8d78c18SKirill A. Shutemov goto out; 2666c8d78c18SKirill A. Shutemov 2667c8d78c18SKirill A. Shutemov if (pgoff == linear_page_index(vma, start)) { 2668c8d78c18SKirill A. Shutemov ret = 0; 2669c8d78c18SKirill A. Shutemov goto out; 2670c8d78c18SKirill A. Shutemov } 2671c8d78c18SKirill A. Shutemov 2672c8d78c18SKirill A. Shutemov prot |= vma->vm_flags & VM_READ ? PROT_READ : 0; 2673c8d78c18SKirill A. Shutemov prot |= vma->vm_flags & VM_WRITE ? PROT_WRITE : 0; 2674c8d78c18SKirill A. Shutemov prot |= vma->vm_flags & VM_EXEC ? PROT_EXEC : 0; 2675c8d78c18SKirill A. Shutemov 2676c8d78c18SKirill A. Shutemov flags &= MAP_NONBLOCK; 2677c8d78c18SKirill A. Shutemov flags |= MAP_SHARED | MAP_FIXED | MAP_POPULATE; 2678c8d78c18SKirill A. Shutemov if (vma->vm_flags & VM_LOCKED) { 2679c8d78c18SKirill A. Shutemov flags |= MAP_LOCKED; 2680c8d78c18SKirill A. Shutemov /* drop PG_Mlocked flag for over-mapped range */ 2681c8d78c18SKirill A. Shutemov munlock_vma_pages_range(vma, start, start + size); 2682c8d78c18SKirill A. Shutemov } 2683c8d78c18SKirill A. Shutemov 2684c8d78c18SKirill A. Shutemov file = get_file(vma->vm_file); 2685c8d78c18SKirill A. Shutemov ret = do_mmap_pgoff(vma->vm_file, start, size, 2686c8d78c18SKirill A. Shutemov prot, flags, pgoff, &populate); 2687c8d78c18SKirill A. Shutemov fput(file); 2688c8d78c18SKirill A. Shutemov out: 2689c8d78c18SKirill A. Shutemov up_write(&mm->mmap_sem); 2690c8d78c18SKirill A. Shutemov if (populate) 2691c8d78c18SKirill A. Shutemov mm_populate(ret, populate); 2692c8d78c18SKirill A. Shutemov if (!IS_ERR_VALUE(ret)) 2693c8d78c18SKirill A. Shutemov ret = 0; 2694c8d78c18SKirill A. Shutemov return ret; 2695c8d78c18SKirill A. Shutemov } 2696c8d78c18SKirill A. Shutemov 26971da177e4SLinus Torvalds static inline void verify_mm_writelocked(struct mm_struct *mm) 26981da177e4SLinus Torvalds { 2699a241ec65SPaul E. McKenney #ifdef CONFIG_DEBUG_VM 27001da177e4SLinus Torvalds if (unlikely(down_read_trylock(&mm->mmap_sem))) { 27011da177e4SLinus Torvalds WARN_ON(1); 27021da177e4SLinus Torvalds up_read(&mm->mmap_sem); 27031da177e4SLinus Torvalds } 27041da177e4SLinus Torvalds #endif 27051da177e4SLinus Torvalds } 27061da177e4SLinus Torvalds 27071da177e4SLinus Torvalds /* 27081da177e4SLinus Torvalds * this is really a simplified "do_mmap". it only handles 27091da177e4SLinus Torvalds * anonymous maps. eventually we may be able to do some 27101da177e4SLinus Torvalds * brk-specific accounting here. 27111da177e4SLinus Torvalds */ 2712e4eb1ff6SLinus Torvalds static unsigned long do_brk(unsigned long addr, unsigned long len) 27131da177e4SLinus Torvalds { 27141da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 27151da177e4SLinus Torvalds struct vm_area_struct *vma, *prev; 27161da177e4SLinus Torvalds unsigned long flags; 27171da177e4SLinus Torvalds struct rb_node **rb_link, *rb_parent; 27181da177e4SLinus Torvalds pgoff_t pgoff = addr >> PAGE_SHIFT; 27193a459756SKirill Korotaev int error; 27201da177e4SLinus Torvalds 27211da177e4SLinus Torvalds len = PAGE_ALIGN(len); 27221da177e4SLinus Torvalds if (!len) 27231da177e4SLinus Torvalds return addr; 27241da177e4SLinus Torvalds 27253a459756SKirill Korotaev flags = VM_DATA_DEFAULT_FLAGS | VM_ACCOUNT | mm->def_flags; 27263a459756SKirill Korotaev 27272c6a1016SAl Viro error = get_unmapped_area(NULL, addr, len, 0, MAP_FIXED); 27282c6a1016SAl Viro if (error & ~PAGE_MASK) 27293a459756SKirill Korotaev return error; 27303a459756SKirill Korotaev 2731363ee17fSDavidlohr Bueso error = mlock_future_check(mm, mm->def_flags, len); 2732363ee17fSDavidlohr Bueso if (error) 2733363ee17fSDavidlohr Bueso return error; 27341da177e4SLinus Torvalds 27351da177e4SLinus Torvalds /* 27361da177e4SLinus Torvalds * mm->mmap_sem is required to protect against another thread 27371da177e4SLinus Torvalds * changing the mappings in case we sleep. 27381da177e4SLinus Torvalds */ 27391da177e4SLinus Torvalds verify_mm_writelocked(mm); 27401da177e4SLinus Torvalds 27411da177e4SLinus Torvalds /* 27421da177e4SLinus Torvalds * Clear old maps. this also does some error checking for us 27431da177e4SLinus Torvalds */ 27441da177e4SLinus Torvalds munmap_back: 27456597d783SHugh Dickins if (find_vma_links(mm, addr, addr + len, &prev, &rb_link, &rb_parent)) { 27461da177e4SLinus Torvalds if (do_munmap(mm, addr, len)) 27471da177e4SLinus Torvalds return -ENOMEM; 27481da177e4SLinus Torvalds goto munmap_back; 27491da177e4SLinus Torvalds } 27501da177e4SLinus Torvalds 27511da177e4SLinus Torvalds /* Check against address space limits *after* clearing old maps... */ 2752119f657cSakpm@osdl.org if (!may_expand_vm(mm, len >> PAGE_SHIFT)) 27531da177e4SLinus Torvalds return -ENOMEM; 27541da177e4SLinus Torvalds 27551da177e4SLinus Torvalds if (mm->map_count > sysctl_max_map_count) 27561da177e4SLinus Torvalds return -ENOMEM; 27571da177e4SLinus Torvalds 2758191c5424SAl Viro if (security_vm_enough_memory_mm(mm, len >> PAGE_SHIFT)) 27591da177e4SLinus Torvalds return -ENOMEM; 27601da177e4SLinus Torvalds 27611da177e4SLinus Torvalds /* Can we just expand an old private anonymous mapping? */ 2762ba470de4SRik van Riel vma = vma_merge(mm, prev, addr, addr + len, flags, 2763ba470de4SRik van Riel NULL, NULL, pgoff, NULL); 2764ba470de4SRik van Riel if (vma) 27651da177e4SLinus Torvalds goto out; 27661da177e4SLinus Torvalds 27671da177e4SLinus Torvalds /* 27681da177e4SLinus Torvalds * create a vma struct for an anonymous mapping 27691da177e4SLinus Torvalds */ 2770c5e3b83eSPekka Enberg vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL); 27711da177e4SLinus Torvalds if (!vma) { 27721da177e4SLinus Torvalds vm_unacct_memory(len >> PAGE_SHIFT); 27731da177e4SLinus Torvalds return -ENOMEM; 27741da177e4SLinus Torvalds } 27751da177e4SLinus Torvalds 27765beb4930SRik van Riel INIT_LIST_HEAD(&vma->anon_vma_chain); 27771da177e4SLinus Torvalds vma->vm_mm = mm; 27781da177e4SLinus Torvalds vma->vm_start = addr; 27791da177e4SLinus Torvalds vma->vm_end = addr + len; 27801da177e4SLinus Torvalds vma->vm_pgoff = pgoff; 27811da177e4SLinus Torvalds vma->vm_flags = flags; 27823ed75eb8SColy Li vma->vm_page_prot = vm_get_page_prot(flags); 27831da177e4SLinus Torvalds vma_link(mm, vma, prev, rb_link, rb_parent); 27841da177e4SLinus Torvalds out: 27853af9e859SEric B Munson perf_event_mmap(vma); 27861da177e4SLinus Torvalds mm->total_vm += len >> PAGE_SHIFT; 2787128557ffSMichel Lespinasse if (flags & VM_LOCKED) 2788ba470de4SRik van Riel mm->locked_vm += (len >> PAGE_SHIFT); 2789d9104d1cSCyrill Gorcunov vma->vm_flags |= VM_SOFTDIRTY; 27901da177e4SLinus Torvalds return addr; 27911da177e4SLinus Torvalds } 27921da177e4SLinus Torvalds 2793e4eb1ff6SLinus Torvalds unsigned long vm_brk(unsigned long addr, unsigned long len) 2794e4eb1ff6SLinus Torvalds { 2795e4eb1ff6SLinus Torvalds struct mm_struct *mm = current->mm; 2796e4eb1ff6SLinus Torvalds unsigned long ret; 2797128557ffSMichel Lespinasse bool populate; 2798e4eb1ff6SLinus Torvalds 2799e4eb1ff6SLinus Torvalds down_write(&mm->mmap_sem); 2800e4eb1ff6SLinus Torvalds ret = do_brk(addr, len); 2801128557ffSMichel Lespinasse populate = ((mm->def_flags & VM_LOCKED) != 0); 2802e4eb1ff6SLinus Torvalds up_write(&mm->mmap_sem); 2803128557ffSMichel Lespinasse if (populate) 2804128557ffSMichel Lespinasse mm_populate(addr, len); 2805e4eb1ff6SLinus Torvalds return ret; 2806e4eb1ff6SLinus Torvalds } 2807e4eb1ff6SLinus Torvalds EXPORT_SYMBOL(vm_brk); 28081da177e4SLinus Torvalds 28091da177e4SLinus Torvalds /* Release all mmaps. */ 28101da177e4SLinus Torvalds void exit_mmap(struct mm_struct *mm) 28111da177e4SLinus Torvalds { 2812d16dfc55SPeter Zijlstra struct mmu_gather tlb; 2813ba470de4SRik van Riel struct vm_area_struct *vma; 28141da177e4SLinus Torvalds unsigned long nr_accounted = 0; 28151da177e4SLinus Torvalds 2816d6dd61c8SJeremy Fitzhardinge /* mm's last user has gone, and its about to be pulled down */ 2817cddb8a5cSAndrea Arcangeli mmu_notifier_release(mm); 2818d6dd61c8SJeremy Fitzhardinge 2819ba470de4SRik van Riel if (mm->locked_vm) { 2820ba470de4SRik van Riel vma = mm->mmap; 2821ba470de4SRik van Riel while (vma) { 2822ba470de4SRik van Riel if (vma->vm_flags & VM_LOCKED) 2823ba470de4SRik van Riel munlock_vma_pages_all(vma); 2824ba470de4SRik van Riel vma = vma->vm_next; 2825ba470de4SRik van Riel } 2826ba470de4SRik van Riel } 28279480c53eSJeremy Fitzhardinge 28289480c53eSJeremy Fitzhardinge arch_exit_mmap(mm); 28299480c53eSJeremy Fitzhardinge 2830ba470de4SRik van Riel vma = mm->mmap; 28319480c53eSJeremy Fitzhardinge if (!vma) /* Can happen if dup_mmap() received an OOM */ 28329480c53eSJeremy Fitzhardinge return; 28339480c53eSJeremy Fitzhardinge 28341da177e4SLinus Torvalds lru_add_drain(); 28351da177e4SLinus Torvalds flush_cache_mm(mm); 28362b047252SLinus Torvalds tlb_gather_mmu(&tlb, mm, 0, -1); 2837901608d9SOleg Nesterov /* update_hiwater_rss(mm) here? but nobody should be looking */ 2838e0da382cSHugh Dickins /* Use -1 here to ensure all VMAs in the mm are unmapped */ 28394f74d2c8SLinus Torvalds unmap_vmas(&tlb, vma, 0, -1); 28409ba69294SHugh Dickins 28416ee8630eSHugh Dickins free_pgtables(&tlb, vma, FIRST_USER_ADDRESS, USER_PGTABLES_CEILING); 2842853f5e26SAl Viro tlb_finish_mmu(&tlb, 0, -1); 28431da177e4SLinus Torvalds 28441da177e4SLinus Torvalds /* 28458f4f8c16SHugh Dickins * Walk the list again, actually closing and freeing it, 28468f4f8c16SHugh Dickins * with preemption enabled, without holding any MM locks. 28471da177e4SLinus Torvalds */ 28484f74d2c8SLinus Torvalds while (vma) { 28494f74d2c8SLinus Torvalds if (vma->vm_flags & VM_ACCOUNT) 28504f74d2c8SLinus Torvalds nr_accounted += vma_pages(vma); 2851a8fb5618SHugh Dickins vma = remove_vma(vma); 28524f74d2c8SLinus Torvalds } 28534f74d2c8SLinus Torvalds vm_unacct_memory(nr_accounted); 2854e0da382cSHugh Dickins 2855e1f56c89SKirill A. Shutemov WARN_ON(atomic_long_read(&mm->nr_ptes) > 2856e1f56c89SKirill A. Shutemov (FIRST_USER_ADDRESS+PMD_SIZE-1)>>PMD_SHIFT); 28571da177e4SLinus Torvalds } 28581da177e4SLinus Torvalds 28591da177e4SLinus Torvalds /* Insert vm structure into process list sorted by address 28601da177e4SLinus Torvalds * and into the inode's i_mmap tree. If vm_file is non-NULL 2861c8c06efaSDavidlohr Bueso * then i_mmap_rwsem is taken here. 28621da177e4SLinus Torvalds */ 28631da177e4SLinus Torvalds int insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma) 28641da177e4SLinus Torvalds { 28656597d783SHugh Dickins struct vm_area_struct *prev; 28661da177e4SLinus Torvalds struct rb_node **rb_link, *rb_parent; 28671da177e4SLinus Torvalds 28681da177e4SLinus Torvalds /* 28691da177e4SLinus Torvalds * The vm_pgoff of a purely anonymous vma should be irrelevant 28701da177e4SLinus Torvalds * until its first write fault, when page's anon_vma and index 28711da177e4SLinus Torvalds * are set. But now set the vm_pgoff it will almost certainly 28721da177e4SLinus Torvalds * end up with (unless mremap moves it elsewhere before that 28731da177e4SLinus Torvalds * first wfault), so /proc/pid/maps tells a consistent story. 28741da177e4SLinus Torvalds * 28751da177e4SLinus Torvalds * By setting it to reflect the virtual start address of the 28761da177e4SLinus Torvalds * vma, merges and splits can happen in a seamless way, just 28771da177e4SLinus Torvalds * using the existing file pgoff checks and manipulations. 28781da177e4SLinus Torvalds * Similarly in do_mmap_pgoff and in do_brk. 28791da177e4SLinus Torvalds */ 28801da177e4SLinus Torvalds if (!vma->vm_file) { 28811da177e4SLinus Torvalds BUG_ON(vma->anon_vma); 28821da177e4SLinus Torvalds vma->vm_pgoff = vma->vm_start >> PAGE_SHIFT; 28831da177e4SLinus Torvalds } 28846597d783SHugh Dickins if (find_vma_links(mm, vma->vm_start, vma->vm_end, 28856597d783SHugh Dickins &prev, &rb_link, &rb_parent)) 28861da177e4SLinus Torvalds return -ENOMEM; 28872fd4ef85SHugh Dickins if ((vma->vm_flags & VM_ACCOUNT) && 288834b4e4aaSAlan Cox security_vm_enough_memory_mm(mm, vma_pages(vma))) 28892fd4ef85SHugh Dickins return -ENOMEM; 28902b144498SSrikar Dronamraju 28911da177e4SLinus Torvalds vma_link(mm, vma, prev, rb_link, rb_parent); 28921da177e4SLinus Torvalds return 0; 28931da177e4SLinus Torvalds } 28941da177e4SLinus Torvalds 28951da177e4SLinus Torvalds /* 28961da177e4SLinus Torvalds * Copy the vma structure to a new location in the same mm, 28971da177e4SLinus Torvalds * prior to moving page table entries, to effect an mremap move. 28981da177e4SLinus Torvalds */ 28991da177e4SLinus Torvalds struct vm_area_struct *copy_vma(struct vm_area_struct **vmap, 290038a76013SMichel Lespinasse unsigned long addr, unsigned long len, pgoff_t pgoff, 290138a76013SMichel Lespinasse bool *need_rmap_locks) 29021da177e4SLinus Torvalds { 29031da177e4SLinus Torvalds struct vm_area_struct *vma = *vmap; 29041da177e4SLinus Torvalds unsigned long vma_start = vma->vm_start; 29051da177e4SLinus Torvalds struct mm_struct *mm = vma->vm_mm; 29061da177e4SLinus Torvalds struct vm_area_struct *new_vma, *prev; 29071da177e4SLinus Torvalds struct rb_node **rb_link, *rb_parent; 2908948f017bSAndrea Arcangeli bool faulted_in_anon_vma = true; 29091da177e4SLinus Torvalds 29101da177e4SLinus Torvalds /* 29111da177e4SLinus Torvalds * If anonymous vma has not yet been faulted, update new pgoff 29121da177e4SLinus Torvalds * to match new location, to increase its chance of merging. 29131da177e4SLinus Torvalds */ 2914948f017bSAndrea Arcangeli if (unlikely(!vma->vm_file && !vma->anon_vma)) { 29151da177e4SLinus Torvalds pgoff = addr >> PAGE_SHIFT; 2916948f017bSAndrea Arcangeli faulted_in_anon_vma = false; 2917948f017bSAndrea Arcangeli } 29181da177e4SLinus Torvalds 29196597d783SHugh Dickins if (find_vma_links(mm, addr, addr + len, &prev, &rb_link, &rb_parent)) 29206597d783SHugh Dickins return NULL; /* should never get here */ 29211da177e4SLinus Torvalds new_vma = vma_merge(mm, prev, addr, addr + len, vma->vm_flags, 29221da177e4SLinus Torvalds vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma)); 29231da177e4SLinus Torvalds if (new_vma) { 29241da177e4SLinus Torvalds /* 29251da177e4SLinus Torvalds * Source vma may have been merged into new_vma 29261da177e4SLinus Torvalds */ 2927948f017bSAndrea Arcangeli if (unlikely(vma_start >= new_vma->vm_start && 2928948f017bSAndrea Arcangeli vma_start < new_vma->vm_end)) { 2929948f017bSAndrea Arcangeli /* 2930948f017bSAndrea Arcangeli * The only way we can get a vma_merge with 2931948f017bSAndrea Arcangeli * self during an mremap is if the vma hasn't 2932948f017bSAndrea Arcangeli * been faulted in yet and we were allowed to 2933948f017bSAndrea Arcangeli * reset the dst vma->vm_pgoff to the 2934948f017bSAndrea Arcangeli * destination address of the mremap to allow 2935948f017bSAndrea Arcangeli * the merge to happen. mremap must change the 2936948f017bSAndrea Arcangeli * vm_pgoff linearity between src and dst vmas 2937948f017bSAndrea Arcangeli * (in turn preventing a vma_merge) to be 2938948f017bSAndrea Arcangeli * safe. It is only safe to keep the vm_pgoff 2939948f017bSAndrea Arcangeli * linear if there are no pages mapped yet. 2940948f017bSAndrea Arcangeli */ 294181d1b09cSSasha Levin VM_BUG_ON_VMA(faulted_in_anon_vma, new_vma); 294238a76013SMichel Lespinasse *vmap = vma = new_vma; 2943108d6642SMichel Lespinasse } 294438a76013SMichel Lespinasse *need_rmap_locks = (new_vma->vm_pgoff <= vma->vm_pgoff); 29451da177e4SLinus Torvalds } else { 2946e94b1766SChristoph Lameter new_vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL); 29471da177e4SLinus Torvalds if (new_vma) { 29481da177e4SLinus Torvalds *new_vma = *vma; 29491da177e4SLinus Torvalds new_vma->vm_start = addr; 29501da177e4SLinus Torvalds new_vma->vm_end = addr + len; 29511da177e4SLinus Torvalds new_vma->vm_pgoff = pgoff; 2952ef0855d3SOleg Nesterov if (vma_dup_policy(vma, new_vma)) 2953523d4e20SMichel Lespinasse goto out_free_vma; 2954523d4e20SMichel Lespinasse INIT_LIST_HEAD(&new_vma->anon_vma_chain); 2955523d4e20SMichel Lespinasse if (anon_vma_clone(new_vma, vma)) 2956523d4e20SMichel Lespinasse goto out_free_mempol; 2957e9714acfSKonstantin Khlebnikov if (new_vma->vm_file) 29581da177e4SLinus Torvalds get_file(new_vma->vm_file); 29591da177e4SLinus Torvalds if (new_vma->vm_ops && new_vma->vm_ops->open) 29601da177e4SLinus Torvalds new_vma->vm_ops->open(new_vma); 29611da177e4SLinus Torvalds vma_link(mm, new_vma, prev, rb_link, rb_parent); 296238a76013SMichel Lespinasse *need_rmap_locks = false; 29631da177e4SLinus Torvalds } 29641da177e4SLinus Torvalds } 29651da177e4SLinus Torvalds return new_vma; 29665beb4930SRik van Riel 29675beb4930SRik van Riel out_free_mempol: 2968ef0855d3SOleg Nesterov mpol_put(vma_policy(new_vma)); 29695beb4930SRik van Riel out_free_vma: 29705beb4930SRik van Riel kmem_cache_free(vm_area_cachep, new_vma); 29715beb4930SRik van Riel return NULL; 29721da177e4SLinus Torvalds } 2973119f657cSakpm@osdl.org 2974119f657cSakpm@osdl.org /* 2975119f657cSakpm@osdl.org * Return true if the calling process may expand its vm space by the passed 2976119f657cSakpm@osdl.org * number of pages 2977119f657cSakpm@osdl.org */ 2978119f657cSakpm@osdl.org int may_expand_vm(struct mm_struct *mm, unsigned long npages) 2979119f657cSakpm@osdl.org { 2980119f657cSakpm@osdl.org unsigned long cur = mm->total_vm; /* pages */ 2981119f657cSakpm@osdl.org unsigned long lim; 2982119f657cSakpm@osdl.org 298359e99e5bSJiri Slaby lim = rlimit(RLIMIT_AS) >> PAGE_SHIFT; 2984119f657cSakpm@osdl.org 2985119f657cSakpm@osdl.org if (cur + npages > lim) 2986119f657cSakpm@osdl.org return 0; 2987119f657cSakpm@osdl.org return 1; 2988119f657cSakpm@osdl.org } 2989fa5dc22fSRoland McGrath 2990a62c34bdSAndy Lutomirski static int special_mapping_fault(struct vm_area_struct *vma, 2991a62c34bdSAndy Lutomirski struct vm_fault *vmf); 2992a62c34bdSAndy Lutomirski 2993a62c34bdSAndy Lutomirski /* 2994a62c34bdSAndy Lutomirski * Having a close hook prevents vma merging regardless of flags. 2995a62c34bdSAndy Lutomirski */ 2996a62c34bdSAndy Lutomirski static void special_mapping_close(struct vm_area_struct *vma) 2997a62c34bdSAndy Lutomirski { 2998a62c34bdSAndy Lutomirski } 2999a62c34bdSAndy Lutomirski 3000a62c34bdSAndy Lutomirski static const char *special_mapping_name(struct vm_area_struct *vma) 3001a62c34bdSAndy Lutomirski { 3002a62c34bdSAndy Lutomirski return ((struct vm_special_mapping *)vma->vm_private_data)->name; 3003a62c34bdSAndy Lutomirski } 3004a62c34bdSAndy Lutomirski 3005a62c34bdSAndy Lutomirski static const struct vm_operations_struct special_mapping_vmops = { 3006a62c34bdSAndy Lutomirski .close = special_mapping_close, 3007a62c34bdSAndy Lutomirski .fault = special_mapping_fault, 3008a62c34bdSAndy Lutomirski .name = special_mapping_name, 3009a62c34bdSAndy Lutomirski }; 3010a62c34bdSAndy Lutomirski 3011a62c34bdSAndy Lutomirski static const struct vm_operations_struct legacy_special_mapping_vmops = { 3012a62c34bdSAndy Lutomirski .close = special_mapping_close, 3013a62c34bdSAndy Lutomirski .fault = special_mapping_fault, 3014a62c34bdSAndy Lutomirski }; 3015fa5dc22fSRoland McGrath 3016b1d0e4f5SNick Piggin static int special_mapping_fault(struct vm_area_struct *vma, 3017b1d0e4f5SNick Piggin struct vm_fault *vmf) 3018fa5dc22fSRoland McGrath { 3019b1d0e4f5SNick Piggin pgoff_t pgoff; 3020fa5dc22fSRoland McGrath struct page **pages; 3021fa5dc22fSRoland McGrath 3022b1d0e4f5SNick Piggin /* 3023b1d0e4f5SNick Piggin * special mappings have no vm_file, and in that case, the mm 3024b1d0e4f5SNick Piggin * uses vm_pgoff internally. So we have to subtract it from here. 3025b1d0e4f5SNick Piggin * We are allowed to do this because we are the mm; do not copy 3026b1d0e4f5SNick Piggin * this code into drivers! 3027b1d0e4f5SNick Piggin */ 3028b1d0e4f5SNick Piggin pgoff = vmf->pgoff - vma->vm_pgoff; 3029fa5dc22fSRoland McGrath 3030a62c34bdSAndy Lutomirski if (vma->vm_ops == &legacy_special_mapping_vmops) 3031a62c34bdSAndy Lutomirski pages = vma->vm_private_data; 3032a62c34bdSAndy Lutomirski else 3033a62c34bdSAndy Lutomirski pages = ((struct vm_special_mapping *)vma->vm_private_data)-> 3034a62c34bdSAndy Lutomirski pages; 3035a62c34bdSAndy Lutomirski 3036a62c34bdSAndy Lutomirski for (; pgoff && *pages; ++pages) 3037b1d0e4f5SNick Piggin pgoff--; 3038fa5dc22fSRoland McGrath 3039fa5dc22fSRoland McGrath if (*pages) { 3040fa5dc22fSRoland McGrath struct page *page = *pages; 3041fa5dc22fSRoland McGrath get_page(page); 3042b1d0e4f5SNick Piggin vmf->page = page; 3043b1d0e4f5SNick Piggin return 0; 3044fa5dc22fSRoland McGrath } 3045fa5dc22fSRoland McGrath 3046b1d0e4f5SNick Piggin return VM_FAULT_SIGBUS; 3047fa5dc22fSRoland McGrath } 3048fa5dc22fSRoland McGrath 3049a62c34bdSAndy Lutomirski static struct vm_area_struct *__install_special_mapping( 3050a62c34bdSAndy Lutomirski struct mm_struct *mm, 3051fa5dc22fSRoland McGrath unsigned long addr, unsigned long len, 3052a62c34bdSAndy Lutomirski unsigned long vm_flags, const struct vm_operations_struct *ops, 3053a62c34bdSAndy Lutomirski void *priv) 3054fa5dc22fSRoland McGrath { 3055462e635eSTavis Ormandy int ret; 3056fa5dc22fSRoland McGrath struct vm_area_struct *vma; 3057fa5dc22fSRoland McGrath 3058fa5dc22fSRoland McGrath vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL); 3059fa5dc22fSRoland McGrath if (unlikely(vma == NULL)) 30603935ed6aSStefani Seibold return ERR_PTR(-ENOMEM); 3061fa5dc22fSRoland McGrath 30625beb4930SRik van Riel INIT_LIST_HEAD(&vma->anon_vma_chain); 3063fa5dc22fSRoland McGrath vma->vm_mm = mm; 3064fa5dc22fSRoland McGrath vma->vm_start = addr; 3065fa5dc22fSRoland McGrath vma->vm_end = addr + len; 3066fa5dc22fSRoland McGrath 3067d9104d1cSCyrill Gorcunov vma->vm_flags = vm_flags | mm->def_flags | VM_DONTEXPAND | VM_SOFTDIRTY; 30683ed75eb8SColy Li vma->vm_page_prot = vm_get_page_prot(vma->vm_flags); 3069fa5dc22fSRoland McGrath 3070a62c34bdSAndy Lutomirski vma->vm_ops = ops; 3071a62c34bdSAndy Lutomirski vma->vm_private_data = priv; 3072fa5dc22fSRoland McGrath 3073462e635eSTavis Ormandy ret = insert_vm_struct(mm, vma); 3074462e635eSTavis Ormandy if (ret) 3075462e635eSTavis Ormandy goto out; 3076fa5dc22fSRoland McGrath 3077fa5dc22fSRoland McGrath mm->total_vm += len >> PAGE_SHIFT; 3078fa5dc22fSRoland McGrath 3079cdd6c482SIngo Molnar perf_event_mmap(vma); 3080089dd79dSPeter Zijlstra 30813935ed6aSStefani Seibold return vma; 3082462e635eSTavis Ormandy 3083462e635eSTavis Ormandy out: 3084462e635eSTavis Ormandy kmem_cache_free(vm_area_cachep, vma); 30853935ed6aSStefani Seibold return ERR_PTR(ret); 30863935ed6aSStefani Seibold } 30873935ed6aSStefani Seibold 3088a62c34bdSAndy Lutomirski /* 3089a62c34bdSAndy Lutomirski * Called with mm->mmap_sem held for writing. 3090a62c34bdSAndy Lutomirski * Insert a new vma covering the given region, with the given flags. 3091a62c34bdSAndy Lutomirski * Its pages are supplied by the given array of struct page *. 3092a62c34bdSAndy Lutomirski * The array can be shorter than len >> PAGE_SHIFT if it's null-terminated. 3093a62c34bdSAndy Lutomirski * The region past the last page supplied will always produce SIGBUS. 3094a62c34bdSAndy Lutomirski * The array pointer and the pages it points to are assumed to stay alive 3095a62c34bdSAndy Lutomirski * for as long as this mapping might exist. 3096a62c34bdSAndy Lutomirski */ 3097a62c34bdSAndy Lutomirski struct vm_area_struct *_install_special_mapping( 3098a62c34bdSAndy Lutomirski struct mm_struct *mm, 3099a62c34bdSAndy Lutomirski unsigned long addr, unsigned long len, 3100a62c34bdSAndy Lutomirski unsigned long vm_flags, const struct vm_special_mapping *spec) 3101a62c34bdSAndy Lutomirski { 3102a62c34bdSAndy Lutomirski return __install_special_mapping(mm, addr, len, vm_flags, 3103a62c34bdSAndy Lutomirski &special_mapping_vmops, (void *)spec); 3104a62c34bdSAndy Lutomirski } 3105a62c34bdSAndy Lutomirski 31063935ed6aSStefani Seibold int install_special_mapping(struct mm_struct *mm, 31073935ed6aSStefani Seibold unsigned long addr, unsigned long len, 31083935ed6aSStefani Seibold unsigned long vm_flags, struct page **pages) 31093935ed6aSStefani Seibold { 3110a62c34bdSAndy Lutomirski struct vm_area_struct *vma = __install_special_mapping( 3111a62c34bdSAndy Lutomirski mm, addr, len, vm_flags, &legacy_special_mapping_vmops, 3112a62c34bdSAndy Lutomirski (void *)pages); 31133935ed6aSStefani Seibold 311414bd5b45SDuan Jiong return PTR_ERR_OR_ZERO(vma); 3115fa5dc22fSRoland McGrath } 31167906d00cSAndrea Arcangeli 31177906d00cSAndrea Arcangeli static DEFINE_MUTEX(mm_all_locks_mutex); 31187906d00cSAndrea Arcangeli 3119454ed842SPeter Zijlstra static void vm_lock_anon_vma(struct mm_struct *mm, struct anon_vma *anon_vma) 31207906d00cSAndrea Arcangeli { 3121bf181b9fSMichel Lespinasse if (!test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_node)) { 31227906d00cSAndrea Arcangeli /* 31237906d00cSAndrea Arcangeli * The LSB of head.next can't change from under us 31247906d00cSAndrea Arcangeli * because we hold the mm_all_locks_mutex. 31257906d00cSAndrea Arcangeli */ 3126572043c9SJiri Kosina down_write_nest_lock(&anon_vma->root->rwsem, &mm->mmap_sem); 31277906d00cSAndrea Arcangeli /* 31287906d00cSAndrea Arcangeli * We can safely modify head.next after taking the 31295a505085SIngo Molnar * anon_vma->root->rwsem. If some other vma in this mm shares 31307906d00cSAndrea Arcangeli * the same anon_vma we won't take it again. 31317906d00cSAndrea Arcangeli * 31327906d00cSAndrea Arcangeli * No need of atomic instructions here, head.next 31337906d00cSAndrea Arcangeli * can't change from under us thanks to the 31345a505085SIngo Molnar * anon_vma->root->rwsem. 31357906d00cSAndrea Arcangeli */ 31367906d00cSAndrea Arcangeli if (__test_and_set_bit(0, (unsigned long *) 3137bf181b9fSMichel Lespinasse &anon_vma->root->rb_root.rb_node)) 31387906d00cSAndrea Arcangeli BUG(); 31397906d00cSAndrea Arcangeli } 31407906d00cSAndrea Arcangeli } 31417906d00cSAndrea Arcangeli 3142454ed842SPeter Zijlstra static void vm_lock_mapping(struct mm_struct *mm, struct address_space *mapping) 31437906d00cSAndrea Arcangeli { 31447906d00cSAndrea Arcangeli if (!test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) { 31457906d00cSAndrea Arcangeli /* 31467906d00cSAndrea Arcangeli * AS_MM_ALL_LOCKS can't change from under us because 31477906d00cSAndrea Arcangeli * we hold the mm_all_locks_mutex. 31487906d00cSAndrea Arcangeli * 31497906d00cSAndrea Arcangeli * Operations on ->flags have to be atomic because 31507906d00cSAndrea Arcangeli * even if AS_MM_ALL_LOCKS is stable thanks to the 31517906d00cSAndrea Arcangeli * mm_all_locks_mutex, there may be other cpus 31527906d00cSAndrea Arcangeli * changing other bitflags in parallel to us. 31537906d00cSAndrea Arcangeli */ 31547906d00cSAndrea Arcangeli if (test_and_set_bit(AS_MM_ALL_LOCKS, &mapping->flags)) 31557906d00cSAndrea Arcangeli BUG(); 3156c8c06efaSDavidlohr Bueso down_write_nest_lock(&mapping->i_mmap_rwsem, &mm->mmap_sem); 31577906d00cSAndrea Arcangeli } 31587906d00cSAndrea Arcangeli } 31597906d00cSAndrea Arcangeli 31607906d00cSAndrea Arcangeli /* 31617906d00cSAndrea Arcangeli * This operation locks against the VM for all pte/vma/mm related 31627906d00cSAndrea Arcangeli * operations that could ever happen on a certain mm. This includes 31637906d00cSAndrea Arcangeli * vmtruncate, try_to_unmap, and all page faults. 31647906d00cSAndrea Arcangeli * 31657906d00cSAndrea Arcangeli * The caller must take the mmap_sem in write mode before calling 31667906d00cSAndrea Arcangeli * mm_take_all_locks(). The caller isn't allowed to release the 31677906d00cSAndrea Arcangeli * mmap_sem until mm_drop_all_locks() returns. 31687906d00cSAndrea Arcangeli * 31697906d00cSAndrea Arcangeli * mmap_sem in write mode is required in order to block all operations 31707906d00cSAndrea Arcangeli * that could modify pagetables and free pages without need of 3171*27ba0644SKirill A. Shutemov * altering the vma layout. It's also needed in write mode to avoid new 31727906d00cSAndrea Arcangeli * anon_vmas to be associated with existing vmas. 31737906d00cSAndrea Arcangeli * 31747906d00cSAndrea Arcangeli * A single task can't take more than one mm_take_all_locks() in a row 31757906d00cSAndrea Arcangeli * or it would deadlock. 31767906d00cSAndrea Arcangeli * 3177bf181b9fSMichel Lespinasse * The LSB in anon_vma->rb_root.rb_node and the AS_MM_ALL_LOCKS bitflag in 31787906d00cSAndrea Arcangeli * mapping->flags avoid to take the same lock twice, if more than one 31797906d00cSAndrea Arcangeli * vma in this mm is backed by the same anon_vma or address_space. 31807906d00cSAndrea Arcangeli * 31817906d00cSAndrea Arcangeli * We can take all the locks in random order because the VM code 3182c8c06efaSDavidlohr Bueso * taking i_mmap_rwsem or anon_vma->rwsem outside the mmap_sem never 31837906d00cSAndrea Arcangeli * takes more than one of them in a row. Secondly we're protected 31847906d00cSAndrea Arcangeli * against a concurrent mm_take_all_locks() by the mm_all_locks_mutex. 31857906d00cSAndrea Arcangeli * 31867906d00cSAndrea Arcangeli * mm_take_all_locks() and mm_drop_all_locks are expensive operations 31877906d00cSAndrea Arcangeli * that may have to take thousand of locks. 31887906d00cSAndrea Arcangeli * 31897906d00cSAndrea Arcangeli * mm_take_all_locks() can fail if it's interrupted by signals. 31907906d00cSAndrea Arcangeli */ 31917906d00cSAndrea Arcangeli int mm_take_all_locks(struct mm_struct *mm) 31927906d00cSAndrea Arcangeli { 31937906d00cSAndrea Arcangeli struct vm_area_struct *vma; 31945beb4930SRik van Riel struct anon_vma_chain *avc; 31957906d00cSAndrea Arcangeli 31967906d00cSAndrea Arcangeli BUG_ON(down_read_trylock(&mm->mmap_sem)); 31977906d00cSAndrea Arcangeli 31987906d00cSAndrea Arcangeli mutex_lock(&mm_all_locks_mutex); 31997906d00cSAndrea Arcangeli 32007906d00cSAndrea Arcangeli for (vma = mm->mmap; vma; vma = vma->vm_next) { 32017906d00cSAndrea Arcangeli if (signal_pending(current)) 32027906d00cSAndrea Arcangeli goto out_unlock; 32037906d00cSAndrea Arcangeli if (vma->vm_file && vma->vm_file->f_mapping) 3204454ed842SPeter Zijlstra vm_lock_mapping(mm, vma->vm_file->f_mapping); 32057906d00cSAndrea Arcangeli } 32067cd5a02fSPeter Zijlstra 32077cd5a02fSPeter Zijlstra for (vma = mm->mmap; vma; vma = vma->vm_next) { 32087cd5a02fSPeter Zijlstra if (signal_pending(current)) 32097cd5a02fSPeter Zijlstra goto out_unlock; 32107cd5a02fSPeter Zijlstra if (vma->anon_vma) 32115beb4930SRik van Riel list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) 32125beb4930SRik van Riel vm_lock_anon_vma(mm, avc->anon_vma); 32137cd5a02fSPeter Zijlstra } 32147cd5a02fSPeter Zijlstra 3215584cff54SKautuk Consul return 0; 32167906d00cSAndrea Arcangeli 32177906d00cSAndrea Arcangeli out_unlock: 32187906d00cSAndrea Arcangeli mm_drop_all_locks(mm); 3219584cff54SKautuk Consul return -EINTR; 32207906d00cSAndrea Arcangeli } 32217906d00cSAndrea Arcangeli 32227906d00cSAndrea Arcangeli static void vm_unlock_anon_vma(struct anon_vma *anon_vma) 32237906d00cSAndrea Arcangeli { 3224bf181b9fSMichel Lespinasse if (test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_node)) { 32257906d00cSAndrea Arcangeli /* 32267906d00cSAndrea Arcangeli * The LSB of head.next can't change to 0 from under 32277906d00cSAndrea Arcangeli * us because we hold the mm_all_locks_mutex. 32287906d00cSAndrea Arcangeli * 32297906d00cSAndrea Arcangeli * We must however clear the bitflag before unlocking 3230bf181b9fSMichel Lespinasse * the vma so the users using the anon_vma->rb_root will 32317906d00cSAndrea Arcangeli * never see our bitflag. 32327906d00cSAndrea Arcangeli * 32337906d00cSAndrea Arcangeli * No need of atomic instructions here, head.next 32347906d00cSAndrea Arcangeli * can't change from under us until we release the 32355a505085SIngo Molnar * anon_vma->root->rwsem. 32367906d00cSAndrea Arcangeli */ 32377906d00cSAndrea Arcangeli if (!__test_and_clear_bit(0, (unsigned long *) 3238bf181b9fSMichel Lespinasse &anon_vma->root->rb_root.rb_node)) 32397906d00cSAndrea Arcangeli BUG(); 324008b52706SKonstantin Khlebnikov anon_vma_unlock_write(anon_vma); 32417906d00cSAndrea Arcangeli } 32427906d00cSAndrea Arcangeli } 32437906d00cSAndrea Arcangeli 32447906d00cSAndrea Arcangeli static void vm_unlock_mapping(struct address_space *mapping) 32457906d00cSAndrea Arcangeli { 32467906d00cSAndrea Arcangeli if (test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) { 32477906d00cSAndrea Arcangeli /* 32487906d00cSAndrea Arcangeli * AS_MM_ALL_LOCKS can't change to 0 from under us 32497906d00cSAndrea Arcangeli * because we hold the mm_all_locks_mutex. 32507906d00cSAndrea Arcangeli */ 325183cde9e8SDavidlohr Bueso i_mmap_unlock_write(mapping); 32527906d00cSAndrea Arcangeli if (!test_and_clear_bit(AS_MM_ALL_LOCKS, 32537906d00cSAndrea Arcangeli &mapping->flags)) 32547906d00cSAndrea Arcangeli BUG(); 32557906d00cSAndrea Arcangeli } 32567906d00cSAndrea Arcangeli } 32577906d00cSAndrea Arcangeli 32587906d00cSAndrea Arcangeli /* 32597906d00cSAndrea Arcangeli * The mmap_sem cannot be released by the caller until 32607906d00cSAndrea Arcangeli * mm_drop_all_locks() returns. 32617906d00cSAndrea Arcangeli */ 32627906d00cSAndrea Arcangeli void mm_drop_all_locks(struct mm_struct *mm) 32637906d00cSAndrea Arcangeli { 32647906d00cSAndrea Arcangeli struct vm_area_struct *vma; 32655beb4930SRik van Riel struct anon_vma_chain *avc; 32667906d00cSAndrea Arcangeli 32677906d00cSAndrea Arcangeli BUG_ON(down_read_trylock(&mm->mmap_sem)); 32687906d00cSAndrea Arcangeli BUG_ON(!mutex_is_locked(&mm_all_locks_mutex)); 32697906d00cSAndrea Arcangeli 32707906d00cSAndrea Arcangeli for (vma = mm->mmap; vma; vma = vma->vm_next) { 32717906d00cSAndrea Arcangeli if (vma->anon_vma) 32725beb4930SRik van Riel list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) 32735beb4930SRik van Riel vm_unlock_anon_vma(avc->anon_vma); 32747906d00cSAndrea Arcangeli if (vma->vm_file && vma->vm_file->f_mapping) 32757906d00cSAndrea Arcangeli vm_unlock_mapping(vma->vm_file->f_mapping); 32767906d00cSAndrea Arcangeli } 32777906d00cSAndrea Arcangeli 32787906d00cSAndrea Arcangeli mutex_unlock(&mm_all_locks_mutex); 32797906d00cSAndrea Arcangeli } 32808feae131SDavid Howells 32818feae131SDavid Howells /* 32828feae131SDavid Howells * initialise the VMA slab 32838feae131SDavid Howells */ 32848feae131SDavid Howells void __init mmap_init(void) 32858feae131SDavid Howells { 328600a62ce9SKOSAKI Motohiro int ret; 328700a62ce9SKOSAKI Motohiro 3288908c7f19STejun Heo ret = percpu_counter_init(&vm_committed_as, 0, GFP_KERNEL); 328900a62ce9SKOSAKI Motohiro VM_BUG_ON(ret); 32908feae131SDavid Howells } 3291c9b1d098SAndrew Shewmaker 3292c9b1d098SAndrew Shewmaker /* 3293c9b1d098SAndrew Shewmaker * Initialise sysctl_user_reserve_kbytes. 3294c9b1d098SAndrew Shewmaker * 3295c9b1d098SAndrew Shewmaker * This is intended to prevent a user from starting a single memory hogging 3296c9b1d098SAndrew Shewmaker * process, such that they cannot recover (kill the hog) in OVERCOMMIT_NEVER 3297c9b1d098SAndrew Shewmaker * mode. 3298c9b1d098SAndrew Shewmaker * 3299c9b1d098SAndrew Shewmaker * The default value is min(3% of free memory, 128MB) 3300c9b1d098SAndrew Shewmaker * 128MB is enough to recover with sshd/login, bash, and top/kill. 3301c9b1d098SAndrew Shewmaker */ 33021640879aSAndrew Shewmaker static int init_user_reserve(void) 3303c9b1d098SAndrew Shewmaker { 3304c9b1d098SAndrew Shewmaker unsigned long free_kbytes; 3305c9b1d098SAndrew Shewmaker 3306c9b1d098SAndrew Shewmaker free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10); 3307c9b1d098SAndrew Shewmaker 3308c9b1d098SAndrew Shewmaker sysctl_user_reserve_kbytes = min(free_kbytes / 32, 1UL << 17); 3309c9b1d098SAndrew Shewmaker return 0; 3310c9b1d098SAndrew Shewmaker } 3311a64fb3cdSPaul Gortmaker subsys_initcall(init_user_reserve); 33124eeab4f5SAndrew Shewmaker 33134eeab4f5SAndrew Shewmaker /* 33144eeab4f5SAndrew Shewmaker * Initialise sysctl_admin_reserve_kbytes. 33154eeab4f5SAndrew Shewmaker * 33164eeab4f5SAndrew Shewmaker * The purpose of sysctl_admin_reserve_kbytes is to allow the sys admin 33174eeab4f5SAndrew Shewmaker * to log in and kill a memory hogging process. 33184eeab4f5SAndrew Shewmaker * 33194eeab4f5SAndrew Shewmaker * Systems with more than 256MB will reserve 8MB, enough to recover 33204eeab4f5SAndrew Shewmaker * with sshd, bash, and top in OVERCOMMIT_GUESS. Smaller systems will 33214eeab4f5SAndrew Shewmaker * only reserve 3% of free pages by default. 33224eeab4f5SAndrew Shewmaker */ 33231640879aSAndrew Shewmaker static int init_admin_reserve(void) 33244eeab4f5SAndrew Shewmaker { 33254eeab4f5SAndrew Shewmaker unsigned long free_kbytes; 33264eeab4f5SAndrew Shewmaker 33274eeab4f5SAndrew Shewmaker free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10); 33284eeab4f5SAndrew Shewmaker 33294eeab4f5SAndrew Shewmaker sysctl_admin_reserve_kbytes = min(free_kbytes / 32, 1UL << 13); 33304eeab4f5SAndrew Shewmaker return 0; 33314eeab4f5SAndrew Shewmaker } 3332a64fb3cdSPaul Gortmaker subsys_initcall(init_admin_reserve); 33331640879aSAndrew Shewmaker 33341640879aSAndrew Shewmaker /* 33351640879aSAndrew Shewmaker * Reinititalise user and admin reserves if memory is added or removed. 33361640879aSAndrew Shewmaker * 33371640879aSAndrew Shewmaker * The default user reserve max is 128MB, and the default max for the 33381640879aSAndrew Shewmaker * admin reserve is 8MB. These are usually, but not always, enough to 33391640879aSAndrew Shewmaker * enable recovery from a memory hogging process using login/sshd, a shell, 33401640879aSAndrew Shewmaker * and tools like top. It may make sense to increase or even disable the 33411640879aSAndrew Shewmaker * reserve depending on the existence of swap or variations in the recovery 33421640879aSAndrew Shewmaker * tools. So, the admin may have changed them. 33431640879aSAndrew Shewmaker * 33441640879aSAndrew Shewmaker * If memory is added and the reserves have been eliminated or increased above 33451640879aSAndrew Shewmaker * the default max, then we'll trust the admin. 33461640879aSAndrew Shewmaker * 33471640879aSAndrew Shewmaker * If memory is removed and there isn't enough free memory, then we 33481640879aSAndrew Shewmaker * need to reset the reserves. 33491640879aSAndrew Shewmaker * 33501640879aSAndrew Shewmaker * Otherwise keep the reserve set by the admin. 33511640879aSAndrew Shewmaker */ 33521640879aSAndrew Shewmaker static int reserve_mem_notifier(struct notifier_block *nb, 33531640879aSAndrew Shewmaker unsigned long action, void *data) 33541640879aSAndrew Shewmaker { 33551640879aSAndrew Shewmaker unsigned long tmp, free_kbytes; 33561640879aSAndrew Shewmaker 33571640879aSAndrew Shewmaker switch (action) { 33581640879aSAndrew Shewmaker case MEM_ONLINE: 33591640879aSAndrew Shewmaker /* Default max is 128MB. Leave alone if modified by operator. */ 33601640879aSAndrew Shewmaker tmp = sysctl_user_reserve_kbytes; 33611640879aSAndrew Shewmaker if (0 < tmp && tmp < (1UL << 17)) 33621640879aSAndrew Shewmaker init_user_reserve(); 33631640879aSAndrew Shewmaker 33641640879aSAndrew Shewmaker /* Default max is 8MB. Leave alone if modified by operator. */ 33651640879aSAndrew Shewmaker tmp = sysctl_admin_reserve_kbytes; 33661640879aSAndrew Shewmaker if (0 < tmp && tmp < (1UL << 13)) 33671640879aSAndrew Shewmaker init_admin_reserve(); 33681640879aSAndrew Shewmaker 33691640879aSAndrew Shewmaker break; 33701640879aSAndrew Shewmaker case MEM_OFFLINE: 33711640879aSAndrew Shewmaker free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10); 33721640879aSAndrew Shewmaker 33731640879aSAndrew Shewmaker if (sysctl_user_reserve_kbytes > free_kbytes) { 33741640879aSAndrew Shewmaker init_user_reserve(); 33751640879aSAndrew Shewmaker pr_info("vm.user_reserve_kbytes reset to %lu\n", 33761640879aSAndrew Shewmaker sysctl_user_reserve_kbytes); 33771640879aSAndrew Shewmaker } 33781640879aSAndrew Shewmaker 33791640879aSAndrew Shewmaker if (sysctl_admin_reserve_kbytes > free_kbytes) { 33801640879aSAndrew Shewmaker init_admin_reserve(); 33811640879aSAndrew Shewmaker pr_info("vm.admin_reserve_kbytes reset to %lu\n", 33821640879aSAndrew Shewmaker sysctl_admin_reserve_kbytes); 33831640879aSAndrew Shewmaker } 33841640879aSAndrew Shewmaker break; 33851640879aSAndrew Shewmaker default: 33861640879aSAndrew Shewmaker break; 33871640879aSAndrew Shewmaker } 33881640879aSAndrew Shewmaker return NOTIFY_OK; 33891640879aSAndrew Shewmaker } 33901640879aSAndrew Shewmaker 33911640879aSAndrew Shewmaker static struct notifier_block reserve_mem_nb = { 33921640879aSAndrew Shewmaker .notifier_call = reserve_mem_notifier, 33931640879aSAndrew Shewmaker }; 33941640879aSAndrew Shewmaker 33951640879aSAndrew Shewmaker static int __meminit init_reserve_notifier(void) 33961640879aSAndrew Shewmaker { 33971640879aSAndrew Shewmaker if (register_hotmemory_notifier(&reserve_mem_nb)) 3398b1de0d13SMitchel Humpherys pr_err("Failed registering memory add/remove notifier for admin reserve\n"); 33991640879aSAndrew Shewmaker 34001640879aSAndrew Shewmaker return 0; 34011640879aSAndrew Shewmaker } 3402a64fb3cdSPaul Gortmaker subsys_initcall(init_reserve_notifier); 3403