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 9234679d7eSShaohua Li int sysctl_overcommit_memory __read_mostly = OVERCOMMIT_GUESS; /* heuristic overcommit */ 9334679d7eSShaohua Li int sysctl_overcommit_ratio __read_mostly = 50; /* default is 50% */ 9449f0ce5fSJerome Marchand unsigned long sysctl_overcommit_kbytes __read_mostly; 95c3d8c141SChristoph Lameter int sysctl_max_map_count __read_mostly = DEFAULT_MAX_MAP_COUNT; 96c9b1d098SAndrew Shewmaker unsigned long sysctl_user_reserve_kbytes __read_mostly = 1UL << 17; /* 128MB */ 974eeab4f5SAndrew Shewmaker unsigned long sysctl_admin_reserve_kbytes __read_mostly = 1UL << 13; /* 8MB */ 9834679d7eSShaohua Li /* 9934679d7eSShaohua Li * Make sure vm_committed_as in one cacheline and not cacheline shared with 10034679d7eSShaohua Li * other variables. It can be updated by several CPUs frequently. 10134679d7eSShaohua Li */ 10234679d7eSShaohua Li struct percpu_counter vm_committed_as ____cacheline_aligned_in_smp; 1031da177e4SLinus Torvalds 1041da177e4SLinus Torvalds /* 105997071bcSK. Y. Srinivasan * The global memory commitment made in the system can be a metric 106997071bcSK. Y. Srinivasan * that can be used to drive ballooning decisions when Linux is hosted 107997071bcSK. Y. Srinivasan * as a guest. On Hyper-V, the host implements a policy engine for dynamically 108997071bcSK. Y. Srinivasan * balancing memory across competing virtual machines that are hosted. 109997071bcSK. Y. Srinivasan * Several metrics drive this policy engine including the guest reported 110997071bcSK. Y. Srinivasan * memory commitment. 111997071bcSK. Y. Srinivasan */ 112997071bcSK. Y. Srinivasan unsigned long vm_memory_committed(void) 113997071bcSK. Y. Srinivasan { 114997071bcSK. Y. Srinivasan return percpu_counter_read_positive(&vm_committed_as); 115997071bcSK. Y. Srinivasan } 116997071bcSK. Y. Srinivasan EXPORT_SYMBOL_GPL(vm_memory_committed); 117997071bcSK. Y. Srinivasan 118997071bcSK. Y. Srinivasan /* 1191da177e4SLinus Torvalds * Check that a process has enough memory to allocate a new virtual 1201da177e4SLinus Torvalds * mapping. 0 means there is enough memory for the allocation to 1211da177e4SLinus Torvalds * succeed and -ENOMEM implies there is not. 1221da177e4SLinus Torvalds * 1231da177e4SLinus Torvalds * We currently support three overcommit policies, which are set via the 1241da177e4SLinus Torvalds * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting 1251da177e4SLinus Torvalds * 1261da177e4SLinus Torvalds * Strict overcommit modes added 2002 Feb 26 by Alan Cox. 1271da177e4SLinus Torvalds * Additional code 2002 Jul 20 by Robert Love. 1281da177e4SLinus Torvalds * 1291da177e4SLinus Torvalds * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise. 1301da177e4SLinus Torvalds * 1311da177e4SLinus Torvalds * Note this is a helper function intended to be used by LSMs which 1321da177e4SLinus Torvalds * wish to use this logic. 1331da177e4SLinus Torvalds */ 13434b4e4aaSAlan Cox int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin) 1351da177e4SLinus Torvalds { 136c9b1d098SAndrew Shewmaker unsigned long free, allowed, reserve; 1371da177e4SLinus Torvalds 13882f71ae4SKonstantin Khlebnikov VM_WARN_ONCE(percpu_counter_read(&vm_committed_as) < 13982f71ae4SKonstantin Khlebnikov -(s64)vm_committed_as_batch * num_online_cpus(), 14082f71ae4SKonstantin Khlebnikov "memory commitment underflow"); 14182f71ae4SKonstantin Khlebnikov 1421da177e4SLinus Torvalds vm_acct_memory(pages); 1431da177e4SLinus Torvalds 1441da177e4SLinus Torvalds /* 1451da177e4SLinus Torvalds * Sometimes we want to use more memory than we have 1461da177e4SLinus Torvalds */ 1471da177e4SLinus Torvalds if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS) 1481da177e4SLinus Torvalds return 0; 1491da177e4SLinus Torvalds 1501da177e4SLinus Torvalds if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) { 151c15bef30SDmitry Fink free = global_page_state(NR_FREE_PAGES); 152c15bef30SDmitry Fink free += global_page_state(NR_FILE_PAGES); 1531da177e4SLinus Torvalds 154c15bef30SDmitry Fink /* 155c15bef30SDmitry Fink * shmem pages shouldn't be counted as free in this 156c15bef30SDmitry Fink * case, they can't be purged, only swapped out, and 157c15bef30SDmitry Fink * that won't affect the overall amount of available 158c15bef30SDmitry Fink * memory in the system. 159c15bef30SDmitry Fink */ 160c15bef30SDmitry Fink free -= global_page_state(NR_SHMEM); 161c15bef30SDmitry Fink 162ec8acf20SShaohua Li free += get_nr_swap_pages(); 1631da177e4SLinus Torvalds 1641da177e4SLinus Torvalds /* 1651da177e4SLinus Torvalds * Any slabs which are created with the 1661da177e4SLinus Torvalds * SLAB_RECLAIM_ACCOUNT flag claim to have contents 1671da177e4SLinus Torvalds * which are reclaimable, under pressure. The dentry 1681da177e4SLinus Torvalds * cache and most inode caches should fall into this 1691da177e4SLinus Torvalds */ 170972d1a7bSChristoph Lameter free += global_page_state(NR_SLAB_RECLAIMABLE); 1711da177e4SLinus Torvalds 1721da177e4SLinus Torvalds /* 173c15bef30SDmitry Fink * Leave reserved pages. The pages are not for anonymous pages. 174c15bef30SDmitry Fink */ 175c15bef30SDmitry Fink if (free <= totalreserve_pages) 176c15bef30SDmitry Fink goto error; 177c15bef30SDmitry Fink else 178c15bef30SDmitry Fink free -= totalreserve_pages; 179c15bef30SDmitry Fink 180c15bef30SDmitry Fink /* 1814eeab4f5SAndrew Shewmaker * Reserve some for root 1821da177e4SLinus Torvalds */ 1831da177e4SLinus Torvalds if (!cap_sys_admin) 1844eeab4f5SAndrew Shewmaker free -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10); 1851da177e4SLinus Torvalds 1861da177e4SLinus Torvalds if (free > pages) 1871da177e4SLinus Torvalds return 0; 1881da177e4SLinus Torvalds 1896d9f7839SHideo AOKI goto error; 1901da177e4SLinus Torvalds } 1911da177e4SLinus Torvalds 19200619bccSJerome Marchand allowed = vm_commit_limit(); 1931da177e4SLinus Torvalds /* 1944eeab4f5SAndrew Shewmaker * Reserve some for root 1951da177e4SLinus Torvalds */ 1961da177e4SLinus Torvalds if (!cap_sys_admin) 1974eeab4f5SAndrew Shewmaker allowed -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10); 1981da177e4SLinus Torvalds 199c9b1d098SAndrew Shewmaker /* 200c9b1d098SAndrew Shewmaker * Don't let a single process grow so big a user can't recover 201c9b1d098SAndrew Shewmaker */ 202c9b1d098SAndrew Shewmaker if (mm) { 203c9b1d098SAndrew Shewmaker reserve = sysctl_user_reserve_kbytes >> (PAGE_SHIFT - 10); 204c9b1d098SAndrew Shewmaker allowed -= min(mm->total_vm / 32, reserve); 205c9b1d098SAndrew Shewmaker } 2061da177e4SLinus Torvalds 20700a62ce9SKOSAKI Motohiro if (percpu_counter_read_positive(&vm_committed_as) < allowed) 2081da177e4SLinus Torvalds return 0; 2096d9f7839SHideo AOKI error: 2101da177e4SLinus Torvalds vm_unacct_memory(pages); 2111da177e4SLinus Torvalds 2121da177e4SLinus Torvalds return -ENOMEM; 2131da177e4SLinus Torvalds } 2141da177e4SLinus Torvalds 2151da177e4SLinus Torvalds /* 2163d48ae45SPeter Zijlstra * Requires inode->i_mapping->i_mmap_mutex 2171da177e4SLinus Torvalds */ 2181da177e4SLinus Torvalds static void __remove_shared_vm_struct(struct vm_area_struct *vma, 2191da177e4SLinus Torvalds struct file *file, struct address_space *mapping) 2201da177e4SLinus Torvalds { 2211da177e4SLinus Torvalds if (vma->vm_flags & VM_DENYWRITE) 222496ad9aaSAl Viro atomic_inc(&file_inode(file)->i_writecount); 2231da177e4SLinus Torvalds if (vma->vm_flags & VM_SHARED) 2244bb5f5d9SDavid Herrmann mapping_unmap_writable(mapping); 2251da177e4SLinus Torvalds 2261da177e4SLinus Torvalds flush_dcache_mmap_lock(mapping); 2271da177e4SLinus Torvalds if (unlikely(vma->vm_flags & VM_NONLINEAR)) 2286b2dbba8SMichel Lespinasse list_del_init(&vma->shared.nonlinear); 2291da177e4SLinus Torvalds else 2306b2dbba8SMichel Lespinasse vma_interval_tree_remove(vma, &mapping->i_mmap); 2311da177e4SLinus Torvalds flush_dcache_mmap_unlock(mapping); 2321da177e4SLinus Torvalds } 2331da177e4SLinus Torvalds 2341da177e4SLinus Torvalds /* 2356b2dbba8SMichel Lespinasse * Unlink a file-based vm structure from its interval tree, to hide 236a8fb5618SHugh Dickins * vma from rmap and vmtruncate before freeing its page tables. 2371da177e4SLinus Torvalds */ 238a8fb5618SHugh Dickins void unlink_file_vma(struct vm_area_struct *vma) 2391da177e4SLinus Torvalds { 2401da177e4SLinus Torvalds struct file *file = vma->vm_file; 2411da177e4SLinus Torvalds 2421da177e4SLinus Torvalds if (file) { 2431da177e4SLinus Torvalds struct address_space *mapping = file->f_mapping; 2443d48ae45SPeter Zijlstra mutex_lock(&mapping->i_mmap_mutex); 2451da177e4SLinus Torvalds __remove_shared_vm_struct(vma, file, mapping); 2463d48ae45SPeter Zijlstra mutex_unlock(&mapping->i_mmap_mutex); 2471da177e4SLinus Torvalds } 248a8fb5618SHugh Dickins } 249a8fb5618SHugh Dickins 250a8fb5618SHugh Dickins /* 251a8fb5618SHugh Dickins * Close a vm structure and free it, returning the next. 252a8fb5618SHugh Dickins */ 253a8fb5618SHugh Dickins static struct vm_area_struct *remove_vma(struct vm_area_struct *vma) 254a8fb5618SHugh Dickins { 255a8fb5618SHugh Dickins struct vm_area_struct *next = vma->vm_next; 256a8fb5618SHugh Dickins 257a8fb5618SHugh Dickins might_sleep(); 2581da177e4SLinus Torvalds if (vma->vm_ops && vma->vm_ops->close) 2591da177e4SLinus Torvalds vma->vm_ops->close(vma); 260e9714acfSKonstantin Khlebnikov if (vma->vm_file) 261a8fb5618SHugh Dickins fput(vma->vm_file); 262f0be3d32SLee Schermerhorn mpol_put(vma_policy(vma)); 2631da177e4SLinus Torvalds kmem_cache_free(vm_area_cachep, vma); 264a8fb5618SHugh Dickins return next; 2651da177e4SLinus Torvalds } 2661da177e4SLinus Torvalds 267e4eb1ff6SLinus Torvalds static unsigned long do_brk(unsigned long addr, unsigned long len); 268e4eb1ff6SLinus Torvalds 2696a6160a7SHeiko Carstens SYSCALL_DEFINE1(brk, unsigned long, brk) 2701da177e4SLinus Torvalds { 2718764b338SCyrill Gorcunov unsigned long retval; 2721da177e4SLinus Torvalds unsigned long newbrk, oldbrk; 2731da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 274a5b4592cSJiri Kosina unsigned long min_brk; 275128557ffSMichel Lespinasse bool populate; 2761da177e4SLinus Torvalds 2771da177e4SLinus Torvalds down_write(&mm->mmap_sem); 2781da177e4SLinus Torvalds 279a5b4592cSJiri Kosina #ifdef CONFIG_COMPAT_BRK 2805520e894SJiri Kosina /* 2815520e894SJiri Kosina * CONFIG_COMPAT_BRK can still be overridden by setting 2825520e894SJiri Kosina * randomize_va_space to 2, which will still cause mm->start_brk 2835520e894SJiri Kosina * to be arbitrarily shifted 2845520e894SJiri Kosina */ 2854471a675SJiri Kosina if (current->brk_randomized) 2865520e894SJiri Kosina min_brk = mm->start_brk; 2875520e894SJiri Kosina else 2885520e894SJiri Kosina min_brk = mm->end_data; 289a5b4592cSJiri Kosina #else 290a5b4592cSJiri Kosina min_brk = mm->start_brk; 291a5b4592cSJiri Kosina #endif 292a5b4592cSJiri Kosina if (brk < min_brk) 2931da177e4SLinus Torvalds goto out; 2941e624196SRam Gupta 2951e624196SRam Gupta /* 2961e624196SRam Gupta * Check against rlimit here. If this check is done later after the test 2971e624196SRam Gupta * of oldbrk with newbrk then it can escape the test and let the data 2981e624196SRam Gupta * segment grow beyond its set limit the in case where the limit is 2991e624196SRam Gupta * not page aligned -Ram Gupta 3001e624196SRam Gupta */ 3018764b338SCyrill Gorcunov if (check_data_rlimit(rlimit(RLIMIT_DATA), brk, mm->start_brk, 3028764b338SCyrill Gorcunov mm->end_data, mm->start_data)) 3031e624196SRam Gupta goto out; 3041e624196SRam Gupta 3051da177e4SLinus Torvalds newbrk = PAGE_ALIGN(brk); 3061da177e4SLinus Torvalds oldbrk = PAGE_ALIGN(mm->brk); 3071da177e4SLinus Torvalds if (oldbrk == newbrk) 3081da177e4SLinus Torvalds goto set_brk; 3091da177e4SLinus Torvalds 3101da177e4SLinus Torvalds /* Always allow shrinking brk. */ 3111da177e4SLinus Torvalds if (brk <= mm->brk) { 3121da177e4SLinus Torvalds if (!do_munmap(mm, newbrk, oldbrk-newbrk)) 3131da177e4SLinus Torvalds goto set_brk; 3141da177e4SLinus Torvalds goto out; 3151da177e4SLinus Torvalds } 3161da177e4SLinus Torvalds 3171da177e4SLinus Torvalds /* Check against existing mmap mappings. */ 3181da177e4SLinus Torvalds if (find_vma_intersection(mm, oldbrk, newbrk+PAGE_SIZE)) 3191da177e4SLinus Torvalds goto out; 3201da177e4SLinus Torvalds 3211da177e4SLinus Torvalds /* Ok, looks good - let it rip. */ 3221da177e4SLinus Torvalds if (do_brk(oldbrk, newbrk-oldbrk) != oldbrk) 3231da177e4SLinus Torvalds goto out; 324128557ffSMichel Lespinasse 3251da177e4SLinus Torvalds set_brk: 3261da177e4SLinus Torvalds mm->brk = brk; 327128557ffSMichel Lespinasse populate = newbrk > oldbrk && (mm->def_flags & VM_LOCKED) != 0; 328128557ffSMichel Lespinasse up_write(&mm->mmap_sem); 329128557ffSMichel Lespinasse if (populate) 330128557ffSMichel Lespinasse mm_populate(oldbrk, newbrk - oldbrk); 331128557ffSMichel Lespinasse return brk; 332128557ffSMichel Lespinasse 3331da177e4SLinus Torvalds out: 3341da177e4SLinus Torvalds retval = mm->brk; 3351da177e4SLinus Torvalds up_write(&mm->mmap_sem); 3361da177e4SLinus Torvalds return retval; 3371da177e4SLinus Torvalds } 3381da177e4SLinus Torvalds 339d3737187SMichel Lespinasse static long vma_compute_subtree_gap(struct vm_area_struct *vma) 340d3737187SMichel Lespinasse { 341d3737187SMichel Lespinasse unsigned long max, subtree_gap; 342d3737187SMichel Lespinasse max = vma->vm_start; 343d3737187SMichel Lespinasse if (vma->vm_prev) 344d3737187SMichel Lespinasse max -= vma->vm_prev->vm_end; 345d3737187SMichel Lespinasse if (vma->vm_rb.rb_left) { 346d3737187SMichel Lespinasse subtree_gap = rb_entry(vma->vm_rb.rb_left, 347d3737187SMichel Lespinasse struct vm_area_struct, vm_rb)->rb_subtree_gap; 348d3737187SMichel Lespinasse if (subtree_gap > max) 349d3737187SMichel Lespinasse max = subtree_gap; 350d3737187SMichel Lespinasse } 351d3737187SMichel Lespinasse if (vma->vm_rb.rb_right) { 352d3737187SMichel Lespinasse subtree_gap = rb_entry(vma->vm_rb.rb_right, 353d3737187SMichel Lespinasse struct vm_area_struct, vm_rb)->rb_subtree_gap; 354d3737187SMichel Lespinasse if (subtree_gap > max) 355d3737187SMichel Lespinasse max = subtree_gap; 356d3737187SMichel Lespinasse } 357d3737187SMichel Lespinasse return max; 358d3737187SMichel Lespinasse } 359d3737187SMichel Lespinasse 360ed8ea815SMichel Lespinasse #ifdef CONFIG_DEBUG_VM_RB 3611da177e4SLinus Torvalds static int browse_rb(struct rb_root *root) 3621da177e4SLinus Torvalds { 3635a0768f6SMichel Lespinasse int i = 0, j, bug = 0; 3641da177e4SLinus Torvalds struct rb_node *nd, *pn = NULL; 3651da177e4SLinus Torvalds unsigned long prev = 0, pend = 0; 3661da177e4SLinus Torvalds 3671da177e4SLinus Torvalds for (nd = rb_first(root); nd; nd = rb_next(nd)) { 3681da177e4SLinus Torvalds struct vm_area_struct *vma; 3691da177e4SLinus Torvalds vma = rb_entry(nd, struct vm_area_struct, vm_rb); 3705a0768f6SMichel Lespinasse if (vma->vm_start < prev) { 371ff26f70fSAndrew Morton pr_emerg("vm_start %lx < prev %lx\n", 372ff26f70fSAndrew Morton vma->vm_start, prev); 3735a0768f6SMichel Lespinasse bug = 1; 3745a0768f6SMichel Lespinasse } 3755a0768f6SMichel Lespinasse if (vma->vm_start < pend) { 376ff26f70fSAndrew Morton pr_emerg("vm_start %lx < pend %lx\n", 377ff26f70fSAndrew Morton vma->vm_start, pend); 3785a0768f6SMichel Lespinasse bug = 1; 3795a0768f6SMichel Lespinasse } 3805a0768f6SMichel Lespinasse if (vma->vm_start > vma->vm_end) { 381ff26f70fSAndrew Morton pr_emerg("vm_start %lx > vm_end %lx\n", 382ff26f70fSAndrew Morton vma->vm_start, vma->vm_end); 3835a0768f6SMichel Lespinasse bug = 1; 3845a0768f6SMichel Lespinasse } 3855a0768f6SMichel Lespinasse if (vma->rb_subtree_gap != vma_compute_subtree_gap(vma)) { 3868542bdfcSSasha Levin pr_emerg("free gap %lx, correct %lx\n", 3875a0768f6SMichel Lespinasse vma->rb_subtree_gap, 3885a0768f6SMichel Lespinasse vma_compute_subtree_gap(vma)); 3895a0768f6SMichel Lespinasse bug = 1; 3905a0768f6SMichel Lespinasse } 3911da177e4SLinus Torvalds i++; 3921da177e4SLinus Torvalds pn = nd; 393d1af65d1SDavid Miller prev = vma->vm_start; 394d1af65d1SDavid Miller pend = vma->vm_end; 3951da177e4SLinus Torvalds } 3961da177e4SLinus Torvalds j = 0; 3975a0768f6SMichel Lespinasse for (nd = pn; nd; nd = rb_prev(nd)) 3981da177e4SLinus Torvalds j++; 3995a0768f6SMichel Lespinasse if (i != j) { 4008542bdfcSSasha Levin pr_emerg("backwards %d, forwards %d\n", j, i); 4015a0768f6SMichel Lespinasse bug = 1; 4021da177e4SLinus Torvalds } 4035a0768f6SMichel Lespinasse return bug ? -1 : i; 4041da177e4SLinus Torvalds } 4051da177e4SLinus Torvalds 406d3737187SMichel Lespinasse static void validate_mm_rb(struct rb_root *root, struct vm_area_struct *ignore) 407d3737187SMichel Lespinasse { 408d3737187SMichel Lespinasse struct rb_node *nd; 409d3737187SMichel Lespinasse 410d3737187SMichel Lespinasse for (nd = rb_first(root); nd; nd = rb_next(nd)) { 411d3737187SMichel Lespinasse struct vm_area_struct *vma; 412d3737187SMichel Lespinasse vma = rb_entry(nd, struct vm_area_struct, vm_rb); 413*96dad67fSSasha Levin VM_BUG_ON_VMA(vma != ignore && 414*96dad67fSSasha Levin vma->rb_subtree_gap != vma_compute_subtree_gap(vma), 415*96dad67fSSasha Levin vma); 416d3737187SMichel Lespinasse } 4171da177e4SLinus Torvalds } 4181da177e4SLinus Torvalds 419eafd4dc4SRashika Kheria static void validate_mm(struct mm_struct *mm) 4201da177e4SLinus Torvalds { 4211da177e4SLinus Torvalds int bug = 0; 4221da177e4SLinus Torvalds int i = 0; 4235a0768f6SMichel Lespinasse unsigned long highest_address = 0; 424ed8ea815SMichel Lespinasse struct vm_area_struct *vma = mm->mmap; 425ff26f70fSAndrew Morton 426ed8ea815SMichel Lespinasse while (vma) { 427ed8ea815SMichel Lespinasse struct anon_vma_chain *avc; 428ff26f70fSAndrew Morton 42963c3b902SMichel Lespinasse vma_lock_anon_vma(vma); 430ed8ea815SMichel Lespinasse list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) 431ed8ea815SMichel Lespinasse anon_vma_interval_tree_verify(avc); 43263c3b902SMichel Lespinasse vma_unlock_anon_vma(vma); 4335a0768f6SMichel Lespinasse highest_address = vma->vm_end; 434ed8ea815SMichel Lespinasse vma = vma->vm_next; 4351da177e4SLinus Torvalds i++; 4361da177e4SLinus Torvalds } 4375a0768f6SMichel Lespinasse if (i != mm->map_count) { 4388542bdfcSSasha Levin pr_emerg("map_count %d vm_next %d\n", mm->map_count, i); 4395a0768f6SMichel Lespinasse bug = 1; 4405a0768f6SMichel Lespinasse } 4415a0768f6SMichel Lespinasse if (highest_address != mm->highest_vm_end) { 4428542bdfcSSasha Levin pr_emerg("mm->highest_vm_end %lx, found %lx\n", 4435a0768f6SMichel Lespinasse mm->highest_vm_end, highest_address); 4445a0768f6SMichel Lespinasse bug = 1; 4455a0768f6SMichel Lespinasse } 4461da177e4SLinus Torvalds i = browse_rb(&mm->mm_rb); 4475a0768f6SMichel Lespinasse if (i != mm->map_count) { 448ff26f70fSAndrew Morton if (i != -1) 4498542bdfcSSasha Levin pr_emerg("map_count %d rb %d\n", mm->map_count, i); 4505a0768f6SMichel Lespinasse bug = 1; 4515a0768f6SMichel Lespinasse } 452*96dad67fSSasha Levin VM_BUG_ON_MM(bug, mm); 4531da177e4SLinus Torvalds } 4541da177e4SLinus Torvalds #else 455d3737187SMichel Lespinasse #define validate_mm_rb(root, ignore) do { } while (0) 4561da177e4SLinus Torvalds #define validate_mm(mm) do { } while (0) 4571da177e4SLinus Torvalds #endif 4581da177e4SLinus Torvalds 459d3737187SMichel Lespinasse RB_DECLARE_CALLBACKS(static, vma_gap_callbacks, struct vm_area_struct, vm_rb, 460d3737187SMichel Lespinasse unsigned long, rb_subtree_gap, vma_compute_subtree_gap) 461d3737187SMichel Lespinasse 462d3737187SMichel Lespinasse /* 463d3737187SMichel Lespinasse * Update augmented rbtree rb_subtree_gap values after vma->vm_start or 464d3737187SMichel Lespinasse * vma->vm_prev->vm_end values changed, without modifying the vma's position 465d3737187SMichel Lespinasse * in the rbtree. 466d3737187SMichel Lespinasse */ 467d3737187SMichel Lespinasse static void vma_gap_update(struct vm_area_struct *vma) 468d3737187SMichel Lespinasse { 469d3737187SMichel Lespinasse /* 470d3737187SMichel Lespinasse * As it turns out, RB_DECLARE_CALLBACKS() already created a callback 471d3737187SMichel Lespinasse * function that does exacltly what we want. 472d3737187SMichel Lespinasse */ 473d3737187SMichel Lespinasse vma_gap_callbacks_propagate(&vma->vm_rb, NULL); 474d3737187SMichel Lespinasse } 475d3737187SMichel Lespinasse 476d3737187SMichel Lespinasse static inline void vma_rb_insert(struct vm_area_struct *vma, 477d3737187SMichel Lespinasse struct rb_root *root) 478d3737187SMichel Lespinasse { 479d3737187SMichel Lespinasse /* All rb_subtree_gap values must be consistent prior to insertion */ 480d3737187SMichel Lespinasse validate_mm_rb(root, NULL); 481d3737187SMichel Lespinasse 482d3737187SMichel Lespinasse rb_insert_augmented(&vma->vm_rb, root, &vma_gap_callbacks); 483d3737187SMichel Lespinasse } 484d3737187SMichel Lespinasse 485d3737187SMichel Lespinasse static void vma_rb_erase(struct vm_area_struct *vma, struct rb_root *root) 486d3737187SMichel Lespinasse { 487d3737187SMichel Lespinasse /* 488d3737187SMichel Lespinasse * All rb_subtree_gap values must be consistent prior to erase, 489d3737187SMichel Lespinasse * with the possible exception of the vma being erased. 490d3737187SMichel Lespinasse */ 491d3737187SMichel Lespinasse validate_mm_rb(root, vma); 492d3737187SMichel Lespinasse 493d3737187SMichel Lespinasse /* 494d3737187SMichel Lespinasse * Note rb_erase_augmented is a fairly large inline function, 495d3737187SMichel Lespinasse * so make sure we instantiate it only once with our desired 496d3737187SMichel Lespinasse * augmented rbtree callbacks. 497d3737187SMichel Lespinasse */ 498d3737187SMichel Lespinasse rb_erase_augmented(&vma->vm_rb, root, &vma_gap_callbacks); 499d3737187SMichel Lespinasse } 500d3737187SMichel Lespinasse 501bf181b9fSMichel Lespinasse /* 502bf181b9fSMichel Lespinasse * vma has some anon_vma assigned, and is already inserted on that 503bf181b9fSMichel Lespinasse * anon_vma's interval trees. 504bf181b9fSMichel Lespinasse * 505bf181b9fSMichel Lespinasse * Before updating the vma's vm_start / vm_end / vm_pgoff fields, the 506bf181b9fSMichel Lespinasse * vma must be removed from the anon_vma's interval trees using 507bf181b9fSMichel Lespinasse * anon_vma_interval_tree_pre_update_vma(). 508bf181b9fSMichel Lespinasse * 509bf181b9fSMichel Lespinasse * After the update, the vma will be reinserted using 510bf181b9fSMichel Lespinasse * anon_vma_interval_tree_post_update_vma(). 511bf181b9fSMichel Lespinasse * 512bf181b9fSMichel Lespinasse * The entire update must be protected by exclusive mmap_sem and by 513bf181b9fSMichel Lespinasse * the root anon_vma's mutex. 514bf181b9fSMichel Lespinasse */ 515bf181b9fSMichel Lespinasse static inline void 516bf181b9fSMichel Lespinasse anon_vma_interval_tree_pre_update_vma(struct vm_area_struct *vma) 517bf181b9fSMichel Lespinasse { 518bf181b9fSMichel Lespinasse struct anon_vma_chain *avc; 519bf181b9fSMichel Lespinasse 520bf181b9fSMichel Lespinasse list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) 521bf181b9fSMichel Lespinasse anon_vma_interval_tree_remove(avc, &avc->anon_vma->rb_root); 522bf181b9fSMichel Lespinasse } 523bf181b9fSMichel Lespinasse 524bf181b9fSMichel Lespinasse static inline void 525bf181b9fSMichel Lespinasse anon_vma_interval_tree_post_update_vma(struct vm_area_struct *vma) 526bf181b9fSMichel Lespinasse { 527bf181b9fSMichel Lespinasse struct anon_vma_chain *avc; 528bf181b9fSMichel Lespinasse 529bf181b9fSMichel Lespinasse list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) 530bf181b9fSMichel Lespinasse anon_vma_interval_tree_insert(avc, &avc->anon_vma->rb_root); 531bf181b9fSMichel Lespinasse } 532bf181b9fSMichel Lespinasse 5336597d783SHugh Dickins static int find_vma_links(struct mm_struct *mm, unsigned long addr, 5346597d783SHugh Dickins unsigned long end, struct vm_area_struct **pprev, 5356597d783SHugh Dickins struct rb_node ***rb_link, struct rb_node **rb_parent) 5361da177e4SLinus Torvalds { 5371da177e4SLinus Torvalds struct rb_node **__rb_link, *__rb_parent, *rb_prev; 5381da177e4SLinus Torvalds 5391da177e4SLinus Torvalds __rb_link = &mm->mm_rb.rb_node; 5401da177e4SLinus Torvalds rb_prev = __rb_parent = NULL; 5411da177e4SLinus Torvalds 5421da177e4SLinus Torvalds while (*__rb_link) { 5431da177e4SLinus Torvalds struct vm_area_struct *vma_tmp; 5441da177e4SLinus Torvalds 5451da177e4SLinus Torvalds __rb_parent = *__rb_link; 5461da177e4SLinus Torvalds vma_tmp = rb_entry(__rb_parent, struct vm_area_struct, vm_rb); 5471da177e4SLinus Torvalds 5481da177e4SLinus Torvalds if (vma_tmp->vm_end > addr) { 5496597d783SHugh Dickins /* Fail if an existing vma overlaps the area */ 5506597d783SHugh Dickins if (vma_tmp->vm_start < end) 5516597d783SHugh Dickins return -ENOMEM; 5521da177e4SLinus Torvalds __rb_link = &__rb_parent->rb_left; 5531da177e4SLinus Torvalds } else { 5541da177e4SLinus Torvalds rb_prev = __rb_parent; 5551da177e4SLinus Torvalds __rb_link = &__rb_parent->rb_right; 5561da177e4SLinus Torvalds } 5571da177e4SLinus Torvalds } 5581da177e4SLinus Torvalds 5591da177e4SLinus Torvalds *pprev = NULL; 5601da177e4SLinus Torvalds if (rb_prev) 5611da177e4SLinus Torvalds *pprev = rb_entry(rb_prev, struct vm_area_struct, vm_rb); 5621da177e4SLinus Torvalds *rb_link = __rb_link; 5631da177e4SLinus Torvalds *rb_parent = __rb_parent; 5646597d783SHugh Dickins return 0; 5651da177e4SLinus Torvalds } 5661da177e4SLinus Torvalds 567e8420a8eSCyril Hrubis static unsigned long count_vma_pages_range(struct mm_struct *mm, 568e8420a8eSCyril Hrubis unsigned long addr, unsigned long end) 569e8420a8eSCyril Hrubis { 570e8420a8eSCyril Hrubis unsigned long nr_pages = 0; 571e8420a8eSCyril Hrubis struct vm_area_struct *vma; 572e8420a8eSCyril Hrubis 573e8420a8eSCyril Hrubis /* Find first overlaping mapping */ 574e8420a8eSCyril Hrubis vma = find_vma_intersection(mm, addr, end); 575e8420a8eSCyril Hrubis if (!vma) 576e8420a8eSCyril Hrubis return 0; 577e8420a8eSCyril Hrubis 578e8420a8eSCyril Hrubis nr_pages = (min(end, vma->vm_end) - 579e8420a8eSCyril Hrubis max(addr, vma->vm_start)) >> PAGE_SHIFT; 580e8420a8eSCyril Hrubis 581e8420a8eSCyril Hrubis /* Iterate over the rest of the overlaps */ 582e8420a8eSCyril Hrubis for (vma = vma->vm_next; vma; vma = vma->vm_next) { 583e8420a8eSCyril Hrubis unsigned long overlap_len; 584e8420a8eSCyril Hrubis 585e8420a8eSCyril Hrubis if (vma->vm_start > end) 586e8420a8eSCyril Hrubis break; 587e8420a8eSCyril Hrubis 588e8420a8eSCyril Hrubis overlap_len = min(end, vma->vm_end) - vma->vm_start; 589e8420a8eSCyril Hrubis nr_pages += overlap_len >> PAGE_SHIFT; 590e8420a8eSCyril Hrubis } 591e8420a8eSCyril Hrubis 592e8420a8eSCyril Hrubis return nr_pages; 593e8420a8eSCyril Hrubis } 594e8420a8eSCyril Hrubis 5951da177e4SLinus Torvalds void __vma_link_rb(struct mm_struct *mm, struct vm_area_struct *vma, 5961da177e4SLinus Torvalds struct rb_node **rb_link, struct rb_node *rb_parent) 5971da177e4SLinus Torvalds { 598d3737187SMichel Lespinasse /* Update tracking information for the gap following the new vma. */ 599d3737187SMichel Lespinasse if (vma->vm_next) 600d3737187SMichel Lespinasse vma_gap_update(vma->vm_next); 601d3737187SMichel Lespinasse else 602d3737187SMichel Lespinasse mm->highest_vm_end = vma->vm_end; 603d3737187SMichel Lespinasse 604d3737187SMichel Lespinasse /* 605d3737187SMichel Lespinasse * vma->vm_prev wasn't known when we followed the rbtree to find the 606d3737187SMichel Lespinasse * correct insertion point for that vma. As a result, we could not 607d3737187SMichel Lespinasse * update the vma vm_rb parents rb_subtree_gap values on the way down. 608d3737187SMichel Lespinasse * So, we first insert the vma with a zero rb_subtree_gap value 609d3737187SMichel Lespinasse * (to be consistent with what we did on the way down), and then 610d3737187SMichel Lespinasse * immediately update the gap to the correct value. Finally we 611d3737187SMichel Lespinasse * rebalance the rbtree after all augmented values have been set. 612d3737187SMichel Lespinasse */ 6131da177e4SLinus Torvalds rb_link_node(&vma->vm_rb, rb_parent, rb_link); 614d3737187SMichel Lespinasse vma->rb_subtree_gap = 0; 615d3737187SMichel Lespinasse vma_gap_update(vma); 616d3737187SMichel Lespinasse vma_rb_insert(vma, &mm->mm_rb); 6171da177e4SLinus Torvalds } 6181da177e4SLinus Torvalds 619cb8f488cSDenys Vlasenko static void __vma_link_file(struct vm_area_struct *vma) 6201da177e4SLinus Torvalds { 6211da177e4SLinus Torvalds struct file *file; 6221da177e4SLinus Torvalds 6231da177e4SLinus Torvalds file = vma->vm_file; 6241da177e4SLinus Torvalds if (file) { 6251da177e4SLinus Torvalds struct address_space *mapping = file->f_mapping; 6261da177e4SLinus Torvalds 6271da177e4SLinus Torvalds if (vma->vm_flags & VM_DENYWRITE) 628496ad9aaSAl Viro atomic_dec(&file_inode(file)->i_writecount); 6291da177e4SLinus Torvalds if (vma->vm_flags & VM_SHARED) 6304bb5f5d9SDavid Herrmann atomic_inc(&mapping->i_mmap_writable); 6311da177e4SLinus Torvalds 6321da177e4SLinus Torvalds flush_dcache_mmap_lock(mapping); 6331da177e4SLinus Torvalds if (unlikely(vma->vm_flags & VM_NONLINEAR)) 6341da177e4SLinus Torvalds vma_nonlinear_insert(vma, &mapping->i_mmap_nonlinear); 6351da177e4SLinus Torvalds else 6366b2dbba8SMichel Lespinasse vma_interval_tree_insert(vma, &mapping->i_mmap); 6371da177e4SLinus Torvalds flush_dcache_mmap_unlock(mapping); 6381da177e4SLinus Torvalds } 6391da177e4SLinus Torvalds } 6401da177e4SLinus Torvalds 6411da177e4SLinus Torvalds static void 6421da177e4SLinus Torvalds __vma_link(struct mm_struct *mm, struct vm_area_struct *vma, 6431da177e4SLinus Torvalds struct vm_area_struct *prev, struct rb_node **rb_link, 6441da177e4SLinus Torvalds struct rb_node *rb_parent) 6451da177e4SLinus Torvalds { 6461da177e4SLinus Torvalds __vma_link_list(mm, vma, prev, rb_parent); 6471da177e4SLinus Torvalds __vma_link_rb(mm, vma, rb_link, rb_parent); 6481da177e4SLinus Torvalds } 6491da177e4SLinus Torvalds 6501da177e4SLinus Torvalds static void vma_link(struct mm_struct *mm, struct vm_area_struct *vma, 6511da177e4SLinus Torvalds struct vm_area_struct *prev, struct rb_node **rb_link, 6521da177e4SLinus Torvalds struct rb_node *rb_parent) 6531da177e4SLinus Torvalds { 6541da177e4SLinus Torvalds struct address_space *mapping = NULL; 6551da177e4SLinus Torvalds 65664ac4940SHuang Shijie if (vma->vm_file) { 6571da177e4SLinus Torvalds mapping = vma->vm_file->f_mapping; 6583d48ae45SPeter Zijlstra mutex_lock(&mapping->i_mmap_mutex); 65964ac4940SHuang Shijie } 6601da177e4SLinus Torvalds 6611da177e4SLinus Torvalds __vma_link(mm, vma, prev, rb_link, rb_parent); 6621da177e4SLinus Torvalds __vma_link_file(vma); 6631da177e4SLinus Torvalds 6641da177e4SLinus Torvalds if (mapping) 6653d48ae45SPeter Zijlstra mutex_unlock(&mapping->i_mmap_mutex); 6661da177e4SLinus Torvalds 6671da177e4SLinus Torvalds mm->map_count++; 6681da177e4SLinus Torvalds validate_mm(mm); 6691da177e4SLinus Torvalds } 6701da177e4SLinus Torvalds 6711da177e4SLinus Torvalds /* 67288f6b4c3SKautuk Consul * Helper for vma_adjust() in the split_vma insert case: insert a vma into the 6736b2dbba8SMichel Lespinasse * mm's list and rbtree. It has already been inserted into the interval tree. 6741da177e4SLinus Torvalds */ 67548aae425SZhenwenXu static void __insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma) 6761da177e4SLinus Torvalds { 6776597d783SHugh Dickins struct vm_area_struct *prev; 6781da177e4SLinus Torvalds struct rb_node **rb_link, *rb_parent; 6791da177e4SLinus Torvalds 6806597d783SHugh Dickins if (find_vma_links(mm, vma->vm_start, vma->vm_end, 6816597d783SHugh Dickins &prev, &rb_link, &rb_parent)) 6826597d783SHugh Dickins BUG(); 6831da177e4SLinus Torvalds __vma_link(mm, vma, prev, rb_link, rb_parent); 6841da177e4SLinus Torvalds mm->map_count++; 6851da177e4SLinus Torvalds } 6861da177e4SLinus Torvalds 6871da177e4SLinus Torvalds static inline void 6881da177e4SLinus Torvalds __vma_unlink(struct mm_struct *mm, struct vm_area_struct *vma, 6891da177e4SLinus Torvalds struct vm_area_struct *prev) 6901da177e4SLinus Torvalds { 691d3737187SMichel Lespinasse struct vm_area_struct *next; 692297c5eeeSLinus Torvalds 693d3737187SMichel Lespinasse vma_rb_erase(vma, &mm->mm_rb); 694d3737187SMichel Lespinasse prev->vm_next = next = vma->vm_next; 695297c5eeeSLinus Torvalds if (next) 696297c5eeeSLinus Torvalds next->vm_prev = prev; 697615d6e87SDavidlohr Bueso 698615d6e87SDavidlohr Bueso /* Kill the cache */ 699615d6e87SDavidlohr Bueso vmacache_invalidate(mm); 7001da177e4SLinus Torvalds } 7011da177e4SLinus Torvalds 7021da177e4SLinus Torvalds /* 7031da177e4SLinus Torvalds * We cannot adjust vm_start, vm_end, vm_pgoff fields of a vma that 7041da177e4SLinus Torvalds * is already present in an i_mmap tree without adjusting the tree. 7051da177e4SLinus Torvalds * The following helper function should be used when such adjustments 7061da177e4SLinus Torvalds * are necessary. The "insert" vma (if any) is to be inserted 7071da177e4SLinus Torvalds * before we drop the necessary locks. 7081da177e4SLinus Torvalds */ 7095beb4930SRik van Riel int vma_adjust(struct vm_area_struct *vma, unsigned long start, 7101da177e4SLinus Torvalds unsigned long end, pgoff_t pgoff, struct vm_area_struct *insert) 7111da177e4SLinus Torvalds { 7121da177e4SLinus Torvalds struct mm_struct *mm = vma->vm_mm; 7131da177e4SLinus Torvalds struct vm_area_struct *next = vma->vm_next; 7141da177e4SLinus Torvalds struct vm_area_struct *importer = NULL; 7151da177e4SLinus Torvalds struct address_space *mapping = NULL; 7166b2dbba8SMichel Lespinasse struct rb_root *root = NULL; 717012f1800SRik van Riel struct anon_vma *anon_vma = NULL; 7181da177e4SLinus Torvalds struct file *file = vma->vm_file; 719d3737187SMichel Lespinasse bool start_changed = false, end_changed = false; 7201da177e4SLinus Torvalds long adjust_next = 0; 7211da177e4SLinus Torvalds int remove_next = 0; 7221da177e4SLinus Torvalds 7231da177e4SLinus Torvalds if (next && !insert) { 724287d97acSLinus Torvalds struct vm_area_struct *exporter = NULL; 725287d97acSLinus Torvalds 7261da177e4SLinus Torvalds if (end >= next->vm_end) { 7271da177e4SLinus Torvalds /* 7281da177e4SLinus Torvalds * vma expands, overlapping all the next, and 7291da177e4SLinus Torvalds * perhaps the one after too (mprotect case 6). 7301da177e4SLinus Torvalds */ 7311da177e4SLinus Torvalds again: remove_next = 1 + (end > next->vm_end); 7321da177e4SLinus Torvalds end = next->vm_end; 733287d97acSLinus Torvalds exporter = next; 7341da177e4SLinus Torvalds importer = vma; 7351da177e4SLinus Torvalds } else if (end > next->vm_start) { 7361da177e4SLinus Torvalds /* 7371da177e4SLinus Torvalds * vma expands, overlapping part of the next: 7381da177e4SLinus Torvalds * mprotect case 5 shifting the boundary up. 7391da177e4SLinus Torvalds */ 7401da177e4SLinus Torvalds adjust_next = (end - next->vm_start) >> PAGE_SHIFT; 741287d97acSLinus Torvalds exporter = next; 7421da177e4SLinus Torvalds importer = vma; 7431da177e4SLinus Torvalds } else if (end < vma->vm_end) { 7441da177e4SLinus Torvalds /* 7451da177e4SLinus Torvalds * vma shrinks, and !insert tells it's not 7461da177e4SLinus Torvalds * split_vma inserting another: so it must be 7471da177e4SLinus Torvalds * mprotect case 4 shifting the boundary down. 7481da177e4SLinus Torvalds */ 7491da177e4SLinus Torvalds adjust_next = -((vma->vm_end - end) >> PAGE_SHIFT); 750287d97acSLinus Torvalds exporter = vma; 7511da177e4SLinus Torvalds importer = next; 7521da177e4SLinus Torvalds } 7531da177e4SLinus Torvalds 7545beb4930SRik van Riel /* 7555beb4930SRik van Riel * Easily overlooked: when mprotect shifts the boundary, 7565beb4930SRik van Riel * make sure the expanding vma has anon_vma set if the 7575beb4930SRik van Riel * shrinking vma had, to cover any anon pages imported. 7585beb4930SRik van Riel */ 759287d97acSLinus Torvalds if (exporter && exporter->anon_vma && !importer->anon_vma) { 760287d97acSLinus Torvalds if (anon_vma_clone(importer, exporter)) 7615beb4930SRik van Riel return -ENOMEM; 762287d97acSLinus Torvalds importer->anon_vma = exporter->anon_vma; 7635beb4930SRik van Riel } 7645beb4930SRik van Riel } 7655beb4930SRik van Riel 7661da177e4SLinus Torvalds if (file) { 7671da177e4SLinus Torvalds mapping = file->f_mapping; 768682968e0SSrikar Dronamraju if (!(vma->vm_flags & VM_NONLINEAR)) { 7691da177e4SLinus Torvalds root = &mapping->i_mmap; 770cbc91f71SSrikar Dronamraju uprobe_munmap(vma, vma->vm_start, vma->vm_end); 771682968e0SSrikar Dronamraju 772682968e0SSrikar Dronamraju if (adjust_next) 773cbc91f71SSrikar Dronamraju uprobe_munmap(next, next->vm_start, 774cbc91f71SSrikar Dronamraju next->vm_end); 775682968e0SSrikar Dronamraju } 776682968e0SSrikar Dronamraju 7773d48ae45SPeter Zijlstra mutex_lock(&mapping->i_mmap_mutex); 7781da177e4SLinus Torvalds if (insert) { 7791da177e4SLinus Torvalds /* 7806b2dbba8SMichel Lespinasse * Put into interval tree now, so instantiated pages 7811da177e4SLinus Torvalds * are visible to arm/parisc __flush_dcache_page 7821da177e4SLinus Torvalds * throughout; but we cannot insert into address 7831da177e4SLinus Torvalds * space until vma start or end is updated. 7841da177e4SLinus Torvalds */ 7851da177e4SLinus Torvalds __vma_link_file(insert); 7861da177e4SLinus Torvalds } 7871da177e4SLinus Torvalds } 7881da177e4SLinus Torvalds 78994fcc585SAndrea Arcangeli vma_adjust_trans_huge(vma, start, end, adjust_next); 79094fcc585SAndrea Arcangeli 791012f1800SRik van Riel anon_vma = vma->anon_vma; 792bf181b9fSMichel Lespinasse if (!anon_vma && adjust_next) 793bf181b9fSMichel Lespinasse anon_vma = next->anon_vma; 794bf181b9fSMichel Lespinasse if (anon_vma) { 79581d1b09cSSasha Levin VM_BUG_ON_VMA(adjust_next && next->anon_vma && 79681d1b09cSSasha Levin anon_vma != next->anon_vma, next); 7974fc3f1d6SIngo Molnar anon_vma_lock_write(anon_vma); 798bf181b9fSMichel Lespinasse anon_vma_interval_tree_pre_update_vma(vma); 799bf181b9fSMichel Lespinasse if (adjust_next) 800bf181b9fSMichel Lespinasse anon_vma_interval_tree_pre_update_vma(next); 801bf181b9fSMichel Lespinasse } 802012f1800SRik van Riel 8031da177e4SLinus Torvalds if (root) { 8041da177e4SLinus Torvalds flush_dcache_mmap_lock(mapping); 8056b2dbba8SMichel Lespinasse vma_interval_tree_remove(vma, root); 8061da177e4SLinus Torvalds if (adjust_next) 8076b2dbba8SMichel Lespinasse vma_interval_tree_remove(next, root); 8081da177e4SLinus Torvalds } 8091da177e4SLinus Torvalds 810d3737187SMichel Lespinasse if (start != vma->vm_start) { 8111da177e4SLinus Torvalds vma->vm_start = start; 812d3737187SMichel Lespinasse start_changed = true; 813d3737187SMichel Lespinasse } 814d3737187SMichel Lespinasse if (end != vma->vm_end) { 8151da177e4SLinus Torvalds vma->vm_end = end; 816d3737187SMichel Lespinasse end_changed = true; 817d3737187SMichel Lespinasse } 8181da177e4SLinus Torvalds vma->vm_pgoff = pgoff; 8191da177e4SLinus Torvalds if (adjust_next) { 8201da177e4SLinus Torvalds next->vm_start += adjust_next << PAGE_SHIFT; 8211da177e4SLinus Torvalds next->vm_pgoff += adjust_next; 8221da177e4SLinus Torvalds } 8231da177e4SLinus Torvalds 8241da177e4SLinus Torvalds if (root) { 8251da177e4SLinus Torvalds if (adjust_next) 8266b2dbba8SMichel Lespinasse vma_interval_tree_insert(next, root); 8276b2dbba8SMichel Lespinasse vma_interval_tree_insert(vma, root); 8281da177e4SLinus Torvalds flush_dcache_mmap_unlock(mapping); 8291da177e4SLinus Torvalds } 8301da177e4SLinus Torvalds 8311da177e4SLinus Torvalds if (remove_next) { 8321da177e4SLinus Torvalds /* 8331da177e4SLinus Torvalds * vma_merge has merged next into vma, and needs 8341da177e4SLinus Torvalds * us to remove next before dropping the locks. 8351da177e4SLinus Torvalds */ 8361da177e4SLinus Torvalds __vma_unlink(mm, next, vma); 8371da177e4SLinus Torvalds if (file) 8381da177e4SLinus Torvalds __remove_shared_vm_struct(next, file, mapping); 8391da177e4SLinus Torvalds } else if (insert) { 8401da177e4SLinus Torvalds /* 8411da177e4SLinus Torvalds * split_vma has split insert from vma, and needs 8421da177e4SLinus Torvalds * us to insert it before dropping the locks 8431da177e4SLinus Torvalds * (it may either follow vma or precede it). 8441da177e4SLinus Torvalds */ 8451da177e4SLinus Torvalds __insert_vm_struct(mm, insert); 846d3737187SMichel Lespinasse } else { 847d3737187SMichel Lespinasse if (start_changed) 848d3737187SMichel Lespinasse vma_gap_update(vma); 849d3737187SMichel Lespinasse if (end_changed) { 850d3737187SMichel Lespinasse if (!next) 851d3737187SMichel Lespinasse mm->highest_vm_end = end; 852d3737187SMichel Lespinasse else if (!adjust_next) 853d3737187SMichel Lespinasse vma_gap_update(next); 854d3737187SMichel Lespinasse } 8551da177e4SLinus Torvalds } 8561da177e4SLinus Torvalds 857bf181b9fSMichel Lespinasse if (anon_vma) { 858bf181b9fSMichel Lespinasse anon_vma_interval_tree_post_update_vma(vma); 859bf181b9fSMichel Lespinasse if (adjust_next) 860bf181b9fSMichel Lespinasse anon_vma_interval_tree_post_update_vma(next); 86108b52706SKonstantin Khlebnikov anon_vma_unlock_write(anon_vma); 862bf181b9fSMichel Lespinasse } 8631da177e4SLinus Torvalds if (mapping) 8643d48ae45SPeter Zijlstra mutex_unlock(&mapping->i_mmap_mutex); 8651da177e4SLinus Torvalds 8662b144498SSrikar Dronamraju if (root) { 8677b2d81d4SIngo Molnar uprobe_mmap(vma); 8682b144498SSrikar Dronamraju 8692b144498SSrikar Dronamraju if (adjust_next) 8707b2d81d4SIngo Molnar uprobe_mmap(next); 8712b144498SSrikar Dronamraju } 8722b144498SSrikar Dronamraju 8731da177e4SLinus Torvalds if (remove_next) { 874925d1c40SMatt Helsley if (file) { 875cbc91f71SSrikar Dronamraju uprobe_munmap(next, next->vm_start, next->vm_end); 8761da177e4SLinus Torvalds fput(file); 877925d1c40SMatt Helsley } 8785beb4930SRik van Riel if (next->anon_vma) 8795beb4930SRik van Riel anon_vma_merge(vma, next); 8801da177e4SLinus Torvalds mm->map_count--; 8813964acd0SOleg Nesterov mpol_put(vma_policy(next)); 8821da177e4SLinus Torvalds kmem_cache_free(vm_area_cachep, next); 8831da177e4SLinus Torvalds /* 8841da177e4SLinus Torvalds * In mprotect's case 6 (see comments on vma_merge), 8851da177e4SLinus Torvalds * we must remove another next too. It would clutter 8861da177e4SLinus Torvalds * up the code too much to do both in one go. 8871da177e4SLinus Torvalds */ 8881da177e4SLinus Torvalds next = vma->vm_next; 889d3737187SMichel Lespinasse if (remove_next == 2) 8901da177e4SLinus Torvalds goto again; 891d3737187SMichel Lespinasse else if (next) 892d3737187SMichel Lespinasse vma_gap_update(next); 893d3737187SMichel Lespinasse else 894d3737187SMichel Lespinasse mm->highest_vm_end = end; 8951da177e4SLinus Torvalds } 8962b144498SSrikar Dronamraju if (insert && file) 8977b2d81d4SIngo Molnar uprobe_mmap(insert); 8981da177e4SLinus Torvalds 8991da177e4SLinus Torvalds validate_mm(mm); 9005beb4930SRik van Riel 9015beb4930SRik van Riel return 0; 9021da177e4SLinus Torvalds } 9031da177e4SLinus Torvalds 9041da177e4SLinus Torvalds /* 9051da177e4SLinus Torvalds * If the vma has a ->close operation then the driver probably needs to release 9061da177e4SLinus Torvalds * per-vma resources, so we don't attempt to merge those. 9071da177e4SLinus Torvalds */ 9081da177e4SLinus Torvalds static inline int is_mergeable_vma(struct vm_area_struct *vma, 9091da177e4SLinus Torvalds struct file *file, unsigned long vm_flags) 9101da177e4SLinus Torvalds { 91134228d47SCyrill Gorcunov /* 91234228d47SCyrill Gorcunov * VM_SOFTDIRTY should not prevent from VMA merging, if we 91334228d47SCyrill Gorcunov * match the flags but dirty bit -- the caller should mark 91434228d47SCyrill Gorcunov * merged VMA as dirty. If dirty bit won't be excluded from 91534228d47SCyrill Gorcunov * comparison, we increase pressue on the memory system forcing 91634228d47SCyrill Gorcunov * the kernel to generate new VMAs when old one could be 91734228d47SCyrill Gorcunov * extended instead. 91834228d47SCyrill Gorcunov */ 91934228d47SCyrill Gorcunov if ((vma->vm_flags ^ vm_flags) & ~VM_SOFTDIRTY) 9201da177e4SLinus Torvalds return 0; 9211da177e4SLinus Torvalds if (vma->vm_file != file) 9221da177e4SLinus Torvalds return 0; 9231da177e4SLinus Torvalds if (vma->vm_ops && vma->vm_ops->close) 9241da177e4SLinus Torvalds return 0; 9251da177e4SLinus Torvalds return 1; 9261da177e4SLinus Torvalds } 9271da177e4SLinus Torvalds 9281da177e4SLinus Torvalds static inline int is_mergeable_anon_vma(struct anon_vma *anon_vma1, 929965f55deSShaohua Li struct anon_vma *anon_vma2, 930965f55deSShaohua Li struct vm_area_struct *vma) 9311da177e4SLinus Torvalds { 932965f55deSShaohua Li /* 933965f55deSShaohua Li * The list_is_singular() test is to avoid merging VMA cloned from 934965f55deSShaohua Li * parents. This can improve scalability caused by anon_vma lock. 935965f55deSShaohua Li */ 936965f55deSShaohua Li if ((!anon_vma1 || !anon_vma2) && (!vma || 937965f55deSShaohua Li list_is_singular(&vma->anon_vma_chain))) 938965f55deSShaohua Li return 1; 939965f55deSShaohua Li return anon_vma1 == anon_vma2; 9401da177e4SLinus Torvalds } 9411da177e4SLinus Torvalds 9421da177e4SLinus Torvalds /* 9431da177e4SLinus Torvalds * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff) 9441da177e4SLinus Torvalds * in front of (at a lower virtual address and file offset than) the vma. 9451da177e4SLinus Torvalds * 9461da177e4SLinus Torvalds * We cannot merge two vmas if they have differently assigned (non-NULL) 9471da177e4SLinus Torvalds * anon_vmas, nor if same anon_vma is assigned but offsets incompatible. 9481da177e4SLinus Torvalds * 9491da177e4SLinus Torvalds * We don't check here for the merged mmap wrapping around the end of pagecache 9501da177e4SLinus Torvalds * indices (16TB on ia32) because do_mmap_pgoff() does not permit mmap's which 9511da177e4SLinus Torvalds * wrap, nor mmaps which cover the final page at index -1UL. 9521da177e4SLinus Torvalds */ 9531da177e4SLinus Torvalds static int 9541da177e4SLinus Torvalds can_vma_merge_before(struct vm_area_struct *vma, unsigned long vm_flags, 9551da177e4SLinus Torvalds struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff) 9561da177e4SLinus Torvalds { 9571da177e4SLinus Torvalds if (is_mergeable_vma(vma, file, vm_flags) && 958965f55deSShaohua Li is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) { 9591da177e4SLinus Torvalds if (vma->vm_pgoff == vm_pgoff) 9601da177e4SLinus Torvalds return 1; 9611da177e4SLinus Torvalds } 9621da177e4SLinus Torvalds return 0; 9631da177e4SLinus Torvalds } 9641da177e4SLinus Torvalds 9651da177e4SLinus Torvalds /* 9661da177e4SLinus Torvalds * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff) 9671da177e4SLinus Torvalds * beyond (at a higher virtual address and file offset than) the vma. 9681da177e4SLinus Torvalds * 9691da177e4SLinus Torvalds * We cannot merge two vmas if they have differently assigned (non-NULL) 9701da177e4SLinus Torvalds * anon_vmas, nor if same anon_vma is assigned but offsets incompatible. 9711da177e4SLinus Torvalds */ 9721da177e4SLinus Torvalds static int 9731da177e4SLinus Torvalds can_vma_merge_after(struct vm_area_struct *vma, unsigned long vm_flags, 9741da177e4SLinus Torvalds struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff) 9751da177e4SLinus Torvalds { 9761da177e4SLinus Torvalds if (is_mergeable_vma(vma, file, vm_flags) && 977965f55deSShaohua Li is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) { 9781da177e4SLinus Torvalds pgoff_t vm_pglen; 979d6e93217SLibin vm_pglen = vma_pages(vma); 9801da177e4SLinus Torvalds if (vma->vm_pgoff + vm_pglen == vm_pgoff) 9811da177e4SLinus Torvalds return 1; 9821da177e4SLinus Torvalds } 9831da177e4SLinus Torvalds return 0; 9841da177e4SLinus Torvalds } 9851da177e4SLinus Torvalds 9861da177e4SLinus Torvalds /* 9871da177e4SLinus Torvalds * Given a mapping request (addr,end,vm_flags,file,pgoff), figure out 9881da177e4SLinus Torvalds * whether that can be merged with its predecessor or its successor. 9891da177e4SLinus Torvalds * Or both (it neatly fills a hole). 9901da177e4SLinus Torvalds * 9911da177e4SLinus Torvalds * In most cases - when called for mmap, brk or mremap - [addr,end) is 9921da177e4SLinus Torvalds * certain not to be mapped by the time vma_merge is called; but when 9931da177e4SLinus Torvalds * called for mprotect, it is certain to be already mapped (either at 9941da177e4SLinus Torvalds * an offset within prev, or at the start of next), and the flags of 9951da177e4SLinus Torvalds * this area are about to be changed to vm_flags - and the no-change 9961da177e4SLinus Torvalds * case has already been eliminated. 9971da177e4SLinus Torvalds * 9981da177e4SLinus Torvalds * The following mprotect cases have to be considered, where AAAA is 9991da177e4SLinus Torvalds * the area passed down from mprotect_fixup, never extending beyond one 10001da177e4SLinus Torvalds * vma, PPPPPP is the prev vma specified, and NNNNNN the next vma after: 10011da177e4SLinus Torvalds * 10021da177e4SLinus Torvalds * AAAA AAAA AAAA AAAA 10031da177e4SLinus Torvalds * PPPPPPNNNNNN PPPPPPNNNNNN PPPPPPNNNNNN PPPPNNNNXXXX 10041da177e4SLinus Torvalds * cannot merge might become might become might become 10051da177e4SLinus Torvalds * PPNNNNNNNNNN PPPPPPPPPPNN PPPPPPPPPPPP 6 or 10061da177e4SLinus Torvalds * mmap, brk or case 4 below case 5 below PPPPPPPPXXXX 7 or 10071da177e4SLinus Torvalds * mremap move: PPPPNNNNNNNN 8 10081da177e4SLinus Torvalds * AAAA 10091da177e4SLinus Torvalds * PPPP NNNN PPPPPPPPPPPP PPPPPPPPNNNN PPPPNNNNNNNN 10101da177e4SLinus Torvalds * might become case 1 below case 2 below case 3 below 10111da177e4SLinus Torvalds * 10121da177e4SLinus Torvalds * Odd one out? Case 8, because it extends NNNN but needs flags of XXXX: 10131da177e4SLinus Torvalds * mprotect_fixup updates vm_flags & vm_page_prot on successful return. 10141da177e4SLinus Torvalds */ 10151da177e4SLinus Torvalds struct vm_area_struct *vma_merge(struct mm_struct *mm, 10161da177e4SLinus Torvalds struct vm_area_struct *prev, unsigned long addr, 10171da177e4SLinus Torvalds unsigned long end, unsigned long vm_flags, 10181da177e4SLinus Torvalds struct anon_vma *anon_vma, struct file *file, 10191da177e4SLinus Torvalds pgoff_t pgoff, struct mempolicy *policy) 10201da177e4SLinus Torvalds { 10211da177e4SLinus Torvalds pgoff_t pglen = (end - addr) >> PAGE_SHIFT; 10221da177e4SLinus Torvalds struct vm_area_struct *area, *next; 10235beb4930SRik van Riel int err; 10241da177e4SLinus Torvalds 10251da177e4SLinus Torvalds /* 10261da177e4SLinus Torvalds * We later require that vma->vm_flags == vm_flags, 10271da177e4SLinus Torvalds * so this tests vma->vm_flags & VM_SPECIAL, too. 10281da177e4SLinus Torvalds */ 10291da177e4SLinus Torvalds if (vm_flags & VM_SPECIAL) 10301da177e4SLinus Torvalds return NULL; 10311da177e4SLinus Torvalds 10321da177e4SLinus Torvalds if (prev) 10331da177e4SLinus Torvalds next = prev->vm_next; 10341da177e4SLinus Torvalds else 10351da177e4SLinus Torvalds next = mm->mmap; 10361da177e4SLinus Torvalds area = next; 10371da177e4SLinus Torvalds if (next && next->vm_end == end) /* cases 6, 7, 8 */ 10381da177e4SLinus Torvalds next = next->vm_next; 10391da177e4SLinus Torvalds 10401da177e4SLinus Torvalds /* 10411da177e4SLinus Torvalds * Can it merge with the predecessor? 10421da177e4SLinus Torvalds */ 10431da177e4SLinus Torvalds if (prev && prev->vm_end == addr && 10441da177e4SLinus Torvalds mpol_equal(vma_policy(prev), policy) && 10451da177e4SLinus Torvalds can_vma_merge_after(prev, vm_flags, 10461da177e4SLinus Torvalds anon_vma, file, pgoff)) { 10471da177e4SLinus Torvalds /* 10481da177e4SLinus Torvalds * OK, it can. Can we now merge in the successor as well? 10491da177e4SLinus Torvalds */ 10501da177e4SLinus Torvalds if (next && end == next->vm_start && 10511da177e4SLinus Torvalds mpol_equal(policy, vma_policy(next)) && 10521da177e4SLinus Torvalds can_vma_merge_before(next, vm_flags, 10531da177e4SLinus Torvalds anon_vma, file, pgoff+pglen) && 10541da177e4SLinus Torvalds is_mergeable_anon_vma(prev->anon_vma, 1055965f55deSShaohua Li next->anon_vma, NULL)) { 10561da177e4SLinus Torvalds /* cases 1, 6 */ 10575beb4930SRik van Riel err = vma_adjust(prev, prev->vm_start, 10581da177e4SLinus Torvalds next->vm_end, prev->vm_pgoff, NULL); 10591da177e4SLinus Torvalds } else /* cases 2, 5, 7 */ 10605beb4930SRik van Riel err = vma_adjust(prev, prev->vm_start, 10611da177e4SLinus Torvalds end, prev->vm_pgoff, NULL); 10625beb4930SRik van Riel if (err) 10635beb4930SRik van Riel return NULL; 1064b15d00b6SAndrea Arcangeli khugepaged_enter_vma_merge(prev); 10651da177e4SLinus Torvalds return prev; 10661da177e4SLinus Torvalds } 10671da177e4SLinus Torvalds 10681da177e4SLinus Torvalds /* 10691da177e4SLinus Torvalds * Can this new request be merged in front of next? 10701da177e4SLinus Torvalds */ 10711da177e4SLinus Torvalds if (next && end == next->vm_start && 10721da177e4SLinus Torvalds mpol_equal(policy, vma_policy(next)) && 10731da177e4SLinus Torvalds can_vma_merge_before(next, vm_flags, 10741da177e4SLinus Torvalds anon_vma, file, pgoff+pglen)) { 10751da177e4SLinus Torvalds if (prev && addr < prev->vm_end) /* case 4 */ 10765beb4930SRik van Riel err = vma_adjust(prev, prev->vm_start, 10771da177e4SLinus Torvalds addr, prev->vm_pgoff, NULL); 10781da177e4SLinus Torvalds else /* cases 3, 8 */ 10795beb4930SRik van Riel err = vma_adjust(area, addr, next->vm_end, 10801da177e4SLinus Torvalds next->vm_pgoff - pglen, NULL); 10815beb4930SRik van Riel if (err) 10825beb4930SRik van Riel return NULL; 1083b15d00b6SAndrea Arcangeli khugepaged_enter_vma_merge(area); 10841da177e4SLinus Torvalds return area; 10851da177e4SLinus Torvalds } 10861da177e4SLinus Torvalds 10871da177e4SLinus Torvalds return NULL; 10881da177e4SLinus Torvalds } 10891da177e4SLinus Torvalds 10901da177e4SLinus Torvalds /* 1091d0e9fe17SLinus Torvalds * Rough compatbility check to quickly see if it's even worth looking 1092d0e9fe17SLinus Torvalds * at sharing an anon_vma. 1093d0e9fe17SLinus Torvalds * 1094d0e9fe17SLinus Torvalds * They need to have the same vm_file, and the flags can only differ 1095d0e9fe17SLinus Torvalds * in things that mprotect may change. 1096d0e9fe17SLinus Torvalds * 1097d0e9fe17SLinus Torvalds * NOTE! The fact that we share an anon_vma doesn't _have_ to mean that 1098d0e9fe17SLinus Torvalds * we can merge the two vma's. For example, we refuse to merge a vma if 1099d0e9fe17SLinus Torvalds * there is a vm_ops->close() function, because that indicates that the 1100d0e9fe17SLinus Torvalds * driver is doing some kind of reference counting. But that doesn't 1101d0e9fe17SLinus Torvalds * really matter for the anon_vma sharing case. 1102d0e9fe17SLinus Torvalds */ 1103d0e9fe17SLinus Torvalds static int anon_vma_compatible(struct vm_area_struct *a, struct vm_area_struct *b) 1104d0e9fe17SLinus Torvalds { 1105d0e9fe17SLinus Torvalds return a->vm_end == b->vm_start && 1106d0e9fe17SLinus Torvalds mpol_equal(vma_policy(a), vma_policy(b)) && 1107d0e9fe17SLinus Torvalds a->vm_file == b->vm_file && 110834228d47SCyrill Gorcunov !((a->vm_flags ^ b->vm_flags) & ~(VM_READ|VM_WRITE|VM_EXEC|VM_SOFTDIRTY)) && 1109d0e9fe17SLinus Torvalds b->vm_pgoff == a->vm_pgoff + ((b->vm_start - a->vm_start) >> PAGE_SHIFT); 1110d0e9fe17SLinus Torvalds } 1111d0e9fe17SLinus Torvalds 1112d0e9fe17SLinus Torvalds /* 1113d0e9fe17SLinus Torvalds * Do some basic sanity checking to see if we can re-use the anon_vma 1114d0e9fe17SLinus Torvalds * from 'old'. The 'a'/'b' vma's are in VM order - one of them will be 1115d0e9fe17SLinus Torvalds * the same as 'old', the other will be the new one that is trying 1116d0e9fe17SLinus Torvalds * to share the anon_vma. 1117d0e9fe17SLinus Torvalds * 1118d0e9fe17SLinus Torvalds * NOTE! This runs with mm_sem held for reading, so it is possible that 1119d0e9fe17SLinus Torvalds * the anon_vma of 'old' is concurrently in the process of being set up 1120d0e9fe17SLinus Torvalds * by another page fault trying to merge _that_. But that's ok: if it 1121d0e9fe17SLinus Torvalds * is being set up, that automatically means that it will be a singleton 1122d0e9fe17SLinus Torvalds * acceptable for merging, so we can do all of this optimistically. But 1123d0e9fe17SLinus Torvalds * we do that ACCESS_ONCE() to make sure that we never re-load the pointer. 1124d0e9fe17SLinus Torvalds * 1125d0e9fe17SLinus Torvalds * IOW: that the "list_is_singular()" test on the anon_vma_chain only 1126d0e9fe17SLinus Torvalds * matters for the 'stable anon_vma' case (ie the thing we want to avoid 1127d0e9fe17SLinus Torvalds * is to return an anon_vma that is "complex" due to having gone through 1128d0e9fe17SLinus Torvalds * a fork). 1129d0e9fe17SLinus Torvalds * 1130d0e9fe17SLinus Torvalds * We also make sure that the two vma's are compatible (adjacent, 1131d0e9fe17SLinus Torvalds * and with the same memory policies). That's all stable, even with just 1132d0e9fe17SLinus Torvalds * a read lock on the mm_sem. 1133d0e9fe17SLinus Torvalds */ 1134d0e9fe17SLinus Torvalds static struct anon_vma *reusable_anon_vma(struct vm_area_struct *old, struct vm_area_struct *a, struct vm_area_struct *b) 1135d0e9fe17SLinus Torvalds { 1136d0e9fe17SLinus Torvalds if (anon_vma_compatible(a, b)) { 1137d0e9fe17SLinus Torvalds struct anon_vma *anon_vma = ACCESS_ONCE(old->anon_vma); 1138d0e9fe17SLinus Torvalds 1139d0e9fe17SLinus Torvalds if (anon_vma && list_is_singular(&old->anon_vma_chain)) 1140d0e9fe17SLinus Torvalds return anon_vma; 1141d0e9fe17SLinus Torvalds } 1142d0e9fe17SLinus Torvalds return NULL; 1143d0e9fe17SLinus Torvalds } 1144d0e9fe17SLinus Torvalds 1145d0e9fe17SLinus Torvalds /* 11461da177e4SLinus Torvalds * find_mergeable_anon_vma is used by anon_vma_prepare, to check 11471da177e4SLinus Torvalds * neighbouring vmas for a suitable anon_vma, before it goes off 11481da177e4SLinus Torvalds * to allocate a new anon_vma. It checks because a repetitive 11491da177e4SLinus Torvalds * sequence of mprotects and faults may otherwise lead to distinct 11501da177e4SLinus Torvalds * anon_vmas being allocated, preventing vma merge in subsequent 11511da177e4SLinus Torvalds * mprotect. 11521da177e4SLinus Torvalds */ 11531da177e4SLinus Torvalds struct anon_vma *find_mergeable_anon_vma(struct vm_area_struct *vma) 11541da177e4SLinus Torvalds { 1155d0e9fe17SLinus Torvalds struct anon_vma *anon_vma; 11561da177e4SLinus Torvalds struct vm_area_struct *near; 11571da177e4SLinus Torvalds 11581da177e4SLinus Torvalds near = vma->vm_next; 11591da177e4SLinus Torvalds if (!near) 11601da177e4SLinus Torvalds goto try_prev; 11611da177e4SLinus Torvalds 1162d0e9fe17SLinus Torvalds anon_vma = reusable_anon_vma(near, vma, near); 1163d0e9fe17SLinus Torvalds if (anon_vma) 1164d0e9fe17SLinus Torvalds return anon_vma; 11651da177e4SLinus Torvalds try_prev: 11669be34c9dSLinus Torvalds near = vma->vm_prev; 11671da177e4SLinus Torvalds if (!near) 11681da177e4SLinus Torvalds goto none; 11691da177e4SLinus Torvalds 1170d0e9fe17SLinus Torvalds anon_vma = reusable_anon_vma(near, near, vma); 1171d0e9fe17SLinus Torvalds if (anon_vma) 1172d0e9fe17SLinus Torvalds return anon_vma; 11731da177e4SLinus Torvalds none: 11741da177e4SLinus Torvalds /* 11751da177e4SLinus Torvalds * There's no absolute need to look only at touching neighbours: 11761da177e4SLinus Torvalds * we could search further afield for "compatible" anon_vmas. 11771da177e4SLinus Torvalds * But it would probably just be a waste of time searching, 11781da177e4SLinus Torvalds * or lead to too many vmas hanging off the same anon_vma. 11791da177e4SLinus Torvalds * We're trying to allow mprotect remerging later on, 11801da177e4SLinus Torvalds * not trying to minimize memory used for anon_vmas. 11811da177e4SLinus Torvalds */ 11821da177e4SLinus Torvalds return NULL; 11831da177e4SLinus Torvalds } 11841da177e4SLinus Torvalds 11851da177e4SLinus Torvalds #ifdef CONFIG_PROC_FS 1186ab50b8edSHugh Dickins void vm_stat_account(struct mm_struct *mm, unsigned long flags, 11871da177e4SLinus Torvalds struct file *file, long pages) 11881da177e4SLinus Torvalds { 11891da177e4SLinus Torvalds const unsigned long stack_flags 11901da177e4SLinus Torvalds = VM_STACK_FLAGS & (VM_GROWSUP|VM_GROWSDOWN); 11911da177e4SLinus Torvalds 119244de9d0cSHuang Shijie mm->total_vm += pages; 119344de9d0cSHuang Shijie 11941da177e4SLinus Torvalds if (file) { 11951da177e4SLinus Torvalds mm->shared_vm += pages; 11961da177e4SLinus Torvalds if ((flags & (VM_EXEC|VM_WRITE)) == VM_EXEC) 11971da177e4SLinus Torvalds mm->exec_vm += pages; 11981da177e4SLinus Torvalds } else if (flags & stack_flags) 11991da177e4SLinus Torvalds mm->stack_vm += pages; 12001da177e4SLinus Torvalds } 12011da177e4SLinus Torvalds #endif /* CONFIG_PROC_FS */ 12021da177e4SLinus Torvalds 12031da177e4SLinus Torvalds /* 120440401530SAl Viro * If a hint addr is less than mmap_min_addr change hint to be as 120540401530SAl Viro * low as possible but still greater than mmap_min_addr 120640401530SAl Viro */ 120740401530SAl Viro static inline unsigned long round_hint_to_min(unsigned long hint) 120840401530SAl Viro { 120940401530SAl Viro hint &= PAGE_MASK; 121040401530SAl Viro if (((void *)hint != NULL) && 121140401530SAl Viro (hint < mmap_min_addr)) 121240401530SAl Viro return PAGE_ALIGN(mmap_min_addr); 121340401530SAl Viro return hint; 121440401530SAl Viro } 121540401530SAl Viro 1216363ee17fSDavidlohr Bueso static inline int mlock_future_check(struct mm_struct *mm, 1217363ee17fSDavidlohr Bueso unsigned long flags, 1218363ee17fSDavidlohr Bueso unsigned long len) 1219363ee17fSDavidlohr Bueso { 1220363ee17fSDavidlohr Bueso unsigned long locked, lock_limit; 1221363ee17fSDavidlohr Bueso 1222363ee17fSDavidlohr Bueso /* mlock MCL_FUTURE? */ 1223363ee17fSDavidlohr Bueso if (flags & VM_LOCKED) { 1224363ee17fSDavidlohr Bueso locked = len >> PAGE_SHIFT; 1225363ee17fSDavidlohr Bueso locked += mm->locked_vm; 1226363ee17fSDavidlohr Bueso lock_limit = rlimit(RLIMIT_MEMLOCK); 1227363ee17fSDavidlohr Bueso lock_limit >>= PAGE_SHIFT; 1228363ee17fSDavidlohr Bueso if (locked > lock_limit && !capable(CAP_IPC_LOCK)) 1229363ee17fSDavidlohr Bueso return -EAGAIN; 1230363ee17fSDavidlohr Bueso } 1231363ee17fSDavidlohr Bueso return 0; 1232363ee17fSDavidlohr Bueso } 1233363ee17fSDavidlohr Bueso 123440401530SAl Viro /* 123527f5de79SJianjun Kong * The caller must hold down_write(¤t->mm->mmap_sem). 12361da177e4SLinus Torvalds */ 12371da177e4SLinus Torvalds 1238e3fc629dSAl Viro unsigned long do_mmap_pgoff(struct file *file, unsigned long addr, 12391da177e4SLinus Torvalds unsigned long len, unsigned long prot, 1240bebeb3d6SMichel Lespinasse unsigned long flags, unsigned long pgoff, 124141badc15SMichel Lespinasse unsigned long *populate) 12421da177e4SLinus Torvalds { 12431da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 1244ca16d140SKOSAKI Motohiro vm_flags_t vm_flags; 12451da177e4SLinus Torvalds 124641badc15SMichel Lespinasse *populate = 0; 1247bebeb3d6SMichel Lespinasse 12481da177e4SLinus Torvalds /* 12491da177e4SLinus Torvalds * Does the application expect PROT_READ to imply PROT_EXEC? 12501da177e4SLinus Torvalds * 12511da177e4SLinus Torvalds * (the exception is when the underlying filesystem is noexec 12521da177e4SLinus Torvalds * mounted, in which case we dont add PROT_EXEC.) 12531da177e4SLinus Torvalds */ 12541da177e4SLinus Torvalds if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC)) 1255d3ac7f89SJosef "Jeff" Sipek if (!(file && (file->f_path.mnt->mnt_flags & MNT_NOEXEC))) 12561da177e4SLinus Torvalds prot |= PROT_EXEC; 12571da177e4SLinus Torvalds 12581da177e4SLinus Torvalds if (!len) 12591da177e4SLinus Torvalds return -EINVAL; 12601da177e4SLinus Torvalds 12617cd94146SEric Paris if (!(flags & MAP_FIXED)) 12627cd94146SEric Paris addr = round_hint_to_min(addr); 12637cd94146SEric Paris 12641da177e4SLinus Torvalds /* Careful about overflows.. */ 12651da177e4SLinus Torvalds len = PAGE_ALIGN(len); 12669206de95SAl Viro if (!len) 12671da177e4SLinus Torvalds return -ENOMEM; 12681da177e4SLinus Torvalds 12691da177e4SLinus Torvalds /* offset overflow? */ 12701da177e4SLinus Torvalds if ((pgoff + (len >> PAGE_SHIFT)) < pgoff) 12711da177e4SLinus Torvalds return -EOVERFLOW; 12721da177e4SLinus Torvalds 12731da177e4SLinus Torvalds /* Too many mappings? */ 12741da177e4SLinus Torvalds if (mm->map_count > sysctl_max_map_count) 12751da177e4SLinus Torvalds return -ENOMEM; 12761da177e4SLinus Torvalds 12771da177e4SLinus Torvalds /* Obtain the address to map to. we verify (or select) it and ensure 12781da177e4SLinus Torvalds * that it represents a valid section of the address space. 12791da177e4SLinus Torvalds */ 12801da177e4SLinus Torvalds addr = get_unmapped_area(file, addr, len, pgoff, flags); 12811da177e4SLinus Torvalds if (addr & ~PAGE_MASK) 12821da177e4SLinus Torvalds return addr; 12831da177e4SLinus Torvalds 12841da177e4SLinus Torvalds /* Do simple checking here so the lower-level routines won't have 12851da177e4SLinus Torvalds * to. we assume access permissions have been handled by the open 12861da177e4SLinus Torvalds * of the memory object, so we don't do any here. 12871da177e4SLinus Torvalds */ 12881da177e4SLinus Torvalds vm_flags = calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags) | 12891da177e4SLinus Torvalds mm->def_flags | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC; 12901da177e4SLinus Torvalds 1291cdf7b341SHuang Shijie if (flags & MAP_LOCKED) 12921da177e4SLinus Torvalds if (!can_do_mlock()) 12931da177e4SLinus Torvalds return -EPERM; 1294ba470de4SRik van Riel 1295363ee17fSDavidlohr Bueso if (mlock_future_check(mm, vm_flags, len)) 12961da177e4SLinus Torvalds return -EAGAIN; 12971da177e4SLinus Torvalds 12981da177e4SLinus Torvalds if (file) { 1299077bf22bSOleg Nesterov struct inode *inode = file_inode(file); 1300077bf22bSOleg Nesterov 13011da177e4SLinus Torvalds switch (flags & MAP_TYPE) { 13021da177e4SLinus Torvalds case MAP_SHARED: 13031da177e4SLinus Torvalds if ((prot&PROT_WRITE) && !(file->f_mode&FMODE_WRITE)) 13041da177e4SLinus Torvalds return -EACCES; 13051da177e4SLinus Torvalds 13061da177e4SLinus Torvalds /* 13071da177e4SLinus Torvalds * Make sure we don't allow writing to an append-only 13081da177e4SLinus Torvalds * file.. 13091da177e4SLinus Torvalds */ 13101da177e4SLinus Torvalds if (IS_APPEND(inode) && (file->f_mode & FMODE_WRITE)) 13111da177e4SLinus Torvalds return -EACCES; 13121da177e4SLinus Torvalds 13131da177e4SLinus Torvalds /* 13141da177e4SLinus Torvalds * Make sure there are no mandatory locks on the file. 13151da177e4SLinus Torvalds */ 1316d7a06983SJeff Layton if (locks_verify_locked(file)) 13171da177e4SLinus Torvalds return -EAGAIN; 13181da177e4SLinus Torvalds 13191da177e4SLinus Torvalds vm_flags |= VM_SHARED | VM_MAYSHARE; 13201da177e4SLinus Torvalds if (!(file->f_mode & FMODE_WRITE)) 13211da177e4SLinus Torvalds vm_flags &= ~(VM_MAYWRITE | VM_SHARED); 13221da177e4SLinus Torvalds 13231da177e4SLinus Torvalds /* fall through */ 13241da177e4SLinus Torvalds case MAP_PRIVATE: 13251da177e4SLinus Torvalds if (!(file->f_mode & FMODE_READ)) 13261da177e4SLinus Torvalds return -EACCES; 1327d3ac7f89SJosef "Jeff" Sipek if (file->f_path.mnt->mnt_flags & MNT_NOEXEC) { 132880c5606cSLinus Torvalds if (vm_flags & VM_EXEC) 132980c5606cSLinus Torvalds return -EPERM; 133080c5606cSLinus Torvalds vm_flags &= ~VM_MAYEXEC; 133180c5606cSLinus Torvalds } 133280c5606cSLinus Torvalds 133372c2d531SAl Viro if (!file->f_op->mmap) 133480c5606cSLinus Torvalds return -ENODEV; 1335b2c56e4fSOleg Nesterov if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP)) 1336b2c56e4fSOleg Nesterov return -EINVAL; 13371da177e4SLinus Torvalds break; 13381da177e4SLinus Torvalds 13391da177e4SLinus Torvalds default: 13401da177e4SLinus Torvalds return -EINVAL; 13411da177e4SLinus Torvalds } 13421da177e4SLinus Torvalds } else { 13431da177e4SLinus Torvalds switch (flags & MAP_TYPE) { 13441da177e4SLinus Torvalds case MAP_SHARED: 1345b2c56e4fSOleg Nesterov if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP)) 1346b2c56e4fSOleg Nesterov return -EINVAL; 1347ce363942STejun Heo /* 1348ce363942STejun Heo * Ignore pgoff. 1349ce363942STejun Heo */ 1350ce363942STejun Heo pgoff = 0; 13511da177e4SLinus Torvalds vm_flags |= VM_SHARED | VM_MAYSHARE; 13521da177e4SLinus Torvalds break; 13531da177e4SLinus Torvalds case MAP_PRIVATE: 13541da177e4SLinus Torvalds /* 13551da177e4SLinus Torvalds * Set pgoff according to addr for anon_vma. 13561da177e4SLinus Torvalds */ 13571da177e4SLinus Torvalds pgoff = addr >> PAGE_SHIFT; 13581da177e4SLinus Torvalds break; 13591da177e4SLinus Torvalds default: 13601da177e4SLinus Torvalds return -EINVAL; 13611da177e4SLinus Torvalds } 13621da177e4SLinus Torvalds } 13631da177e4SLinus Torvalds 1364c22c0d63SMichel Lespinasse /* 1365c22c0d63SMichel Lespinasse * Set 'VM_NORESERVE' if we should not account for the 1366c22c0d63SMichel Lespinasse * memory use of this mapping. 1367c22c0d63SMichel Lespinasse */ 1368c22c0d63SMichel Lespinasse if (flags & MAP_NORESERVE) { 1369c22c0d63SMichel Lespinasse /* We honor MAP_NORESERVE if allowed to overcommit */ 1370c22c0d63SMichel Lespinasse if (sysctl_overcommit_memory != OVERCOMMIT_NEVER) 1371c22c0d63SMichel Lespinasse vm_flags |= VM_NORESERVE; 1372c22c0d63SMichel Lespinasse 1373c22c0d63SMichel Lespinasse /* hugetlb applies strict overcommit unless MAP_NORESERVE */ 1374c22c0d63SMichel Lespinasse if (file && is_file_hugepages(file)) 1375c22c0d63SMichel Lespinasse vm_flags |= VM_NORESERVE; 1376c22c0d63SMichel Lespinasse } 1377c22c0d63SMichel Lespinasse 1378c22c0d63SMichel Lespinasse addr = mmap_region(file, addr, len, vm_flags, pgoff); 137909a9f1d2SMichel Lespinasse if (!IS_ERR_VALUE(addr) && 138009a9f1d2SMichel Lespinasse ((vm_flags & VM_LOCKED) || 138109a9f1d2SMichel Lespinasse (flags & (MAP_POPULATE | MAP_NONBLOCK)) == MAP_POPULATE)) 138241badc15SMichel Lespinasse *populate = len; 1383bebeb3d6SMichel Lespinasse return addr; 13840165ab44SMiklos Szeredi } 13856be5ceb0SLinus Torvalds 138666f0dc48SHugh Dickins SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len, 138766f0dc48SHugh Dickins unsigned long, prot, unsigned long, flags, 138866f0dc48SHugh Dickins unsigned long, fd, unsigned long, pgoff) 138966f0dc48SHugh Dickins { 139066f0dc48SHugh Dickins struct file *file = NULL; 139166f0dc48SHugh Dickins unsigned long retval = -EBADF; 139266f0dc48SHugh Dickins 139366f0dc48SHugh Dickins if (!(flags & MAP_ANONYMOUS)) { 1394120a795dSAl Viro audit_mmap_fd(fd, flags); 139566f0dc48SHugh Dickins file = fget(fd); 139666f0dc48SHugh Dickins if (!file) 139766f0dc48SHugh Dickins goto out; 1398af73e4d9SNaoya Horiguchi if (is_file_hugepages(file)) 1399af73e4d9SNaoya Horiguchi len = ALIGN(len, huge_page_size(hstate_file(file))); 1400493af578SJörn Engel retval = -EINVAL; 1401493af578SJörn Engel if (unlikely(flags & MAP_HUGETLB && !is_file_hugepages(file))) 1402493af578SJörn Engel goto out_fput; 140366f0dc48SHugh Dickins } else if (flags & MAP_HUGETLB) { 140466f0dc48SHugh Dickins struct user_struct *user = NULL; 1405c103a4dcSAndrew Morton struct hstate *hs; 1406af73e4d9SNaoya Horiguchi 1407c103a4dcSAndrew Morton hs = hstate_sizelog((flags >> MAP_HUGE_SHIFT) & SHM_HUGE_MASK); 1408091d0d55SLi Zefan if (!hs) 1409091d0d55SLi Zefan return -EINVAL; 1410091d0d55SLi Zefan 1411091d0d55SLi Zefan len = ALIGN(len, huge_page_size(hs)); 141266f0dc48SHugh Dickins /* 141366f0dc48SHugh Dickins * VM_NORESERVE is used because the reservations will be 141466f0dc48SHugh Dickins * taken when vm_ops->mmap() is called 141566f0dc48SHugh Dickins * A dummy user value is used because we are not locking 141666f0dc48SHugh Dickins * memory so no accounting is necessary 141766f0dc48SHugh Dickins */ 1418af73e4d9SNaoya Horiguchi file = hugetlb_file_setup(HUGETLB_ANON_FILE, len, 141942d7395fSAndi Kleen VM_NORESERVE, 142042d7395fSAndi Kleen &user, HUGETLB_ANONHUGE_INODE, 142142d7395fSAndi Kleen (flags >> MAP_HUGE_SHIFT) & MAP_HUGE_MASK); 142266f0dc48SHugh Dickins if (IS_ERR(file)) 142366f0dc48SHugh Dickins return PTR_ERR(file); 142466f0dc48SHugh Dickins } 142566f0dc48SHugh Dickins 142666f0dc48SHugh Dickins flags &= ~(MAP_EXECUTABLE | MAP_DENYWRITE); 142766f0dc48SHugh Dickins 1428eb36c587SAl Viro retval = vm_mmap_pgoff(file, addr, len, prot, flags, pgoff); 1429493af578SJörn Engel out_fput: 143066f0dc48SHugh Dickins if (file) 143166f0dc48SHugh Dickins fput(file); 143266f0dc48SHugh Dickins out: 143366f0dc48SHugh Dickins return retval; 143466f0dc48SHugh Dickins } 143566f0dc48SHugh Dickins 1436a4679373SChristoph Hellwig #ifdef __ARCH_WANT_SYS_OLD_MMAP 1437a4679373SChristoph Hellwig struct mmap_arg_struct { 1438a4679373SChristoph Hellwig unsigned long addr; 1439a4679373SChristoph Hellwig unsigned long len; 1440a4679373SChristoph Hellwig unsigned long prot; 1441a4679373SChristoph Hellwig unsigned long flags; 1442a4679373SChristoph Hellwig unsigned long fd; 1443a4679373SChristoph Hellwig unsigned long offset; 1444a4679373SChristoph Hellwig }; 1445a4679373SChristoph Hellwig 1446a4679373SChristoph Hellwig SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg) 1447a4679373SChristoph Hellwig { 1448a4679373SChristoph Hellwig struct mmap_arg_struct a; 1449a4679373SChristoph Hellwig 1450a4679373SChristoph Hellwig if (copy_from_user(&a, arg, sizeof(a))) 1451a4679373SChristoph Hellwig return -EFAULT; 1452a4679373SChristoph Hellwig if (a.offset & ~PAGE_MASK) 1453a4679373SChristoph Hellwig return -EINVAL; 1454a4679373SChristoph Hellwig 1455a4679373SChristoph Hellwig return sys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd, 1456a4679373SChristoph Hellwig a.offset >> PAGE_SHIFT); 1457a4679373SChristoph Hellwig } 1458a4679373SChristoph Hellwig #endif /* __ARCH_WANT_SYS_OLD_MMAP */ 1459a4679373SChristoph Hellwig 14604e950f6fSAlexey Dobriyan /* 14614e950f6fSAlexey Dobriyan * Some shared mappigns will want the pages marked read-only 14624e950f6fSAlexey Dobriyan * to track write events. If so, we'll downgrade vm_page_prot 14634e950f6fSAlexey Dobriyan * to the private version (using protection_map[] without the 14644e950f6fSAlexey Dobriyan * VM_SHARED bit). 14654e950f6fSAlexey Dobriyan */ 14664e950f6fSAlexey Dobriyan int vma_wants_writenotify(struct vm_area_struct *vma) 14674e950f6fSAlexey Dobriyan { 1468ca16d140SKOSAKI Motohiro vm_flags_t vm_flags = vma->vm_flags; 14694e950f6fSAlexey Dobriyan 14704e950f6fSAlexey Dobriyan /* If it was private or non-writable, the write bit is already clear */ 14714e950f6fSAlexey Dobriyan if ((vm_flags & (VM_WRITE|VM_SHARED)) != ((VM_WRITE|VM_SHARED))) 14724e950f6fSAlexey Dobriyan return 0; 14734e950f6fSAlexey Dobriyan 14744e950f6fSAlexey Dobriyan /* The backer wishes to know when pages are first written to? */ 14754e950f6fSAlexey Dobriyan if (vma->vm_ops && vma->vm_ops->page_mkwrite) 14764e950f6fSAlexey Dobriyan return 1; 14774e950f6fSAlexey Dobriyan 14784e950f6fSAlexey Dobriyan /* The open routine did something to the protections already? */ 14794e950f6fSAlexey Dobriyan if (pgprot_val(vma->vm_page_prot) != 14803ed75eb8SColy Li pgprot_val(vm_get_page_prot(vm_flags))) 14814e950f6fSAlexey Dobriyan return 0; 14824e950f6fSAlexey Dobriyan 14834e950f6fSAlexey Dobriyan /* Specialty mapping? */ 14844b6e1e37SKonstantin Khlebnikov if (vm_flags & VM_PFNMAP) 14854e950f6fSAlexey Dobriyan return 0; 14864e950f6fSAlexey Dobriyan 14874e950f6fSAlexey Dobriyan /* Can the mapping track the dirty pages? */ 14884e950f6fSAlexey Dobriyan return vma->vm_file && vma->vm_file->f_mapping && 14894e950f6fSAlexey Dobriyan mapping_cap_account_dirty(vma->vm_file->f_mapping); 14904e950f6fSAlexey Dobriyan } 14914e950f6fSAlexey Dobriyan 1492fc8744adSLinus Torvalds /* 1493fc8744adSLinus Torvalds * We account for memory if it's a private writeable mapping, 14945a6fe125SMel Gorman * not hugepages and VM_NORESERVE wasn't set. 1495fc8744adSLinus Torvalds */ 1496ca16d140SKOSAKI Motohiro static inline int accountable_mapping(struct file *file, vm_flags_t vm_flags) 1497fc8744adSLinus Torvalds { 14985a6fe125SMel Gorman /* 14995a6fe125SMel Gorman * hugetlb has its own accounting separate from the core VM 15005a6fe125SMel Gorman * VM_HUGETLB may not be set yet so we cannot check for that flag. 15015a6fe125SMel Gorman */ 15025a6fe125SMel Gorman if (file && is_file_hugepages(file)) 15035a6fe125SMel Gorman return 0; 15045a6fe125SMel Gorman 1505fc8744adSLinus Torvalds return (vm_flags & (VM_NORESERVE | VM_SHARED | VM_WRITE)) == VM_WRITE; 1506fc8744adSLinus Torvalds } 1507fc8744adSLinus Torvalds 15080165ab44SMiklos Szeredi unsigned long mmap_region(struct file *file, unsigned long addr, 1509c22c0d63SMichel Lespinasse unsigned long len, vm_flags_t vm_flags, unsigned long pgoff) 15100165ab44SMiklos Szeredi { 15110165ab44SMiklos Szeredi struct mm_struct *mm = current->mm; 15120165ab44SMiklos Szeredi struct vm_area_struct *vma, *prev; 15130165ab44SMiklos Szeredi int error; 15140165ab44SMiklos Szeredi struct rb_node **rb_link, *rb_parent; 15150165ab44SMiklos Szeredi unsigned long charged = 0; 15160165ab44SMiklos Szeredi 1517e8420a8eSCyril Hrubis /* Check against address space limit. */ 1518e8420a8eSCyril Hrubis if (!may_expand_vm(mm, len >> PAGE_SHIFT)) { 1519e8420a8eSCyril Hrubis unsigned long nr_pages; 1520e8420a8eSCyril Hrubis 1521e8420a8eSCyril Hrubis /* 1522e8420a8eSCyril Hrubis * MAP_FIXED may remove pages of mappings that intersects with 1523e8420a8eSCyril Hrubis * requested mapping. Account for the pages it would unmap. 1524e8420a8eSCyril Hrubis */ 1525e8420a8eSCyril Hrubis if (!(vm_flags & MAP_FIXED)) 1526e8420a8eSCyril Hrubis return -ENOMEM; 1527e8420a8eSCyril Hrubis 1528e8420a8eSCyril Hrubis nr_pages = count_vma_pages_range(mm, addr, addr + len); 1529e8420a8eSCyril Hrubis 1530e8420a8eSCyril Hrubis if (!may_expand_vm(mm, (len >> PAGE_SHIFT) - nr_pages)) 1531e8420a8eSCyril Hrubis return -ENOMEM; 1532e8420a8eSCyril Hrubis } 1533e8420a8eSCyril Hrubis 15341da177e4SLinus Torvalds /* Clear old maps */ 15351da177e4SLinus Torvalds error = -ENOMEM; 15361da177e4SLinus Torvalds munmap_back: 15376597d783SHugh Dickins if (find_vma_links(mm, addr, addr + len, &prev, &rb_link, &rb_parent)) { 15381da177e4SLinus Torvalds if (do_munmap(mm, addr, len)) 15391da177e4SLinus Torvalds return -ENOMEM; 15401da177e4SLinus Torvalds goto munmap_back; 15411da177e4SLinus Torvalds } 15421da177e4SLinus Torvalds 1543fc8744adSLinus Torvalds /* 15441da177e4SLinus Torvalds * Private writable mapping: check memory availability 15451da177e4SLinus Torvalds */ 15465a6fe125SMel Gorman if (accountable_mapping(file, vm_flags)) { 15471da177e4SLinus Torvalds charged = len >> PAGE_SHIFT; 1548191c5424SAl Viro if (security_vm_enough_memory_mm(mm, charged)) 15491da177e4SLinus Torvalds return -ENOMEM; 15501da177e4SLinus Torvalds vm_flags |= VM_ACCOUNT; 15511da177e4SLinus Torvalds } 15521da177e4SLinus Torvalds 15531da177e4SLinus Torvalds /* 1554de33c8dbSLinus Torvalds * Can we just expand an old mapping? 15551da177e4SLinus Torvalds */ 1556de33c8dbSLinus Torvalds vma = vma_merge(mm, prev, addr, addr + len, vm_flags, NULL, file, pgoff, NULL); 1557ba470de4SRik van Riel if (vma) 15581da177e4SLinus Torvalds goto out; 15591da177e4SLinus Torvalds 15601da177e4SLinus Torvalds /* 15611da177e4SLinus Torvalds * Determine the object being mapped and call the appropriate 15621da177e4SLinus Torvalds * specific mapper. the address has already been validated, but 15631da177e4SLinus Torvalds * not unmapped, but the maps are removed from the list. 15641da177e4SLinus Torvalds */ 1565c5e3b83eSPekka Enberg vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL); 15661da177e4SLinus Torvalds if (!vma) { 15671da177e4SLinus Torvalds error = -ENOMEM; 15681da177e4SLinus Torvalds goto unacct_error; 15691da177e4SLinus Torvalds } 15701da177e4SLinus Torvalds 15711da177e4SLinus Torvalds vma->vm_mm = mm; 15721da177e4SLinus Torvalds vma->vm_start = addr; 15731da177e4SLinus Torvalds vma->vm_end = addr + len; 15741da177e4SLinus Torvalds vma->vm_flags = vm_flags; 15753ed75eb8SColy Li vma->vm_page_prot = vm_get_page_prot(vm_flags); 15761da177e4SLinus Torvalds vma->vm_pgoff = pgoff; 15775beb4930SRik van Riel INIT_LIST_HEAD(&vma->anon_vma_chain); 15781da177e4SLinus Torvalds 15791da177e4SLinus Torvalds if (file) { 15801da177e4SLinus Torvalds if (vm_flags & VM_DENYWRITE) { 15811da177e4SLinus Torvalds error = deny_write_access(file); 15821da177e4SLinus Torvalds if (error) 15831da177e4SLinus Torvalds goto free_vma; 15841da177e4SLinus Torvalds } 15854bb5f5d9SDavid Herrmann if (vm_flags & VM_SHARED) { 15864bb5f5d9SDavid Herrmann error = mapping_map_writable(file->f_mapping); 15874bb5f5d9SDavid Herrmann if (error) 15884bb5f5d9SDavid Herrmann goto allow_write_and_free_vma; 15894bb5f5d9SDavid Herrmann } 15904bb5f5d9SDavid Herrmann 15914bb5f5d9SDavid Herrmann /* ->mmap() can change vma->vm_file, but must guarantee that 15924bb5f5d9SDavid Herrmann * vma_link() below can deny write-access if VM_DENYWRITE is set 15934bb5f5d9SDavid Herrmann * and map writably if VM_SHARED is set. This usually means the 15944bb5f5d9SDavid Herrmann * new file must not have been exposed to user-space, yet. 15954bb5f5d9SDavid Herrmann */ 1596cb0942b8SAl Viro vma->vm_file = get_file(file); 15971da177e4SLinus Torvalds error = file->f_op->mmap(file, vma); 15981da177e4SLinus Torvalds if (error) 15991da177e4SLinus Torvalds goto unmap_and_free_vma; 16001da177e4SLinus Torvalds 16011da177e4SLinus Torvalds /* Can addr have changed?? 16021da177e4SLinus Torvalds * 16031da177e4SLinus Torvalds * Answer: Yes, several device drivers can do it in their 16041da177e4SLinus Torvalds * f_op->mmap method. -DaveM 16052897b4d2SJoonsoo Kim * Bug: If addr is changed, prev, rb_link, rb_parent should 16062897b4d2SJoonsoo Kim * be updated for vma_link() 16071da177e4SLinus Torvalds */ 16082897b4d2SJoonsoo Kim WARN_ON_ONCE(addr != vma->vm_start); 16092897b4d2SJoonsoo Kim 16101da177e4SLinus Torvalds addr = vma->vm_start; 16111da177e4SLinus Torvalds vm_flags = vma->vm_flags; 1612f8dbf0a7SHuang Shijie } else if (vm_flags & VM_SHARED) { 1613f8dbf0a7SHuang Shijie error = shmem_zero_setup(vma); 1614f8dbf0a7SHuang Shijie if (error) 1615f8dbf0a7SHuang Shijie goto free_vma; 1616f8dbf0a7SHuang Shijie } 16171da177e4SLinus Torvalds 1618c9d0bf24SMagnus Damm if (vma_wants_writenotify(vma)) { 1619c9d0bf24SMagnus Damm pgprot_t pprot = vma->vm_page_prot; 1620c9d0bf24SMagnus Damm 1621c9d0bf24SMagnus Damm /* Can vma->vm_page_prot have changed?? 1622c9d0bf24SMagnus Damm * 1623c9d0bf24SMagnus Damm * Answer: Yes, drivers may have changed it in their 1624c9d0bf24SMagnus Damm * f_op->mmap method. 1625c9d0bf24SMagnus Damm * 1626c9d0bf24SMagnus Damm * Ensures that vmas marked as uncached stay that way. 1627c9d0bf24SMagnus Damm */ 16281ddd439eSHugh Dickins vma->vm_page_prot = vm_get_page_prot(vm_flags & ~VM_SHARED); 1629c9d0bf24SMagnus Damm if (pgprot_val(pprot) == pgprot_val(pgprot_noncached(pprot))) 1630c9d0bf24SMagnus Damm vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot); 1631c9d0bf24SMagnus Damm } 1632d08b3851SPeter Zijlstra 16334d3d5b41SOleg Nesterov vma_link(mm, vma, prev, rb_link, rb_parent); 16344d3d5b41SOleg Nesterov /* Once vma denies write, undo our temporary denial count */ 16354bb5f5d9SDavid Herrmann if (file) { 16364bb5f5d9SDavid Herrmann if (vm_flags & VM_SHARED) 16374bb5f5d9SDavid Herrmann mapping_unmap_writable(file->f_mapping); 1638e8686772SOleg Nesterov if (vm_flags & VM_DENYWRITE) 1639e8686772SOleg Nesterov allow_write_access(file); 16404bb5f5d9SDavid Herrmann } 1641e8686772SOleg Nesterov file = vma->vm_file; 16421da177e4SLinus Torvalds out: 1643cdd6c482SIngo Molnar perf_event_mmap(vma); 16440a4a9391SPeter Zijlstra 1645ab50b8edSHugh Dickins vm_stat_account(mm, vm_flags, file, len >> PAGE_SHIFT); 16461da177e4SLinus Torvalds if (vm_flags & VM_LOCKED) { 1647bebeb3d6SMichel Lespinasse if (!((vm_flags & VM_SPECIAL) || is_vm_hugetlb_page(vma) || 1648bebeb3d6SMichel Lespinasse vma == get_gate_vma(current->mm))) 164906f9d8c2SKOSAKI Motohiro mm->locked_vm += (len >> PAGE_SHIFT); 1650bebeb3d6SMichel Lespinasse else 1651bebeb3d6SMichel Lespinasse vma->vm_flags &= ~VM_LOCKED; 1652bebeb3d6SMichel Lespinasse } 16532b144498SSrikar Dronamraju 1654c7a3a88cSOleg Nesterov if (file) 1655c7a3a88cSOleg Nesterov uprobe_mmap(vma); 16562b144498SSrikar Dronamraju 1657d9104d1cSCyrill Gorcunov /* 1658d9104d1cSCyrill Gorcunov * New (or expanded) vma always get soft dirty status. 1659d9104d1cSCyrill Gorcunov * Otherwise user-space soft-dirty page tracker won't 1660d9104d1cSCyrill Gorcunov * be able to distinguish situation when vma area unmapped, 1661d9104d1cSCyrill Gorcunov * then new mapped in-place (which must be aimed as 1662d9104d1cSCyrill Gorcunov * a completely new data area). 1663d9104d1cSCyrill Gorcunov */ 1664d9104d1cSCyrill Gorcunov vma->vm_flags |= VM_SOFTDIRTY; 1665d9104d1cSCyrill Gorcunov 16661da177e4SLinus Torvalds return addr; 16671da177e4SLinus Torvalds 16681da177e4SLinus Torvalds unmap_and_free_vma: 16691da177e4SLinus Torvalds vma->vm_file = NULL; 16701da177e4SLinus Torvalds fput(file); 16711da177e4SLinus Torvalds 16721da177e4SLinus Torvalds /* Undo any partial mapping done by a device driver. */ 1673e0da382cSHugh Dickins unmap_region(mm, vma, prev, vma->vm_start, vma->vm_end); 1674e0da382cSHugh Dickins charged = 0; 16754bb5f5d9SDavid Herrmann if (vm_flags & VM_SHARED) 16764bb5f5d9SDavid Herrmann mapping_unmap_writable(file->f_mapping); 16774bb5f5d9SDavid Herrmann allow_write_and_free_vma: 16784bb5f5d9SDavid Herrmann if (vm_flags & VM_DENYWRITE) 16794bb5f5d9SDavid Herrmann allow_write_access(file); 16801da177e4SLinus Torvalds free_vma: 16811da177e4SLinus Torvalds kmem_cache_free(vm_area_cachep, vma); 16821da177e4SLinus Torvalds unacct_error: 16831da177e4SLinus Torvalds if (charged) 16841da177e4SLinus Torvalds vm_unacct_memory(charged); 16851da177e4SLinus Torvalds return error; 16861da177e4SLinus Torvalds } 16871da177e4SLinus Torvalds 1688db4fbfb9SMichel Lespinasse unsigned long unmapped_area(struct vm_unmapped_area_info *info) 1689db4fbfb9SMichel Lespinasse { 1690db4fbfb9SMichel Lespinasse /* 1691db4fbfb9SMichel Lespinasse * We implement the search by looking for an rbtree node that 1692db4fbfb9SMichel Lespinasse * immediately follows a suitable gap. That is, 1693db4fbfb9SMichel Lespinasse * - gap_start = vma->vm_prev->vm_end <= info->high_limit - length; 1694db4fbfb9SMichel Lespinasse * - gap_end = vma->vm_start >= info->low_limit + length; 1695db4fbfb9SMichel Lespinasse * - gap_end - gap_start >= length 1696db4fbfb9SMichel Lespinasse */ 1697db4fbfb9SMichel Lespinasse 1698db4fbfb9SMichel Lespinasse struct mm_struct *mm = current->mm; 1699db4fbfb9SMichel Lespinasse struct vm_area_struct *vma; 1700db4fbfb9SMichel Lespinasse unsigned long length, low_limit, high_limit, gap_start, gap_end; 1701db4fbfb9SMichel Lespinasse 1702db4fbfb9SMichel Lespinasse /* Adjust search length to account for worst case alignment overhead */ 1703db4fbfb9SMichel Lespinasse length = info->length + info->align_mask; 1704db4fbfb9SMichel Lespinasse if (length < info->length) 1705db4fbfb9SMichel Lespinasse return -ENOMEM; 1706db4fbfb9SMichel Lespinasse 1707db4fbfb9SMichel Lespinasse /* Adjust search limits by the desired length */ 1708db4fbfb9SMichel Lespinasse if (info->high_limit < length) 1709db4fbfb9SMichel Lespinasse return -ENOMEM; 1710db4fbfb9SMichel Lespinasse high_limit = info->high_limit - length; 1711db4fbfb9SMichel Lespinasse 1712db4fbfb9SMichel Lespinasse if (info->low_limit > high_limit) 1713db4fbfb9SMichel Lespinasse return -ENOMEM; 1714db4fbfb9SMichel Lespinasse low_limit = info->low_limit + length; 1715db4fbfb9SMichel Lespinasse 1716db4fbfb9SMichel Lespinasse /* Check if rbtree root looks promising */ 1717db4fbfb9SMichel Lespinasse if (RB_EMPTY_ROOT(&mm->mm_rb)) 1718db4fbfb9SMichel Lespinasse goto check_highest; 1719db4fbfb9SMichel Lespinasse vma = rb_entry(mm->mm_rb.rb_node, struct vm_area_struct, vm_rb); 1720db4fbfb9SMichel Lespinasse if (vma->rb_subtree_gap < length) 1721db4fbfb9SMichel Lespinasse goto check_highest; 1722db4fbfb9SMichel Lespinasse 1723db4fbfb9SMichel Lespinasse while (true) { 1724db4fbfb9SMichel Lespinasse /* Visit left subtree if it looks promising */ 1725db4fbfb9SMichel Lespinasse gap_end = vma->vm_start; 1726db4fbfb9SMichel Lespinasse if (gap_end >= low_limit && vma->vm_rb.rb_left) { 1727db4fbfb9SMichel Lespinasse struct vm_area_struct *left = 1728db4fbfb9SMichel Lespinasse rb_entry(vma->vm_rb.rb_left, 1729db4fbfb9SMichel Lespinasse struct vm_area_struct, vm_rb); 1730db4fbfb9SMichel Lespinasse if (left->rb_subtree_gap >= length) { 1731db4fbfb9SMichel Lespinasse vma = left; 1732db4fbfb9SMichel Lespinasse continue; 1733db4fbfb9SMichel Lespinasse } 1734db4fbfb9SMichel Lespinasse } 1735db4fbfb9SMichel Lespinasse 1736db4fbfb9SMichel Lespinasse gap_start = vma->vm_prev ? vma->vm_prev->vm_end : 0; 1737db4fbfb9SMichel Lespinasse check_current: 1738db4fbfb9SMichel Lespinasse /* Check if current node has a suitable gap */ 1739db4fbfb9SMichel Lespinasse if (gap_start > high_limit) 1740db4fbfb9SMichel Lespinasse return -ENOMEM; 1741db4fbfb9SMichel Lespinasse if (gap_end >= low_limit && gap_end - gap_start >= length) 1742db4fbfb9SMichel Lespinasse goto found; 1743db4fbfb9SMichel Lespinasse 1744db4fbfb9SMichel Lespinasse /* Visit right subtree if it looks promising */ 1745db4fbfb9SMichel Lespinasse if (vma->vm_rb.rb_right) { 1746db4fbfb9SMichel Lespinasse struct vm_area_struct *right = 1747db4fbfb9SMichel Lespinasse rb_entry(vma->vm_rb.rb_right, 1748db4fbfb9SMichel Lespinasse struct vm_area_struct, vm_rb); 1749db4fbfb9SMichel Lespinasse if (right->rb_subtree_gap >= length) { 1750db4fbfb9SMichel Lespinasse vma = right; 1751db4fbfb9SMichel Lespinasse continue; 1752db4fbfb9SMichel Lespinasse } 1753db4fbfb9SMichel Lespinasse } 1754db4fbfb9SMichel Lespinasse 1755db4fbfb9SMichel Lespinasse /* Go back up the rbtree to find next candidate node */ 1756db4fbfb9SMichel Lespinasse while (true) { 1757db4fbfb9SMichel Lespinasse struct rb_node *prev = &vma->vm_rb; 1758db4fbfb9SMichel Lespinasse if (!rb_parent(prev)) 1759db4fbfb9SMichel Lespinasse goto check_highest; 1760db4fbfb9SMichel Lespinasse vma = rb_entry(rb_parent(prev), 1761db4fbfb9SMichel Lespinasse struct vm_area_struct, vm_rb); 1762db4fbfb9SMichel Lespinasse if (prev == vma->vm_rb.rb_left) { 1763db4fbfb9SMichel Lespinasse gap_start = vma->vm_prev->vm_end; 1764db4fbfb9SMichel Lespinasse gap_end = vma->vm_start; 1765db4fbfb9SMichel Lespinasse goto check_current; 1766db4fbfb9SMichel Lespinasse } 1767db4fbfb9SMichel Lespinasse } 1768db4fbfb9SMichel Lespinasse } 1769db4fbfb9SMichel Lespinasse 1770db4fbfb9SMichel Lespinasse check_highest: 1771db4fbfb9SMichel Lespinasse /* Check highest gap, which does not precede any rbtree node */ 1772db4fbfb9SMichel Lespinasse gap_start = mm->highest_vm_end; 1773db4fbfb9SMichel Lespinasse gap_end = ULONG_MAX; /* Only for VM_BUG_ON below */ 1774db4fbfb9SMichel Lespinasse if (gap_start > high_limit) 1775db4fbfb9SMichel Lespinasse return -ENOMEM; 1776db4fbfb9SMichel Lespinasse 1777db4fbfb9SMichel Lespinasse found: 1778db4fbfb9SMichel Lespinasse /* We found a suitable gap. Clip it with the original low_limit. */ 1779db4fbfb9SMichel Lespinasse if (gap_start < info->low_limit) 1780db4fbfb9SMichel Lespinasse gap_start = info->low_limit; 1781db4fbfb9SMichel Lespinasse 1782db4fbfb9SMichel Lespinasse /* Adjust gap address to the desired alignment */ 1783db4fbfb9SMichel Lespinasse gap_start += (info->align_offset - gap_start) & info->align_mask; 1784db4fbfb9SMichel Lespinasse 1785db4fbfb9SMichel Lespinasse VM_BUG_ON(gap_start + info->length > info->high_limit); 1786db4fbfb9SMichel Lespinasse VM_BUG_ON(gap_start + info->length > gap_end); 1787db4fbfb9SMichel Lespinasse return gap_start; 1788db4fbfb9SMichel Lespinasse } 1789db4fbfb9SMichel Lespinasse 1790db4fbfb9SMichel Lespinasse unsigned long unmapped_area_topdown(struct vm_unmapped_area_info *info) 1791db4fbfb9SMichel Lespinasse { 1792db4fbfb9SMichel Lespinasse struct mm_struct *mm = current->mm; 1793db4fbfb9SMichel Lespinasse struct vm_area_struct *vma; 1794db4fbfb9SMichel Lespinasse unsigned long length, low_limit, high_limit, gap_start, gap_end; 1795db4fbfb9SMichel Lespinasse 1796db4fbfb9SMichel Lespinasse /* Adjust search length to account for worst case alignment overhead */ 1797db4fbfb9SMichel Lespinasse length = info->length + info->align_mask; 1798db4fbfb9SMichel Lespinasse if (length < info->length) 1799db4fbfb9SMichel Lespinasse return -ENOMEM; 1800db4fbfb9SMichel Lespinasse 1801db4fbfb9SMichel Lespinasse /* 1802db4fbfb9SMichel Lespinasse * Adjust search limits by the desired length. 1803db4fbfb9SMichel Lespinasse * See implementation comment at top of unmapped_area(). 1804db4fbfb9SMichel Lespinasse */ 1805db4fbfb9SMichel Lespinasse gap_end = info->high_limit; 1806db4fbfb9SMichel Lespinasse if (gap_end < length) 1807db4fbfb9SMichel Lespinasse return -ENOMEM; 1808db4fbfb9SMichel Lespinasse high_limit = gap_end - length; 1809db4fbfb9SMichel Lespinasse 1810db4fbfb9SMichel Lespinasse if (info->low_limit > high_limit) 1811db4fbfb9SMichel Lespinasse return -ENOMEM; 1812db4fbfb9SMichel Lespinasse low_limit = info->low_limit + length; 1813db4fbfb9SMichel Lespinasse 1814db4fbfb9SMichel Lespinasse /* Check highest gap, which does not precede any rbtree node */ 1815db4fbfb9SMichel Lespinasse gap_start = mm->highest_vm_end; 1816db4fbfb9SMichel Lespinasse if (gap_start <= high_limit) 1817db4fbfb9SMichel Lespinasse goto found_highest; 1818db4fbfb9SMichel Lespinasse 1819db4fbfb9SMichel Lespinasse /* Check if rbtree root looks promising */ 1820db4fbfb9SMichel Lespinasse if (RB_EMPTY_ROOT(&mm->mm_rb)) 1821db4fbfb9SMichel Lespinasse return -ENOMEM; 1822db4fbfb9SMichel Lespinasse vma = rb_entry(mm->mm_rb.rb_node, struct vm_area_struct, vm_rb); 1823db4fbfb9SMichel Lespinasse if (vma->rb_subtree_gap < length) 1824db4fbfb9SMichel Lespinasse return -ENOMEM; 1825db4fbfb9SMichel Lespinasse 1826db4fbfb9SMichel Lespinasse while (true) { 1827db4fbfb9SMichel Lespinasse /* Visit right subtree if it looks promising */ 1828db4fbfb9SMichel Lespinasse gap_start = vma->vm_prev ? vma->vm_prev->vm_end : 0; 1829db4fbfb9SMichel Lespinasse if (gap_start <= high_limit && vma->vm_rb.rb_right) { 1830db4fbfb9SMichel Lespinasse struct vm_area_struct *right = 1831db4fbfb9SMichel Lespinasse rb_entry(vma->vm_rb.rb_right, 1832db4fbfb9SMichel Lespinasse struct vm_area_struct, vm_rb); 1833db4fbfb9SMichel Lespinasse if (right->rb_subtree_gap >= length) { 1834db4fbfb9SMichel Lespinasse vma = right; 1835db4fbfb9SMichel Lespinasse continue; 1836db4fbfb9SMichel Lespinasse } 1837db4fbfb9SMichel Lespinasse } 1838db4fbfb9SMichel Lespinasse 1839db4fbfb9SMichel Lespinasse check_current: 1840db4fbfb9SMichel Lespinasse /* Check if current node has a suitable gap */ 1841db4fbfb9SMichel Lespinasse gap_end = vma->vm_start; 1842db4fbfb9SMichel Lespinasse if (gap_end < low_limit) 1843db4fbfb9SMichel Lespinasse return -ENOMEM; 1844db4fbfb9SMichel Lespinasse if (gap_start <= high_limit && gap_end - gap_start >= length) 1845db4fbfb9SMichel Lespinasse goto found; 1846db4fbfb9SMichel Lespinasse 1847db4fbfb9SMichel Lespinasse /* Visit left subtree if it looks promising */ 1848db4fbfb9SMichel Lespinasse if (vma->vm_rb.rb_left) { 1849db4fbfb9SMichel Lespinasse struct vm_area_struct *left = 1850db4fbfb9SMichel Lespinasse rb_entry(vma->vm_rb.rb_left, 1851db4fbfb9SMichel Lespinasse struct vm_area_struct, vm_rb); 1852db4fbfb9SMichel Lespinasse if (left->rb_subtree_gap >= length) { 1853db4fbfb9SMichel Lespinasse vma = left; 1854db4fbfb9SMichel Lespinasse continue; 1855db4fbfb9SMichel Lespinasse } 1856db4fbfb9SMichel Lespinasse } 1857db4fbfb9SMichel Lespinasse 1858db4fbfb9SMichel Lespinasse /* Go back up the rbtree to find next candidate node */ 1859db4fbfb9SMichel Lespinasse while (true) { 1860db4fbfb9SMichel Lespinasse struct rb_node *prev = &vma->vm_rb; 1861db4fbfb9SMichel Lespinasse if (!rb_parent(prev)) 1862db4fbfb9SMichel Lespinasse return -ENOMEM; 1863db4fbfb9SMichel Lespinasse vma = rb_entry(rb_parent(prev), 1864db4fbfb9SMichel Lespinasse struct vm_area_struct, vm_rb); 1865db4fbfb9SMichel Lespinasse if (prev == vma->vm_rb.rb_right) { 1866db4fbfb9SMichel Lespinasse gap_start = vma->vm_prev ? 1867db4fbfb9SMichel Lespinasse vma->vm_prev->vm_end : 0; 1868db4fbfb9SMichel Lespinasse goto check_current; 1869db4fbfb9SMichel Lespinasse } 1870db4fbfb9SMichel Lespinasse } 1871db4fbfb9SMichel Lespinasse } 1872db4fbfb9SMichel Lespinasse 1873db4fbfb9SMichel Lespinasse found: 1874db4fbfb9SMichel Lespinasse /* We found a suitable gap. Clip it with the original high_limit. */ 1875db4fbfb9SMichel Lespinasse if (gap_end > info->high_limit) 1876db4fbfb9SMichel Lespinasse gap_end = info->high_limit; 1877db4fbfb9SMichel Lespinasse 1878db4fbfb9SMichel Lespinasse found_highest: 1879db4fbfb9SMichel Lespinasse /* Compute highest gap address at the desired alignment */ 1880db4fbfb9SMichel Lespinasse gap_end -= info->length; 1881db4fbfb9SMichel Lespinasse gap_end -= (gap_end - info->align_offset) & info->align_mask; 1882db4fbfb9SMichel Lespinasse 1883db4fbfb9SMichel Lespinasse VM_BUG_ON(gap_end < info->low_limit); 1884db4fbfb9SMichel Lespinasse VM_BUG_ON(gap_end < gap_start); 1885db4fbfb9SMichel Lespinasse return gap_end; 1886db4fbfb9SMichel Lespinasse } 1887db4fbfb9SMichel Lespinasse 18881da177e4SLinus Torvalds /* Get an address range which is currently unmapped. 18891da177e4SLinus Torvalds * For shmat() with addr=0. 18901da177e4SLinus Torvalds * 18911da177e4SLinus Torvalds * Ugly calling convention alert: 18921da177e4SLinus Torvalds * Return value with the low bits set means error value, 18931da177e4SLinus Torvalds * ie 18941da177e4SLinus Torvalds * if (ret & ~PAGE_MASK) 18951da177e4SLinus Torvalds * error = ret; 18961da177e4SLinus Torvalds * 18971da177e4SLinus Torvalds * This function "knows" that -ENOMEM has the bits set. 18981da177e4SLinus Torvalds */ 18991da177e4SLinus Torvalds #ifndef HAVE_ARCH_UNMAPPED_AREA 19001da177e4SLinus Torvalds unsigned long 19011da177e4SLinus Torvalds arch_get_unmapped_area(struct file *filp, unsigned long addr, 19021da177e4SLinus Torvalds unsigned long len, unsigned long pgoff, unsigned long flags) 19031da177e4SLinus Torvalds { 19041da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 19051da177e4SLinus Torvalds struct vm_area_struct *vma; 1906db4fbfb9SMichel Lespinasse struct vm_unmapped_area_info info; 19071da177e4SLinus Torvalds 19082afc745fSAkira Takeuchi if (len > TASK_SIZE - mmap_min_addr) 19091da177e4SLinus Torvalds return -ENOMEM; 19101da177e4SLinus Torvalds 191106abdfb4SBenjamin Herrenschmidt if (flags & MAP_FIXED) 191206abdfb4SBenjamin Herrenschmidt return addr; 191306abdfb4SBenjamin Herrenschmidt 19141da177e4SLinus Torvalds if (addr) { 19151da177e4SLinus Torvalds addr = PAGE_ALIGN(addr); 19161da177e4SLinus Torvalds vma = find_vma(mm, addr); 19172afc745fSAkira Takeuchi if (TASK_SIZE - len >= addr && addr >= mmap_min_addr && 19181da177e4SLinus Torvalds (!vma || addr + len <= vma->vm_start)) 19191da177e4SLinus Torvalds return addr; 19201da177e4SLinus Torvalds } 19211da177e4SLinus Torvalds 1922db4fbfb9SMichel Lespinasse info.flags = 0; 1923db4fbfb9SMichel Lespinasse info.length = len; 19244e99b021SHeiko Carstens info.low_limit = mm->mmap_base; 1925db4fbfb9SMichel Lespinasse info.high_limit = TASK_SIZE; 1926db4fbfb9SMichel Lespinasse info.align_mask = 0; 1927db4fbfb9SMichel Lespinasse return vm_unmapped_area(&info); 19281da177e4SLinus Torvalds } 19291da177e4SLinus Torvalds #endif 19301da177e4SLinus Torvalds 19311da177e4SLinus Torvalds /* 19321da177e4SLinus Torvalds * This mmap-allocator allocates new areas top-down from below the 19331da177e4SLinus Torvalds * stack's low limit (the base): 19341da177e4SLinus Torvalds */ 19351da177e4SLinus Torvalds #ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN 19361da177e4SLinus Torvalds unsigned long 19371da177e4SLinus Torvalds arch_get_unmapped_area_topdown(struct file *filp, const unsigned long addr0, 19381da177e4SLinus Torvalds const unsigned long len, const unsigned long pgoff, 19391da177e4SLinus Torvalds const unsigned long flags) 19401da177e4SLinus Torvalds { 19411da177e4SLinus Torvalds struct vm_area_struct *vma; 19421da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 1943db4fbfb9SMichel Lespinasse unsigned long addr = addr0; 1944db4fbfb9SMichel Lespinasse struct vm_unmapped_area_info info; 19451da177e4SLinus Torvalds 19461da177e4SLinus Torvalds /* requested length too big for entire address space */ 19472afc745fSAkira Takeuchi if (len > TASK_SIZE - mmap_min_addr) 19481da177e4SLinus Torvalds return -ENOMEM; 19491da177e4SLinus Torvalds 195006abdfb4SBenjamin Herrenschmidt if (flags & MAP_FIXED) 195106abdfb4SBenjamin Herrenschmidt return addr; 195206abdfb4SBenjamin Herrenschmidt 19531da177e4SLinus Torvalds /* requesting a specific address */ 19541da177e4SLinus Torvalds if (addr) { 19551da177e4SLinus Torvalds addr = PAGE_ALIGN(addr); 19561da177e4SLinus Torvalds vma = find_vma(mm, addr); 19572afc745fSAkira Takeuchi if (TASK_SIZE - len >= addr && addr >= mmap_min_addr && 19581da177e4SLinus Torvalds (!vma || addr + len <= vma->vm_start)) 19591da177e4SLinus Torvalds return addr; 19601da177e4SLinus Torvalds } 19611da177e4SLinus Torvalds 1962db4fbfb9SMichel Lespinasse info.flags = VM_UNMAPPED_AREA_TOPDOWN; 1963db4fbfb9SMichel Lespinasse info.length = len; 19642afc745fSAkira Takeuchi info.low_limit = max(PAGE_SIZE, mmap_min_addr); 1965db4fbfb9SMichel Lespinasse info.high_limit = mm->mmap_base; 1966db4fbfb9SMichel Lespinasse info.align_mask = 0; 1967db4fbfb9SMichel Lespinasse addr = vm_unmapped_area(&info); 1968b716ad95SXiao Guangrong 19691da177e4SLinus Torvalds /* 19701da177e4SLinus Torvalds * A failed mmap() very likely causes application failure, 19711da177e4SLinus Torvalds * so fall back to the bottom-up function here. This scenario 19721da177e4SLinus Torvalds * can happen with large stack limits and large mmap() 19731da177e4SLinus Torvalds * allocations. 19741da177e4SLinus Torvalds */ 1975db4fbfb9SMichel Lespinasse if (addr & ~PAGE_MASK) { 1976db4fbfb9SMichel Lespinasse VM_BUG_ON(addr != -ENOMEM); 1977db4fbfb9SMichel Lespinasse info.flags = 0; 1978db4fbfb9SMichel Lespinasse info.low_limit = TASK_UNMAPPED_BASE; 1979db4fbfb9SMichel Lespinasse info.high_limit = TASK_SIZE; 1980db4fbfb9SMichel Lespinasse addr = vm_unmapped_area(&info); 1981db4fbfb9SMichel Lespinasse } 19821da177e4SLinus Torvalds 19831da177e4SLinus Torvalds return addr; 19841da177e4SLinus Torvalds } 19851da177e4SLinus Torvalds #endif 19861da177e4SLinus Torvalds 19871da177e4SLinus Torvalds unsigned long 19881da177e4SLinus Torvalds get_unmapped_area(struct file *file, unsigned long addr, unsigned long len, 19891da177e4SLinus Torvalds unsigned long pgoff, unsigned long flags) 19901da177e4SLinus Torvalds { 199106abdfb4SBenjamin Herrenschmidt unsigned long (*get_area)(struct file *, unsigned long, 199206abdfb4SBenjamin Herrenschmidt unsigned long, unsigned long, unsigned long); 199307ab67c8SLinus Torvalds 19949206de95SAl Viro unsigned long error = arch_mmap_check(addr, len, flags); 19959206de95SAl Viro if (error) 19969206de95SAl Viro return error; 19979206de95SAl Viro 19989206de95SAl Viro /* Careful about overflows.. */ 19999206de95SAl Viro if (len > TASK_SIZE) 20009206de95SAl Viro return -ENOMEM; 20019206de95SAl Viro 200207ab67c8SLinus Torvalds get_area = current->mm->get_unmapped_area; 200372c2d531SAl Viro if (file && file->f_op->get_unmapped_area) 200407ab67c8SLinus Torvalds get_area = file->f_op->get_unmapped_area; 200507ab67c8SLinus Torvalds addr = get_area(file, addr, len, pgoff, flags); 200607ab67c8SLinus Torvalds if (IS_ERR_VALUE(addr)) 200707ab67c8SLinus Torvalds return addr; 200807ab67c8SLinus Torvalds 20091da177e4SLinus Torvalds if (addr > TASK_SIZE - len) 20101da177e4SLinus Torvalds return -ENOMEM; 20111da177e4SLinus Torvalds if (addr & ~PAGE_MASK) 20121da177e4SLinus Torvalds return -EINVAL; 201306abdfb4SBenjamin Herrenschmidt 20149ac4ed4bSAl Viro addr = arch_rebalance_pgtables(addr, len); 20159ac4ed4bSAl Viro error = security_mmap_addr(addr); 20169ac4ed4bSAl Viro return error ? error : addr; 20171da177e4SLinus Torvalds } 20181da177e4SLinus Torvalds 20191da177e4SLinus Torvalds EXPORT_SYMBOL(get_unmapped_area); 20201da177e4SLinus Torvalds 20211da177e4SLinus Torvalds /* Look up the first VMA which satisfies addr < vm_end, NULL if none. */ 20221da177e4SLinus Torvalds struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr) 20231da177e4SLinus Torvalds { 2024615d6e87SDavidlohr Bueso struct rb_node *rb_node; 2025615d6e87SDavidlohr Bueso struct vm_area_struct *vma; 20261da177e4SLinus Torvalds 20271da177e4SLinus Torvalds /* Check the cache first. */ 2028615d6e87SDavidlohr Bueso vma = vmacache_find(mm, addr); 2029615d6e87SDavidlohr Bueso if (likely(vma)) 2030615d6e87SDavidlohr Bueso return vma; 20311da177e4SLinus Torvalds 20321da177e4SLinus Torvalds rb_node = mm->mm_rb.rb_node; 20331da177e4SLinus Torvalds vma = NULL; 20341da177e4SLinus Torvalds 20351da177e4SLinus Torvalds while (rb_node) { 2036615d6e87SDavidlohr Bueso struct vm_area_struct *tmp; 20371da177e4SLinus Torvalds 2038615d6e87SDavidlohr Bueso tmp = rb_entry(rb_node, struct vm_area_struct, vm_rb); 20391da177e4SLinus Torvalds 2040615d6e87SDavidlohr Bueso if (tmp->vm_end > addr) { 2041615d6e87SDavidlohr Bueso vma = tmp; 2042615d6e87SDavidlohr Bueso if (tmp->vm_start <= addr) 20431da177e4SLinus Torvalds break; 20441da177e4SLinus Torvalds rb_node = rb_node->rb_left; 20451da177e4SLinus Torvalds } else 20461da177e4SLinus Torvalds rb_node = rb_node->rb_right; 20471da177e4SLinus Torvalds } 2048615d6e87SDavidlohr Bueso 20491da177e4SLinus Torvalds if (vma) 2050615d6e87SDavidlohr Bueso vmacache_update(addr, vma); 20511da177e4SLinus Torvalds return vma; 20521da177e4SLinus Torvalds } 20531da177e4SLinus Torvalds 20541da177e4SLinus Torvalds EXPORT_SYMBOL(find_vma); 20551da177e4SLinus Torvalds 20566bd4837dSKOSAKI Motohiro /* 20576bd4837dSKOSAKI Motohiro * Same as find_vma, but also return a pointer to the previous VMA in *pprev. 20586bd4837dSKOSAKI Motohiro */ 20591da177e4SLinus Torvalds struct vm_area_struct * 20601da177e4SLinus Torvalds find_vma_prev(struct mm_struct *mm, unsigned long addr, 20611da177e4SLinus Torvalds struct vm_area_struct **pprev) 20621da177e4SLinus Torvalds { 20636bd4837dSKOSAKI Motohiro struct vm_area_struct *vma; 20641da177e4SLinus Torvalds 20656bd4837dSKOSAKI Motohiro vma = find_vma(mm, addr); 206683cd904dSMikulas Patocka if (vma) { 206783cd904dSMikulas Patocka *pprev = vma->vm_prev; 206883cd904dSMikulas Patocka } else { 206983cd904dSMikulas Patocka struct rb_node *rb_node = mm->mm_rb.rb_node; 207083cd904dSMikulas Patocka *pprev = NULL; 207183cd904dSMikulas Patocka while (rb_node) { 207283cd904dSMikulas Patocka *pprev = rb_entry(rb_node, struct vm_area_struct, vm_rb); 207383cd904dSMikulas Patocka rb_node = rb_node->rb_right; 207483cd904dSMikulas Patocka } 207583cd904dSMikulas Patocka } 20766bd4837dSKOSAKI Motohiro return vma; 20771da177e4SLinus Torvalds } 20781da177e4SLinus Torvalds 20791da177e4SLinus Torvalds /* 20801da177e4SLinus Torvalds * Verify that the stack growth is acceptable and 20811da177e4SLinus Torvalds * update accounting. This is shared with both the 20821da177e4SLinus Torvalds * grow-up and grow-down cases. 20831da177e4SLinus Torvalds */ 20841da177e4SLinus Torvalds static int acct_stack_growth(struct vm_area_struct *vma, unsigned long size, unsigned long grow) 20851da177e4SLinus Torvalds { 20861da177e4SLinus Torvalds struct mm_struct *mm = vma->vm_mm; 20871da177e4SLinus Torvalds struct rlimit *rlim = current->signal->rlim; 20880d59a01bSAdam Litke unsigned long new_start; 20891da177e4SLinus Torvalds 20901da177e4SLinus Torvalds /* address space limit tests */ 2091119f657cSakpm@osdl.org if (!may_expand_vm(mm, grow)) 20921da177e4SLinus Torvalds return -ENOMEM; 20931da177e4SLinus Torvalds 20941da177e4SLinus Torvalds /* Stack limit test */ 209559e99e5bSJiri Slaby if (size > ACCESS_ONCE(rlim[RLIMIT_STACK].rlim_cur)) 20961da177e4SLinus Torvalds return -ENOMEM; 20971da177e4SLinus Torvalds 20981da177e4SLinus Torvalds /* mlock limit tests */ 20991da177e4SLinus Torvalds if (vma->vm_flags & VM_LOCKED) { 21001da177e4SLinus Torvalds unsigned long locked; 21011da177e4SLinus Torvalds unsigned long limit; 21021da177e4SLinus Torvalds locked = mm->locked_vm + grow; 210359e99e5bSJiri Slaby limit = ACCESS_ONCE(rlim[RLIMIT_MEMLOCK].rlim_cur); 210459e99e5bSJiri Slaby limit >>= PAGE_SHIFT; 21051da177e4SLinus Torvalds if (locked > limit && !capable(CAP_IPC_LOCK)) 21061da177e4SLinus Torvalds return -ENOMEM; 21071da177e4SLinus Torvalds } 21081da177e4SLinus Torvalds 21090d59a01bSAdam Litke /* Check to ensure the stack will not grow into a hugetlb-only region */ 21100d59a01bSAdam Litke new_start = (vma->vm_flags & VM_GROWSUP) ? vma->vm_start : 21110d59a01bSAdam Litke vma->vm_end - size; 21120d59a01bSAdam Litke if (is_hugepage_only_range(vma->vm_mm, new_start, size)) 21130d59a01bSAdam Litke return -EFAULT; 21140d59a01bSAdam Litke 21151da177e4SLinus Torvalds /* 21161da177e4SLinus Torvalds * Overcommit.. This must be the final test, as it will 21171da177e4SLinus Torvalds * update security statistics. 21181da177e4SLinus Torvalds */ 211905fa199dSHugh Dickins if (security_vm_enough_memory_mm(mm, grow)) 21201da177e4SLinus Torvalds return -ENOMEM; 21211da177e4SLinus Torvalds 21221da177e4SLinus Torvalds /* Ok, everything looks good - let it rip */ 21231da177e4SLinus Torvalds if (vma->vm_flags & VM_LOCKED) 21241da177e4SLinus Torvalds mm->locked_vm += grow; 2125ab50b8edSHugh Dickins vm_stat_account(mm, vma->vm_flags, vma->vm_file, grow); 21261da177e4SLinus Torvalds return 0; 21271da177e4SLinus Torvalds } 21281da177e4SLinus Torvalds 212946dea3d0SHugh Dickins #if defined(CONFIG_STACK_GROWSUP) || defined(CONFIG_IA64) 21301da177e4SLinus Torvalds /* 213146dea3d0SHugh Dickins * PA-RISC uses this for its stack; IA64 for its Register Backing Store. 213246dea3d0SHugh Dickins * vma is the last one with address > vma->vm_end. Have to extend vma. 21331da177e4SLinus Torvalds */ 213446dea3d0SHugh Dickins int expand_upwards(struct vm_area_struct *vma, unsigned long address) 21351da177e4SLinus Torvalds { 21361da177e4SLinus Torvalds int error; 21371da177e4SLinus Torvalds 21381da177e4SLinus Torvalds if (!(vma->vm_flags & VM_GROWSUP)) 21391da177e4SLinus Torvalds return -EFAULT; 21401da177e4SLinus Torvalds 21411da177e4SLinus Torvalds /* 21421da177e4SLinus Torvalds * We must make sure the anon_vma is allocated 21431da177e4SLinus Torvalds * so that the anon_vma locking is not a noop. 21441da177e4SLinus Torvalds */ 21451da177e4SLinus Torvalds if (unlikely(anon_vma_prepare(vma))) 21461da177e4SLinus Torvalds return -ENOMEM; 2147bb4a340eSRik van Riel vma_lock_anon_vma(vma); 21481da177e4SLinus Torvalds 21491da177e4SLinus Torvalds /* 21501da177e4SLinus Torvalds * vma->vm_start/vm_end cannot change under us because the caller 21511da177e4SLinus Torvalds * is required to hold the mmap_sem in read mode. We need the 21521da177e4SLinus Torvalds * anon_vma lock to serialize against concurrent expand_stacks. 215306b32f3aSHelge Deller * Also guard against wrapping around to address 0. 21541da177e4SLinus Torvalds */ 215506b32f3aSHelge Deller if (address < PAGE_ALIGN(address+4)) 215606b32f3aSHelge Deller address = PAGE_ALIGN(address+4); 215706b32f3aSHelge Deller else { 2158bb4a340eSRik van Riel vma_unlock_anon_vma(vma); 215906b32f3aSHelge Deller return -ENOMEM; 216006b32f3aSHelge Deller } 21611da177e4SLinus Torvalds error = 0; 21621da177e4SLinus Torvalds 21631da177e4SLinus Torvalds /* Somebody else might have raced and expanded it already */ 21641da177e4SLinus Torvalds if (address > vma->vm_end) { 21651da177e4SLinus Torvalds unsigned long size, grow; 21661da177e4SLinus Torvalds 21671da177e4SLinus Torvalds size = address - vma->vm_start; 21681da177e4SLinus Torvalds grow = (address - vma->vm_end) >> PAGE_SHIFT; 21691da177e4SLinus Torvalds 217042c36f63SHugh Dickins error = -ENOMEM; 217142c36f63SHugh Dickins if (vma->vm_pgoff + (size >> PAGE_SHIFT) >= vma->vm_pgoff) { 21721da177e4SLinus Torvalds error = acct_stack_growth(vma, size, grow); 21733af9e859SEric B Munson if (!error) { 21744128997bSMichel Lespinasse /* 21754128997bSMichel Lespinasse * vma_gap_update() doesn't support concurrent 21764128997bSMichel Lespinasse * updates, but we only hold a shared mmap_sem 21774128997bSMichel Lespinasse * lock here, so we need to protect against 21784128997bSMichel Lespinasse * concurrent vma expansions. 21794128997bSMichel Lespinasse * vma_lock_anon_vma() doesn't help here, as 21804128997bSMichel Lespinasse * we don't guarantee that all growable vmas 21814128997bSMichel Lespinasse * in a mm share the same root anon vma. 21824128997bSMichel Lespinasse * So, we reuse mm->page_table_lock to guard 21834128997bSMichel Lespinasse * against concurrent vma expansions. 21844128997bSMichel Lespinasse */ 21854128997bSMichel Lespinasse spin_lock(&vma->vm_mm->page_table_lock); 2186bf181b9fSMichel Lespinasse anon_vma_interval_tree_pre_update_vma(vma); 21871da177e4SLinus Torvalds vma->vm_end = address; 2188bf181b9fSMichel Lespinasse anon_vma_interval_tree_post_update_vma(vma); 2189d3737187SMichel Lespinasse if (vma->vm_next) 2190d3737187SMichel Lespinasse vma_gap_update(vma->vm_next); 2191d3737187SMichel Lespinasse else 2192d3737187SMichel Lespinasse vma->vm_mm->highest_vm_end = address; 21934128997bSMichel Lespinasse spin_unlock(&vma->vm_mm->page_table_lock); 21944128997bSMichel Lespinasse 21953af9e859SEric B Munson perf_event_mmap(vma); 21963af9e859SEric B Munson } 21971da177e4SLinus Torvalds } 219842c36f63SHugh Dickins } 2199bb4a340eSRik van Riel vma_unlock_anon_vma(vma); 2200b15d00b6SAndrea Arcangeli khugepaged_enter_vma_merge(vma); 2201ed8ea815SMichel Lespinasse validate_mm(vma->vm_mm); 22021da177e4SLinus Torvalds return error; 22031da177e4SLinus Torvalds } 220446dea3d0SHugh Dickins #endif /* CONFIG_STACK_GROWSUP || CONFIG_IA64 */ 220546dea3d0SHugh Dickins 22061da177e4SLinus Torvalds /* 22071da177e4SLinus Torvalds * vma is the first one with address < vma->vm_start. Have to extend vma. 22081da177e4SLinus Torvalds */ 2209d05f3169SMichal Hocko int expand_downwards(struct vm_area_struct *vma, 2210b6a2fea3SOllie Wild unsigned long address) 22111da177e4SLinus Torvalds { 22121da177e4SLinus Torvalds int error; 22131da177e4SLinus Torvalds 22141da177e4SLinus Torvalds /* 22151da177e4SLinus Torvalds * We must make sure the anon_vma is allocated 22161da177e4SLinus Torvalds * so that the anon_vma locking is not a noop. 22171da177e4SLinus Torvalds */ 22181da177e4SLinus Torvalds if (unlikely(anon_vma_prepare(vma))) 22191da177e4SLinus Torvalds return -ENOMEM; 22208869477aSEric Paris 22218869477aSEric Paris address &= PAGE_MASK; 2222e5467859SAl Viro error = security_mmap_addr(address); 22238869477aSEric Paris if (error) 22248869477aSEric Paris return error; 22258869477aSEric Paris 2226bb4a340eSRik van Riel vma_lock_anon_vma(vma); 22271da177e4SLinus Torvalds 22281da177e4SLinus Torvalds /* 22291da177e4SLinus Torvalds * vma->vm_start/vm_end cannot change under us because the caller 22301da177e4SLinus Torvalds * is required to hold the mmap_sem in read mode. We need the 22311da177e4SLinus Torvalds * anon_vma lock to serialize against concurrent expand_stacks. 22321da177e4SLinus Torvalds */ 22331da177e4SLinus Torvalds 22341da177e4SLinus Torvalds /* Somebody else might have raced and expanded it already */ 22351da177e4SLinus Torvalds if (address < vma->vm_start) { 22361da177e4SLinus Torvalds unsigned long size, grow; 22371da177e4SLinus Torvalds 22381da177e4SLinus Torvalds size = vma->vm_end - address; 22391da177e4SLinus Torvalds grow = (vma->vm_start - address) >> PAGE_SHIFT; 22401da177e4SLinus Torvalds 2241a626ca6aSLinus Torvalds error = -ENOMEM; 2242a626ca6aSLinus Torvalds if (grow <= vma->vm_pgoff) { 22431da177e4SLinus Torvalds error = acct_stack_growth(vma, size, grow); 22441da177e4SLinus Torvalds if (!error) { 22454128997bSMichel Lespinasse /* 22464128997bSMichel Lespinasse * vma_gap_update() doesn't support concurrent 22474128997bSMichel Lespinasse * updates, but we only hold a shared mmap_sem 22484128997bSMichel Lespinasse * lock here, so we need to protect against 22494128997bSMichel Lespinasse * concurrent vma expansions. 22504128997bSMichel Lespinasse * vma_lock_anon_vma() doesn't help here, as 22514128997bSMichel Lespinasse * we don't guarantee that all growable vmas 22524128997bSMichel Lespinasse * in a mm share the same root anon vma. 22534128997bSMichel Lespinasse * So, we reuse mm->page_table_lock to guard 22544128997bSMichel Lespinasse * against concurrent vma expansions. 22554128997bSMichel Lespinasse */ 22564128997bSMichel Lespinasse spin_lock(&vma->vm_mm->page_table_lock); 2257bf181b9fSMichel Lespinasse anon_vma_interval_tree_pre_update_vma(vma); 22581da177e4SLinus Torvalds vma->vm_start = address; 22591da177e4SLinus Torvalds vma->vm_pgoff -= grow; 2260bf181b9fSMichel Lespinasse anon_vma_interval_tree_post_update_vma(vma); 2261d3737187SMichel Lespinasse vma_gap_update(vma); 22624128997bSMichel Lespinasse spin_unlock(&vma->vm_mm->page_table_lock); 22634128997bSMichel Lespinasse 22643af9e859SEric B Munson perf_event_mmap(vma); 22651da177e4SLinus Torvalds } 22661da177e4SLinus Torvalds } 2267a626ca6aSLinus Torvalds } 2268bb4a340eSRik van Riel vma_unlock_anon_vma(vma); 2269b15d00b6SAndrea Arcangeli khugepaged_enter_vma_merge(vma); 2270ed8ea815SMichel Lespinasse validate_mm(vma->vm_mm); 22711da177e4SLinus Torvalds return error; 22721da177e4SLinus Torvalds } 22731da177e4SLinus Torvalds 227409884964SLinus Torvalds /* 227509884964SLinus Torvalds * Note how expand_stack() refuses to expand the stack all the way to 227609884964SLinus Torvalds * abut the next virtual mapping, *unless* that mapping itself is also 227709884964SLinus Torvalds * a stack mapping. We want to leave room for a guard page, after all 227809884964SLinus Torvalds * (the guard page itself is not added here, that is done by the 227909884964SLinus Torvalds * actual page faulting logic) 228009884964SLinus Torvalds * 228109884964SLinus Torvalds * This matches the behavior of the guard page logic (see mm/memory.c: 228209884964SLinus Torvalds * check_stack_guard_page()), which only allows the guard page to be 228309884964SLinus Torvalds * removed under these circumstances. 228409884964SLinus Torvalds */ 2285b6a2fea3SOllie Wild #ifdef CONFIG_STACK_GROWSUP 2286b6a2fea3SOllie Wild int expand_stack(struct vm_area_struct *vma, unsigned long address) 2287b6a2fea3SOllie Wild { 228809884964SLinus Torvalds struct vm_area_struct *next; 228909884964SLinus Torvalds 229009884964SLinus Torvalds address &= PAGE_MASK; 229109884964SLinus Torvalds next = vma->vm_next; 229209884964SLinus Torvalds if (next && next->vm_start == address + PAGE_SIZE) { 229309884964SLinus Torvalds if (!(next->vm_flags & VM_GROWSUP)) 229409884964SLinus Torvalds return -ENOMEM; 229509884964SLinus Torvalds } 2296b6a2fea3SOllie Wild return expand_upwards(vma, address); 2297b6a2fea3SOllie Wild } 2298b6a2fea3SOllie Wild 2299b6a2fea3SOllie Wild struct vm_area_struct * 2300b6a2fea3SOllie Wild find_extend_vma(struct mm_struct *mm, unsigned long addr) 2301b6a2fea3SOllie Wild { 2302b6a2fea3SOllie Wild struct vm_area_struct *vma, *prev; 2303b6a2fea3SOllie Wild 2304b6a2fea3SOllie Wild addr &= PAGE_MASK; 2305b6a2fea3SOllie Wild vma = find_vma_prev(mm, addr, &prev); 2306b6a2fea3SOllie Wild if (vma && (vma->vm_start <= addr)) 2307b6a2fea3SOllie Wild return vma; 23081c127185SDenys Vlasenko if (!prev || expand_stack(prev, addr)) 2309b6a2fea3SOllie Wild return NULL; 2310cea10a19SMichel Lespinasse if (prev->vm_flags & VM_LOCKED) 2311cea10a19SMichel Lespinasse __mlock_vma_pages_range(prev, addr, prev->vm_end, NULL); 2312b6a2fea3SOllie Wild return prev; 2313b6a2fea3SOllie Wild } 2314b6a2fea3SOllie Wild #else 2315b6a2fea3SOllie Wild int expand_stack(struct vm_area_struct *vma, unsigned long address) 2316b6a2fea3SOllie Wild { 231709884964SLinus Torvalds struct vm_area_struct *prev; 231809884964SLinus Torvalds 231909884964SLinus Torvalds address &= PAGE_MASK; 232009884964SLinus Torvalds prev = vma->vm_prev; 232109884964SLinus Torvalds if (prev && prev->vm_end == address) { 232209884964SLinus Torvalds if (!(prev->vm_flags & VM_GROWSDOWN)) 232309884964SLinus Torvalds return -ENOMEM; 232409884964SLinus Torvalds } 2325b6a2fea3SOllie Wild return expand_downwards(vma, address); 2326b6a2fea3SOllie Wild } 2327b6a2fea3SOllie Wild 23281da177e4SLinus Torvalds struct vm_area_struct * 23291da177e4SLinus Torvalds find_extend_vma(struct mm_struct *mm, unsigned long addr) 23301da177e4SLinus Torvalds { 23311da177e4SLinus Torvalds struct vm_area_struct *vma; 23321da177e4SLinus Torvalds unsigned long start; 23331da177e4SLinus Torvalds 23341da177e4SLinus Torvalds addr &= PAGE_MASK; 23351da177e4SLinus Torvalds vma = find_vma(mm, addr); 23361da177e4SLinus Torvalds if (!vma) 23371da177e4SLinus Torvalds return NULL; 23381da177e4SLinus Torvalds if (vma->vm_start <= addr) 23391da177e4SLinus Torvalds return vma; 23401da177e4SLinus Torvalds if (!(vma->vm_flags & VM_GROWSDOWN)) 23411da177e4SLinus Torvalds return NULL; 23421da177e4SLinus Torvalds start = vma->vm_start; 23431da177e4SLinus Torvalds if (expand_stack(vma, addr)) 23441da177e4SLinus Torvalds return NULL; 2345cea10a19SMichel Lespinasse if (vma->vm_flags & VM_LOCKED) 2346cea10a19SMichel Lespinasse __mlock_vma_pages_range(vma, addr, start, NULL); 23471da177e4SLinus Torvalds return vma; 23481da177e4SLinus Torvalds } 23491da177e4SLinus Torvalds #endif 23501da177e4SLinus Torvalds 23512c0b3814SHugh Dickins /* 23522c0b3814SHugh Dickins * Ok - we have the memory areas we should free on the vma list, 23532c0b3814SHugh Dickins * so release them, and do the vma updates. 23541da177e4SLinus Torvalds * 23552c0b3814SHugh Dickins * Called with the mm semaphore held. 23561da177e4SLinus Torvalds */ 23572c0b3814SHugh Dickins static void remove_vma_list(struct mm_struct *mm, struct vm_area_struct *vma) 23581da177e4SLinus Torvalds { 23594f74d2c8SLinus Torvalds unsigned long nr_accounted = 0; 23604f74d2c8SLinus Torvalds 2361365e9c87SHugh Dickins /* Update high watermark before we lower total_vm */ 2362365e9c87SHugh Dickins update_hiwater_vm(mm); 23632c0b3814SHugh Dickins do { 2364ab50b8edSHugh Dickins long nrpages = vma_pages(vma); 23651da177e4SLinus Torvalds 23664f74d2c8SLinus Torvalds if (vma->vm_flags & VM_ACCOUNT) 23674f74d2c8SLinus Torvalds nr_accounted += nrpages; 2368ab50b8edSHugh Dickins vm_stat_account(mm, vma->vm_flags, vma->vm_file, -nrpages); 2369a8fb5618SHugh Dickins vma = remove_vma(vma); 2370146425a3SHugh Dickins } while (vma); 23714f74d2c8SLinus Torvalds vm_unacct_memory(nr_accounted); 23721da177e4SLinus Torvalds validate_mm(mm); 23731da177e4SLinus Torvalds } 23741da177e4SLinus Torvalds 23751da177e4SLinus Torvalds /* 23761da177e4SLinus Torvalds * Get rid of page table information in the indicated region. 23771da177e4SLinus Torvalds * 2378f10df686SPaolo 'Blaisorblade' Giarrusso * Called with the mm semaphore held. 23791da177e4SLinus Torvalds */ 23801da177e4SLinus Torvalds static void unmap_region(struct mm_struct *mm, 2381e0da382cSHugh Dickins struct vm_area_struct *vma, struct vm_area_struct *prev, 2382e0da382cSHugh Dickins unsigned long start, unsigned long end) 23831da177e4SLinus Torvalds { 2384e0da382cSHugh Dickins struct vm_area_struct *next = prev ? prev->vm_next : mm->mmap; 2385d16dfc55SPeter Zijlstra struct mmu_gather tlb; 23861da177e4SLinus Torvalds 23871da177e4SLinus Torvalds lru_add_drain(); 23882b047252SLinus Torvalds tlb_gather_mmu(&tlb, mm, start, end); 2389365e9c87SHugh Dickins update_hiwater_rss(mm); 23904f74d2c8SLinus Torvalds unmap_vmas(&tlb, vma, start, end); 2391d16dfc55SPeter Zijlstra free_pgtables(&tlb, vma, prev ? prev->vm_end : FIRST_USER_ADDRESS, 23926ee8630eSHugh Dickins next ? next->vm_start : USER_PGTABLES_CEILING); 2393d16dfc55SPeter Zijlstra tlb_finish_mmu(&tlb, start, end); 23941da177e4SLinus Torvalds } 23951da177e4SLinus Torvalds 23961da177e4SLinus Torvalds /* 23971da177e4SLinus Torvalds * Create a list of vma's touched by the unmap, removing them from the mm's 23981da177e4SLinus Torvalds * vma list as we go.. 23991da177e4SLinus Torvalds */ 24001da177e4SLinus Torvalds static void 24011da177e4SLinus Torvalds detach_vmas_to_be_unmapped(struct mm_struct *mm, struct vm_area_struct *vma, 24021da177e4SLinus Torvalds struct vm_area_struct *prev, unsigned long end) 24031da177e4SLinus Torvalds { 24041da177e4SLinus Torvalds struct vm_area_struct **insertion_point; 24051da177e4SLinus Torvalds struct vm_area_struct *tail_vma = NULL; 24061da177e4SLinus Torvalds 24071da177e4SLinus Torvalds insertion_point = (prev ? &prev->vm_next : &mm->mmap); 2408297c5eeeSLinus Torvalds vma->vm_prev = NULL; 24091da177e4SLinus Torvalds do { 2410d3737187SMichel Lespinasse vma_rb_erase(vma, &mm->mm_rb); 24111da177e4SLinus Torvalds mm->map_count--; 24121da177e4SLinus Torvalds tail_vma = vma; 24131da177e4SLinus Torvalds vma = vma->vm_next; 24141da177e4SLinus Torvalds } while (vma && vma->vm_start < end); 24151da177e4SLinus Torvalds *insertion_point = vma; 2416d3737187SMichel Lespinasse if (vma) { 2417297c5eeeSLinus Torvalds vma->vm_prev = prev; 2418d3737187SMichel Lespinasse vma_gap_update(vma); 2419d3737187SMichel Lespinasse } else 2420d3737187SMichel Lespinasse mm->highest_vm_end = prev ? prev->vm_end : 0; 24211da177e4SLinus Torvalds tail_vma->vm_next = NULL; 2422615d6e87SDavidlohr Bueso 2423615d6e87SDavidlohr Bueso /* Kill the cache */ 2424615d6e87SDavidlohr Bueso vmacache_invalidate(mm); 24251da177e4SLinus Torvalds } 24261da177e4SLinus Torvalds 24271da177e4SLinus Torvalds /* 2428659ace58SKOSAKI Motohiro * __split_vma() bypasses sysctl_max_map_count checking. We use this on the 2429659ace58SKOSAKI Motohiro * munmap path where it doesn't make sense to fail. 24301da177e4SLinus Torvalds */ 2431659ace58SKOSAKI Motohiro static int __split_vma(struct mm_struct *mm, struct vm_area_struct *vma, 24321da177e4SLinus Torvalds unsigned long addr, int new_below) 24331da177e4SLinus Torvalds { 24341da177e4SLinus Torvalds struct vm_area_struct *new; 24355beb4930SRik van Riel int err = -ENOMEM; 24361da177e4SLinus Torvalds 2437a5516438SAndi Kleen if (is_vm_hugetlb_page(vma) && (addr & 2438a5516438SAndi Kleen ~(huge_page_mask(hstate_vma(vma))))) 24391da177e4SLinus Torvalds return -EINVAL; 24401da177e4SLinus Torvalds 2441e94b1766SChristoph Lameter new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL); 24421da177e4SLinus Torvalds if (!new) 24435beb4930SRik van Riel goto out_err; 24441da177e4SLinus Torvalds 24451da177e4SLinus Torvalds /* most fields are the same, copy all, and then fixup */ 24461da177e4SLinus Torvalds *new = *vma; 24471da177e4SLinus Torvalds 24485beb4930SRik van Riel INIT_LIST_HEAD(&new->anon_vma_chain); 24495beb4930SRik van Riel 24501da177e4SLinus Torvalds if (new_below) 24511da177e4SLinus Torvalds new->vm_end = addr; 24521da177e4SLinus Torvalds else { 24531da177e4SLinus Torvalds new->vm_start = addr; 24541da177e4SLinus Torvalds new->vm_pgoff += ((addr - vma->vm_start) >> PAGE_SHIFT); 24551da177e4SLinus Torvalds } 24561da177e4SLinus Torvalds 2457ef0855d3SOleg Nesterov err = vma_dup_policy(vma, new); 2458ef0855d3SOleg Nesterov if (err) 24595beb4930SRik van Riel goto out_free_vma; 24601da177e4SLinus Torvalds 24615beb4930SRik van Riel if (anon_vma_clone(new, vma)) 24625beb4930SRik van Riel goto out_free_mpol; 24635beb4930SRik van Riel 2464e9714acfSKonstantin Khlebnikov if (new->vm_file) 24651da177e4SLinus Torvalds get_file(new->vm_file); 24661da177e4SLinus Torvalds 24671da177e4SLinus Torvalds if (new->vm_ops && new->vm_ops->open) 24681da177e4SLinus Torvalds new->vm_ops->open(new); 24691da177e4SLinus Torvalds 24701da177e4SLinus Torvalds if (new_below) 24715beb4930SRik van Riel err = vma_adjust(vma, addr, vma->vm_end, vma->vm_pgoff + 24721da177e4SLinus Torvalds ((addr - new->vm_start) >> PAGE_SHIFT), new); 24731da177e4SLinus Torvalds else 24745beb4930SRik van Riel err = vma_adjust(vma, vma->vm_start, addr, vma->vm_pgoff, new); 24751da177e4SLinus Torvalds 24765beb4930SRik van Riel /* Success. */ 24775beb4930SRik van Riel if (!err) 24781da177e4SLinus Torvalds return 0; 24795beb4930SRik van Riel 24805beb4930SRik van Riel /* Clean everything up if vma_adjust failed. */ 248158927533SRik van Riel if (new->vm_ops && new->vm_ops->close) 24825beb4930SRik van Riel new->vm_ops->close(new); 2483e9714acfSKonstantin Khlebnikov if (new->vm_file) 24845beb4930SRik van Riel fput(new->vm_file); 24852aeadc30SAndrea Arcangeli unlink_anon_vmas(new); 24865beb4930SRik van Riel out_free_mpol: 2487ef0855d3SOleg Nesterov mpol_put(vma_policy(new)); 24885beb4930SRik van Riel out_free_vma: 24895beb4930SRik van Riel kmem_cache_free(vm_area_cachep, new); 24905beb4930SRik van Riel out_err: 24915beb4930SRik van Riel return err; 24921da177e4SLinus Torvalds } 24931da177e4SLinus Torvalds 2494659ace58SKOSAKI Motohiro /* 2495659ace58SKOSAKI Motohiro * Split a vma into two pieces at address 'addr', a new vma is allocated 2496659ace58SKOSAKI Motohiro * either for the first part or the tail. 2497659ace58SKOSAKI Motohiro */ 2498659ace58SKOSAKI Motohiro int split_vma(struct mm_struct *mm, struct vm_area_struct *vma, 2499659ace58SKOSAKI Motohiro unsigned long addr, int new_below) 2500659ace58SKOSAKI Motohiro { 2501659ace58SKOSAKI Motohiro if (mm->map_count >= sysctl_max_map_count) 2502659ace58SKOSAKI Motohiro return -ENOMEM; 2503659ace58SKOSAKI Motohiro 2504659ace58SKOSAKI Motohiro return __split_vma(mm, vma, addr, new_below); 2505659ace58SKOSAKI Motohiro } 2506659ace58SKOSAKI Motohiro 25071da177e4SLinus Torvalds /* Munmap is split into 2 main parts -- this part which finds 25081da177e4SLinus Torvalds * what needs doing, and the areas themselves, which do the 25091da177e4SLinus Torvalds * work. This now handles partial unmappings. 25101da177e4SLinus Torvalds * Jeremy Fitzhardinge <jeremy@goop.org> 25111da177e4SLinus Torvalds */ 25121da177e4SLinus Torvalds int do_munmap(struct mm_struct *mm, unsigned long start, size_t len) 25131da177e4SLinus Torvalds { 25141da177e4SLinus Torvalds unsigned long end; 2515146425a3SHugh Dickins struct vm_area_struct *vma, *prev, *last; 25161da177e4SLinus Torvalds 25171da177e4SLinus Torvalds if ((start & ~PAGE_MASK) || start > TASK_SIZE || len > TASK_SIZE-start) 25181da177e4SLinus Torvalds return -EINVAL; 25191da177e4SLinus Torvalds 2520cc71aba3Svishnu.ps len = PAGE_ALIGN(len); 2521cc71aba3Svishnu.ps if (len == 0) 25221da177e4SLinus Torvalds return -EINVAL; 25231da177e4SLinus Torvalds 25241da177e4SLinus Torvalds /* Find the first overlapping VMA */ 25259be34c9dSLinus Torvalds vma = find_vma(mm, start); 2526146425a3SHugh Dickins if (!vma) 25271da177e4SLinus Torvalds return 0; 25289be34c9dSLinus Torvalds prev = vma->vm_prev; 2529146425a3SHugh Dickins /* we have start < vma->vm_end */ 25301da177e4SLinus Torvalds 25311da177e4SLinus Torvalds /* if it doesn't overlap, we have nothing.. */ 25321da177e4SLinus Torvalds end = start + len; 2533146425a3SHugh Dickins if (vma->vm_start >= end) 25341da177e4SLinus Torvalds return 0; 25351da177e4SLinus Torvalds 25361da177e4SLinus Torvalds /* 25371da177e4SLinus Torvalds * If we need to split any vma, do it now to save pain later. 25381da177e4SLinus Torvalds * 25391da177e4SLinus Torvalds * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially 25401da177e4SLinus Torvalds * unmapped vm_area_struct will remain in use: so lower split_vma 25411da177e4SLinus Torvalds * places tmp vma above, and higher split_vma places tmp vma below. 25421da177e4SLinus Torvalds */ 2543146425a3SHugh Dickins if (start > vma->vm_start) { 2544659ace58SKOSAKI Motohiro int error; 2545659ace58SKOSAKI Motohiro 2546659ace58SKOSAKI Motohiro /* 2547659ace58SKOSAKI Motohiro * Make sure that map_count on return from munmap() will 2548659ace58SKOSAKI Motohiro * not exceed its limit; but let map_count go just above 2549659ace58SKOSAKI Motohiro * its limit temporarily, to help free resources as expected. 2550659ace58SKOSAKI Motohiro */ 2551659ace58SKOSAKI Motohiro if (end < vma->vm_end && mm->map_count >= sysctl_max_map_count) 2552659ace58SKOSAKI Motohiro return -ENOMEM; 2553659ace58SKOSAKI Motohiro 2554659ace58SKOSAKI Motohiro error = __split_vma(mm, vma, start, 0); 25551da177e4SLinus Torvalds if (error) 25561da177e4SLinus Torvalds return error; 2557146425a3SHugh Dickins prev = vma; 25581da177e4SLinus Torvalds } 25591da177e4SLinus Torvalds 25601da177e4SLinus Torvalds /* Does it split the last one? */ 25611da177e4SLinus Torvalds last = find_vma(mm, end); 25621da177e4SLinus Torvalds if (last && end > last->vm_start) { 2563659ace58SKOSAKI Motohiro int error = __split_vma(mm, last, end, 1); 25641da177e4SLinus Torvalds if (error) 25651da177e4SLinus Torvalds return error; 25661da177e4SLinus Torvalds } 2567146425a3SHugh Dickins vma = prev ? prev->vm_next : mm->mmap; 25681da177e4SLinus Torvalds 25691da177e4SLinus Torvalds /* 2570ba470de4SRik van Riel * unlock any mlock()ed ranges before detaching vmas 2571ba470de4SRik van Riel */ 2572ba470de4SRik van Riel if (mm->locked_vm) { 2573ba470de4SRik van Riel struct vm_area_struct *tmp = vma; 2574ba470de4SRik van Riel while (tmp && tmp->vm_start < end) { 2575ba470de4SRik van Riel if (tmp->vm_flags & VM_LOCKED) { 2576ba470de4SRik van Riel mm->locked_vm -= vma_pages(tmp); 2577ba470de4SRik van Riel munlock_vma_pages_all(tmp); 2578ba470de4SRik van Riel } 2579ba470de4SRik van Riel tmp = tmp->vm_next; 2580ba470de4SRik van Riel } 2581ba470de4SRik van Riel } 2582ba470de4SRik van Riel 2583ba470de4SRik van Riel /* 25841da177e4SLinus Torvalds * Remove the vma's, and unmap the actual pages 25851da177e4SLinus Torvalds */ 2586146425a3SHugh Dickins detach_vmas_to_be_unmapped(mm, vma, prev, end); 2587146425a3SHugh Dickins unmap_region(mm, vma, prev, start, end); 25881da177e4SLinus Torvalds 25891da177e4SLinus Torvalds /* Fix up all other VM information */ 25902c0b3814SHugh Dickins remove_vma_list(mm, vma); 25911da177e4SLinus Torvalds 25921da177e4SLinus Torvalds return 0; 25931da177e4SLinus Torvalds } 25941da177e4SLinus Torvalds 2595bfce281cSAl Viro int vm_munmap(unsigned long start, size_t len) 2596a46ef99dSLinus Torvalds { 2597a46ef99dSLinus Torvalds int ret; 2598bfce281cSAl Viro struct mm_struct *mm = current->mm; 2599a46ef99dSLinus Torvalds 2600a46ef99dSLinus Torvalds down_write(&mm->mmap_sem); 2601a46ef99dSLinus Torvalds ret = do_munmap(mm, start, len); 2602a46ef99dSLinus Torvalds up_write(&mm->mmap_sem); 2603a46ef99dSLinus Torvalds return ret; 2604a46ef99dSLinus Torvalds } 2605a46ef99dSLinus Torvalds EXPORT_SYMBOL(vm_munmap); 2606a46ef99dSLinus Torvalds 26076a6160a7SHeiko Carstens SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len) 26081da177e4SLinus Torvalds { 26091da177e4SLinus Torvalds profile_munmap(addr); 2610bfce281cSAl Viro return vm_munmap(addr, len); 26111da177e4SLinus Torvalds } 26121da177e4SLinus Torvalds 26131da177e4SLinus Torvalds static inline void verify_mm_writelocked(struct mm_struct *mm) 26141da177e4SLinus Torvalds { 2615a241ec65SPaul E. McKenney #ifdef CONFIG_DEBUG_VM 26161da177e4SLinus Torvalds if (unlikely(down_read_trylock(&mm->mmap_sem))) { 26171da177e4SLinus Torvalds WARN_ON(1); 26181da177e4SLinus Torvalds up_read(&mm->mmap_sem); 26191da177e4SLinus Torvalds } 26201da177e4SLinus Torvalds #endif 26211da177e4SLinus Torvalds } 26221da177e4SLinus Torvalds 26231da177e4SLinus Torvalds /* 26241da177e4SLinus Torvalds * this is really a simplified "do_mmap". it only handles 26251da177e4SLinus Torvalds * anonymous maps. eventually we may be able to do some 26261da177e4SLinus Torvalds * brk-specific accounting here. 26271da177e4SLinus Torvalds */ 2628e4eb1ff6SLinus Torvalds static unsigned long do_brk(unsigned long addr, unsigned long len) 26291da177e4SLinus Torvalds { 26301da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 26311da177e4SLinus Torvalds struct vm_area_struct *vma, *prev; 26321da177e4SLinus Torvalds unsigned long flags; 26331da177e4SLinus Torvalds struct rb_node **rb_link, *rb_parent; 26341da177e4SLinus Torvalds pgoff_t pgoff = addr >> PAGE_SHIFT; 26353a459756SKirill Korotaev int error; 26361da177e4SLinus Torvalds 26371da177e4SLinus Torvalds len = PAGE_ALIGN(len); 26381da177e4SLinus Torvalds if (!len) 26391da177e4SLinus Torvalds return addr; 26401da177e4SLinus Torvalds 26413a459756SKirill Korotaev flags = VM_DATA_DEFAULT_FLAGS | VM_ACCOUNT | mm->def_flags; 26423a459756SKirill Korotaev 26432c6a1016SAl Viro error = get_unmapped_area(NULL, addr, len, 0, MAP_FIXED); 26442c6a1016SAl Viro if (error & ~PAGE_MASK) 26453a459756SKirill Korotaev return error; 26463a459756SKirill Korotaev 2647363ee17fSDavidlohr Bueso error = mlock_future_check(mm, mm->def_flags, len); 2648363ee17fSDavidlohr Bueso if (error) 2649363ee17fSDavidlohr Bueso return error; 26501da177e4SLinus Torvalds 26511da177e4SLinus Torvalds /* 26521da177e4SLinus Torvalds * mm->mmap_sem is required to protect against another thread 26531da177e4SLinus Torvalds * changing the mappings in case we sleep. 26541da177e4SLinus Torvalds */ 26551da177e4SLinus Torvalds verify_mm_writelocked(mm); 26561da177e4SLinus Torvalds 26571da177e4SLinus Torvalds /* 26581da177e4SLinus Torvalds * Clear old maps. this also does some error checking for us 26591da177e4SLinus Torvalds */ 26601da177e4SLinus Torvalds munmap_back: 26616597d783SHugh Dickins if (find_vma_links(mm, addr, addr + len, &prev, &rb_link, &rb_parent)) { 26621da177e4SLinus Torvalds if (do_munmap(mm, addr, len)) 26631da177e4SLinus Torvalds return -ENOMEM; 26641da177e4SLinus Torvalds goto munmap_back; 26651da177e4SLinus Torvalds } 26661da177e4SLinus Torvalds 26671da177e4SLinus Torvalds /* Check against address space limits *after* clearing old maps... */ 2668119f657cSakpm@osdl.org if (!may_expand_vm(mm, len >> PAGE_SHIFT)) 26691da177e4SLinus Torvalds return -ENOMEM; 26701da177e4SLinus Torvalds 26711da177e4SLinus Torvalds if (mm->map_count > sysctl_max_map_count) 26721da177e4SLinus Torvalds return -ENOMEM; 26731da177e4SLinus Torvalds 2674191c5424SAl Viro if (security_vm_enough_memory_mm(mm, len >> PAGE_SHIFT)) 26751da177e4SLinus Torvalds return -ENOMEM; 26761da177e4SLinus Torvalds 26771da177e4SLinus Torvalds /* Can we just expand an old private anonymous mapping? */ 2678ba470de4SRik van Riel vma = vma_merge(mm, prev, addr, addr + len, flags, 2679ba470de4SRik van Riel NULL, NULL, pgoff, NULL); 2680ba470de4SRik van Riel if (vma) 26811da177e4SLinus Torvalds goto out; 26821da177e4SLinus Torvalds 26831da177e4SLinus Torvalds /* 26841da177e4SLinus Torvalds * create a vma struct for an anonymous mapping 26851da177e4SLinus Torvalds */ 2686c5e3b83eSPekka Enberg vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL); 26871da177e4SLinus Torvalds if (!vma) { 26881da177e4SLinus Torvalds vm_unacct_memory(len >> PAGE_SHIFT); 26891da177e4SLinus Torvalds return -ENOMEM; 26901da177e4SLinus Torvalds } 26911da177e4SLinus Torvalds 26925beb4930SRik van Riel INIT_LIST_HEAD(&vma->anon_vma_chain); 26931da177e4SLinus Torvalds vma->vm_mm = mm; 26941da177e4SLinus Torvalds vma->vm_start = addr; 26951da177e4SLinus Torvalds vma->vm_end = addr + len; 26961da177e4SLinus Torvalds vma->vm_pgoff = pgoff; 26971da177e4SLinus Torvalds vma->vm_flags = flags; 26983ed75eb8SColy Li vma->vm_page_prot = vm_get_page_prot(flags); 26991da177e4SLinus Torvalds vma_link(mm, vma, prev, rb_link, rb_parent); 27001da177e4SLinus Torvalds out: 27013af9e859SEric B Munson perf_event_mmap(vma); 27021da177e4SLinus Torvalds mm->total_vm += len >> PAGE_SHIFT; 2703128557ffSMichel Lespinasse if (flags & VM_LOCKED) 2704ba470de4SRik van Riel mm->locked_vm += (len >> PAGE_SHIFT); 2705d9104d1cSCyrill Gorcunov vma->vm_flags |= VM_SOFTDIRTY; 27061da177e4SLinus Torvalds return addr; 27071da177e4SLinus Torvalds } 27081da177e4SLinus Torvalds 2709e4eb1ff6SLinus Torvalds unsigned long vm_brk(unsigned long addr, unsigned long len) 2710e4eb1ff6SLinus Torvalds { 2711e4eb1ff6SLinus Torvalds struct mm_struct *mm = current->mm; 2712e4eb1ff6SLinus Torvalds unsigned long ret; 2713128557ffSMichel Lespinasse bool populate; 2714e4eb1ff6SLinus Torvalds 2715e4eb1ff6SLinus Torvalds down_write(&mm->mmap_sem); 2716e4eb1ff6SLinus Torvalds ret = do_brk(addr, len); 2717128557ffSMichel Lespinasse populate = ((mm->def_flags & VM_LOCKED) != 0); 2718e4eb1ff6SLinus Torvalds up_write(&mm->mmap_sem); 2719128557ffSMichel Lespinasse if (populate) 2720128557ffSMichel Lespinasse mm_populate(addr, len); 2721e4eb1ff6SLinus Torvalds return ret; 2722e4eb1ff6SLinus Torvalds } 2723e4eb1ff6SLinus Torvalds EXPORT_SYMBOL(vm_brk); 27241da177e4SLinus Torvalds 27251da177e4SLinus Torvalds /* Release all mmaps. */ 27261da177e4SLinus Torvalds void exit_mmap(struct mm_struct *mm) 27271da177e4SLinus Torvalds { 2728d16dfc55SPeter Zijlstra struct mmu_gather tlb; 2729ba470de4SRik van Riel struct vm_area_struct *vma; 27301da177e4SLinus Torvalds unsigned long nr_accounted = 0; 27311da177e4SLinus Torvalds 2732d6dd61c8SJeremy Fitzhardinge /* mm's last user has gone, and its about to be pulled down */ 2733cddb8a5cSAndrea Arcangeli mmu_notifier_release(mm); 2734d6dd61c8SJeremy Fitzhardinge 2735ba470de4SRik van Riel if (mm->locked_vm) { 2736ba470de4SRik van Riel vma = mm->mmap; 2737ba470de4SRik van Riel while (vma) { 2738ba470de4SRik van Riel if (vma->vm_flags & VM_LOCKED) 2739ba470de4SRik van Riel munlock_vma_pages_all(vma); 2740ba470de4SRik van Riel vma = vma->vm_next; 2741ba470de4SRik van Riel } 2742ba470de4SRik van Riel } 27439480c53eSJeremy Fitzhardinge 27449480c53eSJeremy Fitzhardinge arch_exit_mmap(mm); 27459480c53eSJeremy Fitzhardinge 2746ba470de4SRik van Riel vma = mm->mmap; 27479480c53eSJeremy Fitzhardinge if (!vma) /* Can happen if dup_mmap() received an OOM */ 27489480c53eSJeremy Fitzhardinge return; 27499480c53eSJeremy Fitzhardinge 27501da177e4SLinus Torvalds lru_add_drain(); 27511da177e4SLinus Torvalds flush_cache_mm(mm); 27522b047252SLinus Torvalds tlb_gather_mmu(&tlb, mm, 0, -1); 2753901608d9SOleg Nesterov /* update_hiwater_rss(mm) here? but nobody should be looking */ 2754e0da382cSHugh Dickins /* Use -1 here to ensure all VMAs in the mm are unmapped */ 27554f74d2c8SLinus Torvalds unmap_vmas(&tlb, vma, 0, -1); 27569ba69294SHugh Dickins 27576ee8630eSHugh Dickins free_pgtables(&tlb, vma, FIRST_USER_ADDRESS, USER_PGTABLES_CEILING); 2758853f5e26SAl Viro tlb_finish_mmu(&tlb, 0, -1); 27591da177e4SLinus Torvalds 27601da177e4SLinus Torvalds /* 27618f4f8c16SHugh Dickins * Walk the list again, actually closing and freeing it, 27628f4f8c16SHugh Dickins * with preemption enabled, without holding any MM locks. 27631da177e4SLinus Torvalds */ 27644f74d2c8SLinus Torvalds while (vma) { 27654f74d2c8SLinus Torvalds if (vma->vm_flags & VM_ACCOUNT) 27664f74d2c8SLinus Torvalds nr_accounted += vma_pages(vma); 2767a8fb5618SHugh Dickins vma = remove_vma(vma); 27684f74d2c8SLinus Torvalds } 27694f74d2c8SLinus Torvalds vm_unacct_memory(nr_accounted); 2770e0da382cSHugh Dickins 2771e1f56c89SKirill A. Shutemov WARN_ON(atomic_long_read(&mm->nr_ptes) > 2772e1f56c89SKirill A. Shutemov (FIRST_USER_ADDRESS+PMD_SIZE-1)>>PMD_SHIFT); 27731da177e4SLinus Torvalds } 27741da177e4SLinus Torvalds 27751da177e4SLinus Torvalds /* Insert vm structure into process list sorted by address 27761da177e4SLinus Torvalds * and into the inode's i_mmap tree. If vm_file is non-NULL 27773d48ae45SPeter Zijlstra * then i_mmap_mutex is taken here. 27781da177e4SLinus Torvalds */ 27791da177e4SLinus Torvalds int insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma) 27801da177e4SLinus Torvalds { 27816597d783SHugh Dickins struct vm_area_struct *prev; 27821da177e4SLinus Torvalds struct rb_node **rb_link, *rb_parent; 27831da177e4SLinus Torvalds 27841da177e4SLinus Torvalds /* 27851da177e4SLinus Torvalds * The vm_pgoff of a purely anonymous vma should be irrelevant 27861da177e4SLinus Torvalds * until its first write fault, when page's anon_vma and index 27871da177e4SLinus Torvalds * are set. But now set the vm_pgoff it will almost certainly 27881da177e4SLinus Torvalds * end up with (unless mremap moves it elsewhere before that 27891da177e4SLinus Torvalds * first wfault), so /proc/pid/maps tells a consistent story. 27901da177e4SLinus Torvalds * 27911da177e4SLinus Torvalds * By setting it to reflect the virtual start address of the 27921da177e4SLinus Torvalds * vma, merges and splits can happen in a seamless way, just 27931da177e4SLinus Torvalds * using the existing file pgoff checks and manipulations. 27941da177e4SLinus Torvalds * Similarly in do_mmap_pgoff and in do_brk. 27951da177e4SLinus Torvalds */ 27961da177e4SLinus Torvalds if (!vma->vm_file) { 27971da177e4SLinus Torvalds BUG_ON(vma->anon_vma); 27981da177e4SLinus Torvalds vma->vm_pgoff = vma->vm_start >> PAGE_SHIFT; 27991da177e4SLinus Torvalds } 28006597d783SHugh Dickins if (find_vma_links(mm, vma->vm_start, vma->vm_end, 28016597d783SHugh Dickins &prev, &rb_link, &rb_parent)) 28021da177e4SLinus Torvalds return -ENOMEM; 28032fd4ef85SHugh Dickins if ((vma->vm_flags & VM_ACCOUNT) && 280434b4e4aaSAlan Cox security_vm_enough_memory_mm(mm, vma_pages(vma))) 28052fd4ef85SHugh Dickins return -ENOMEM; 28062b144498SSrikar Dronamraju 28071da177e4SLinus Torvalds vma_link(mm, vma, prev, rb_link, rb_parent); 28081da177e4SLinus Torvalds return 0; 28091da177e4SLinus Torvalds } 28101da177e4SLinus Torvalds 28111da177e4SLinus Torvalds /* 28121da177e4SLinus Torvalds * Copy the vma structure to a new location in the same mm, 28131da177e4SLinus Torvalds * prior to moving page table entries, to effect an mremap move. 28141da177e4SLinus Torvalds */ 28151da177e4SLinus Torvalds struct vm_area_struct *copy_vma(struct vm_area_struct **vmap, 281638a76013SMichel Lespinasse unsigned long addr, unsigned long len, pgoff_t pgoff, 281738a76013SMichel Lespinasse bool *need_rmap_locks) 28181da177e4SLinus Torvalds { 28191da177e4SLinus Torvalds struct vm_area_struct *vma = *vmap; 28201da177e4SLinus Torvalds unsigned long vma_start = vma->vm_start; 28211da177e4SLinus Torvalds struct mm_struct *mm = vma->vm_mm; 28221da177e4SLinus Torvalds struct vm_area_struct *new_vma, *prev; 28231da177e4SLinus Torvalds struct rb_node **rb_link, *rb_parent; 2824948f017bSAndrea Arcangeli bool faulted_in_anon_vma = true; 28251da177e4SLinus Torvalds 28261da177e4SLinus Torvalds /* 28271da177e4SLinus Torvalds * If anonymous vma has not yet been faulted, update new pgoff 28281da177e4SLinus Torvalds * to match new location, to increase its chance of merging. 28291da177e4SLinus Torvalds */ 2830948f017bSAndrea Arcangeli if (unlikely(!vma->vm_file && !vma->anon_vma)) { 28311da177e4SLinus Torvalds pgoff = addr >> PAGE_SHIFT; 2832948f017bSAndrea Arcangeli faulted_in_anon_vma = false; 2833948f017bSAndrea Arcangeli } 28341da177e4SLinus Torvalds 28356597d783SHugh Dickins if (find_vma_links(mm, addr, addr + len, &prev, &rb_link, &rb_parent)) 28366597d783SHugh Dickins return NULL; /* should never get here */ 28371da177e4SLinus Torvalds new_vma = vma_merge(mm, prev, addr, addr + len, vma->vm_flags, 28381da177e4SLinus Torvalds vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma)); 28391da177e4SLinus Torvalds if (new_vma) { 28401da177e4SLinus Torvalds /* 28411da177e4SLinus Torvalds * Source vma may have been merged into new_vma 28421da177e4SLinus Torvalds */ 2843948f017bSAndrea Arcangeli if (unlikely(vma_start >= new_vma->vm_start && 2844948f017bSAndrea Arcangeli vma_start < new_vma->vm_end)) { 2845948f017bSAndrea Arcangeli /* 2846948f017bSAndrea Arcangeli * The only way we can get a vma_merge with 2847948f017bSAndrea Arcangeli * self during an mremap is if the vma hasn't 2848948f017bSAndrea Arcangeli * been faulted in yet and we were allowed to 2849948f017bSAndrea Arcangeli * reset the dst vma->vm_pgoff to the 2850948f017bSAndrea Arcangeli * destination address of the mremap to allow 2851948f017bSAndrea Arcangeli * the merge to happen. mremap must change the 2852948f017bSAndrea Arcangeli * vm_pgoff linearity between src and dst vmas 2853948f017bSAndrea Arcangeli * (in turn preventing a vma_merge) to be 2854948f017bSAndrea Arcangeli * safe. It is only safe to keep the vm_pgoff 2855948f017bSAndrea Arcangeli * linear if there are no pages mapped yet. 2856948f017bSAndrea Arcangeli */ 285781d1b09cSSasha Levin VM_BUG_ON_VMA(faulted_in_anon_vma, new_vma); 285838a76013SMichel Lespinasse *vmap = vma = new_vma; 2859108d6642SMichel Lespinasse } 286038a76013SMichel Lespinasse *need_rmap_locks = (new_vma->vm_pgoff <= vma->vm_pgoff); 28611da177e4SLinus Torvalds } else { 2862e94b1766SChristoph Lameter new_vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL); 28631da177e4SLinus Torvalds if (new_vma) { 28641da177e4SLinus Torvalds *new_vma = *vma; 28651da177e4SLinus Torvalds new_vma->vm_start = addr; 28661da177e4SLinus Torvalds new_vma->vm_end = addr + len; 28671da177e4SLinus Torvalds new_vma->vm_pgoff = pgoff; 2868ef0855d3SOleg Nesterov if (vma_dup_policy(vma, new_vma)) 2869523d4e20SMichel Lespinasse goto out_free_vma; 2870523d4e20SMichel Lespinasse INIT_LIST_HEAD(&new_vma->anon_vma_chain); 2871523d4e20SMichel Lespinasse if (anon_vma_clone(new_vma, vma)) 2872523d4e20SMichel Lespinasse goto out_free_mempol; 2873e9714acfSKonstantin Khlebnikov if (new_vma->vm_file) 28741da177e4SLinus Torvalds get_file(new_vma->vm_file); 28751da177e4SLinus Torvalds if (new_vma->vm_ops && new_vma->vm_ops->open) 28761da177e4SLinus Torvalds new_vma->vm_ops->open(new_vma); 28771da177e4SLinus Torvalds vma_link(mm, new_vma, prev, rb_link, rb_parent); 287838a76013SMichel Lespinasse *need_rmap_locks = false; 28791da177e4SLinus Torvalds } 28801da177e4SLinus Torvalds } 28811da177e4SLinus Torvalds return new_vma; 28825beb4930SRik van Riel 28835beb4930SRik van Riel out_free_mempol: 2884ef0855d3SOleg Nesterov mpol_put(vma_policy(new_vma)); 28855beb4930SRik van Riel out_free_vma: 28865beb4930SRik van Riel kmem_cache_free(vm_area_cachep, new_vma); 28875beb4930SRik van Riel return NULL; 28881da177e4SLinus Torvalds } 2889119f657cSakpm@osdl.org 2890119f657cSakpm@osdl.org /* 2891119f657cSakpm@osdl.org * Return true if the calling process may expand its vm space by the passed 2892119f657cSakpm@osdl.org * number of pages 2893119f657cSakpm@osdl.org */ 2894119f657cSakpm@osdl.org int may_expand_vm(struct mm_struct *mm, unsigned long npages) 2895119f657cSakpm@osdl.org { 2896119f657cSakpm@osdl.org unsigned long cur = mm->total_vm; /* pages */ 2897119f657cSakpm@osdl.org unsigned long lim; 2898119f657cSakpm@osdl.org 289959e99e5bSJiri Slaby lim = rlimit(RLIMIT_AS) >> PAGE_SHIFT; 2900119f657cSakpm@osdl.org 2901119f657cSakpm@osdl.org if (cur + npages > lim) 2902119f657cSakpm@osdl.org return 0; 2903119f657cSakpm@osdl.org return 1; 2904119f657cSakpm@osdl.org } 2905fa5dc22fSRoland McGrath 2906a62c34bdSAndy Lutomirski static int special_mapping_fault(struct vm_area_struct *vma, 2907a62c34bdSAndy Lutomirski struct vm_fault *vmf); 2908a62c34bdSAndy Lutomirski 2909a62c34bdSAndy Lutomirski /* 2910a62c34bdSAndy Lutomirski * Having a close hook prevents vma merging regardless of flags. 2911a62c34bdSAndy Lutomirski */ 2912a62c34bdSAndy Lutomirski static void special_mapping_close(struct vm_area_struct *vma) 2913a62c34bdSAndy Lutomirski { 2914a62c34bdSAndy Lutomirski } 2915a62c34bdSAndy Lutomirski 2916a62c34bdSAndy Lutomirski static const char *special_mapping_name(struct vm_area_struct *vma) 2917a62c34bdSAndy Lutomirski { 2918a62c34bdSAndy Lutomirski return ((struct vm_special_mapping *)vma->vm_private_data)->name; 2919a62c34bdSAndy Lutomirski } 2920a62c34bdSAndy Lutomirski 2921a62c34bdSAndy Lutomirski static const struct vm_operations_struct special_mapping_vmops = { 2922a62c34bdSAndy Lutomirski .close = special_mapping_close, 2923a62c34bdSAndy Lutomirski .fault = special_mapping_fault, 2924a62c34bdSAndy Lutomirski .name = special_mapping_name, 2925a62c34bdSAndy Lutomirski }; 2926a62c34bdSAndy Lutomirski 2927a62c34bdSAndy Lutomirski static const struct vm_operations_struct legacy_special_mapping_vmops = { 2928a62c34bdSAndy Lutomirski .close = special_mapping_close, 2929a62c34bdSAndy Lutomirski .fault = special_mapping_fault, 2930a62c34bdSAndy Lutomirski }; 2931fa5dc22fSRoland McGrath 2932b1d0e4f5SNick Piggin static int special_mapping_fault(struct vm_area_struct *vma, 2933b1d0e4f5SNick Piggin struct vm_fault *vmf) 2934fa5dc22fSRoland McGrath { 2935b1d0e4f5SNick Piggin pgoff_t pgoff; 2936fa5dc22fSRoland McGrath struct page **pages; 2937fa5dc22fSRoland McGrath 2938b1d0e4f5SNick Piggin /* 2939b1d0e4f5SNick Piggin * special mappings have no vm_file, and in that case, the mm 2940b1d0e4f5SNick Piggin * uses vm_pgoff internally. So we have to subtract it from here. 2941b1d0e4f5SNick Piggin * We are allowed to do this because we are the mm; do not copy 2942b1d0e4f5SNick Piggin * this code into drivers! 2943b1d0e4f5SNick Piggin */ 2944b1d0e4f5SNick Piggin pgoff = vmf->pgoff - vma->vm_pgoff; 2945fa5dc22fSRoland McGrath 2946a62c34bdSAndy Lutomirski if (vma->vm_ops == &legacy_special_mapping_vmops) 2947a62c34bdSAndy Lutomirski pages = vma->vm_private_data; 2948a62c34bdSAndy Lutomirski else 2949a62c34bdSAndy Lutomirski pages = ((struct vm_special_mapping *)vma->vm_private_data)-> 2950a62c34bdSAndy Lutomirski pages; 2951a62c34bdSAndy Lutomirski 2952a62c34bdSAndy Lutomirski for (; pgoff && *pages; ++pages) 2953b1d0e4f5SNick Piggin pgoff--; 2954fa5dc22fSRoland McGrath 2955fa5dc22fSRoland McGrath if (*pages) { 2956fa5dc22fSRoland McGrath struct page *page = *pages; 2957fa5dc22fSRoland McGrath get_page(page); 2958b1d0e4f5SNick Piggin vmf->page = page; 2959b1d0e4f5SNick Piggin return 0; 2960fa5dc22fSRoland McGrath } 2961fa5dc22fSRoland McGrath 2962b1d0e4f5SNick Piggin return VM_FAULT_SIGBUS; 2963fa5dc22fSRoland McGrath } 2964fa5dc22fSRoland McGrath 2965a62c34bdSAndy Lutomirski static struct vm_area_struct *__install_special_mapping( 2966a62c34bdSAndy Lutomirski struct mm_struct *mm, 2967fa5dc22fSRoland McGrath unsigned long addr, unsigned long len, 2968a62c34bdSAndy Lutomirski unsigned long vm_flags, const struct vm_operations_struct *ops, 2969a62c34bdSAndy Lutomirski void *priv) 2970fa5dc22fSRoland McGrath { 2971462e635eSTavis Ormandy int ret; 2972fa5dc22fSRoland McGrath struct vm_area_struct *vma; 2973fa5dc22fSRoland McGrath 2974fa5dc22fSRoland McGrath vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL); 2975fa5dc22fSRoland McGrath if (unlikely(vma == NULL)) 29763935ed6aSStefani Seibold return ERR_PTR(-ENOMEM); 2977fa5dc22fSRoland McGrath 29785beb4930SRik van Riel INIT_LIST_HEAD(&vma->anon_vma_chain); 2979fa5dc22fSRoland McGrath vma->vm_mm = mm; 2980fa5dc22fSRoland McGrath vma->vm_start = addr; 2981fa5dc22fSRoland McGrath vma->vm_end = addr + len; 2982fa5dc22fSRoland McGrath 2983d9104d1cSCyrill Gorcunov vma->vm_flags = vm_flags | mm->def_flags | VM_DONTEXPAND | VM_SOFTDIRTY; 29843ed75eb8SColy Li vma->vm_page_prot = vm_get_page_prot(vma->vm_flags); 2985fa5dc22fSRoland McGrath 2986a62c34bdSAndy Lutomirski vma->vm_ops = ops; 2987a62c34bdSAndy Lutomirski vma->vm_private_data = priv; 2988fa5dc22fSRoland McGrath 2989462e635eSTavis Ormandy ret = insert_vm_struct(mm, vma); 2990462e635eSTavis Ormandy if (ret) 2991462e635eSTavis Ormandy goto out; 2992fa5dc22fSRoland McGrath 2993fa5dc22fSRoland McGrath mm->total_vm += len >> PAGE_SHIFT; 2994fa5dc22fSRoland McGrath 2995cdd6c482SIngo Molnar perf_event_mmap(vma); 2996089dd79dSPeter Zijlstra 29973935ed6aSStefani Seibold return vma; 2998462e635eSTavis Ormandy 2999462e635eSTavis Ormandy out: 3000462e635eSTavis Ormandy kmem_cache_free(vm_area_cachep, vma); 30013935ed6aSStefani Seibold return ERR_PTR(ret); 30023935ed6aSStefani Seibold } 30033935ed6aSStefani Seibold 3004a62c34bdSAndy Lutomirski /* 3005a62c34bdSAndy Lutomirski * Called with mm->mmap_sem held for writing. 3006a62c34bdSAndy Lutomirski * Insert a new vma covering the given region, with the given flags. 3007a62c34bdSAndy Lutomirski * Its pages are supplied by the given array of struct page *. 3008a62c34bdSAndy Lutomirski * The array can be shorter than len >> PAGE_SHIFT if it's null-terminated. 3009a62c34bdSAndy Lutomirski * The region past the last page supplied will always produce SIGBUS. 3010a62c34bdSAndy Lutomirski * The array pointer and the pages it points to are assumed to stay alive 3011a62c34bdSAndy Lutomirski * for as long as this mapping might exist. 3012a62c34bdSAndy Lutomirski */ 3013a62c34bdSAndy Lutomirski struct vm_area_struct *_install_special_mapping( 3014a62c34bdSAndy Lutomirski struct mm_struct *mm, 3015a62c34bdSAndy Lutomirski unsigned long addr, unsigned long len, 3016a62c34bdSAndy Lutomirski unsigned long vm_flags, const struct vm_special_mapping *spec) 3017a62c34bdSAndy Lutomirski { 3018a62c34bdSAndy Lutomirski return __install_special_mapping(mm, addr, len, vm_flags, 3019a62c34bdSAndy Lutomirski &special_mapping_vmops, (void *)spec); 3020a62c34bdSAndy Lutomirski } 3021a62c34bdSAndy Lutomirski 30223935ed6aSStefani Seibold int install_special_mapping(struct mm_struct *mm, 30233935ed6aSStefani Seibold unsigned long addr, unsigned long len, 30243935ed6aSStefani Seibold unsigned long vm_flags, struct page **pages) 30253935ed6aSStefani Seibold { 3026a62c34bdSAndy Lutomirski struct vm_area_struct *vma = __install_special_mapping( 3027a62c34bdSAndy Lutomirski mm, addr, len, vm_flags, &legacy_special_mapping_vmops, 3028a62c34bdSAndy Lutomirski (void *)pages); 30293935ed6aSStefani Seibold 303014bd5b45SDuan Jiong return PTR_ERR_OR_ZERO(vma); 3031fa5dc22fSRoland McGrath } 30327906d00cSAndrea Arcangeli 30337906d00cSAndrea Arcangeli static DEFINE_MUTEX(mm_all_locks_mutex); 30347906d00cSAndrea Arcangeli 3035454ed842SPeter Zijlstra static void vm_lock_anon_vma(struct mm_struct *mm, struct anon_vma *anon_vma) 30367906d00cSAndrea Arcangeli { 3037bf181b9fSMichel Lespinasse if (!test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_node)) { 30387906d00cSAndrea Arcangeli /* 30397906d00cSAndrea Arcangeli * The LSB of head.next can't change from under us 30407906d00cSAndrea Arcangeli * because we hold the mm_all_locks_mutex. 30417906d00cSAndrea Arcangeli */ 3042572043c9SJiri Kosina down_write_nest_lock(&anon_vma->root->rwsem, &mm->mmap_sem); 30437906d00cSAndrea Arcangeli /* 30447906d00cSAndrea Arcangeli * We can safely modify head.next after taking the 30455a505085SIngo Molnar * anon_vma->root->rwsem. If some other vma in this mm shares 30467906d00cSAndrea Arcangeli * the same anon_vma we won't take it again. 30477906d00cSAndrea Arcangeli * 30487906d00cSAndrea Arcangeli * No need of atomic instructions here, head.next 30497906d00cSAndrea Arcangeli * can't change from under us thanks to the 30505a505085SIngo Molnar * anon_vma->root->rwsem. 30517906d00cSAndrea Arcangeli */ 30527906d00cSAndrea Arcangeli if (__test_and_set_bit(0, (unsigned long *) 3053bf181b9fSMichel Lespinasse &anon_vma->root->rb_root.rb_node)) 30547906d00cSAndrea Arcangeli BUG(); 30557906d00cSAndrea Arcangeli } 30567906d00cSAndrea Arcangeli } 30577906d00cSAndrea Arcangeli 3058454ed842SPeter Zijlstra static void vm_lock_mapping(struct mm_struct *mm, struct address_space *mapping) 30597906d00cSAndrea Arcangeli { 30607906d00cSAndrea Arcangeli if (!test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) { 30617906d00cSAndrea Arcangeli /* 30627906d00cSAndrea Arcangeli * AS_MM_ALL_LOCKS can't change from under us because 30637906d00cSAndrea Arcangeli * we hold the mm_all_locks_mutex. 30647906d00cSAndrea Arcangeli * 30657906d00cSAndrea Arcangeli * Operations on ->flags have to be atomic because 30667906d00cSAndrea Arcangeli * even if AS_MM_ALL_LOCKS is stable thanks to the 30677906d00cSAndrea Arcangeli * mm_all_locks_mutex, there may be other cpus 30687906d00cSAndrea Arcangeli * changing other bitflags in parallel to us. 30697906d00cSAndrea Arcangeli */ 30707906d00cSAndrea Arcangeli if (test_and_set_bit(AS_MM_ALL_LOCKS, &mapping->flags)) 30717906d00cSAndrea Arcangeli BUG(); 30723d48ae45SPeter Zijlstra mutex_lock_nest_lock(&mapping->i_mmap_mutex, &mm->mmap_sem); 30737906d00cSAndrea Arcangeli } 30747906d00cSAndrea Arcangeli } 30757906d00cSAndrea Arcangeli 30767906d00cSAndrea Arcangeli /* 30777906d00cSAndrea Arcangeli * This operation locks against the VM for all pte/vma/mm related 30787906d00cSAndrea Arcangeli * operations that could ever happen on a certain mm. This includes 30797906d00cSAndrea Arcangeli * vmtruncate, try_to_unmap, and all page faults. 30807906d00cSAndrea Arcangeli * 30817906d00cSAndrea Arcangeli * The caller must take the mmap_sem in write mode before calling 30827906d00cSAndrea Arcangeli * mm_take_all_locks(). The caller isn't allowed to release the 30837906d00cSAndrea Arcangeli * mmap_sem until mm_drop_all_locks() returns. 30847906d00cSAndrea Arcangeli * 30857906d00cSAndrea Arcangeli * mmap_sem in write mode is required in order to block all operations 30867906d00cSAndrea Arcangeli * that could modify pagetables and free pages without need of 30877906d00cSAndrea Arcangeli * altering the vma layout (for example populate_range() with 30887906d00cSAndrea Arcangeli * nonlinear vmas). It's also needed in write mode to avoid new 30897906d00cSAndrea Arcangeli * anon_vmas to be associated with existing vmas. 30907906d00cSAndrea Arcangeli * 30917906d00cSAndrea Arcangeli * A single task can't take more than one mm_take_all_locks() in a row 30927906d00cSAndrea Arcangeli * or it would deadlock. 30937906d00cSAndrea Arcangeli * 3094bf181b9fSMichel Lespinasse * The LSB in anon_vma->rb_root.rb_node and the AS_MM_ALL_LOCKS bitflag in 30957906d00cSAndrea Arcangeli * mapping->flags avoid to take the same lock twice, if more than one 30967906d00cSAndrea Arcangeli * vma in this mm is backed by the same anon_vma or address_space. 30977906d00cSAndrea Arcangeli * 30987906d00cSAndrea Arcangeli * We can take all the locks in random order because the VM code 3099631b0cfdSYuanhan Liu * taking i_mmap_mutex or anon_vma->rwsem outside the mmap_sem never 31007906d00cSAndrea Arcangeli * takes more than one of them in a row. Secondly we're protected 31017906d00cSAndrea Arcangeli * against a concurrent mm_take_all_locks() by the mm_all_locks_mutex. 31027906d00cSAndrea Arcangeli * 31037906d00cSAndrea Arcangeli * mm_take_all_locks() and mm_drop_all_locks are expensive operations 31047906d00cSAndrea Arcangeli * that may have to take thousand of locks. 31057906d00cSAndrea Arcangeli * 31067906d00cSAndrea Arcangeli * mm_take_all_locks() can fail if it's interrupted by signals. 31077906d00cSAndrea Arcangeli */ 31087906d00cSAndrea Arcangeli int mm_take_all_locks(struct mm_struct *mm) 31097906d00cSAndrea Arcangeli { 31107906d00cSAndrea Arcangeli struct vm_area_struct *vma; 31115beb4930SRik van Riel struct anon_vma_chain *avc; 31127906d00cSAndrea Arcangeli 31137906d00cSAndrea Arcangeli BUG_ON(down_read_trylock(&mm->mmap_sem)); 31147906d00cSAndrea Arcangeli 31157906d00cSAndrea Arcangeli mutex_lock(&mm_all_locks_mutex); 31167906d00cSAndrea Arcangeli 31177906d00cSAndrea Arcangeli for (vma = mm->mmap; vma; vma = vma->vm_next) { 31187906d00cSAndrea Arcangeli if (signal_pending(current)) 31197906d00cSAndrea Arcangeli goto out_unlock; 31207906d00cSAndrea Arcangeli if (vma->vm_file && vma->vm_file->f_mapping) 3121454ed842SPeter Zijlstra vm_lock_mapping(mm, vma->vm_file->f_mapping); 31227906d00cSAndrea Arcangeli } 31237cd5a02fSPeter Zijlstra 31247cd5a02fSPeter Zijlstra for (vma = mm->mmap; vma; vma = vma->vm_next) { 31257cd5a02fSPeter Zijlstra if (signal_pending(current)) 31267cd5a02fSPeter Zijlstra goto out_unlock; 31277cd5a02fSPeter Zijlstra if (vma->anon_vma) 31285beb4930SRik van Riel list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) 31295beb4930SRik van Riel vm_lock_anon_vma(mm, avc->anon_vma); 31307cd5a02fSPeter Zijlstra } 31317cd5a02fSPeter Zijlstra 3132584cff54SKautuk Consul return 0; 31337906d00cSAndrea Arcangeli 31347906d00cSAndrea Arcangeli out_unlock: 31357906d00cSAndrea Arcangeli mm_drop_all_locks(mm); 3136584cff54SKautuk Consul return -EINTR; 31377906d00cSAndrea Arcangeli } 31387906d00cSAndrea Arcangeli 31397906d00cSAndrea Arcangeli static void vm_unlock_anon_vma(struct anon_vma *anon_vma) 31407906d00cSAndrea Arcangeli { 3141bf181b9fSMichel Lespinasse if (test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_node)) { 31427906d00cSAndrea Arcangeli /* 31437906d00cSAndrea Arcangeli * The LSB of head.next can't change to 0 from under 31447906d00cSAndrea Arcangeli * us because we hold the mm_all_locks_mutex. 31457906d00cSAndrea Arcangeli * 31467906d00cSAndrea Arcangeli * We must however clear the bitflag before unlocking 3147bf181b9fSMichel Lespinasse * the vma so the users using the anon_vma->rb_root will 31487906d00cSAndrea Arcangeli * never see our bitflag. 31497906d00cSAndrea Arcangeli * 31507906d00cSAndrea Arcangeli * No need of atomic instructions here, head.next 31517906d00cSAndrea Arcangeli * can't change from under us until we release the 31525a505085SIngo Molnar * anon_vma->root->rwsem. 31537906d00cSAndrea Arcangeli */ 31547906d00cSAndrea Arcangeli if (!__test_and_clear_bit(0, (unsigned long *) 3155bf181b9fSMichel Lespinasse &anon_vma->root->rb_root.rb_node)) 31567906d00cSAndrea Arcangeli BUG(); 315708b52706SKonstantin Khlebnikov anon_vma_unlock_write(anon_vma); 31587906d00cSAndrea Arcangeli } 31597906d00cSAndrea Arcangeli } 31607906d00cSAndrea Arcangeli 31617906d00cSAndrea Arcangeli static void vm_unlock_mapping(struct address_space *mapping) 31627906d00cSAndrea Arcangeli { 31637906d00cSAndrea Arcangeli if (test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) { 31647906d00cSAndrea Arcangeli /* 31657906d00cSAndrea Arcangeli * AS_MM_ALL_LOCKS can't change to 0 from under us 31667906d00cSAndrea Arcangeli * because we hold the mm_all_locks_mutex. 31677906d00cSAndrea Arcangeli */ 31683d48ae45SPeter Zijlstra mutex_unlock(&mapping->i_mmap_mutex); 31697906d00cSAndrea Arcangeli if (!test_and_clear_bit(AS_MM_ALL_LOCKS, 31707906d00cSAndrea Arcangeli &mapping->flags)) 31717906d00cSAndrea Arcangeli BUG(); 31727906d00cSAndrea Arcangeli } 31737906d00cSAndrea Arcangeli } 31747906d00cSAndrea Arcangeli 31757906d00cSAndrea Arcangeli /* 31767906d00cSAndrea Arcangeli * The mmap_sem cannot be released by the caller until 31777906d00cSAndrea Arcangeli * mm_drop_all_locks() returns. 31787906d00cSAndrea Arcangeli */ 31797906d00cSAndrea Arcangeli void mm_drop_all_locks(struct mm_struct *mm) 31807906d00cSAndrea Arcangeli { 31817906d00cSAndrea Arcangeli struct vm_area_struct *vma; 31825beb4930SRik van Riel struct anon_vma_chain *avc; 31837906d00cSAndrea Arcangeli 31847906d00cSAndrea Arcangeli BUG_ON(down_read_trylock(&mm->mmap_sem)); 31857906d00cSAndrea Arcangeli BUG_ON(!mutex_is_locked(&mm_all_locks_mutex)); 31867906d00cSAndrea Arcangeli 31877906d00cSAndrea Arcangeli for (vma = mm->mmap; vma; vma = vma->vm_next) { 31887906d00cSAndrea Arcangeli if (vma->anon_vma) 31895beb4930SRik van Riel list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) 31905beb4930SRik van Riel vm_unlock_anon_vma(avc->anon_vma); 31917906d00cSAndrea Arcangeli if (vma->vm_file && vma->vm_file->f_mapping) 31927906d00cSAndrea Arcangeli vm_unlock_mapping(vma->vm_file->f_mapping); 31937906d00cSAndrea Arcangeli } 31947906d00cSAndrea Arcangeli 31957906d00cSAndrea Arcangeli mutex_unlock(&mm_all_locks_mutex); 31967906d00cSAndrea Arcangeli } 31978feae131SDavid Howells 31988feae131SDavid Howells /* 31998feae131SDavid Howells * initialise the VMA slab 32008feae131SDavid Howells */ 32018feae131SDavid Howells void __init mmap_init(void) 32028feae131SDavid Howells { 320300a62ce9SKOSAKI Motohiro int ret; 320400a62ce9SKOSAKI Motohiro 320500a62ce9SKOSAKI Motohiro ret = percpu_counter_init(&vm_committed_as, 0); 320600a62ce9SKOSAKI Motohiro VM_BUG_ON(ret); 32078feae131SDavid Howells } 3208c9b1d098SAndrew Shewmaker 3209c9b1d098SAndrew Shewmaker /* 3210c9b1d098SAndrew Shewmaker * Initialise sysctl_user_reserve_kbytes. 3211c9b1d098SAndrew Shewmaker * 3212c9b1d098SAndrew Shewmaker * This is intended to prevent a user from starting a single memory hogging 3213c9b1d098SAndrew Shewmaker * process, such that they cannot recover (kill the hog) in OVERCOMMIT_NEVER 3214c9b1d098SAndrew Shewmaker * mode. 3215c9b1d098SAndrew Shewmaker * 3216c9b1d098SAndrew Shewmaker * The default value is min(3% of free memory, 128MB) 3217c9b1d098SAndrew Shewmaker * 128MB is enough to recover with sshd/login, bash, and top/kill. 3218c9b1d098SAndrew Shewmaker */ 32191640879aSAndrew Shewmaker static int init_user_reserve(void) 3220c9b1d098SAndrew Shewmaker { 3221c9b1d098SAndrew Shewmaker unsigned long free_kbytes; 3222c9b1d098SAndrew Shewmaker 3223c9b1d098SAndrew Shewmaker free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10); 3224c9b1d098SAndrew Shewmaker 3225c9b1d098SAndrew Shewmaker sysctl_user_reserve_kbytes = min(free_kbytes / 32, 1UL << 17); 3226c9b1d098SAndrew Shewmaker return 0; 3227c9b1d098SAndrew Shewmaker } 3228a64fb3cdSPaul Gortmaker subsys_initcall(init_user_reserve); 32294eeab4f5SAndrew Shewmaker 32304eeab4f5SAndrew Shewmaker /* 32314eeab4f5SAndrew Shewmaker * Initialise sysctl_admin_reserve_kbytes. 32324eeab4f5SAndrew Shewmaker * 32334eeab4f5SAndrew Shewmaker * The purpose of sysctl_admin_reserve_kbytes is to allow the sys admin 32344eeab4f5SAndrew Shewmaker * to log in and kill a memory hogging process. 32354eeab4f5SAndrew Shewmaker * 32364eeab4f5SAndrew Shewmaker * Systems with more than 256MB will reserve 8MB, enough to recover 32374eeab4f5SAndrew Shewmaker * with sshd, bash, and top in OVERCOMMIT_GUESS. Smaller systems will 32384eeab4f5SAndrew Shewmaker * only reserve 3% of free pages by default. 32394eeab4f5SAndrew Shewmaker */ 32401640879aSAndrew Shewmaker static int init_admin_reserve(void) 32414eeab4f5SAndrew Shewmaker { 32424eeab4f5SAndrew Shewmaker unsigned long free_kbytes; 32434eeab4f5SAndrew Shewmaker 32444eeab4f5SAndrew Shewmaker free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10); 32454eeab4f5SAndrew Shewmaker 32464eeab4f5SAndrew Shewmaker sysctl_admin_reserve_kbytes = min(free_kbytes / 32, 1UL << 13); 32474eeab4f5SAndrew Shewmaker return 0; 32484eeab4f5SAndrew Shewmaker } 3249a64fb3cdSPaul Gortmaker subsys_initcall(init_admin_reserve); 32501640879aSAndrew Shewmaker 32511640879aSAndrew Shewmaker /* 32521640879aSAndrew Shewmaker * Reinititalise user and admin reserves if memory is added or removed. 32531640879aSAndrew Shewmaker * 32541640879aSAndrew Shewmaker * The default user reserve max is 128MB, and the default max for the 32551640879aSAndrew Shewmaker * admin reserve is 8MB. These are usually, but not always, enough to 32561640879aSAndrew Shewmaker * enable recovery from a memory hogging process using login/sshd, a shell, 32571640879aSAndrew Shewmaker * and tools like top. It may make sense to increase or even disable the 32581640879aSAndrew Shewmaker * reserve depending on the existence of swap or variations in the recovery 32591640879aSAndrew Shewmaker * tools. So, the admin may have changed them. 32601640879aSAndrew Shewmaker * 32611640879aSAndrew Shewmaker * If memory is added and the reserves have been eliminated or increased above 32621640879aSAndrew Shewmaker * the default max, then we'll trust the admin. 32631640879aSAndrew Shewmaker * 32641640879aSAndrew Shewmaker * If memory is removed and there isn't enough free memory, then we 32651640879aSAndrew Shewmaker * need to reset the reserves. 32661640879aSAndrew Shewmaker * 32671640879aSAndrew Shewmaker * Otherwise keep the reserve set by the admin. 32681640879aSAndrew Shewmaker */ 32691640879aSAndrew Shewmaker static int reserve_mem_notifier(struct notifier_block *nb, 32701640879aSAndrew Shewmaker unsigned long action, void *data) 32711640879aSAndrew Shewmaker { 32721640879aSAndrew Shewmaker unsigned long tmp, free_kbytes; 32731640879aSAndrew Shewmaker 32741640879aSAndrew Shewmaker switch (action) { 32751640879aSAndrew Shewmaker case MEM_ONLINE: 32761640879aSAndrew Shewmaker /* Default max is 128MB. Leave alone if modified by operator. */ 32771640879aSAndrew Shewmaker tmp = sysctl_user_reserve_kbytes; 32781640879aSAndrew Shewmaker if (0 < tmp && tmp < (1UL << 17)) 32791640879aSAndrew Shewmaker init_user_reserve(); 32801640879aSAndrew Shewmaker 32811640879aSAndrew Shewmaker /* Default max is 8MB. Leave alone if modified by operator. */ 32821640879aSAndrew Shewmaker tmp = sysctl_admin_reserve_kbytes; 32831640879aSAndrew Shewmaker if (0 < tmp && tmp < (1UL << 13)) 32841640879aSAndrew Shewmaker init_admin_reserve(); 32851640879aSAndrew Shewmaker 32861640879aSAndrew Shewmaker break; 32871640879aSAndrew Shewmaker case MEM_OFFLINE: 32881640879aSAndrew Shewmaker free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10); 32891640879aSAndrew Shewmaker 32901640879aSAndrew Shewmaker if (sysctl_user_reserve_kbytes > free_kbytes) { 32911640879aSAndrew Shewmaker init_user_reserve(); 32921640879aSAndrew Shewmaker pr_info("vm.user_reserve_kbytes reset to %lu\n", 32931640879aSAndrew Shewmaker sysctl_user_reserve_kbytes); 32941640879aSAndrew Shewmaker } 32951640879aSAndrew Shewmaker 32961640879aSAndrew Shewmaker if (sysctl_admin_reserve_kbytes > free_kbytes) { 32971640879aSAndrew Shewmaker init_admin_reserve(); 32981640879aSAndrew Shewmaker pr_info("vm.admin_reserve_kbytes reset to %lu\n", 32991640879aSAndrew Shewmaker sysctl_admin_reserve_kbytes); 33001640879aSAndrew Shewmaker } 33011640879aSAndrew Shewmaker break; 33021640879aSAndrew Shewmaker default: 33031640879aSAndrew Shewmaker break; 33041640879aSAndrew Shewmaker } 33051640879aSAndrew Shewmaker return NOTIFY_OK; 33061640879aSAndrew Shewmaker } 33071640879aSAndrew Shewmaker 33081640879aSAndrew Shewmaker static struct notifier_block reserve_mem_nb = { 33091640879aSAndrew Shewmaker .notifier_call = reserve_mem_notifier, 33101640879aSAndrew Shewmaker }; 33111640879aSAndrew Shewmaker 33121640879aSAndrew Shewmaker static int __meminit init_reserve_notifier(void) 33131640879aSAndrew Shewmaker { 33141640879aSAndrew Shewmaker if (register_hotmemory_notifier(&reserve_mem_nb)) 3315b1de0d13SMitchel Humpherys pr_err("Failed registering memory add/remove notifier for admin reserve\n"); 33161640879aSAndrew Shewmaker 33171640879aSAndrew Shewmaker return 0; 33181640879aSAndrew Shewmaker } 3319a64fb3cdSPaul Gortmaker subsys_initcall(init_reserve_notifier); 3320