11da177e4SLinus Torvalds /* 21da177e4SLinus Torvalds * mm/mmap.c 31da177e4SLinus Torvalds * 41da177e4SLinus Torvalds * Written by obz. 51da177e4SLinus Torvalds * 6046c6884SAlan Cox * Address space accounting code <alan@lxorguk.ukuu.org.uk> 71da177e4SLinus Torvalds */ 81da177e4SLinus Torvalds 9b1de0d13SMitchel Humpherys #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt 10b1de0d13SMitchel Humpherys 11e8420a8eSCyril Hrubis #include <linux/kernel.h> 121da177e4SLinus Torvalds #include <linux/slab.h> 134af3c9ccSAlexey Dobriyan #include <linux/backing-dev.h> 141da177e4SLinus Torvalds #include <linux/mm.h> 15615d6e87SDavidlohr Bueso #include <linux/vmacache.h> 161da177e4SLinus Torvalds #include <linux/shm.h> 171da177e4SLinus Torvalds #include <linux/mman.h> 181da177e4SLinus Torvalds #include <linux/pagemap.h> 191da177e4SLinus Torvalds #include <linux/swap.h> 201da177e4SLinus Torvalds #include <linux/syscalls.h> 21c59ede7bSRandy.Dunlap #include <linux/capability.h> 221da177e4SLinus Torvalds #include <linux/init.h> 231da177e4SLinus Torvalds #include <linux/file.h> 241da177e4SLinus Torvalds #include <linux/fs.h> 251da177e4SLinus Torvalds #include <linux/personality.h> 261da177e4SLinus Torvalds #include <linux/security.h> 271da177e4SLinus Torvalds #include <linux/hugetlb.h> 281da177e4SLinus Torvalds #include <linux/profile.h> 29b95f1b31SPaul Gortmaker #include <linux/export.h> 301da177e4SLinus Torvalds #include <linux/mount.h> 311da177e4SLinus Torvalds #include <linux/mempolicy.h> 321da177e4SLinus Torvalds #include <linux/rmap.h> 33cddb8a5cSAndrea Arcangeli #include <linux/mmu_notifier.h> 3482f71ae4SKonstantin Khlebnikov #include <linux/mmdebug.h> 35cdd6c482SIngo Molnar #include <linux/perf_event.h> 36120a795dSAl Viro #include <linux/audit.h> 37b15d00b6SAndrea Arcangeli #include <linux/khugepaged.h> 382b144498SSrikar Dronamraju #include <linux/uprobes.h> 39d3737187SMichel Lespinasse #include <linux/rbtree_augmented.h> 40cf4aebc2SClark Williams #include <linux/sched/sysctl.h> 411640879aSAndrew Shewmaker #include <linux/notifier.h> 421640879aSAndrew Shewmaker #include <linux/memory.h> 43b1de0d13SMitchel Humpherys #include <linux/printk.h> 4419a809afSAndrea Arcangeli #include <linux/userfaultfd_k.h> 451da177e4SLinus Torvalds 461da177e4SLinus Torvalds #include <asm/uaccess.h> 471da177e4SLinus Torvalds #include <asm/cacheflush.h> 481da177e4SLinus Torvalds #include <asm/tlb.h> 49d6dd61c8SJeremy Fitzhardinge #include <asm/mmu_context.h> 501da177e4SLinus Torvalds 5142b77728SJan Beulich #include "internal.h" 5242b77728SJan Beulich 533a459756SKirill Korotaev #ifndef arch_mmap_check 543a459756SKirill Korotaev #define arch_mmap_check(addr, len, flags) (0) 553a459756SKirill Korotaev #endif 563a459756SKirill Korotaev 5708e7d9b5SMartin Schwidefsky #ifndef arch_rebalance_pgtables 5808e7d9b5SMartin Schwidefsky #define arch_rebalance_pgtables(addr, len) (addr) 5908e7d9b5SMartin Schwidefsky #endif 6008e7d9b5SMartin Schwidefsky 61*d07e2259SDaniel Cashman #ifdef CONFIG_HAVE_ARCH_MMAP_RND_BITS 62*d07e2259SDaniel Cashman const int mmap_rnd_bits_min = CONFIG_ARCH_MMAP_RND_BITS_MIN; 63*d07e2259SDaniel Cashman const int mmap_rnd_bits_max = CONFIG_ARCH_MMAP_RND_BITS_MAX; 64*d07e2259SDaniel Cashman int mmap_rnd_bits __read_mostly = CONFIG_ARCH_MMAP_RND_BITS; 65*d07e2259SDaniel Cashman #endif 66*d07e2259SDaniel Cashman #ifdef CONFIG_HAVE_ARCH_MMAP_RND_COMPAT_BITS 67*d07e2259SDaniel Cashman const int mmap_rnd_compat_bits_min = CONFIG_ARCH_MMAP_RND_COMPAT_BITS_MIN; 68*d07e2259SDaniel Cashman const int mmap_rnd_compat_bits_max = CONFIG_ARCH_MMAP_RND_COMPAT_BITS_MAX; 69*d07e2259SDaniel Cashman int mmap_rnd_compat_bits __read_mostly = CONFIG_ARCH_MMAP_RND_COMPAT_BITS; 70*d07e2259SDaniel Cashman #endif 71*d07e2259SDaniel Cashman 72*d07e2259SDaniel Cashman 73e0da382cSHugh Dickins static void unmap_region(struct mm_struct *mm, 74e0da382cSHugh Dickins struct vm_area_struct *vma, struct vm_area_struct *prev, 75e0da382cSHugh Dickins unsigned long start, unsigned long end); 76e0da382cSHugh Dickins 771da177e4SLinus Torvalds /* description of effects of mapping type and prot in current implementation. 781da177e4SLinus Torvalds * this is due to the limited x86 page protection hardware. The expected 791da177e4SLinus Torvalds * behavior is in parens: 801da177e4SLinus Torvalds * 811da177e4SLinus Torvalds * map_type prot 821da177e4SLinus Torvalds * PROT_NONE PROT_READ PROT_WRITE PROT_EXEC 831da177e4SLinus Torvalds * MAP_SHARED r: (no) no r: (yes) yes r: (no) yes r: (no) yes 841da177e4SLinus Torvalds * w: (no) no w: (no) no w: (yes) yes w: (no) no 851da177e4SLinus Torvalds * x: (no) no x: (no) yes x: (no) yes x: (yes) yes 861da177e4SLinus Torvalds * 871da177e4SLinus Torvalds * MAP_PRIVATE r: (no) no r: (yes) yes r: (no) yes r: (no) yes 881da177e4SLinus Torvalds * w: (no) no w: (no) no w: (copy) copy w: (no) no 891da177e4SLinus Torvalds * x: (no) no x: (no) yes x: (no) yes x: (yes) yes 901da177e4SLinus Torvalds * 911da177e4SLinus Torvalds */ 921da177e4SLinus Torvalds pgprot_t protection_map[16] = { 931da177e4SLinus Torvalds __P000, __P001, __P010, __P011, __P100, __P101, __P110, __P111, 941da177e4SLinus Torvalds __S000, __S001, __S010, __S011, __S100, __S101, __S110, __S111 951da177e4SLinus Torvalds }; 961da177e4SLinus Torvalds 97804af2cfSHugh Dickins pgprot_t vm_get_page_prot(unsigned long vm_flags) 98804af2cfSHugh Dickins { 99b845f313SDave Kleikamp return __pgprot(pgprot_val(protection_map[vm_flags & 100b845f313SDave Kleikamp (VM_READ|VM_WRITE|VM_EXEC|VM_SHARED)]) | 101b845f313SDave Kleikamp pgprot_val(arch_vm_get_page_prot(vm_flags))); 102804af2cfSHugh Dickins } 103804af2cfSHugh Dickins EXPORT_SYMBOL(vm_get_page_prot); 104804af2cfSHugh Dickins 10564e45507SPeter Feiner static pgprot_t vm_pgprot_modify(pgprot_t oldprot, unsigned long vm_flags) 10664e45507SPeter Feiner { 10764e45507SPeter Feiner return pgprot_modify(oldprot, vm_get_page_prot(vm_flags)); 10864e45507SPeter Feiner } 10964e45507SPeter Feiner 11064e45507SPeter Feiner /* Update vma->vm_page_prot to reflect vma->vm_flags. */ 11164e45507SPeter Feiner void vma_set_page_prot(struct vm_area_struct *vma) 11264e45507SPeter Feiner { 11364e45507SPeter Feiner unsigned long vm_flags = vma->vm_flags; 11464e45507SPeter Feiner 11564e45507SPeter Feiner vma->vm_page_prot = vm_pgprot_modify(vma->vm_page_prot, vm_flags); 11664e45507SPeter Feiner if (vma_wants_writenotify(vma)) { 11764e45507SPeter Feiner vm_flags &= ~VM_SHARED; 11864e45507SPeter Feiner vma->vm_page_prot = vm_pgprot_modify(vma->vm_page_prot, 11964e45507SPeter Feiner vm_flags); 12064e45507SPeter Feiner } 12164e45507SPeter Feiner } 12264e45507SPeter Feiner 12364e45507SPeter Feiner 12434679d7eSShaohua Li int sysctl_overcommit_memory __read_mostly = OVERCOMMIT_GUESS; /* heuristic overcommit */ 12534679d7eSShaohua Li int sysctl_overcommit_ratio __read_mostly = 50; /* default is 50% */ 12649f0ce5fSJerome Marchand unsigned long sysctl_overcommit_kbytes __read_mostly; 127c3d8c141SChristoph Lameter int sysctl_max_map_count __read_mostly = DEFAULT_MAX_MAP_COUNT; 128c9b1d098SAndrew Shewmaker unsigned long sysctl_user_reserve_kbytes __read_mostly = 1UL << 17; /* 128MB */ 1294eeab4f5SAndrew Shewmaker unsigned long sysctl_admin_reserve_kbytes __read_mostly = 1UL << 13; /* 8MB */ 13034679d7eSShaohua Li /* 13134679d7eSShaohua Li * Make sure vm_committed_as in one cacheline and not cacheline shared with 13234679d7eSShaohua Li * other variables. It can be updated by several CPUs frequently. 13334679d7eSShaohua Li */ 13434679d7eSShaohua Li struct percpu_counter vm_committed_as ____cacheline_aligned_in_smp; 1351da177e4SLinus Torvalds 1361da177e4SLinus Torvalds /* 137997071bcSK. Y. Srinivasan * The global memory commitment made in the system can be a metric 138997071bcSK. Y. Srinivasan * that can be used to drive ballooning decisions when Linux is hosted 139997071bcSK. Y. Srinivasan * as a guest. On Hyper-V, the host implements a policy engine for dynamically 140997071bcSK. Y. Srinivasan * balancing memory across competing virtual machines that are hosted. 141997071bcSK. Y. Srinivasan * Several metrics drive this policy engine including the guest reported 142997071bcSK. Y. Srinivasan * memory commitment. 143997071bcSK. Y. Srinivasan */ 144997071bcSK. Y. Srinivasan unsigned long vm_memory_committed(void) 145997071bcSK. Y. Srinivasan { 146997071bcSK. Y. Srinivasan return percpu_counter_read_positive(&vm_committed_as); 147997071bcSK. Y. Srinivasan } 148997071bcSK. Y. Srinivasan EXPORT_SYMBOL_GPL(vm_memory_committed); 149997071bcSK. Y. Srinivasan 150997071bcSK. Y. Srinivasan /* 1511da177e4SLinus Torvalds * Check that a process has enough memory to allocate a new virtual 1521da177e4SLinus Torvalds * mapping. 0 means there is enough memory for the allocation to 1531da177e4SLinus Torvalds * succeed and -ENOMEM implies there is not. 1541da177e4SLinus Torvalds * 1551da177e4SLinus Torvalds * We currently support three overcommit policies, which are set via the 1561da177e4SLinus Torvalds * vm.overcommit_memory sysctl. See Documentation/vm/overcommit-accounting 1571da177e4SLinus Torvalds * 1581da177e4SLinus Torvalds * Strict overcommit modes added 2002 Feb 26 by Alan Cox. 1591da177e4SLinus Torvalds * Additional code 2002 Jul 20 by Robert Love. 1601da177e4SLinus Torvalds * 1611da177e4SLinus Torvalds * cap_sys_admin is 1 if the process has admin privileges, 0 otherwise. 1621da177e4SLinus Torvalds * 1631da177e4SLinus Torvalds * Note this is a helper function intended to be used by LSMs which 1641da177e4SLinus Torvalds * wish to use this logic. 1651da177e4SLinus Torvalds */ 16634b4e4aaSAlan Cox int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin) 1671da177e4SLinus Torvalds { 1685703b087SRoman Gushchin long free, allowed, reserve; 1691da177e4SLinus Torvalds 17082f71ae4SKonstantin Khlebnikov VM_WARN_ONCE(percpu_counter_read(&vm_committed_as) < 17182f71ae4SKonstantin Khlebnikov -(s64)vm_committed_as_batch * num_online_cpus(), 17282f71ae4SKonstantin Khlebnikov "memory commitment underflow"); 17382f71ae4SKonstantin Khlebnikov 1741da177e4SLinus Torvalds vm_acct_memory(pages); 1751da177e4SLinus Torvalds 1761da177e4SLinus Torvalds /* 1771da177e4SLinus Torvalds * Sometimes we want to use more memory than we have 1781da177e4SLinus Torvalds */ 1791da177e4SLinus Torvalds if (sysctl_overcommit_memory == OVERCOMMIT_ALWAYS) 1801da177e4SLinus Torvalds return 0; 1811da177e4SLinus Torvalds 1821da177e4SLinus Torvalds if (sysctl_overcommit_memory == OVERCOMMIT_GUESS) { 183c15bef30SDmitry Fink free = global_page_state(NR_FREE_PAGES); 184c15bef30SDmitry Fink free += global_page_state(NR_FILE_PAGES); 1851da177e4SLinus Torvalds 186c15bef30SDmitry Fink /* 187c15bef30SDmitry Fink * shmem pages shouldn't be counted as free in this 188c15bef30SDmitry Fink * case, they can't be purged, only swapped out, and 189c15bef30SDmitry Fink * that won't affect the overall amount of available 190c15bef30SDmitry Fink * memory in the system. 191c15bef30SDmitry Fink */ 192c15bef30SDmitry Fink free -= global_page_state(NR_SHMEM); 193c15bef30SDmitry Fink 194ec8acf20SShaohua Li free += get_nr_swap_pages(); 1951da177e4SLinus Torvalds 1961da177e4SLinus Torvalds /* 1971da177e4SLinus Torvalds * Any slabs which are created with the 1981da177e4SLinus Torvalds * SLAB_RECLAIM_ACCOUNT flag claim to have contents 1991da177e4SLinus Torvalds * which are reclaimable, under pressure. The dentry 2001da177e4SLinus Torvalds * cache and most inode caches should fall into this 2011da177e4SLinus Torvalds */ 202972d1a7bSChristoph Lameter free += global_page_state(NR_SLAB_RECLAIMABLE); 2031da177e4SLinus Torvalds 2041da177e4SLinus Torvalds /* 205c15bef30SDmitry Fink * Leave reserved pages. The pages are not for anonymous pages. 206c15bef30SDmitry Fink */ 207c15bef30SDmitry Fink if (free <= totalreserve_pages) 208c15bef30SDmitry Fink goto error; 209c15bef30SDmitry Fink else 210c15bef30SDmitry Fink free -= totalreserve_pages; 211c15bef30SDmitry Fink 212c15bef30SDmitry Fink /* 2134eeab4f5SAndrew Shewmaker * Reserve some for root 2141da177e4SLinus Torvalds */ 2151da177e4SLinus Torvalds if (!cap_sys_admin) 2164eeab4f5SAndrew Shewmaker free -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10); 2171da177e4SLinus Torvalds 2181da177e4SLinus Torvalds if (free > pages) 2191da177e4SLinus Torvalds return 0; 2201da177e4SLinus Torvalds 2216d9f7839SHideo AOKI goto error; 2221da177e4SLinus Torvalds } 2231da177e4SLinus Torvalds 22400619bccSJerome Marchand allowed = vm_commit_limit(); 2251da177e4SLinus Torvalds /* 2264eeab4f5SAndrew Shewmaker * Reserve some for root 2271da177e4SLinus Torvalds */ 2281da177e4SLinus Torvalds if (!cap_sys_admin) 2294eeab4f5SAndrew Shewmaker allowed -= sysctl_admin_reserve_kbytes >> (PAGE_SHIFT - 10); 2301da177e4SLinus Torvalds 231c9b1d098SAndrew Shewmaker /* 232c9b1d098SAndrew Shewmaker * Don't let a single process grow so big a user can't recover 233c9b1d098SAndrew Shewmaker */ 234c9b1d098SAndrew Shewmaker if (mm) { 235c9b1d098SAndrew Shewmaker reserve = sysctl_user_reserve_kbytes >> (PAGE_SHIFT - 10); 2365703b087SRoman Gushchin allowed -= min_t(long, mm->total_vm / 32, reserve); 237c9b1d098SAndrew Shewmaker } 2381da177e4SLinus Torvalds 23900a62ce9SKOSAKI Motohiro if (percpu_counter_read_positive(&vm_committed_as) < allowed) 2401da177e4SLinus Torvalds return 0; 2416d9f7839SHideo AOKI error: 2421da177e4SLinus Torvalds vm_unacct_memory(pages); 2431da177e4SLinus Torvalds 2441da177e4SLinus Torvalds return -ENOMEM; 2451da177e4SLinus Torvalds } 2461da177e4SLinus Torvalds 2471da177e4SLinus Torvalds /* 248c8c06efaSDavidlohr Bueso * Requires inode->i_mapping->i_mmap_rwsem 2491da177e4SLinus Torvalds */ 2501da177e4SLinus Torvalds static void __remove_shared_vm_struct(struct vm_area_struct *vma, 2511da177e4SLinus Torvalds struct file *file, struct address_space *mapping) 2521da177e4SLinus Torvalds { 2531da177e4SLinus Torvalds if (vma->vm_flags & VM_DENYWRITE) 254496ad9aaSAl Viro atomic_inc(&file_inode(file)->i_writecount); 2551da177e4SLinus Torvalds if (vma->vm_flags & VM_SHARED) 2564bb5f5d9SDavid Herrmann mapping_unmap_writable(mapping); 2571da177e4SLinus Torvalds 2581da177e4SLinus Torvalds flush_dcache_mmap_lock(mapping); 2596b2dbba8SMichel Lespinasse vma_interval_tree_remove(vma, &mapping->i_mmap); 2601da177e4SLinus Torvalds flush_dcache_mmap_unlock(mapping); 2611da177e4SLinus Torvalds } 2621da177e4SLinus Torvalds 2631da177e4SLinus Torvalds /* 2646b2dbba8SMichel Lespinasse * Unlink a file-based vm structure from its interval tree, to hide 265a8fb5618SHugh Dickins * vma from rmap and vmtruncate before freeing its page tables. 2661da177e4SLinus Torvalds */ 267a8fb5618SHugh Dickins void unlink_file_vma(struct vm_area_struct *vma) 2681da177e4SLinus Torvalds { 2691da177e4SLinus Torvalds struct file *file = vma->vm_file; 2701da177e4SLinus Torvalds 2711da177e4SLinus Torvalds if (file) { 2721da177e4SLinus Torvalds struct address_space *mapping = file->f_mapping; 27383cde9e8SDavidlohr Bueso i_mmap_lock_write(mapping); 2741da177e4SLinus Torvalds __remove_shared_vm_struct(vma, file, mapping); 27583cde9e8SDavidlohr Bueso i_mmap_unlock_write(mapping); 2761da177e4SLinus Torvalds } 277a8fb5618SHugh Dickins } 278a8fb5618SHugh Dickins 279a8fb5618SHugh Dickins /* 280a8fb5618SHugh Dickins * Close a vm structure and free it, returning the next. 281a8fb5618SHugh Dickins */ 282a8fb5618SHugh Dickins static struct vm_area_struct *remove_vma(struct vm_area_struct *vma) 283a8fb5618SHugh Dickins { 284a8fb5618SHugh Dickins struct vm_area_struct *next = vma->vm_next; 285a8fb5618SHugh Dickins 286a8fb5618SHugh Dickins might_sleep(); 2871da177e4SLinus Torvalds if (vma->vm_ops && vma->vm_ops->close) 2881da177e4SLinus Torvalds vma->vm_ops->close(vma); 289e9714acfSKonstantin Khlebnikov if (vma->vm_file) 290a8fb5618SHugh Dickins fput(vma->vm_file); 291f0be3d32SLee Schermerhorn mpol_put(vma_policy(vma)); 2921da177e4SLinus Torvalds kmem_cache_free(vm_area_cachep, vma); 293a8fb5618SHugh Dickins return next; 2941da177e4SLinus Torvalds } 2951da177e4SLinus Torvalds 296e4eb1ff6SLinus Torvalds static unsigned long do_brk(unsigned long addr, unsigned long len); 297e4eb1ff6SLinus Torvalds 2986a6160a7SHeiko Carstens SYSCALL_DEFINE1(brk, unsigned long, brk) 2991da177e4SLinus Torvalds { 3008764b338SCyrill Gorcunov unsigned long retval; 3011da177e4SLinus Torvalds unsigned long newbrk, oldbrk; 3021da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 303a5b4592cSJiri Kosina unsigned long min_brk; 304128557ffSMichel Lespinasse bool populate; 3051da177e4SLinus Torvalds 3061da177e4SLinus Torvalds down_write(&mm->mmap_sem); 3071da177e4SLinus Torvalds 308a5b4592cSJiri Kosina #ifdef CONFIG_COMPAT_BRK 3095520e894SJiri Kosina /* 3105520e894SJiri Kosina * CONFIG_COMPAT_BRK can still be overridden by setting 3115520e894SJiri Kosina * randomize_va_space to 2, which will still cause mm->start_brk 3125520e894SJiri Kosina * to be arbitrarily shifted 3135520e894SJiri Kosina */ 3144471a675SJiri Kosina if (current->brk_randomized) 3155520e894SJiri Kosina min_brk = mm->start_brk; 3165520e894SJiri Kosina else 3175520e894SJiri Kosina min_brk = mm->end_data; 318a5b4592cSJiri Kosina #else 319a5b4592cSJiri Kosina min_brk = mm->start_brk; 320a5b4592cSJiri Kosina #endif 321a5b4592cSJiri Kosina if (brk < min_brk) 3221da177e4SLinus Torvalds goto out; 3231e624196SRam Gupta 3241e624196SRam Gupta /* 3251e624196SRam Gupta * Check against rlimit here. If this check is done later after the test 3261e624196SRam Gupta * of oldbrk with newbrk then it can escape the test and let the data 3271e624196SRam Gupta * segment grow beyond its set limit the in case where the limit is 3281e624196SRam Gupta * not page aligned -Ram Gupta 3291e624196SRam Gupta */ 3308764b338SCyrill Gorcunov if (check_data_rlimit(rlimit(RLIMIT_DATA), brk, mm->start_brk, 3318764b338SCyrill Gorcunov mm->end_data, mm->start_data)) 3321e624196SRam Gupta goto out; 3331e624196SRam Gupta 3341da177e4SLinus Torvalds newbrk = PAGE_ALIGN(brk); 3351da177e4SLinus Torvalds oldbrk = PAGE_ALIGN(mm->brk); 3361da177e4SLinus Torvalds if (oldbrk == newbrk) 3371da177e4SLinus Torvalds goto set_brk; 3381da177e4SLinus Torvalds 3391da177e4SLinus Torvalds /* Always allow shrinking brk. */ 3401da177e4SLinus Torvalds if (brk <= mm->brk) { 3411da177e4SLinus Torvalds if (!do_munmap(mm, newbrk, oldbrk-newbrk)) 3421da177e4SLinus Torvalds goto set_brk; 3431da177e4SLinus Torvalds goto out; 3441da177e4SLinus Torvalds } 3451da177e4SLinus Torvalds 3461da177e4SLinus Torvalds /* Check against existing mmap mappings. */ 3471da177e4SLinus Torvalds if (find_vma_intersection(mm, oldbrk, newbrk+PAGE_SIZE)) 3481da177e4SLinus Torvalds goto out; 3491da177e4SLinus Torvalds 3501da177e4SLinus Torvalds /* Ok, looks good - let it rip. */ 3511da177e4SLinus Torvalds if (do_brk(oldbrk, newbrk-oldbrk) != oldbrk) 3521da177e4SLinus Torvalds goto out; 353128557ffSMichel Lespinasse 3541da177e4SLinus Torvalds set_brk: 3551da177e4SLinus Torvalds mm->brk = brk; 356128557ffSMichel Lespinasse populate = newbrk > oldbrk && (mm->def_flags & VM_LOCKED) != 0; 357128557ffSMichel Lespinasse up_write(&mm->mmap_sem); 358128557ffSMichel Lespinasse if (populate) 359128557ffSMichel Lespinasse mm_populate(oldbrk, newbrk - oldbrk); 360128557ffSMichel Lespinasse return brk; 361128557ffSMichel Lespinasse 3621da177e4SLinus Torvalds out: 3631da177e4SLinus Torvalds retval = mm->brk; 3641da177e4SLinus Torvalds up_write(&mm->mmap_sem); 3651da177e4SLinus Torvalds return retval; 3661da177e4SLinus Torvalds } 3671da177e4SLinus Torvalds 368d3737187SMichel Lespinasse static long vma_compute_subtree_gap(struct vm_area_struct *vma) 369d3737187SMichel Lespinasse { 370d3737187SMichel Lespinasse unsigned long max, subtree_gap; 371d3737187SMichel Lespinasse max = vma->vm_start; 372d3737187SMichel Lespinasse if (vma->vm_prev) 373d3737187SMichel Lespinasse max -= vma->vm_prev->vm_end; 374d3737187SMichel Lespinasse if (vma->vm_rb.rb_left) { 375d3737187SMichel Lespinasse subtree_gap = rb_entry(vma->vm_rb.rb_left, 376d3737187SMichel Lespinasse struct vm_area_struct, vm_rb)->rb_subtree_gap; 377d3737187SMichel Lespinasse if (subtree_gap > max) 378d3737187SMichel Lespinasse max = subtree_gap; 379d3737187SMichel Lespinasse } 380d3737187SMichel Lespinasse if (vma->vm_rb.rb_right) { 381d3737187SMichel Lespinasse subtree_gap = rb_entry(vma->vm_rb.rb_right, 382d3737187SMichel Lespinasse struct vm_area_struct, vm_rb)->rb_subtree_gap; 383d3737187SMichel Lespinasse if (subtree_gap > max) 384d3737187SMichel Lespinasse max = subtree_gap; 385d3737187SMichel Lespinasse } 386d3737187SMichel Lespinasse return max; 387d3737187SMichel Lespinasse } 388d3737187SMichel Lespinasse 389ed8ea815SMichel Lespinasse #ifdef CONFIG_DEBUG_VM_RB 3901da177e4SLinus Torvalds static int browse_rb(struct rb_root *root) 3911da177e4SLinus Torvalds { 3925a0768f6SMichel Lespinasse int i = 0, j, bug = 0; 3931da177e4SLinus Torvalds struct rb_node *nd, *pn = NULL; 3941da177e4SLinus Torvalds unsigned long prev = 0, pend = 0; 3951da177e4SLinus Torvalds 3961da177e4SLinus Torvalds for (nd = rb_first(root); nd; nd = rb_next(nd)) { 3971da177e4SLinus Torvalds struct vm_area_struct *vma; 3981da177e4SLinus Torvalds vma = rb_entry(nd, struct vm_area_struct, vm_rb); 3995a0768f6SMichel Lespinasse if (vma->vm_start < prev) { 400ff26f70fSAndrew Morton pr_emerg("vm_start %lx < prev %lx\n", 401ff26f70fSAndrew Morton vma->vm_start, prev); 4025a0768f6SMichel Lespinasse bug = 1; 4035a0768f6SMichel Lespinasse } 4045a0768f6SMichel Lespinasse if (vma->vm_start < pend) { 405ff26f70fSAndrew Morton pr_emerg("vm_start %lx < pend %lx\n", 406ff26f70fSAndrew Morton vma->vm_start, pend); 4075a0768f6SMichel Lespinasse bug = 1; 4085a0768f6SMichel Lespinasse } 4095a0768f6SMichel Lespinasse if (vma->vm_start > vma->vm_end) { 410ff26f70fSAndrew Morton pr_emerg("vm_start %lx > vm_end %lx\n", 411ff26f70fSAndrew Morton vma->vm_start, vma->vm_end); 4125a0768f6SMichel Lespinasse bug = 1; 4135a0768f6SMichel Lespinasse } 4145a0768f6SMichel Lespinasse if (vma->rb_subtree_gap != vma_compute_subtree_gap(vma)) { 4158542bdfcSSasha Levin pr_emerg("free gap %lx, correct %lx\n", 4165a0768f6SMichel Lespinasse vma->rb_subtree_gap, 4175a0768f6SMichel Lespinasse vma_compute_subtree_gap(vma)); 4185a0768f6SMichel Lespinasse bug = 1; 4195a0768f6SMichel Lespinasse } 4201da177e4SLinus Torvalds i++; 4211da177e4SLinus Torvalds pn = nd; 422d1af65d1SDavid Miller prev = vma->vm_start; 423d1af65d1SDavid Miller pend = vma->vm_end; 4241da177e4SLinus Torvalds } 4251da177e4SLinus Torvalds j = 0; 4265a0768f6SMichel Lespinasse for (nd = pn; nd; nd = rb_prev(nd)) 4271da177e4SLinus Torvalds j++; 4285a0768f6SMichel Lespinasse if (i != j) { 4298542bdfcSSasha Levin pr_emerg("backwards %d, forwards %d\n", j, i); 4305a0768f6SMichel Lespinasse bug = 1; 4311da177e4SLinus Torvalds } 4325a0768f6SMichel Lespinasse return bug ? -1 : i; 4331da177e4SLinus Torvalds } 4341da177e4SLinus Torvalds 435d3737187SMichel Lespinasse static void validate_mm_rb(struct rb_root *root, struct vm_area_struct *ignore) 436d3737187SMichel Lespinasse { 437d3737187SMichel Lespinasse struct rb_node *nd; 438d3737187SMichel Lespinasse 439d3737187SMichel Lespinasse for (nd = rb_first(root); nd; nd = rb_next(nd)) { 440d3737187SMichel Lespinasse struct vm_area_struct *vma; 441d3737187SMichel Lespinasse vma = rb_entry(nd, struct vm_area_struct, vm_rb); 44296dad67fSSasha Levin VM_BUG_ON_VMA(vma != ignore && 44396dad67fSSasha Levin vma->rb_subtree_gap != vma_compute_subtree_gap(vma), 44496dad67fSSasha Levin vma); 445d3737187SMichel Lespinasse } 4461da177e4SLinus Torvalds } 4471da177e4SLinus Torvalds 448eafd4dc4SRashika Kheria static void validate_mm(struct mm_struct *mm) 4491da177e4SLinus Torvalds { 4501da177e4SLinus Torvalds int bug = 0; 4511da177e4SLinus Torvalds int i = 0; 4525a0768f6SMichel Lespinasse unsigned long highest_address = 0; 453ed8ea815SMichel Lespinasse struct vm_area_struct *vma = mm->mmap; 454ff26f70fSAndrew Morton 455ed8ea815SMichel Lespinasse while (vma) { 456ed8ea815SMichel Lespinasse struct anon_vma_chain *avc; 457ff26f70fSAndrew Morton 45863c3b902SMichel Lespinasse vma_lock_anon_vma(vma); 459ed8ea815SMichel Lespinasse list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) 460ed8ea815SMichel Lespinasse anon_vma_interval_tree_verify(avc); 46163c3b902SMichel Lespinasse vma_unlock_anon_vma(vma); 4625a0768f6SMichel Lespinasse highest_address = vma->vm_end; 463ed8ea815SMichel Lespinasse vma = vma->vm_next; 4641da177e4SLinus Torvalds i++; 4651da177e4SLinus Torvalds } 4665a0768f6SMichel Lespinasse if (i != mm->map_count) { 4678542bdfcSSasha Levin pr_emerg("map_count %d vm_next %d\n", mm->map_count, i); 4685a0768f6SMichel Lespinasse bug = 1; 4695a0768f6SMichel Lespinasse } 4705a0768f6SMichel Lespinasse if (highest_address != mm->highest_vm_end) { 4718542bdfcSSasha Levin pr_emerg("mm->highest_vm_end %lx, found %lx\n", 4725a0768f6SMichel Lespinasse mm->highest_vm_end, highest_address); 4735a0768f6SMichel Lespinasse bug = 1; 4745a0768f6SMichel Lespinasse } 4751da177e4SLinus Torvalds i = browse_rb(&mm->mm_rb); 4765a0768f6SMichel Lespinasse if (i != mm->map_count) { 477ff26f70fSAndrew Morton if (i != -1) 4788542bdfcSSasha Levin pr_emerg("map_count %d rb %d\n", mm->map_count, i); 4795a0768f6SMichel Lespinasse bug = 1; 4805a0768f6SMichel Lespinasse } 48196dad67fSSasha Levin VM_BUG_ON_MM(bug, mm); 4821da177e4SLinus Torvalds } 4831da177e4SLinus Torvalds #else 484d3737187SMichel Lespinasse #define validate_mm_rb(root, ignore) do { } while (0) 4851da177e4SLinus Torvalds #define validate_mm(mm) do { } while (0) 4861da177e4SLinus Torvalds #endif 4871da177e4SLinus Torvalds 488d3737187SMichel Lespinasse RB_DECLARE_CALLBACKS(static, vma_gap_callbacks, struct vm_area_struct, vm_rb, 489d3737187SMichel Lespinasse unsigned long, rb_subtree_gap, vma_compute_subtree_gap) 490d3737187SMichel Lespinasse 491d3737187SMichel Lespinasse /* 492d3737187SMichel Lespinasse * Update augmented rbtree rb_subtree_gap values after vma->vm_start or 493d3737187SMichel Lespinasse * vma->vm_prev->vm_end values changed, without modifying the vma's position 494d3737187SMichel Lespinasse * in the rbtree. 495d3737187SMichel Lespinasse */ 496d3737187SMichel Lespinasse static void vma_gap_update(struct vm_area_struct *vma) 497d3737187SMichel Lespinasse { 498d3737187SMichel Lespinasse /* 499d3737187SMichel Lespinasse * As it turns out, RB_DECLARE_CALLBACKS() already created a callback 500d3737187SMichel Lespinasse * function that does exacltly what we want. 501d3737187SMichel Lespinasse */ 502d3737187SMichel Lespinasse vma_gap_callbacks_propagate(&vma->vm_rb, NULL); 503d3737187SMichel Lespinasse } 504d3737187SMichel Lespinasse 505d3737187SMichel Lespinasse static inline void vma_rb_insert(struct vm_area_struct *vma, 506d3737187SMichel Lespinasse struct rb_root *root) 507d3737187SMichel Lespinasse { 508d3737187SMichel Lespinasse /* All rb_subtree_gap values must be consistent prior to insertion */ 509d3737187SMichel Lespinasse validate_mm_rb(root, NULL); 510d3737187SMichel Lespinasse 511d3737187SMichel Lespinasse rb_insert_augmented(&vma->vm_rb, root, &vma_gap_callbacks); 512d3737187SMichel Lespinasse } 513d3737187SMichel Lespinasse 514d3737187SMichel Lespinasse static void vma_rb_erase(struct vm_area_struct *vma, struct rb_root *root) 515d3737187SMichel Lespinasse { 516d3737187SMichel Lespinasse /* 517d3737187SMichel Lespinasse * All rb_subtree_gap values must be consistent prior to erase, 518d3737187SMichel Lespinasse * with the possible exception of the vma being erased. 519d3737187SMichel Lespinasse */ 520d3737187SMichel Lespinasse validate_mm_rb(root, vma); 521d3737187SMichel Lespinasse 522d3737187SMichel Lespinasse /* 523d3737187SMichel Lespinasse * Note rb_erase_augmented is a fairly large inline function, 524d3737187SMichel Lespinasse * so make sure we instantiate it only once with our desired 525d3737187SMichel Lespinasse * augmented rbtree callbacks. 526d3737187SMichel Lespinasse */ 527d3737187SMichel Lespinasse rb_erase_augmented(&vma->vm_rb, root, &vma_gap_callbacks); 528d3737187SMichel Lespinasse } 529d3737187SMichel Lespinasse 530bf181b9fSMichel Lespinasse /* 531bf181b9fSMichel Lespinasse * vma has some anon_vma assigned, and is already inserted on that 532bf181b9fSMichel Lespinasse * anon_vma's interval trees. 533bf181b9fSMichel Lespinasse * 534bf181b9fSMichel Lespinasse * Before updating the vma's vm_start / vm_end / vm_pgoff fields, the 535bf181b9fSMichel Lespinasse * vma must be removed from the anon_vma's interval trees using 536bf181b9fSMichel Lespinasse * anon_vma_interval_tree_pre_update_vma(). 537bf181b9fSMichel Lespinasse * 538bf181b9fSMichel Lespinasse * After the update, the vma will be reinserted using 539bf181b9fSMichel Lespinasse * anon_vma_interval_tree_post_update_vma(). 540bf181b9fSMichel Lespinasse * 541bf181b9fSMichel Lespinasse * The entire update must be protected by exclusive mmap_sem and by 542bf181b9fSMichel Lespinasse * the root anon_vma's mutex. 543bf181b9fSMichel Lespinasse */ 544bf181b9fSMichel Lespinasse static inline void 545bf181b9fSMichel Lespinasse anon_vma_interval_tree_pre_update_vma(struct vm_area_struct *vma) 546bf181b9fSMichel Lespinasse { 547bf181b9fSMichel Lespinasse struct anon_vma_chain *avc; 548bf181b9fSMichel Lespinasse 549bf181b9fSMichel Lespinasse list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) 550bf181b9fSMichel Lespinasse anon_vma_interval_tree_remove(avc, &avc->anon_vma->rb_root); 551bf181b9fSMichel Lespinasse } 552bf181b9fSMichel Lespinasse 553bf181b9fSMichel Lespinasse static inline void 554bf181b9fSMichel Lespinasse anon_vma_interval_tree_post_update_vma(struct vm_area_struct *vma) 555bf181b9fSMichel Lespinasse { 556bf181b9fSMichel Lespinasse struct anon_vma_chain *avc; 557bf181b9fSMichel Lespinasse 558bf181b9fSMichel Lespinasse list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) 559bf181b9fSMichel Lespinasse anon_vma_interval_tree_insert(avc, &avc->anon_vma->rb_root); 560bf181b9fSMichel Lespinasse } 561bf181b9fSMichel Lespinasse 5626597d783SHugh Dickins static int find_vma_links(struct mm_struct *mm, unsigned long addr, 5636597d783SHugh Dickins unsigned long end, struct vm_area_struct **pprev, 5646597d783SHugh Dickins struct rb_node ***rb_link, struct rb_node **rb_parent) 5651da177e4SLinus Torvalds { 5661da177e4SLinus Torvalds struct rb_node **__rb_link, *__rb_parent, *rb_prev; 5671da177e4SLinus Torvalds 5681da177e4SLinus Torvalds __rb_link = &mm->mm_rb.rb_node; 5691da177e4SLinus Torvalds rb_prev = __rb_parent = NULL; 5701da177e4SLinus Torvalds 5711da177e4SLinus Torvalds while (*__rb_link) { 5721da177e4SLinus Torvalds struct vm_area_struct *vma_tmp; 5731da177e4SLinus Torvalds 5741da177e4SLinus Torvalds __rb_parent = *__rb_link; 5751da177e4SLinus Torvalds vma_tmp = rb_entry(__rb_parent, struct vm_area_struct, vm_rb); 5761da177e4SLinus Torvalds 5771da177e4SLinus Torvalds if (vma_tmp->vm_end > addr) { 5786597d783SHugh Dickins /* Fail if an existing vma overlaps the area */ 5796597d783SHugh Dickins if (vma_tmp->vm_start < end) 5806597d783SHugh Dickins return -ENOMEM; 5811da177e4SLinus Torvalds __rb_link = &__rb_parent->rb_left; 5821da177e4SLinus Torvalds } else { 5831da177e4SLinus Torvalds rb_prev = __rb_parent; 5841da177e4SLinus Torvalds __rb_link = &__rb_parent->rb_right; 5851da177e4SLinus Torvalds } 5861da177e4SLinus Torvalds } 5871da177e4SLinus Torvalds 5881da177e4SLinus Torvalds *pprev = NULL; 5891da177e4SLinus Torvalds if (rb_prev) 5901da177e4SLinus Torvalds *pprev = rb_entry(rb_prev, struct vm_area_struct, vm_rb); 5911da177e4SLinus Torvalds *rb_link = __rb_link; 5921da177e4SLinus Torvalds *rb_parent = __rb_parent; 5936597d783SHugh Dickins return 0; 5941da177e4SLinus Torvalds } 5951da177e4SLinus Torvalds 596e8420a8eSCyril Hrubis static unsigned long count_vma_pages_range(struct mm_struct *mm, 597e8420a8eSCyril Hrubis unsigned long addr, unsigned long end) 598e8420a8eSCyril Hrubis { 599e8420a8eSCyril Hrubis unsigned long nr_pages = 0; 600e8420a8eSCyril Hrubis struct vm_area_struct *vma; 601e8420a8eSCyril Hrubis 602e8420a8eSCyril Hrubis /* Find first overlaping mapping */ 603e8420a8eSCyril Hrubis vma = find_vma_intersection(mm, addr, end); 604e8420a8eSCyril Hrubis if (!vma) 605e8420a8eSCyril Hrubis return 0; 606e8420a8eSCyril Hrubis 607e8420a8eSCyril Hrubis nr_pages = (min(end, vma->vm_end) - 608e8420a8eSCyril Hrubis max(addr, vma->vm_start)) >> PAGE_SHIFT; 609e8420a8eSCyril Hrubis 610e8420a8eSCyril Hrubis /* Iterate over the rest of the overlaps */ 611e8420a8eSCyril Hrubis for (vma = vma->vm_next; vma; vma = vma->vm_next) { 612e8420a8eSCyril Hrubis unsigned long overlap_len; 613e8420a8eSCyril Hrubis 614e8420a8eSCyril Hrubis if (vma->vm_start > end) 615e8420a8eSCyril Hrubis break; 616e8420a8eSCyril Hrubis 617e8420a8eSCyril Hrubis overlap_len = min(end, vma->vm_end) - vma->vm_start; 618e8420a8eSCyril Hrubis nr_pages += overlap_len >> PAGE_SHIFT; 619e8420a8eSCyril Hrubis } 620e8420a8eSCyril Hrubis 621e8420a8eSCyril Hrubis return nr_pages; 622e8420a8eSCyril Hrubis } 623e8420a8eSCyril Hrubis 6241da177e4SLinus Torvalds void __vma_link_rb(struct mm_struct *mm, struct vm_area_struct *vma, 6251da177e4SLinus Torvalds struct rb_node **rb_link, struct rb_node *rb_parent) 6261da177e4SLinus Torvalds { 627d3737187SMichel Lespinasse /* Update tracking information for the gap following the new vma. */ 628d3737187SMichel Lespinasse if (vma->vm_next) 629d3737187SMichel Lespinasse vma_gap_update(vma->vm_next); 630d3737187SMichel Lespinasse else 631d3737187SMichel Lespinasse mm->highest_vm_end = vma->vm_end; 632d3737187SMichel Lespinasse 633d3737187SMichel Lespinasse /* 634d3737187SMichel Lespinasse * vma->vm_prev wasn't known when we followed the rbtree to find the 635d3737187SMichel Lespinasse * correct insertion point for that vma. As a result, we could not 636d3737187SMichel Lespinasse * update the vma vm_rb parents rb_subtree_gap values on the way down. 637d3737187SMichel Lespinasse * So, we first insert the vma with a zero rb_subtree_gap value 638d3737187SMichel Lespinasse * (to be consistent with what we did on the way down), and then 639d3737187SMichel Lespinasse * immediately update the gap to the correct value. Finally we 640d3737187SMichel Lespinasse * rebalance the rbtree after all augmented values have been set. 641d3737187SMichel Lespinasse */ 6421da177e4SLinus Torvalds rb_link_node(&vma->vm_rb, rb_parent, rb_link); 643d3737187SMichel Lespinasse vma->rb_subtree_gap = 0; 644d3737187SMichel Lespinasse vma_gap_update(vma); 645d3737187SMichel Lespinasse vma_rb_insert(vma, &mm->mm_rb); 6461da177e4SLinus Torvalds } 6471da177e4SLinus Torvalds 648cb8f488cSDenys Vlasenko static void __vma_link_file(struct vm_area_struct *vma) 6491da177e4SLinus Torvalds { 6501da177e4SLinus Torvalds struct file *file; 6511da177e4SLinus Torvalds 6521da177e4SLinus Torvalds file = vma->vm_file; 6531da177e4SLinus Torvalds if (file) { 6541da177e4SLinus Torvalds struct address_space *mapping = file->f_mapping; 6551da177e4SLinus Torvalds 6561da177e4SLinus Torvalds if (vma->vm_flags & VM_DENYWRITE) 657496ad9aaSAl Viro atomic_dec(&file_inode(file)->i_writecount); 6581da177e4SLinus Torvalds if (vma->vm_flags & VM_SHARED) 6594bb5f5d9SDavid Herrmann atomic_inc(&mapping->i_mmap_writable); 6601da177e4SLinus Torvalds 6611da177e4SLinus Torvalds flush_dcache_mmap_lock(mapping); 6626b2dbba8SMichel Lespinasse vma_interval_tree_insert(vma, &mapping->i_mmap); 6631da177e4SLinus Torvalds flush_dcache_mmap_unlock(mapping); 6641da177e4SLinus Torvalds } 6651da177e4SLinus Torvalds } 6661da177e4SLinus Torvalds 6671da177e4SLinus Torvalds static void 6681da177e4SLinus Torvalds __vma_link(struct mm_struct *mm, struct vm_area_struct *vma, 6691da177e4SLinus Torvalds struct vm_area_struct *prev, struct rb_node **rb_link, 6701da177e4SLinus Torvalds struct rb_node *rb_parent) 6711da177e4SLinus Torvalds { 6721da177e4SLinus Torvalds __vma_link_list(mm, vma, prev, rb_parent); 6731da177e4SLinus Torvalds __vma_link_rb(mm, vma, rb_link, rb_parent); 6741da177e4SLinus Torvalds } 6751da177e4SLinus Torvalds 6761da177e4SLinus Torvalds static void vma_link(struct mm_struct *mm, struct vm_area_struct *vma, 6771da177e4SLinus Torvalds struct vm_area_struct *prev, struct rb_node **rb_link, 6781da177e4SLinus Torvalds struct rb_node *rb_parent) 6791da177e4SLinus Torvalds { 6801da177e4SLinus Torvalds struct address_space *mapping = NULL; 6811da177e4SLinus Torvalds 68264ac4940SHuang Shijie if (vma->vm_file) { 6831da177e4SLinus Torvalds mapping = vma->vm_file->f_mapping; 68483cde9e8SDavidlohr Bueso i_mmap_lock_write(mapping); 68564ac4940SHuang Shijie } 6861da177e4SLinus Torvalds 6871da177e4SLinus Torvalds __vma_link(mm, vma, prev, rb_link, rb_parent); 6881da177e4SLinus Torvalds __vma_link_file(vma); 6891da177e4SLinus Torvalds 6901da177e4SLinus Torvalds if (mapping) 69183cde9e8SDavidlohr Bueso i_mmap_unlock_write(mapping); 6921da177e4SLinus Torvalds 6931da177e4SLinus Torvalds mm->map_count++; 6941da177e4SLinus Torvalds validate_mm(mm); 6951da177e4SLinus Torvalds } 6961da177e4SLinus Torvalds 6971da177e4SLinus Torvalds /* 69888f6b4c3SKautuk Consul * Helper for vma_adjust() in the split_vma insert case: insert a vma into the 6996b2dbba8SMichel Lespinasse * mm's list and rbtree. It has already been inserted into the interval tree. 7001da177e4SLinus Torvalds */ 70148aae425SZhenwenXu static void __insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma) 7021da177e4SLinus Torvalds { 7036597d783SHugh Dickins struct vm_area_struct *prev; 7041da177e4SLinus Torvalds struct rb_node **rb_link, *rb_parent; 7051da177e4SLinus Torvalds 7066597d783SHugh Dickins if (find_vma_links(mm, vma->vm_start, vma->vm_end, 7076597d783SHugh Dickins &prev, &rb_link, &rb_parent)) 7086597d783SHugh Dickins BUG(); 7091da177e4SLinus Torvalds __vma_link(mm, vma, prev, rb_link, rb_parent); 7101da177e4SLinus Torvalds mm->map_count++; 7111da177e4SLinus Torvalds } 7121da177e4SLinus Torvalds 7131da177e4SLinus Torvalds static inline void 7141da177e4SLinus Torvalds __vma_unlink(struct mm_struct *mm, struct vm_area_struct *vma, 7151da177e4SLinus Torvalds struct vm_area_struct *prev) 7161da177e4SLinus Torvalds { 717d3737187SMichel Lespinasse struct vm_area_struct *next; 718297c5eeeSLinus Torvalds 719d3737187SMichel Lespinasse vma_rb_erase(vma, &mm->mm_rb); 720d3737187SMichel Lespinasse prev->vm_next = next = vma->vm_next; 721297c5eeeSLinus Torvalds if (next) 722297c5eeeSLinus Torvalds next->vm_prev = prev; 723615d6e87SDavidlohr Bueso 724615d6e87SDavidlohr Bueso /* Kill the cache */ 725615d6e87SDavidlohr Bueso vmacache_invalidate(mm); 7261da177e4SLinus Torvalds } 7271da177e4SLinus Torvalds 7281da177e4SLinus Torvalds /* 7291da177e4SLinus Torvalds * We cannot adjust vm_start, vm_end, vm_pgoff fields of a vma that 7301da177e4SLinus Torvalds * is already present in an i_mmap tree without adjusting the tree. 7311da177e4SLinus Torvalds * The following helper function should be used when such adjustments 7321da177e4SLinus Torvalds * are necessary. The "insert" vma (if any) is to be inserted 7331da177e4SLinus Torvalds * before we drop the necessary locks. 7341da177e4SLinus Torvalds */ 7355beb4930SRik van Riel int vma_adjust(struct vm_area_struct *vma, unsigned long start, 7361da177e4SLinus Torvalds unsigned long end, pgoff_t pgoff, struct vm_area_struct *insert) 7371da177e4SLinus Torvalds { 7381da177e4SLinus Torvalds struct mm_struct *mm = vma->vm_mm; 7391da177e4SLinus Torvalds struct vm_area_struct *next = vma->vm_next; 7401da177e4SLinus Torvalds struct vm_area_struct *importer = NULL; 7411da177e4SLinus Torvalds struct address_space *mapping = NULL; 7426b2dbba8SMichel Lespinasse struct rb_root *root = NULL; 743012f1800SRik van Riel struct anon_vma *anon_vma = NULL; 7441da177e4SLinus Torvalds struct file *file = vma->vm_file; 745d3737187SMichel Lespinasse bool start_changed = false, end_changed = false; 7461da177e4SLinus Torvalds long adjust_next = 0; 7471da177e4SLinus Torvalds int remove_next = 0; 7481da177e4SLinus Torvalds 7491da177e4SLinus Torvalds if (next && !insert) { 750287d97acSLinus Torvalds struct vm_area_struct *exporter = NULL; 751287d97acSLinus Torvalds 7521da177e4SLinus Torvalds if (end >= next->vm_end) { 7531da177e4SLinus Torvalds /* 7541da177e4SLinus Torvalds * vma expands, overlapping all the next, and 7551da177e4SLinus Torvalds * perhaps the one after too (mprotect case 6). 7561da177e4SLinus Torvalds */ 7571da177e4SLinus Torvalds again: remove_next = 1 + (end > next->vm_end); 7581da177e4SLinus Torvalds end = next->vm_end; 759287d97acSLinus Torvalds exporter = next; 7601da177e4SLinus Torvalds importer = vma; 7611da177e4SLinus Torvalds } else if (end > next->vm_start) { 7621da177e4SLinus Torvalds /* 7631da177e4SLinus Torvalds * vma expands, overlapping part of the next: 7641da177e4SLinus Torvalds * mprotect case 5 shifting the boundary up. 7651da177e4SLinus Torvalds */ 7661da177e4SLinus Torvalds adjust_next = (end - next->vm_start) >> PAGE_SHIFT; 767287d97acSLinus Torvalds exporter = next; 7681da177e4SLinus Torvalds importer = vma; 7691da177e4SLinus Torvalds } else if (end < vma->vm_end) { 7701da177e4SLinus Torvalds /* 7711da177e4SLinus Torvalds * vma shrinks, and !insert tells it's not 7721da177e4SLinus Torvalds * split_vma inserting another: so it must be 7731da177e4SLinus Torvalds * mprotect case 4 shifting the boundary down. 7741da177e4SLinus Torvalds */ 7751da177e4SLinus Torvalds adjust_next = -((vma->vm_end - end) >> PAGE_SHIFT); 776287d97acSLinus Torvalds exporter = vma; 7771da177e4SLinus Torvalds importer = next; 7781da177e4SLinus Torvalds } 7791da177e4SLinus Torvalds 7805beb4930SRik van Riel /* 7815beb4930SRik van Riel * Easily overlooked: when mprotect shifts the boundary, 7825beb4930SRik van Riel * make sure the expanding vma has anon_vma set if the 7835beb4930SRik van Riel * shrinking vma had, to cover any anon pages imported. 7845beb4930SRik van Riel */ 785287d97acSLinus Torvalds if (exporter && exporter->anon_vma && !importer->anon_vma) { 786c4ea95d7SDaniel Forrest int error; 787c4ea95d7SDaniel Forrest 788287d97acSLinus Torvalds importer->anon_vma = exporter->anon_vma; 789b800c91aSKonstantin Khlebnikov error = anon_vma_clone(importer, exporter); 7903fe89b3eSLeon Yu if (error) 791b800c91aSKonstantin Khlebnikov return error; 792b800c91aSKonstantin Khlebnikov } 7935beb4930SRik van Riel } 7945beb4930SRik van Riel 7951da177e4SLinus Torvalds if (file) { 7961da177e4SLinus Torvalds mapping = file->f_mapping; 7971da177e4SLinus Torvalds root = &mapping->i_mmap; 798cbc91f71SSrikar Dronamraju uprobe_munmap(vma, vma->vm_start, vma->vm_end); 799682968e0SSrikar Dronamraju 800682968e0SSrikar Dronamraju if (adjust_next) 80127ba0644SKirill A. Shutemov uprobe_munmap(next, next->vm_start, next->vm_end); 802682968e0SSrikar Dronamraju 80383cde9e8SDavidlohr Bueso i_mmap_lock_write(mapping); 8041da177e4SLinus Torvalds if (insert) { 8051da177e4SLinus Torvalds /* 8066b2dbba8SMichel Lespinasse * Put into interval tree now, so instantiated pages 8071da177e4SLinus Torvalds * are visible to arm/parisc __flush_dcache_page 8081da177e4SLinus Torvalds * throughout; but we cannot insert into address 8091da177e4SLinus Torvalds * space until vma start or end is updated. 8101da177e4SLinus Torvalds */ 8111da177e4SLinus Torvalds __vma_link_file(insert); 8121da177e4SLinus Torvalds } 8131da177e4SLinus Torvalds } 8141da177e4SLinus Torvalds 81594fcc585SAndrea Arcangeli vma_adjust_trans_huge(vma, start, end, adjust_next); 81694fcc585SAndrea Arcangeli 817012f1800SRik van Riel anon_vma = vma->anon_vma; 818bf181b9fSMichel Lespinasse if (!anon_vma && adjust_next) 819bf181b9fSMichel Lespinasse anon_vma = next->anon_vma; 820bf181b9fSMichel Lespinasse if (anon_vma) { 82181d1b09cSSasha Levin VM_BUG_ON_VMA(adjust_next && next->anon_vma && 82281d1b09cSSasha Levin anon_vma != next->anon_vma, next); 8234fc3f1d6SIngo Molnar anon_vma_lock_write(anon_vma); 824bf181b9fSMichel Lespinasse anon_vma_interval_tree_pre_update_vma(vma); 825bf181b9fSMichel Lespinasse if (adjust_next) 826bf181b9fSMichel Lespinasse anon_vma_interval_tree_pre_update_vma(next); 827bf181b9fSMichel Lespinasse } 828012f1800SRik van Riel 8291da177e4SLinus Torvalds if (root) { 8301da177e4SLinus Torvalds flush_dcache_mmap_lock(mapping); 8316b2dbba8SMichel Lespinasse vma_interval_tree_remove(vma, root); 8321da177e4SLinus Torvalds if (adjust_next) 8336b2dbba8SMichel Lespinasse vma_interval_tree_remove(next, root); 8341da177e4SLinus Torvalds } 8351da177e4SLinus Torvalds 836d3737187SMichel Lespinasse if (start != vma->vm_start) { 8371da177e4SLinus Torvalds vma->vm_start = start; 838d3737187SMichel Lespinasse start_changed = true; 839d3737187SMichel Lespinasse } 840d3737187SMichel Lespinasse if (end != vma->vm_end) { 8411da177e4SLinus Torvalds vma->vm_end = end; 842d3737187SMichel Lespinasse end_changed = true; 843d3737187SMichel Lespinasse } 8441da177e4SLinus Torvalds vma->vm_pgoff = pgoff; 8451da177e4SLinus Torvalds if (adjust_next) { 8461da177e4SLinus Torvalds next->vm_start += adjust_next << PAGE_SHIFT; 8471da177e4SLinus Torvalds next->vm_pgoff += adjust_next; 8481da177e4SLinus Torvalds } 8491da177e4SLinus Torvalds 8501da177e4SLinus Torvalds if (root) { 8511da177e4SLinus Torvalds if (adjust_next) 8526b2dbba8SMichel Lespinasse vma_interval_tree_insert(next, root); 8536b2dbba8SMichel Lespinasse vma_interval_tree_insert(vma, root); 8541da177e4SLinus Torvalds flush_dcache_mmap_unlock(mapping); 8551da177e4SLinus Torvalds } 8561da177e4SLinus Torvalds 8571da177e4SLinus Torvalds if (remove_next) { 8581da177e4SLinus Torvalds /* 8591da177e4SLinus Torvalds * vma_merge has merged next into vma, and needs 8601da177e4SLinus Torvalds * us to remove next before dropping the locks. 8611da177e4SLinus Torvalds */ 8621da177e4SLinus Torvalds __vma_unlink(mm, next, vma); 8631da177e4SLinus Torvalds if (file) 8641da177e4SLinus Torvalds __remove_shared_vm_struct(next, file, mapping); 8651da177e4SLinus Torvalds } else if (insert) { 8661da177e4SLinus Torvalds /* 8671da177e4SLinus Torvalds * split_vma has split insert from vma, and needs 8681da177e4SLinus Torvalds * us to insert it before dropping the locks 8691da177e4SLinus Torvalds * (it may either follow vma or precede it). 8701da177e4SLinus Torvalds */ 8711da177e4SLinus Torvalds __insert_vm_struct(mm, insert); 872d3737187SMichel Lespinasse } else { 873d3737187SMichel Lespinasse if (start_changed) 874d3737187SMichel Lespinasse vma_gap_update(vma); 875d3737187SMichel Lespinasse if (end_changed) { 876d3737187SMichel Lespinasse if (!next) 877d3737187SMichel Lespinasse mm->highest_vm_end = end; 878d3737187SMichel Lespinasse else if (!adjust_next) 879d3737187SMichel Lespinasse vma_gap_update(next); 880d3737187SMichel Lespinasse } 8811da177e4SLinus Torvalds } 8821da177e4SLinus Torvalds 883bf181b9fSMichel Lespinasse if (anon_vma) { 884bf181b9fSMichel Lespinasse anon_vma_interval_tree_post_update_vma(vma); 885bf181b9fSMichel Lespinasse if (adjust_next) 886bf181b9fSMichel Lespinasse anon_vma_interval_tree_post_update_vma(next); 88708b52706SKonstantin Khlebnikov anon_vma_unlock_write(anon_vma); 888bf181b9fSMichel Lespinasse } 8891da177e4SLinus Torvalds if (mapping) 89083cde9e8SDavidlohr Bueso i_mmap_unlock_write(mapping); 8911da177e4SLinus Torvalds 8922b144498SSrikar Dronamraju if (root) { 8937b2d81d4SIngo Molnar uprobe_mmap(vma); 8942b144498SSrikar Dronamraju 8952b144498SSrikar Dronamraju if (adjust_next) 8967b2d81d4SIngo Molnar uprobe_mmap(next); 8972b144498SSrikar Dronamraju } 8982b144498SSrikar Dronamraju 8991da177e4SLinus Torvalds if (remove_next) { 900925d1c40SMatt Helsley if (file) { 901cbc91f71SSrikar Dronamraju uprobe_munmap(next, next->vm_start, next->vm_end); 9021da177e4SLinus Torvalds fput(file); 903925d1c40SMatt Helsley } 9045beb4930SRik van Riel if (next->anon_vma) 9055beb4930SRik van Riel anon_vma_merge(vma, next); 9061da177e4SLinus Torvalds mm->map_count--; 9073964acd0SOleg Nesterov mpol_put(vma_policy(next)); 9081da177e4SLinus Torvalds kmem_cache_free(vm_area_cachep, next); 9091da177e4SLinus Torvalds /* 9101da177e4SLinus Torvalds * In mprotect's case 6 (see comments on vma_merge), 9111da177e4SLinus Torvalds * we must remove another next too. It would clutter 9121da177e4SLinus Torvalds * up the code too much to do both in one go. 9131da177e4SLinus Torvalds */ 9141da177e4SLinus Torvalds next = vma->vm_next; 915d3737187SMichel Lespinasse if (remove_next == 2) 9161da177e4SLinus Torvalds goto again; 917d3737187SMichel Lespinasse else if (next) 918d3737187SMichel Lespinasse vma_gap_update(next); 919d3737187SMichel Lespinasse else 920d3737187SMichel Lespinasse mm->highest_vm_end = end; 9211da177e4SLinus Torvalds } 9222b144498SSrikar Dronamraju if (insert && file) 9237b2d81d4SIngo Molnar uprobe_mmap(insert); 9241da177e4SLinus Torvalds 9251da177e4SLinus Torvalds validate_mm(mm); 9265beb4930SRik van Riel 9275beb4930SRik van Riel return 0; 9281da177e4SLinus Torvalds } 9291da177e4SLinus Torvalds 9301da177e4SLinus Torvalds /* 9311da177e4SLinus Torvalds * If the vma has a ->close operation then the driver probably needs to release 9321da177e4SLinus Torvalds * per-vma resources, so we don't attempt to merge those. 9331da177e4SLinus Torvalds */ 9341da177e4SLinus Torvalds static inline int is_mergeable_vma(struct vm_area_struct *vma, 93519a809afSAndrea Arcangeli struct file *file, unsigned long vm_flags, 93619a809afSAndrea Arcangeli struct vm_userfaultfd_ctx vm_userfaultfd_ctx) 9371da177e4SLinus Torvalds { 93834228d47SCyrill Gorcunov /* 93934228d47SCyrill Gorcunov * VM_SOFTDIRTY should not prevent from VMA merging, if we 94034228d47SCyrill Gorcunov * match the flags but dirty bit -- the caller should mark 94134228d47SCyrill Gorcunov * merged VMA as dirty. If dirty bit won't be excluded from 94234228d47SCyrill Gorcunov * comparison, we increase pressue on the memory system forcing 94334228d47SCyrill Gorcunov * the kernel to generate new VMAs when old one could be 94434228d47SCyrill Gorcunov * extended instead. 94534228d47SCyrill Gorcunov */ 94634228d47SCyrill Gorcunov if ((vma->vm_flags ^ vm_flags) & ~VM_SOFTDIRTY) 9471da177e4SLinus Torvalds return 0; 9481da177e4SLinus Torvalds if (vma->vm_file != file) 9491da177e4SLinus Torvalds return 0; 9501da177e4SLinus Torvalds if (vma->vm_ops && vma->vm_ops->close) 9511da177e4SLinus Torvalds return 0; 95219a809afSAndrea Arcangeli if (!is_mergeable_vm_userfaultfd_ctx(vma, vm_userfaultfd_ctx)) 95319a809afSAndrea Arcangeli return 0; 9541da177e4SLinus Torvalds return 1; 9551da177e4SLinus Torvalds } 9561da177e4SLinus Torvalds 9571da177e4SLinus Torvalds static inline int is_mergeable_anon_vma(struct anon_vma *anon_vma1, 958965f55deSShaohua Li struct anon_vma *anon_vma2, 959965f55deSShaohua Li struct vm_area_struct *vma) 9601da177e4SLinus Torvalds { 961965f55deSShaohua Li /* 962965f55deSShaohua Li * The list_is_singular() test is to avoid merging VMA cloned from 963965f55deSShaohua Li * parents. This can improve scalability caused by anon_vma lock. 964965f55deSShaohua Li */ 965965f55deSShaohua Li if ((!anon_vma1 || !anon_vma2) && (!vma || 966965f55deSShaohua Li list_is_singular(&vma->anon_vma_chain))) 967965f55deSShaohua Li return 1; 968965f55deSShaohua Li return anon_vma1 == anon_vma2; 9691da177e4SLinus Torvalds } 9701da177e4SLinus Torvalds 9711da177e4SLinus Torvalds /* 9721da177e4SLinus Torvalds * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff) 9731da177e4SLinus Torvalds * in front of (at a lower virtual address and file offset than) the vma. 9741da177e4SLinus Torvalds * 9751da177e4SLinus Torvalds * We cannot merge two vmas if they have differently assigned (non-NULL) 9761da177e4SLinus Torvalds * anon_vmas, nor if same anon_vma is assigned but offsets incompatible. 9771da177e4SLinus Torvalds * 9781da177e4SLinus Torvalds * We don't check here for the merged mmap wrapping around the end of pagecache 9791da177e4SLinus Torvalds * indices (16TB on ia32) because do_mmap_pgoff() does not permit mmap's which 9801da177e4SLinus Torvalds * wrap, nor mmaps which cover the final page at index -1UL. 9811da177e4SLinus Torvalds */ 9821da177e4SLinus Torvalds static int 9831da177e4SLinus Torvalds can_vma_merge_before(struct vm_area_struct *vma, unsigned long vm_flags, 98419a809afSAndrea Arcangeli struct anon_vma *anon_vma, struct file *file, 98519a809afSAndrea Arcangeli pgoff_t vm_pgoff, 98619a809afSAndrea Arcangeli struct vm_userfaultfd_ctx vm_userfaultfd_ctx) 9871da177e4SLinus Torvalds { 98819a809afSAndrea Arcangeli if (is_mergeable_vma(vma, file, vm_flags, vm_userfaultfd_ctx) && 989965f55deSShaohua Li is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) { 9901da177e4SLinus Torvalds if (vma->vm_pgoff == vm_pgoff) 9911da177e4SLinus Torvalds return 1; 9921da177e4SLinus Torvalds } 9931da177e4SLinus Torvalds return 0; 9941da177e4SLinus Torvalds } 9951da177e4SLinus Torvalds 9961da177e4SLinus Torvalds /* 9971da177e4SLinus Torvalds * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff) 9981da177e4SLinus Torvalds * beyond (at a higher virtual address and file offset than) the vma. 9991da177e4SLinus Torvalds * 10001da177e4SLinus Torvalds * We cannot merge two vmas if they have differently assigned (non-NULL) 10011da177e4SLinus Torvalds * anon_vmas, nor if same anon_vma is assigned but offsets incompatible. 10021da177e4SLinus Torvalds */ 10031da177e4SLinus Torvalds static int 10041da177e4SLinus Torvalds can_vma_merge_after(struct vm_area_struct *vma, unsigned long vm_flags, 100519a809afSAndrea Arcangeli struct anon_vma *anon_vma, struct file *file, 100619a809afSAndrea Arcangeli pgoff_t vm_pgoff, 100719a809afSAndrea Arcangeli struct vm_userfaultfd_ctx vm_userfaultfd_ctx) 10081da177e4SLinus Torvalds { 100919a809afSAndrea Arcangeli if (is_mergeable_vma(vma, file, vm_flags, vm_userfaultfd_ctx) && 1010965f55deSShaohua Li is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) { 10111da177e4SLinus Torvalds pgoff_t vm_pglen; 1012d6e93217SLibin vm_pglen = vma_pages(vma); 10131da177e4SLinus Torvalds if (vma->vm_pgoff + vm_pglen == vm_pgoff) 10141da177e4SLinus Torvalds return 1; 10151da177e4SLinus Torvalds } 10161da177e4SLinus Torvalds return 0; 10171da177e4SLinus Torvalds } 10181da177e4SLinus Torvalds 10191da177e4SLinus Torvalds /* 10201da177e4SLinus Torvalds * Given a mapping request (addr,end,vm_flags,file,pgoff), figure out 10211da177e4SLinus Torvalds * whether that can be merged with its predecessor or its successor. 10221da177e4SLinus Torvalds * Or both (it neatly fills a hole). 10231da177e4SLinus Torvalds * 10241da177e4SLinus Torvalds * In most cases - when called for mmap, brk or mremap - [addr,end) is 10251da177e4SLinus Torvalds * certain not to be mapped by the time vma_merge is called; but when 10261da177e4SLinus Torvalds * called for mprotect, it is certain to be already mapped (either at 10271da177e4SLinus Torvalds * an offset within prev, or at the start of next), and the flags of 10281da177e4SLinus Torvalds * this area are about to be changed to vm_flags - and the no-change 10291da177e4SLinus Torvalds * case has already been eliminated. 10301da177e4SLinus Torvalds * 10311da177e4SLinus Torvalds * The following mprotect cases have to be considered, where AAAA is 10321da177e4SLinus Torvalds * the area passed down from mprotect_fixup, never extending beyond one 10331da177e4SLinus Torvalds * vma, PPPPPP is the prev vma specified, and NNNNNN the next vma after: 10341da177e4SLinus Torvalds * 10351da177e4SLinus Torvalds * AAAA AAAA AAAA AAAA 10361da177e4SLinus Torvalds * PPPPPPNNNNNN PPPPPPNNNNNN PPPPPPNNNNNN PPPPNNNNXXXX 10371da177e4SLinus Torvalds * cannot merge might become might become might become 10381da177e4SLinus Torvalds * PPNNNNNNNNNN PPPPPPPPPPNN PPPPPPPPPPPP 6 or 10391da177e4SLinus Torvalds * mmap, brk or case 4 below case 5 below PPPPPPPPXXXX 7 or 10401da177e4SLinus Torvalds * mremap move: PPPPNNNNNNNN 8 10411da177e4SLinus Torvalds * AAAA 10421da177e4SLinus Torvalds * PPPP NNNN PPPPPPPPPPPP PPPPPPPPNNNN PPPPNNNNNNNN 10431da177e4SLinus Torvalds * might become case 1 below case 2 below case 3 below 10441da177e4SLinus Torvalds * 10451da177e4SLinus Torvalds * Odd one out? Case 8, because it extends NNNN but needs flags of XXXX: 10461da177e4SLinus Torvalds * mprotect_fixup updates vm_flags & vm_page_prot on successful return. 10471da177e4SLinus Torvalds */ 10481da177e4SLinus Torvalds struct vm_area_struct *vma_merge(struct mm_struct *mm, 10491da177e4SLinus Torvalds struct vm_area_struct *prev, unsigned long addr, 10501da177e4SLinus Torvalds unsigned long end, unsigned long vm_flags, 10511da177e4SLinus Torvalds struct anon_vma *anon_vma, struct file *file, 105219a809afSAndrea Arcangeli pgoff_t pgoff, struct mempolicy *policy, 105319a809afSAndrea Arcangeli struct vm_userfaultfd_ctx vm_userfaultfd_ctx) 10541da177e4SLinus Torvalds { 10551da177e4SLinus Torvalds pgoff_t pglen = (end - addr) >> PAGE_SHIFT; 10561da177e4SLinus Torvalds struct vm_area_struct *area, *next; 10575beb4930SRik van Riel int err; 10581da177e4SLinus Torvalds 10591da177e4SLinus Torvalds /* 10601da177e4SLinus Torvalds * We later require that vma->vm_flags == vm_flags, 10611da177e4SLinus Torvalds * so this tests vma->vm_flags & VM_SPECIAL, too. 10621da177e4SLinus Torvalds */ 10631da177e4SLinus Torvalds if (vm_flags & VM_SPECIAL) 10641da177e4SLinus Torvalds return NULL; 10651da177e4SLinus Torvalds 10661da177e4SLinus Torvalds if (prev) 10671da177e4SLinus Torvalds next = prev->vm_next; 10681da177e4SLinus Torvalds else 10691da177e4SLinus Torvalds next = mm->mmap; 10701da177e4SLinus Torvalds area = next; 10711da177e4SLinus Torvalds if (next && next->vm_end == end) /* cases 6, 7, 8 */ 10721da177e4SLinus Torvalds next = next->vm_next; 10731da177e4SLinus Torvalds 10741da177e4SLinus Torvalds /* 10751da177e4SLinus Torvalds * Can it merge with the predecessor? 10761da177e4SLinus Torvalds */ 10771da177e4SLinus Torvalds if (prev && prev->vm_end == addr && 10781da177e4SLinus Torvalds mpol_equal(vma_policy(prev), policy) && 10791da177e4SLinus Torvalds can_vma_merge_after(prev, vm_flags, 108019a809afSAndrea Arcangeli anon_vma, file, pgoff, 108119a809afSAndrea Arcangeli vm_userfaultfd_ctx)) { 10821da177e4SLinus Torvalds /* 10831da177e4SLinus Torvalds * OK, it can. Can we now merge in the successor as well? 10841da177e4SLinus Torvalds */ 10851da177e4SLinus Torvalds if (next && end == next->vm_start && 10861da177e4SLinus Torvalds mpol_equal(policy, vma_policy(next)) && 10871da177e4SLinus Torvalds can_vma_merge_before(next, vm_flags, 108819a809afSAndrea Arcangeli anon_vma, file, 108919a809afSAndrea Arcangeli pgoff+pglen, 109019a809afSAndrea Arcangeli vm_userfaultfd_ctx) && 10911da177e4SLinus Torvalds is_mergeable_anon_vma(prev->anon_vma, 1092965f55deSShaohua Li next->anon_vma, NULL)) { 10931da177e4SLinus Torvalds /* cases 1, 6 */ 10945beb4930SRik van Riel err = vma_adjust(prev, prev->vm_start, 10951da177e4SLinus Torvalds next->vm_end, prev->vm_pgoff, NULL); 10961da177e4SLinus Torvalds } else /* cases 2, 5, 7 */ 10975beb4930SRik van Riel err = vma_adjust(prev, prev->vm_start, 10981da177e4SLinus Torvalds end, prev->vm_pgoff, NULL); 10995beb4930SRik van Riel if (err) 11005beb4930SRik van Riel return NULL; 11016d50e60cSDavid Rientjes khugepaged_enter_vma_merge(prev, vm_flags); 11021da177e4SLinus Torvalds return prev; 11031da177e4SLinus Torvalds } 11041da177e4SLinus Torvalds 11051da177e4SLinus Torvalds /* 11061da177e4SLinus Torvalds * Can this new request be merged in front of next? 11071da177e4SLinus Torvalds */ 11081da177e4SLinus Torvalds if (next && end == next->vm_start && 11091da177e4SLinus Torvalds mpol_equal(policy, vma_policy(next)) && 11101da177e4SLinus Torvalds can_vma_merge_before(next, vm_flags, 111119a809afSAndrea Arcangeli anon_vma, file, pgoff+pglen, 111219a809afSAndrea Arcangeli vm_userfaultfd_ctx)) { 11131da177e4SLinus Torvalds if (prev && addr < prev->vm_end) /* case 4 */ 11145beb4930SRik van Riel err = vma_adjust(prev, prev->vm_start, 11151da177e4SLinus Torvalds addr, prev->vm_pgoff, NULL); 11161da177e4SLinus Torvalds else /* cases 3, 8 */ 11175beb4930SRik van Riel err = vma_adjust(area, addr, next->vm_end, 11181da177e4SLinus Torvalds next->vm_pgoff - pglen, NULL); 11195beb4930SRik van Riel if (err) 11205beb4930SRik van Riel return NULL; 11216d50e60cSDavid Rientjes khugepaged_enter_vma_merge(area, vm_flags); 11221da177e4SLinus Torvalds return area; 11231da177e4SLinus Torvalds } 11241da177e4SLinus Torvalds 11251da177e4SLinus Torvalds return NULL; 11261da177e4SLinus Torvalds } 11271da177e4SLinus Torvalds 11281da177e4SLinus Torvalds /* 1129d0e9fe17SLinus Torvalds * Rough compatbility check to quickly see if it's even worth looking 1130d0e9fe17SLinus Torvalds * at sharing an anon_vma. 1131d0e9fe17SLinus Torvalds * 1132d0e9fe17SLinus Torvalds * They need to have the same vm_file, and the flags can only differ 1133d0e9fe17SLinus Torvalds * in things that mprotect may change. 1134d0e9fe17SLinus Torvalds * 1135d0e9fe17SLinus Torvalds * NOTE! The fact that we share an anon_vma doesn't _have_ to mean that 1136d0e9fe17SLinus Torvalds * we can merge the two vma's. For example, we refuse to merge a vma if 1137d0e9fe17SLinus Torvalds * there is a vm_ops->close() function, because that indicates that the 1138d0e9fe17SLinus Torvalds * driver is doing some kind of reference counting. But that doesn't 1139d0e9fe17SLinus Torvalds * really matter for the anon_vma sharing case. 1140d0e9fe17SLinus Torvalds */ 1141d0e9fe17SLinus Torvalds static int anon_vma_compatible(struct vm_area_struct *a, struct vm_area_struct *b) 1142d0e9fe17SLinus Torvalds { 1143d0e9fe17SLinus Torvalds return a->vm_end == b->vm_start && 1144d0e9fe17SLinus Torvalds mpol_equal(vma_policy(a), vma_policy(b)) && 1145d0e9fe17SLinus Torvalds a->vm_file == b->vm_file && 114634228d47SCyrill Gorcunov !((a->vm_flags ^ b->vm_flags) & ~(VM_READ|VM_WRITE|VM_EXEC|VM_SOFTDIRTY)) && 1147d0e9fe17SLinus Torvalds b->vm_pgoff == a->vm_pgoff + ((b->vm_start - a->vm_start) >> PAGE_SHIFT); 1148d0e9fe17SLinus Torvalds } 1149d0e9fe17SLinus Torvalds 1150d0e9fe17SLinus Torvalds /* 1151d0e9fe17SLinus Torvalds * Do some basic sanity checking to see if we can re-use the anon_vma 1152d0e9fe17SLinus Torvalds * from 'old'. The 'a'/'b' vma's are in VM order - one of them will be 1153d0e9fe17SLinus Torvalds * the same as 'old', the other will be the new one that is trying 1154d0e9fe17SLinus Torvalds * to share the anon_vma. 1155d0e9fe17SLinus Torvalds * 1156d0e9fe17SLinus Torvalds * NOTE! This runs with mm_sem held for reading, so it is possible that 1157d0e9fe17SLinus Torvalds * the anon_vma of 'old' is concurrently in the process of being set up 1158d0e9fe17SLinus Torvalds * by another page fault trying to merge _that_. But that's ok: if it 1159d0e9fe17SLinus Torvalds * is being set up, that automatically means that it will be a singleton 1160d0e9fe17SLinus Torvalds * acceptable for merging, so we can do all of this optimistically. But 11614db0c3c2SJason Low * we do that READ_ONCE() to make sure that we never re-load the pointer. 1162d0e9fe17SLinus Torvalds * 1163d0e9fe17SLinus Torvalds * IOW: that the "list_is_singular()" test on the anon_vma_chain only 1164d0e9fe17SLinus Torvalds * matters for the 'stable anon_vma' case (ie the thing we want to avoid 1165d0e9fe17SLinus Torvalds * is to return an anon_vma that is "complex" due to having gone through 1166d0e9fe17SLinus Torvalds * a fork). 1167d0e9fe17SLinus Torvalds * 1168d0e9fe17SLinus Torvalds * We also make sure that the two vma's are compatible (adjacent, 1169d0e9fe17SLinus Torvalds * and with the same memory policies). That's all stable, even with just 1170d0e9fe17SLinus Torvalds * a read lock on the mm_sem. 1171d0e9fe17SLinus Torvalds */ 1172d0e9fe17SLinus Torvalds static struct anon_vma *reusable_anon_vma(struct vm_area_struct *old, struct vm_area_struct *a, struct vm_area_struct *b) 1173d0e9fe17SLinus Torvalds { 1174d0e9fe17SLinus Torvalds if (anon_vma_compatible(a, b)) { 11754db0c3c2SJason Low struct anon_vma *anon_vma = READ_ONCE(old->anon_vma); 1176d0e9fe17SLinus Torvalds 1177d0e9fe17SLinus Torvalds if (anon_vma && list_is_singular(&old->anon_vma_chain)) 1178d0e9fe17SLinus Torvalds return anon_vma; 1179d0e9fe17SLinus Torvalds } 1180d0e9fe17SLinus Torvalds return NULL; 1181d0e9fe17SLinus Torvalds } 1182d0e9fe17SLinus Torvalds 1183d0e9fe17SLinus Torvalds /* 11841da177e4SLinus Torvalds * find_mergeable_anon_vma is used by anon_vma_prepare, to check 11851da177e4SLinus Torvalds * neighbouring vmas for a suitable anon_vma, before it goes off 11861da177e4SLinus Torvalds * to allocate a new anon_vma. It checks because a repetitive 11871da177e4SLinus Torvalds * sequence of mprotects and faults may otherwise lead to distinct 11881da177e4SLinus Torvalds * anon_vmas being allocated, preventing vma merge in subsequent 11891da177e4SLinus Torvalds * mprotect. 11901da177e4SLinus Torvalds */ 11911da177e4SLinus Torvalds struct anon_vma *find_mergeable_anon_vma(struct vm_area_struct *vma) 11921da177e4SLinus Torvalds { 1193d0e9fe17SLinus Torvalds struct anon_vma *anon_vma; 11941da177e4SLinus Torvalds struct vm_area_struct *near; 11951da177e4SLinus Torvalds 11961da177e4SLinus Torvalds near = vma->vm_next; 11971da177e4SLinus Torvalds if (!near) 11981da177e4SLinus Torvalds goto try_prev; 11991da177e4SLinus Torvalds 1200d0e9fe17SLinus Torvalds anon_vma = reusable_anon_vma(near, vma, near); 1201d0e9fe17SLinus Torvalds if (anon_vma) 1202d0e9fe17SLinus Torvalds return anon_vma; 12031da177e4SLinus Torvalds try_prev: 12049be34c9dSLinus Torvalds near = vma->vm_prev; 12051da177e4SLinus Torvalds if (!near) 12061da177e4SLinus Torvalds goto none; 12071da177e4SLinus Torvalds 1208d0e9fe17SLinus Torvalds anon_vma = reusable_anon_vma(near, near, vma); 1209d0e9fe17SLinus Torvalds if (anon_vma) 1210d0e9fe17SLinus Torvalds return anon_vma; 12111da177e4SLinus Torvalds none: 12121da177e4SLinus Torvalds /* 12131da177e4SLinus Torvalds * There's no absolute need to look only at touching neighbours: 12141da177e4SLinus Torvalds * we could search further afield for "compatible" anon_vmas. 12151da177e4SLinus Torvalds * But it would probably just be a waste of time searching, 12161da177e4SLinus Torvalds * or lead to too many vmas hanging off the same anon_vma. 12171da177e4SLinus Torvalds * We're trying to allow mprotect remerging later on, 12181da177e4SLinus Torvalds * not trying to minimize memory used for anon_vmas. 12191da177e4SLinus Torvalds */ 12201da177e4SLinus Torvalds return NULL; 12211da177e4SLinus Torvalds } 12221da177e4SLinus Torvalds 12231da177e4SLinus Torvalds #ifdef CONFIG_PROC_FS 1224ab50b8edSHugh Dickins void vm_stat_account(struct mm_struct *mm, unsigned long flags, 12251da177e4SLinus Torvalds struct file *file, long pages) 12261da177e4SLinus Torvalds { 12271da177e4SLinus Torvalds const unsigned long stack_flags 12281da177e4SLinus Torvalds = VM_STACK_FLAGS & (VM_GROWSUP|VM_GROWSDOWN); 12291da177e4SLinus Torvalds 123044de9d0cSHuang Shijie mm->total_vm += pages; 123144de9d0cSHuang Shijie 12321da177e4SLinus Torvalds if (file) { 12331da177e4SLinus Torvalds mm->shared_vm += pages; 12341da177e4SLinus Torvalds if ((flags & (VM_EXEC|VM_WRITE)) == VM_EXEC) 12351da177e4SLinus Torvalds mm->exec_vm += pages; 12361da177e4SLinus Torvalds } else if (flags & stack_flags) 12371da177e4SLinus Torvalds mm->stack_vm += pages; 12381da177e4SLinus Torvalds } 12391da177e4SLinus Torvalds #endif /* CONFIG_PROC_FS */ 12401da177e4SLinus Torvalds 12411da177e4SLinus Torvalds /* 124240401530SAl Viro * If a hint addr is less than mmap_min_addr change hint to be as 124340401530SAl Viro * low as possible but still greater than mmap_min_addr 124440401530SAl Viro */ 124540401530SAl Viro static inline unsigned long round_hint_to_min(unsigned long hint) 124640401530SAl Viro { 124740401530SAl Viro hint &= PAGE_MASK; 124840401530SAl Viro if (((void *)hint != NULL) && 124940401530SAl Viro (hint < mmap_min_addr)) 125040401530SAl Viro return PAGE_ALIGN(mmap_min_addr); 125140401530SAl Viro return hint; 125240401530SAl Viro } 125340401530SAl Viro 1254363ee17fSDavidlohr Bueso static inline int mlock_future_check(struct mm_struct *mm, 1255363ee17fSDavidlohr Bueso unsigned long flags, 1256363ee17fSDavidlohr Bueso unsigned long len) 1257363ee17fSDavidlohr Bueso { 1258363ee17fSDavidlohr Bueso unsigned long locked, lock_limit; 1259363ee17fSDavidlohr Bueso 1260363ee17fSDavidlohr Bueso /* mlock MCL_FUTURE? */ 1261363ee17fSDavidlohr Bueso if (flags & VM_LOCKED) { 1262363ee17fSDavidlohr Bueso locked = len >> PAGE_SHIFT; 1263363ee17fSDavidlohr Bueso locked += mm->locked_vm; 1264363ee17fSDavidlohr Bueso lock_limit = rlimit(RLIMIT_MEMLOCK); 1265363ee17fSDavidlohr Bueso lock_limit >>= PAGE_SHIFT; 1266363ee17fSDavidlohr Bueso if (locked > lock_limit && !capable(CAP_IPC_LOCK)) 1267363ee17fSDavidlohr Bueso return -EAGAIN; 1268363ee17fSDavidlohr Bueso } 1269363ee17fSDavidlohr Bueso return 0; 1270363ee17fSDavidlohr Bueso } 1271363ee17fSDavidlohr Bueso 127240401530SAl Viro /* 127327f5de79SJianjun Kong * The caller must hold down_write(¤t->mm->mmap_sem). 12741da177e4SLinus Torvalds */ 12751fcfd8dbSOleg Nesterov unsigned long do_mmap(struct file *file, unsigned long addr, 12761da177e4SLinus Torvalds unsigned long len, unsigned long prot, 12771fcfd8dbSOleg Nesterov unsigned long flags, vm_flags_t vm_flags, 12781fcfd8dbSOleg Nesterov unsigned long pgoff, unsigned long *populate) 12791da177e4SLinus Torvalds { 12801da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 12811da177e4SLinus Torvalds 128241badc15SMichel Lespinasse *populate = 0; 1283bebeb3d6SMichel Lespinasse 1284e37609bbSPiotr Kwapulinski if (!len) 1285e37609bbSPiotr Kwapulinski return -EINVAL; 1286e37609bbSPiotr Kwapulinski 12871da177e4SLinus Torvalds /* 12881da177e4SLinus Torvalds * Does the application expect PROT_READ to imply PROT_EXEC? 12891da177e4SLinus Torvalds * 12901da177e4SLinus Torvalds * (the exception is when the underlying filesystem is noexec 12911da177e4SLinus Torvalds * mounted, in which case we dont add PROT_EXEC.) 12921da177e4SLinus Torvalds */ 12931da177e4SLinus Torvalds if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC)) 129490f8572bSEric W. Biederman if (!(file && path_noexec(&file->f_path))) 12951da177e4SLinus Torvalds prot |= PROT_EXEC; 12961da177e4SLinus Torvalds 12977cd94146SEric Paris if (!(flags & MAP_FIXED)) 12987cd94146SEric Paris addr = round_hint_to_min(addr); 12997cd94146SEric Paris 13001da177e4SLinus Torvalds /* Careful about overflows.. */ 13011da177e4SLinus Torvalds len = PAGE_ALIGN(len); 13029206de95SAl Viro if (!len) 13031da177e4SLinus Torvalds return -ENOMEM; 13041da177e4SLinus Torvalds 13051da177e4SLinus Torvalds /* offset overflow? */ 13061da177e4SLinus Torvalds if ((pgoff + (len >> PAGE_SHIFT)) < pgoff) 13071da177e4SLinus Torvalds return -EOVERFLOW; 13081da177e4SLinus Torvalds 13091da177e4SLinus Torvalds /* Too many mappings? */ 13101da177e4SLinus Torvalds if (mm->map_count > sysctl_max_map_count) 13111da177e4SLinus Torvalds return -ENOMEM; 13121da177e4SLinus Torvalds 13131da177e4SLinus Torvalds /* Obtain the address to map to. we verify (or select) it and ensure 13141da177e4SLinus Torvalds * that it represents a valid section of the address space. 13151da177e4SLinus Torvalds */ 13161da177e4SLinus Torvalds addr = get_unmapped_area(file, addr, len, pgoff, flags); 1317de1741a1SAlexander Kuleshov if (offset_in_page(addr)) 13181da177e4SLinus Torvalds return addr; 13191da177e4SLinus Torvalds 13201da177e4SLinus Torvalds /* Do simple checking here so the lower-level routines won't have 13211da177e4SLinus Torvalds * to. we assume access permissions have been handled by the open 13221da177e4SLinus Torvalds * of the memory object, so we don't do any here. 13231da177e4SLinus Torvalds */ 13241fcfd8dbSOleg Nesterov vm_flags |= calc_vm_prot_bits(prot) | calc_vm_flag_bits(flags) | 13251da177e4SLinus Torvalds mm->def_flags | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC; 13261da177e4SLinus Torvalds 1327cdf7b341SHuang Shijie if (flags & MAP_LOCKED) 13281da177e4SLinus Torvalds if (!can_do_mlock()) 13291da177e4SLinus Torvalds return -EPERM; 1330ba470de4SRik van Riel 1331363ee17fSDavidlohr Bueso if (mlock_future_check(mm, vm_flags, len)) 13321da177e4SLinus Torvalds return -EAGAIN; 13331da177e4SLinus Torvalds 13341da177e4SLinus Torvalds if (file) { 1335077bf22bSOleg Nesterov struct inode *inode = file_inode(file); 1336077bf22bSOleg Nesterov 13371da177e4SLinus Torvalds switch (flags & MAP_TYPE) { 13381da177e4SLinus Torvalds case MAP_SHARED: 13391da177e4SLinus Torvalds if ((prot&PROT_WRITE) && !(file->f_mode&FMODE_WRITE)) 13401da177e4SLinus Torvalds return -EACCES; 13411da177e4SLinus Torvalds 13421da177e4SLinus Torvalds /* 13431da177e4SLinus Torvalds * Make sure we don't allow writing to an append-only 13441da177e4SLinus Torvalds * file.. 13451da177e4SLinus Torvalds */ 13461da177e4SLinus Torvalds if (IS_APPEND(inode) && (file->f_mode & FMODE_WRITE)) 13471da177e4SLinus Torvalds return -EACCES; 13481da177e4SLinus Torvalds 13491da177e4SLinus Torvalds /* 13501da177e4SLinus Torvalds * Make sure there are no mandatory locks on the file. 13511da177e4SLinus Torvalds */ 1352d7a06983SJeff Layton if (locks_verify_locked(file)) 13531da177e4SLinus Torvalds return -EAGAIN; 13541da177e4SLinus Torvalds 13551da177e4SLinus Torvalds vm_flags |= VM_SHARED | VM_MAYSHARE; 13561da177e4SLinus Torvalds if (!(file->f_mode & FMODE_WRITE)) 13571da177e4SLinus Torvalds vm_flags &= ~(VM_MAYWRITE | VM_SHARED); 13581da177e4SLinus Torvalds 13591da177e4SLinus Torvalds /* fall through */ 13601da177e4SLinus Torvalds case MAP_PRIVATE: 13611da177e4SLinus Torvalds if (!(file->f_mode & FMODE_READ)) 13621da177e4SLinus Torvalds return -EACCES; 136390f8572bSEric W. Biederman if (path_noexec(&file->f_path)) { 136480c5606cSLinus Torvalds if (vm_flags & VM_EXEC) 136580c5606cSLinus Torvalds return -EPERM; 136680c5606cSLinus Torvalds vm_flags &= ~VM_MAYEXEC; 136780c5606cSLinus Torvalds } 136880c5606cSLinus Torvalds 136972c2d531SAl Viro if (!file->f_op->mmap) 137080c5606cSLinus Torvalds return -ENODEV; 1371b2c56e4fSOleg Nesterov if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP)) 1372b2c56e4fSOleg Nesterov return -EINVAL; 13731da177e4SLinus Torvalds break; 13741da177e4SLinus Torvalds 13751da177e4SLinus Torvalds default: 13761da177e4SLinus Torvalds return -EINVAL; 13771da177e4SLinus Torvalds } 13781da177e4SLinus Torvalds } else { 13791da177e4SLinus Torvalds switch (flags & MAP_TYPE) { 13801da177e4SLinus Torvalds case MAP_SHARED: 1381b2c56e4fSOleg Nesterov if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP)) 1382b2c56e4fSOleg Nesterov return -EINVAL; 1383ce363942STejun Heo /* 1384ce363942STejun Heo * Ignore pgoff. 1385ce363942STejun Heo */ 1386ce363942STejun Heo pgoff = 0; 13871da177e4SLinus Torvalds vm_flags |= VM_SHARED | VM_MAYSHARE; 13881da177e4SLinus Torvalds break; 13891da177e4SLinus Torvalds case MAP_PRIVATE: 13901da177e4SLinus Torvalds /* 13911da177e4SLinus Torvalds * Set pgoff according to addr for anon_vma. 13921da177e4SLinus Torvalds */ 13931da177e4SLinus Torvalds pgoff = addr >> PAGE_SHIFT; 13941da177e4SLinus Torvalds break; 13951da177e4SLinus Torvalds default: 13961da177e4SLinus Torvalds return -EINVAL; 13971da177e4SLinus Torvalds } 13981da177e4SLinus Torvalds } 13991da177e4SLinus Torvalds 1400c22c0d63SMichel Lespinasse /* 1401c22c0d63SMichel Lespinasse * Set 'VM_NORESERVE' if we should not account for the 1402c22c0d63SMichel Lespinasse * memory use of this mapping. 1403c22c0d63SMichel Lespinasse */ 1404c22c0d63SMichel Lespinasse if (flags & MAP_NORESERVE) { 1405c22c0d63SMichel Lespinasse /* We honor MAP_NORESERVE if allowed to overcommit */ 1406c22c0d63SMichel Lespinasse if (sysctl_overcommit_memory != OVERCOMMIT_NEVER) 1407c22c0d63SMichel Lespinasse vm_flags |= VM_NORESERVE; 1408c22c0d63SMichel Lespinasse 1409c22c0d63SMichel Lespinasse /* hugetlb applies strict overcommit unless MAP_NORESERVE */ 1410c22c0d63SMichel Lespinasse if (file && is_file_hugepages(file)) 1411c22c0d63SMichel Lespinasse vm_flags |= VM_NORESERVE; 1412c22c0d63SMichel Lespinasse } 1413c22c0d63SMichel Lespinasse 1414c22c0d63SMichel Lespinasse addr = mmap_region(file, addr, len, vm_flags, pgoff); 141509a9f1d2SMichel Lespinasse if (!IS_ERR_VALUE(addr) && 141609a9f1d2SMichel Lespinasse ((vm_flags & VM_LOCKED) || 141709a9f1d2SMichel Lespinasse (flags & (MAP_POPULATE | MAP_NONBLOCK)) == MAP_POPULATE)) 141841badc15SMichel Lespinasse *populate = len; 1419bebeb3d6SMichel Lespinasse return addr; 14200165ab44SMiklos Szeredi } 14216be5ceb0SLinus Torvalds 142266f0dc48SHugh Dickins SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len, 142366f0dc48SHugh Dickins unsigned long, prot, unsigned long, flags, 142466f0dc48SHugh Dickins unsigned long, fd, unsigned long, pgoff) 142566f0dc48SHugh Dickins { 142666f0dc48SHugh Dickins struct file *file = NULL; 14271e3ee14bSChen Gang unsigned long retval; 142866f0dc48SHugh Dickins 142966f0dc48SHugh Dickins if (!(flags & MAP_ANONYMOUS)) { 1430120a795dSAl Viro audit_mmap_fd(fd, flags); 143166f0dc48SHugh Dickins file = fget(fd); 143266f0dc48SHugh Dickins if (!file) 14331e3ee14bSChen Gang return -EBADF; 1434af73e4d9SNaoya Horiguchi if (is_file_hugepages(file)) 1435af73e4d9SNaoya Horiguchi len = ALIGN(len, huge_page_size(hstate_file(file))); 1436493af578SJörn Engel retval = -EINVAL; 1437493af578SJörn Engel if (unlikely(flags & MAP_HUGETLB && !is_file_hugepages(file))) 1438493af578SJörn Engel goto out_fput; 143966f0dc48SHugh Dickins } else if (flags & MAP_HUGETLB) { 144066f0dc48SHugh Dickins struct user_struct *user = NULL; 1441c103a4dcSAndrew Morton struct hstate *hs; 1442af73e4d9SNaoya Horiguchi 1443c103a4dcSAndrew Morton hs = hstate_sizelog((flags >> MAP_HUGE_SHIFT) & SHM_HUGE_MASK); 1444091d0d55SLi Zefan if (!hs) 1445091d0d55SLi Zefan return -EINVAL; 1446091d0d55SLi Zefan 1447091d0d55SLi Zefan len = ALIGN(len, huge_page_size(hs)); 144866f0dc48SHugh Dickins /* 144966f0dc48SHugh Dickins * VM_NORESERVE is used because the reservations will be 145066f0dc48SHugh Dickins * taken when vm_ops->mmap() is called 145166f0dc48SHugh Dickins * A dummy user value is used because we are not locking 145266f0dc48SHugh Dickins * memory so no accounting is necessary 145366f0dc48SHugh Dickins */ 1454af73e4d9SNaoya Horiguchi file = hugetlb_file_setup(HUGETLB_ANON_FILE, len, 145542d7395fSAndi Kleen VM_NORESERVE, 145642d7395fSAndi Kleen &user, HUGETLB_ANONHUGE_INODE, 145742d7395fSAndi Kleen (flags >> MAP_HUGE_SHIFT) & MAP_HUGE_MASK); 145866f0dc48SHugh Dickins if (IS_ERR(file)) 145966f0dc48SHugh Dickins return PTR_ERR(file); 146066f0dc48SHugh Dickins } 146166f0dc48SHugh Dickins 146266f0dc48SHugh Dickins flags &= ~(MAP_EXECUTABLE | MAP_DENYWRITE); 146366f0dc48SHugh Dickins 1464eb36c587SAl Viro retval = vm_mmap_pgoff(file, addr, len, prot, flags, pgoff); 1465493af578SJörn Engel out_fput: 146666f0dc48SHugh Dickins if (file) 146766f0dc48SHugh Dickins fput(file); 146866f0dc48SHugh Dickins return retval; 146966f0dc48SHugh Dickins } 147066f0dc48SHugh Dickins 1471a4679373SChristoph Hellwig #ifdef __ARCH_WANT_SYS_OLD_MMAP 1472a4679373SChristoph Hellwig struct mmap_arg_struct { 1473a4679373SChristoph Hellwig unsigned long addr; 1474a4679373SChristoph Hellwig unsigned long len; 1475a4679373SChristoph Hellwig unsigned long prot; 1476a4679373SChristoph Hellwig unsigned long flags; 1477a4679373SChristoph Hellwig unsigned long fd; 1478a4679373SChristoph Hellwig unsigned long offset; 1479a4679373SChristoph Hellwig }; 1480a4679373SChristoph Hellwig 1481a4679373SChristoph Hellwig SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg) 1482a4679373SChristoph Hellwig { 1483a4679373SChristoph Hellwig struct mmap_arg_struct a; 1484a4679373SChristoph Hellwig 1485a4679373SChristoph Hellwig if (copy_from_user(&a, arg, sizeof(a))) 1486a4679373SChristoph Hellwig return -EFAULT; 1487de1741a1SAlexander Kuleshov if (offset_in_page(a.offset)) 1488a4679373SChristoph Hellwig return -EINVAL; 1489a4679373SChristoph Hellwig 1490a4679373SChristoph Hellwig return sys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd, 1491a4679373SChristoph Hellwig a.offset >> PAGE_SHIFT); 1492a4679373SChristoph Hellwig } 1493a4679373SChristoph Hellwig #endif /* __ARCH_WANT_SYS_OLD_MMAP */ 1494a4679373SChristoph Hellwig 14954e950f6fSAlexey Dobriyan /* 14964e950f6fSAlexey Dobriyan * Some shared mappigns will want the pages marked read-only 14974e950f6fSAlexey Dobriyan * to track write events. If so, we'll downgrade vm_page_prot 14984e950f6fSAlexey Dobriyan * to the private version (using protection_map[] without the 14994e950f6fSAlexey Dobriyan * VM_SHARED bit). 15004e950f6fSAlexey Dobriyan */ 15014e950f6fSAlexey Dobriyan int vma_wants_writenotify(struct vm_area_struct *vma) 15024e950f6fSAlexey Dobriyan { 1503ca16d140SKOSAKI Motohiro vm_flags_t vm_flags = vma->vm_flags; 15048a04446aSKirill A. Shutemov const struct vm_operations_struct *vm_ops = vma->vm_ops; 15054e950f6fSAlexey Dobriyan 15064e950f6fSAlexey Dobriyan /* If it was private or non-writable, the write bit is already clear */ 15074e950f6fSAlexey Dobriyan if ((vm_flags & (VM_WRITE|VM_SHARED)) != ((VM_WRITE|VM_SHARED))) 15084e950f6fSAlexey Dobriyan return 0; 15094e950f6fSAlexey Dobriyan 15104e950f6fSAlexey Dobriyan /* The backer wishes to know when pages are first written to? */ 15118a04446aSKirill A. Shutemov if (vm_ops && (vm_ops->page_mkwrite || vm_ops->pfn_mkwrite)) 15124e950f6fSAlexey Dobriyan return 1; 15134e950f6fSAlexey Dobriyan 151464e45507SPeter Feiner /* The open routine did something to the protections that pgprot_modify 151564e45507SPeter Feiner * won't preserve? */ 15164e950f6fSAlexey Dobriyan if (pgprot_val(vma->vm_page_prot) != 151764e45507SPeter Feiner pgprot_val(vm_pgprot_modify(vma->vm_page_prot, vm_flags))) 15184e950f6fSAlexey Dobriyan return 0; 15194e950f6fSAlexey Dobriyan 152064e45507SPeter Feiner /* Do we need to track softdirty? */ 152164e45507SPeter Feiner if (IS_ENABLED(CONFIG_MEM_SOFT_DIRTY) && !(vm_flags & VM_SOFTDIRTY)) 152264e45507SPeter Feiner return 1; 152364e45507SPeter Feiner 15244e950f6fSAlexey Dobriyan /* Specialty mapping? */ 15254b6e1e37SKonstantin Khlebnikov if (vm_flags & VM_PFNMAP) 15264e950f6fSAlexey Dobriyan return 0; 15274e950f6fSAlexey Dobriyan 15284e950f6fSAlexey Dobriyan /* Can the mapping track the dirty pages? */ 15294e950f6fSAlexey Dobriyan return vma->vm_file && vma->vm_file->f_mapping && 15304e950f6fSAlexey Dobriyan mapping_cap_account_dirty(vma->vm_file->f_mapping); 15314e950f6fSAlexey Dobriyan } 15324e950f6fSAlexey Dobriyan 1533fc8744adSLinus Torvalds /* 1534fc8744adSLinus Torvalds * We account for memory if it's a private writeable mapping, 15355a6fe125SMel Gorman * not hugepages and VM_NORESERVE wasn't set. 1536fc8744adSLinus Torvalds */ 1537ca16d140SKOSAKI Motohiro static inline int accountable_mapping(struct file *file, vm_flags_t vm_flags) 1538fc8744adSLinus Torvalds { 15395a6fe125SMel Gorman /* 15405a6fe125SMel Gorman * hugetlb has its own accounting separate from the core VM 15415a6fe125SMel Gorman * VM_HUGETLB may not be set yet so we cannot check for that flag. 15425a6fe125SMel Gorman */ 15435a6fe125SMel Gorman if (file && is_file_hugepages(file)) 15445a6fe125SMel Gorman return 0; 15455a6fe125SMel Gorman 1546fc8744adSLinus Torvalds return (vm_flags & (VM_NORESERVE | VM_SHARED | VM_WRITE)) == VM_WRITE; 1547fc8744adSLinus Torvalds } 1548fc8744adSLinus Torvalds 15490165ab44SMiklos Szeredi unsigned long mmap_region(struct file *file, unsigned long addr, 1550c22c0d63SMichel Lespinasse unsigned long len, vm_flags_t vm_flags, unsigned long pgoff) 15510165ab44SMiklos Szeredi { 15520165ab44SMiklos Szeredi struct mm_struct *mm = current->mm; 15530165ab44SMiklos Szeredi struct vm_area_struct *vma, *prev; 15540165ab44SMiklos Szeredi int error; 15550165ab44SMiklos Szeredi struct rb_node **rb_link, *rb_parent; 15560165ab44SMiklos Szeredi unsigned long charged = 0; 15570165ab44SMiklos Szeredi 1558e8420a8eSCyril Hrubis /* Check against address space limit. */ 1559e8420a8eSCyril Hrubis if (!may_expand_vm(mm, len >> PAGE_SHIFT)) { 1560e8420a8eSCyril Hrubis unsigned long nr_pages; 1561e8420a8eSCyril Hrubis 1562e8420a8eSCyril Hrubis /* 1563e8420a8eSCyril Hrubis * MAP_FIXED may remove pages of mappings that intersects with 1564e8420a8eSCyril Hrubis * requested mapping. Account for the pages it would unmap. 1565e8420a8eSCyril Hrubis */ 1566e8420a8eSCyril Hrubis nr_pages = count_vma_pages_range(mm, addr, addr + len); 1567e8420a8eSCyril Hrubis 1568e8420a8eSCyril Hrubis if (!may_expand_vm(mm, (len >> PAGE_SHIFT) - nr_pages)) 1569e8420a8eSCyril Hrubis return -ENOMEM; 1570e8420a8eSCyril Hrubis } 1571e8420a8eSCyril Hrubis 15721da177e4SLinus Torvalds /* Clear old maps */ 15739fcd1457SRasmus Villemoes while (find_vma_links(mm, addr, addr + len, &prev, &rb_link, 15749fcd1457SRasmus Villemoes &rb_parent)) { 15751da177e4SLinus Torvalds if (do_munmap(mm, addr, len)) 15761da177e4SLinus Torvalds return -ENOMEM; 15771da177e4SLinus Torvalds } 15781da177e4SLinus Torvalds 1579fc8744adSLinus Torvalds /* 15801da177e4SLinus Torvalds * Private writable mapping: check memory availability 15811da177e4SLinus Torvalds */ 15825a6fe125SMel Gorman if (accountable_mapping(file, vm_flags)) { 15831da177e4SLinus Torvalds charged = len >> PAGE_SHIFT; 1584191c5424SAl Viro if (security_vm_enough_memory_mm(mm, charged)) 15851da177e4SLinus Torvalds return -ENOMEM; 15861da177e4SLinus Torvalds vm_flags |= VM_ACCOUNT; 15871da177e4SLinus Torvalds } 15881da177e4SLinus Torvalds 15891da177e4SLinus Torvalds /* 1590de33c8dbSLinus Torvalds * Can we just expand an old mapping? 15911da177e4SLinus Torvalds */ 159219a809afSAndrea Arcangeli vma = vma_merge(mm, prev, addr, addr + len, vm_flags, 159319a809afSAndrea Arcangeli NULL, file, pgoff, NULL, NULL_VM_UFFD_CTX); 1594ba470de4SRik van Riel if (vma) 15951da177e4SLinus Torvalds goto out; 15961da177e4SLinus Torvalds 15971da177e4SLinus Torvalds /* 15981da177e4SLinus Torvalds * Determine the object being mapped and call the appropriate 15991da177e4SLinus Torvalds * specific mapper. the address has already been validated, but 16001da177e4SLinus Torvalds * not unmapped, but the maps are removed from the list. 16011da177e4SLinus Torvalds */ 1602c5e3b83eSPekka Enberg vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL); 16031da177e4SLinus Torvalds if (!vma) { 16041da177e4SLinus Torvalds error = -ENOMEM; 16051da177e4SLinus Torvalds goto unacct_error; 16061da177e4SLinus Torvalds } 16071da177e4SLinus Torvalds 16081da177e4SLinus Torvalds vma->vm_mm = mm; 16091da177e4SLinus Torvalds vma->vm_start = addr; 16101da177e4SLinus Torvalds vma->vm_end = addr + len; 16111da177e4SLinus Torvalds vma->vm_flags = vm_flags; 16123ed75eb8SColy Li vma->vm_page_prot = vm_get_page_prot(vm_flags); 16131da177e4SLinus Torvalds vma->vm_pgoff = pgoff; 16145beb4930SRik van Riel INIT_LIST_HEAD(&vma->anon_vma_chain); 16151da177e4SLinus Torvalds 16161da177e4SLinus Torvalds if (file) { 16171da177e4SLinus Torvalds if (vm_flags & VM_DENYWRITE) { 16181da177e4SLinus Torvalds error = deny_write_access(file); 16191da177e4SLinus Torvalds if (error) 16201da177e4SLinus Torvalds goto free_vma; 16211da177e4SLinus Torvalds } 16224bb5f5d9SDavid Herrmann if (vm_flags & VM_SHARED) { 16234bb5f5d9SDavid Herrmann error = mapping_map_writable(file->f_mapping); 16244bb5f5d9SDavid Herrmann if (error) 16254bb5f5d9SDavid Herrmann goto allow_write_and_free_vma; 16264bb5f5d9SDavid Herrmann } 16274bb5f5d9SDavid Herrmann 16284bb5f5d9SDavid Herrmann /* ->mmap() can change vma->vm_file, but must guarantee that 16294bb5f5d9SDavid Herrmann * vma_link() below can deny write-access if VM_DENYWRITE is set 16304bb5f5d9SDavid Herrmann * and map writably if VM_SHARED is set. This usually means the 16314bb5f5d9SDavid Herrmann * new file must not have been exposed to user-space, yet. 16324bb5f5d9SDavid Herrmann */ 1633cb0942b8SAl Viro vma->vm_file = get_file(file); 16341da177e4SLinus Torvalds error = file->f_op->mmap(file, vma); 16351da177e4SLinus Torvalds if (error) 16361da177e4SLinus Torvalds goto unmap_and_free_vma; 16371da177e4SLinus Torvalds 16381da177e4SLinus Torvalds /* Can addr have changed?? 16391da177e4SLinus Torvalds * 16401da177e4SLinus Torvalds * Answer: Yes, several device drivers can do it in their 16411da177e4SLinus Torvalds * f_op->mmap method. -DaveM 16422897b4d2SJoonsoo Kim * Bug: If addr is changed, prev, rb_link, rb_parent should 16432897b4d2SJoonsoo Kim * be updated for vma_link() 16441da177e4SLinus Torvalds */ 16452897b4d2SJoonsoo Kim WARN_ON_ONCE(addr != vma->vm_start); 16462897b4d2SJoonsoo Kim 16471da177e4SLinus Torvalds addr = vma->vm_start; 16481da177e4SLinus Torvalds vm_flags = vma->vm_flags; 1649f8dbf0a7SHuang Shijie } else if (vm_flags & VM_SHARED) { 1650f8dbf0a7SHuang Shijie error = shmem_zero_setup(vma); 1651f8dbf0a7SHuang Shijie if (error) 1652f8dbf0a7SHuang Shijie goto free_vma; 1653f8dbf0a7SHuang Shijie } 16541da177e4SLinus Torvalds 16554d3d5b41SOleg Nesterov vma_link(mm, vma, prev, rb_link, rb_parent); 16564d3d5b41SOleg Nesterov /* Once vma denies write, undo our temporary denial count */ 16574bb5f5d9SDavid Herrmann if (file) { 16584bb5f5d9SDavid Herrmann if (vm_flags & VM_SHARED) 16594bb5f5d9SDavid Herrmann mapping_unmap_writable(file->f_mapping); 1660e8686772SOleg Nesterov if (vm_flags & VM_DENYWRITE) 1661e8686772SOleg Nesterov allow_write_access(file); 16624bb5f5d9SDavid Herrmann } 1663e8686772SOleg Nesterov file = vma->vm_file; 16641da177e4SLinus Torvalds out: 1665cdd6c482SIngo Molnar perf_event_mmap(vma); 16660a4a9391SPeter Zijlstra 1667ab50b8edSHugh Dickins vm_stat_account(mm, vm_flags, file, len >> PAGE_SHIFT); 16681da177e4SLinus Torvalds if (vm_flags & VM_LOCKED) { 1669bebeb3d6SMichel Lespinasse if (!((vm_flags & VM_SPECIAL) || is_vm_hugetlb_page(vma) || 1670bebeb3d6SMichel Lespinasse vma == get_gate_vma(current->mm))) 167106f9d8c2SKOSAKI Motohiro mm->locked_vm += (len >> PAGE_SHIFT); 1672bebeb3d6SMichel Lespinasse else 1673de60f5f1SEric B Munson vma->vm_flags &= VM_LOCKED_CLEAR_MASK; 1674bebeb3d6SMichel Lespinasse } 16752b144498SSrikar Dronamraju 1676c7a3a88cSOleg Nesterov if (file) 1677c7a3a88cSOleg Nesterov uprobe_mmap(vma); 16782b144498SSrikar Dronamraju 1679d9104d1cSCyrill Gorcunov /* 1680d9104d1cSCyrill Gorcunov * New (or expanded) vma always get soft dirty status. 1681d9104d1cSCyrill Gorcunov * Otherwise user-space soft-dirty page tracker won't 1682d9104d1cSCyrill Gorcunov * be able to distinguish situation when vma area unmapped, 1683d9104d1cSCyrill Gorcunov * then new mapped in-place (which must be aimed as 1684d9104d1cSCyrill Gorcunov * a completely new data area). 1685d9104d1cSCyrill Gorcunov */ 1686d9104d1cSCyrill Gorcunov vma->vm_flags |= VM_SOFTDIRTY; 1687d9104d1cSCyrill Gorcunov 168864e45507SPeter Feiner vma_set_page_prot(vma); 168964e45507SPeter Feiner 16901da177e4SLinus Torvalds return addr; 16911da177e4SLinus Torvalds 16921da177e4SLinus Torvalds unmap_and_free_vma: 16931da177e4SLinus Torvalds vma->vm_file = NULL; 16941da177e4SLinus Torvalds fput(file); 16951da177e4SLinus Torvalds 16961da177e4SLinus Torvalds /* Undo any partial mapping done by a device driver. */ 1697e0da382cSHugh Dickins unmap_region(mm, vma, prev, vma->vm_start, vma->vm_end); 1698e0da382cSHugh Dickins charged = 0; 16994bb5f5d9SDavid Herrmann if (vm_flags & VM_SHARED) 17004bb5f5d9SDavid Herrmann mapping_unmap_writable(file->f_mapping); 17014bb5f5d9SDavid Herrmann allow_write_and_free_vma: 17024bb5f5d9SDavid Herrmann if (vm_flags & VM_DENYWRITE) 17034bb5f5d9SDavid Herrmann allow_write_access(file); 17041da177e4SLinus Torvalds free_vma: 17051da177e4SLinus Torvalds kmem_cache_free(vm_area_cachep, vma); 17061da177e4SLinus Torvalds unacct_error: 17071da177e4SLinus Torvalds if (charged) 17081da177e4SLinus Torvalds vm_unacct_memory(charged); 17091da177e4SLinus Torvalds return error; 17101da177e4SLinus Torvalds } 17111da177e4SLinus Torvalds 1712db4fbfb9SMichel Lespinasse unsigned long unmapped_area(struct vm_unmapped_area_info *info) 1713db4fbfb9SMichel Lespinasse { 1714db4fbfb9SMichel Lespinasse /* 1715db4fbfb9SMichel Lespinasse * We implement the search by looking for an rbtree node that 1716db4fbfb9SMichel Lespinasse * immediately follows a suitable gap. That is, 1717db4fbfb9SMichel Lespinasse * - gap_start = vma->vm_prev->vm_end <= info->high_limit - length; 1718db4fbfb9SMichel Lespinasse * - gap_end = vma->vm_start >= info->low_limit + length; 1719db4fbfb9SMichel Lespinasse * - gap_end - gap_start >= length 1720db4fbfb9SMichel Lespinasse */ 1721db4fbfb9SMichel Lespinasse 1722db4fbfb9SMichel Lespinasse struct mm_struct *mm = current->mm; 1723db4fbfb9SMichel Lespinasse struct vm_area_struct *vma; 1724db4fbfb9SMichel Lespinasse unsigned long length, low_limit, high_limit, gap_start, gap_end; 1725db4fbfb9SMichel Lespinasse 1726db4fbfb9SMichel Lespinasse /* Adjust search length to account for worst case alignment overhead */ 1727db4fbfb9SMichel Lespinasse length = info->length + info->align_mask; 1728db4fbfb9SMichel Lespinasse if (length < info->length) 1729db4fbfb9SMichel Lespinasse return -ENOMEM; 1730db4fbfb9SMichel Lespinasse 1731db4fbfb9SMichel Lespinasse /* Adjust search limits by the desired length */ 1732db4fbfb9SMichel Lespinasse if (info->high_limit < length) 1733db4fbfb9SMichel Lespinasse return -ENOMEM; 1734db4fbfb9SMichel Lespinasse high_limit = info->high_limit - length; 1735db4fbfb9SMichel Lespinasse 1736db4fbfb9SMichel Lespinasse if (info->low_limit > high_limit) 1737db4fbfb9SMichel Lespinasse return -ENOMEM; 1738db4fbfb9SMichel Lespinasse low_limit = info->low_limit + length; 1739db4fbfb9SMichel Lespinasse 1740db4fbfb9SMichel Lespinasse /* Check if rbtree root looks promising */ 1741db4fbfb9SMichel Lespinasse if (RB_EMPTY_ROOT(&mm->mm_rb)) 1742db4fbfb9SMichel Lespinasse goto check_highest; 1743db4fbfb9SMichel Lespinasse vma = rb_entry(mm->mm_rb.rb_node, struct vm_area_struct, vm_rb); 1744db4fbfb9SMichel Lespinasse if (vma->rb_subtree_gap < length) 1745db4fbfb9SMichel Lespinasse goto check_highest; 1746db4fbfb9SMichel Lespinasse 1747db4fbfb9SMichel Lespinasse while (true) { 1748db4fbfb9SMichel Lespinasse /* Visit left subtree if it looks promising */ 1749db4fbfb9SMichel Lespinasse gap_end = vma->vm_start; 1750db4fbfb9SMichel Lespinasse if (gap_end >= low_limit && vma->vm_rb.rb_left) { 1751db4fbfb9SMichel Lespinasse struct vm_area_struct *left = 1752db4fbfb9SMichel Lespinasse rb_entry(vma->vm_rb.rb_left, 1753db4fbfb9SMichel Lespinasse struct vm_area_struct, vm_rb); 1754db4fbfb9SMichel Lespinasse if (left->rb_subtree_gap >= length) { 1755db4fbfb9SMichel Lespinasse vma = left; 1756db4fbfb9SMichel Lespinasse continue; 1757db4fbfb9SMichel Lespinasse } 1758db4fbfb9SMichel Lespinasse } 1759db4fbfb9SMichel Lespinasse 1760db4fbfb9SMichel Lespinasse gap_start = vma->vm_prev ? vma->vm_prev->vm_end : 0; 1761db4fbfb9SMichel Lespinasse check_current: 1762db4fbfb9SMichel Lespinasse /* Check if current node has a suitable gap */ 1763db4fbfb9SMichel Lespinasse if (gap_start > high_limit) 1764db4fbfb9SMichel Lespinasse return -ENOMEM; 1765db4fbfb9SMichel Lespinasse if (gap_end >= low_limit && gap_end - gap_start >= length) 1766db4fbfb9SMichel Lespinasse goto found; 1767db4fbfb9SMichel Lespinasse 1768db4fbfb9SMichel Lespinasse /* Visit right subtree if it looks promising */ 1769db4fbfb9SMichel Lespinasse if (vma->vm_rb.rb_right) { 1770db4fbfb9SMichel Lespinasse struct vm_area_struct *right = 1771db4fbfb9SMichel Lespinasse rb_entry(vma->vm_rb.rb_right, 1772db4fbfb9SMichel Lespinasse struct vm_area_struct, vm_rb); 1773db4fbfb9SMichel Lespinasse if (right->rb_subtree_gap >= length) { 1774db4fbfb9SMichel Lespinasse vma = right; 1775db4fbfb9SMichel Lespinasse continue; 1776db4fbfb9SMichel Lespinasse } 1777db4fbfb9SMichel Lespinasse } 1778db4fbfb9SMichel Lespinasse 1779db4fbfb9SMichel Lespinasse /* Go back up the rbtree to find next candidate node */ 1780db4fbfb9SMichel Lespinasse while (true) { 1781db4fbfb9SMichel Lespinasse struct rb_node *prev = &vma->vm_rb; 1782db4fbfb9SMichel Lespinasse if (!rb_parent(prev)) 1783db4fbfb9SMichel Lespinasse goto check_highest; 1784db4fbfb9SMichel Lespinasse vma = rb_entry(rb_parent(prev), 1785db4fbfb9SMichel Lespinasse struct vm_area_struct, vm_rb); 1786db4fbfb9SMichel Lespinasse if (prev == vma->vm_rb.rb_left) { 1787db4fbfb9SMichel Lespinasse gap_start = vma->vm_prev->vm_end; 1788db4fbfb9SMichel Lespinasse gap_end = vma->vm_start; 1789db4fbfb9SMichel Lespinasse goto check_current; 1790db4fbfb9SMichel Lespinasse } 1791db4fbfb9SMichel Lespinasse } 1792db4fbfb9SMichel Lespinasse } 1793db4fbfb9SMichel Lespinasse 1794db4fbfb9SMichel Lespinasse check_highest: 1795db4fbfb9SMichel Lespinasse /* Check highest gap, which does not precede any rbtree node */ 1796db4fbfb9SMichel Lespinasse gap_start = mm->highest_vm_end; 1797db4fbfb9SMichel Lespinasse gap_end = ULONG_MAX; /* Only for VM_BUG_ON below */ 1798db4fbfb9SMichel Lespinasse if (gap_start > high_limit) 1799db4fbfb9SMichel Lespinasse return -ENOMEM; 1800db4fbfb9SMichel Lespinasse 1801db4fbfb9SMichel Lespinasse found: 1802db4fbfb9SMichel Lespinasse /* We found a suitable gap. Clip it with the original low_limit. */ 1803db4fbfb9SMichel Lespinasse if (gap_start < info->low_limit) 1804db4fbfb9SMichel Lespinasse gap_start = info->low_limit; 1805db4fbfb9SMichel Lespinasse 1806db4fbfb9SMichel Lespinasse /* Adjust gap address to the desired alignment */ 1807db4fbfb9SMichel Lespinasse gap_start += (info->align_offset - gap_start) & info->align_mask; 1808db4fbfb9SMichel Lespinasse 1809db4fbfb9SMichel Lespinasse VM_BUG_ON(gap_start + info->length > info->high_limit); 1810db4fbfb9SMichel Lespinasse VM_BUG_ON(gap_start + info->length > gap_end); 1811db4fbfb9SMichel Lespinasse return gap_start; 1812db4fbfb9SMichel Lespinasse } 1813db4fbfb9SMichel Lespinasse 1814db4fbfb9SMichel Lespinasse unsigned long unmapped_area_topdown(struct vm_unmapped_area_info *info) 1815db4fbfb9SMichel Lespinasse { 1816db4fbfb9SMichel Lespinasse struct mm_struct *mm = current->mm; 1817db4fbfb9SMichel Lespinasse struct vm_area_struct *vma; 1818db4fbfb9SMichel Lespinasse unsigned long length, low_limit, high_limit, gap_start, gap_end; 1819db4fbfb9SMichel Lespinasse 1820db4fbfb9SMichel Lespinasse /* Adjust search length to account for worst case alignment overhead */ 1821db4fbfb9SMichel Lespinasse length = info->length + info->align_mask; 1822db4fbfb9SMichel Lespinasse if (length < info->length) 1823db4fbfb9SMichel Lespinasse return -ENOMEM; 1824db4fbfb9SMichel Lespinasse 1825db4fbfb9SMichel Lespinasse /* 1826db4fbfb9SMichel Lespinasse * Adjust search limits by the desired length. 1827db4fbfb9SMichel Lespinasse * See implementation comment at top of unmapped_area(). 1828db4fbfb9SMichel Lespinasse */ 1829db4fbfb9SMichel Lespinasse gap_end = info->high_limit; 1830db4fbfb9SMichel Lespinasse if (gap_end < length) 1831db4fbfb9SMichel Lespinasse return -ENOMEM; 1832db4fbfb9SMichel Lespinasse high_limit = gap_end - length; 1833db4fbfb9SMichel Lespinasse 1834db4fbfb9SMichel Lespinasse if (info->low_limit > high_limit) 1835db4fbfb9SMichel Lespinasse return -ENOMEM; 1836db4fbfb9SMichel Lespinasse low_limit = info->low_limit + length; 1837db4fbfb9SMichel Lespinasse 1838db4fbfb9SMichel Lespinasse /* Check highest gap, which does not precede any rbtree node */ 1839db4fbfb9SMichel Lespinasse gap_start = mm->highest_vm_end; 1840db4fbfb9SMichel Lespinasse if (gap_start <= high_limit) 1841db4fbfb9SMichel Lespinasse goto found_highest; 1842db4fbfb9SMichel Lespinasse 1843db4fbfb9SMichel Lespinasse /* Check if rbtree root looks promising */ 1844db4fbfb9SMichel Lespinasse if (RB_EMPTY_ROOT(&mm->mm_rb)) 1845db4fbfb9SMichel Lespinasse return -ENOMEM; 1846db4fbfb9SMichel Lespinasse vma = rb_entry(mm->mm_rb.rb_node, struct vm_area_struct, vm_rb); 1847db4fbfb9SMichel Lespinasse if (vma->rb_subtree_gap < length) 1848db4fbfb9SMichel Lespinasse return -ENOMEM; 1849db4fbfb9SMichel Lespinasse 1850db4fbfb9SMichel Lespinasse while (true) { 1851db4fbfb9SMichel Lespinasse /* Visit right subtree if it looks promising */ 1852db4fbfb9SMichel Lespinasse gap_start = vma->vm_prev ? vma->vm_prev->vm_end : 0; 1853db4fbfb9SMichel Lespinasse if (gap_start <= high_limit && vma->vm_rb.rb_right) { 1854db4fbfb9SMichel Lespinasse struct vm_area_struct *right = 1855db4fbfb9SMichel Lespinasse rb_entry(vma->vm_rb.rb_right, 1856db4fbfb9SMichel Lespinasse struct vm_area_struct, vm_rb); 1857db4fbfb9SMichel Lespinasse if (right->rb_subtree_gap >= length) { 1858db4fbfb9SMichel Lespinasse vma = right; 1859db4fbfb9SMichel Lespinasse continue; 1860db4fbfb9SMichel Lespinasse } 1861db4fbfb9SMichel Lespinasse } 1862db4fbfb9SMichel Lespinasse 1863db4fbfb9SMichel Lespinasse check_current: 1864db4fbfb9SMichel Lespinasse /* Check if current node has a suitable gap */ 1865db4fbfb9SMichel Lespinasse gap_end = vma->vm_start; 1866db4fbfb9SMichel Lespinasse if (gap_end < low_limit) 1867db4fbfb9SMichel Lespinasse return -ENOMEM; 1868db4fbfb9SMichel Lespinasse if (gap_start <= high_limit && gap_end - gap_start >= length) 1869db4fbfb9SMichel Lespinasse goto found; 1870db4fbfb9SMichel Lespinasse 1871db4fbfb9SMichel Lespinasse /* Visit left subtree if it looks promising */ 1872db4fbfb9SMichel Lespinasse if (vma->vm_rb.rb_left) { 1873db4fbfb9SMichel Lespinasse struct vm_area_struct *left = 1874db4fbfb9SMichel Lespinasse rb_entry(vma->vm_rb.rb_left, 1875db4fbfb9SMichel Lespinasse struct vm_area_struct, vm_rb); 1876db4fbfb9SMichel Lespinasse if (left->rb_subtree_gap >= length) { 1877db4fbfb9SMichel Lespinasse vma = left; 1878db4fbfb9SMichel Lespinasse continue; 1879db4fbfb9SMichel Lespinasse } 1880db4fbfb9SMichel Lespinasse } 1881db4fbfb9SMichel Lespinasse 1882db4fbfb9SMichel Lespinasse /* Go back up the rbtree to find next candidate node */ 1883db4fbfb9SMichel Lespinasse while (true) { 1884db4fbfb9SMichel Lespinasse struct rb_node *prev = &vma->vm_rb; 1885db4fbfb9SMichel Lespinasse if (!rb_parent(prev)) 1886db4fbfb9SMichel Lespinasse return -ENOMEM; 1887db4fbfb9SMichel Lespinasse vma = rb_entry(rb_parent(prev), 1888db4fbfb9SMichel Lespinasse struct vm_area_struct, vm_rb); 1889db4fbfb9SMichel Lespinasse if (prev == vma->vm_rb.rb_right) { 1890db4fbfb9SMichel Lespinasse gap_start = vma->vm_prev ? 1891db4fbfb9SMichel Lespinasse vma->vm_prev->vm_end : 0; 1892db4fbfb9SMichel Lespinasse goto check_current; 1893db4fbfb9SMichel Lespinasse } 1894db4fbfb9SMichel Lespinasse } 1895db4fbfb9SMichel Lespinasse } 1896db4fbfb9SMichel Lespinasse 1897db4fbfb9SMichel Lespinasse found: 1898db4fbfb9SMichel Lespinasse /* We found a suitable gap. Clip it with the original high_limit. */ 1899db4fbfb9SMichel Lespinasse if (gap_end > info->high_limit) 1900db4fbfb9SMichel Lespinasse gap_end = info->high_limit; 1901db4fbfb9SMichel Lespinasse 1902db4fbfb9SMichel Lespinasse found_highest: 1903db4fbfb9SMichel Lespinasse /* Compute highest gap address at the desired alignment */ 1904db4fbfb9SMichel Lespinasse gap_end -= info->length; 1905db4fbfb9SMichel Lespinasse gap_end -= (gap_end - info->align_offset) & info->align_mask; 1906db4fbfb9SMichel Lespinasse 1907db4fbfb9SMichel Lespinasse VM_BUG_ON(gap_end < info->low_limit); 1908db4fbfb9SMichel Lespinasse VM_BUG_ON(gap_end < gap_start); 1909db4fbfb9SMichel Lespinasse return gap_end; 1910db4fbfb9SMichel Lespinasse } 1911db4fbfb9SMichel Lespinasse 19121da177e4SLinus Torvalds /* Get an address range which is currently unmapped. 19131da177e4SLinus Torvalds * For shmat() with addr=0. 19141da177e4SLinus Torvalds * 19151da177e4SLinus Torvalds * Ugly calling convention alert: 19161da177e4SLinus Torvalds * Return value with the low bits set means error value, 19171da177e4SLinus Torvalds * ie 19181da177e4SLinus Torvalds * if (ret & ~PAGE_MASK) 19191da177e4SLinus Torvalds * error = ret; 19201da177e4SLinus Torvalds * 19211da177e4SLinus Torvalds * This function "knows" that -ENOMEM has the bits set. 19221da177e4SLinus Torvalds */ 19231da177e4SLinus Torvalds #ifndef HAVE_ARCH_UNMAPPED_AREA 19241da177e4SLinus Torvalds unsigned long 19251da177e4SLinus Torvalds arch_get_unmapped_area(struct file *filp, unsigned long addr, 19261da177e4SLinus Torvalds unsigned long len, unsigned long pgoff, unsigned long flags) 19271da177e4SLinus Torvalds { 19281da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 19291da177e4SLinus Torvalds struct vm_area_struct *vma; 1930db4fbfb9SMichel Lespinasse struct vm_unmapped_area_info info; 19311da177e4SLinus Torvalds 19322afc745fSAkira Takeuchi if (len > TASK_SIZE - mmap_min_addr) 19331da177e4SLinus Torvalds return -ENOMEM; 19341da177e4SLinus Torvalds 193506abdfb4SBenjamin Herrenschmidt if (flags & MAP_FIXED) 193606abdfb4SBenjamin Herrenschmidt return addr; 193706abdfb4SBenjamin Herrenschmidt 19381da177e4SLinus Torvalds if (addr) { 19391da177e4SLinus Torvalds addr = PAGE_ALIGN(addr); 19401da177e4SLinus Torvalds vma = find_vma(mm, addr); 19412afc745fSAkira Takeuchi if (TASK_SIZE - len >= addr && addr >= mmap_min_addr && 19421da177e4SLinus Torvalds (!vma || addr + len <= vma->vm_start)) 19431da177e4SLinus Torvalds return addr; 19441da177e4SLinus Torvalds } 19451da177e4SLinus Torvalds 1946db4fbfb9SMichel Lespinasse info.flags = 0; 1947db4fbfb9SMichel Lespinasse info.length = len; 19484e99b021SHeiko Carstens info.low_limit = mm->mmap_base; 1949db4fbfb9SMichel Lespinasse info.high_limit = TASK_SIZE; 1950db4fbfb9SMichel Lespinasse info.align_mask = 0; 1951db4fbfb9SMichel Lespinasse return vm_unmapped_area(&info); 19521da177e4SLinus Torvalds } 19531da177e4SLinus Torvalds #endif 19541da177e4SLinus Torvalds 19551da177e4SLinus Torvalds /* 19561da177e4SLinus Torvalds * This mmap-allocator allocates new areas top-down from below the 19571da177e4SLinus Torvalds * stack's low limit (the base): 19581da177e4SLinus Torvalds */ 19591da177e4SLinus Torvalds #ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN 19601da177e4SLinus Torvalds unsigned long 19611da177e4SLinus Torvalds arch_get_unmapped_area_topdown(struct file *filp, const unsigned long addr0, 19621da177e4SLinus Torvalds const unsigned long len, const unsigned long pgoff, 19631da177e4SLinus Torvalds const unsigned long flags) 19641da177e4SLinus Torvalds { 19651da177e4SLinus Torvalds struct vm_area_struct *vma; 19661da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 1967db4fbfb9SMichel Lespinasse unsigned long addr = addr0; 1968db4fbfb9SMichel Lespinasse struct vm_unmapped_area_info info; 19691da177e4SLinus Torvalds 19701da177e4SLinus Torvalds /* requested length too big for entire address space */ 19712afc745fSAkira Takeuchi if (len > TASK_SIZE - mmap_min_addr) 19721da177e4SLinus Torvalds return -ENOMEM; 19731da177e4SLinus Torvalds 197406abdfb4SBenjamin Herrenschmidt if (flags & MAP_FIXED) 197506abdfb4SBenjamin Herrenschmidt return addr; 197606abdfb4SBenjamin Herrenschmidt 19771da177e4SLinus Torvalds /* requesting a specific address */ 19781da177e4SLinus Torvalds if (addr) { 19791da177e4SLinus Torvalds addr = PAGE_ALIGN(addr); 19801da177e4SLinus Torvalds vma = find_vma(mm, addr); 19812afc745fSAkira Takeuchi if (TASK_SIZE - len >= addr && addr >= mmap_min_addr && 19821da177e4SLinus Torvalds (!vma || addr + len <= vma->vm_start)) 19831da177e4SLinus Torvalds return addr; 19841da177e4SLinus Torvalds } 19851da177e4SLinus Torvalds 1986db4fbfb9SMichel Lespinasse info.flags = VM_UNMAPPED_AREA_TOPDOWN; 1987db4fbfb9SMichel Lespinasse info.length = len; 19882afc745fSAkira Takeuchi info.low_limit = max(PAGE_SIZE, mmap_min_addr); 1989db4fbfb9SMichel Lespinasse info.high_limit = mm->mmap_base; 1990db4fbfb9SMichel Lespinasse info.align_mask = 0; 1991db4fbfb9SMichel Lespinasse addr = vm_unmapped_area(&info); 1992b716ad95SXiao Guangrong 19931da177e4SLinus Torvalds /* 19941da177e4SLinus Torvalds * A failed mmap() very likely causes application failure, 19951da177e4SLinus Torvalds * so fall back to the bottom-up function here. This scenario 19961da177e4SLinus Torvalds * can happen with large stack limits and large mmap() 19971da177e4SLinus Torvalds * allocations. 19981da177e4SLinus Torvalds */ 1999de1741a1SAlexander Kuleshov if (offset_in_page(addr)) { 2000db4fbfb9SMichel Lespinasse VM_BUG_ON(addr != -ENOMEM); 2001db4fbfb9SMichel Lespinasse info.flags = 0; 2002db4fbfb9SMichel Lespinasse info.low_limit = TASK_UNMAPPED_BASE; 2003db4fbfb9SMichel Lespinasse info.high_limit = TASK_SIZE; 2004db4fbfb9SMichel Lespinasse addr = vm_unmapped_area(&info); 2005db4fbfb9SMichel Lespinasse } 20061da177e4SLinus Torvalds 20071da177e4SLinus Torvalds return addr; 20081da177e4SLinus Torvalds } 20091da177e4SLinus Torvalds #endif 20101da177e4SLinus Torvalds 20111da177e4SLinus Torvalds unsigned long 20121da177e4SLinus Torvalds get_unmapped_area(struct file *file, unsigned long addr, unsigned long len, 20131da177e4SLinus Torvalds unsigned long pgoff, unsigned long flags) 20141da177e4SLinus Torvalds { 201506abdfb4SBenjamin Herrenschmidt unsigned long (*get_area)(struct file *, unsigned long, 201606abdfb4SBenjamin Herrenschmidt unsigned long, unsigned long, unsigned long); 201707ab67c8SLinus Torvalds 20189206de95SAl Viro unsigned long error = arch_mmap_check(addr, len, flags); 20199206de95SAl Viro if (error) 20209206de95SAl Viro return error; 20219206de95SAl Viro 20229206de95SAl Viro /* Careful about overflows.. */ 20239206de95SAl Viro if (len > TASK_SIZE) 20249206de95SAl Viro return -ENOMEM; 20259206de95SAl Viro 202607ab67c8SLinus Torvalds get_area = current->mm->get_unmapped_area; 202772c2d531SAl Viro if (file && file->f_op->get_unmapped_area) 202807ab67c8SLinus Torvalds get_area = file->f_op->get_unmapped_area; 202907ab67c8SLinus Torvalds addr = get_area(file, addr, len, pgoff, flags); 203007ab67c8SLinus Torvalds if (IS_ERR_VALUE(addr)) 203107ab67c8SLinus Torvalds return addr; 203207ab67c8SLinus Torvalds 20331da177e4SLinus Torvalds if (addr > TASK_SIZE - len) 20341da177e4SLinus Torvalds return -ENOMEM; 2035de1741a1SAlexander Kuleshov if (offset_in_page(addr)) 20361da177e4SLinus Torvalds return -EINVAL; 203706abdfb4SBenjamin Herrenschmidt 20389ac4ed4bSAl Viro addr = arch_rebalance_pgtables(addr, len); 20399ac4ed4bSAl Viro error = security_mmap_addr(addr); 20409ac4ed4bSAl Viro return error ? error : addr; 20411da177e4SLinus Torvalds } 20421da177e4SLinus Torvalds 20431da177e4SLinus Torvalds EXPORT_SYMBOL(get_unmapped_area); 20441da177e4SLinus Torvalds 20451da177e4SLinus Torvalds /* Look up the first VMA which satisfies addr < vm_end, NULL if none. */ 20461da177e4SLinus Torvalds struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr) 20471da177e4SLinus Torvalds { 2048615d6e87SDavidlohr Bueso struct rb_node *rb_node; 2049615d6e87SDavidlohr Bueso struct vm_area_struct *vma; 20501da177e4SLinus Torvalds 20511da177e4SLinus Torvalds /* Check the cache first. */ 2052615d6e87SDavidlohr Bueso vma = vmacache_find(mm, addr); 2053615d6e87SDavidlohr Bueso if (likely(vma)) 2054615d6e87SDavidlohr Bueso return vma; 20551da177e4SLinus Torvalds 20561da177e4SLinus Torvalds rb_node = mm->mm_rb.rb_node; 20571da177e4SLinus Torvalds 20581da177e4SLinus Torvalds while (rb_node) { 2059615d6e87SDavidlohr Bueso struct vm_area_struct *tmp; 20601da177e4SLinus Torvalds 2061615d6e87SDavidlohr Bueso tmp = rb_entry(rb_node, struct vm_area_struct, vm_rb); 20621da177e4SLinus Torvalds 2063615d6e87SDavidlohr Bueso if (tmp->vm_end > addr) { 2064615d6e87SDavidlohr Bueso vma = tmp; 2065615d6e87SDavidlohr Bueso if (tmp->vm_start <= addr) 20661da177e4SLinus Torvalds break; 20671da177e4SLinus Torvalds rb_node = rb_node->rb_left; 20681da177e4SLinus Torvalds } else 20691da177e4SLinus Torvalds rb_node = rb_node->rb_right; 20701da177e4SLinus Torvalds } 2071615d6e87SDavidlohr Bueso 20721da177e4SLinus Torvalds if (vma) 2073615d6e87SDavidlohr Bueso vmacache_update(addr, vma); 20741da177e4SLinus Torvalds return vma; 20751da177e4SLinus Torvalds } 20761da177e4SLinus Torvalds 20771da177e4SLinus Torvalds EXPORT_SYMBOL(find_vma); 20781da177e4SLinus Torvalds 20796bd4837dSKOSAKI Motohiro /* 20806bd4837dSKOSAKI Motohiro * Same as find_vma, but also return a pointer to the previous VMA in *pprev. 20816bd4837dSKOSAKI Motohiro */ 20821da177e4SLinus Torvalds struct vm_area_struct * 20831da177e4SLinus Torvalds find_vma_prev(struct mm_struct *mm, unsigned long addr, 20841da177e4SLinus Torvalds struct vm_area_struct **pprev) 20851da177e4SLinus Torvalds { 20866bd4837dSKOSAKI Motohiro struct vm_area_struct *vma; 20871da177e4SLinus Torvalds 20886bd4837dSKOSAKI Motohiro vma = find_vma(mm, addr); 208983cd904dSMikulas Patocka if (vma) { 209083cd904dSMikulas Patocka *pprev = vma->vm_prev; 209183cd904dSMikulas Patocka } else { 209283cd904dSMikulas Patocka struct rb_node *rb_node = mm->mm_rb.rb_node; 209383cd904dSMikulas Patocka *pprev = NULL; 209483cd904dSMikulas Patocka while (rb_node) { 209583cd904dSMikulas Patocka *pprev = rb_entry(rb_node, struct vm_area_struct, vm_rb); 209683cd904dSMikulas Patocka rb_node = rb_node->rb_right; 209783cd904dSMikulas Patocka } 209883cd904dSMikulas Patocka } 20996bd4837dSKOSAKI Motohiro return vma; 21001da177e4SLinus Torvalds } 21011da177e4SLinus Torvalds 21021da177e4SLinus Torvalds /* 21031da177e4SLinus Torvalds * Verify that the stack growth is acceptable and 21041da177e4SLinus Torvalds * update accounting. This is shared with both the 21051da177e4SLinus Torvalds * grow-up and grow-down cases. 21061da177e4SLinus Torvalds */ 21071da177e4SLinus Torvalds static int acct_stack_growth(struct vm_area_struct *vma, unsigned long size, unsigned long grow) 21081da177e4SLinus Torvalds { 21091da177e4SLinus Torvalds struct mm_struct *mm = vma->vm_mm; 21101da177e4SLinus Torvalds struct rlimit *rlim = current->signal->rlim; 2111690eac53SLinus Torvalds unsigned long new_start, actual_size; 21121da177e4SLinus Torvalds 21131da177e4SLinus Torvalds /* address space limit tests */ 2114119f657cSakpm@osdl.org if (!may_expand_vm(mm, grow)) 21151da177e4SLinus Torvalds return -ENOMEM; 21161da177e4SLinus Torvalds 21171da177e4SLinus Torvalds /* Stack limit test */ 2118690eac53SLinus Torvalds actual_size = size; 2119690eac53SLinus Torvalds if (size && (vma->vm_flags & (VM_GROWSUP | VM_GROWSDOWN))) 2120690eac53SLinus Torvalds actual_size -= PAGE_SIZE; 21214db0c3c2SJason Low if (actual_size > READ_ONCE(rlim[RLIMIT_STACK].rlim_cur)) 21221da177e4SLinus Torvalds return -ENOMEM; 21231da177e4SLinus Torvalds 21241da177e4SLinus Torvalds /* mlock limit tests */ 21251da177e4SLinus Torvalds if (vma->vm_flags & VM_LOCKED) { 21261da177e4SLinus Torvalds unsigned long locked; 21271da177e4SLinus Torvalds unsigned long limit; 21281da177e4SLinus Torvalds locked = mm->locked_vm + grow; 21294db0c3c2SJason Low limit = READ_ONCE(rlim[RLIMIT_MEMLOCK].rlim_cur); 213059e99e5bSJiri Slaby limit >>= PAGE_SHIFT; 21311da177e4SLinus Torvalds if (locked > limit && !capable(CAP_IPC_LOCK)) 21321da177e4SLinus Torvalds return -ENOMEM; 21331da177e4SLinus Torvalds } 21341da177e4SLinus Torvalds 21350d59a01bSAdam Litke /* Check to ensure the stack will not grow into a hugetlb-only region */ 21360d59a01bSAdam Litke new_start = (vma->vm_flags & VM_GROWSUP) ? vma->vm_start : 21370d59a01bSAdam Litke vma->vm_end - size; 21380d59a01bSAdam Litke if (is_hugepage_only_range(vma->vm_mm, new_start, size)) 21390d59a01bSAdam Litke return -EFAULT; 21400d59a01bSAdam Litke 21411da177e4SLinus Torvalds /* 21421da177e4SLinus Torvalds * Overcommit.. This must be the final test, as it will 21431da177e4SLinus Torvalds * update security statistics. 21441da177e4SLinus Torvalds */ 214505fa199dSHugh Dickins if (security_vm_enough_memory_mm(mm, grow)) 21461da177e4SLinus Torvalds return -ENOMEM; 21471da177e4SLinus Torvalds 21481da177e4SLinus Torvalds return 0; 21491da177e4SLinus Torvalds } 21501da177e4SLinus Torvalds 215146dea3d0SHugh Dickins #if defined(CONFIG_STACK_GROWSUP) || defined(CONFIG_IA64) 21521da177e4SLinus Torvalds /* 215346dea3d0SHugh Dickins * PA-RISC uses this for its stack; IA64 for its Register Backing Store. 215446dea3d0SHugh Dickins * vma is the last one with address > vma->vm_end. Have to extend vma. 21551da177e4SLinus Torvalds */ 215646dea3d0SHugh Dickins int expand_upwards(struct vm_area_struct *vma, unsigned long address) 21571da177e4SLinus Torvalds { 215809357814SOleg Nesterov struct mm_struct *mm = vma->vm_mm; 21591da177e4SLinus Torvalds int error; 21601da177e4SLinus Torvalds 21611da177e4SLinus Torvalds if (!(vma->vm_flags & VM_GROWSUP)) 21621da177e4SLinus Torvalds return -EFAULT; 21631da177e4SLinus Torvalds 21641da177e4SLinus Torvalds /* 21651da177e4SLinus Torvalds * We must make sure the anon_vma is allocated 21661da177e4SLinus Torvalds * so that the anon_vma locking is not a noop. 21671da177e4SLinus Torvalds */ 21681da177e4SLinus Torvalds if (unlikely(anon_vma_prepare(vma))) 21691da177e4SLinus Torvalds return -ENOMEM; 2170bb4a340eSRik van Riel vma_lock_anon_vma(vma); 21711da177e4SLinus Torvalds 21721da177e4SLinus Torvalds /* 21731da177e4SLinus Torvalds * vma->vm_start/vm_end cannot change under us because the caller 21741da177e4SLinus Torvalds * is required to hold the mmap_sem in read mode. We need the 21751da177e4SLinus Torvalds * anon_vma lock to serialize against concurrent expand_stacks. 217606b32f3aSHelge Deller * Also guard against wrapping around to address 0. 21771da177e4SLinus Torvalds */ 217806b32f3aSHelge Deller if (address < PAGE_ALIGN(address+4)) 217906b32f3aSHelge Deller address = PAGE_ALIGN(address+4); 218006b32f3aSHelge Deller else { 2181bb4a340eSRik van Riel vma_unlock_anon_vma(vma); 218206b32f3aSHelge Deller return -ENOMEM; 218306b32f3aSHelge Deller } 21841da177e4SLinus Torvalds error = 0; 21851da177e4SLinus Torvalds 21861da177e4SLinus Torvalds /* Somebody else might have raced and expanded it already */ 21871da177e4SLinus Torvalds if (address > vma->vm_end) { 21881da177e4SLinus Torvalds unsigned long size, grow; 21891da177e4SLinus Torvalds 21901da177e4SLinus Torvalds size = address - vma->vm_start; 21911da177e4SLinus Torvalds grow = (address - vma->vm_end) >> PAGE_SHIFT; 21921da177e4SLinus Torvalds 219342c36f63SHugh Dickins error = -ENOMEM; 219442c36f63SHugh Dickins if (vma->vm_pgoff + (size >> PAGE_SHIFT) >= vma->vm_pgoff) { 21951da177e4SLinus Torvalds error = acct_stack_growth(vma, size, grow); 21963af9e859SEric B Munson if (!error) { 21974128997bSMichel Lespinasse /* 21984128997bSMichel Lespinasse * vma_gap_update() doesn't support concurrent 21994128997bSMichel Lespinasse * updates, but we only hold a shared mmap_sem 22004128997bSMichel Lespinasse * lock here, so we need to protect against 22014128997bSMichel Lespinasse * concurrent vma expansions. 22024128997bSMichel Lespinasse * vma_lock_anon_vma() doesn't help here, as 22034128997bSMichel Lespinasse * we don't guarantee that all growable vmas 22044128997bSMichel Lespinasse * in a mm share the same root anon vma. 22054128997bSMichel Lespinasse * So, we reuse mm->page_table_lock to guard 22064128997bSMichel Lespinasse * against concurrent vma expansions. 22074128997bSMichel Lespinasse */ 220809357814SOleg Nesterov spin_lock(&mm->page_table_lock); 220987e8827bSOleg Nesterov if (vma->vm_flags & VM_LOCKED) 221009357814SOleg Nesterov mm->locked_vm += grow; 221109357814SOleg Nesterov vm_stat_account(mm, vma->vm_flags, 221287e8827bSOleg Nesterov vma->vm_file, grow); 2213bf181b9fSMichel Lespinasse anon_vma_interval_tree_pre_update_vma(vma); 22141da177e4SLinus Torvalds vma->vm_end = address; 2215bf181b9fSMichel Lespinasse anon_vma_interval_tree_post_update_vma(vma); 2216d3737187SMichel Lespinasse if (vma->vm_next) 2217d3737187SMichel Lespinasse vma_gap_update(vma->vm_next); 2218d3737187SMichel Lespinasse else 221909357814SOleg Nesterov mm->highest_vm_end = address; 222009357814SOleg Nesterov spin_unlock(&mm->page_table_lock); 22214128997bSMichel Lespinasse 22223af9e859SEric B Munson perf_event_mmap(vma); 22233af9e859SEric B Munson } 22241da177e4SLinus Torvalds } 222542c36f63SHugh Dickins } 2226bb4a340eSRik van Riel vma_unlock_anon_vma(vma); 22276d50e60cSDavid Rientjes khugepaged_enter_vma_merge(vma, vma->vm_flags); 222809357814SOleg Nesterov validate_mm(mm); 22291da177e4SLinus Torvalds return error; 22301da177e4SLinus Torvalds } 223146dea3d0SHugh Dickins #endif /* CONFIG_STACK_GROWSUP || CONFIG_IA64 */ 223246dea3d0SHugh Dickins 22331da177e4SLinus Torvalds /* 22341da177e4SLinus Torvalds * vma is the first one with address < vma->vm_start. Have to extend vma. 22351da177e4SLinus Torvalds */ 2236d05f3169SMichal Hocko int expand_downwards(struct vm_area_struct *vma, 2237b6a2fea3SOllie Wild unsigned long address) 22381da177e4SLinus Torvalds { 223909357814SOleg Nesterov struct mm_struct *mm = vma->vm_mm; 22401da177e4SLinus Torvalds int error; 22411da177e4SLinus Torvalds 22421da177e4SLinus Torvalds /* 22431da177e4SLinus Torvalds * We must make sure the anon_vma is allocated 22441da177e4SLinus Torvalds * so that the anon_vma locking is not a noop. 22451da177e4SLinus Torvalds */ 22461da177e4SLinus Torvalds if (unlikely(anon_vma_prepare(vma))) 22471da177e4SLinus Torvalds return -ENOMEM; 22488869477aSEric Paris 22498869477aSEric Paris address &= PAGE_MASK; 2250e5467859SAl Viro error = security_mmap_addr(address); 22518869477aSEric Paris if (error) 22528869477aSEric Paris return error; 22538869477aSEric Paris 2254bb4a340eSRik van Riel vma_lock_anon_vma(vma); 22551da177e4SLinus Torvalds 22561da177e4SLinus Torvalds /* 22571da177e4SLinus Torvalds * vma->vm_start/vm_end cannot change under us because the caller 22581da177e4SLinus Torvalds * is required to hold the mmap_sem in read mode. We need the 22591da177e4SLinus Torvalds * anon_vma lock to serialize against concurrent expand_stacks. 22601da177e4SLinus Torvalds */ 22611da177e4SLinus Torvalds 22621da177e4SLinus Torvalds /* Somebody else might have raced and expanded it already */ 22631da177e4SLinus Torvalds if (address < vma->vm_start) { 22641da177e4SLinus Torvalds unsigned long size, grow; 22651da177e4SLinus Torvalds 22661da177e4SLinus Torvalds size = vma->vm_end - address; 22671da177e4SLinus Torvalds grow = (vma->vm_start - address) >> PAGE_SHIFT; 22681da177e4SLinus Torvalds 2269a626ca6aSLinus Torvalds error = -ENOMEM; 2270a626ca6aSLinus Torvalds if (grow <= vma->vm_pgoff) { 22711da177e4SLinus Torvalds error = acct_stack_growth(vma, size, grow); 22721da177e4SLinus Torvalds if (!error) { 22734128997bSMichel Lespinasse /* 22744128997bSMichel Lespinasse * vma_gap_update() doesn't support concurrent 22754128997bSMichel Lespinasse * updates, but we only hold a shared mmap_sem 22764128997bSMichel Lespinasse * lock here, so we need to protect against 22774128997bSMichel Lespinasse * concurrent vma expansions. 22784128997bSMichel Lespinasse * vma_lock_anon_vma() doesn't help here, as 22794128997bSMichel Lespinasse * we don't guarantee that all growable vmas 22804128997bSMichel Lespinasse * in a mm share the same root anon vma. 22814128997bSMichel Lespinasse * So, we reuse mm->page_table_lock to guard 22824128997bSMichel Lespinasse * against concurrent vma expansions. 22834128997bSMichel Lespinasse */ 228409357814SOleg Nesterov spin_lock(&mm->page_table_lock); 228587e8827bSOleg Nesterov if (vma->vm_flags & VM_LOCKED) 228609357814SOleg Nesterov mm->locked_vm += grow; 228709357814SOleg Nesterov vm_stat_account(mm, vma->vm_flags, 228887e8827bSOleg Nesterov vma->vm_file, grow); 2289bf181b9fSMichel Lespinasse anon_vma_interval_tree_pre_update_vma(vma); 22901da177e4SLinus Torvalds vma->vm_start = address; 22911da177e4SLinus Torvalds vma->vm_pgoff -= grow; 2292bf181b9fSMichel Lespinasse anon_vma_interval_tree_post_update_vma(vma); 2293d3737187SMichel Lespinasse vma_gap_update(vma); 229409357814SOleg Nesterov spin_unlock(&mm->page_table_lock); 22954128997bSMichel Lespinasse 22963af9e859SEric B Munson perf_event_mmap(vma); 22971da177e4SLinus Torvalds } 22981da177e4SLinus Torvalds } 2299a626ca6aSLinus Torvalds } 2300bb4a340eSRik van Riel vma_unlock_anon_vma(vma); 23016d50e60cSDavid Rientjes khugepaged_enter_vma_merge(vma, vma->vm_flags); 230209357814SOleg Nesterov validate_mm(mm); 23031da177e4SLinus Torvalds return error; 23041da177e4SLinus Torvalds } 23051da177e4SLinus Torvalds 230609884964SLinus Torvalds /* 230709884964SLinus Torvalds * Note how expand_stack() refuses to expand the stack all the way to 230809884964SLinus Torvalds * abut the next virtual mapping, *unless* that mapping itself is also 230909884964SLinus Torvalds * a stack mapping. We want to leave room for a guard page, after all 231009884964SLinus Torvalds * (the guard page itself is not added here, that is done by the 231109884964SLinus Torvalds * actual page faulting logic) 231209884964SLinus Torvalds * 231309884964SLinus Torvalds * This matches the behavior of the guard page logic (see mm/memory.c: 231409884964SLinus Torvalds * check_stack_guard_page()), which only allows the guard page to be 231509884964SLinus Torvalds * removed under these circumstances. 231609884964SLinus Torvalds */ 2317b6a2fea3SOllie Wild #ifdef CONFIG_STACK_GROWSUP 2318b6a2fea3SOllie Wild int expand_stack(struct vm_area_struct *vma, unsigned long address) 2319b6a2fea3SOllie Wild { 232009884964SLinus Torvalds struct vm_area_struct *next; 232109884964SLinus Torvalds 232209884964SLinus Torvalds address &= PAGE_MASK; 232309884964SLinus Torvalds next = vma->vm_next; 232409884964SLinus Torvalds if (next && next->vm_start == address + PAGE_SIZE) { 232509884964SLinus Torvalds if (!(next->vm_flags & VM_GROWSUP)) 232609884964SLinus Torvalds return -ENOMEM; 232709884964SLinus Torvalds } 2328b6a2fea3SOllie Wild return expand_upwards(vma, address); 2329b6a2fea3SOllie Wild } 2330b6a2fea3SOllie Wild 2331b6a2fea3SOllie Wild struct vm_area_struct * 2332b6a2fea3SOllie Wild find_extend_vma(struct mm_struct *mm, unsigned long addr) 2333b6a2fea3SOllie Wild { 2334b6a2fea3SOllie Wild struct vm_area_struct *vma, *prev; 2335b6a2fea3SOllie Wild 2336b6a2fea3SOllie Wild addr &= PAGE_MASK; 2337b6a2fea3SOllie Wild vma = find_vma_prev(mm, addr, &prev); 2338b6a2fea3SOllie Wild if (vma && (vma->vm_start <= addr)) 2339b6a2fea3SOllie Wild return vma; 23401c127185SDenys Vlasenko if (!prev || expand_stack(prev, addr)) 2341b6a2fea3SOllie Wild return NULL; 2342cea10a19SMichel Lespinasse if (prev->vm_flags & VM_LOCKED) 2343fc05f566SKirill A. Shutemov populate_vma_page_range(prev, addr, prev->vm_end, NULL); 2344b6a2fea3SOllie Wild return prev; 2345b6a2fea3SOllie Wild } 2346b6a2fea3SOllie Wild #else 2347b6a2fea3SOllie Wild int expand_stack(struct vm_area_struct *vma, unsigned long address) 2348b6a2fea3SOllie Wild { 234909884964SLinus Torvalds struct vm_area_struct *prev; 235009884964SLinus Torvalds 235109884964SLinus Torvalds address &= PAGE_MASK; 235209884964SLinus Torvalds prev = vma->vm_prev; 235309884964SLinus Torvalds if (prev && prev->vm_end == address) { 235409884964SLinus Torvalds if (!(prev->vm_flags & VM_GROWSDOWN)) 235509884964SLinus Torvalds return -ENOMEM; 235609884964SLinus Torvalds } 2357b6a2fea3SOllie Wild return expand_downwards(vma, address); 2358b6a2fea3SOllie Wild } 2359b6a2fea3SOllie Wild 23601da177e4SLinus Torvalds struct vm_area_struct * 23611da177e4SLinus Torvalds find_extend_vma(struct mm_struct *mm, unsigned long addr) 23621da177e4SLinus Torvalds { 23631da177e4SLinus Torvalds struct vm_area_struct *vma; 23641da177e4SLinus Torvalds unsigned long start; 23651da177e4SLinus Torvalds 23661da177e4SLinus Torvalds addr &= PAGE_MASK; 23671da177e4SLinus Torvalds vma = find_vma(mm, addr); 23681da177e4SLinus Torvalds if (!vma) 23691da177e4SLinus Torvalds return NULL; 23701da177e4SLinus Torvalds if (vma->vm_start <= addr) 23711da177e4SLinus Torvalds return vma; 23721da177e4SLinus Torvalds if (!(vma->vm_flags & VM_GROWSDOWN)) 23731da177e4SLinus Torvalds return NULL; 23741da177e4SLinus Torvalds start = vma->vm_start; 23751da177e4SLinus Torvalds if (expand_stack(vma, addr)) 23761da177e4SLinus Torvalds return NULL; 2377cea10a19SMichel Lespinasse if (vma->vm_flags & VM_LOCKED) 2378fc05f566SKirill A. Shutemov populate_vma_page_range(vma, addr, start, NULL); 23791da177e4SLinus Torvalds return vma; 23801da177e4SLinus Torvalds } 23811da177e4SLinus Torvalds #endif 23821da177e4SLinus Torvalds 2383e1d6d01aSJesse Barnes EXPORT_SYMBOL_GPL(find_extend_vma); 2384e1d6d01aSJesse Barnes 23852c0b3814SHugh Dickins /* 23862c0b3814SHugh Dickins * Ok - we have the memory areas we should free on the vma list, 23872c0b3814SHugh Dickins * so release them, and do the vma updates. 23881da177e4SLinus Torvalds * 23892c0b3814SHugh Dickins * Called with the mm semaphore held. 23901da177e4SLinus Torvalds */ 23912c0b3814SHugh Dickins static void remove_vma_list(struct mm_struct *mm, struct vm_area_struct *vma) 23921da177e4SLinus Torvalds { 23934f74d2c8SLinus Torvalds unsigned long nr_accounted = 0; 23944f74d2c8SLinus Torvalds 2395365e9c87SHugh Dickins /* Update high watermark before we lower total_vm */ 2396365e9c87SHugh Dickins update_hiwater_vm(mm); 23972c0b3814SHugh Dickins do { 2398ab50b8edSHugh Dickins long nrpages = vma_pages(vma); 23991da177e4SLinus Torvalds 24004f74d2c8SLinus Torvalds if (vma->vm_flags & VM_ACCOUNT) 24014f74d2c8SLinus Torvalds nr_accounted += nrpages; 2402ab50b8edSHugh Dickins vm_stat_account(mm, vma->vm_flags, vma->vm_file, -nrpages); 2403a8fb5618SHugh Dickins vma = remove_vma(vma); 2404146425a3SHugh Dickins } while (vma); 24054f74d2c8SLinus Torvalds vm_unacct_memory(nr_accounted); 24061da177e4SLinus Torvalds validate_mm(mm); 24071da177e4SLinus Torvalds } 24081da177e4SLinus Torvalds 24091da177e4SLinus Torvalds /* 24101da177e4SLinus Torvalds * Get rid of page table information in the indicated region. 24111da177e4SLinus Torvalds * 2412f10df686SPaolo 'Blaisorblade' Giarrusso * Called with the mm semaphore held. 24131da177e4SLinus Torvalds */ 24141da177e4SLinus Torvalds static void unmap_region(struct mm_struct *mm, 2415e0da382cSHugh Dickins struct vm_area_struct *vma, struct vm_area_struct *prev, 2416e0da382cSHugh Dickins unsigned long start, unsigned long end) 24171da177e4SLinus Torvalds { 2418e0da382cSHugh Dickins struct vm_area_struct *next = prev ? prev->vm_next : mm->mmap; 2419d16dfc55SPeter Zijlstra struct mmu_gather tlb; 24201da177e4SLinus Torvalds 24211da177e4SLinus Torvalds lru_add_drain(); 24222b047252SLinus Torvalds tlb_gather_mmu(&tlb, mm, start, end); 2423365e9c87SHugh Dickins update_hiwater_rss(mm); 24244f74d2c8SLinus Torvalds unmap_vmas(&tlb, vma, start, end); 2425d16dfc55SPeter Zijlstra free_pgtables(&tlb, vma, prev ? prev->vm_end : FIRST_USER_ADDRESS, 24266ee8630eSHugh Dickins next ? next->vm_start : USER_PGTABLES_CEILING); 2427d16dfc55SPeter Zijlstra tlb_finish_mmu(&tlb, start, end); 24281da177e4SLinus Torvalds } 24291da177e4SLinus Torvalds 24301da177e4SLinus Torvalds /* 24311da177e4SLinus Torvalds * Create a list of vma's touched by the unmap, removing them from the mm's 24321da177e4SLinus Torvalds * vma list as we go.. 24331da177e4SLinus Torvalds */ 24341da177e4SLinus Torvalds static void 24351da177e4SLinus Torvalds detach_vmas_to_be_unmapped(struct mm_struct *mm, struct vm_area_struct *vma, 24361da177e4SLinus Torvalds struct vm_area_struct *prev, unsigned long end) 24371da177e4SLinus Torvalds { 24381da177e4SLinus Torvalds struct vm_area_struct **insertion_point; 24391da177e4SLinus Torvalds struct vm_area_struct *tail_vma = NULL; 24401da177e4SLinus Torvalds 24411da177e4SLinus Torvalds insertion_point = (prev ? &prev->vm_next : &mm->mmap); 2442297c5eeeSLinus Torvalds vma->vm_prev = NULL; 24431da177e4SLinus Torvalds do { 2444d3737187SMichel Lespinasse vma_rb_erase(vma, &mm->mm_rb); 24451da177e4SLinus Torvalds mm->map_count--; 24461da177e4SLinus Torvalds tail_vma = vma; 24471da177e4SLinus Torvalds vma = vma->vm_next; 24481da177e4SLinus Torvalds } while (vma && vma->vm_start < end); 24491da177e4SLinus Torvalds *insertion_point = vma; 2450d3737187SMichel Lespinasse if (vma) { 2451297c5eeeSLinus Torvalds vma->vm_prev = prev; 2452d3737187SMichel Lespinasse vma_gap_update(vma); 2453d3737187SMichel Lespinasse } else 2454d3737187SMichel Lespinasse mm->highest_vm_end = prev ? prev->vm_end : 0; 24551da177e4SLinus Torvalds tail_vma->vm_next = NULL; 2456615d6e87SDavidlohr Bueso 2457615d6e87SDavidlohr Bueso /* Kill the cache */ 2458615d6e87SDavidlohr Bueso vmacache_invalidate(mm); 24591da177e4SLinus Torvalds } 24601da177e4SLinus Torvalds 24611da177e4SLinus Torvalds /* 2462659ace58SKOSAKI Motohiro * __split_vma() bypasses sysctl_max_map_count checking. We use this on the 2463659ace58SKOSAKI Motohiro * munmap path where it doesn't make sense to fail. 24641da177e4SLinus Torvalds */ 2465659ace58SKOSAKI Motohiro static int __split_vma(struct mm_struct *mm, struct vm_area_struct *vma, 24661da177e4SLinus Torvalds unsigned long addr, int new_below) 24671da177e4SLinus Torvalds { 24681da177e4SLinus Torvalds struct vm_area_struct *new; 2469e3975891SChen Gang int err; 24701da177e4SLinus Torvalds 2471a5516438SAndi Kleen if (is_vm_hugetlb_page(vma) && (addr & 2472a5516438SAndi Kleen ~(huge_page_mask(hstate_vma(vma))))) 24731da177e4SLinus Torvalds return -EINVAL; 24741da177e4SLinus Torvalds 2475e94b1766SChristoph Lameter new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL); 24761da177e4SLinus Torvalds if (!new) 2477e3975891SChen Gang return -ENOMEM; 24781da177e4SLinus Torvalds 24791da177e4SLinus Torvalds /* most fields are the same, copy all, and then fixup */ 24801da177e4SLinus Torvalds *new = *vma; 24811da177e4SLinus Torvalds 24825beb4930SRik van Riel INIT_LIST_HEAD(&new->anon_vma_chain); 24835beb4930SRik van Riel 24841da177e4SLinus Torvalds if (new_below) 24851da177e4SLinus Torvalds new->vm_end = addr; 24861da177e4SLinus Torvalds else { 24871da177e4SLinus Torvalds new->vm_start = addr; 24881da177e4SLinus Torvalds new->vm_pgoff += ((addr - vma->vm_start) >> PAGE_SHIFT); 24891da177e4SLinus Torvalds } 24901da177e4SLinus Torvalds 2491ef0855d3SOleg Nesterov err = vma_dup_policy(vma, new); 2492ef0855d3SOleg Nesterov if (err) 24935beb4930SRik van Riel goto out_free_vma; 24941da177e4SLinus Torvalds 2495c4ea95d7SDaniel Forrest err = anon_vma_clone(new, vma); 2496c4ea95d7SDaniel Forrest if (err) 24975beb4930SRik van Riel goto out_free_mpol; 24985beb4930SRik van Riel 2499e9714acfSKonstantin Khlebnikov if (new->vm_file) 25001da177e4SLinus Torvalds get_file(new->vm_file); 25011da177e4SLinus Torvalds 25021da177e4SLinus Torvalds if (new->vm_ops && new->vm_ops->open) 25031da177e4SLinus Torvalds new->vm_ops->open(new); 25041da177e4SLinus Torvalds 25051da177e4SLinus Torvalds if (new_below) 25065beb4930SRik van Riel err = vma_adjust(vma, addr, vma->vm_end, vma->vm_pgoff + 25071da177e4SLinus Torvalds ((addr - new->vm_start) >> PAGE_SHIFT), new); 25081da177e4SLinus Torvalds else 25095beb4930SRik van Riel err = vma_adjust(vma, vma->vm_start, addr, vma->vm_pgoff, new); 25101da177e4SLinus Torvalds 25115beb4930SRik van Riel /* Success. */ 25125beb4930SRik van Riel if (!err) 25131da177e4SLinus Torvalds return 0; 25145beb4930SRik van Riel 25155beb4930SRik van Riel /* Clean everything up if vma_adjust failed. */ 251658927533SRik van Riel if (new->vm_ops && new->vm_ops->close) 25175beb4930SRik van Riel new->vm_ops->close(new); 2518e9714acfSKonstantin Khlebnikov if (new->vm_file) 25195beb4930SRik van Riel fput(new->vm_file); 25202aeadc30SAndrea Arcangeli unlink_anon_vmas(new); 25215beb4930SRik van Riel out_free_mpol: 2522ef0855d3SOleg Nesterov mpol_put(vma_policy(new)); 25235beb4930SRik van Riel out_free_vma: 25245beb4930SRik van Riel kmem_cache_free(vm_area_cachep, new); 25255beb4930SRik van Riel return err; 25261da177e4SLinus Torvalds } 25271da177e4SLinus Torvalds 2528659ace58SKOSAKI Motohiro /* 2529659ace58SKOSAKI Motohiro * Split a vma into two pieces at address 'addr', a new vma is allocated 2530659ace58SKOSAKI Motohiro * either for the first part or the tail. 2531659ace58SKOSAKI Motohiro */ 2532659ace58SKOSAKI Motohiro int split_vma(struct mm_struct *mm, struct vm_area_struct *vma, 2533659ace58SKOSAKI Motohiro unsigned long addr, int new_below) 2534659ace58SKOSAKI Motohiro { 2535659ace58SKOSAKI Motohiro if (mm->map_count >= sysctl_max_map_count) 2536659ace58SKOSAKI Motohiro return -ENOMEM; 2537659ace58SKOSAKI Motohiro 2538659ace58SKOSAKI Motohiro return __split_vma(mm, vma, addr, new_below); 2539659ace58SKOSAKI Motohiro } 2540659ace58SKOSAKI Motohiro 25411da177e4SLinus Torvalds /* Munmap is split into 2 main parts -- this part which finds 25421da177e4SLinus Torvalds * what needs doing, and the areas themselves, which do the 25431da177e4SLinus Torvalds * work. This now handles partial unmappings. 25441da177e4SLinus Torvalds * Jeremy Fitzhardinge <jeremy@goop.org> 25451da177e4SLinus Torvalds */ 25461da177e4SLinus Torvalds int do_munmap(struct mm_struct *mm, unsigned long start, size_t len) 25471da177e4SLinus Torvalds { 25481da177e4SLinus Torvalds unsigned long end; 2549146425a3SHugh Dickins struct vm_area_struct *vma, *prev, *last; 25501da177e4SLinus Torvalds 2551de1741a1SAlexander Kuleshov if ((offset_in_page(start)) || start > TASK_SIZE || len > TASK_SIZE-start) 25521da177e4SLinus Torvalds return -EINVAL; 25531da177e4SLinus Torvalds 2554cc71aba3Svishnu.ps len = PAGE_ALIGN(len); 2555cc71aba3Svishnu.ps if (len == 0) 25561da177e4SLinus Torvalds return -EINVAL; 25571da177e4SLinus Torvalds 25581da177e4SLinus Torvalds /* Find the first overlapping VMA */ 25599be34c9dSLinus Torvalds vma = find_vma(mm, start); 2560146425a3SHugh Dickins if (!vma) 25611da177e4SLinus Torvalds return 0; 25629be34c9dSLinus Torvalds prev = vma->vm_prev; 2563146425a3SHugh Dickins /* we have start < vma->vm_end */ 25641da177e4SLinus Torvalds 25651da177e4SLinus Torvalds /* if it doesn't overlap, we have nothing.. */ 25661da177e4SLinus Torvalds end = start + len; 2567146425a3SHugh Dickins if (vma->vm_start >= end) 25681da177e4SLinus Torvalds return 0; 25691da177e4SLinus Torvalds 25701da177e4SLinus Torvalds /* 25711da177e4SLinus Torvalds * If we need to split any vma, do it now to save pain later. 25721da177e4SLinus Torvalds * 25731da177e4SLinus Torvalds * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially 25741da177e4SLinus Torvalds * unmapped vm_area_struct will remain in use: so lower split_vma 25751da177e4SLinus Torvalds * places tmp vma above, and higher split_vma places tmp vma below. 25761da177e4SLinus Torvalds */ 2577146425a3SHugh Dickins if (start > vma->vm_start) { 2578659ace58SKOSAKI Motohiro int error; 2579659ace58SKOSAKI Motohiro 2580659ace58SKOSAKI Motohiro /* 2581659ace58SKOSAKI Motohiro * Make sure that map_count on return from munmap() will 2582659ace58SKOSAKI Motohiro * not exceed its limit; but let map_count go just above 2583659ace58SKOSAKI Motohiro * its limit temporarily, to help free resources as expected. 2584659ace58SKOSAKI Motohiro */ 2585659ace58SKOSAKI Motohiro if (end < vma->vm_end && mm->map_count >= sysctl_max_map_count) 2586659ace58SKOSAKI Motohiro return -ENOMEM; 2587659ace58SKOSAKI Motohiro 2588659ace58SKOSAKI Motohiro error = __split_vma(mm, vma, start, 0); 25891da177e4SLinus Torvalds if (error) 25901da177e4SLinus Torvalds return error; 2591146425a3SHugh Dickins prev = vma; 25921da177e4SLinus Torvalds } 25931da177e4SLinus Torvalds 25941da177e4SLinus Torvalds /* Does it split the last one? */ 25951da177e4SLinus Torvalds last = find_vma(mm, end); 25961da177e4SLinus Torvalds if (last && end > last->vm_start) { 2597659ace58SKOSAKI Motohiro int error = __split_vma(mm, last, end, 1); 25981da177e4SLinus Torvalds if (error) 25991da177e4SLinus Torvalds return error; 26001da177e4SLinus Torvalds } 2601146425a3SHugh Dickins vma = prev ? prev->vm_next : mm->mmap; 26021da177e4SLinus Torvalds 26031da177e4SLinus Torvalds /* 2604ba470de4SRik van Riel * unlock any mlock()ed ranges before detaching vmas 2605ba470de4SRik van Riel */ 2606ba470de4SRik van Riel if (mm->locked_vm) { 2607ba470de4SRik van Riel struct vm_area_struct *tmp = vma; 2608ba470de4SRik van Riel while (tmp && tmp->vm_start < end) { 2609ba470de4SRik van Riel if (tmp->vm_flags & VM_LOCKED) { 2610ba470de4SRik van Riel mm->locked_vm -= vma_pages(tmp); 2611ba470de4SRik van Riel munlock_vma_pages_all(tmp); 2612ba470de4SRik van Riel } 2613ba470de4SRik van Riel tmp = tmp->vm_next; 2614ba470de4SRik van Riel } 2615ba470de4SRik van Riel } 2616ba470de4SRik van Riel 2617ba470de4SRik van Riel /* 26181da177e4SLinus Torvalds * Remove the vma's, and unmap the actual pages 26191da177e4SLinus Torvalds */ 2620146425a3SHugh Dickins detach_vmas_to_be_unmapped(mm, vma, prev, end); 2621146425a3SHugh Dickins unmap_region(mm, vma, prev, start, end); 26221da177e4SLinus Torvalds 26231de4fa14SDave Hansen arch_unmap(mm, vma, start, end); 26241de4fa14SDave Hansen 26251da177e4SLinus Torvalds /* Fix up all other VM information */ 26262c0b3814SHugh Dickins remove_vma_list(mm, vma); 26271da177e4SLinus Torvalds 26281da177e4SLinus Torvalds return 0; 26291da177e4SLinus Torvalds } 26301da177e4SLinus Torvalds 2631bfce281cSAl Viro int vm_munmap(unsigned long start, size_t len) 2632a46ef99dSLinus Torvalds { 2633a46ef99dSLinus Torvalds int ret; 2634bfce281cSAl Viro struct mm_struct *mm = current->mm; 2635a46ef99dSLinus Torvalds 2636a46ef99dSLinus Torvalds down_write(&mm->mmap_sem); 2637a46ef99dSLinus Torvalds ret = do_munmap(mm, start, len); 2638a46ef99dSLinus Torvalds up_write(&mm->mmap_sem); 2639a46ef99dSLinus Torvalds return ret; 2640a46ef99dSLinus Torvalds } 2641a46ef99dSLinus Torvalds EXPORT_SYMBOL(vm_munmap); 2642a46ef99dSLinus Torvalds 26436a6160a7SHeiko Carstens SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len) 26441da177e4SLinus Torvalds { 26451da177e4SLinus Torvalds profile_munmap(addr); 2646bfce281cSAl Viro return vm_munmap(addr, len); 26471da177e4SLinus Torvalds } 26481da177e4SLinus Torvalds 2649c8d78c18SKirill A. Shutemov 2650c8d78c18SKirill A. Shutemov /* 2651c8d78c18SKirill A. Shutemov * Emulation of deprecated remap_file_pages() syscall. 2652c8d78c18SKirill A. Shutemov */ 2653c8d78c18SKirill A. Shutemov SYSCALL_DEFINE5(remap_file_pages, unsigned long, start, unsigned long, size, 2654c8d78c18SKirill A. Shutemov unsigned long, prot, unsigned long, pgoff, unsigned long, flags) 2655c8d78c18SKirill A. Shutemov { 2656c8d78c18SKirill A. Shutemov 2657c8d78c18SKirill A. Shutemov struct mm_struct *mm = current->mm; 2658c8d78c18SKirill A. Shutemov struct vm_area_struct *vma; 2659c8d78c18SKirill A. Shutemov unsigned long populate = 0; 2660c8d78c18SKirill A. Shutemov unsigned long ret = -EINVAL; 2661c8d78c18SKirill A. Shutemov struct file *file; 2662c8d78c18SKirill A. Shutemov 2663c8d78c18SKirill A. Shutemov pr_warn_once("%s (%d) uses deprecated remap_file_pages() syscall. " 2664c8d78c18SKirill A. Shutemov "See Documentation/vm/remap_file_pages.txt.\n", 2665c8d78c18SKirill A. Shutemov current->comm, current->pid); 2666c8d78c18SKirill A. Shutemov 2667c8d78c18SKirill A. Shutemov if (prot) 2668c8d78c18SKirill A. Shutemov return ret; 2669c8d78c18SKirill A. Shutemov start = start & PAGE_MASK; 2670c8d78c18SKirill A. Shutemov size = size & PAGE_MASK; 2671c8d78c18SKirill A. Shutemov 2672c8d78c18SKirill A. Shutemov if (start + size <= start) 2673c8d78c18SKirill A. Shutemov return ret; 2674c8d78c18SKirill A. Shutemov 2675c8d78c18SKirill A. Shutemov /* Does pgoff wrap? */ 2676c8d78c18SKirill A. Shutemov if (pgoff + (size >> PAGE_SHIFT) < pgoff) 2677c8d78c18SKirill A. Shutemov return ret; 2678c8d78c18SKirill A. Shutemov 2679c8d78c18SKirill A. Shutemov down_write(&mm->mmap_sem); 2680c8d78c18SKirill A. Shutemov vma = find_vma(mm, start); 2681c8d78c18SKirill A. Shutemov 2682c8d78c18SKirill A. Shutemov if (!vma || !(vma->vm_flags & VM_SHARED)) 2683c8d78c18SKirill A. Shutemov goto out; 2684c8d78c18SKirill A. Shutemov 2685c8d78c18SKirill A. Shutemov if (start < vma->vm_start || start + size > vma->vm_end) 2686c8d78c18SKirill A. Shutemov goto out; 2687c8d78c18SKirill A. Shutemov 2688c8d78c18SKirill A. Shutemov if (pgoff == linear_page_index(vma, start)) { 2689c8d78c18SKirill A. Shutemov ret = 0; 2690c8d78c18SKirill A. Shutemov goto out; 2691c8d78c18SKirill A. Shutemov } 2692c8d78c18SKirill A. Shutemov 2693c8d78c18SKirill A. Shutemov prot |= vma->vm_flags & VM_READ ? PROT_READ : 0; 2694c8d78c18SKirill A. Shutemov prot |= vma->vm_flags & VM_WRITE ? PROT_WRITE : 0; 2695c8d78c18SKirill A. Shutemov prot |= vma->vm_flags & VM_EXEC ? PROT_EXEC : 0; 2696c8d78c18SKirill A. Shutemov 2697c8d78c18SKirill A. Shutemov flags &= MAP_NONBLOCK; 2698c8d78c18SKirill A. Shutemov flags |= MAP_SHARED | MAP_FIXED | MAP_POPULATE; 2699c8d78c18SKirill A. Shutemov if (vma->vm_flags & VM_LOCKED) { 2700c8d78c18SKirill A. Shutemov flags |= MAP_LOCKED; 2701c8d78c18SKirill A. Shutemov /* drop PG_Mlocked flag for over-mapped range */ 2702c8d78c18SKirill A. Shutemov munlock_vma_pages_range(vma, start, start + size); 2703c8d78c18SKirill A. Shutemov } 2704c8d78c18SKirill A. Shutemov 2705c8d78c18SKirill A. Shutemov file = get_file(vma->vm_file); 2706c8d78c18SKirill A. Shutemov ret = do_mmap_pgoff(vma->vm_file, start, size, 2707c8d78c18SKirill A. Shutemov prot, flags, pgoff, &populate); 2708c8d78c18SKirill A. Shutemov fput(file); 2709c8d78c18SKirill A. Shutemov out: 2710c8d78c18SKirill A. Shutemov up_write(&mm->mmap_sem); 2711c8d78c18SKirill A. Shutemov if (populate) 2712c8d78c18SKirill A. Shutemov mm_populate(ret, populate); 2713c8d78c18SKirill A. Shutemov if (!IS_ERR_VALUE(ret)) 2714c8d78c18SKirill A. Shutemov ret = 0; 2715c8d78c18SKirill A. Shutemov return ret; 2716c8d78c18SKirill A. Shutemov } 2717c8d78c18SKirill A. Shutemov 27181da177e4SLinus Torvalds static inline void verify_mm_writelocked(struct mm_struct *mm) 27191da177e4SLinus Torvalds { 2720a241ec65SPaul E. McKenney #ifdef CONFIG_DEBUG_VM 27211da177e4SLinus Torvalds if (unlikely(down_read_trylock(&mm->mmap_sem))) { 27221da177e4SLinus Torvalds WARN_ON(1); 27231da177e4SLinus Torvalds up_read(&mm->mmap_sem); 27241da177e4SLinus Torvalds } 27251da177e4SLinus Torvalds #endif 27261da177e4SLinus Torvalds } 27271da177e4SLinus Torvalds 27281da177e4SLinus Torvalds /* 27291da177e4SLinus Torvalds * this is really a simplified "do_mmap". it only handles 27301da177e4SLinus Torvalds * anonymous maps. eventually we may be able to do some 27311da177e4SLinus Torvalds * brk-specific accounting here. 27321da177e4SLinus Torvalds */ 2733e4eb1ff6SLinus Torvalds static unsigned long do_brk(unsigned long addr, unsigned long len) 27341da177e4SLinus Torvalds { 27351da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 27361da177e4SLinus Torvalds struct vm_area_struct *vma, *prev; 27371da177e4SLinus Torvalds unsigned long flags; 27381da177e4SLinus Torvalds struct rb_node **rb_link, *rb_parent; 27391da177e4SLinus Torvalds pgoff_t pgoff = addr >> PAGE_SHIFT; 27403a459756SKirill Korotaev int error; 27411da177e4SLinus Torvalds 27421da177e4SLinus Torvalds len = PAGE_ALIGN(len); 27431da177e4SLinus Torvalds if (!len) 27441da177e4SLinus Torvalds return addr; 27451da177e4SLinus Torvalds 27463a459756SKirill Korotaev flags = VM_DATA_DEFAULT_FLAGS | VM_ACCOUNT | mm->def_flags; 27473a459756SKirill Korotaev 27482c6a1016SAl Viro error = get_unmapped_area(NULL, addr, len, 0, MAP_FIXED); 2749de1741a1SAlexander Kuleshov if (offset_in_page(error)) 27503a459756SKirill Korotaev return error; 27513a459756SKirill Korotaev 2752363ee17fSDavidlohr Bueso error = mlock_future_check(mm, mm->def_flags, len); 2753363ee17fSDavidlohr Bueso if (error) 2754363ee17fSDavidlohr Bueso return error; 27551da177e4SLinus Torvalds 27561da177e4SLinus Torvalds /* 27571da177e4SLinus Torvalds * mm->mmap_sem is required to protect against another thread 27581da177e4SLinus Torvalds * changing the mappings in case we sleep. 27591da177e4SLinus Torvalds */ 27601da177e4SLinus Torvalds verify_mm_writelocked(mm); 27611da177e4SLinus Torvalds 27621da177e4SLinus Torvalds /* 27631da177e4SLinus Torvalds * Clear old maps. this also does some error checking for us 27641da177e4SLinus Torvalds */ 27659fcd1457SRasmus Villemoes while (find_vma_links(mm, addr, addr + len, &prev, &rb_link, 27669fcd1457SRasmus Villemoes &rb_parent)) { 27671da177e4SLinus Torvalds if (do_munmap(mm, addr, len)) 27681da177e4SLinus Torvalds return -ENOMEM; 27691da177e4SLinus Torvalds } 27701da177e4SLinus Torvalds 27711da177e4SLinus Torvalds /* Check against address space limits *after* clearing old maps... */ 2772119f657cSakpm@osdl.org if (!may_expand_vm(mm, len >> PAGE_SHIFT)) 27731da177e4SLinus Torvalds return -ENOMEM; 27741da177e4SLinus Torvalds 27751da177e4SLinus Torvalds if (mm->map_count > sysctl_max_map_count) 27761da177e4SLinus Torvalds return -ENOMEM; 27771da177e4SLinus Torvalds 2778191c5424SAl Viro if (security_vm_enough_memory_mm(mm, len >> PAGE_SHIFT)) 27791da177e4SLinus Torvalds return -ENOMEM; 27801da177e4SLinus Torvalds 27811da177e4SLinus Torvalds /* Can we just expand an old private anonymous mapping? */ 2782ba470de4SRik van Riel vma = vma_merge(mm, prev, addr, addr + len, flags, 278319a809afSAndrea Arcangeli NULL, NULL, pgoff, NULL, NULL_VM_UFFD_CTX); 2784ba470de4SRik van Riel if (vma) 27851da177e4SLinus Torvalds goto out; 27861da177e4SLinus Torvalds 27871da177e4SLinus Torvalds /* 27881da177e4SLinus Torvalds * create a vma struct for an anonymous mapping 27891da177e4SLinus Torvalds */ 2790c5e3b83eSPekka Enberg vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL); 27911da177e4SLinus Torvalds if (!vma) { 27921da177e4SLinus Torvalds vm_unacct_memory(len >> PAGE_SHIFT); 27931da177e4SLinus Torvalds return -ENOMEM; 27941da177e4SLinus Torvalds } 27951da177e4SLinus Torvalds 27965beb4930SRik van Riel INIT_LIST_HEAD(&vma->anon_vma_chain); 27971da177e4SLinus Torvalds vma->vm_mm = mm; 27981da177e4SLinus Torvalds vma->vm_start = addr; 27991da177e4SLinus Torvalds vma->vm_end = addr + len; 28001da177e4SLinus Torvalds vma->vm_pgoff = pgoff; 28011da177e4SLinus Torvalds vma->vm_flags = flags; 28023ed75eb8SColy Li vma->vm_page_prot = vm_get_page_prot(flags); 28031da177e4SLinus Torvalds vma_link(mm, vma, prev, rb_link, rb_parent); 28041da177e4SLinus Torvalds out: 28053af9e859SEric B Munson perf_event_mmap(vma); 28061da177e4SLinus Torvalds mm->total_vm += len >> PAGE_SHIFT; 2807128557ffSMichel Lespinasse if (flags & VM_LOCKED) 2808ba470de4SRik van Riel mm->locked_vm += (len >> PAGE_SHIFT); 2809d9104d1cSCyrill Gorcunov vma->vm_flags |= VM_SOFTDIRTY; 28101da177e4SLinus Torvalds return addr; 28111da177e4SLinus Torvalds } 28121da177e4SLinus Torvalds 2813e4eb1ff6SLinus Torvalds unsigned long vm_brk(unsigned long addr, unsigned long len) 2814e4eb1ff6SLinus Torvalds { 2815e4eb1ff6SLinus Torvalds struct mm_struct *mm = current->mm; 2816e4eb1ff6SLinus Torvalds unsigned long ret; 2817128557ffSMichel Lespinasse bool populate; 2818e4eb1ff6SLinus Torvalds 2819e4eb1ff6SLinus Torvalds down_write(&mm->mmap_sem); 2820e4eb1ff6SLinus Torvalds ret = do_brk(addr, len); 2821128557ffSMichel Lespinasse populate = ((mm->def_flags & VM_LOCKED) != 0); 2822e4eb1ff6SLinus Torvalds up_write(&mm->mmap_sem); 2823128557ffSMichel Lespinasse if (populate) 2824128557ffSMichel Lespinasse mm_populate(addr, len); 2825e4eb1ff6SLinus Torvalds return ret; 2826e4eb1ff6SLinus Torvalds } 2827e4eb1ff6SLinus Torvalds EXPORT_SYMBOL(vm_brk); 28281da177e4SLinus Torvalds 28291da177e4SLinus Torvalds /* Release all mmaps. */ 28301da177e4SLinus Torvalds void exit_mmap(struct mm_struct *mm) 28311da177e4SLinus Torvalds { 2832d16dfc55SPeter Zijlstra struct mmu_gather tlb; 2833ba470de4SRik van Riel struct vm_area_struct *vma; 28341da177e4SLinus Torvalds unsigned long nr_accounted = 0; 28351da177e4SLinus Torvalds 2836d6dd61c8SJeremy Fitzhardinge /* mm's last user has gone, and its about to be pulled down */ 2837cddb8a5cSAndrea Arcangeli mmu_notifier_release(mm); 2838d6dd61c8SJeremy Fitzhardinge 2839ba470de4SRik van Riel if (mm->locked_vm) { 2840ba470de4SRik van Riel vma = mm->mmap; 2841ba470de4SRik van Riel while (vma) { 2842ba470de4SRik van Riel if (vma->vm_flags & VM_LOCKED) 2843ba470de4SRik van Riel munlock_vma_pages_all(vma); 2844ba470de4SRik van Riel vma = vma->vm_next; 2845ba470de4SRik van Riel } 2846ba470de4SRik van Riel } 28479480c53eSJeremy Fitzhardinge 28489480c53eSJeremy Fitzhardinge arch_exit_mmap(mm); 28499480c53eSJeremy Fitzhardinge 2850ba470de4SRik van Riel vma = mm->mmap; 28519480c53eSJeremy Fitzhardinge if (!vma) /* Can happen if dup_mmap() received an OOM */ 28529480c53eSJeremy Fitzhardinge return; 28539480c53eSJeremy Fitzhardinge 28541da177e4SLinus Torvalds lru_add_drain(); 28551da177e4SLinus Torvalds flush_cache_mm(mm); 28562b047252SLinus Torvalds tlb_gather_mmu(&tlb, mm, 0, -1); 2857901608d9SOleg Nesterov /* update_hiwater_rss(mm) here? but nobody should be looking */ 2858e0da382cSHugh Dickins /* Use -1 here to ensure all VMAs in the mm are unmapped */ 28594f74d2c8SLinus Torvalds unmap_vmas(&tlb, vma, 0, -1); 28609ba69294SHugh Dickins 28616ee8630eSHugh Dickins free_pgtables(&tlb, vma, FIRST_USER_ADDRESS, USER_PGTABLES_CEILING); 2862853f5e26SAl Viro tlb_finish_mmu(&tlb, 0, -1); 28631da177e4SLinus Torvalds 28641da177e4SLinus Torvalds /* 28658f4f8c16SHugh Dickins * Walk the list again, actually closing and freeing it, 28668f4f8c16SHugh Dickins * with preemption enabled, without holding any MM locks. 28671da177e4SLinus Torvalds */ 28684f74d2c8SLinus Torvalds while (vma) { 28694f74d2c8SLinus Torvalds if (vma->vm_flags & VM_ACCOUNT) 28704f74d2c8SLinus Torvalds nr_accounted += vma_pages(vma); 2871a8fb5618SHugh Dickins vma = remove_vma(vma); 28724f74d2c8SLinus Torvalds } 28734f74d2c8SLinus Torvalds vm_unacct_memory(nr_accounted); 28741da177e4SLinus Torvalds } 28751da177e4SLinus Torvalds 28761da177e4SLinus Torvalds /* Insert vm structure into process list sorted by address 28771da177e4SLinus Torvalds * and into the inode's i_mmap tree. If vm_file is non-NULL 2878c8c06efaSDavidlohr Bueso * then i_mmap_rwsem is taken here. 28791da177e4SLinus Torvalds */ 28801da177e4SLinus Torvalds int insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma) 28811da177e4SLinus Torvalds { 28826597d783SHugh Dickins struct vm_area_struct *prev; 28831da177e4SLinus Torvalds struct rb_node **rb_link, *rb_parent; 28841da177e4SLinus Torvalds 2885c9d13f5fSChen Gang if (find_vma_links(mm, vma->vm_start, vma->vm_end, 2886c9d13f5fSChen Gang &prev, &rb_link, &rb_parent)) 2887c9d13f5fSChen Gang return -ENOMEM; 2888c9d13f5fSChen Gang if ((vma->vm_flags & VM_ACCOUNT) && 2889c9d13f5fSChen Gang security_vm_enough_memory_mm(mm, vma_pages(vma))) 2890c9d13f5fSChen Gang return -ENOMEM; 2891c9d13f5fSChen Gang 28921da177e4SLinus Torvalds /* 28931da177e4SLinus Torvalds * The vm_pgoff of a purely anonymous vma should be irrelevant 28941da177e4SLinus Torvalds * until its first write fault, when page's anon_vma and index 28951da177e4SLinus Torvalds * are set. But now set the vm_pgoff it will almost certainly 28961da177e4SLinus Torvalds * end up with (unless mremap moves it elsewhere before that 28971da177e4SLinus Torvalds * first wfault), so /proc/pid/maps tells a consistent story. 28981da177e4SLinus Torvalds * 28991da177e4SLinus Torvalds * By setting it to reflect the virtual start address of the 29001da177e4SLinus Torvalds * vma, merges and splits can happen in a seamless way, just 29011da177e4SLinus Torvalds * using the existing file pgoff checks and manipulations. 29021da177e4SLinus Torvalds * Similarly in do_mmap_pgoff and in do_brk. 29031da177e4SLinus Torvalds */ 29048a9cc3b5SOleg Nesterov if (vma_is_anonymous(vma)) { 29051da177e4SLinus Torvalds BUG_ON(vma->anon_vma); 29061da177e4SLinus Torvalds vma->vm_pgoff = vma->vm_start >> PAGE_SHIFT; 29071da177e4SLinus Torvalds } 29082b144498SSrikar Dronamraju 29091da177e4SLinus Torvalds vma_link(mm, vma, prev, rb_link, rb_parent); 29101da177e4SLinus Torvalds return 0; 29111da177e4SLinus Torvalds } 29121da177e4SLinus Torvalds 29131da177e4SLinus Torvalds /* 29141da177e4SLinus Torvalds * Copy the vma structure to a new location in the same mm, 29151da177e4SLinus Torvalds * prior to moving page table entries, to effect an mremap move. 29161da177e4SLinus Torvalds */ 29171da177e4SLinus Torvalds struct vm_area_struct *copy_vma(struct vm_area_struct **vmap, 291838a76013SMichel Lespinasse unsigned long addr, unsigned long len, pgoff_t pgoff, 291938a76013SMichel Lespinasse bool *need_rmap_locks) 29201da177e4SLinus Torvalds { 29211da177e4SLinus Torvalds struct vm_area_struct *vma = *vmap; 29221da177e4SLinus Torvalds unsigned long vma_start = vma->vm_start; 29231da177e4SLinus Torvalds struct mm_struct *mm = vma->vm_mm; 29241da177e4SLinus Torvalds struct vm_area_struct *new_vma, *prev; 29251da177e4SLinus Torvalds struct rb_node **rb_link, *rb_parent; 2926948f017bSAndrea Arcangeli bool faulted_in_anon_vma = true; 29271da177e4SLinus Torvalds 29281da177e4SLinus Torvalds /* 29291da177e4SLinus Torvalds * If anonymous vma has not yet been faulted, update new pgoff 29301da177e4SLinus Torvalds * to match new location, to increase its chance of merging. 29311da177e4SLinus Torvalds */ 2932ce75799bSOleg Nesterov if (unlikely(vma_is_anonymous(vma) && !vma->anon_vma)) { 29331da177e4SLinus Torvalds pgoff = addr >> PAGE_SHIFT; 2934948f017bSAndrea Arcangeli faulted_in_anon_vma = false; 2935948f017bSAndrea Arcangeli } 29361da177e4SLinus Torvalds 29376597d783SHugh Dickins if (find_vma_links(mm, addr, addr + len, &prev, &rb_link, &rb_parent)) 29386597d783SHugh Dickins return NULL; /* should never get here */ 29391da177e4SLinus Torvalds new_vma = vma_merge(mm, prev, addr, addr + len, vma->vm_flags, 294019a809afSAndrea Arcangeli vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma), 294119a809afSAndrea Arcangeli vma->vm_userfaultfd_ctx); 29421da177e4SLinus Torvalds if (new_vma) { 29431da177e4SLinus Torvalds /* 29441da177e4SLinus Torvalds * Source vma may have been merged into new_vma 29451da177e4SLinus Torvalds */ 2946948f017bSAndrea Arcangeli if (unlikely(vma_start >= new_vma->vm_start && 2947948f017bSAndrea Arcangeli vma_start < new_vma->vm_end)) { 2948948f017bSAndrea Arcangeli /* 2949948f017bSAndrea Arcangeli * The only way we can get a vma_merge with 2950948f017bSAndrea Arcangeli * self during an mremap is if the vma hasn't 2951948f017bSAndrea Arcangeli * been faulted in yet and we were allowed to 2952948f017bSAndrea Arcangeli * reset the dst vma->vm_pgoff to the 2953948f017bSAndrea Arcangeli * destination address of the mremap to allow 2954948f017bSAndrea Arcangeli * the merge to happen. mremap must change the 2955948f017bSAndrea Arcangeli * vm_pgoff linearity between src and dst vmas 2956948f017bSAndrea Arcangeli * (in turn preventing a vma_merge) to be 2957948f017bSAndrea Arcangeli * safe. It is only safe to keep the vm_pgoff 2958948f017bSAndrea Arcangeli * linear if there are no pages mapped yet. 2959948f017bSAndrea Arcangeli */ 296081d1b09cSSasha Levin VM_BUG_ON_VMA(faulted_in_anon_vma, new_vma); 296138a76013SMichel Lespinasse *vmap = vma = new_vma; 2962108d6642SMichel Lespinasse } 296338a76013SMichel Lespinasse *need_rmap_locks = (new_vma->vm_pgoff <= vma->vm_pgoff); 29641da177e4SLinus Torvalds } else { 2965e94b1766SChristoph Lameter new_vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL); 2966e3975891SChen Gang if (!new_vma) 2967e3975891SChen Gang goto out; 29681da177e4SLinus Torvalds *new_vma = *vma; 29691da177e4SLinus Torvalds new_vma->vm_start = addr; 29701da177e4SLinus Torvalds new_vma->vm_end = addr + len; 29711da177e4SLinus Torvalds new_vma->vm_pgoff = pgoff; 2972ef0855d3SOleg Nesterov if (vma_dup_policy(vma, new_vma)) 2973523d4e20SMichel Lespinasse goto out_free_vma; 2974523d4e20SMichel Lespinasse INIT_LIST_HEAD(&new_vma->anon_vma_chain); 2975523d4e20SMichel Lespinasse if (anon_vma_clone(new_vma, vma)) 2976523d4e20SMichel Lespinasse goto out_free_mempol; 2977e9714acfSKonstantin Khlebnikov if (new_vma->vm_file) 29781da177e4SLinus Torvalds get_file(new_vma->vm_file); 29791da177e4SLinus Torvalds if (new_vma->vm_ops && new_vma->vm_ops->open) 29801da177e4SLinus Torvalds new_vma->vm_ops->open(new_vma); 29811da177e4SLinus Torvalds vma_link(mm, new_vma, prev, rb_link, rb_parent); 298238a76013SMichel Lespinasse *need_rmap_locks = false; 29831da177e4SLinus Torvalds } 29841da177e4SLinus Torvalds return new_vma; 29855beb4930SRik van Riel 29865beb4930SRik van Riel out_free_mempol: 2987ef0855d3SOleg Nesterov mpol_put(vma_policy(new_vma)); 29885beb4930SRik van Riel out_free_vma: 29895beb4930SRik van Riel kmem_cache_free(vm_area_cachep, new_vma); 2990e3975891SChen Gang out: 29915beb4930SRik van Riel return NULL; 29921da177e4SLinus Torvalds } 2993119f657cSakpm@osdl.org 2994119f657cSakpm@osdl.org /* 2995119f657cSakpm@osdl.org * Return true if the calling process may expand its vm space by the passed 2996119f657cSakpm@osdl.org * number of pages 2997119f657cSakpm@osdl.org */ 2998119f657cSakpm@osdl.org int may_expand_vm(struct mm_struct *mm, unsigned long npages) 2999119f657cSakpm@osdl.org { 30000b57d6baSChen Gang return mm->total_vm + npages <= rlimit(RLIMIT_AS) >> PAGE_SHIFT; 3001119f657cSakpm@osdl.org } 3002fa5dc22fSRoland McGrath 3003a62c34bdSAndy Lutomirski static int special_mapping_fault(struct vm_area_struct *vma, 3004a62c34bdSAndy Lutomirski struct vm_fault *vmf); 3005a62c34bdSAndy Lutomirski 3006a62c34bdSAndy Lutomirski /* 3007a62c34bdSAndy Lutomirski * Having a close hook prevents vma merging regardless of flags. 3008a62c34bdSAndy Lutomirski */ 3009a62c34bdSAndy Lutomirski static void special_mapping_close(struct vm_area_struct *vma) 3010a62c34bdSAndy Lutomirski { 3011a62c34bdSAndy Lutomirski } 3012a62c34bdSAndy Lutomirski 3013a62c34bdSAndy Lutomirski static const char *special_mapping_name(struct vm_area_struct *vma) 3014a62c34bdSAndy Lutomirski { 3015a62c34bdSAndy Lutomirski return ((struct vm_special_mapping *)vma->vm_private_data)->name; 3016a62c34bdSAndy Lutomirski } 3017a62c34bdSAndy Lutomirski 3018a62c34bdSAndy Lutomirski static const struct vm_operations_struct special_mapping_vmops = { 3019a62c34bdSAndy Lutomirski .close = special_mapping_close, 3020a62c34bdSAndy Lutomirski .fault = special_mapping_fault, 3021a62c34bdSAndy Lutomirski .name = special_mapping_name, 3022a62c34bdSAndy Lutomirski }; 3023a62c34bdSAndy Lutomirski 3024a62c34bdSAndy Lutomirski static const struct vm_operations_struct legacy_special_mapping_vmops = { 3025a62c34bdSAndy Lutomirski .close = special_mapping_close, 3026a62c34bdSAndy Lutomirski .fault = special_mapping_fault, 3027a62c34bdSAndy Lutomirski }; 3028fa5dc22fSRoland McGrath 3029b1d0e4f5SNick Piggin static int special_mapping_fault(struct vm_area_struct *vma, 3030b1d0e4f5SNick Piggin struct vm_fault *vmf) 3031fa5dc22fSRoland McGrath { 3032b1d0e4f5SNick Piggin pgoff_t pgoff; 3033fa5dc22fSRoland McGrath struct page **pages; 3034fa5dc22fSRoland McGrath 3035a62c34bdSAndy Lutomirski if (vma->vm_ops == &legacy_special_mapping_vmops) 3036a62c34bdSAndy Lutomirski pages = vma->vm_private_data; 3037a62c34bdSAndy Lutomirski else 3038a62c34bdSAndy Lutomirski pages = ((struct vm_special_mapping *)vma->vm_private_data)-> 3039a62c34bdSAndy Lutomirski pages; 3040a62c34bdSAndy Lutomirski 30418a9cc3b5SOleg Nesterov for (pgoff = vmf->pgoff; pgoff && *pages; ++pages) 3042b1d0e4f5SNick Piggin pgoff--; 3043fa5dc22fSRoland McGrath 3044fa5dc22fSRoland McGrath if (*pages) { 3045fa5dc22fSRoland McGrath struct page *page = *pages; 3046fa5dc22fSRoland McGrath get_page(page); 3047b1d0e4f5SNick Piggin vmf->page = page; 3048b1d0e4f5SNick Piggin return 0; 3049fa5dc22fSRoland McGrath } 3050fa5dc22fSRoland McGrath 3051b1d0e4f5SNick Piggin return VM_FAULT_SIGBUS; 3052fa5dc22fSRoland McGrath } 3053fa5dc22fSRoland McGrath 3054a62c34bdSAndy Lutomirski static struct vm_area_struct *__install_special_mapping( 3055a62c34bdSAndy Lutomirski struct mm_struct *mm, 3056fa5dc22fSRoland McGrath unsigned long addr, unsigned long len, 305727f28b97SChen Gang unsigned long vm_flags, void *priv, 305827f28b97SChen Gang const struct vm_operations_struct *ops) 3059fa5dc22fSRoland McGrath { 3060462e635eSTavis Ormandy int ret; 3061fa5dc22fSRoland McGrath struct vm_area_struct *vma; 3062fa5dc22fSRoland McGrath 3063fa5dc22fSRoland McGrath vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL); 3064fa5dc22fSRoland McGrath if (unlikely(vma == NULL)) 30653935ed6aSStefani Seibold return ERR_PTR(-ENOMEM); 3066fa5dc22fSRoland McGrath 30675beb4930SRik van Riel INIT_LIST_HEAD(&vma->anon_vma_chain); 3068fa5dc22fSRoland McGrath vma->vm_mm = mm; 3069fa5dc22fSRoland McGrath vma->vm_start = addr; 3070fa5dc22fSRoland McGrath vma->vm_end = addr + len; 3071fa5dc22fSRoland McGrath 3072d9104d1cSCyrill Gorcunov vma->vm_flags = vm_flags | mm->def_flags | VM_DONTEXPAND | VM_SOFTDIRTY; 30733ed75eb8SColy Li vma->vm_page_prot = vm_get_page_prot(vma->vm_flags); 3074fa5dc22fSRoland McGrath 3075a62c34bdSAndy Lutomirski vma->vm_ops = ops; 3076a62c34bdSAndy Lutomirski vma->vm_private_data = priv; 3077fa5dc22fSRoland McGrath 3078462e635eSTavis Ormandy ret = insert_vm_struct(mm, vma); 3079462e635eSTavis Ormandy if (ret) 3080462e635eSTavis Ormandy goto out; 3081fa5dc22fSRoland McGrath 3082fa5dc22fSRoland McGrath mm->total_vm += len >> PAGE_SHIFT; 3083fa5dc22fSRoland McGrath 3084cdd6c482SIngo Molnar perf_event_mmap(vma); 3085089dd79dSPeter Zijlstra 30863935ed6aSStefani Seibold return vma; 3087462e635eSTavis Ormandy 3088462e635eSTavis Ormandy out: 3089462e635eSTavis Ormandy kmem_cache_free(vm_area_cachep, vma); 30903935ed6aSStefani Seibold return ERR_PTR(ret); 30913935ed6aSStefani Seibold } 30923935ed6aSStefani Seibold 3093a62c34bdSAndy Lutomirski /* 3094a62c34bdSAndy Lutomirski * Called with mm->mmap_sem held for writing. 3095a62c34bdSAndy Lutomirski * Insert a new vma covering the given region, with the given flags. 3096a62c34bdSAndy Lutomirski * Its pages are supplied by the given array of struct page *. 3097a62c34bdSAndy Lutomirski * The array can be shorter than len >> PAGE_SHIFT if it's null-terminated. 3098a62c34bdSAndy Lutomirski * The region past the last page supplied will always produce SIGBUS. 3099a62c34bdSAndy Lutomirski * The array pointer and the pages it points to are assumed to stay alive 3100a62c34bdSAndy Lutomirski * for as long as this mapping might exist. 3101a62c34bdSAndy Lutomirski */ 3102a62c34bdSAndy Lutomirski struct vm_area_struct *_install_special_mapping( 3103a62c34bdSAndy Lutomirski struct mm_struct *mm, 3104a62c34bdSAndy Lutomirski unsigned long addr, unsigned long len, 3105a62c34bdSAndy Lutomirski unsigned long vm_flags, const struct vm_special_mapping *spec) 3106a62c34bdSAndy Lutomirski { 310727f28b97SChen Gang return __install_special_mapping(mm, addr, len, vm_flags, (void *)spec, 310827f28b97SChen Gang &special_mapping_vmops); 3109a62c34bdSAndy Lutomirski } 3110a62c34bdSAndy Lutomirski 31113935ed6aSStefani Seibold int install_special_mapping(struct mm_struct *mm, 31123935ed6aSStefani Seibold unsigned long addr, unsigned long len, 31133935ed6aSStefani Seibold unsigned long vm_flags, struct page **pages) 31143935ed6aSStefani Seibold { 3115a62c34bdSAndy Lutomirski struct vm_area_struct *vma = __install_special_mapping( 311627f28b97SChen Gang mm, addr, len, vm_flags, (void *)pages, 311727f28b97SChen Gang &legacy_special_mapping_vmops); 31183935ed6aSStefani Seibold 311914bd5b45SDuan Jiong return PTR_ERR_OR_ZERO(vma); 3120fa5dc22fSRoland McGrath } 31217906d00cSAndrea Arcangeli 31227906d00cSAndrea Arcangeli static DEFINE_MUTEX(mm_all_locks_mutex); 31237906d00cSAndrea Arcangeli 3124454ed842SPeter Zijlstra static void vm_lock_anon_vma(struct mm_struct *mm, struct anon_vma *anon_vma) 31257906d00cSAndrea Arcangeli { 3126bf181b9fSMichel Lespinasse if (!test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_node)) { 31277906d00cSAndrea Arcangeli /* 31287906d00cSAndrea Arcangeli * The LSB of head.next can't change from under us 31297906d00cSAndrea Arcangeli * because we hold the mm_all_locks_mutex. 31307906d00cSAndrea Arcangeli */ 3131572043c9SJiri Kosina down_write_nest_lock(&anon_vma->root->rwsem, &mm->mmap_sem); 31327906d00cSAndrea Arcangeli /* 31337906d00cSAndrea Arcangeli * We can safely modify head.next after taking the 31345a505085SIngo Molnar * anon_vma->root->rwsem. If some other vma in this mm shares 31357906d00cSAndrea Arcangeli * the same anon_vma we won't take it again. 31367906d00cSAndrea Arcangeli * 31377906d00cSAndrea Arcangeli * No need of atomic instructions here, head.next 31387906d00cSAndrea Arcangeli * can't change from under us thanks to the 31395a505085SIngo Molnar * anon_vma->root->rwsem. 31407906d00cSAndrea Arcangeli */ 31417906d00cSAndrea Arcangeli if (__test_and_set_bit(0, (unsigned long *) 3142bf181b9fSMichel Lespinasse &anon_vma->root->rb_root.rb_node)) 31437906d00cSAndrea Arcangeli BUG(); 31447906d00cSAndrea Arcangeli } 31457906d00cSAndrea Arcangeli } 31467906d00cSAndrea Arcangeli 3147454ed842SPeter Zijlstra static void vm_lock_mapping(struct mm_struct *mm, struct address_space *mapping) 31487906d00cSAndrea Arcangeli { 31497906d00cSAndrea Arcangeli if (!test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) { 31507906d00cSAndrea Arcangeli /* 31517906d00cSAndrea Arcangeli * AS_MM_ALL_LOCKS can't change from under us because 31527906d00cSAndrea Arcangeli * we hold the mm_all_locks_mutex. 31537906d00cSAndrea Arcangeli * 31547906d00cSAndrea Arcangeli * Operations on ->flags have to be atomic because 31557906d00cSAndrea Arcangeli * even if AS_MM_ALL_LOCKS is stable thanks to the 31567906d00cSAndrea Arcangeli * mm_all_locks_mutex, there may be other cpus 31577906d00cSAndrea Arcangeli * changing other bitflags in parallel to us. 31587906d00cSAndrea Arcangeli */ 31597906d00cSAndrea Arcangeli if (test_and_set_bit(AS_MM_ALL_LOCKS, &mapping->flags)) 31607906d00cSAndrea Arcangeli BUG(); 3161c8c06efaSDavidlohr Bueso down_write_nest_lock(&mapping->i_mmap_rwsem, &mm->mmap_sem); 31627906d00cSAndrea Arcangeli } 31637906d00cSAndrea Arcangeli } 31647906d00cSAndrea Arcangeli 31657906d00cSAndrea Arcangeli /* 31667906d00cSAndrea Arcangeli * This operation locks against the VM for all pte/vma/mm related 31677906d00cSAndrea Arcangeli * operations that could ever happen on a certain mm. This includes 31687906d00cSAndrea Arcangeli * vmtruncate, try_to_unmap, and all page faults. 31697906d00cSAndrea Arcangeli * 31707906d00cSAndrea Arcangeli * The caller must take the mmap_sem in write mode before calling 31717906d00cSAndrea Arcangeli * mm_take_all_locks(). The caller isn't allowed to release the 31727906d00cSAndrea Arcangeli * mmap_sem until mm_drop_all_locks() returns. 31737906d00cSAndrea Arcangeli * 31747906d00cSAndrea Arcangeli * mmap_sem in write mode is required in order to block all operations 31757906d00cSAndrea Arcangeli * that could modify pagetables and free pages without need of 317627ba0644SKirill A. Shutemov * altering the vma layout. It's also needed in write mode to avoid new 31777906d00cSAndrea Arcangeli * anon_vmas to be associated with existing vmas. 31787906d00cSAndrea Arcangeli * 31797906d00cSAndrea Arcangeli * A single task can't take more than one mm_take_all_locks() in a row 31807906d00cSAndrea Arcangeli * or it would deadlock. 31817906d00cSAndrea Arcangeli * 3182bf181b9fSMichel Lespinasse * The LSB in anon_vma->rb_root.rb_node and the AS_MM_ALL_LOCKS bitflag in 31837906d00cSAndrea Arcangeli * mapping->flags avoid to take the same lock twice, if more than one 31847906d00cSAndrea Arcangeli * vma in this mm is backed by the same anon_vma or address_space. 31857906d00cSAndrea Arcangeli * 31867906d00cSAndrea Arcangeli * We can take all the locks in random order because the VM code 3187c8c06efaSDavidlohr Bueso * taking i_mmap_rwsem or anon_vma->rwsem outside the mmap_sem never 31887906d00cSAndrea Arcangeli * takes more than one of them in a row. Secondly we're protected 31897906d00cSAndrea Arcangeli * against a concurrent mm_take_all_locks() by the mm_all_locks_mutex. 31907906d00cSAndrea Arcangeli * 31917906d00cSAndrea Arcangeli * mm_take_all_locks() and mm_drop_all_locks are expensive operations 31927906d00cSAndrea Arcangeli * that may have to take thousand of locks. 31937906d00cSAndrea Arcangeli * 31947906d00cSAndrea Arcangeli * mm_take_all_locks() can fail if it's interrupted by signals. 31957906d00cSAndrea Arcangeli */ 31967906d00cSAndrea Arcangeli int mm_take_all_locks(struct mm_struct *mm) 31977906d00cSAndrea Arcangeli { 31987906d00cSAndrea Arcangeli struct vm_area_struct *vma; 31995beb4930SRik van Riel struct anon_vma_chain *avc; 32007906d00cSAndrea Arcangeli 32017906d00cSAndrea Arcangeli BUG_ON(down_read_trylock(&mm->mmap_sem)); 32027906d00cSAndrea Arcangeli 32037906d00cSAndrea Arcangeli mutex_lock(&mm_all_locks_mutex); 32047906d00cSAndrea Arcangeli 32057906d00cSAndrea Arcangeli for (vma = mm->mmap; vma; vma = vma->vm_next) { 32067906d00cSAndrea Arcangeli if (signal_pending(current)) 32077906d00cSAndrea Arcangeli goto out_unlock; 32087906d00cSAndrea Arcangeli if (vma->vm_file && vma->vm_file->f_mapping) 3209454ed842SPeter Zijlstra vm_lock_mapping(mm, vma->vm_file->f_mapping); 32107906d00cSAndrea Arcangeli } 32117cd5a02fSPeter Zijlstra 32127cd5a02fSPeter Zijlstra for (vma = mm->mmap; vma; vma = vma->vm_next) { 32137cd5a02fSPeter Zijlstra if (signal_pending(current)) 32147cd5a02fSPeter Zijlstra goto out_unlock; 32157cd5a02fSPeter Zijlstra if (vma->anon_vma) 32165beb4930SRik van Riel list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) 32175beb4930SRik van Riel vm_lock_anon_vma(mm, avc->anon_vma); 32187cd5a02fSPeter Zijlstra } 32197cd5a02fSPeter Zijlstra 3220584cff54SKautuk Consul return 0; 32217906d00cSAndrea Arcangeli 32227906d00cSAndrea Arcangeli out_unlock: 32237906d00cSAndrea Arcangeli mm_drop_all_locks(mm); 3224584cff54SKautuk Consul return -EINTR; 32257906d00cSAndrea Arcangeli } 32267906d00cSAndrea Arcangeli 32277906d00cSAndrea Arcangeli static void vm_unlock_anon_vma(struct anon_vma *anon_vma) 32287906d00cSAndrea Arcangeli { 3229bf181b9fSMichel Lespinasse if (test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_node)) { 32307906d00cSAndrea Arcangeli /* 32317906d00cSAndrea Arcangeli * The LSB of head.next can't change to 0 from under 32327906d00cSAndrea Arcangeli * us because we hold the mm_all_locks_mutex. 32337906d00cSAndrea Arcangeli * 32347906d00cSAndrea Arcangeli * We must however clear the bitflag before unlocking 3235bf181b9fSMichel Lespinasse * the vma so the users using the anon_vma->rb_root will 32367906d00cSAndrea Arcangeli * never see our bitflag. 32377906d00cSAndrea Arcangeli * 32387906d00cSAndrea Arcangeli * No need of atomic instructions here, head.next 32397906d00cSAndrea Arcangeli * can't change from under us until we release the 32405a505085SIngo Molnar * anon_vma->root->rwsem. 32417906d00cSAndrea Arcangeli */ 32427906d00cSAndrea Arcangeli if (!__test_and_clear_bit(0, (unsigned long *) 3243bf181b9fSMichel Lespinasse &anon_vma->root->rb_root.rb_node)) 32447906d00cSAndrea Arcangeli BUG(); 324508b52706SKonstantin Khlebnikov anon_vma_unlock_write(anon_vma); 32467906d00cSAndrea Arcangeli } 32477906d00cSAndrea Arcangeli } 32487906d00cSAndrea Arcangeli 32497906d00cSAndrea Arcangeli static void vm_unlock_mapping(struct address_space *mapping) 32507906d00cSAndrea Arcangeli { 32517906d00cSAndrea Arcangeli if (test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) { 32527906d00cSAndrea Arcangeli /* 32537906d00cSAndrea Arcangeli * AS_MM_ALL_LOCKS can't change to 0 from under us 32547906d00cSAndrea Arcangeli * because we hold the mm_all_locks_mutex. 32557906d00cSAndrea Arcangeli */ 325683cde9e8SDavidlohr Bueso i_mmap_unlock_write(mapping); 32577906d00cSAndrea Arcangeli if (!test_and_clear_bit(AS_MM_ALL_LOCKS, 32587906d00cSAndrea Arcangeli &mapping->flags)) 32597906d00cSAndrea Arcangeli BUG(); 32607906d00cSAndrea Arcangeli } 32617906d00cSAndrea Arcangeli } 32627906d00cSAndrea Arcangeli 32637906d00cSAndrea Arcangeli /* 32647906d00cSAndrea Arcangeli * The mmap_sem cannot be released by the caller until 32657906d00cSAndrea Arcangeli * mm_drop_all_locks() returns. 32667906d00cSAndrea Arcangeli */ 32677906d00cSAndrea Arcangeli void mm_drop_all_locks(struct mm_struct *mm) 32687906d00cSAndrea Arcangeli { 32697906d00cSAndrea Arcangeli struct vm_area_struct *vma; 32705beb4930SRik van Riel struct anon_vma_chain *avc; 32717906d00cSAndrea Arcangeli 32727906d00cSAndrea Arcangeli BUG_ON(down_read_trylock(&mm->mmap_sem)); 32737906d00cSAndrea Arcangeli BUG_ON(!mutex_is_locked(&mm_all_locks_mutex)); 32747906d00cSAndrea Arcangeli 32757906d00cSAndrea Arcangeli for (vma = mm->mmap; vma; vma = vma->vm_next) { 32767906d00cSAndrea Arcangeli if (vma->anon_vma) 32775beb4930SRik van Riel list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) 32785beb4930SRik van Riel vm_unlock_anon_vma(avc->anon_vma); 32797906d00cSAndrea Arcangeli if (vma->vm_file && vma->vm_file->f_mapping) 32807906d00cSAndrea Arcangeli vm_unlock_mapping(vma->vm_file->f_mapping); 32817906d00cSAndrea Arcangeli } 32827906d00cSAndrea Arcangeli 32837906d00cSAndrea Arcangeli mutex_unlock(&mm_all_locks_mutex); 32847906d00cSAndrea Arcangeli } 32858feae131SDavid Howells 32868feae131SDavid Howells /* 32878feae131SDavid Howells * initialise the VMA slab 32888feae131SDavid Howells */ 32898feae131SDavid Howells void __init mmap_init(void) 32908feae131SDavid Howells { 329100a62ce9SKOSAKI Motohiro int ret; 329200a62ce9SKOSAKI Motohiro 3293908c7f19STejun Heo ret = percpu_counter_init(&vm_committed_as, 0, GFP_KERNEL); 329400a62ce9SKOSAKI Motohiro VM_BUG_ON(ret); 32958feae131SDavid Howells } 3296c9b1d098SAndrew Shewmaker 3297c9b1d098SAndrew Shewmaker /* 3298c9b1d098SAndrew Shewmaker * Initialise sysctl_user_reserve_kbytes. 3299c9b1d098SAndrew Shewmaker * 3300c9b1d098SAndrew Shewmaker * This is intended to prevent a user from starting a single memory hogging 3301c9b1d098SAndrew Shewmaker * process, such that they cannot recover (kill the hog) in OVERCOMMIT_NEVER 3302c9b1d098SAndrew Shewmaker * mode. 3303c9b1d098SAndrew Shewmaker * 3304c9b1d098SAndrew Shewmaker * The default value is min(3% of free memory, 128MB) 3305c9b1d098SAndrew Shewmaker * 128MB is enough to recover with sshd/login, bash, and top/kill. 3306c9b1d098SAndrew Shewmaker */ 33071640879aSAndrew Shewmaker static int init_user_reserve(void) 3308c9b1d098SAndrew Shewmaker { 3309c9b1d098SAndrew Shewmaker unsigned long free_kbytes; 3310c9b1d098SAndrew Shewmaker 3311c9b1d098SAndrew Shewmaker free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10); 3312c9b1d098SAndrew Shewmaker 3313c9b1d098SAndrew Shewmaker sysctl_user_reserve_kbytes = min(free_kbytes / 32, 1UL << 17); 3314c9b1d098SAndrew Shewmaker return 0; 3315c9b1d098SAndrew Shewmaker } 3316a64fb3cdSPaul Gortmaker subsys_initcall(init_user_reserve); 33174eeab4f5SAndrew Shewmaker 33184eeab4f5SAndrew Shewmaker /* 33194eeab4f5SAndrew Shewmaker * Initialise sysctl_admin_reserve_kbytes. 33204eeab4f5SAndrew Shewmaker * 33214eeab4f5SAndrew Shewmaker * The purpose of sysctl_admin_reserve_kbytes is to allow the sys admin 33224eeab4f5SAndrew Shewmaker * to log in and kill a memory hogging process. 33234eeab4f5SAndrew Shewmaker * 33244eeab4f5SAndrew Shewmaker * Systems with more than 256MB will reserve 8MB, enough to recover 33254eeab4f5SAndrew Shewmaker * with sshd, bash, and top in OVERCOMMIT_GUESS. Smaller systems will 33264eeab4f5SAndrew Shewmaker * only reserve 3% of free pages by default. 33274eeab4f5SAndrew Shewmaker */ 33281640879aSAndrew Shewmaker static int init_admin_reserve(void) 33294eeab4f5SAndrew Shewmaker { 33304eeab4f5SAndrew Shewmaker unsigned long free_kbytes; 33314eeab4f5SAndrew Shewmaker 33324eeab4f5SAndrew Shewmaker free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10); 33334eeab4f5SAndrew Shewmaker 33344eeab4f5SAndrew Shewmaker sysctl_admin_reserve_kbytes = min(free_kbytes / 32, 1UL << 13); 33354eeab4f5SAndrew Shewmaker return 0; 33364eeab4f5SAndrew Shewmaker } 3337a64fb3cdSPaul Gortmaker subsys_initcall(init_admin_reserve); 33381640879aSAndrew Shewmaker 33391640879aSAndrew Shewmaker /* 33401640879aSAndrew Shewmaker * Reinititalise user and admin reserves if memory is added or removed. 33411640879aSAndrew Shewmaker * 33421640879aSAndrew Shewmaker * The default user reserve max is 128MB, and the default max for the 33431640879aSAndrew Shewmaker * admin reserve is 8MB. These are usually, but not always, enough to 33441640879aSAndrew Shewmaker * enable recovery from a memory hogging process using login/sshd, a shell, 33451640879aSAndrew Shewmaker * and tools like top. It may make sense to increase or even disable the 33461640879aSAndrew Shewmaker * reserve depending on the existence of swap or variations in the recovery 33471640879aSAndrew Shewmaker * tools. So, the admin may have changed them. 33481640879aSAndrew Shewmaker * 33491640879aSAndrew Shewmaker * If memory is added and the reserves have been eliminated or increased above 33501640879aSAndrew Shewmaker * the default max, then we'll trust the admin. 33511640879aSAndrew Shewmaker * 33521640879aSAndrew Shewmaker * If memory is removed and there isn't enough free memory, then we 33531640879aSAndrew Shewmaker * need to reset the reserves. 33541640879aSAndrew Shewmaker * 33551640879aSAndrew Shewmaker * Otherwise keep the reserve set by the admin. 33561640879aSAndrew Shewmaker */ 33571640879aSAndrew Shewmaker static int reserve_mem_notifier(struct notifier_block *nb, 33581640879aSAndrew Shewmaker unsigned long action, void *data) 33591640879aSAndrew Shewmaker { 33601640879aSAndrew Shewmaker unsigned long tmp, free_kbytes; 33611640879aSAndrew Shewmaker 33621640879aSAndrew Shewmaker switch (action) { 33631640879aSAndrew Shewmaker case MEM_ONLINE: 33641640879aSAndrew Shewmaker /* Default max is 128MB. Leave alone if modified by operator. */ 33651640879aSAndrew Shewmaker tmp = sysctl_user_reserve_kbytes; 33661640879aSAndrew Shewmaker if (0 < tmp && tmp < (1UL << 17)) 33671640879aSAndrew Shewmaker init_user_reserve(); 33681640879aSAndrew Shewmaker 33691640879aSAndrew Shewmaker /* Default max is 8MB. Leave alone if modified by operator. */ 33701640879aSAndrew Shewmaker tmp = sysctl_admin_reserve_kbytes; 33711640879aSAndrew Shewmaker if (0 < tmp && tmp < (1UL << 13)) 33721640879aSAndrew Shewmaker init_admin_reserve(); 33731640879aSAndrew Shewmaker 33741640879aSAndrew Shewmaker break; 33751640879aSAndrew Shewmaker case MEM_OFFLINE: 33761640879aSAndrew Shewmaker free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10); 33771640879aSAndrew Shewmaker 33781640879aSAndrew Shewmaker if (sysctl_user_reserve_kbytes > free_kbytes) { 33791640879aSAndrew Shewmaker init_user_reserve(); 33801640879aSAndrew Shewmaker pr_info("vm.user_reserve_kbytes reset to %lu\n", 33811640879aSAndrew Shewmaker sysctl_user_reserve_kbytes); 33821640879aSAndrew Shewmaker } 33831640879aSAndrew Shewmaker 33841640879aSAndrew Shewmaker if (sysctl_admin_reserve_kbytes > free_kbytes) { 33851640879aSAndrew Shewmaker init_admin_reserve(); 33861640879aSAndrew Shewmaker pr_info("vm.admin_reserve_kbytes reset to %lu\n", 33871640879aSAndrew Shewmaker sysctl_admin_reserve_kbytes); 33881640879aSAndrew Shewmaker } 33891640879aSAndrew Shewmaker break; 33901640879aSAndrew Shewmaker default: 33911640879aSAndrew Shewmaker break; 33921640879aSAndrew Shewmaker } 33931640879aSAndrew Shewmaker return NOTIFY_OK; 33941640879aSAndrew Shewmaker } 33951640879aSAndrew Shewmaker 33961640879aSAndrew Shewmaker static struct notifier_block reserve_mem_nb = { 33971640879aSAndrew Shewmaker .notifier_call = reserve_mem_notifier, 33981640879aSAndrew Shewmaker }; 33991640879aSAndrew Shewmaker 34001640879aSAndrew Shewmaker static int __meminit init_reserve_notifier(void) 34011640879aSAndrew Shewmaker { 34021640879aSAndrew Shewmaker if (register_hotmemory_notifier(&reserve_mem_nb)) 3403b1de0d13SMitchel Humpherys pr_err("Failed registering memory add/remove notifier for admin reserve\n"); 34041640879aSAndrew Shewmaker 34051640879aSAndrew Shewmaker return 0; 34061640879aSAndrew Shewmaker } 3407a64fb3cdSPaul Gortmaker subsys_initcall(init_reserve_notifier); 3408