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> 28c01d5b30SHugh Dickins #include <linux/shmem_fs.h> 291da177e4SLinus Torvalds #include <linux/profile.h> 30b95f1b31SPaul Gortmaker #include <linux/export.h> 311da177e4SLinus Torvalds #include <linux/mount.h> 321da177e4SLinus Torvalds #include <linux/mempolicy.h> 331da177e4SLinus Torvalds #include <linux/rmap.h> 34cddb8a5cSAndrea Arcangeli #include <linux/mmu_notifier.h> 3582f71ae4SKonstantin Khlebnikov #include <linux/mmdebug.h> 36cdd6c482SIngo Molnar #include <linux/perf_event.h> 37120a795dSAl Viro #include <linux/audit.h> 38b15d00b6SAndrea Arcangeli #include <linux/khugepaged.h> 392b144498SSrikar Dronamraju #include <linux/uprobes.h> 40d3737187SMichel Lespinasse #include <linux/rbtree_augmented.h> 411640879aSAndrew Shewmaker #include <linux/notifier.h> 421640879aSAndrew Shewmaker #include <linux/memory.h> 43b1de0d13SMitchel Humpherys #include <linux/printk.h> 4419a809afSAndrea Arcangeli #include <linux/userfaultfd_k.h> 45d977d56cSKonstantin Khlebnikov #include <linux/moduleparam.h> 4662b5f7d0SDave Hansen #include <linux/pkeys.h> 47*21292580SAndrea Arcangeli #include <linux/oom.h> 481da177e4SLinus Torvalds 497c0f6ba6SLinus Torvalds #include <linux/uaccess.h> 501da177e4SLinus Torvalds #include <asm/cacheflush.h> 511da177e4SLinus Torvalds #include <asm/tlb.h> 52d6dd61c8SJeremy Fitzhardinge #include <asm/mmu_context.h> 531da177e4SLinus Torvalds 5442b77728SJan Beulich #include "internal.h" 5542b77728SJan Beulich 563a459756SKirill Korotaev #ifndef arch_mmap_check 573a459756SKirill Korotaev #define arch_mmap_check(addr, len, flags) (0) 583a459756SKirill Korotaev #endif 593a459756SKirill Korotaev 60d07e2259SDaniel Cashman #ifdef CONFIG_HAVE_ARCH_MMAP_RND_BITS 61d07e2259SDaniel Cashman const int mmap_rnd_bits_min = CONFIG_ARCH_MMAP_RND_BITS_MIN; 62d07e2259SDaniel Cashman const int mmap_rnd_bits_max = CONFIG_ARCH_MMAP_RND_BITS_MAX; 63d07e2259SDaniel Cashman int mmap_rnd_bits __read_mostly = CONFIG_ARCH_MMAP_RND_BITS; 64d07e2259SDaniel Cashman #endif 65d07e2259SDaniel Cashman #ifdef CONFIG_HAVE_ARCH_MMAP_RND_COMPAT_BITS 66d07e2259SDaniel Cashman const int mmap_rnd_compat_bits_min = CONFIG_ARCH_MMAP_RND_COMPAT_BITS_MIN; 67d07e2259SDaniel Cashman const int mmap_rnd_compat_bits_max = CONFIG_ARCH_MMAP_RND_COMPAT_BITS_MAX; 68d07e2259SDaniel Cashman int mmap_rnd_compat_bits __read_mostly = CONFIG_ARCH_MMAP_RND_COMPAT_BITS; 69d07e2259SDaniel Cashman #endif 70d07e2259SDaniel Cashman 71f4fcd558SKonstantin Khlebnikov static bool ignore_rlimit_data; 72d977d56cSKonstantin Khlebnikov core_param(ignore_rlimit_data, ignore_rlimit_data, bool, 0644); 73d07e2259SDaniel Cashman 74e0da382cSHugh Dickins static void unmap_region(struct mm_struct *mm, 75e0da382cSHugh Dickins struct vm_area_struct *vma, struct vm_area_struct *prev, 76e0da382cSHugh Dickins unsigned long start, unsigned long end); 77e0da382cSHugh Dickins 781da177e4SLinus Torvalds /* description of effects of mapping type and prot in current implementation. 791da177e4SLinus Torvalds * this is due to the limited x86 page protection hardware. The expected 801da177e4SLinus Torvalds * behavior is in parens: 811da177e4SLinus Torvalds * 821da177e4SLinus Torvalds * map_type prot 831da177e4SLinus Torvalds * PROT_NONE PROT_READ PROT_WRITE PROT_EXEC 841da177e4SLinus Torvalds * MAP_SHARED r: (no) no r: (yes) yes r: (no) yes r: (no) yes 851da177e4SLinus Torvalds * w: (no) no w: (no) no w: (yes) yes w: (no) no 861da177e4SLinus Torvalds * x: (no) no x: (no) yes x: (no) yes x: (yes) yes 871da177e4SLinus Torvalds * 881da177e4SLinus Torvalds * MAP_PRIVATE r: (no) no r: (yes) yes r: (no) yes r: (no) yes 891da177e4SLinus Torvalds * w: (no) no w: (no) no w: (copy) copy w: (no) no 901da177e4SLinus Torvalds * x: (no) no x: (no) yes x: (no) yes x: (yes) yes 911da177e4SLinus Torvalds * 92cab15ce6SCatalin Marinas * On arm64, PROT_EXEC has the following behaviour for both MAP_SHARED and 93cab15ce6SCatalin Marinas * MAP_PRIVATE: 94cab15ce6SCatalin Marinas * r: (no) no 95cab15ce6SCatalin Marinas * w: (no) no 96cab15ce6SCatalin Marinas * x: (yes) yes 971da177e4SLinus Torvalds */ 98ac34ceafSDaniel Micay pgprot_t protection_map[16] __ro_after_init = { 991da177e4SLinus Torvalds __P000, __P001, __P010, __P011, __P100, __P101, __P110, __P111, 1001da177e4SLinus Torvalds __S000, __S001, __S010, __S011, __S100, __S101, __S110, __S111 1011da177e4SLinus Torvalds }; 1021da177e4SLinus Torvalds 103804af2cfSHugh Dickins pgprot_t vm_get_page_prot(unsigned long vm_flags) 104804af2cfSHugh Dickins { 105b845f313SDave Kleikamp return __pgprot(pgprot_val(protection_map[vm_flags & 106b845f313SDave Kleikamp (VM_READ|VM_WRITE|VM_EXEC|VM_SHARED)]) | 107b845f313SDave Kleikamp pgprot_val(arch_vm_get_page_prot(vm_flags))); 108804af2cfSHugh Dickins } 109804af2cfSHugh Dickins EXPORT_SYMBOL(vm_get_page_prot); 110804af2cfSHugh Dickins 11164e45507SPeter Feiner static pgprot_t vm_pgprot_modify(pgprot_t oldprot, unsigned long vm_flags) 11264e45507SPeter Feiner { 11364e45507SPeter Feiner return pgprot_modify(oldprot, vm_get_page_prot(vm_flags)); 11464e45507SPeter Feiner } 11564e45507SPeter Feiner 11664e45507SPeter Feiner /* Update vma->vm_page_prot to reflect vma->vm_flags. */ 11764e45507SPeter Feiner void vma_set_page_prot(struct vm_area_struct *vma) 11864e45507SPeter Feiner { 11964e45507SPeter Feiner unsigned long vm_flags = vma->vm_flags; 1206d2329f8SAndrea Arcangeli pgprot_t vm_page_prot; 12164e45507SPeter Feiner 1226d2329f8SAndrea Arcangeli vm_page_prot = vm_pgprot_modify(vma->vm_page_prot, vm_flags); 1236d2329f8SAndrea Arcangeli if (vma_wants_writenotify(vma, vm_page_prot)) { 12464e45507SPeter Feiner vm_flags &= ~VM_SHARED; 1256d2329f8SAndrea Arcangeli vm_page_prot = vm_pgprot_modify(vm_page_prot, vm_flags); 12664e45507SPeter Feiner } 1276d2329f8SAndrea Arcangeli /* remove_protection_ptes reads vma->vm_page_prot without mmap_sem */ 1286d2329f8SAndrea Arcangeli WRITE_ONCE(vma->vm_page_prot, vm_page_prot); 12964e45507SPeter Feiner } 13064e45507SPeter Feiner 1311da177e4SLinus Torvalds /* 132c8c06efaSDavidlohr Bueso * Requires inode->i_mapping->i_mmap_rwsem 1331da177e4SLinus Torvalds */ 1341da177e4SLinus Torvalds static void __remove_shared_vm_struct(struct vm_area_struct *vma, 1351da177e4SLinus Torvalds struct file *file, struct address_space *mapping) 1361da177e4SLinus Torvalds { 1371da177e4SLinus Torvalds if (vma->vm_flags & VM_DENYWRITE) 138496ad9aaSAl Viro atomic_inc(&file_inode(file)->i_writecount); 1391da177e4SLinus Torvalds if (vma->vm_flags & VM_SHARED) 1404bb5f5d9SDavid Herrmann mapping_unmap_writable(mapping); 1411da177e4SLinus Torvalds 1421da177e4SLinus Torvalds flush_dcache_mmap_lock(mapping); 1436b2dbba8SMichel Lespinasse vma_interval_tree_remove(vma, &mapping->i_mmap); 1441da177e4SLinus Torvalds flush_dcache_mmap_unlock(mapping); 1451da177e4SLinus Torvalds } 1461da177e4SLinus Torvalds 1471da177e4SLinus Torvalds /* 1486b2dbba8SMichel Lespinasse * Unlink a file-based vm structure from its interval tree, to hide 149a8fb5618SHugh Dickins * vma from rmap and vmtruncate before freeing its page tables. 1501da177e4SLinus Torvalds */ 151a8fb5618SHugh Dickins void unlink_file_vma(struct vm_area_struct *vma) 1521da177e4SLinus Torvalds { 1531da177e4SLinus Torvalds struct file *file = vma->vm_file; 1541da177e4SLinus Torvalds 1551da177e4SLinus Torvalds if (file) { 1561da177e4SLinus Torvalds struct address_space *mapping = file->f_mapping; 15783cde9e8SDavidlohr Bueso i_mmap_lock_write(mapping); 1581da177e4SLinus Torvalds __remove_shared_vm_struct(vma, file, mapping); 15983cde9e8SDavidlohr Bueso i_mmap_unlock_write(mapping); 1601da177e4SLinus Torvalds } 161a8fb5618SHugh Dickins } 162a8fb5618SHugh Dickins 163a8fb5618SHugh Dickins /* 164a8fb5618SHugh Dickins * Close a vm structure and free it, returning the next. 165a8fb5618SHugh Dickins */ 166a8fb5618SHugh Dickins static struct vm_area_struct *remove_vma(struct vm_area_struct *vma) 167a8fb5618SHugh Dickins { 168a8fb5618SHugh Dickins struct vm_area_struct *next = vma->vm_next; 169a8fb5618SHugh Dickins 170a8fb5618SHugh Dickins might_sleep(); 1711da177e4SLinus Torvalds if (vma->vm_ops && vma->vm_ops->close) 1721da177e4SLinus Torvalds vma->vm_ops->close(vma); 173e9714acfSKonstantin Khlebnikov if (vma->vm_file) 174a8fb5618SHugh Dickins fput(vma->vm_file); 175f0be3d32SLee Schermerhorn mpol_put(vma_policy(vma)); 1761da177e4SLinus Torvalds kmem_cache_free(vm_area_cachep, vma); 177a8fb5618SHugh Dickins return next; 1781da177e4SLinus Torvalds } 1791da177e4SLinus Torvalds 180897ab3e0SMike Rapoport static int do_brk(unsigned long addr, unsigned long len, struct list_head *uf); 181e4eb1ff6SLinus Torvalds 1826a6160a7SHeiko Carstens SYSCALL_DEFINE1(brk, unsigned long, brk) 1831da177e4SLinus Torvalds { 1848764b338SCyrill Gorcunov unsigned long retval; 1851da177e4SLinus Torvalds unsigned long newbrk, oldbrk; 1861da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 1871be7107fSHugh Dickins struct vm_area_struct *next; 188a5b4592cSJiri Kosina unsigned long min_brk; 189128557ffSMichel Lespinasse bool populate; 190897ab3e0SMike Rapoport LIST_HEAD(uf); 1911da177e4SLinus Torvalds 192dc0ef0dfSMichal Hocko if (down_write_killable(&mm->mmap_sem)) 193dc0ef0dfSMichal Hocko return -EINTR; 1941da177e4SLinus Torvalds 195a5b4592cSJiri Kosina #ifdef CONFIG_COMPAT_BRK 1965520e894SJiri Kosina /* 1975520e894SJiri Kosina * CONFIG_COMPAT_BRK can still be overridden by setting 1985520e894SJiri Kosina * randomize_va_space to 2, which will still cause mm->start_brk 1995520e894SJiri Kosina * to be arbitrarily shifted 2005520e894SJiri Kosina */ 2014471a675SJiri Kosina if (current->brk_randomized) 2025520e894SJiri Kosina min_brk = mm->start_brk; 2035520e894SJiri Kosina else 2045520e894SJiri Kosina min_brk = mm->end_data; 205a5b4592cSJiri Kosina #else 206a5b4592cSJiri Kosina min_brk = mm->start_brk; 207a5b4592cSJiri Kosina #endif 208a5b4592cSJiri Kosina if (brk < min_brk) 2091da177e4SLinus Torvalds goto out; 2101e624196SRam Gupta 2111e624196SRam Gupta /* 2121e624196SRam Gupta * Check against rlimit here. If this check is done later after the test 2131e624196SRam Gupta * of oldbrk with newbrk then it can escape the test and let the data 2141e624196SRam Gupta * segment grow beyond its set limit the in case where the limit is 2151e624196SRam Gupta * not page aligned -Ram Gupta 2161e624196SRam Gupta */ 2178764b338SCyrill Gorcunov if (check_data_rlimit(rlimit(RLIMIT_DATA), brk, mm->start_brk, 2188764b338SCyrill Gorcunov mm->end_data, mm->start_data)) 2191e624196SRam Gupta goto out; 2201e624196SRam Gupta 2211da177e4SLinus Torvalds newbrk = PAGE_ALIGN(brk); 2221da177e4SLinus Torvalds oldbrk = PAGE_ALIGN(mm->brk); 2231da177e4SLinus Torvalds if (oldbrk == newbrk) 2241da177e4SLinus Torvalds goto set_brk; 2251da177e4SLinus Torvalds 2261da177e4SLinus Torvalds /* Always allow shrinking brk. */ 2271da177e4SLinus Torvalds if (brk <= mm->brk) { 228897ab3e0SMike Rapoport if (!do_munmap(mm, newbrk, oldbrk-newbrk, &uf)) 2291da177e4SLinus Torvalds goto set_brk; 2301da177e4SLinus Torvalds goto out; 2311da177e4SLinus Torvalds } 2321da177e4SLinus Torvalds 2331da177e4SLinus Torvalds /* Check against existing mmap mappings. */ 2341be7107fSHugh Dickins next = find_vma(mm, oldbrk); 2351be7107fSHugh Dickins if (next && newbrk + PAGE_SIZE > vm_start_gap(next)) 2361da177e4SLinus Torvalds goto out; 2371da177e4SLinus Torvalds 2381da177e4SLinus Torvalds /* Ok, looks good - let it rip. */ 239897ab3e0SMike Rapoport if (do_brk(oldbrk, newbrk-oldbrk, &uf) < 0) 2401da177e4SLinus Torvalds goto out; 241128557ffSMichel Lespinasse 2421da177e4SLinus Torvalds set_brk: 2431da177e4SLinus Torvalds mm->brk = brk; 244128557ffSMichel Lespinasse populate = newbrk > oldbrk && (mm->def_flags & VM_LOCKED) != 0; 245128557ffSMichel Lespinasse up_write(&mm->mmap_sem); 246897ab3e0SMike Rapoport userfaultfd_unmap_complete(mm, &uf); 247128557ffSMichel Lespinasse if (populate) 248128557ffSMichel Lespinasse mm_populate(oldbrk, newbrk - oldbrk); 249128557ffSMichel Lespinasse return brk; 250128557ffSMichel Lespinasse 2511da177e4SLinus Torvalds out: 2521da177e4SLinus Torvalds retval = mm->brk; 2531da177e4SLinus Torvalds up_write(&mm->mmap_sem); 2541da177e4SLinus Torvalds return retval; 2551da177e4SLinus Torvalds } 2561da177e4SLinus Torvalds 257d3737187SMichel Lespinasse static long vma_compute_subtree_gap(struct vm_area_struct *vma) 258d3737187SMichel Lespinasse { 2591be7107fSHugh Dickins unsigned long max, prev_end, subtree_gap; 2601be7107fSHugh Dickins 2611be7107fSHugh Dickins /* 2621be7107fSHugh Dickins * Note: in the rare case of a VM_GROWSDOWN above a VM_GROWSUP, we 2631be7107fSHugh Dickins * allow two stack_guard_gaps between them here, and when choosing 2641be7107fSHugh Dickins * an unmapped area; whereas when expanding we only require one. 2651be7107fSHugh Dickins * That's a little inconsistent, but keeps the code here simpler. 2661be7107fSHugh Dickins */ 2671be7107fSHugh Dickins max = vm_start_gap(vma); 2681be7107fSHugh Dickins if (vma->vm_prev) { 2691be7107fSHugh Dickins prev_end = vm_end_gap(vma->vm_prev); 2701be7107fSHugh Dickins if (max > prev_end) 2711be7107fSHugh Dickins max -= prev_end; 2721be7107fSHugh Dickins else 2731be7107fSHugh Dickins max = 0; 2741be7107fSHugh Dickins } 275d3737187SMichel Lespinasse if (vma->vm_rb.rb_left) { 276d3737187SMichel Lespinasse subtree_gap = rb_entry(vma->vm_rb.rb_left, 277d3737187SMichel Lespinasse struct vm_area_struct, vm_rb)->rb_subtree_gap; 278d3737187SMichel Lespinasse if (subtree_gap > max) 279d3737187SMichel Lespinasse max = subtree_gap; 280d3737187SMichel Lespinasse } 281d3737187SMichel Lespinasse if (vma->vm_rb.rb_right) { 282d3737187SMichel Lespinasse subtree_gap = rb_entry(vma->vm_rb.rb_right, 283d3737187SMichel Lespinasse struct vm_area_struct, vm_rb)->rb_subtree_gap; 284d3737187SMichel Lespinasse if (subtree_gap > max) 285d3737187SMichel Lespinasse max = subtree_gap; 286d3737187SMichel Lespinasse } 287d3737187SMichel Lespinasse return max; 288d3737187SMichel Lespinasse } 289d3737187SMichel Lespinasse 290ed8ea815SMichel Lespinasse #ifdef CONFIG_DEBUG_VM_RB 291acf128d0SAndrea Arcangeli static int browse_rb(struct mm_struct *mm) 2921da177e4SLinus Torvalds { 293acf128d0SAndrea Arcangeli struct rb_root *root = &mm->mm_rb; 2945a0768f6SMichel Lespinasse int i = 0, j, bug = 0; 2951da177e4SLinus Torvalds struct rb_node *nd, *pn = NULL; 2961da177e4SLinus Torvalds unsigned long prev = 0, pend = 0; 2971da177e4SLinus Torvalds 2981da177e4SLinus Torvalds for (nd = rb_first(root); nd; nd = rb_next(nd)) { 2991da177e4SLinus Torvalds struct vm_area_struct *vma; 3001da177e4SLinus Torvalds vma = rb_entry(nd, struct vm_area_struct, vm_rb); 3015a0768f6SMichel Lespinasse if (vma->vm_start < prev) { 302ff26f70fSAndrew Morton pr_emerg("vm_start %lx < prev %lx\n", 303ff26f70fSAndrew Morton vma->vm_start, prev); 3045a0768f6SMichel Lespinasse bug = 1; 3055a0768f6SMichel Lespinasse } 3065a0768f6SMichel Lespinasse if (vma->vm_start < pend) { 307ff26f70fSAndrew Morton pr_emerg("vm_start %lx < pend %lx\n", 308ff26f70fSAndrew Morton vma->vm_start, pend); 3095a0768f6SMichel Lespinasse bug = 1; 3105a0768f6SMichel Lespinasse } 3115a0768f6SMichel Lespinasse if (vma->vm_start > vma->vm_end) { 312ff26f70fSAndrew Morton pr_emerg("vm_start %lx > vm_end %lx\n", 313ff26f70fSAndrew Morton vma->vm_start, vma->vm_end); 3145a0768f6SMichel Lespinasse bug = 1; 3155a0768f6SMichel Lespinasse } 316acf128d0SAndrea Arcangeli spin_lock(&mm->page_table_lock); 3175a0768f6SMichel Lespinasse if (vma->rb_subtree_gap != vma_compute_subtree_gap(vma)) { 3188542bdfcSSasha Levin pr_emerg("free gap %lx, correct %lx\n", 3195a0768f6SMichel Lespinasse vma->rb_subtree_gap, 3205a0768f6SMichel Lespinasse vma_compute_subtree_gap(vma)); 3215a0768f6SMichel Lespinasse bug = 1; 3225a0768f6SMichel Lespinasse } 323acf128d0SAndrea Arcangeli spin_unlock(&mm->page_table_lock); 3241da177e4SLinus Torvalds i++; 3251da177e4SLinus Torvalds pn = nd; 326d1af65d1SDavid Miller prev = vma->vm_start; 327d1af65d1SDavid Miller pend = vma->vm_end; 3281da177e4SLinus Torvalds } 3291da177e4SLinus Torvalds j = 0; 3305a0768f6SMichel Lespinasse for (nd = pn; nd; nd = rb_prev(nd)) 3311da177e4SLinus Torvalds j++; 3325a0768f6SMichel Lespinasse if (i != j) { 3338542bdfcSSasha Levin pr_emerg("backwards %d, forwards %d\n", j, i); 3345a0768f6SMichel Lespinasse bug = 1; 3351da177e4SLinus Torvalds } 3365a0768f6SMichel Lespinasse return bug ? -1 : i; 3371da177e4SLinus Torvalds } 3381da177e4SLinus Torvalds 339d3737187SMichel Lespinasse static void validate_mm_rb(struct rb_root *root, struct vm_area_struct *ignore) 340d3737187SMichel Lespinasse { 341d3737187SMichel Lespinasse struct rb_node *nd; 342d3737187SMichel Lespinasse 343d3737187SMichel Lespinasse for (nd = rb_first(root); nd; nd = rb_next(nd)) { 344d3737187SMichel Lespinasse struct vm_area_struct *vma; 345d3737187SMichel Lespinasse vma = rb_entry(nd, struct vm_area_struct, vm_rb); 34696dad67fSSasha Levin VM_BUG_ON_VMA(vma != ignore && 34796dad67fSSasha Levin vma->rb_subtree_gap != vma_compute_subtree_gap(vma), 34896dad67fSSasha Levin vma); 349d3737187SMichel Lespinasse } 3501da177e4SLinus Torvalds } 3511da177e4SLinus Torvalds 352eafd4dc4SRashika Kheria static void validate_mm(struct mm_struct *mm) 3531da177e4SLinus Torvalds { 3541da177e4SLinus Torvalds int bug = 0; 3551da177e4SLinus Torvalds int i = 0; 3565a0768f6SMichel Lespinasse unsigned long highest_address = 0; 357ed8ea815SMichel Lespinasse struct vm_area_struct *vma = mm->mmap; 358ff26f70fSAndrew Morton 359ed8ea815SMichel Lespinasse while (vma) { 36012352d3cSKonstantin Khlebnikov struct anon_vma *anon_vma = vma->anon_vma; 361ed8ea815SMichel Lespinasse struct anon_vma_chain *avc; 362ff26f70fSAndrew Morton 36312352d3cSKonstantin Khlebnikov if (anon_vma) { 36412352d3cSKonstantin Khlebnikov anon_vma_lock_read(anon_vma); 365ed8ea815SMichel Lespinasse list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) 366ed8ea815SMichel Lespinasse anon_vma_interval_tree_verify(avc); 36712352d3cSKonstantin Khlebnikov anon_vma_unlock_read(anon_vma); 36812352d3cSKonstantin Khlebnikov } 36912352d3cSKonstantin Khlebnikov 3701be7107fSHugh Dickins highest_address = vm_end_gap(vma); 371ed8ea815SMichel Lespinasse vma = vma->vm_next; 3721da177e4SLinus Torvalds i++; 3731da177e4SLinus Torvalds } 3745a0768f6SMichel Lespinasse if (i != mm->map_count) { 3758542bdfcSSasha Levin pr_emerg("map_count %d vm_next %d\n", mm->map_count, i); 3765a0768f6SMichel Lespinasse bug = 1; 3775a0768f6SMichel Lespinasse } 3785a0768f6SMichel Lespinasse if (highest_address != mm->highest_vm_end) { 3798542bdfcSSasha Levin pr_emerg("mm->highest_vm_end %lx, found %lx\n", 3805a0768f6SMichel Lespinasse mm->highest_vm_end, highest_address); 3815a0768f6SMichel Lespinasse bug = 1; 3825a0768f6SMichel Lespinasse } 383acf128d0SAndrea Arcangeli i = browse_rb(mm); 3845a0768f6SMichel Lespinasse if (i != mm->map_count) { 385ff26f70fSAndrew Morton if (i != -1) 3868542bdfcSSasha Levin pr_emerg("map_count %d rb %d\n", mm->map_count, i); 3875a0768f6SMichel Lespinasse bug = 1; 3885a0768f6SMichel Lespinasse } 38996dad67fSSasha Levin VM_BUG_ON_MM(bug, mm); 3901da177e4SLinus Torvalds } 3911da177e4SLinus Torvalds #else 392d3737187SMichel Lespinasse #define validate_mm_rb(root, ignore) do { } while (0) 3931da177e4SLinus Torvalds #define validate_mm(mm) do { } while (0) 3941da177e4SLinus Torvalds #endif 3951da177e4SLinus Torvalds 396d3737187SMichel Lespinasse RB_DECLARE_CALLBACKS(static, vma_gap_callbacks, struct vm_area_struct, vm_rb, 397d3737187SMichel Lespinasse unsigned long, rb_subtree_gap, vma_compute_subtree_gap) 398d3737187SMichel Lespinasse 399d3737187SMichel Lespinasse /* 400d3737187SMichel Lespinasse * Update augmented rbtree rb_subtree_gap values after vma->vm_start or 401d3737187SMichel Lespinasse * vma->vm_prev->vm_end values changed, without modifying the vma's position 402d3737187SMichel Lespinasse * in the rbtree. 403d3737187SMichel Lespinasse */ 404d3737187SMichel Lespinasse static void vma_gap_update(struct vm_area_struct *vma) 405d3737187SMichel Lespinasse { 406d3737187SMichel Lespinasse /* 407d3737187SMichel Lespinasse * As it turns out, RB_DECLARE_CALLBACKS() already created a callback 408d3737187SMichel Lespinasse * function that does exacltly what we want. 409d3737187SMichel Lespinasse */ 410d3737187SMichel Lespinasse vma_gap_callbacks_propagate(&vma->vm_rb, NULL); 411d3737187SMichel Lespinasse } 412d3737187SMichel Lespinasse 413d3737187SMichel Lespinasse static inline void vma_rb_insert(struct vm_area_struct *vma, 414d3737187SMichel Lespinasse struct rb_root *root) 415d3737187SMichel Lespinasse { 416d3737187SMichel Lespinasse /* All rb_subtree_gap values must be consistent prior to insertion */ 417d3737187SMichel Lespinasse validate_mm_rb(root, NULL); 418d3737187SMichel Lespinasse 419d3737187SMichel Lespinasse rb_insert_augmented(&vma->vm_rb, root, &vma_gap_callbacks); 420d3737187SMichel Lespinasse } 421d3737187SMichel Lespinasse 4228f26e0b1SAndrea Arcangeli static void __vma_rb_erase(struct vm_area_struct *vma, struct rb_root *root) 4238f26e0b1SAndrea Arcangeli { 4248f26e0b1SAndrea Arcangeli /* 4258f26e0b1SAndrea Arcangeli * Note rb_erase_augmented is a fairly large inline function, 4268f26e0b1SAndrea Arcangeli * so make sure we instantiate it only once with our desired 4278f26e0b1SAndrea Arcangeli * augmented rbtree callbacks. 4288f26e0b1SAndrea Arcangeli */ 4298f26e0b1SAndrea Arcangeli rb_erase_augmented(&vma->vm_rb, root, &vma_gap_callbacks); 4308f26e0b1SAndrea Arcangeli } 4318f26e0b1SAndrea Arcangeli 4328f26e0b1SAndrea Arcangeli static __always_inline void vma_rb_erase_ignore(struct vm_area_struct *vma, 4338f26e0b1SAndrea Arcangeli struct rb_root *root, 4348f26e0b1SAndrea Arcangeli struct vm_area_struct *ignore) 4358f26e0b1SAndrea Arcangeli { 4368f26e0b1SAndrea Arcangeli /* 4378f26e0b1SAndrea Arcangeli * All rb_subtree_gap values must be consistent prior to erase, 4388f26e0b1SAndrea Arcangeli * with the possible exception of the "next" vma being erased if 4398f26e0b1SAndrea Arcangeli * next->vm_start was reduced. 4408f26e0b1SAndrea Arcangeli */ 4418f26e0b1SAndrea Arcangeli validate_mm_rb(root, ignore); 4428f26e0b1SAndrea Arcangeli 4438f26e0b1SAndrea Arcangeli __vma_rb_erase(vma, root); 4448f26e0b1SAndrea Arcangeli } 4458f26e0b1SAndrea Arcangeli 4468f26e0b1SAndrea Arcangeli static __always_inline void vma_rb_erase(struct vm_area_struct *vma, 4478f26e0b1SAndrea Arcangeli struct rb_root *root) 448d3737187SMichel Lespinasse { 449d3737187SMichel Lespinasse /* 450d3737187SMichel Lespinasse * All rb_subtree_gap values must be consistent prior to erase, 451d3737187SMichel Lespinasse * with the possible exception of the vma being erased. 452d3737187SMichel Lespinasse */ 453d3737187SMichel Lespinasse validate_mm_rb(root, vma); 454d3737187SMichel Lespinasse 4558f26e0b1SAndrea Arcangeli __vma_rb_erase(vma, root); 456d3737187SMichel Lespinasse } 457d3737187SMichel Lespinasse 458bf181b9fSMichel Lespinasse /* 459bf181b9fSMichel Lespinasse * vma has some anon_vma assigned, and is already inserted on that 460bf181b9fSMichel Lespinasse * anon_vma's interval trees. 461bf181b9fSMichel Lespinasse * 462bf181b9fSMichel Lespinasse * Before updating the vma's vm_start / vm_end / vm_pgoff fields, the 463bf181b9fSMichel Lespinasse * vma must be removed from the anon_vma's interval trees using 464bf181b9fSMichel Lespinasse * anon_vma_interval_tree_pre_update_vma(). 465bf181b9fSMichel Lespinasse * 466bf181b9fSMichel Lespinasse * After the update, the vma will be reinserted using 467bf181b9fSMichel Lespinasse * anon_vma_interval_tree_post_update_vma(). 468bf181b9fSMichel Lespinasse * 469bf181b9fSMichel Lespinasse * The entire update must be protected by exclusive mmap_sem and by 470bf181b9fSMichel Lespinasse * the root anon_vma's mutex. 471bf181b9fSMichel Lespinasse */ 472bf181b9fSMichel Lespinasse static inline void 473bf181b9fSMichel Lespinasse anon_vma_interval_tree_pre_update_vma(struct vm_area_struct *vma) 474bf181b9fSMichel Lespinasse { 475bf181b9fSMichel Lespinasse struct anon_vma_chain *avc; 476bf181b9fSMichel Lespinasse 477bf181b9fSMichel Lespinasse list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) 478bf181b9fSMichel Lespinasse anon_vma_interval_tree_remove(avc, &avc->anon_vma->rb_root); 479bf181b9fSMichel Lespinasse } 480bf181b9fSMichel Lespinasse 481bf181b9fSMichel Lespinasse static inline void 482bf181b9fSMichel Lespinasse anon_vma_interval_tree_post_update_vma(struct vm_area_struct *vma) 483bf181b9fSMichel Lespinasse { 484bf181b9fSMichel Lespinasse struct anon_vma_chain *avc; 485bf181b9fSMichel Lespinasse 486bf181b9fSMichel Lespinasse list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) 487bf181b9fSMichel Lespinasse anon_vma_interval_tree_insert(avc, &avc->anon_vma->rb_root); 488bf181b9fSMichel Lespinasse } 489bf181b9fSMichel Lespinasse 4906597d783SHugh Dickins static int find_vma_links(struct mm_struct *mm, unsigned long addr, 4916597d783SHugh Dickins unsigned long end, struct vm_area_struct **pprev, 4926597d783SHugh Dickins struct rb_node ***rb_link, struct rb_node **rb_parent) 4931da177e4SLinus Torvalds { 4941da177e4SLinus Torvalds struct rb_node **__rb_link, *__rb_parent, *rb_prev; 4951da177e4SLinus Torvalds 4961da177e4SLinus Torvalds __rb_link = &mm->mm_rb.rb_node; 4971da177e4SLinus Torvalds rb_prev = __rb_parent = NULL; 4981da177e4SLinus Torvalds 4991da177e4SLinus Torvalds while (*__rb_link) { 5001da177e4SLinus Torvalds struct vm_area_struct *vma_tmp; 5011da177e4SLinus Torvalds 5021da177e4SLinus Torvalds __rb_parent = *__rb_link; 5031da177e4SLinus Torvalds vma_tmp = rb_entry(__rb_parent, struct vm_area_struct, vm_rb); 5041da177e4SLinus Torvalds 5051da177e4SLinus Torvalds if (vma_tmp->vm_end > addr) { 5066597d783SHugh Dickins /* Fail if an existing vma overlaps the area */ 5076597d783SHugh Dickins if (vma_tmp->vm_start < end) 5086597d783SHugh Dickins return -ENOMEM; 5091da177e4SLinus Torvalds __rb_link = &__rb_parent->rb_left; 5101da177e4SLinus Torvalds } else { 5111da177e4SLinus Torvalds rb_prev = __rb_parent; 5121da177e4SLinus Torvalds __rb_link = &__rb_parent->rb_right; 5131da177e4SLinus Torvalds } 5141da177e4SLinus Torvalds } 5151da177e4SLinus Torvalds 5161da177e4SLinus Torvalds *pprev = NULL; 5171da177e4SLinus Torvalds if (rb_prev) 5181da177e4SLinus Torvalds *pprev = rb_entry(rb_prev, struct vm_area_struct, vm_rb); 5191da177e4SLinus Torvalds *rb_link = __rb_link; 5201da177e4SLinus Torvalds *rb_parent = __rb_parent; 5216597d783SHugh Dickins return 0; 5221da177e4SLinus Torvalds } 5231da177e4SLinus Torvalds 524e8420a8eSCyril Hrubis static unsigned long count_vma_pages_range(struct mm_struct *mm, 525e8420a8eSCyril Hrubis unsigned long addr, unsigned long end) 526e8420a8eSCyril Hrubis { 527e8420a8eSCyril Hrubis unsigned long nr_pages = 0; 528e8420a8eSCyril Hrubis struct vm_area_struct *vma; 529e8420a8eSCyril Hrubis 530e8420a8eSCyril Hrubis /* Find first overlaping mapping */ 531e8420a8eSCyril Hrubis vma = find_vma_intersection(mm, addr, end); 532e8420a8eSCyril Hrubis if (!vma) 533e8420a8eSCyril Hrubis return 0; 534e8420a8eSCyril Hrubis 535e8420a8eSCyril Hrubis nr_pages = (min(end, vma->vm_end) - 536e8420a8eSCyril Hrubis max(addr, vma->vm_start)) >> PAGE_SHIFT; 537e8420a8eSCyril Hrubis 538e8420a8eSCyril Hrubis /* Iterate over the rest of the overlaps */ 539e8420a8eSCyril Hrubis for (vma = vma->vm_next; vma; vma = vma->vm_next) { 540e8420a8eSCyril Hrubis unsigned long overlap_len; 541e8420a8eSCyril Hrubis 542e8420a8eSCyril Hrubis if (vma->vm_start > end) 543e8420a8eSCyril Hrubis break; 544e8420a8eSCyril Hrubis 545e8420a8eSCyril Hrubis overlap_len = min(end, vma->vm_end) - vma->vm_start; 546e8420a8eSCyril Hrubis nr_pages += overlap_len >> PAGE_SHIFT; 547e8420a8eSCyril Hrubis } 548e8420a8eSCyril Hrubis 549e8420a8eSCyril Hrubis return nr_pages; 550e8420a8eSCyril Hrubis } 551e8420a8eSCyril Hrubis 5521da177e4SLinus Torvalds void __vma_link_rb(struct mm_struct *mm, struct vm_area_struct *vma, 5531da177e4SLinus Torvalds struct rb_node **rb_link, struct rb_node *rb_parent) 5541da177e4SLinus Torvalds { 555d3737187SMichel Lespinasse /* Update tracking information for the gap following the new vma. */ 556d3737187SMichel Lespinasse if (vma->vm_next) 557d3737187SMichel Lespinasse vma_gap_update(vma->vm_next); 558d3737187SMichel Lespinasse else 5591be7107fSHugh Dickins mm->highest_vm_end = vm_end_gap(vma); 560d3737187SMichel Lespinasse 561d3737187SMichel Lespinasse /* 562d3737187SMichel Lespinasse * vma->vm_prev wasn't known when we followed the rbtree to find the 563d3737187SMichel Lespinasse * correct insertion point for that vma. As a result, we could not 564d3737187SMichel Lespinasse * update the vma vm_rb parents rb_subtree_gap values on the way down. 565d3737187SMichel Lespinasse * So, we first insert the vma with a zero rb_subtree_gap value 566d3737187SMichel Lespinasse * (to be consistent with what we did on the way down), and then 567d3737187SMichel Lespinasse * immediately update the gap to the correct value. Finally we 568d3737187SMichel Lespinasse * rebalance the rbtree after all augmented values have been set. 569d3737187SMichel Lespinasse */ 5701da177e4SLinus Torvalds rb_link_node(&vma->vm_rb, rb_parent, rb_link); 571d3737187SMichel Lespinasse vma->rb_subtree_gap = 0; 572d3737187SMichel Lespinasse vma_gap_update(vma); 573d3737187SMichel Lespinasse vma_rb_insert(vma, &mm->mm_rb); 5741da177e4SLinus Torvalds } 5751da177e4SLinus Torvalds 576cb8f488cSDenys Vlasenko static void __vma_link_file(struct vm_area_struct *vma) 5771da177e4SLinus Torvalds { 5781da177e4SLinus Torvalds struct file *file; 5791da177e4SLinus Torvalds 5801da177e4SLinus Torvalds file = vma->vm_file; 5811da177e4SLinus Torvalds if (file) { 5821da177e4SLinus Torvalds struct address_space *mapping = file->f_mapping; 5831da177e4SLinus Torvalds 5841da177e4SLinus Torvalds if (vma->vm_flags & VM_DENYWRITE) 585496ad9aaSAl Viro atomic_dec(&file_inode(file)->i_writecount); 5861da177e4SLinus Torvalds if (vma->vm_flags & VM_SHARED) 5874bb5f5d9SDavid Herrmann atomic_inc(&mapping->i_mmap_writable); 5881da177e4SLinus Torvalds 5891da177e4SLinus Torvalds flush_dcache_mmap_lock(mapping); 5906b2dbba8SMichel Lespinasse vma_interval_tree_insert(vma, &mapping->i_mmap); 5911da177e4SLinus Torvalds flush_dcache_mmap_unlock(mapping); 5921da177e4SLinus Torvalds } 5931da177e4SLinus Torvalds } 5941da177e4SLinus Torvalds 5951da177e4SLinus Torvalds static void 5961da177e4SLinus Torvalds __vma_link(struct mm_struct *mm, struct vm_area_struct *vma, 5971da177e4SLinus Torvalds struct vm_area_struct *prev, struct rb_node **rb_link, 5981da177e4SLinus Torvalds struct rb_node *rb_parent) 5991da177e4SLinus Torvalds { 6001da177e4SLinus Torvalds __vma_link_list(mm, vma, prev, rb_parent); 6011da177e4SLinus Torvalds __vma_link_rb(mm, vma, rb_link, rb_parent); 6021da177e4SLinus Torvalds } 6031da177e4SLinus Torvalds 6041da177e4SLinus Torvalds static void vma_link(struct mm_struct *mm, struct vm_area_struct *vma, 6051da177e4SLinus Torvalds struct vm_area_struct *prev, struct rb_node **rb_link, 6061da177e4SLinus Torvalds struct rb_node *rb_parent) 6071da177e4SLinus Torvalds { 6081da177e4SLinus Torvalds struct address_space *mapping = NULL; 6091da177e4SLinus Torvalds 61064ac4940SHuang Shijie if (vma->vm_file) { 6111da177e4SLinus Torvalds mapping = vma->vm_file->f_mapping; 61283cde9e8SDavidlohr Bueso i_mmap_lock_write(mapping); 61364ac4940SHuang Shijie } 6141da177e4SLinus Torvalds 6151da177e4SLinus Torvalds __vma_link(mm, vma, prev, rb_link, rb_parent); 6161da177e4SLinus Torvalds __vma_link_file(vma); 6171da177e4SLinus Torvalds 6181da177e4SLinus Torvalds if (mapping) 61983cde9e8SDavidlohr Bueso i_mmap_unlock_write(mapping); 6201da177e4SLinus Torvalds 6211da177e4SLinus Torvalds mm->map_count++; 6221da177e4SLinus Torvalds validate_mm(mm); 6231da177e4SLinus Torvalds } 6241da177e4SLinus Torvalds 6251da177e4SLinus Torvalds /* 62688f6b4c3SKautuk Consul * Helper for vma_adjust() in the split_vma insert case: insert a vma into the 6276b2dbba8SMichel Lespinasse * mm's list and rbtree. It has already been inserted into the interval tree. 6281da177e4SLinus Torvalds */ 62948aae425SZhenwenXu static void __insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma) 6301da177e4SLinus Torvalds { 6316597d783SHugh Dickins struct vm_area_struct *prev; 6321da177e4SLinus Torvalds struct rb_node **rb_link, *rb_parent; 6331da177e4SLinus Torvalds 6346597d783SHugh Dickins if (find_vma_links(mm, vma->vm_start, vma->vm_end, 6356597d783SHugh Dickins &prev, &rb_link, &rb_parent)) 6366597d783SHugh Dickins BUG(); 6371da177e4SLinus Torvalds __vma_link(mm, vma, prev, rb_link, rb_parent); 6381da177e4SLinus Torvalds mm->map_count++; 6391da177e4SLinus Torvalds } 6401da177e4SLinus Torvalds 641e86f15eeSAndrea Arcangeli static __always_inline void __vma_unlink_common(struct mm_struct *mm, 642e86f15eeSAndrea Arcangeli struct vm_area_struct *vma, 643e86f15eeSAndrea Arcangeli struct vm_area_struct *prev, 6448f26e0b1SAndrea Arcangeli bool has_prev, 6458f26e0b1SAndrea Arcangeli struct vm_area_struct *ignore) 6461da177e4SLinus Torvalds { 647d3737187SMichel Lespinasse struct vm_area_struct *next; 648297c5eeeSLinus Torvalds 6498f26e0b1SAndrea Arcangeli vma_rb_erase_ignore(vma, &mm->mm_rb, ignore); 650e86f15eeSAndrea Arcangeli next = vma->vm_next; 651e86f15eeSAndrea Arcangeli if (has_prev) 652e86f15eeSAndrea Arcangeli prev->vm_next = next; 653e86f15eeSAndrea Arcangeli else { 654e86f15eeSAndrea Arcangeli prev = vma->vm_prev; 655e86f15eeSAndrea Arcangeli if (prev) 656e86f15eeSAndrea Arcangeli prev->vm_next = next; 657e86f15eeSAndrea Arcangeli else 658e86f15eeSAndrea Arcangeli mm->mmap = next; 659e86f15eeSAndrea Arcangeli } 660297c5eeeSLinus Torvalds if (next) 661297c5eeeSLinus Torvalds next->vm_prev = prev; 662615d6e87SDavidlohr Bueso 663615d6e87SDavidlohr Bueso /* Kill the cache */ 664615d6e87SDavidlohr Bueso vmacache_invalidate(mm); 6651da177e4SLinus Torvalds } 6661da177e4SLinus Torvalds 667e86f15eeSAndrea Arcangeli static inline void __vma_unlink_prev(struct mm_struct *mm, 668e86f15eeSAndrea Arcangeli struct vm_area_struct *vma, 669e86f15eeSAndrea Arcangeli struct vm_area_struct *prev) 670e86f15eeSAndrea Arcangeli { 6718f26e0b1SAndrea Arcangeli __vma_unlink_common(mm, vma, prev, true, vma); 672e86f15eeSAndrea Arcangeli } 673e86f15eeSAndrea Arcangeli 6741da177e4SLinus Torvalds /* 6751da177e4SLinus Torvalds * We cannot adjust vm_start, vm_end, vm_pgoff fields of a vma that 6761da177e4SLinus Torvalds * is already present in an i_mmap tree without adjusting the tree. 6771da177e4SLinus Torvalds * The following helper function should be used when such adjustments 6781da177e4SLinus Torvalds * are necessary. The "insert" vma (if any) is to be inserted 6791da177e4SLinus Torvalds * before we drop the necessary locks. 6801da177e4SLinus Torvalds */ 681e86f15eeSAndrea Arcangeli int __vma_adjust(struct vm_area_struct *vma, unsigned long start, 682e86f15eeSAndrea Arcangeli unsigned long end, pgoff_t pgoff, struct vm_area_struct *insert, 683e86f15eeSAndrea Arcangeli struct vm_area_struct *expand) 6841da177e4SLinus Torvalds { 6851da177e4SLinus Torvalds struct mm_struct *mm = vma->vm_mm; 686e86f15eeSAndrea Arcangeli struct vm_area_struct *next = vma->vm_next, *orig_vma = vma; 6871da177e4SLinus Torvalds struct address_space *mapping = NULL; 6886b2dbba8SMichel Lespinasse struct rb_root *root = NULL; 689012f1800SRik van Riel struct anon_vma *anon_vma = NULL; 6901da177e4SLinus Torvalds struct file *file = vma->vm_file; 691d3737187SMichel Lespinasse bool start_changed = false, end_changed = false; 6921da177e4SLinus Torvalds long adjust_next = 0; 6931da177e4SLinus Torvalds int remove_next = 0; 6941da177e4SLinus Torvalds 6951da177e4SLinus Torvalds if (next && !insert) { 696734537c9SKirill A. Shutemov struct vm_area_struct *exporter = NULL, *importer = NULL; 697287d97acSLinus Torvalds 6981da177e4SLinus Torvalds if (end >= next->vm_end) { 6991da177e4SLinus Torvalds /* 7001da177e4SLinus Torvalds * vma expands, overlapping all the next, and 7011da177e4SLinus Torvalds * perhaps the one after too (mprotect case 6). 70286d12e47SAndrea Arcangeli * The only other cases that gets here are 703e86f15eeSAndrea Arcangeli * case 1, case 7 and case 8. 704e86f15eeSAndrea Arcangeli */ 705e86f15eeSAndrea Arcangeli if (next == expand) { 706e86f15eeSAndrea Arcangeli /* 707e86f15eeSAndrea Arcangeli * The only case where we don't expand "vma" 708e86f15eeSAndrea Arcangeli * and we expand "next" instead is case 8. 709e86f15eeSAndrea Arcangeli */ 710e86f15eeSAndrea Arcangeli VM_WARN_ON(end != next->vm_end); 711e86f15eeSAndrea Arcangeli /* 712e86f15eeSAndrea Arcangeli * remove_next == 3 means we're 713e86f15eeSAndrea Arcangeli * removing "vma" and that to do so we 714e86f15eeSAndrea Arcangeli * swapped "vma" and "next". 715e86f15eeSAndrea Arcangeli */ 716e86f15eeSAndrea Arcangeli remove_next = 3; 717e86f15eeSAndrea Arcangeli VM_WARN_ON(file != next->vm_file); 718e86f15eeSAndrea Arcangeli swap(vma, next); 719e86f15eeSAndrea Arcangeli } else { 720e86f15eeSAndrea Arcangeli VM_WARN_ON(expand != vma); 721e86f15eeSAndrea Arcangeli /* 722e86f15eeSAndrea Arcangeli * case 1, 6, 7, remove_next == 2 is case 6, 723e86f15eeSAndrea Arcangeli * remove_next == 1 is case 1 or 7. 7241da177e4SLinus Torvalds */ 725734537c9SKirill A. Shutemov remove_next = 1 + (end > next->vm_end); 726e86f15eeSAndrea Arcangeli VM_WARN_ON(remove_next == 2 && 727e86f15eeSAndrea Arcangeli end != next->vm_next->vm_end); 728e86f15eeSAndrea Arcangeli VM_WARN_ON(remove_next == 1 && 729e86f15eeSAndrea Arcangeli end != next->vm_end); 730e86f15eeSAndrea Arcangeli /* trim end to next, for case 6 first pass */ 7311da177e4SLinus Torvalds end = next->vm_end; 732e86f15eeSAndrea Arcangeli } 733e86f15eeSAndrea Arcangeli 734287d97acSLinus Torvalds exporter = next; 7351da177e4SLinus Torvalds importer = vma; 736734537c9SKirill A. Shutemov 737734537c9SKirill A. Shutemov /* 738734537c9SKirill A. Shutemov * If next doesn't have anon_vma, import from vma after 739734537c9SKirill A. Shutemov * next, if the vma overlaps with it. 740734537c9SKirill A. Shutemov */ 74197a42cd4SAndrea Arcangeli if (remove_next == 2 && !next->anon_vma) 742734537c9SKirill A. Shutemov exporter = next->vm_next; 743734537c9SKirill A. Shutemov 7441da177e4SLinus Torvalds } else if (end > next->vm_start) { 7451da177e4SLinus Torvalds /* 7461da177e4SLinus Torvalds * vma expands, overlapping part of the next: 7471da177e4SLinus Torvalds * mprotect case 5 shifting the boundary up. 7481da177e4SLinus Torvalds */ 7491da177e4SLinus Torvalds adjust_next = (end - next->vm_start) >> PAGE_SHIFT; 750287d97acSLinus Torvalds exporter = next; 7511da177e4SLinus Torvalds importer = vma; 752e86f15eeSAndrea Arcangeli VM_WARN_ON(expand != importer); 7531da177e4SLinus Torvalds } else if (end < vma->vm_end) { 7541da177e4SLinus Torvalds /* 7551da177e4SLinus Torvalds * vma shrinks, and !insert tells it's not 7561da177e4SLinus Torvalds * split_vma inserting another: so it must be 7571da177e4SLinus Torvalds * mprotect case 4 shifting the boundary down. 7581da177e4SLinus Torvalds */ 7591da177e4SLinus Torvalds adjust_next = -((vma->vm_end - end) >> PAGE_SHIFT); 760287d97acSLinus Torvalds exporter = vma; 7611da177e4SLinus Torvalds importer = next; 762e86f15eeSAndrea Arcangeli VM_WARN_ON(expand != importer); 7631da177e4SLinus Torvalds } 7641da177e4SLinus Torvalds 7655beb4930SRik van Riel /* 7665beb4930SRik van Riel * Easily overlooked: when mprotect shifts the boundary, 7675beb4930SRik van Riel * make sure the expanding vma has anon_vma set if the 7685beb4930SRik van Riel * shrinking vma had, to cover any anon pages imported. 7695beb4930SRik van Riel */ 770287d97acSLinus Torvalds if (exporter && exporter->anon_vma && !importer->anon_vma) { 771c4ea95d7SDaniel Forrest int error; 772c4ea95d7SDaniel Forrest 773287d97acSLinus Torvalds importer->anon_vma = exporter->anon_vma; 774b800c91aSKonstantin Khlebnikov error = anon_vma_clone(importer, exporter); 7753fe89b3eSLeon Yu if (error) 776b800c91aSKonstantin Khlebnikov return error; 777b800c91aSKonstantin Khlebnikov } 7785beb4930SRik van Riel } 779734537c9SKirill A. Shutemov again: 780e86f15eeSAndrea Arcangeli vma_adjust_trans_huge(orig_vma, start, end, adjust_next); 78137f9f559SKirill A. Shutemov 7821da177e4SLinus Torvalds if (file) { 7831da177e4SLinus Torvalds mapping = file->f_mapping; 7841da177e4SLinus Torvalds root = &mapping->i_mmap; 785cbc91f71SSrikar Dronamraju uprobe_munmap(vma, vma->vm_start, vma->vm_end); 786682968e0SSrikar Dronamraju 787682968e0SSrikar Dronamraju if (adjust_next) 78827ba0644SKirill A. Shutemov uprobe_munmap(next, next->vm_start, next->vm_end); 789682968e0SSrikar Dronamraju 79083cde9e8SDavidlohr Bueso i_mmap_lock_write(mapping); 7911da177e4SLinus Torvalds if (insert) { 7921da177e4SLinus Torvalds /* 7936b2dbba8SMichel Lespinasse * Put into interval tree now, so instantiated pages 7941da177e4SLinus Torvalds * are visible to arm/parisc __flush_dcache_page 7951da177e4SLinus Torvalds * throughout; but we cannot insert into address 7961da177e4SLinus Torvalds * space until vma start or end is updated. 7971da177e4SLinus Torvalds */ 7981da177e4SLinus Torvalds __vma_link_file(insert); 7991da177e4SLinus Torvalds } 8001da177e4SLinus Torvalds } 8011da177e4SLinus Torvalds 802012f1800SRik van Riel anon_vma = vma->anon_vma; 803bf181b9fSMichel Lespinasse if (!anon_vma && adjust_next) 804bf181b9fSMichel Lespinasse anon_vma = next->anon_vma; 805bf181b9fSMichel Lespinasse if (anon_vma) { 806e86f15eeSAndrea Arcangeli VM_WARN_ON(adjust_next && next->anon_vma && 807e86f15eeSAndrea Arcangeli anon_vma != next->anon_vma); 8084fc3f1d6SIngo Molnar anon_vma_lock_write(anon_vma); 809bf181b9fSMichel Lespinasse anon_vma_interval_tree_pre_update_vma(vma); 810bf181b9fSMichel Lespinasse if (adjust_next) 811bf181b9fSMichel Lespinasse anon_vma_interval_tree_pre_update_vma(next); 812bf181b9fSMichel Lespinasse } 813012f1800SRik van Riel 8141da177e4SLinus Torvalds if (root) { 8151da177e4SLinus Torvalds flush_dcache_mmap_lock(mapping); 8166b2dbba8SMichel Lespinasse vma_interval_tree_remove(vma, root); 8171da177e4SLinus Torvalds if (adjust_next) 8186b2dbba8SMichel Lespinasse vma_interval_tree_remove(next, root); 8191da177e4SLinus Torvalds } 8201da177e4SLinus Torvalds 821d3737187SMichel Lespinasse if (start != vma->vm_start) { 8221da177e4SLinus Torvalds vma->vm_start = start; 823d3737187SMichel Lespinasse start_changed = true; 824d3737187SMichel Lespinasse } 825d3737187SMichel Lespinasse if (end != vma->vm_end) { 8261da177e4SLinus Torvalds vma->vm_end = end; 827d3737187SMichel Lespinasse end_changed = true; 828d3737187SMichel Lespinasse } 8291da177e4SLinus Torvalds vma->vm_pgoff = pgoff; 8301da177e4SLinus Torvalds if (adjust_next) { 8311da177e4SLinus Torvalds next->vm_start += adjust_next << PAGE_SHIFT; 8321da177e4SLinus Torvalds next->vm_pgoff += adjust_next; 8331da177e4SLinus Torvalds } 8341da177e4SLinus Torvalds 8351da177e4SLinus Torvalds if (root) { 8361da177e4SLinus Torvalds if (adjust_next) 8376b2dbba8SMichel Lespinasse vma_interval_tree_insert(next, root); 8386b2dbba8SMichel Lespinasse vma_interval_tree_insert(vma, root); 8391da177e4SLinus Torvalds flush_dcache_mmap_unlock(mapping); 8401da177e4SLinus Torvalds } 8411da177e4SLinus Torvalds 8421da177e4SLinus Torvalds if (remove_next) { 8431da177e4SLinus Torvalds /* 8441da177e4SLinus Torvalds * vma_merge has merged next into vma, and needs 8451da177e4SLinus Torvalds * us to remove next before dropping the locks. 8461da177e4SLinus Torvalds */ 847e86f15eeSAndrea Arcangeli if (remove_next != 3) 848e86f15eeSAndrea Arcangeli __vma_unlink_prev(mm, next, vma); 849e86f15eeSAndrea Arcangeli else 8508f26e0b1SAndrea Arcangeli /* 8518f26e0b1SAndrea Arcangeli * vma is not before next if they've been 8528f26e0b1SAndrea Arcangeli * swapped. 8538f26e0b1SAndrea Arcangeli * 8548f26e0b1SAndrea Arcangeli * pre-swap() next->vm_start was reduced so 8558f26e0b1SAndrea Arcangeli * tell validate_mm_rb to ignore pre-swap() 8568f26e0b1SAndrea Arcangeli * "next" (which is stored in post-swap() 8578f26e0b1SAndrea Arcangeli * "vma"). 8588f26e0b1SAndrea Arcangeli */ 8598f26e0b1SAndrea Arcangeli __vma_unlink_common(mm, next, NULL, false, vma); 8601da177e4SLinus Torvalds if (file) 8611da177e4SLinus Torvalds __remove_shared_vm_struct(next, file, mapping); 8621da177e4SLinus Torvalds } else if (insert) { 8631da177e4SLinus Torvalds /* 8641da177e4SLinus Torvalds * split_vma has split insert from vma, and needs 8651da177e4SLinus Torvalds * us to insert it before dropping the locks 8661da177e4SLinus Torvalds * (it may either follow vma or precede it). 8671da177e4SLinus Torvalds */ 8681da177e4SLinus Torvalds __insert_vm_struct(mm, insert); 869d3737187SMichel Lespinasse } else { 870d3737187SMichel Lespinasse if (start_changed) 871d3737187SMichel Lespinasse vma_gap_update(vma); 872d3737187SMichel Lespinasse if (end_changed) { 873d3737187SMichel Lespinasse if (!next) 8741be7107fSHugh Dickins mm->highest_vm_end = vm_end_gap(vma); 875d3737187SMichel Lespinasse else if (!adjust_next) 876d3737187SMichel Lespinasse vma_gap_update(next); 877d3737187SMichel Lespinasse } 8781da177e4SLinus Torvalds } 8791da177e4SLinus Torvalds 880bf181b9fSMichel Lespinasse if (anon_vma) { 881bf181b9fSMichel Lespinasse anon_vma_interval_tree_post_update_vma(vma); 882bf181b9fSMichel Lespinasse if (adjust_next) 883bf181b9fSMichel Lespinasse anon_vma_interval_tree_post_update_vma(next); 88408b52706SKonstantin Khlebnikov anon_vma_unlock_write(anon_vma); 885bf181b9fSMichel Lespinasse } 8861da177e4SLinus Torvalds if (mapping) 88783cde9e8SDavidlohr Bueso i_mmap_unlock_write(mapping); 8881da177e4SLinus Torvalds 8892b144498SSrikar Dronamraju if (root) { 8907b2d81d4SIngo Molnar uprobe_mmap(vma); 8912b144498SSrikar Dronamraju 8922b144498SSrikar Dronamraju if (adjust_next) 8937b2d81d4SIngo Molnar uprobe_mmap(next); 8942b144498SSrikar Dronamraju } 8952b144498SSrikar Dronamraju 8961da177e4SLinus Torvalds if (remove_next) { 897925d1c40SMatt Helsley if (file) { 898cbc91f71SSrikar Dronamraju uprobe_munmap(next, next->vm_start, next->vm_end); 8991da177e4SLinus Torvalds fput(file); 900925d1c40SMatt Helsley } 9015beb4930SRik van Riel if (next->anon_vma) 9025beb4930SRik van Riel anon_vma_merge(vma, next); 9031da177e4SLinus Torvalds mm->map_count--; 9043964acd0SOleg Nesterov mpol_put(vma_policy(next)); 9051da177e4SLinus Torvalds kmem_cache_free(vm_area_cachep, next); 9061da177e4SLinus Torvalds /* 9071da177e4SLinus Torvalds * In mprotect's case 6 (see comments on vma_merge), 9081da177e4SLinus Torvalds * we must remove another next too. It would clutter 9091da177e4SLinus Torvalds * up the code too much to do both in one go. 9101da177e4SLinus Torvalds */ 911e86f15eeSAndrea Arcangeli if (remove_next != 3) { 912e86f15eeSAndrea Arcangeli /* 913e86f15eeSAndrea Arcangeli * If "next" was removed and vma->vm_end was 914e86f15eeSAndrea Arcangeli * expanded (up) over it, in turn 915e86f15eeSAndrea Arcangeli * "next->vm_prev->vm_end" changed and the 916e86f15eeSAndrea Arcangeli * "vma->vm_next" gap must be updated. 917e86f15eeSAndrea Arcangeli */ 9181da177e4SLinus Torvalds next = vma->vm_next; 919e86f15eeSAndrea Arcangeli } else { 920e86f15eeSAndrea Arcangeli /* 921e86f15eeSAndrea Arcangeli * For the scope of the comment "next" and 922e86f15eeSAndrea Arcangeli * "vma" considered pre-swap(): if "vma" was 923e86f15eeSAndrea Arcangeli * removed, next->vm_start was expanded (down) 924e86f15eeSAndrea Arcangeli * over it and the "next" gap must be updated. 925e86f15eeSAndrea Arcangeli * Because of the swap() the post-swap() "vma" 926e86f15eeSAndrea Arcangeli * actually points to pre-swap() "next" 927e86f15eeSAndrea Arcangeli * (post-swap() "next" as opposed is now a 928e86f15eeSAndrea Arcangeli * dangling pointer). 929e86f15eeSAndrea Arcangeli */ 930e86f15eeSAndrea Arcangeli next = vma; 931e86f15eeSAndrea Arcangeli } 932734537c9SKirill A. Shutemov if (remove_next == 2) { 933734537c9SKirill A. Shutemov remove_next = 1; 934734537c9SKirill A. Shutemov end = next->vm_end; 9351da177e4SLinus Torvalds goto again; 936734537c9SKirill A. Shutemov } 937d3737187SMichel Lespinasse else if (next) 938d3737187SMichel Lespinasse vma_gap_update(next); 939fb8c41e9SAndrea Arcangeli else { 940fb8c41e9SAndrea Arcangeli /* 941fb8c41e9SAndrea Arcangeli * If remove_next == 2 we obviously can't 942fb8c41e9SAndrea Arcangeli * reach this path. 943fb8c41e9SAndrea Arcangeli * 944fb8c41e9SAndrea Arcangeli * If remove_next == 3 we can't reach this 945fb8c41e9SAndrea Arcangeli * path because pre-swap() next is always not 946fb8c41e9SAndrea Arcangeli * NULL. pre-swap() "next" is not being 947fb8c41e9SAndrea Arcangeli * removed and its next->vm_end is not altered 948fb8c41e9SAndrea Arcangeli * (and furthermore "end" already matches 949fb8c41e9SAndrea Arcangeli * next->vm_end in remove_next == 3). 950fb8c41e9SAndrea Arcangeli * 951fb8c41e9SAndrea Arcangeli * We reach this only in the remove_next == 1 952fb8c41e9SAndrea Arcangeli * case if the "next" vma that was removed was 953fb8c41e9SAndrea Arcangeli * the highest vma of the mm. However in such 954fb8c41e9SAndrea Arcangeli * case next->vm_end == "end" and the extended 955fb8c41e9SAndrea Arcangeli * "vma" has vma->vm_end == next->vm_end so 956fb8c41e9SAndrea Arcangeli * mm->highest_vm_end doesn't need any update 957fb8c41e9SAndrea Arcangeli * in remove_next == 1 case. 958fb8c41e9SAndrea Arcangeli */ 9591be7107fSHugh Dickins VM_WARN_ON(mm->highest_vm_end != vm_end_gap(vma)); 960fb8c41e9SAndrea Arcangeli } 9611da177e4SLinus Torvalds } 9622b144498SSrikar Dronamraju if (insert && file) 9637b2d81d4SIngo Molnar uprobe_mmap(insert); 9641da177e4SLinus Torvalds 9651da177e4SLinus Torvalds validate_mm(mm); 9665beb4930SRik van Riel 9675beb4930SRik van Riel return 0; 9681da177e4SLinus Torvalds } 9691da177e4SLinus Torvalds 9701da177e4SLinus Torvalds /* 9711da177e4SLinus Torvalds * If the vma has a ->close operation then the driver probably needs to release 9721da177e4SLinus Torvalds * per-vma resources, so we don't attempt to merge those. 9731da177e4SLinus Torvalds */ 9741da177e4SLinus Torvalds static inline int is_mergeable_vma(struct vm_area_struct *vma, 97519a809afSAndrea Arcangeli struct file *file, unsigned long vm_flags, 97619a809afSAndrea Arcangeli struct vm_userfaultfd_ctx vm_userfaultfd_ctx) 9771da177e4SLinus Torvalds { 97834228d47SCyrill Gorcunov /* 97934228d47SCyrill Gorcunov * VM_SOFTDIRTY should not prevent from VMA merging, if we 98034228d47SCyrill Gorcunov * match the flags but dirty bit -- the caller should mark 98134228d47SCyrill Gorcunov * merged VMA as dirty. If dirty bit won't be excluded from 98234228d47SCyrill Gorcunov * comparison, we increase pressue on the memory system forcing 98334228d47SCyrill Gorcunov * the kernel to generate new VMAs when old one could be 98434228d47SCyrill Gorcunov * extended instead. 98534228d47SCyrill Gorcunov */ 98634228d47SCyrill Gorcunov if ((vma->vm_flags ^ vm_flags) & ~VM_SOFTDIRTY) 9871da177e4SLinus Torvalds return 0; 9881da177e4SLinus Torvalds if (vma->vm_file != file) 9891da177e4SLinus Torvalds return 0; 9901da177e4SLinus Torvalds if (vma->vm_ops && vma->vm_ops->close) 9911da177e4SLinus Torvalds return 0; 99219a809afSAndrea Arcangeli if (!is_mergeable_vm_userfaultfd_ctx(vma, vm_userfaultfd_ctx)) 99319a809afSAndrea Arcangeli return 0; 9941da177e4SLinus Torvalds return 1; 9951da177e4SLinus Torvalds } 9961da177e4SLinus Torvalds 9971da177e4SLinus Torvalds static inline int is_mergeable_anon_vma(struct anon_vma *anon_vma1, 998965f55deSShaohua Li struct anon_vma *anon_vma2, 999965f55deSShaohua Li struct vm_area_struct *vma) 10001da177e4SLinus Torvalds { 1001965f55deSShaohua Li /* 1002965f55deSShaohua Li * The list_is_singular() test is to avoid merging VMA cloned from 1003965f55deSShaohua Li * parents. This can improve scalability caused by anon_vma lock. 1004965f55deSShaohua Li */ 1005965f55deSShaohua Li if ((!anon_vma1 || !anon_vma2) && (!vma || 1006965f55deSShaohua Li list_is_singular(&vma->anon_vma_chain))) 1007965f55deSShaohua Li return 1; 1008965f55deSShaohua Li return anon_vma1 == anon_vma2; 10091da177e4SLinus Torvalds } 10101da177e4SLinus Torvalds 10111da177e4SLinus Torvalds /* 10121da177e4SLinus Torvalds * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff) 10131da177e4SLinus Torvalds * in front of (at a lower virtual address and file offset than) the vma. 10141da177e4SLinus Torvalds * 10151da177e4SLinus Torvalds * We cannot merge two vmas if they have differently assigned (non-NULL) 10161da177e4SLinus Torvalds * anon_vmas, nor if same anon_vma is assigned but offsets incompatible. 10171da177e4SLinus Torvalds * 10181da177e4SLinus Torvalds * We don't check here for the merged mmap wrapping around the end of pagecache 10191da177e4SLinus Torvalds * indices (16TB on ia32) because do_mmap_pgoff() does not permit mmap's which 10201da177e4SLinus Torvalds * wrap, nor mmaps which cover the final page at index -1UL. 10211da177e4SLinus Torvalds */ 10221da177e4SLinus Torvalds static int 10231da177e4SLinus Torvalds can_vma_merge_before(struct vm_area_struct *vma, unsigned long vm_flags, 102419a809afSAndrea Arcangeli struct anon_vma *anon_vma, struct file *file, 102519a809afSAndrea Arcangeli pgoff_t vm_pgoff, 102619a809afSAndrea Arcangeli struct vm_userfaultfd_ctx vm_userfaultfd_ctx) 10271da177e4SLinus Torvalds { 102819a809afSAndrea Arcangeli if (is_mergeable_vma(vma, file, vm_flags, vm_userfaultfd_ctx) && 1029965f55deSShaohua Li is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) { 10301da177e4SLinus Torvalds if (vma->vm_pgoff == vm_pgoff) 10311da177e4SLinus Torvalds return 1; 10321da177e4SLinus Torvalds } 10331da177e4SLinus Torvalds return 0; 10341da177e4SLinus Torvalds } 10351da177e4SLinus Torvalds 10361da177e4SLinus Torvalds /* 10371da177e4SLinus Torvalds * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff) 10381da177e4SLinus Torvalds * beyond (at a higher virtual address and file offset than) the vma. 10391da177e4SLinus Torvalds * 10401da177e4SLinus Torvalds * We cannot merge two vmas if they have differently assigned (non-NULL) 10411da177e4SLinus Torvalds * anon_vmas, nor if same anon_vma is assigned but offsets incompatible. 10421da177e4SLinus Torvalds */ 10431da177e4SLinus Torvalds static int 10441da177e4SLinus Torvalds can_vma_merge_after(struct vm_area_struct *vma, unsigned long vm_flags, 104519a809afSAndrea Arcangeli struct anon_vma *anon_vma, struct file *file, 104619a809afSAndrea Arcangeli pgoff_t vm_pgoff, 104719a809afSAndrea Arcangeli struct vm_userfaultfd_ctx vm_userfaultfd_ctx) 10481da177e4SLinus Torvalds { 104919a809afSAndrea Arcangeli if (is_mergeable_vma(vma, file, vm_flags, vm_userfaultfd_ctx) && 1050965f55deSShaohua Li is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) { 10511da177e4SLinus Torvalds pgoff_t vm_pglen; 1052d6e93217SLibin vm_pglen = vma_pages(vma); 10531da177e4SLinus Torvalds if (vma->vm_pgoff + vm_pglen == vm_pgoff) 10541da177e4SLinus Torvalds return 1; 10551da177e4SLinus Torvalds } 10561da177e4SLinus Torvalds return 0; 10571da177e4SLinus Torvalds } 10581da177e4SLinus Torvalds 10591da177e4SLinus Torvalds /* 10601da177e4SLinus Torvalds * Given a mapping request (addr,end,vm_flags,file,pgoff), figure out 10611da177e4SLinus Torvalds * whether that can be merged with its predecessor or its successor. 10621da177e4SLinus Torvalds * Or both (it neatly fills a hole). 10631da177e4SLinus Torvalds * 10641da177e4SLinus Torvalds * In most cases - when called for mmap, brk or mremap - [addr,end) is 10651da177e4SLinus Torvalds * certain not to be mapped by the time vma_merge is called; but when 10661da177e4SLinus Torvalds * called for mprotect, it is certain to be already mapped (either at 10671da177e4SLinus Torvalds * an offset within prev, or at the start of next), and the flags of 10681da177e4SLinus Torvalds * this area are about to be changed to vm_flags - and the no-change 10691da177e4SLinus Torvalds * case has already been eliminated. 10701da177e4SLinus Torvalds * 10711da177e4SLinus Torvalds * The following mprotect cases have to be considered, where AAAA is 10721da177e4SLinus Torvalds * the area passed down from mprotect_fixup, never extending beyond one 10731da177e4SLinus Torvalds * vma, PPPPPP is the prev vma specified, and NNNNNN the next vma after: 10741da177e4SLinus Torvalds * 10751da177e4SLinus Torvalds * AAAA AAAA AAAA AAAA 10761da177e4SLinus Torvalds * PPPPPPNNNNNN PPPPPPNNNNNN PPPPPPNNNNNN PPPPNNNNXXXX 10771da177e4SLinus Torvalds * cannot merge might become might become might become 10781da177e4SLinus Torvalds * PPNNNNNNNNNN PPPPPPPPPPNN PPPPPPPPPPPP 6 or 10791da177e4SLinus Torvalds * mmap, brk or case 4 below case 5 below PPPPPPPPXXXX 7 or 1080e86f15eeSAndrea Arcangeli * mremap move: PPPPXXXXXXXX 8 10811da177e4SLinus Torvalds * AAAA 10821da177e4SLinus Torvalds * PPPP NNNN PPPPPPPPPPPP PPPPPPPPNNNN PPPPNNNNNNNN 10831da177e4SLinus Torvalds * might become case 1 below case 2 below case 3 below 10841da177e4SLinus Torvalds * 1085e86f15eeSAndrea Arcangeli * It is important for case 8 that the the vma NNNN overlapping the 1086e86f15eeSAndrea Arcangeli * region AAAA is never going to extended over XXXX. Instead XXXX must 1087e86f15eeSAndrea Arcangeli * be extended in region AAAA and NNNN must be removed. This way in 1088e86f15eeSAndrea Arcangeli * all cases where vma_merge succeeds, the moment vma_adjust drops the 1089e86f15eeSAndrea Arcangeli * rmap_locks, the properties of the merged vma will be already 1090e86f15eeSAndrea Arcangeli * correct for the whole merged range. Some of those properties like 1091e86f15eeSAndrea Arcangeli * vm_page_prot/vm_flags may be accessed by rmap_walks and they must 1092e86f15eeSAndrea Arcangeli * be correct for the whole merged range immediately after the 1093e86f15eeSAndrea Arcangeli * rmap_locks are released. Otherwise if XXXX would be removed and 1094e86f15eeSAndrea Arcangeli * NNNN would be extended over the XXXX range, remove_migration_ptes 1095e86f15eeSAndrea Arcangeli * or other rmap walkers (if working on addresses beyond the "end" 1096e86f15eeSAndrea Arcangeli * parameter) may establish ptes with the wrong permissions of NNNN 1097e86f15eeSAndrea Arcangeli * instead of the right permissions of XXXX. 10981da177e4SLinus Torvalds */ 10991da177e4SLinus Torvalds struct vm_area_struct *vma_merge(struct mm_struct *mm, 11001da177e4SLinus Torvalds struct vm_area_struct *prev, unsigned long addr, 11011da177e4SLinus Torvalds unsigned long end, unsigned long vm_flags, 11021da177e4SLinus Torvalds struct anon_vma *anon_vma, struct file *file, 110319a809afSAndrea Arcangeli pgoff_t pgoff, struct mempolicy *policy, 110419a809afSAndrea Arcangeli struct vm_userfaultfd_ctx vm_userfaultfd_ctx) 11051da177e4SLinus Torvalds { 11061da177e4SLinus Torvalds pgoff_t pglen = (end - addr) >> PAGE_SHIFT; 11071da177e4SLinus Torvalds struct vm_area_struct *area, *next; 11085beb4930SRik van Riel int err; 11091da177e4SLinus Torvalds 11101da177e4SLinus Torvalds /* 11111da177e4SLinus Torvalds * We later require that vma->vm_flags == vm_flags, 11121da177e4SLinus Torvalds * so this tests vma->vm_flags & VM_SPECIAL, too. 11131da177e4SLinus Torvalds */ 11141da177e4SLinus Torvalds if (vm_flags & VM_SPECIAL) 11151da177e4SLinus Torvalds return NULL; 11161da177e4SLinus Torvalds 11171da177e4SLinus Torvalds if (prev) 11181da177e4SLinus Torvalds next = prev->vm_next; 11191da177e4SLinus Torvalds else 11201da177e4SLinus Torvalds next = mm->mmap; 11211da177e4SLinus Torvalds area = next; 1122e86f15eeSAndrea Arcangeli if (area && area->vm_end == end) /* cases 6, 7, 8 */ 11231da177e4SLinus Torvalds next = next->vm_next; 11241da177e4SLinus Torvalds 1125e86f15eeSAndrea Arcangeli /* verify some invariant that must be enforced by the caller */ 1126e86f15eeSAndrea Arcangeli VM_WARN_ON(prev && addr <= prev->vm_start); 1127e86f15eeSAndrea Arcangeli VM_WARN_ON(area && end > area->vm_end); 1128e86f15eeSAndrea Arcangeli VM_WARN_ON(addr >= end); 1129e86f15eeSAndrea Arcangeli 11301da177e4SLinus Torvalds /* 11311da177e4SLinus Torvalds * Can it merge with the predecessor? 11321da177e4SLinus Torvalds */ 11331da177e4SLinus Torvalds if (prev && prev->vm_end == addr && 11341da177e4SLinus Torvalds mpol_equal(vma_policy(prev), policy) && 11351da177e4SLinus Torvalds can_vma_merge_after(prev, vm_flags, 113619a809afSAndrea Arcangeli anon_vma, file, pgoff, 113719a809afSAndrea Arcangeli vm_userfaultfd_ctx)) { 11381da177e4SLinus Torvalds /* 11391da177e4SLinus Torvalds * OK, it can. Can we now merge in the successor as well? 11401da177e4SLinus Torvalds */ 11411da177e4SLinus Torvalds if (next && end == next->vm_start && 11421da177e4SLinus Torvalds mpol_equal(policy, vma_policy(next)) && 11431da177e4SLinus Torvalds can_vma_merge_before(next, vm_flags, 114419a809afSAndrea Arcangeli anon_vma, file, 114519a809afSAndrea Arcangeli pgoff+pglen, 114619a809afSAndrea Arcangeli vm_userfaultfd_ctx) && 11471da177e4SLinus Torvalds is_mergeable_anon_vma(prev->anon_vma, 1148965f55deSShaohua Li next->anon_vma, NULL)) { 11491da177e4SLinus Torvalds /* cases 1, 6 */ 1150e86f15eeSAndrea Arcangeli err = __vma_adjust(prev, prev->vm_start, 1151e86f15eeSAndrea Arcangeli next->vm_end, prev->vm_pgoff, NULL, 1152e86f15eeSAndrea Arcangeli prev); 11531da177e4SLinus Torvalds } else /* cases 2, 5, 7 */ 1154e86f15eeSAndrea Arcangeli err = __vma_adjust(prev, prev->vm_start, 1155e86f15eeSAndrea Arcangeli end, prev->vm_pgoff, NULL, prev); 11565beb4930SRik van Riel if (err) 11575beb4930SRik van Riel return NULL; 11586d50e60cSDavid Rientjes khugepaged_enter_vma_merge(prev, vm_flags); 11591da177e4SLinus Torvalds return prev; 11601da177e4SLinus Torvalds } 11611da177e4SLinus Torvalds 11621da177e4SLinus Torvalds /* 11631da177e4SLinus Torvalds * Can this new request be merged in front of next? 11641da177e4SLinus Torvalds */ 11651da177e4SLinus Torvalds if (next && end == next->vm_start && 11661da177e4SLinus Torvalds mpol_equal(policy, vma_policy(next)) && 11671da177e4SLinus Torvalds can_vma_merge_before(next, vm_flags, 116819a809afSAndrea Arcangeli anon_vma, file, pgoff+pglen, 116919a809afSAndrea Arcangeli vm_userfaultfd_ctx)) { 11701da177e4SLinus Torvalds if (prev && addr < prev->vm_end) /* case 4 */ 1171e86f15eeSAndrea Arcangeli err = __vma_adjust(prev, prev->vm_start, 1172e86f15eeSAndrea Arcangeli addr, prev->vm_pgoff, NULL, next); 1173e86f15eeSAndrea Arcangeli else { /* cases 3, 8 */ 1174e86f15eeSAndrea Arcangeli err = __vma_adjust(area, addr, next->vm_end, 1175e86f15eeSAndrea Arcangeli next->vm_pgoff - pglen, NULL, next); 1176e86f15eeSAndrea Arcangeli /* 1177e86f15eeSAndrea Arcangeli * In case 3 area is already equal to next and 1178e86f15eeSAndrea Arcangeli * this is a noop, but in case 8 "area" has 1179e86f15eeSAndrea Arcangeli * been removed and next was expanded over it. 1180e86f15eeSAndrea Arcangeli */ 1181e86f15eeSAndrea Arcangeli area = next; 1182e86f15eeSAndrea Arcangeli } 11835beb4930SRik van Riel if (err) 11845beb4930SRik van Riel return NULL; 11856d50e60cSDavid Rientjes khugepaged_enter_vma_merge(area, vm_flags); 11861da177e4SLinus Torvalds return area; 11871da177e4SLinus Torvalds } 11881da177e4SLinus Torvalds 11891da177e4SLinus Torvalds return NULL; 11901da177e4SLinus Torvalds } 11911da177e4SLinus Torvalds 11921da177e4SLinus Torvalds /* 1193d0e9fe17SLinus Torvalds * Rough compatbility check to quickly see if it's even worth looking 1194d0e9fe17SLinus Torvalds * at sharing an anon_vma. 1195d0e9fe17SLinus Torvalds * 1196d0e9fe17SLinus Torvalds * They need to have the same vm_file, and the flags can only differ 1197d0e9fe17SLinus Torvalds * in things that mprotect may change. 1198d0e9fe17SLinus Torvalds * 1199d0e9fe17SLinus Torvalds * NOTE! The fact that we share an anon_vma doesn't _have_ to mean that 1200d0e9fe17SLinus Torvalds * we can merge the two vma's. For example, we refuse to merge a vma if 1201d0e9fe17SLinus Torvalds * there is a vm_ops->close() function, because that indicates that the 1202d0e9fe17SLinus Torvalds * driver is doing some kind of reference counting. But that doesn't 1203d0e9fe17SLinus Torvalds * really matter for the anon_vma sharing case. 1204d0e9fe17SLinus Torvalds */ 1205d0e9fe17SLinus Torvalds static int anon_vma_compatible(struct vm_area_struct *a, struct vm_area_struct *b) 1206d0e9fe17SLinus Torvalds { 1207d0e9fe17SLinus Torvalds return a->vm_end == b->vm_start && 1208d0e9fe17SLinus Torvalds mpol_equal(vma_policy(a), vma_policy(b)) && 1209d0e9fe17SLinus Torvalds a->vm_file == b->vm_file && 121034228d47SCyrill Gorcunov !((a->vm_flags ^ b->vm_flags) & ~(VM_READ|VM_WRITE|VM_EXEC|VM_SOFTDIRTY)) && 1211d0e9fe17SLinus Torvalds b->vm_pgoff == a->vm_pgoff + ((b->vm_start - a->vm_start) >> PAGE_SHIFT); 1212d0e9fe17SLinus Torvalds } 1213d0e9fe17SLinus Torvalds 1214d0e9fe17SLinus Torvalds /* 1215d0e9fe17SLinus Torvalds * Do some basic sanity checking to see if we can re-use the anon_vma 1216d0e9fe17SLinus Torvalds * from 'old'. The 'a'/'b' vma's are in VM order - one of them will be 1217d0e9fe17SLinus Torvalds * the same as 'old', the other will be the new one that is trying 1218d0e9fe17SLinus Torvalds * to share the anon_vma. 1219d0e9fe17SLinus Torvalds * 1220d0e9fe17SLinus Torvalds * NOTE! This runs with mm_sem held for reading, so it is possible that 1221d0e9fe17SLinus Torvalds * the anon_vma of 'old' is concurrently in the process of being set up 1222d0e9fe17SLinus Torvalds * by another page fault trying to merge _that_. But that's ok: if it 1223d0e9fe17SLinus Torvalds * is being set up, that automatically means that it will be a singleton 1224d0e9fe17SLinus Torvalds * acceptable for merging, so we can do all of this optimistically. But 12254db0c3c2SJason Low * we do that READ_ONCE() to make sure that we never re-load the pointer. 1226d0e9fe17SLinus Torvalds * 1227d0e9fe17SLinus Torvalds * IOW: that the "list_is_singular()" test on the anon_vma_chain only 1228d0e9fe17SLinus Torvalds * matters for the 'stable anon_vma' case (ie the thing we want to avoid 1229d0e9fe17SLinus Torvalds * is to return an anon_vma that is "complex" due to having gone through 1230d0e9fe17SLinus Torvalds * a fork). 1231d0e9fe17SLinus Torvalds * 1232d0e9fe17SLinus Torvalds * We also make sure that the two vma's are compatible (adjacent, 1233d0e9fe17SLinus Torvalds * and with the same memory policies). That's all stable, even with just 1234d0e9fe17SLinus Torvalds * a read lock on the mm_sem. 1235d0e9fe17SLinus Torvalds */ 1236d0e9fe17SLinus Torvalds static struct anon_vma *reusable_anon_vma(struct vm_area_struct *old, struct vm_area_struct *a, struct vm_area_struct *b) 1237d0e9fe17SLinus Torvalds { 1238d0e9fe17SLinus Torvalds if (anon_vma_compatible(a, b)) { 12394db0c3c2SJason Low struct anon_vma *anon_vma = READ_ONCE(old->anon_vma); 1240d0e9fe17SLinus Torvalds 1241d0e9fe17SLinus Torvalds if (anon_vma && list_is_singular(&old->anon_vma_chain)) 1242d0e9fe17SLinus Torvalds return anon_vma; 1243d0e9fe17SLinus Torvalds } 1244d0e9fe17SLinus Torvalds return NULL; 1245d0e9fe17SLinus Torvalds } 1246d0e9fe17SLinus Torvalds 1247d0e9fe17SLinus Torvalds /* 12481da177e4SLinus Torvalds * find_mergeable_anon_vma is used by anon_vma_prepare, to check 12491da177e4SLinus Torvalds * neighbouring vmas for a suitable anon_vma, before it goes off 12501da177e4SLinus Torvalds * to allocate a new anon_vma. It checks because a repetitive 12511da177e4SLinus Torvalds * sequence of mprotects and faults may otherwise lead to distinct 12521da177e4SLinus Torvalds * anon_vmas being allocated, preventing vma merge in subsequent 12531da177e4SLinus Torvalds * mprotect. 12541da177e4SLinus Torvalds */ 12551da177e4SLinus Torvalds struct anon_vma *find_mergeable_anon_vma(struct vm_area_struct *vma) 12561da177e4SLinus Torvalds { 1257d0e9fe17SLinus Torvalds struct anon_vma *anon_vma; 12581da177e4SLinus Torvalds struct vm_area_struct *near; 12591da177e4SLinus Torvalds 12601da177e4SLinus Torvalds near = vma->vm_next; 12611da177e4SLinus Torvalds if (!near) 12621da177e4SLinus Torvalds goto try_prev; 12631da177e4SLinus Torvalds 1264d0e9fe17SLinus Torvalds anon_vma = reusable_anon_vma(near, vma, near); 1265d0e9fe17SLinus Torvalds if (anon_vma) 1266d0e9fe17SLinus Torvalds return anon_vma; 12671da177e4SLinus Torvalds try_prev: 12689be34c9dSLinus Torvalds near = vma->vm_prev; 12691da177e4SLinus Torvalds if (!near) 12701da177e4SLinus Torvalds goto none; 12711da177e4SLinus Torvalds 1272d0e9fe17SLinus Torvalds anon_vma = reusable_anon_vma(near, near, vma); 1273d0e9fe17SLinus Torvalds if (anon_vma) 1274d0e9fe17SLinus Torvalds return anon_vma; 12751da177e4SLinus Torvalds none: 12761da177e4SLinus Torvalds /* 12771da177e4SLinus Torvalds * There's no absolute need to look only at touching neighbours: 12781da177e4SLinus Torvalds * we could search further afield for "compatible" anon_vmas. 12791da177e4SLinus Torvalds * But it would probably just be a waste of time searching, 12801da177e4SLinus Torvalds * or lead to too many vmas hanging off the same anon_vma. 12811da177e4SLinus Torvalds * We're trying to allow mprotect remerging later on, 12821da177e4SLinus Torvalds * not trying to minimize memory used for anon_vmas. 12831da177e4SLinus Torvalds */ 12841da177e4SLinus Torvalds return NULL; 12851da177e4SLinus Torvalds } 12861da177e4SLinus Torvalds 12871da177e4SLinus Torvalds /* 128840401530SAl Viro * If a hint addr is less than mmap_min_addr change hint to be as 128940401530SAl Viro * low as possible but still greater than mmap_min_addr 129040401530SAl Viro */ 129140401530SAl Viro static inline unsigned long round_hint_to_min(unsigned long hint) 129240401530SAl Viro { 129340401530SAl Viro hint &= PAGE_MASK; 129440401530SAl Viro if (((void *)hint != NULL) && 129540401530SAl Viro (hint < mmap_min_addr)) 129640401530SAl Viro return PAGE_ALIGN(mmap_min_addr); 129740401530SAl Viro return hint; 129840401530SAl Viro } 129940401530SAl Viro 1300363ee17fSDavidlohr Bueso static inline int mlock_future_check(struct mm_struct *mm, 1301363ee17fSDavidlohr Bueso unsigned long flags, 1302363ee17fSDavidlohr Bueso unsigned long len) 1303363ee17fSDavidlohr Bueso { 1304363ee17fSDavidlohr Bueso unsigned long locked, lock_limit; 1305363ee17fSDavidlohr Bueso 1306363ee17fSDavidlohr Bueso /* mlock MCL_FUTURE? */ 1307363ee17fSDavidlohr Bueso if (flags & VM_LOCKED) { 1308363ee17fSDavidlohr Bueso locked = len >> PAGE_SHIFT; 1309363ee17fSDavidlohr Bueso locked += mm->locked_vm; 1310363ee17fSDavidlohr Bueso lock_limit = rlimit(RLIMIT_MEMLOCK); 1311363ee17fSDavidlohr Bueso lock_limit >>= PAGE_SHIFT; 1312363ee17fSDavidlohr Bueso if (locked > lock_limit && !capable(CAP_IPC_LOCK)) 1313363ee17fSDavidlohr Bueso return -EAGAIN; 1314363ee17fSDavidlohr Bueso } 1315363ee17fSDavidlohr Bueso return 0; 1316363ee17fSDavidlohr Bueso } 1317363ee17fSDavidlohr Bueso 131840401530SAl Viro /* 131927f5de79SJianjun Kong * The caller must hold down_write(¤t->mm->mmap_sem). 13201da177e4SLinus Torvalds */ 13211fcfd8dbSOleg Nesterov unsigned long do_mmap(struct file *file, unsigned long addr, 13221da177e4SLinus Torvalds unsigned long len, unsigned long prot, 13231fcfd8dbSOleg Nesterov unsigned long flags, vm_flags_t vm_flags, 1324897ab3e0SMike Rapoport unsigned long pgoff, unsigned long *populate, 1325897ab3e0SMike Rapoport struct list_head *uf) 13261da177e4SLinus Torvalds { 13271da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 132862b5f7d0SDave Hansen int pkey = 0; 13291da177e4SLinus Torvalds 133041badc15SMichel Lespinasse *populate = 0; 1331bebeb3d6SMichel Lespinasse 1332e37609bbSPiotr Kwapulinski if (!len) 1333e37609bbSPiotr Kwapulinski return -EINVAL; 1334e37609bbSPiotr Kwapulinski 13351da177e4SLinus Torvalds /* 13361da177e4SLinus Torvalds * Does the application expect PROT_READ to imply PROT_EXEC? 13371da177e4SLinus Torvalds * 13381da177e4SLinus Torvalds * (the exception is when the underlying filesystem is noexec 13391da177e4SLinus Torvalds * mounted, in which case we dont add PROT_EXEC.) 13401da177e4SLinus Torvalds */ 13411da177e4SLinus Torvalds if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC)) 134290f8572bSEric W. Biederman if (!(file && path_noexec(&file->f_path))) 13431da177e4SLinus Torvalds prot |= PROT_EXEC; 13441da177e4SLinus Torvalds 13457cd94146SEric Paris if (!(flags & MAP_FIXED)) 13467cd94146SEric Paris addr = round_hint_to_min(addr); 13477cd94146SEric Paris 13481da177e4SLinus Torvalds /* Careful about overflows.. */ 13491da177e4SLinus Torvalds len = PAGE_ALIGN(len); 13509206de95SAl Viro if (!len) 13511da177e4SLinus Torvalds return -ENOMEM; 13521da177e4SLinus Torvalds 13531da177e4SLinus Torvalds /* offset overflow? */ 13541da177e4SLinus Torvalds if ((pgoff + (len >> PAGE_SHIFT)) < pgoff) 13551da177e4SLinus Torvalds return -EOVERFLOW; 13561da177e4SLinus Torvalds 13571da177e4SLinus Torvalds /* Too many mappings? */ 13581da177e4SLinus Torvalds if (mm->map_count > sysctl_max_map_count) 13591da177e4SLinus Torvalds return -ENOMEM; 13601da177e4SLinus Torvalds 13611da177e4SLinus Torvalds /* Obtain the address to map to. we verify (or select) it and ensure 13621da177e4SLinus Torvalds * that it represents a valid section of the address space. 13631da177e4SLinus Torvalds */ 13641da177e4SLinus Torvalds addr = get_unmapped_area(file, addr, len, pgoff, flags); 1365de1741a1SAlexander Kuleshov if (offset_in_page(addr)) 13661da177e4SLinus Torvalds return addr; 13671da177e4SLinus Torvalds 136862b5f7d0SDave Hansen if (prot == PROT_EXEC) { 136962b5f7d0SDave Hansen pkey = execute_only_pkey(mm); 137062b5f7d0SDave Hansen if (pkey < 0) 137162b5f7d0SDave Hansen pkey = 0; 137262b5f7d0SDave Hansen } 137362b5f7d0SDave Hansen 13741da177e4SLinus Torvalds /* Do simple checking here so the lower-level routines won't have 13751da177e4SLinus Torvalds * to. we assume access permissions have been handled by the open 13761da177e4SLinus Torvalds * of the memory object, so we don't do any here. 13771da177e4SLinus Torvalds */ 137862b5f7d0SDave Hansen vm_flags |= calc_vm_prot_bits(prot, pkey) | calc_vm_flag_bits(flags) | 13791da177e4SLinus Torvalds mm->def_flags | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC; 13801da177e4SLinus Torvalds 1381cdf7b341SHuang Shijie if (flags & MAP_LOCKED) 13821da177e4SLinus Torvalds if (!can_do_mlock()) 13831da177e4SLinus Torvalds return -EPERM; 1384ba470de4SRik van Riel 1385363ee17fSDavidlohr Bueso if (mlock_future_check(mm, vm_flags, len)) 13861da177e4SLinus Torvalds return -EAGAIN; 13871da177e4SLinus Torvalds 13881da177e4SLinus Torvalds if (file) { 1389077bf22bSOleg Nesterov struct inode *inode = file_inode(file); 1390077bf22bSOleg Nesterov 13911da177e4SLinus Torvalds switch (flags & MAP_TYPE) { 13921da177e4SLinus Torvalds case MAP_SHARED: 13931da177e4SLinus Torvalds if ((prot&PROT_WRITE) && !(file->f_mode&FMODE_WRITE)) 13941da177e4SLinus Torvalds return -EACCES; 13951da177e4SLinus Torvalds 13961da177e4SLinus Torvalds /* 13971da177e4SLinus Torvalds * Make sure we don't allow writing to an append-only 13981da177e4SLinus Torvalds * file.. 13991da177e4SLinus Torvalds */ 14001da177e4SLinus Torvalds if (IS_APPEND(inode) && (file->f_mode & FMODE_WRITE)) 14011da177e4SLinus Torvalds return -EACCES; 14021da177e4SLinus Torvalds 14031da177e4SLinus Torvalds /* 14041da177e4SLinus Torvalds * Make sure there are no mandatory locks on the file. 14051da177e4SLinus Torvalds */ 1406d7a06983SJeff Layton if (locks_verify_locked(file)) 14071da177e4SLinus Torvalds return -EAGAIN; 14081da177e4SLinus Torvalds 14091da177e4SLinus Torvalds vm_flags |= VM_SHARED | VM_MAYSHARE; 14101da177e4SLinus Torvalds if (!(file->f_mode & FMODE_WRITE)) 14111da177e4SLinus Torvalds vm_flags &= ~(VM_MAYWRITE | VM_SHARED); 14121da177e4SLinus Torvalds 14131da177e4SLinus Torvalds /* fall through */ 14141da177e4SLinus Torvalds case MAP_PRIVATE: 14151da177e4SLinus Torvalds if (!(file->f_mode & FMODE_READ)) 14161da177e4SLinus Torvalds return -EACCES; 141790f8572bSEric W. Biederman if (path_noexec(&file->f_path)) { 141880c5606cSLinus Torvalds if (vm_flags & VM_EXEC) 141980c5606cSLinus Torvalds return -EPERM; 142080c5606cSLinus Torvalds vm_flags &= ~VM_MAYEXEC; 142180c5606cSLinus Torvalds } 142280c5606cSLinus Torvalds 142372c2d531SAl Viro if (!file->f_op->mmap) 142480c5606cSLinus Torvalds return -ENODEV; 1425b2c56e4fSOleg Nesterov if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP)) 1426b2c56e4fSOleg Nesterov return -EINVAL; 14271da177e4SLinus Torvalds break; 14281da177e4SLinus Torvalds 14291da177e4SLinus Torvalds default: 14301da177e4SLinus Torvalds return -EINVAL; 14311da177e4SLinus Torvalds } 14321da177e4SLinus Torvalds } else { 14331da177e4SLinus Torvalds switch (flags & MAP_TYPE) { 14341da177e4SLinus Torvalds case MAP_SHARED: 1435b2c56e4fSOleg Nesterov if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP)) 1436b2c56e4fSOleg Nesterov return -EINVAL; 1437ce363942STejun Heo /* 1438ce363942STejun Heo * Ignore pgoff. 1439ce363942STejun Heo */ 1440ce363942STejun Heo pgoff = 0; 14411da177e4SLinus Torvalds vm_flags |= VM_SHARED | VM_MAYSHARE; 14421da177e4SLinus Torvalds break; 14431da177e4SLinus Torvalds case MAP_PRIVATE: 14441da177e4SLinus Torvalds /* 14451da177e4SLinus Torvalds * Set pgoff according to addr for anon_vma. 14461da177e4SLinus Torvalds */ 14471da177e4SLinus Torvalds pgoff = addr >> PAGE_SHIFT; 14481da177e4SLinus Torvalds break; 14491da177e4SLinus Torvalds default: 14501da177e4SLinus Torvalds return -EINVAL; 14511da177e4SLinus Torvalds } 14521da177e4SLinus Torvalds } 14531da177e4SLinus Torvalds 1454c22c0d63SMichel Lespinasse /* 1455c22c0d63SMichel Lespinasse * Set 'VM_NORESERVE' if we should not account for the 1456c22c0d63SMichel Lespinasse * memory use of this mapping. 1457c22c0d63SMichel Lespinasse */ 1458c22c0d63SMichel Lespinasse if (flags & MAP_NORESERVE) { 1459c22c0d63SMichel Lespinasse /* We honor MAP_NORESERVE if allowed to overcommit */ 1460c22c0d63SMichel Lespinasse if (sysctl_overcommit_memory != OVERCOMMIT_NEVER) 1461c22c0d63SMichel Lespinasse vm_flags |= VM_NORESERVE; 1462c22c0d63SMichel Lespinasse 1463c22c0d63SMichel Lespinasse /* hugetlb applies strict overcommit unless MAP_NORESERVE */ 1464c22c0d63SMichel Lespinasse if (file && is_file_hugepages(file)) 1465c22c0d63SMichel Lespinasse vm_flags |= VM_NORESERVE; 1466c22c0d63SMichel Lespinasse } 1467c22c0d63SMichel Lespinasse 1468897ab3e0SMike Rapoport addr = mmap_region(file, addr, len, vm_flags, pgoff, uf); 146909a9f1d2SMichel Lespinasse if (!IS_ERR_VALUE(addr) && 147009a9f1d2SMichel Lespinasse ((vm_flags & VM_LOCKED) || 147109a9f1d2SMichel Lespinasse (flags & (MAP_POPULATE | MAP_NONBLOCK)) == MAP_POPULATE)) 147241badc15SMichel Lespinasse *populate = len; 1473bebeb3d6SMichel Lespinasse return addr; 14740165ab44SMiklos Szeredi } 14756be5ceb0SLinus Torvalds 147666f0dc48SHugh Dickins SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len, 147766f0dc48SHugh Dickins unsigned long, prot, unsigned long, flags, 147866f0dc48SHugh Dickins unsigned long, fd, unsigned long, pgoff) 147966f0dc48SHugh Dickins { 148066f0dc48SHugh Dickins struct file *file = NULL; 14811e3ee14bSChen Gang unsigned long retval; 148266f0dc48SHugh Dickins 148366f0dc48SHugh Dickins if (!(flags & MAP_ANONYMOUS)) { 1484120a795dSAl Viro audit_mmap_fd(fd, flags); 148566f0dc48SHugh Dickins file = fget(fd); 148666f0dc48SHugh Dickins if (!file) 14871e3ee14bSChen Gang return -EBADF; 1488af73e4d9SNaoya Horiguchi if (is_file_hugepages(file)) 1489af73e4d9SNaoya Horiguchi len = ALIGN(len, huge_page_size(hstate_file(file))); 1490493af578SJörn Engel retval = -EINVAL; 1491493af578SJörn Engel if (unlikely(flags & MAP_HUGETLB && !is_file_hugepages(file))) 1492493af578SJörn Engel goto out_fput; 149366f0dc48SHugh Dickins } else if (flags & MAP_HUGETLB) { 149466f0dc48SHugh Dickins struct user_struct *user = NULL; 1495c103a4dcSAndrew Morton struct hstate *hs; 1496af73e4d9SNaoya Horiguchi 149720ac2893SAnshuman Khandual hs = hstate_sizelog((flags >> MAP_HUGE_SHIFT) & MAP_HUGE_MASK); 1498091d0d55SLi Zefan if (!hs) 1499091d0d55SLi Zefan return -EINVAL; 1500091d0d55SLi Zefan 1501091d0d55SLi Zefan len = ALIGN(len, huge_page_size(hs)); 150266f0dc48SHugh Dickins /* 150366f0dc48SHugh Dickins * VM_NORESERVE is used because the reservations will be 150466f0dc48SHugh Dickins * taken when vm_ops->mmap() is called 150566f0dc48SHugh Dickins * A dummy user value is used because we are not locking 150666f0dc48SHugh Dickins * memory so no accounting is necessary 150766f0dc48SHugh Dickins */ 1508af73e4d9SNaoya Horiguchi file = hugetlb_file_setup(HUGETLB_ANON_FILE, len, 150942d7395fSAndi Kleen VM_NORESERVE, 151042d7395fSAndi Kleen &user, HUGETLB_ANONHUGE_INODE, 151142d7395fSAndi Kleen (flags >> MAP_HUGE_SHIFT) & MAP_HUGE_MASK); 151266f0dc48SHugh Dickins if (IS_ERR(file)) 151366f0dc48SHugh Dickins return PTR_ERR(file); 151466f0dc48SHugh Dickins } 151566f0dc48SHugh Dickins 151666f0dc48SHugh Dickins flags &= ~(MAP_EXECUTABLE | MAP_DENYWRITE); 151766f0dc48SHugh Dickins 15189fbeb5abSMichal Hocko retval = vm_mmap_pgoff(file, addr, len, prot, flags, pgoff); 1519493af578SJörn Engel out_fput: 152066f0dc48SHugh Dickins if (file) 152166f0dc48SHugh Dickins fput(file); 152266f0dc48SHugh Dickins return retval; 152366f0dc48SHugh Dickins } 152466f0dc48SHugh Dickins 1525a4679373SChristoph Hellwig #ifdef __ARCH_WANT_SYS_OLD_MMAP 1526a4679373SChristoph Hellwig struct mmap_arg_struct { 1527a4679373SChristoph Hellwig unsigned long addr; 1528a4679373SChristoph Hellwig unsigned long len; 1529a4679373SChristoph Hellwig unsigned long prot; 1530a4679373SChristoph Hellwig unsigned long flags; 1531a4679373SChristoph Hellwig unsigned long fd; 1532a4679373SChristoph Hellwig unsigned long offset; 1533a4679373SChristoph Hellwig }; 1534a4679373SChristoph Hellwig 1535a4679373SChristoph Hellwig SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg) 1536a4679373SChristoph Hellwig { 1537a4679373SChristoph Hellwig struct mmap_arg_struct a; 1538a4679373SChristoph Hellwig 1539a4679373SChristoph Hellwig if (copy_from_user(&a, arg, sizeof(a))) 1540a4679373SChristoph Hellwig return -EFAULT; 1541de1741a1SAlexander Kuleshov if (offset_in_page(a.offset)) 1542a4679373SChristoph Hellwig return -EINVAL; 1543a4679373SChristoph Hellwig 1544a4679373SChristoph Hellwig return sys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd, 1545a4679373SChristoph Hellwig a.offset >> PAGE_SHIFT); 1546a4679373SChristoph Hellwig } 1547a4679373SChristoph Hellwig #endif /* __ARCH_WANT_SYS_OLD_MMAP */ 1548a4679373SChristoph Hellwig 15494e950f6fSAlexey Dobriyan /* 15504e950f6fSAlexey Dobriyan * Some shared mappigns will want the pages marked read-only 15514e950f6fSAlexey Dobriyan * to track write events. If so, we'll downgrade vm_page_prot 15524e950f6fSAlexey Dobriyan * to the private version (using protection_map[] without the 15534e950f6fSAlexey Dobriyan * VM_SHARED bit). 15544e950f6fSAlexey Dobriyan */ 15556d2329f8SAndrea Arcangeli int vma_wants_writenotify(struct vm_area_struct *vma, pgprot_t vm_page_prot) 15564e950f6fSAlexey Dobriyan { 1557ca16d140SKOSAKI Motohiro vm_flags_t vm_flags = vma->vm_flags; 15588a04446aSKirill A. Shutemov const struct vm_operations_struct *vm_ops = vma->vm_ops; 15594e950f6fSAlexey Dobriyan 15604e950f6fSAlexey Dobriyan /* If it was private or non-writable, the write bit is already clear */ 15614e950f6fSAlexey Dobriyan if ((vm_flags & (VM_WRITE|VM_SHARED)) != ((VM_WRITE|VM_SHARED))) 15624e950f6fSAlexey Dobriyan return 0; 15634e950f6fSAlexey Dobriyan 15644e950f6fSAlexey Dobriyan /* The backer wishes to know when pages are first written to? */ 15658a04446aSKirill A. Shutemov if (vm_ops && (vm_ops->page_mkwrite || vm_ops->pfn_mkwrite)) 15664e950f6fSAlexey Dobriyan return 1; 15674e950f6fSAlexey Dobriyan 156864e45507SPeter Feiner /* The open routine did something to the protections that pgprot_modify 156964e45507SPeter Feiner * won't preserve? */ 15706d2329f8SAndrea Arcangeli if (pgprot_val(vm_page_prot) != 15716d2329f8SAndrea Arcangeli pgprot_val(vm_pgprot_modify(vm_page_prot, vm_flags))) 15724e950f6fSAlexey Dobriyan return 0; 15734e950f6fSAlexey Dobriyan 157464e45507SPeter Feiner /* Do we need to track softdirty? */ 157564e45507SPeter Feiner if (IS_ENABLED(CONFIG_MEM_SOFT_DIRTY) && !(vm_flags & VM_SOFTDIRTY)) 157664e45507SPeter Feiner return 1; 157764e45507SPeter Feiner 15784e950f6fSAlexey Dobriyan /* Specialty mapping? */ 15794b6e1e37SKonstantin Khlebnikov if (vm_flags & VM_PFNMAP) 15804e950f6fSAlexey Dobriyan return 0; 15814e950f6fSAlexey Dobriyan 15824e950f6fSAlexey Dobriyan /* Can the mapping track the dirty pages? */ 15834e950f6fSAlexey Dobriyan return vma->vm_file && vma->vm_file->f_mapping && 15844e950f6fSAlexey Dobriyan mapping_cap_account_dirty(vma->vm_file->f_mapping); 15854e950f6fSAlexey Dobriyan } 15864e950f6fSAlexey Dobriyan 1587fc8744adSLinus Torvalds /* 1588fc8744adSLinus Torvalds * We account for memory if it's a private writeable mapping, 15895a6fe125SMel Gorman * not hugepages and VM_NORESERVE wasn't set. 1590fc8744adSLinus Torvalds */ 1591ca16d140SKOSAKI Motohiro static inline int accountable_mapping(struct file *file, vm_flags_t vm_flags) 1592fc8744adSLinus Torvalds { 15935a6fe125SMel Gorman /* 15945a6fe125SMel Gorman * hugetlb has its own accounting separate from the core VM 15955a6fe125SMel Gorman * VM_HUGETLB may not be set yet so we cannot check for that flag. 15965a6fe125SMel Gorman */ 15975a6fe125SMel Gorman if (file && is_file_hugepages(file)) 15985a6fe125SMel Gorman return 0; 15995a6fe125SMel Gorman 1600fc8744adSLinus Torvalds return (vm_flags & (VM_NORESERVE | VM_SHARED | VM_WRITE)) == VM_WRITE; 1601fc8744adSLinus Torvalds } 1602fc8744adSLinus Torvalds 16030165ab44SMiklos Szeredi unsigned long mmap_region(struct file *file, unsigned long addr, 1604897ab3e0SMike Rapoport unsigned long len, vm_flags_t vm_flags, unsigned long pgoff, 1605897ab3e0SMike Rapoport struct list_head *uf) 16060165ab44SMiklos Szeredi { 16070165ab44SMiklos Szeredi struct mm_struct *mm = current->mm; 16080165ab44SMiklos Szeredi struct vm_area_struct *vma, *prev; 16090165ab44SMiklos Szeredi int error; 16100165ab44SMiklos Szeredi struct rb_node **rb_link, *rb_parent; 16110165ab44SMiklos Szeredi unsigned long charged = 0; 16120165ab44SMiklos Szeredi 1613e8420a8eSCyril Hrubis /* Check against address space limit. */ 161484638335SKonstantin Khlebnikov if (!may_expand_vm(mm, vm_flags, len >> PAGE_SHIFT)) { 1615e8420a8eSCyril Hrubis unsigned long nr_pages; 1616e8420a8eSCyril Hrubis 1617e8420a8eSCyril Hrubis /* 1618e8420a8eSCyril Hrubis * MAP_FIXED may remove pages of mappings that intersects with 1619e8420a8eSCyril Hrubis * requested mapping. Account for the pages it would unmap. 1620e8420a8eSCyril Hrubis */ 1621e8420a8eSCyril Hrubis nr_pages = count_vma_pages_range(mm, addr, addr + len); 1622e8420a8eSCyril Hrubis 162384638335SKonstantin Khlebnikov if (!may_expand_vm(mm, vm_flags, 162484638335SKonstantin Khlebnikov (len >> PAGE_SHIFT) - nr_pages)) 1625e8420a8eSCyril Hrubis return -ENOMEM; 1626e8420a8eSCyril Hrubis } 1627e8420a8eSCyril Hrubis 16281da177e4SLinus Torvalds /* Clear old maps */ 16299fcd1457SRasmus Villemoes while (find_vma_links(mm, addr, addr + len, &prev, &rb_link, 16309fcd1457SRasmus Villemoes &rb_parent)) { 1631897ab3e0SMike Rapoport if (do_munmap(mm, addr, len, uf)) 16321da177e4SLinus Torvalds return -ENOMEM; 16331da177e4SLinus Torvalds } 16341da177e4SLinus Torvalds 1635fc8744adSLinus Torvalds /* 16361da177e4SLinus Torvalds * Private writable mapping: check memory availability 16371da177e4SLinus Torvalds */ 16385a6fe125SMel Gorman if (accountable_mapping(file, vm_flags)) { 16391da177e4SLinus Torvalds charged = len >> PAGE_SHIFT; 1640191c5424SAl Viro if (security_vm_enough_memory_mm(mm, charged)) 16411da177e4SLinus Torvalds return -ENOMEM; 16421da177e4SLinus Torvalds vm_flags |= VM_ACCOUNT; 16431da177e4SLinus Torvalds } 16441da177e4SLinus Torvalds 16451da177e4SLinus Torvalds /* 1646de33c8dbSLinus Torvalds * Can we just expand an old mapping? 16471da177e4SLinus Torvalds */ 164819a809afSAndrea Arcangeli vma = vma_merge(mm, prev, addr, addr + len, vm_flags, 164919a809afSAndrea Arcangeli NULL, file, pgoff, NULL, NULL_VM_UFFD_CTX); 1650ba470de4SRik van Riel if (vma) 16511da177e4SLinus Torvalds goto out; 16521da177e4SLinus Torvalds 16531da177e4SLinus Torvalds /* 16541da177e4SLinus Torvalds * Determine the object being mapped and call the appropriate 16551da177e4SLinus Torvalds * specific mapper. the address has already been validated, but 16561da177e4SLinus Torvalds * not unmapped, but the maps are removed from the list. 16571da177e4SLinus Torvalds */ 1658c5e3b83eSPekka Enberg vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL); 16591da177e4SLinus Torvalds if (!vma) { 16601da177e4SLinus Torvalds error = -ENOMEM; 16611da177e4SLinus Torvalds goto unacct_error; 16621da177e4SLinus Torvalds } 16631da177e4SLinus Torvalds 16641da177e4SLinus Torvalds vma->vm_mm = mm; 16651da177e4SLinus Torvalds vma->vm_start = addr; 16661da177e4SLinus Torvalds vma->vm_end = addr + len; 16671da177e4SLinus Torvalds vma->vm_flags = vm_flags; 16683ed75eb8SColy Li vma->vm_page_prot = vm_get_page_prot(vm_flags); 16691da177e4SLinus Torvalds vma->vm_pgoff = pgoff; 16705beb4930SRik van Riel INIT_LIST_HEAD(&vma->anon_vma_chain); 16711da177e4SLinus Torvalds 16721da177e4SLinus Torvalds if (file) { 16731da177e4SLinus Torvalds if (vm_flags & VM_DENYWRITE) { 16741da177e4SLinus Torvalds error = deny_write_access(file); 16751da177e4SLinus Torvalds if (error) 16761da177e4SLinus Torvalds goto free_vma; 16771da177e4SLinus Torvalds } 16784bb5f5d9SDavid Herrmann if (vm_flags & VM_SHARED) { 16794bb5f5d9SDavid Herrmann error = mapping_map_writable(file->f_mapping); 16804bb5f5d9SDavid Herrmann if (error) 16814bb5f5d9SDavid Herrmann goto allow_write_and_free_vma; 16824bb5f5d9SDavid Herrmann } 16834bb5f5d9SDavid Herrmann 16844bb5f5d9SDavid Herrmann /* ->mmap() can change vma->vm_file, but must guarantee that 16854bb5f5d9SDavid Herrmann * vma_link() below can deny write-access if VM_DENYWRITE is set 16864bb5f5d9SDavid Herrmann * and map writably if VM_SHARED is set. This usually means the 16874bb5f5d9SDavid Herrmann * new file must not have been exposed to user-space, yet. 16884bb5f5d9SDavid Herrmann */ 1689cb0942b8SAl Viro vma->vm_file = get_file(file); 1690f74ac015SMiklos Szeredi error = call_mmap(file, vma); 16911da177e4SLinus Torvalds if (error) 16921da177e4SLinus Torvalds goto unmap_and_free_vma; 16931da177e4SLinus Torvalds 16941da177e4SLinus Torvalds /* Can addr have changed?? 16951da177e4SLinus Torvalds * 16961da177e4SLinus Torvalds * Answer: Yes, several device drivers can do it in their 16971da177e4SLinus Torvalds * f_op->mmap method. -DaveM 16982897b4d2SJoonsoo Kim * Bug: If addr is changed, prev, rb_link, rb_parent should 16992897b4d2SJoonsoo Kim * be updated for vma_link() 17001da177e4SLinus Torvalds */ 17012897b4d2SJoonsoo Kim WARN_ON_ONCE(addr != vma->vm_start); 17022897b4d2SJoonsoo Kim 17031da177e4SLinus Torvalds addr = vma->vm_start; 17041da177e4SLinus Torvalds vm_flags = vma->vm_flags; 1705f8dbf0a7SHuang Shijie } else if (vm_flags & VM_SHARED) { 1706f8dbf0a7SHuang Shijie error = shmem_zero_setup(vma); 1707f8dbf0a7SHuang Shijie if (error) 1708f8dbf0a7SHuang Shijie goto free_vma; 1709f8dbf0a7SHuang Shijie } 17101da177e4SLinus Torvalds 17114d3d5b41SOleg Nesterov vma_link(mm, vma, prev, rb_link, rb_parent); 17124d3d5b41SOleg Nesterov /* Once vma denies write, undo our temporary denial count */ 17134bb5f5d9SDavid Herrmann if (file) { 17144bb5f5d9SDavid Herrmann if (vm_flags & VM_SHARED) 17154bb5f5d9SDavid Herrmann mapping_unmap_writable(file->f_mapping); 1716e8686772SOleg Nesterov if (vm_flags & VM_DENYWRITE) 1717e8686772SOleg Nesterov allow_write_access(file); 17184bb5f5d9SDavid Herrmann } 1719e8686772SOleg Nesterov file = vma->vm_file; 17201da177e4SLinus Torvalds out: 1721cdd6c482SIngo Molnar perf_event_mmap(vma); 17220a4a9391SPeter Zijlstra 172384638335SKonstantin Khlebnikov vm_stat_account(mm, vm_flags, len >> PAGE_SHIFT); 17241da177e4SLinus Torvalds if (vm_flags & VM_LOCKED) { 1725bebeb3d6SMichel Lespinasse if (!((vm_flags & VM_SPECIAL) || is_vm_hugetlb_page(vma) || 1726bebeb3d6SMichel Lespinasse vma == get_gate_vma(current->mm))) 172706f9d8c2SKOSAKI Motohiro mm->locked_vm += (len >> PAGE_SHIFT); 1728bebeb3d6SMichel Lespinasse else 1729de60f5f1SEric B Munson vma->vm_flags &= VM_LOCKED_CLEAR_MASK; 1730bebeb3d6SMichel Lespinasse } 17312b144498SSrikar Dronamraju 1732c7a3a88cSOleg Nesterov if (file) 1733c7a3a88cSOleg Nesterov uprobe_mmap(vma); 17342b144498SSrikar Dronamraju 1735d9104d1cSCyrill Gorcunov /* 1736d9104d1cSCyrill Gorcunov * New (or expanded) vma always get soft dirty status. 1737d9104d1cSCyrill Gorcunov * Otherwise user-space soft-dirty page tracker won't 1738d9104d1cSCyrill Gorcunov * be able to distinguish situation when vma area unmapped, 1739d9104d1cSCyrill Gorcunov * then new mapped in-place (which must be aimed as 1740d9104d1cSCyrill Gorcunov * a completely new data area). 1741d9104d1cSCyrill Gorcunov */ 1742d9104d1cSCyrill Gorcunov vma->vm_flags |= VM_SOFTDIRTY; 1743d9104d1cSCyrill Gorcunov 174464e45507SPeter Feiner vma_set_page_prot(vma); 174564e45507SPeter Feiner 17461da177e4SLinus Torvalds return addr; 17471da177e4SLinus Torvalds 17481da177e4SLinus Torvalds unmap_and_free_vma: 17491da177e4SLinus Torvalds vma->vm_file = NULL; 17501da177e4SLinus Torvalds fput(file); 17511da177e4SLinus Torvalds 17521da177e4SLinus Torvalds /* Undo any partial mapping done by a device driver. */ 1753e0da382cSHugh Dickins unmap_region(mm, vma, prev, vma->vm_start, vma->vm_end); 1754e0da382cSHugh Dickins charged = 0; 17554bb5f5d9SDavid Herrmann if (vm_flags & VM_SHARED) 17564bb5f5d9SDavid Herrmann mapping_unmap_writable(file->f_mapping); 17574bb5f5d9SDavid Herrmann allow_write_and_free_vma: 17584bb5f5d9SDavid Herrmann if (vm_flags & VM_DENYWRITE) 17594bb5f5d9SDavid Herrmann allow_write_access(file); 17601da177e4SLinus Torvalds free_vma: 17611da177e4SLinus Torvalds kmem_cache_free(vm_area_cachep, vma); 17621da177e4SLinus Torvalds unacct_error: 17631da177e4SLinus Torvalds if (charged) 17641da177e4SLinus Torvalds vm_unacct_memory(charged); 17651da177e4SLinus Torvalds return error; 17661da177e4SLinus Torvalds } 17671da177e4SLinus Torvalds 1768db4fbfb9SMichel Lespinasse unsigned long unmapped_area(struct vm_unmapped_area_info *info) 1769db4fbfb9SMichel Lespinasse { 1770db4fbfb9SMichel Lespinasse /* 1771db4fbfb9SMichel Lespinasse * We implement the search by looking for an rbtree node that 1772db4fbfb9SMichel Lespinasse * immediately follows a suitable gap. That is, 1773db4fbfb9SMichel Lespinasse * - gap_start = vma->vm_prev->vm_end <= info->high_limit - length; 1774db4fbfb9SMichel Lespinasse * - gap_end = vma->vm_start >= info->low_limit + length; 1775db4fbfb9SMichel Lespinasse * - gap_end - gap_start >= length 1776db4fbfb9SMichel Lespinasse */ 1777db4fbfb9SMichel Lespinasse 1778db4fbfb9SMichel Lespinasse struct mm_struct *mm = current->mm; 1779db4fbfb9SMichel Lespinasse struct vm_area_struct *vma; 1780db4fbfb9SMichel Lespinasse unsigned long length, low_limit, high_limit, gap_start, gap_end; 1781db4fbfb9SMichel Lespinasse 1782db4fbfb9SMichel Lespinasse /* Adjust search length to account for worst case alignment overhead */ 1783db4fbfb9SMichel Lespinasse length = info->length + info->align_mask; 1784db4fbfb9SMichel Lespinasse if (length < info->length) 1785db4fbfb9SMichel Lespinasse return -ENOMEM; 1786db4fbfb9SMichel Lespinasse 1787db4fbfb9SMichel Lespinasse /* Adjust search limits by the desired length */ 1788db4fbfb9SMichel Lespinasse if (info->high_limit < length) 1789db4fbfb9SMichel Lespinasse return -ENOMEM; 1790db4fbfb9SMichel Lespinasse high_limit = info->high_limit - length; 1791db4fbfb9SMichel Lespinasse 1792db4fbfb9SMichel Lespinasse if (info->low_limit > high_limit) 1793db4fbfb9SMichel Lespinasse return -ENOMEM; 1794db4fbfb9SMichel Lespinasse low_limit = info->low_limit + length; 1795db4fbfb9SMichel Lespinasse 1796db4fbfb9SMichel Lespinasse /* Check if rbtree root looks promising */ 1797db4fbfb9SMichel Lespinasse if (RB_EMPTY_ROOT(&mm->mm_rb)) 1798db4fbfb9SMichel Lespinasse goto check_highest; 1799db4fbfb9SMichel Lespinasse vma = rb_entry(mm->mm_rb.rb_node, struct vm_area_struct, vm_rb); 1800db4fbfb9SMichel Lespinasse if (vma->rb_subtree_gap < length) 1801db4fbfb9SMichel Lespinasse goto check_highest; 1802db4fbfb9SMichel Lespinasse 1803db4fbfb9SMichel Lespinasse while (true) { 1804db4fbfb9SMichel Lespinasse /* Visit left subtree if it looks promising */ 18051be7107fSHugh Dickins gap_end = vm_start_gap(vma); 1806db4fbfb9SMichel Lespinasse if (gap_end >= low_limit && vma->vm_rb.rb_left) { 1807db4fbfb9SMichel Lespinasse struct vm_area_struct *left = 1808db4fbfb9SMichel Lespinasse rb_entry(vma->vm_rb.rb_left, 1809db4fbfb9SMichel Lespinasse struct vm_area_struct, vm_rb); 1810db4fbfb9SMichel Lespinasse if (left->rb_subtree_gap >= length) { 1811db4fbfb9SMichel Lespinasse vma = left; 1812db4fbfb9SMichel Lespinasse continue; 1813db4fbfb9SMichel Lespinasse } 1814db4fbfb9SMichel Lespinasse } 1815db4fbfb9SMichel Lespinasse 18161be7107fSHugh Dickins gap_start = vma->vm_prev ? vm_end_gap(vma->vm_prev) : 0; 1817db4fbfb9SMichel Lespinasse check_current: 1818db4fbfb9SMichel Lespinasse /* Check if current node has a suitable gap */ 1819db4fbfb9SMichel Lespinasse if (gap_start > high_limit) 1820db4fbfb9SMichel Lespinasse return -ENOMEM; 1821f4cb767dSHugh Dickins if (gap_end >= low_limit && 1822f4cb767dSHugh Dickins gap_end > gap_start && gap_end - gap_start >= length) 1823db4fbfb9SMichel Lespinasse goto found; 1824db4fbfb9SMichel Lespinasse 1825db4fbfb9SMichel Lespinasse /* Visit right subtree if it looks promising */ 1826db4fbfb9SMichel Lespinasse if (vma->vm_rb.rb_right) { 1827db4fbfb9SMichel Lespinasse struct vm_area_struct *right = 1828db4fbfb9SMichel Lespinasse rb_entry(vma->vm_rb.rb_right, 1829db4fbfb9SMichel Lespinasse struct vm_area_struct, vm_rb); 1830db4fbfb9SMichel Lespinasse if (right->rb_subtree_gap >= length) { 1831db4fbfb9SMichel Lespinasse vma = right; 1832db4fbfb9SMichel Lespinasse continue; 1833db4fbfb9SMichel Lespinasse } 1834db4fbfb9SMichel Lespinasse } 1835db4fbfb9SMichel Lespinasse 1836db4fbfb9SMichel Lespinasse /* Go back up the rbtree to find next candidate node */ 1837db4fbfb9SMichel Lespinasse while (true) { 1838db4fbfb9SMichel Lespinasse struct rb_node *prev = &vma->vm_rb; 1839db4fbfb9SMichel Lespinasse if (!rb_parent(prev)) 1840db4fbfb9SMichel Lespinasse goto check_highest; 1841db4fbfb9SMichel Lespinasse vma = rb_entry(rb_parent(prev), 1842db4fbfb9SMichel Lespinasse struct vm_area_struct, vm_rb); 1843db4fbfb9SMichel Lespinasse if (prev == vma->vm_rb.rb_left) { 18441be7107fSHugh Dickins gap_start = vm_end_gap(vma->vm_prev); 18451be7107fSHugh Dickins gap_end = vm_start_gap(vma); 1846db4fbfb9SMichel Lespinasse goto check_current; 1847db4fbfb9SMichel Lespinasse } 1848db4fbfb9SMichel Lespinasse } 1849db4fbfb9SMichel Lespinasse } 1850db4fbfb9SMichel Lespinasse 1851db4fbfb9SMichel Lespinasse check_highest: 1852db4fbfb9SMichel Lespinasse /* Check highest gap, which does not precede any rbtree node */ 1853db4fbfb9SMichel Lespinasse gap_start = mm->highest_vm_end; 1854db4fbfb9SMichel Lespinasse gap_end = ULONG_MAX; /* Only for VM_BUG_ON below */ 1855db4fbfb9SMichel Lespinasse if (gap_start > high_limit) 1856db4fbfb9SMichel Lespinasse return -ENOMEM; 1857db4fbfb9SMichel Lespinasse 1858db4fbfb9SMichel Lespinasse found: 1859db4fbfb9SMichel Lespinasse /* We found a suitable gap. Clip it with the original low_limit. */ 1860db4fbfb9SMichel Lespinasse if (gap_start < info->low_limit) 1861db4fbfb9SMichel Lespinasse gap_start = info->low_limit; 1862db4fbfb9SMichel Lespinasse 1863db4fbfb9SMichel Lespinasse /* Adjust gap address to the desired alignment */ 1864db4fbfb9SMichel Lespinasse gap_start += (info->align_offset - gap_start) & info->align_mask; 1865db4fbfb9SMichel Lespinasse 1866db4fbfb9SMichel Lespinasse VM_BUG_ON(gap_start + info->length > info->high_limit); 1867db4fbfb9SMichel Lespinasse VM_BUG_ON(gap_start + info->length > gap_end); 1868db4fbfb9SMichel Lespinasse return gap_start; 1869db4fbfb9SMichel Lespinasse } 1870db4fbfb9SMichel Lespinasse 1871db4fbfb9SMichel Lespinasse unsigned long unmapped_area_topdown(struct vm_unmapped_area_info *info) 1872db4fbfb9SMichel Lespinasse { 1873db4fbfb9SMichel Lespinasse struct mm_struct *mm = current->mm; 1874db4fbfb9SMichel Lespinasse struct vm_area_struct *vma; 1875db4fbfb9SMichel Lespinasse unsigned long length, low_limit, high_limit, gap_start, gap_end; 1876db4fbfb9SMichel Lespinasse 1877db4fbfb9SMichel Lespinasse /* Adjust search length to account for worst case alignment overhead */ 1878db4fbfb9SMichel Lespinasse length = info->length + info->align_mask; 1879db4fbfb9SMichel Lespinasse if (length < info->length) 1880db4fbfb9SMichel Lespinasse return -ENOMEM; 1881db4fbfb9SMichel Lespinasse 1882db4fbfb9SMichel Lespinasse /* 1883db4fbfb9SMichel Lespinasse * Adjust search limits by the desired length. 1884db4fbfb9SMichel Lespinasse * See implementation comment at top of unmapped_area(). 1885db4fbfb9SMichel Lespinasse */ 1886db4fbfb9SMichel Lespinasse gap_end = info->high_limit; 1887db4fbfb9SMichel Lespinasse if (gap_end < length) 1888db4fbfb9SMichel Lespinasse return -ENOMEM; 1889db4fbfb9SMichel Lespinasse high_limit = gap_end - length; 1890db4fbfb9SMichel Lespinasse 1891db4fbfb9SMichel Lespinasse if (info->low_limit > high_limit) 1892db4fbfb9SMichel Lespinasse return -ENOMEM; 1893db4fbfb9SMichel Lespinasse low_limit = info->low_limit + length; 1894db4fbfb9SMichel Lespinasse 1895db4fbfb9SMichel Lespinasse /* Check highest gap, which does not precede any rbtree node */ 1896db4fbfb9SMichel Lespinasse gap_start = mm->highest_vm_end; 1897db4fbfb9SMichel Lespinasse if (gap_start <= high_limit) 1898db4fbfb9SMichel Lespinasse goto found_highest; 1899db4fbfb9SMichel Lespinasse 1900db4fbfb9SMichel Lespinasse /* Check if rbtree root looks promising */ 1901db4fbfb9SMichel Lespinasse if (RB_EMPTY_ROOT(&mm->mm_rb)) 1902db4fbfb9SMichel Lespinasse return -ENOMEM; 1903db4fbfb9SMichel Lespinasse vma = rb_entry(mm->mm_rb.rb_node, struct vm_area_struct, vm_rb); 1904db4fbfb9SMichel Lespinasse if (vma->rb_subtree_gap < length) 1905db4fbfb9SMichel Lespinasse return -ENOMEM; 1906db4fbfb9SMichel Lespinasse 1907db4fbfb9SMichel Lespinasse while (true) { 1908db4fbfb9SMichel Lespinasse /* Visit right subtree if it looks promising */ 19091be7107fSHugh Dickins gap_start = vma->vm_prev ? vm_end_gap(vma->vm_prev) : 0; 1910db4fbfb9SMichel Lespinasse if (gap_start <= high_limit && vma->vm_rb.rb_right) { 1911db4fbfb9SMichel Lespinasse struct vm_area_struct *right = 1912db4fbfb9SMichel Lespinasse rb_entry(vma->vm_rb.rb_right, 1913db4fbfb9SMichel Lespinasse struct vm_area_struct, vm_rb); 1914db4fbfb9SMichel Lespinasse if (right->rb_subtree_gap >= length) { 1915db4fbfb9SMichel Lespinasse vma = right; 1916db4fbfb9SMichel Lespinasse continue; 1917db4fbfb9SMichel Lespinasse } 1918db4fbfb9SMichel Lespinasse } 1919db4fbfb9SMichel Lespinasse 1920db4fbfb9SMichel Lespinasse check_current: 1921db4fbfb9SMichel Lespinasse /* Check if current node has a suitable gap */ 19221be7107fSHugh Dickins gap_end = vm_start_gap(vma); 1923db4fbfb9SMichel Lespinasse if (gap_end < low_limit) 1924db4fbfb9SMichel Lespinasse return -ENOMEM; 1925f4cb767dSHugh Dickins if (gap_start <= high_limit && 1926f4cb767dSHugh Dickins gap_end > gap_start && gap_end - gap_start >= length) 1927db4fbfb9SMichel Lespinasse goto found; 1928db4fbfb9SMichel Lespinasse 1929db4fbfb9SMichel Lespinasse /* Visit left subtree if it looks promising */ 1930db4fbfb9SMichel Lespinasse if (vma->vm_rb.rb_left) { 1931db4fbfb9SMichel Lespinasse struct vm_area_struct *left = 1932db4fbfb9SMichel Lespinasse rb_entry(vma->vm_rb.rb_left, 1933db4fbfb9SMichel Lespinasse struct vm_area_struct, vm_rb); 1934db4fbfb9SMichel Lespinasse if (left->rb_subtree_gap >= length) { 1935db4fbfb9SMichel Lespinasse vma = left; 1936db4fbfb9SMichel Lespinasse continue; 1937db4fbfb9SMichel Lespinasse } 1938db4fbfb9SMichel Lespinasse } 1939db4fbfb9SMichel Lespinasse 1940db4fbfb9SMichel Lespinasse /* Go back up the rbtree to find next candidate node */ 1941db4fbfb9SMichel Lespinasse while (true) { 1942db4fbfb9SMichel Lespinasse struct rb_node *prev = &vma->vm_rb; 1943db4fbfb9SMichel Lespinasse if (!rb_parent(prev)) 1944db4fbfb9SMichel Lespinasse return -ENOMEM; 1945db4fbfb9SMichel Lespinasse vma = rb_entry(rb_parent(prev), 1946db4fbfb9SMichel Lespinasse struct vm_area_struct, vm_rb); 1947db4fbfb9SMichel Lespinasse if (prev == vma->vm_rb.rb_right) { 1948db4fbfb9SMichel Lespinasse gap_start = vma->vm_prev ? 19491be7107fSHugh Dickins vm_end_gap(vma->vm_prev) : 0; 1950db4fbfb9SMichel Lespinasse goto check_current; 1951db4fbfb9SMichel Lespinasse } 1952db4fbfb9SMichel Lespinasse } 1953db4fbfb9SMichel Lespinasse } 1954db4fbfb9SMichel Lespinasse 1955db4fbfb9SMichel Lespinasse found: 1956db4fbfb9SMichel Lespinasse /* We found a suitable gap. Clip it with the original high_limit. */ 1957db4fbfb9SMichel Lespinasse if (gap_end > info->high_limit) 1958db4fbfb9SMichel Lespinasse gap_end = info->high_limit; 1959db4fbfb9SMichel Lespinasse 1960db4fbfb9SMichel Lespinasse found_highest: 1961db4fbfb9SMichel Lespinasse /* Compute highest gap address at the desired alignment */ 1962db4fbfb9SMichel Lespinasse gap_end -= info->length; 1963db4fbfb9SMichel Lespinasse gap_end -= (gap_end - info->align_offset) & info->align_mask; 1964db4fbfb9SMichel Lespinasse 1965db4fbfb9SMichel Lespinasse VM_BUG_ON(gap_end < info->low_limit); 1966db4fbfb9SMichel Lespinasse VM_BUG_ON(gap_end < gap_start); 1967db4fbfb9SMichel Lespinasse return gap_end; 1968db4fbfb9SMichel Lespinasse } 1969db4fbfb9SMichel Lespinasse 19701da177e4SLinus Torvalds /* Get an address range which is currently unmapped. 19711da177e4SLinus Torvalds * For shmat() with addr=0. 19721da177e4SLinus Torvalds * 19731da177e4SLinus Torvalds * Ugly calling convention alert: 19741da177e4SLinus Torvalds * Return value with the low bits set means error value, 19751da177e4SLinus Torvalds * ie 19761da177e4SLinus Torvalds * if (ret & ~PAGE_MASK) 19771da177e4SLinus Torvalds * error = ret; 19781da177e4SLinus Torvalds * 19791da177e4SLinus Torvalds * This function "knows" that -ENOMEM has the bits set. 19801da177e4SLinus Torvalds */ 19811da177e4SLinus Torvalds #ifndef HAVE_ARCH_UNMAPPED_AREA 19821da177e4SLinus Torvalds unsigned long 19831da177e4SLinus Torvalds arch_get_unmapped_area(struct file *filp, unsigned long addr, 19841da177e4SLinus Torvalds unsigned long len, unsigned long pgoff, unsigned long flags) 19851da177e4SLinus Torvalds { 19861da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 19871be7107fSHugh Dickins struct vm_area_struct *vma, *prev; 1988db4fbfb9SMichel Lespinasse struct vm_unmapped_area_info info; 19891da177e4SLinus Torvalds 19902afc745fSAkira Takeuchi if (len > TASK_SIZE - mmap_min_addr) 19911da177e4SLinus Torvalds return -ENOMEM; 19921da177e4SLinus Torvalds 199306abdfb4SBenjamin Herrenschmidt if (flags & MAP_FIXED) 199406abdfb4SBenjamin Herrenschmidt return addr; 199506abdfb4SBenjamin Herrenschmidt 19961da177e4SLinus Torvalds if (addr) { 19971da177e4SLinus Torvalds addr = PAGE_ALIGN(addr); 19981be7107fSHugh Dickins vma = find_vma_prev(mm, addr, &prev); 19992afc745fSAkira Takeuchi if (TASK_SIZE - len >= addr && addr >= mmap_min_addr && 20001be7107fSHugh Dickins (!vma || addr + len <= vm_start_gap(vma)) && 20011be7107fSHugh Dickins (!prev || addr >= vm_end_gap(prev))) 20021da177e4SLinus Torvalds return addr; 20031da177e4SLinus Torvalds } 20041da177e4SLinus Torvalds 2005db4fbfb9SMichel Lespinasse info.flags = 0; 2006db4fbfb9SMichel Lespinasse info.length = len; 20074e99b021SHeiko Carstens info.low_limit = mm->mmap_base; 2008db4fbfb9SMichel Lespinasse info.high_limit = TASK_SIZE; 2009db4fbfb9SMichel Lespinasse info.align_mask = 0; 2010db4fbfb9SMichel Lespinasse return vm_unmapped_area(&info); 20111da177e4SLinus Torvalds } 20121da177e4SLinus Torvalds #endif 20131da177e4SLinus Torvalds 20141da177e4SLinus Torvalds /* 20151da177e4SLinus Torvalds * This mmap-allocator allocates new areas top-down from below the 20161da177e4SLinus Torvalds * stack's low limit (the base): 20171da177e4SLinus Torvalds */ 20181da177e4SLinus Torvalds #ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN 20191da177e4SLinus Torvalds unsigned long 20201da177e4SLinus Torvalds arch_get_unmapped_area_topdown(struct file *filp, const unsigned long addr0, 20211da177e4SLinus Torvalds const unsigned long len, const unsigned long pgoff, 20221da177e4SLinus Torvalds const unsigned long flags) 20231da177e4SLinus Torvalds { 20241be7107fSHugh Dickins struct vm_area_struct *vma, *prev; 20251da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 2026db4fbfb9SMichel Lespinasse unsigned long addr = addr0; 2027db4fbfb9SMichel Lespinasse struct vm_unmapped_area_info info; 20281da177e4SLinus Torvalds 20291da177e4SLinus Torvalds /* requested length too big for entire address space */ 20302afc745fSAkira Takeuchi if (len > TASK_SIZE - mmap_min_addr) 20311da177e4SLinus Torvalds return -ENOMEM; 20321da177e4SLinus Torvalds 203306abdfb4SBenjamin Herrenschmidt if (flags & MAP_FIXED) 203406abdfb4SBenjamin Herrenschmidt return addr; 203506abdfb4SBenjamin Herrenschmidt 20361da177e4SLinus Torvalds /* requesting a specific address */ 20371da177e4SLinus Torvalds if (addr) { 20381da177e4SLinus Torvalds addr = PAGE_ALIGN(addr); 20391be7107fSHugh Dickins vma = find_vma_prev(mm, addr, &prev); 20402afc745fSAkira Takeuchi if (TASK_SIZE - len >= addr && addr >= mmap_min_addr && 20411be7107fSHugh Dickins (!vma || addr + len <= vm_start_gap(vma)) && 20421be7107fSHugh Dickins (!prev || addr >= vm_end_gap(prev))) 20431da177e4SLinus Torvalds return addr; 20441da177e4SLinus Torvalds } 20451da177e4SLinus Torvalds 2046db4fbfb9SMichel Lespinasse info.flags = VM_UNMAPPED_AREA_TOPDOWN; 2047db4fbfb9SMichel Lespinasse info.length = len; 20482afc745fSAkira Takeuchi info.low_limit = max(PAGE_SIZE, mmap_min_addr); 2049db4fbfb9SMichel Lespinasse info.high_limit = mm->mmap_base; 2050db4fbfb9SMichel Lespinasse info.align_mask = 0; 2051db4fbfb9SMichel Lespinasse addr = vm_unmapped_area(&info); 2052b716ad95SXiao Guangrong 20531da177e4SLinus Torvalds /* 20541da177e4SLinus Torvalds * A failed mmap() very likely causes application failure, 20551da177e4SLinus Torvalds * so fall back to the bottom-up function here. This scenario 20561da177e4SLinus Torvalds * can happen with large stack limits and large mmap() 20571da177e4SLinus Torvalds * allocations. 20581da177e4SLinus Torvalds */ 2059de1741a1SAlexander Kuleshov if (offset_in_page(addr)) { 2060db4fbfb9SMichel Lespinasse VM_BUG_ON(addr != -ENOMEM); 2061db4fbfb9SMichel Lespinasse info.flags = 0; 2062db4fbfb9SMichel Lespinasse info.low_limit = TASK_UNMAPPED_BASE; 2063db4fbfb9SMichel Lespinasse info.high_limit = TASK_SIZE; 2064db4fbfb9SMichel Lespinasse addr = vm_unmapped_area(&info); 2065db4fbfb9SMichel Lespinasse } 20661da177e4SLinus Torvalds 20671da177e4SLinus Torvalds return addr; 20681da177e4SLinus Torvalds } 20691da177e4SLinus Torvalds #endif 20701da177e4SLinus Torvalds 20711da177e4SLinus Torvalds unsigned long 20721da177e4SLinus Torvalds get_unmapped_area(struct file *file, unsigned long addr, unsigned long len, 20731da177e4SLinus Torvalds unsigned long pgoff, unsigned long flags) 20741da177e4SLinus Torvalds { 207506abdfb4SBenjamin Herrenschmidt unsigned long (*get_area)(struct file *, unsigned long, 207606abdfb4SBenjamin Herrenschmidt unsigned long, unsigned long, unsigned long); 207707ab67c8SLinus Torvalds 20789206de95SAl Viro unsigned long error = arch_mmap_check(addr, len, flags); 20799206de95SAl Viro if (error) 20809206de95SAl Viro return error; 20819206de95SAl Viro 20829206de95SAl Viro /* Careful about overflows.. */ 20839206de95SAl Viro if (len > TASK_SIZE) 20849206de95SAl Viro return -ENOMEM; 20859206de95SAl Viro 208607ab67c8SLinus Torvalds get_area = current->mm->get_unmapped_area; 2087c01d5b30SHugh Dickins if (file) { 2088c01d5b30SHugh Dickins if (file->f_op->get_unmapped_area) 208907ab67c8SLinus Torvalds get_area = file->f_op->get_unmapped_area; 2090c01d5b30SHugh Dickins } else if (flags & MAP_SHARED) { 2091c01d5b30SHugh Dickins /* 2092c01d5b30SHugh Dickins * mmap_region() will call shmem_zero_setup() to create a file, 2093c01d5b30SHugh Dickins * so use shmem's get_unmapped_area in case it can be huge. 2094c01d5b30SHugh Dickins * do_mmap_pgoff() will clear pgoff, so match alignment. 2095c01d5b30SHugh Dickins */ 2096c01d5b30SHugh Dickins pgoff = 0; 2097c01d5b30SHugh Dickins get_area = shmem_get_unmapped_area; 2098c01d5b30SHugh Dickins } 2099c01d5b30SHugh Dickins 210007ab67c8SLinus Torvalds addr = get_area(file, addr, len, pgoff, flags); 210107ab67c8SLinus Torvalds if (IS_ERR_VALUE(addr)) 210207ab67c8SLinus Torvalds return addr; 210307ab67c8SLinus Torvalds 21041da177e4SLinus Torvalds if (addr > TASK_SIZE - len) 21051da177e4SLinus Torvalds return -ENOMEM; 2106de1741a1SAlexander Kuleshov if (offset_in_page(addr)) 21071da177e4SLinus Torvalds return -EINVAL; 210806abdfb4SBenjamin Herrenschmidt 21099ac4ed4bSAl Viro error = security_mmap_addr(addr); 21109ac4ed4bSAl Viro return error ? error : addr; 21111da177e4SLinus Torvalds } 21121da177e4SLinus Torvalds 21131da177e4SLinus Torvalds EXPORT_SYMBOL(get_unmapped_area); 21141da177e4SLinus Torvalds 21151da177e4SLinus Torvalds /* Look up the first VMA which satisfies addr < vm_end, NULL if none. */ 21161da177e4SLinus Torvalds struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr) 21171da177e4SLinus Torvalds { 2118615d6e87SDavidlohr Bueso struct rb_node *rb_node; 2119615d6e87SDavidlohr Bueso struct vm_area_struct *vma; 21201da177e4SLinus Torvalds 21211da177e4SLinus Torvalds /* Check the cache first. */ 2122615d6e87SDavidlohr Bueso vma = vmacache_find(mm, addr); 2123615d6e87SDavidlohr Bueso if (likely(vma)) 2124615d6e87SDavidlohr Bueso return vma; 21251da177e4SLinus Torvalds 21261da177e4SLinus Torvalds rb_node = mm->mm_rb.rb_node; 21271da177e4SLinus Torvalds 21281da177e4SLinus Torvalds while (rb_node) { 2129615d6e87SDavidlohr Bueso struct vm_area_struct *tmp; 21301da177e4SLinus Torvalds 2131615d6e87SDavidlohr Bueso tmp = rb_entry(rb_node, struct vm_area_struct, vm_rb); 21321da177e4SLinus Torvalds 2133615d6e87SDavidlohr Bueso if (tmp->vm_end > addr) { 2134615d6e87SDavidlohr Bueso vma = tmp; 2135615d6e87SDavidlohr Bueso if (tmp->vm_start <= addr) 21361da177e4SLinus Torvalds break; 21371da177e4SLinus Torvalds rb_node = rb_node->rb_left; 21381da177e4SLinus Torvalds } else 21391da177e4SLinus Torvalds rb_node = rb_node->rb_right; 21401da177e4SLinus Torvalds } 2141615d6e87SDavidlohr Bueso 21421da177e4SLinus Torvalds if (vma) 2143615d6e87SDavidlohr Bueso vmacache_update(addr, vma); 21441da177e4SLinus Torvalds return vma; 21451da177e4SLinus Torvalds } 21461da177e4SLinus Torvalds 21471da177e4SLinus Torvalds EXPORT_SYMBOL(find_vma); 21481da177e4SLinus Torvalds 21496bd4837dSKOSAKI Motohiro /* 21506bd4837dSKOSAKI Motohiro * Same as find_vma, but also return a pointer to the previous VMA in *pprev. 21516bd4837dSKOSAKI Motohiro */ 21521da177e4SLinus Torvalds struct vm_area_struct * 21531da177e4SLinus Torvalds find_vma_prev(struct mm_struct *mm, unsigned long addr, 21541da177e4SLinus Torvalds struct vm_area_struct **pprev) 21551da177e4SLinus Torvalds { 21566bd4837dSKOSAKI Motohiro struct vm_area_struct *vma; 21571da177e4SLinus Torvalds 21586bd4837dSKOSAKI Motohiro vma = find_vma(mm, addr); 215983cd904dSMikulas Patocka if (vma) { 216083cd904dSMikulas Patocka *pprev = vma->vm_prev; 216183cd904dSMikulas Patocka } else { 216283cd904dSMikulas Patocka struct rb_node *rb_node = mm->mm_rb.rb_node; 216383cd904dSMikulas Patocka *pprev = NULL; 216483cd904dSMikulas Patocka while (rb_node) { 216583cd904dSMikulas Patocka *pprev = rb_entry(rb_node, struct vm_area_struct, vm_rb); 216683cd904dSMikulas Patocka rb_node = rb_node->rb_right; 216783cd904dSMikulas Patocka } 216883cd904dSMikulas Patocka } 21696bd4837dSKOSAKI Motohiro return vma; 21701da177e4SLinus Torvalds } 21711da177e4SLinus Torvalds 21721da177e4SLinus Torvalds /* 21731da177e4SLinus Torvalds * Verify that the stack growth is acceptable and 21741da177e4SLinus Torvalds * update accounting. This is shared with both the 21751da177e4SLinus Torvalds * grow-up and grow-down cases. 21761da177e4SLinus Torvalds */ 21771be7107fSHugh Dickins static int acct_stack_growth(struct vm_area_struct *vma, 21781be7107fSHugh Dickins unsigned long size, unsigned long grow) 21791da177e4SLinus Torvalds { 21801da177e4SLinus Torvalds struct mm_struct *mm = vma->vm_mm; 21811be7107fSHugh Dickins unsigned long new_start; 21821da177e4SLinus Torvalds 21831da177e4SLinus Torvalds /* address space limit tests */ 218484638335SKonstantin Khlebnikov if (!may_expand_vm(mm, vma->vm_flags, grow)) 21851da177e4SLinus Torvalds return -ENOMEM; 21861da177e4SLinus Torvalds 21871da177e4SLinus Torvalds /* Stack limit test */ 218824c79d8eSKrzysztof Opasiak if (size > rlimit(RLIMIT_STACK)) 21891da177e4SLinus Torvalds return -ENOMEM; 21901da177e4SLinus Torvalds 21911da177e4SLinus Torvalds /* mlock limit tests */ 21921da177e4SLinus Torvalds if (vma->vm_flags & VM_LOCKED) { 21931da177e4SLinus Torvalds unsigned long locked; 21941da177e4SLinus Torvalds unsigned long limit; 21951da177e4SLinus Torvalds locked = mm->locked_vm + grow; 219624c79d8eSKrzysztof Opasiak limit = rlimit(RLIMIT_MEMLOCK); 219759e99e5bSJiri Slaby limit >>= PAGE_SHIFT; 21981da177e4SLinus Torvalds if (locked > limit && !capable(CAP_IPC_LOCK)) 21991da177e4SLinus Torvalds return -ENOMEM; 22001da177e4SLinus Torvalds } 22011da177e4SLinus Torvalds 22020d59a01bSAdam Litke /* Check to ensure the stack will not grow into a hugetlb-only region */ 22030d59a01bSAdam Litke new_start = (vma->vm_flags & VM_GROWSUP) ? vma->vm_start : 22040d59a01bSAdam Litke vma->vm_end - size; 22050d59a01bSAdam Litke if (is_hugepage_only_range(vma->vm_mm, new_start, size)) 22060d59a01bSAdam Litke return -EFAULT; 22070d59a01bSAdam Litke 22081da177e4SLinus Torvalds /* 22091da177e4SLinus Torvalds * Overcommit.. This must be the final test, as it will 22101da177e4SLinus Torvalds * update security statistics. 22111da177e4SLinus Torvalds */ 221205fa199dSHugh Dickins if (security_vm_enough_memory_mm(mm, grow)) 22131da177e4SLinus Torvalds return -ENOMEM; 22141da177e4SLinus Torvalds 22151da177e4SLinus Torvalds return 0; 22161da177e4SLinus Torvalds } 22171da177e4SLinus Torvalds 221846dea3d0SHugh Dickins #if defined(CONFIG_STACK_GROWSUP) || defined(CONFIG_IA64) 22191da177e4SLinus Torvalds /* 222046dea3d0SHugh Dickins * PA-RISC uses this for its stack; IA64 for its Register Backing Store. 222146dea3d0SHugh Dickins * vma is the last one with address > vma->vm_end. Have to extend vma. 22221da177e4SLinus Torvalds */ 222346dea3d0SHugh Dickins int expand_upwards(struct vm_area_struct *vma, unsigned long address) 22241da177e4SLinus Torvalds { 222509357814SOleg Nesterov struct mm_struct *mm = vma->vm_mm; 22261be7107fSHugh Dickins struct vm_area_struct *next; 22271be7107fSHugh Dickins unsigned long gap_addr; 222812352d3cSKonstantin Khlebnikov int error = 0; 22291da177e4SLinus Torvalds 22301da177e4SLinus Torvalds if (!(vma->vm_flags & VM_GROWSUP)) 22311da177e4SLinus Torvalds return -EFAULT; 22321da177e4SLinus Torvalds 2233bd726c90SHelge Deller /* Guard against exceeding limits of the address space. */ 22341be7107fSHugh Dickins address &= PAGE_MASK; 223537511fb5SHelge Deller if (address >= (TASK_SIZE & PAGE_MASK)) 223612352d3cSKonstantin Khlebnikov return -ENOMEM; 2237bd726c90SHelge Deller address += PAGE_SIZE; 223812352d3cSKonstantin Khlebnikov 22391be7107fSHugh Dickins /* Enforce stack_guard_gap */ 22401be7107fSHugh Dickins gap_addr = address + stack_guard_gap; 2241bd726c90SHelge Deller 2242bd726c90SHelge Deller /* Guard against overflow */ 2243bd726c90SHelge Deller if (gap_addr < address || gap_addr > TASK_SIZE) 2244bd726c90SHelge Deller gap_addr = TASK_SIZE; 2245bd726c90SHelge Deller 22461be7107fSHugh Dickins next = vma->vm_next; 2247561b5e07SMichal Hocko if (next && next->vm_start < gap_addr && 2248561b5e07SMichal Hocko (next->vm_flags & (VM_WRITE|VM_READ|VM_EXEC))) { 22491be7107fSHugh Dickins if (!(next->vm_flags & VM_GROWSUP)) 22501be7107fSHugh Dickins return -ENOMEM; 22511be7107fSHugh Dickins /* Check that both stack segments have the same anon_vma? */ 22521be7107fSHugh Dickins } 22531be7107fSHugh Dickins 225412352d3cSKonstantin Khlebnikov /* We must make sure the anon_vma is allocated. */ 22551da177e4SLinus Torvalds if (unlikely(anon_vma_prepare(vma))) 22561da177e4SLinus Torvalds return -ENOMEM; 22571da177e4SLinus Torvalds 22581da177e4SLinus Torvalds /* 22591da177e4SLinus Torvalds * vma->vm_start/vm_end cannot change under us because the caller 22601da177e4SLinus Torvalds * is required to hold the mmap_sem in read mode. We need the 22611da177e4SLinus Torvalds * anon_vma lock to serialize against concurrent expand_stacks. 22621da177e4SLinus Torvalds */ 226312352d3cSKonstantin Khlebnikov anon_vma_lock_write(vma->anon_vma); 22641da177e4SLinus Torvalds 22651da177e4SLinus Torvalds /* Somebody else might have raced and expanded it already */ 22661da177e4SLinus Torvalds if (address > vma->vm_end) { 22671da177e4SLinus Torvalds unsigned long size, grow; 22681da177e4SLinus Torvalds 22691da177e4SLinus Torvalds size = address - vma->vm_start; 22701da177e4SLinus Torvalds grow = (address - vma->vm_end) >> PAGE_SHIFT; 22711da177e4SLinus Torvalds 227242c36f63SHugh Dickins error = -ENOMEM; 227342c36f63SHugh Dickins if (vma->vm_pgoff + (size >> PAGE_SHIFT) >= vma->vm_pgoff) { 22741da177e4SLinus Torvalds error = acct_stack_growth(vma, size, grow); 22753af9e859SEric B Munson if (!error) { 22764128997bSMichel Lespinasse /* 22774128997bSMichel Lespinasse * vma_gap_update() doesn't support concurrent 22784128997bSMichel Lespinasse * updates, but we only hold a shared mmap_sem 22794128997bSMichel Lespinasse * lock here, so we need to protect against 22804128997bSMichel Lespinasse * concurrent vma expansions. 228112352d3cSKonstantin Khlebnikov * anon_vma_lock_write() doesn't help here, as 22824128997bSMichel Lespinasse * we don't guarantee that all growable vmas 22834128997bSMichel Lespinasse * in a mm share the same root anon vma. 22844128997bSMichel Lespinasse * So, we reuse mm->page_table_lock to guard 22854128997bSMichel Lespinasse * against concurrent vma expansions. 22864128997bSMichel Lespinasse */ 228709357814SOleg Nesterov spin_lock(&mm->page_table_lock); 228887e8827bSOleg Nesterov if (vma->vm_flags & VM_LOCKED) 228909357814SOleg Nesterov mm->locked_vm += grow; 229084638335SKonstantin Khlebnikov vm_stat_account(mm, vma->vm_flags, grow); 2291bf181b9fSMichel Lespinasse anon_vma_interval_tree_pre_update_vma(vma); 22921da177e4SLinus Torvalds vma->vm_end = address; 2293bf181b9fSMichel Lespinasse anon_vma_interval_tree_post_update_vma(vma); 2294d3737187SMichel Lespinasse if (vma->vm_next) 2295d3737187SMichel Lespinasse vma_gap_update(vma->vm_next); 2296d3737187SMichel Lespinasse else 22971be7107fSHugh Dickins mm->highest_vm_end = vm_end_gap(vma); 229809357814SOleg Nesterov spin_unlock(&mm->page_table_lock); 22994128997bSMichel Lespinasse 23003af9e859SEric B Munson perf_event_mmap(vma); 23013af9e859SEric B Munson } 23021da177e4SLinus Torvalds } 230342c36f63SHugh Dickins } 230412352d3cSKonstantin Khlebnikov anon_vma_unlock_write(vma->anon_vma); 23056d50e60cSDavid Rientjes khugepaged_enter_vma_merge(vma, vma->vm_flags); 230609357814SOleg Nesterov validate_mm(mm); 23071da177e4SLinus Torvalds return error; 23081da177e4SLinus Torvalds } 230946dea3d0SHugh Dickins #endif /* CONFIG_STACK_GROWSUP || CONFIG_IA64 */ 231046dea3d0SHugh Dickins 23111da177e4SLinus Torvalds /* 23121da177e4SLinus Torvalds * vma is the first one with address < vma->vm_start. Have to extend vma. 23131da177e4SLinus Torvalds */ 2314d05f3169SMichal Hocko int expand_downwards(struct vm_area_struct *vma, 2315b6a2fea3SOllie Wild unsigned long address) 23161da177e4SLinus Torvalds { 231709357814SOleg Nesterov struct mm_struct *mm = vma->vm_mm; 23181be7107fSHugh Dickins struct vm_area_struct *prev; 23191da177e4SLinus Torvalds int error; 23201da177e4SLinus Torvalds 23218869477aSEric Paris address &= PAGE_MASK; 2322e5467859SAl Viro error = security_mmap_addr(address); 23238869477aSEric Paris if (error) 23248869477aSEric Paris return error; 23258869477aSEric Paris 23261be7107fSHugh Dickins /* Enforce stack_guard_gap */ 23271be7107fSHugh Dickins prev = vma->vm_prev; 23281be7107fSHugh Dickins /* Check that both stack segments have the same anon_vma? */ 232932e4e6d5SOleg Nesterov if (prev && !(prev->vm_flags & VM_GROWSDOWN) && 233032e4e6d5SOleg Nesterov (prev->vm_flags & (VM_WRITE|VM_READ|VM_EXEC))) { 233132e4e6d5SOleg Nesterov if (address - prev->vm_end < stack_guard_gap) 233232e4e6d5SOleg Nesterov return -ENOMEM; 23331be7107fSHugh Dickins } 23341be7107fSHugh Dickins 233512352d3cSKonstantin Khlebnikov /* We must make sure the anon_vma is allocated. */ 233612352d3cSKonstantin Khlebnikov if (unlikely(anon_vma_prepare(vma))) 233712352d3cSKonstantin Khlebnikov return -ENOMEM; 23381da177e4SLinus Torvalds 23391da177e4SLinus Torvalds /* 23401da177e4SLinus Torvalds * vma->vm_start/vm_end cannot change under us because the caller 23411da177e4SLinus Torvalds * is required to hold the mmap_sem in read mode. We need the 23421da177e4SLinus Torvalds * anon_vma lock to serialize against concurrent expand_stacks. 23431da177e4SLinus Torvalds */ 234412352d3cSKonstantin Khlebnikov anon_vma_lock_write(vma->anon_vma); 23451da177e4SLinus Torvalds 23461da177e4SLinus Torvalds /* Somebody else might have raced and expanded it already */ 23471da177e4SLinus Torvalds if (address < vma->vm_start) { 23481da177e4SLinus Torvalds unsigned long size, grow; 23491da177e4SLinus Torvalds 23501da177e4SLinus Torvalds size = vma->vm_end - address; 23511da177e4SLinus Torvalds grow = (vma->vm_start - address) >> PAGE_SHIFT; 23521da177e4SLinus Torvalds 2353a626ca6aSLinus Torvalds error = -ENOMEM; 2354a626ca6aSLinus Torvalds if (grow <= vma->vm_pgoff) { 23551da177e4SLinus Torvalds error = acct_stack_growth(vma, size, grow); 23561da177e4SLinus Torvalds if (!error) { 23574128997bSMichel Lespinasse /* 23584128997bSMichel Lespinasse * vma_gap_update() doesn't support concurrent 23594128997bSMichel Lespinasse * updates, but we only hold a shared mmap_sem 23604128997bSMichel Lespinasse * lock here, so we need to protect against 23614128997bSMichel Lespinasse * concurrent vma expansions. 236212352d3cSKonstantin Khlebnikov * anon_vma_lock_write() doesn't help here, as 23634128997bSMichel Lespinasse * we don't guarantee that all growable vmas 23644128997bSMichel Lespinasse * in a mm share the same root anon vma. 23654128997bSMichel Lespinasse * So, we reuse mm->page_table_lock to guard 23664128997bSMichel Lespinasse * against concurrent vma expansions. 23674128997bSMichel Lespinasse */ 236809357814SOleg Nesterov spin_lock(&mm->page_table_lock); 236987e8827bSOleg Nesterov if (vma->vm_flags & VM_LOCKED) 237009357814SOleg Nesterov mm->locked_vm += grow; 237184638335SKonstantin Khlebnikov vm_stat_account(mm, vma->vm_flags, grow); 2372bf181b9fSMichel Lespinasse anon_vma_interval_tree_pre_update_vma(vma); 23731da177e4SLinus Torvalds vma->vm_start = address; 23741da177e4SLinus Torvalds vma->vm_pgoff -= grow; 2375bf181b9fSMichel Lespinasse anon_vma_interval_tree_post_update_vma(vma); 2376d3737187SMichel Lespinasse vma_gap_update(vma); 237709357814SOleg Nesterov spin_unlock(&mm->page_table_lock); 23784128997bSMichel Lespinasse 23793af9e859SEric B Munson perf_event_mmap(vma); 23801da177e4SLinus Torvalds } 23811da177e4SLinus Torvalds } 2382a626ca6aSLinus Torvalds } 238312352d3cSKonstantin Khlebnikov anon_vma_unlock_write(vma->anon_vma); 23846d50e60cSDavid Rientjes khugepaged_enter_vma_merge(vma, vma->vm_flags); 238509357814SOleg Nesterov validate_mm(mm); 23861da177e4SLinus Torvalds return error; 23871da177e4SLinus Torvalds } 23881da177e4SLinus Torvalds 23891be7107fSHugh Dickins /* enforced gap between the expanding stack and other mappings. */ 23901be7107fSHugh Dickins unsigned long stack_guard_gap = 256UL<<PAGE_SHIFT; 23911be7107fSHugh Dickins 23921be7107fSHugh Dickins static int __init cmdline_parse_stack_guard_gap(char *p) 23931be7107fSHugh Dickins { 23941be7107fSHugh Dickins unsigned long val; 23951be7107fSHugh Dickins char *endptr; 23961be7107fSHugh Dickins 23971be7107fSHugh Dickins val = simple_strtoul(p, &endptr, 10); 23981be7107fSHugh Dickins if (!*endptr) 23991be7107fSHugh Dickins stack_guard_gap = val << PAGE_SHIFT; 24001be7107fSHugh Dickins 24011be7107fSHugh Dickins return 0; 24021be7107fSHugh Dickins } 24031be7107fSHugh Dickins __setup("stack_guard_gap=", cmdline_parse_stack_guard_gap); 24041be7107fSHugh Dickins 2405b6a2fea3SOllie Wild #ifdef CONFIG_STACK_GROWSUP 2406b6a2fea3SOllie Wild int expand_stack(struct vm_area_struct *vma, unsigned long address) 2407b6a2fea3SOllie Wild { 2408b6a2fea3SOllie Wild return expand_upwards(vma, address); 2409b6a2fea3SOllie Wild } 2410b6a2fea3SOllie Wild 2411b6a2fea3SOllie Wild struct vm_area_struct * 2412b6a2fea3SOllie Wild find_extend_vma(struct mm_struct *mm, unsigned long addr) 2413b6a2fea3SOllie Wild { 2414b6a2fea3SOllie Wild struct vm_area_struct *vma, *prev; 2415b6a2fea3SOllie Wild 2416b6a2fea3SOllie Wild addr &= PAGE_MASK; 2417b6a2fea3SOllie Wild vma = find_vma_prev(mm, addr, &prev); 2418b6a2fea3SOllie Wild if (vma && (vma->vm_start <= addr)) 2419b6a2fea3SOllie Wild return vma; 24201c127185SDenys Vlasenko if (!prev || expand_stack(prev, addr)) 2421b6a2fea3SOllie Wild return NULL; 2422cea10a19SMichel Lespinasse if (prev->vm_flags & VM_LOCKED) 2423fc05f566SKirill A. Shutemov populate_vma_page_range(prev, addr, prev->vm_end, NULL); 2424b6a2fea3SOllie Wild return prev; 2425b6a2fea3SOllie Wild } 2426b6a2fea3SOllie Wild #else 2427b6a2fea3SOllie Wild int expand_stack(struct vm_area_struct *vma, unsigned long address) 2428b6a2fea3SOllie Wild { 2429b6a2fea3SOllie Wild return expand_downwards(vma, address); 2430b6a2fea3SOllie Wild } 2431b6a2fea3SOllie Wild 24321da177e4SLinus Torvalds struct vm_area_struct * 24331da177e4SLinus Torvalds find_extend_vma(struct mm_struct *mm, unsigned long addr) 24341da177e4SLinus Torvalds { 24351da177e4SLinus Torvalds struct vm_area_struct *vma; 24361da177e4SLinus Torvalds unsigned long start; 24371da177e4SLinus Torvalds 24381da177e4SLinus Torvalds addr &= PAGE_MASK; 24391da177e4SLinus Torvalds vma = find_vma(mm, addr); 24401da177e4SLinus Torvalds if (!vma) 24411da177e4SLinus Torvalds return NULL; 24421da177e4SLinus Torvalds if (vma->vm_start <= addr) 24431da177e4SLinus Torvalds return vma; 24441da177e4SLinus Torvalds if (!(vma->vm_flags & VM_GROWSDOWN)) 24451da177e4SLinus Torvalds return NULL; 24461da177e4SLinus Torvalds start = vma->vm_start; 24471da177e4SLinus Torvalds if (expand_stack(vma, addr)) 24481da177e4SLinus Torvalds return NULL; 2449cea10a19SMichel Lespinasse if (vma->vm_flags & VM_LOCKED) 2450fc05f566SKirill A. Shutemov populate_vma_page_range(vma, addr, start, NULL); 24511da177e4SLinus Torvalds return vma; 24521da177e4SLinus Torvalds } 24531da177e4SLinus Torvalds #endif 24541da177e4SLinus Torvalds 2455e1d6d01aSJesse Barnes EXPORT_SYMBOL_GPL(find_extend_vma); 2456e1d6d01aSJesse Barnes 24572c0b3814SHugh Dickins /* 24582c0b3814SHugh Dickins * Ok - we have the memory areas we should free on the vma list, 24592c0b3814SHugh Dickins * so release them, and do the vma updates. 24601da177e4SLinus Torvalds * 24612c0b3814SHugh Dickins * Called with the mm semaphore held. 24621da177e4SLinus Torvalds */ 24632c0b3814SHugh Dickins static void remove_vma_list(struct mm_struct *mm, struct vm_area_struct *vma) 24641da177e4SLinus Torvalds { 24654f74d2c8SLinus Torvalds unsigned long nr_accounted = 0; 24664f74d2c8SLinus Torvalds 2467365e9c87SHugh Dickins /* Update high watermark before we lower total_vm */ 2468365e9c87SHugh Dickins update_hiwater_vm(mm); 24692c0b3814SHugh Dickins do { 2470ab50b8edSHugh Dickins long nrpages = vma_pages(vma); 24711da177e4SLinus Torvalds 24724f74d2c8SLinus Torvalds if (vma->vm_flags & VM_ACCOUNT) 24734f74d2c8SLinus Torvalds nr_accounted += nrpages; 247484638335SKonstantin Khlebnikov vm_stat_account(mm, vma->vm_flags, -nrpages); 2475a8fb5618SHugh Dickins vma = remove_vma(vma); 2476146425a3SHugh Dickins } while (vma); 24774f74d2c8SLinus Torvalds vm_unacct_memory(nr_accounted); 24781da177e4SLinus Torvalds validate_mm(mm); 24791da177e4SLinus Torvalds } 24801da177e4SLinus Torvalds 24811da177e4SLinus Torvalds /* 24821da177e4SLinus Torvalds * Get rid of page table information in the indicated region. 24831da177e4SLinus Torvalds * 2484f10df686SPaolo 'Blaisorblade' Giarrusso * Called with the mm semaphore held. 24851da177e4SLinus Torvalds */ 24861da177e4SLinus Torvalds static void unmap_region(struct mm_struct *mm, 2487e0da382cSHugh Dickins struct vm_area_struct *vma, struct vm_area_struct *prev, 2488e0da382cSHugh Dickins unsigned long start, unsigned long end) 24891da177e4SLinus Torvalds { 2490e0da382cSHugh Dickins struct vm_area_struct *next = prev ? prev->vm_next : mm->mmap; 2491d16dfc55SPeter Zijlstra struct mmu_gather tlb; 24921da177e4SLinus Torvalds 24931da177e4SLinus Torvalds lru_add_drain(); 24942b047252SLinus Torvalds tlb_gather_mmu(&tlb, mm, start, end); 2495365e9c87SHugh Dickins update_hiwater_rss(mm); 24964f74d2c8SLinus Torvalds unmap_vmas(&tlb, vma, start, end); 2497d16dfc55SPeter Zijlstra free_pgtables(&tlb, vma, prev ? prev->vm_end : FIRST_USER_ADDRESS, 24986ee8630eSHugh Dickins next ? next->vm_start : USER_PGTABLES_CEILING); 2499d16dfc55SPeter Zijlstra tlb_finish_mmu(&tlb, start, end); 25001da177e4SLinus Torvalds } 25011da177e4SLinus Torvalds 25021da177e4SLinus Torvalds /* 25031da177e4SLinus Torvalds * Create a list of vma's touched by the unmap, removing them from the mm's 25041da177e4SLinus Torvalds * vma list as we go.. 25051da177e4SLinus Torvalds */ 25061da177e4SLinus Torvalds static void 25071da177e4SLinus Torvalds detach_vmas_to_be_unmapped(struct mm_struct *mm, struct vm_area_struct *vma, 25081da177e4SLinus Torvalds struct vm_area_struct *prev, unsigned long end) 25091da177e4SLinus Torvalds { 25101da177e4SLinus Torvalds struct vm_area_struct **insertion_point; 25111da177e4SLinus Torvalds struct vm_area_struct *tail_vma = NULL; 25121da177e4SLinus Torvalds 25131da177e4SLinus Torvalds insertion_point = (prev ? &prev->vm_next : &mm->mmap); 2514297c5eeeSLinus Torvalds vma->vm_prev = NULL; 25151da177e4SLinus Torvalds do { 2516d3737187SMichel Lespinasse vma_rb_erase(vma, &mm->mm_rb); 25171da177e4SLinus Torvalds mm->map_count--; 25181da177e4SLinus Torvalds tail_vma = vma; 25191da177e4SLinus Torvalds vma = vma->vm_next; 25201da177e4SLinus Torvalds } while (vma && vma->vm_start < end); 25211da177e4SLinus Torvalds *insertion_point = vma; 2522d3737187SMichel Lespinasse if (vma) { 2523297c5eeeSLinus Torvalds vma->vm_prev = prev; 2524d3737187SMichel Lespinasse vma_gap_update(vma); 2525d3737187SMichel Lespinasse } else 25261be7107fSHugh Dickins mm->highest_vm_end = prev ? vm_end_gap(prev) : 0; 25271da177e4SLinus Torvalds tail_vma->vm_next = NULL; 2528615d6e87SDavidlohr Bueso 2529615d6e87SDavidlohr Bueso /* Kill the cache */ 2530615d6e87SDavidlohr Bueso vmacache_invalidate(mm); 25311da177e4SLinus Torvalds } 25321da177e4SLinus Torvalds 25331da177e4SLinus Torvalds /* 2534def5efe0SDavid Rientjes * __split_vma() bypasses sysctl_max_map_count checking. We use this where it 2535def5efe0SDavid Rientjes * has already been checked or doesn't make sense to fail. 25361da177e4SLinus Torvalds */ 2537def5efe0SDavid Rientjes int __split_vma(struct mm_struct *mm, struct vm_area_struct *vma, 25381da177e4SLinus Torvalds unsigned long addr, int new_below) 25391da177e4SLinus Torvalds { 25401da177e4SLinus Torvalds struct vm_area_struct *new; 2541e3975891SChen Gang int err; 25421da177e4SLinus Torvalds 2543a5516438SAndi Kleen if (is_vm_hugetlb_page(vma) && (addr & 2544a5516438SAndi Kleen ~(huge_page_mask(hstate_vma(vma))))) 25451da177e4SLinus Torvalds return -EINVAL; 25461da177e4SLinus Torvalds 2547e94b1766SChristoph Lameter new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL); 25481da177e4SLinus Torvalds if (!new) 2549e3975891SChen Gang return -ENOMEM; 25501da177e4SLinus Torvalds 25511da177e4SLinus Torvalds /* most fields are the same, copy all, and then fixup */ 25521da177e4SLinus Torvalds *new = *vma; 25531da177e4SLinus Torvalds 25545beb4930SRik van Riel INIT_LIST_HEAD(&new->anon_vma_chain); 25555beb4930SRik van Riel 25561da177e4SLinus Torvalds if (new_below) 25571da177e4SLinus Torvalds new->vm_end = addr; 25581da177e4SLinus Torvalds else { 25591da177e4SLinus Torvalds new->vm_start = addr; 25601da177e4SLinus Torvalds new->vm_pgoff += ((addr - vma->vm_start) >> PAGE_SHIFT); 25611da177e4SLinus Torvalds } 25621da177e4SLinus Torvalds 2563ef0855d3SOleg Nesterov err = vma_dup_policy(vma, new); 2564ef0855d3SOleg Nesterov if (err) 25655beb4930SRik van Riel goto out_free_vma; 25661da177e4SLinus Torvalds 2567c4ea95d7SDaniel Forrest err = anon_vma_clone(new, vma); 2568c4ea95d7SDaniel Forrest if (err) 25695beb4930SRik van Riel goto out_free_mpol; 25705beb4930SRik van Riel 2571e9714acfSKonstantin Khlebnikov if (new->vm_file) 25721da177e4SLinus Torvalds get_file(new->vm_file); 25731da177e4SLinus Torvalds 25741da177e4SLinus Torvalds if (new->vm_ops && new->vm_ops->open) 25751da177e4SLinus Torvalds new->vm_ops->open(new); 25761da177e4SLinus Torvalds 25771da177e4SLinus Torvalds if (new_below) 25785beb4930SRik van Riel err = vma_adjust(vma, addr, vma->vm_end, vma->vm_pgoff + 25791da177e4SLinus Torvalds ((addr - new->vm_start) >> PAGE_SHIFT), new); 25801da177e4SLinus Torvalds else 25815beb4930SRik van Riel err = vma_adjust(vma, vma->vm_start, addr, vma->vm_pgoff, new); 25821da177e4SLinus Torvalds 25835beb4930SRik van Riel /* Success. */ 25845beb4930SRik van Riel if (!err) 25851da177e4SLinus Torvalds return 0; 25865beb4930SRik van Riel 25875beb4930SRik van Riel /* Clean everything up if vma_adjust failed. */ 258858927533SRik van Riel if (new->vm_ops && new->vm_ops->close) 25895beb4930SRik van Riel new->vm_ops->close(new); 2590e9714acfSKonstantin Khlebnikov if (new->vm_file) 25915beb4930SRik van Riel fput(new->vm_file); 25922aeadc30SAndrea Arcangeli unlink_anon_vmas(new); 25935beb4930SRik van Riel out_free_mpol: 2594ef0855d3SOleg Nesterov mpol_put(vma_policy(new)); 25955beb4930SRik van Riel out_free_vma: 25965beb4930SRik van Riel kmem_cache_free(vm_area_cachep, new); 25975beb4930SRik van Riel return err; 25981da177e4SLinus Torvalds } 25991da177e4SLinus Torvalds 2600659ace58SKOSAKI Motohiro /* 2601659ace58SKOSAKI Motohiro * Split a vma into two pieces at address 'addr', a new vma is allocated 2602659ace58SKOSAKI Motohiro * either for the first part or the tail. 2603659ace58SKOSAKI Motohiro */ 2604659ace58SKOSAKI Motohiro int split_vma(struct mm_struct *mm, struct vm_area_struct *vma, 2605659ace58SKOSAKI Motohiro unsigned long addr, int new_below) 2606659ace58SKOSAKI Motohiro { 2607659ace58SKOSAKI Motohiro if (mm->map_count >= sysctl_max_map_count) 2608659ace58SKOSAKI Motohiro return -ENOMEM; 2609659ace58SKOSAKI Motohiro 2610659ace58SKOSAKI Motohiro return __split_vma(mm, vma, addr, new_below); 2611659ace58SKOSAKI Motohiro } 2612659ace58SKOSAKI Motohiro 26131da177e4SLinus Torvalds /* Munmap is split into 2 main parts -- this part which finds 26141da177e4SLinus Torvalds * what needs doing, and the areas themselves, which do the 26151da177e4SLinus Torvalds * work. This now handles partial unmappings. 26161da177e4SLinus Torvalds * Jeremy Fitzhardinge <jeremy@goop.org> 26171da177e4SLinus Torvalds */ 2618897ab3e0SMike Rapoport int do_munmap(struct mm_struct *mm, unsigned long start, size_t len, 2619897ab3e0SMike Rapoport struct list_head *uf) 26201da177e4SLinus Torvalds { 26211da177e4SLinus Torvalds unsigned long end; 2622146425a3SHugh Dickins struct vm_area_struct *vma, *prev, *last; 26231da177e4SLinus Torvalds 2624de1741a1SAlexander Kuleshov if ((offset_in_page(start)) || start > TASK_SIZE || len > TASK_SIZE-start) 26251da177e4SLinus Torvalds return -EINVAL; 26261da177e4SLinus Torvalds 2627cc71aba3Svishnu.ps len = PAGE_ALIGN(len); 2628cc71aba3Svishnu.ps if (len == 0) 26291da177e4SLinus Torvalds return -EINVAL; 26301da177e4SLinus Torvalds 26311da177e4SLinus Torvalds /* Find the first overlapping VMA */ 26329be34c9dSLinus Torvalds vma = find_vma(mm, start); 2633146425a3SHugh Dickins if (!vma) 26341da177e4SLinus Torvalds return 0; 26359be34c9dSLinus Torvalds prev = vma->vm_prev; 2636146425a3SHugh Dickins /* we have start < vma->vm_end */ 26371da177e4SLinus Torvalds 26381da177e4SLinus Torvalds /* if it doesn't overlap, we have nothing.. */ 26391da177e4SLinus Torvalds end = start + len; 2640146425a3SHugh Dickins if (vma->vm_start >= end) 26411da177e4SLinus Torvalds return 0; 26421da177e4SLinus Torvalds 26431da177e4SLinus Torvalds /* 26441da177e4SLinus Torvalds * If we need to split any vma, do it now to save pain later. 26451da177e4SLinus Torvalds * 26461da177e4SLinus Torvalds * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially 26471da177e4SLinus Torvalds * unmapped vm_area_struct will remain in use: so lower split_vma 26481da177e4SLinus Torvalds * places tmp vma above, and higher split_vma places tmp vma below. 26491da177e4SLinus Torvalds */ 2650146425a3SHugh Dickins if (start > vma->vm_start) { 2651659ace58SKOSAKI Motohiro int error; 2652659ace58SKOSAKI Motohiro 2653659ace58SKOSAKI Motohiro /* 2654659ace58SKOSAKI Motohiro * Make sure that map_count on return from munmap() will 2655659ace58SKOSAKI Motohiro * not exceed its limit; but let map_count go just above 2656659ace58SKOSAKI Motohiro * its limit temporarily, to help free resources as expected. 2657659ace58SKOSAKI Motohiro */ 2658659ace58SKOSAKI Motohiro if (end < vma->vm_end && mm->map_count >= sysctl_max_map_count) 2659659ace58SKOSAKI Motohiro return -ENOMEM; 2660659ace58SKOSAKI Motohiro 2661659ace58SKOSAKI Motohiro error = __split_vma(mm, vma, start, 0); 26621da177e4SLinus Torvalds if (error) 26631da177e4SLinus Torvalds return error; 2664146425a3SHugh Dickins prev = vma; 26651da177e4SLinus Torvalds } 26661da177e4SLinus Torvalds 26671da177e4SLinus Torvalds /* Does it split the last one? */ 26681da177e4SLinus Torvalds last = find_vma(mm, end); 26691da177e4SLinus Torvalds if (last && end > last->vm_start) { 2670659ace58SKOSAKI Motohiro int error = __split_vma(mm, last, end, 1); 26711da177e4SLinus Torvalds if (error) 26721da177e4SLinus Torvalds return error; 26731da177e4SLinus Torvalds } 2674146425a3SHugh Dickins vma = prev ? prev->vm_next : mm->mmap; 26751da177e4SLinus Torvalds 26762376dd7cSAndrea Arcangeli if (unlikely(uf)) { 26772376dd7cSAndrea Arcangeli /* 26782376dd7cSAndrea Arcangeli * If userfaultfd_unmap_prep returns an error the vmas 26792376dd7cSAndrea Arcangeli * will remain splitted, but userland will get a 26802376dd7cSAndrea Arcangeli * highly unexpected error anyway. This is no 26812376dd7cSAndrea Arcangeli * different than the case where the first of the two 26822376dd7cSAndrea Arcangeli * __split_vma fails, but we don't undo the first 26832376dd7cSAndrea Arcangeli * split, despite we could. This is unlikely enough 26842376dd7cSAndrea Arcangeli * failure that it's not worth optimizing it for. 26852376dd7cSAndrea Arcangeli */ 26862376dd7cSAndrea Arcangeli int error = userfaultfd_unmap_prep(vma, start, end, uf); 26872376dd7cSAndrea Arcangeli if (error) 26882376dd7cSAndrea Arcangeli return error; 26892376dd7cSAndrea Arcangeli } 26902376dd7cSAndrea Arcangeli 26911da177e4SLinus Torvalds /* 2692ba470de4SRik van Riel * unlock any mlock()ed ranges before detaching vmas 2693ba470de4SRik van Riel */ 2694ba470de4SRik van Riel if (mm->locked_vm) { 2695ba470de4SRik van Riel struct vm_area_struct *tmp = vma; 2696ba470de4SRik van Riel while (tmp && tmp->vm_start < end) { 2697ba470de4SRik van Riel if (tmp->vm_flags & VM_LOCKED) { 2698ba470de4SRik van Riel mm->locked_vm -= vma_pages(tmp); 2699ba470de4SRik van Riel munlock_vma_pages_all(tmp); 2700ba470de4SRik van Riel } 2701ba470de4SRik van Riel tmp = tmp->vm_next; 2702ba470de4SRik van Riel } 2703ba470de4SRik van Riel } 2704ba470de4SRik van Riel 2705ba470de4SRik van Riel /* 27061da177e4SLinus Torvalds * Remove the vma's, and unmap the actual pages 27071da177e4SLinus Torvalds */ 2708146425a3SHugh Dickins detach_vmas_to_be_unmapped(mm, vma, prev, end); 2709146425a3SHugh Dickins unmap_region(mm, vma, prev, start, end); 27101da177e4SLinus Torvalds 27111de4fa14SDave Hansen arch_unmap(mm, vma, start, end); 27121de4fa14SDave Hansen 27131da177e4SLinus Torvalds /* Fix up all other VM information */ 27142c0b3814SHugh Dickins remove_vma_list(mm, vma); 27151da177e4SLinus Torvalds 27161da177e4SLinus Torvalds return 0; 27171da177e4SLinus Torvalds } 27181da177e4SLinus Torvalds 2719bfce281cSAl Viro int vm_munmap(unsigned long start, size_t len) 2720a46ef99dSLinus Torvalds { 2721a46ef99dSLinus Torvalds int ret; 2722bfce281cSAl Viro struct mm_struct *mm = current->mm; 2723897ab3e0SMike Rapoport LIST_HEAD(uf); 2724a46ef99dSLinus Torvalds 2725ae798783SMichal Hocko if (down_write_killable(&mm->mmap_sem)) 2726ae798783SMichal Hocko return -EINTR; 2727ae798783SMichal Hocko 2728897ab3e0SMike Rapoport ret = do_munmap(mm, start, len, &uf); 2729a46ef99dSLinus Torvalds up_write(&mm->mmap_sem); 2730897ab3e0SMike Rapoport userfaultfd_unmap_complete(mm, &uf); 2731a46ef99dSLinus Torvalds return ret; 2732a46ef99dSLinus Torvalds } 2733a46ef99dSLinus Torvalds EXPORT_SYMBOL(vm_munmap); 2734a46ef99dSLinus Torvalds 27356a6160a7SHeiko Carstens SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len) 27361da177e4SLinus Torvalds { 27371da177e4SLinus Torvalds profile_munmap(addr); 2738846b1a0fSMike Rapoport return vm_munmap(addr, len); 27391da177e4SLinus Torvalds } 27401da177e4SLinus Torvalds 2741c8d78c18SKirill A. Shutemov 2742c8d78c18SKirill A. Shutemov /* 2743c8d78c18SKirill A. Shutemov * Emulation of deprecated remap_file_pages() syscall. 2744c8d78c18SKirill A. Shutemov */ 2745c8d78c18SKirill A. Shutemov SYSCALL_DEFINE5(remap_file_pages, unsigned long, start, unsigned long, size, 2746c8d78c18SKirill A. Shutemov unsigned long, prot, unsigned long, pgoff, unsigned long, flags) 2747c8d78c18SKirill A. Shutemov { 2748c8d78c18SKirill A. Shutemov 2749c8d78c18SKirill A. Shutemov struct mm_struct *mm = current->mm; 2750c8d78c18SKirill A. Shutemov struct vm_area_struct *vma; 2751c8d78c18SKirill A. Shutemov unsigned long populate = 0; 2752c8d78c18SKirill A. Shutemov unsigned long ret = -EINVAL; 2753c8d78c18SKirill A. Shutemov struct file *file; 2754c8d78c18SKirill A. Shutemov 2755756a025fSJoe Perches pr_warn_once("%s (%d) uses deprecated remap_file_pages() syscall. See Documentation/vm/remap_file_pages.txt.\n", 2756c8d78c18SKirill A. Shutemov current->comm, current->pid); 2757c8d78c18SKirill A. Shutemov 2758c8d78c18SKirill A. Shutemov if (prot) 2759c8d78c18SKirill A. Shutemov return ret; 2760c8d78c18SKirill A. Shutemov start = start & PAGE_MASK; 2761c8d78c18SKirill A. Shutemov size = size & PAGE_MASK; 2762c8d78c18SKirill A. Shutemov 2763c8d78c18SKirill A. Shutemov if (start + size <= start) 2764c8d78c18SKirill A. Shutemov return ret; 2765c8d78c18SKirill A. Shutemov 2766c8d78c18SKirill A. Shutemov /* Does pgoff wrap? */ 2767c8d78c18SKirill A. Shutemov if (pgoff + (size >> PAGE_SHIFT) < pgoff) 2768c8d78c18SKirill A. Shutemov return ret; 2769c8d78c18SKirill A. Shutemov 2770dc0ef0dfSMichal Hocko if (down_write_killable(&mm->mmap_sem)) 2771dc0ef0dfSMichal Hocko return -EINTR; 2772dc0ef0dfSMichal Hocko 2773c8d78c18SKirill A. Shutemov vma = find_vma(mm, start); 2774c8d78c18SKirill A. Shutemov 2775c8d78c18SKirill A. Shutemov if (!vma || !(vma->vm_flags & VM_SHARED)) 2776c8d78c18SKirill A. Shutemov goto out; 2777c8d78c18SKirill A. Shutemov 277848f7df32SKirill A. Shutemov if (start < vma->vm_start) 2779c8d78c18SKirill A. Shutemov goto out; 2780c8d78c18SKirill A. Shutemov 278148f7df32SKirill A. Shutemov if (start + size > vma->vm_end) { 278248f7df32SKirill A. Shutemov struct vm_area_struct *next; 278348f7df32SKirill A. Shutemov 278448f7df32SKirill A. Shutemov for (next = vma->vm_next; next; next = next->vm_next) { 278548f7df32SKirill A. Shutemov /* hole between vmas ? */ 278648f7df32SKirill A. Shutemov if (next->vm_start != next->vm_prev->vm_end) 278748f7df32SKirill A. Shutemov goto out; 278848f7df32SKirill A. Shutemov 278948f7df32SKirill A. Shutemov if (next->vm_file != vma->vm_file) 279048f7df32SKirill A. Shutemov goto out; 279148f7df32SKirill A. Shutemov 279248f7df32SKirill A. Shutemov if (next->vm_flags != vma->vm_flags) 279348f7df32SKirill A. Shutemov goto out; 279448f7df32SKirill A. Shutemov 279548f7df32SKirill A. Shutemov if (start + size <= next->vm_end) 279648f7df32SKirill A. Shutemov break; 279748f7df32SKirill A. Shutemov } 279848f7df32SKirill A. Shutemov 279948f7df32SKirill A. Shutemov if (!next) 2800c8d78c18SKirill A. Shutemov goto out; 2801c8d78c18SKirill A. Shutemov } 2802c8d78c18SKirill A. Shutemov 2803c8d78c18SKirill A. Shutemov prot |= vma->vm_flags & VM_READ ? PROT_READ : 0; 2804c8d78c18SKirill A. Shutemov prot |= vma->vm_flags & VM_WRITE ? PROT_WRITE : 0; 2805c8d78c18SKirill A. Shutemov prot |= vma->vm_flags & VM_EXEC ? PROT_EXEC : 0; 2806c8d78c18SKirill A. Shutemov 2807c8d78c18SKirill A. Shutemov flags &= MAP_NONBLOCK; 2808c8d78c18SKirill A. Shutemov flags |= MAP_SHARED | MAP_FIXED | MAP_POPULATE; 2809c8d78c18SKirill A. Shutemov if (vma->vm_flags & VM_LOCKED) { 281048f7df32SKirill A. Shutemov struct vm_area_struct *tmp; 2811c8d78c18SKirill A. Shutemov flags |= MAP_LOCKED; 281248f7df32SKirill A. Shutemov 2813c8d78c18SKirill A. Shutemov /* drop PG_Mlocked flag for over-mapped range */ 281448f7df32SKirill A. Shutemov for (tmp = vma; tmp->vm_start >= start + size; 281548f7df32SKirill A. Shutemov tmp = tmp->vm_next) { 28169a73f61bSKirill A. Shutemov /* 28179a73f61bSKirill A. Shutemov * Split pmd and munlock page on the border 28189a73f61bSKirill A. Shutemov * of the range. 28199a73f61bSKirill A. Shutemov */ 28209a73f61bSKirill A. Shutemov vma_adjust_trans_huge(tmp, start, start + size, 0); 28219a73f61bSKirill A. Shutemov 282248f7df32SKirill A. Shutemov munlock_vma_pages_range(tmp, 282348f7df32SKirill A. Shutemov max(tmp->vm_start, start), 282448f7df32SKirill A. Shutemov min(tmp->vm_end, start + size)); 282548f7df32SKirill A. Shutemov } 2826c8d78c18SKirill A. Shutemov } 2827c8d78c18SKirill A. Shutemov 2828c8d78c18SKirill A. Shutemov file = get_file(vma->vm_file); 2829c8d78c18SKirill A. Shutemov ret = do_mmap_pgoff(vma->vm_file, start, size, 2830897ab3e0SMike Rapoport prot, flags, pgoff, &populate, NULL); 2831c8d78c18SKirill A. Shutemov fput(file); 2832c8d78c18SKirill A. Shutemov out: 2833c8d78c18SKirill A. Shutemov up_write(&mm->mmap_sem); 2834c8d78c18SKirill A. Shutemov if (populate) 2835c8d78c18SKirill A. Shutemov mm_populate(ret, populate); 2836c8d78c18SKirill A. Shutemov if (!IS_ERR_VALUE(ret)) 2837c8d78c18SKirill A. Shutemov ret = 0; 2838c8d78c18SKirill A. Shutemov return ret; 2839c8d78c18SKirill A. Shutemov } 2840c8d78c18SKirill A. Shutemov 28411da177e4SLinus Torvalds static inline void verify_mm_writelocked(struct mm_struct *mm) 28421da177e4SLinus Torvalds { 2843a241ec65SPaul E. McKenney #ifdef CONFIG_DEBUG_VM 28441da177e4SLinus Torvalds if (unlikely(down_read_trylock(&mm->mmap_sem))) { 28451da177e4SLinus Torvalds WARN_ON(1); 28461da177e4SLinus Torvalds up_read(&mm->mmap_sem); 28471da177e4SLinus Torvalds } 28481da177e4SLinus Torvalds #endif 28491da177e4SLinus Torvalds } 28501da177e4SLinus Torvalds 28511da177e4SLinus Torvalds /* 28521da177e4SLinus Torvalds * this is really a simplified "do_mmap". it only handles 28531da177e4SLinus Torvalds * anonymous maps. eventually we may be able to do some 28541da177e4SLinus Torvalds * brk-specific accounting here. 28551da177e4SLinus Torvalds */ 2856897ab3e0SMike Rapoport static int do_brk_flags(unsigned long addr, unsigned long request, unsigned long flags, struct list_head *uf) 28571da177e4SLinus Torvalds { 28581da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 28591da177e4SLinus Torvalds struct vm_area_struct *vma, *prev; 286016e72e9bSDenys Vlasenko unsigned long len; 28611da177e4SLinus Torvalds struct rb_node **rb_link, *rb_parent; 28621da177e4SLinus Torvalds pgoff_t pgoff = addr >> PAGE_SHIFT; 28633a459756SKirill Korotaev int error; 28641da177e4SLinus Torvalds 2865ba093a6dSKees Cook len = PAGE_ALIGN(request); 2866ba093a6dSKees Cook if (len < request) 2867ba093a6dSKees Cook return -ENOMEM; 28681da177e4SLinus Torvalds if (!len) 28695d22fc25SLinus Torvalds return 0; 28701da177e4SLinus Torvalds 287116e72e9bSDenys Vlasenko /* Until we need other flags, refuse anything except VM_EXEC. */ 287216e72e9bSDenys Vlasenko if ((flags & (~VM_EXEC)) != 0) 287316e72e9bSDenys Vlasenko return -EINVAL; 287416e72e9bSDenys Vlasenko flags |= VM_DATA_DEFAULT_FLAGS | VM_ACCOUNT | mm->def_flags; 28753a459756SKirill Korotaev 28762c6a1016SAl Viro error = get_unmapped_area(NULL, addr, len, 0, MAP_FIXED); 2877de1741a1SAlexander Kuleshov if (offset_in_page(error)) 28783a459756SKirill Korotaev return error; 28793a459756SKirill Korotaev 2880363ee17fSDavidlohr Bueso error = mlock_future_check(mm, mm->def_flags, len); 2881363ee17fSDavidlohr Bueso if (error) 2882363ee17fSDavidlohr Bueso return error; 28831da177e4SLinus Torvalds 28841da177e4SLinus Torvalds /* 28851da177e4SLinus Torvalds * mm->mmap_sem is required to protect against another thread 28861da177e4SLinus Torvalds * changing the mappings in case we sleep. 28871da177e4SLinus Torvalds */ 28881da177e4SLinus Torvalds verify_mm_writelocked(mm); 28891da177e4SLinus Torvalds 28901da177e4SLinus Torvalds /* 28911da177e4SLinus Torvalds * Clear old maps. this also does some error checking for us 28921da177e4SLinus Torvalds */ 28939fcd1457SRasmus Villemoes while (find_vma_links(mm, addr, addr + len, &prev, &rb_link, 28949fcd1457SRasmus Villemoes &rb_parent)) { 2895897ab3e0SMike Rapoport if (do_munmap(mm, addr, len, uf)) 28961da177e4SLinus Torvalds return -ENOMEM; 28971da177e4SLinus Torvalds } 28981da177e4SLinus Torvalds 28991da177e4SLinus Torvalds /* Check against address space limits *after* clearing old maps... */ 290084638335SKonstantin Khlebnikov if (!may_expand_vm(mm, flags, len >> PAGE_SHIFT)) 29011da177e4SLinus Torvalds return -ENOMEM; 29021da177e4SLinus Torvalds 29031da177e4SLinus Torvalds if (mm->map_count > sysctl_max_map_count) 29041da177e4SLinus Torvalds return -ENOMEM; 29051da177e4SLinus Torvalds 2906191c5424SAl Viro if (security_vm_enough_memory_mm(mm, len >> PAGE_SHIFT)) 29071da177e4SLinus Torvalds return -ENOMEM; 29081da177e4SLinus Torvalds 29091da177e4SLinus Torvalds /* Can we just expand an old private anonymous mapping? */ 2910ba470de4SRik van Riel vma = vma_merge(mm, prev, addr, addr + len, flags, 291119a809afSAndrea Arcangeli NULL, NULL, pgoff, NULL, NULL_VM_UFFD_CTX); 2912ba470de4SRik van Riel if (vma) 29131da177e4SLinus Torvalds goto out; 29141da177e4SLinus Torvalds 29151da177e4SLinus Torvalds /* 29161da177e4SLinus Torvalds * create a vma struct for an anonymous mapping 29171da177e4SLinus Torvalds */ 2918c5e3b83eSPekka Enberg vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL); 29191da177e4SLinus Torvalds if (!vma) { 29201da177e4SLinus Torvalds vm_unacct_memory(len >> PAGE_SHIFT); 29211da177e4SLinus Torvalds return -ENOMEM; 29221da177e4SLinus Torvalds } 29231da177e4SLinus Torvalds 29245beb4930SRik van Riel INIT_LIST_HEAD(&vma->anon_vma_chain); 29251da177e4SLinus Torvalds vma->vm_mm = mm; 29261da177e4SLinus Torvalds vma->vm_start = addr; 29271da177e4SLinus Torvalds vma->vm_end = addr + len; 29281da177e4SLinus Torvalds vma->vm_pgoff = pgoff; 29291da177e4SLinus Torvalds vma->vm_flags = flags; 29303ed75eb8SColy Li vma->vm_page_prot = vm_get_page_prot(flags); 29311da177e4SLinus Torvalds vma_link(mm, vma, prev, rb_link, rb_parent); 29321da177e4SLinus Torvalds out: 29333af9e859SEric B Munson perf_event_mmap(vma); 29341da177e4SLinus Torvalds mm->total_vm += len >> PAGE_SHIFT; 293584638335SKonstantin Khlebnikov mm->data_vm += len >> PAGE_SHIFT; 2936128557ffSMichel Lespinasse if (flags & VM_LOCKED) 2937ba470de4SRik van Riel mm->locked_vm += (len >> PAGE_SHIFT); 2938d9104d1cSCyrill Gorcunov vma->vm_flags |= VM_SOFTDIRTY; 29395d22fc25SLinus Torvalds return 0; 29401da177e4SLinus Torvalds } 29411da177e4SLinus Torvalds 2942897ab3e0SMike Rapoport static int do_brk(unsigned long addr, unsigned long len, struct list_head *uf) 294316e72e9bSDenys Vlasenko { 2944897ab3e0SMike Rapoport return do_brk_flags(addr, len, 0, uf); 294516e72e9bSDenys Vlasenko } 294616e72e9bSDenys Vlasenko 294716e72e9bSDenys Vlasenko int vm_brk_flags(unsigned long addr, unsigned long len, unsigned long flags) 2948e4eb1ff6SLinus Torvalds { 2949e4eb1ff6SLinus Torvalds struct mm_struct *mm = current->mm; 29505d22fc25SLinus Torvalds int ret; 2951128557ffSMichel Lespinasse bool populate; 2952897ab3e0SMike Rapoport LIST_HEAD(uf); 2953e4eb1ff6SLinus Torvalds 29542d6c9282SMichal Hocko if (down_write_killable(&mm->mmap_sem)) 29552d6c9282SMichal Hocko return -EINTR; 29562d6c9282SMichal Hocko 2957897ab3e0SMike Rapoport ret = do_brk_flags(addr, len, flags, &uf); 2958128557ffSMichel Lespinasse populate = ((mm->def_flags & VM_LOCKED) != 0); 2959e4eb1ff6SLinus Torvalds up_write(&mm->mmap_sem); 2960897ab3e0SMike Rapoport userfaultfd_unmap_complete(mm, &uf); 29615d22fc25SLinus Torvalds if (populate && !ret) 2962128557ffSMichel Lespinasse mm_populate(addr, len); 2963e4eb1ff6SLinus Torvalds return ret; 2964e4eb1ff6SLinus Torvalds } 296516e72e9bSDenys Vlasenko EXPORT_SYMBOL(vm_brk_flags); 296616e72e9bSDenys Vlasenko 296716e72e9bSDenys Vlasenko int vm_brk(unsigned long addr, unsigned long len) 296816e72e9bSDenys Vlasenko { 296916e72e9bSDenys Vlasenko return vm_brk_flags(addr, len, 0); 297016e72e9bSDenys Vlasenko } 2971e4eb1ff6SLinus Torvalds EXPORT_SYMBOL(vm_brk); 29721da177e4SLinus Torvalds 29731da177e4SLinus Torvalds /* Release all mmaps. */ 29741da177e4SLinus Torvalds void exit_mmap(struct mm_struct *mm) 29751da177e4SLinus Torvalds { 2976d16dfc55SPeter Zijlstra struct mmu_gather tlb; 2977ba470de4SRik van Riel struct vm_area_struct *vma; 29781da177e4SLinus Torvalds unsigned long nr_accounted = 0; 29791da177e4SLinus Torvalds 2980d6dd61c8SJeremy Fitzhardinge /* mm's last user has gone, and its about to be pulled down */ 2981cddb8a5cSAndrea Arcangeli mmu_notifier_release(mm); 2982d6dd61c8SJeremy Fitzhardinge 2983ba470de4SRik van Riel if (mm->locked_vm) { 2984ba470de4SRik van Riel vma = mm->mmap; 2985ba470de4SRik van Riel while (vma) { 2986ba470de4SRik van Riel if (vma->vm_flags & VM_LOCKED) 2987ba470de4SRik van Riel munlock_vma_pages_all(vma); 2988ba470de4SRik van Riel vma = vma->vm_next; 2989ba470de4SRik van Riel } 2990ba470de4SRik van Riel } 29919480c53eSJeremy Fitzhardinge 29929480c53eSJeremy Fitzhardinge arch_exit_mmap(mm); 29939480c53eSJeremy Fitzhardinge 2994ba470de4SRik van Riel vma = mm->mmap; 29959480c53eSJeremy Fitzhardinge if (!vma) /* Can happen if dup_mmap() received an OOM */ 29969480c53eSJeremy Fitzhardinge return; 29979480c53eSJeremy Fitzhardinge 29981da177e4SLinus Torvalds lru_add_drain(); 29991da177e4SLinus Torvalds flush_cache_mm(mm); 30002b047252SLinus Torvalds tlb_gather_mmu(&tlb, mm, 0, -1); 3001901608d9SOleg Nesterov /* update_hiwater_rss(mm) here? but nobody should be looking */ 3002e0da382cSHugh Dickins /* Use -1 here to ensure all VMAs in the mm are unmapped */ 30034f74d2c8SLinus Torvalds unmap_vmas(&tlb, vma, 0, -1); 30049ba69294SHugh Dickins 3005*21292580SAndrea Arcangeli set_bit(MMF_OOM_SKIP, &mm->flags); 3006*21292580SAndrea Arcangeli if (unlikely(tsk_is_oom_victim(current))) { 3007*21292580SAndrea Arcangeli /* 3008*21292580SAndrea Arcangeli * Wait for oom_reap_task() to stop working on this 3009*21292580SAndrea Arcangeli * mm. Because MMF_OOM_SKIP is already set before 3010*21292580SAndrea Arcangeli * calling down_read(), oom_reap_task() will not run 3011*21292580SAndrea Arcangeli * on this "mm" post up_write(). 3012*21292580SAndrea Arcangeli * 3013*21292580SAndrea Arcangeli * tsk_is_oom_victim() cannot be set from under us 3014*21292580SAndrea Arcangeli * either because current->mm is already set to NULL 3015*21292580SAndrea Arcangeli * under task_lock before calling mmput and oom_mm is 3016*21292580SAndrea Arcangeli * set not NULL by the OOM killer only if current->mm 3017*21292580SAndrea Arcangeli * is found not NULL while holding the task_lock. 3018*21292580SAndrea Arcangeli */ 3019*21292580SAndrea Arcangeli down_write(&mm->mmap_sem); 3020*21292580SAndrea Arcangeli up_write(&mm->mmap_sem); 3021*21292580SAndrea Arcangeli } 30226ee8630eSHugh Dickins free_pgtables(&tlb, vma, FIRST_USER_ADDRESS, USER_PGTABLES_CEILING); 3023853f5e26SAl Viro tlb_finish_mmu(&tlb, 0, -1); 30241da177e4SLinus Torvalds 30251da177e4SLinus Torvalds /* 30268f4f8c16SHugh Dickins * Walk the list again, actually closing and freeing it, 30278f4f8c16SHugh Dickins * with preemption enabled, without holding any MM locks. 30281da177e4SLinus Torvalds */ 30294f74d2c8SLinus Torvalds while (vma) { 30304f74d2c8SLinus Torvalds if (vma->vm_flags & VM_ACCOUNT) 30314f74d2c8SLinus Torvalds nr_accounted += vma_pages(vma); 3032a8fb5618SHugh Dickins vma = remove_vma(vma); 30334f74d2c8SLinus Torvalds } 30344f74d2c8SLinus Torvalds vm_unacct_memory(nr_accounted); 30351da177e4SLinus Torvalds } 30361da177e4SLinus Torvalds 30371da177e4SLinus Torvalds /* Insert vm structure into process list sorted by address 30381da177e4SLinus Torvalds * and into the inode's i_mmap tree. If vm_file is non-NULL 3039c8c06efaSDavidlohr Bueso * then i_mmap_rwsem is taken here. 30401da177e4SLinus Torvalds */ 30411da177e4SLinus Torvalds int insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma) 30421da177e4SLinus Torvalds { 30436597d783SHugh Dickins struct vm_area_struct *prev; 30441da177e4SLinus Torvalds struct rb_node **rb_link, *rb_parent; 30451da177e4SLinus Torvalds 3046c9d13f5fSChen Gang if (find_vma_links(mm, vma->vm_start, vma->vm_end, 3047c9d13f5fSChen Gang &prev, &rb_link, &rb_parent)) 3048c9d13f5fSChen Gang return -ENOMEM; 3049c9d13f5fSChen Gang if ((vma->vm_flags & VM_ACCOUNT) && 3050c9d13f5fSChen Gang security_vm_enough_memory_mm(mm, vma_pages(vma))) 3051c9d13f5fSChen Gang return -ENOMEM; 3052c9d13f5fSChen Gang 30531da177e4SLinus Torvalds /* 30541da177e4SLinus Torvalds * The vm_pgoff of a purely anonymous vma should be irrelevant 30551da177e4SLinus Torvalds * until its first write fault, when page's anon_vma and index 30561da177e4SLinus Torvalds * are set. But now set the vm_pgoff it will almost certainly 30571da177e4SLinus Torvalds * end up with (unless mremap moves it elsewhere before that 30581da177e4SLinus Torvalds * first wfault), so /proc/pid/maps tells a consistent story. 30591da177e4SLinus Torvalds * 30601da177e4SLinus Torvalds * By setting it to reflect the virtual start address of the 30611da177e4SLinus Torvalds * vma, merges and splits can happen in a seamless way, just 30621da177e4SLinus Torvalds * using the existing file pgoff checks and manipulations. 30631da177e4SLinus Torvalds * Similarly in do_mmap_pgoff and in do_brk. 30641da177e4SLinus Torvalds */ 30658a9cc3b5SOleg Nesterov if (vma_is_anonymous(vma)) { 30661da177e4SLinus Torvalds BUG_ON(vma->anon_vma); 30671da177e4SLinus Torvalds vma->vm_pgoff = vma->vm_start >> PAGE_SHIFT; 30681da177e4SLinus Torvalds } 30692b144498SSrikar Dronamraju 30701da177e4SLinus Torvalds vma_link(mm, vma, prev, rb_link, rb_parent); 30711da177e4SLinus Torvalds return 0; 30721da177e4SLinus Torvalds } 30731da177e4SLinus Torvalds 30741da177e4SLinus Torvalds /* 30751da177e4SLinus Torvalds * Copy the vma structure to a new location in the same mm, 30761da177e4SLinus Torvalds * prior to moving page table entries, to effect an mremap move. 30771da177e4SLinus Torvalds */ 30781da177e4SLinus Torvalds struct vm_area_struct *copy_vma(struct vm_area_struct **vmap, 307938a76013SMichel Lespinasse unsigned long addr, unsigned long len, pgoff_t pgoff, 308038a76013SMichel Lespinasse bool *need_rmap_locks) 30811da177e4SLinus Torvalds { 30821da177e4SLinus Torvalds struct vm_area_struct *vma = *vmap; 30831da177e4SLinus Torvalds unsigned long vma_start = vma->vm_start; 30841da177e4SLinus Torvalds struct mm_struct *mm = vma->vm_mm; 30851da177e4SLinus Torvalds struct vm_area_struct *new_vma, *prev; 30861da177e4SLinus Torvalds struct rb_node **rb_link, *rb_parent; 3087948f017bSAndrea Arcangeli bool faulted_in_anon_vma = true; 30881da177e4SLinus Torvalds 30891da177e4SLinus Torvalds /* 30901da177e4SLinus Torvalds * If anonymous vma has not yet been faulted, update new pgoff 30911da177e4SLinus Torvalds * to match new location, to increase its chance of merging. 30921da177e4SLinus Torvalds */ 3093ce75799bSOleg Nesterov if (unlikely(vma_is_anonymous(vma) && !vma->anon_vma)) { 30941da177e4SLinus Torvalds pgoff = addr >> PAGE_SHIFT; 3095948f017bSAndrea Arcangeli faulted_in_anon_vma = false; 3096948f017bSAndrea Arcangeli } 30971da177e4SLinus Torvalds 30986597d783SHugh Dickins if (find_vma_links(mm, addr, addr + len, &prev, &rb_link, &rb_parent)) 30996597d783SHugh Dickins return NULL; /* should never get here */ 31001da177e4SLinus Torvalds new_vma = vma_merge(mm, prev, addr, addr + len, vma->vm_flags, 310119a809afSAndrea Arcangeli vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma), 310219a809afSAndrea Arcangeli vma->vm_userfaultfd_ctx); 31031da177e4SLinus Torvalds if (new_vma) { 31041da177e4SLinus Torvalds /* 31051da177e4SLinus Torvalds * Source vma may have been merged into new_vma 31061da177e4SLinus Torvalds */ 3107948f017bSAndrea Arcangeli if (unlikely(vma_start >= new_vma->vm_start && 3108948f017bSAndrea Arcangeli vma_start < new_vma->vm_end)) { 3109948f017bSAndrea Arcangeli /* 3110948f017bSAndrea Arcangeli * The only way we can get a vma_merge with 3111948f017bSAndrea Arcangeli * self during an mremap is if the vma hasn't 3112948f017bSAndrea Arcangeli * been faulted in yet and we were allowed to 3113948f017bSAndrea Arcangeli * reset the dst vma->vm_pgoff to the 3114948f017bSAndrea Arcangeli * destination address of the mremap to allow 3115948f017bSAndrea Arcangeli * the merge to happen. mremap must change the 3116948f017bSAndrea Arcangeli * vm_pgoff linearity between src and dst vmas 3117948f017bSAndrea Arcangeli * (in turn preventing a vma_merge) to be 3118948f017bSAndrea Arcangeli * safe. It is only safe to keep the vm_pgoff 3119948f017bSAndrea Arcangeli * linear if there are no pages mapped yet. 3120948f017bSAndrea Arcangeli */ 312181d1b09cSSasha Levin VM_BUG_ON_VMA(faulted_in_anon_vma, new_vma); 312238a76013SMichel Lespinasse *vmap = vma = new_vma; 3123108d6642SMichel Lespinasse } 312438a76013SMichel Lespinasse *need_rmap_locks = (new_vma->vm_pgoff <= vma->vm_pgoff); 31251da177e4SLinus Torvalds } else { 3126e94b1766SChristoph Lameter new_vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL); 3127e3975891SChen Gang if (!new_vma) 3128e3975891SChen Gang goto out; 31291da177e4SLinus Torvalds *new_vma = *vma; 31301da177e4SLinus Torvalds new_vma->vm_start = addr; 31311da177e4SLinus Torvalds new_vma->vm_end = addr + len; 31321da177e4SLinus Torvalds new_vma->vm_pgoff = pgoff; 3133ef0855d3SOleg Nesterov if (vma_dup_policy(vma, new_vma)) 3134523d4e20SMichel Lespinasse goto out_free_vma; 3135523d4e20SMichel Lespinasse INIT_LIST_HEAD(&new_vma->anon_vma_chain); 3136523d4e20SMichel Lespinasse if (anon_vma_clone(new_vma, vma)) 3137523d4e20SMichel Lespinasse goto out_free_mempol; 3138e9714acfSKonstantin Khlebnikov if (new_vma->vm_file) 31391da177e4SLinus Torvalds get_file(new_vma->vm_file); 31401da177e4SLinus Torvalds if (new_vma->vm_ops && new_vma->vm_ops->open) 31411da177e4SLinus Torvalds new_vma->vm_ops->open(new_vma); 31421da177e4SLinus Torvalds vma_link(mm, new_vma, prev, rb_link, rb_parent); 314338a76013SMichel Lespinasse *need_rmap_locks = false; 31441da177e4SLinus Torvalds } 31451da177e4SLinus Torvalds return new_vma; 31465beb4930SRik van Riel 31475beb4930SRik van Riel out_free_mempol: 3148ef0855d3SOleg Nesterov mpol_put(vma_policy(new_vma)); 31495beb4930SRik van Riel out_free_vma: 31505beb4930SRik van Riel kmem_cache_free(vm_area_cachep, new_vma); 3151e3975891SChen Gang out: 31525beb4930SRik van Riel return NULL; 31531da177e4SLinus Torvalds } 3154119f657cSakpm@osdl.org 3155119f657cSakpm@osdl.org /* 3156119f657cSakpm@osdl.org * Return true if the calling process may expand its vm space by the passed 3157119f657cSakpm@osdl.org * number of pages 3158119f657cSakpm@osdl.org */ 315984638335SKonstantin Khlebnikov bool may_expand_vm(struct mm_struct *mm, vm_flags_t flags, unsigned long npages) 3160119f657cSakpm@osdl.org { 316184638335SKonstantin Khlebnikov if (mm->total_vm + npages > rlimit(RLIMIT_AS) >> PAGE_SHIFT) 316284638335SKonstantin Khlebnikov return false; 3163119f657cSakpm@osdl.org 3164d977d56cSKonstantin Khlebnikov if (is_data_mapping(flags) && 3165d977d56cSKonstantin Khlebnikov mm->data_vm + npages > rlimit(RLIMIT_DATA) >> PAGE_SHIFT) { 3166f4fcd558SKonstantin Khlebnikov /* Workaround for Valgrind */ 3167f4fcd558SKonstantin Khlebnikov if (rlimit(RLIMIT_DATA) == 0 && 3168f4fcd558SKonstantin Khlebnikov mm->data_vm + npages <= rlimit_max(RLIMIT_DATA) >> PAGE_SHIFT) 3169f4fcd558SKonstantin Khlebnikov return true; 3170f4fcd558SKonstantin Khlebnikov if (!ignore_rlimit_data) { 3171f4fcd558SKonstantin Khlebnikov pr_warn_once("%s (%d): VmData %lu exceed data ulimit %lu. Update limits or use boot option ignore_rlimit_data.\n", 3172d977d56cSKonstantin Khlebnikov current->comm, current->pid, 3173d977d56cSKonstantin Khlebnikov (mm->data_vm + npages) << PAGE_SHIFT, 3174d977d56cSKonstantin Khlebnikov rlimit(RLIMIT_DATA)); 3175d977d56cSKonstantin Khlebnikov return false; 3176d977d56cSKonstantin Khlebnikov } 3177f4fcd558SKonstantin Khlebnikov } 3178119f657cSakpm@osdl.org 317984638335SKonstantin Khlebnikov return true; 318084638335SKonstantin Khlebnikov } 318184638335SKonstantin Khlebnikov 318284638335SKonstantin Khlebnikov void vm_stat_account(struct mm_struct *mm, vm_flags_t flags, long npages) 318384638335SKonstantin Khlebnikov { 318484638335SKonstantin Khlebnikov mm->total_vm += npages; 318584638335SKonstantin Khlebnikov 3186d977d56cSKonstantin Khlebnikov if (is_exec_mapping(flags)) 318784638335SKonstantin Khlebnikov mm->exec_vm += npages; 3188d977d56cSKonstantin Khlebnikov else if (is_stack_mapping(flags)) 318984638335SKonstantin Khlebnikov mm->stack_vm += npages; 3190d977d56cSKonstantin Khlebnikov else if (is_data_mapping(flags)) 319184638335SKonstantin Khlebnikov mm->data_vm += npages; 3192119f657cSakpm@osdl.org } 3193fa5dc22fSRoland McGrath 319411bac800SDave Jiang static int special_mapping_fault(struct vm_fault *vmf); 3195a62c34bdSAndy Lutomirski 3196a62c34bdSAndy Lutomirski /* 3197a62c34bdSAndy Lutomirski * Having a close hook prevents vma merging regardless of flags. 3198a62c34bdSAndy Lutomirski */ 3199a62c34bdSAndy Lutomirski static void special_mapping_close(struct vm_area_struct *vma) 3200a62c34bdSAndy Lutomirski { 3201a62c34bdSAndy Lutomirski } 3202a62c34bdSAndy Lutomirski 3203a62c34bdSAndy Lutomirski static const char *special_mapping_name(struct vm_area_struct *vma) 3204a62c34bdSAndy Lutomirski { 3205a62c34bdSAndy Lutomirski return ((struct vm_special_mapping *)vma->vm_private_data)->name; 3206a62c34bdSAndy Lutomirski } 3207a62c34bdSAndy Lutomirski 3208b059a453SDmitry Safonov static int special_mapping_mremap(struct vm_area_struct *new_vma) 3209b059a453SDmitry Safonov { 3210b059a453SDmitry Safonov struct vm_special_mapping *sm = new_vma->vm_private_data; 3211b059a453SDmitry Safonov 3212280e87e9SDmitry Safonov if (WARN_ON_ONCE(current->mm != new_vma->vm_mm)) 3213280e87e9SDmitry Safonov return -EFAULT; 3214280e87e9SDmitry Safonov 3215b059a453SDmitry Safonov if (sm->mremap) 3216b059a453SDmitry Safonov return sm->mremap(sm, new_vma); 3217280e87e9SDmitry Safonov 3218b059a453SDmitry Safonov return 0; 3219b059a453SDmitry Safonov } 3220b059a453SDmitry Safonov 3221a62c34bdSAndy Lutomirski static const struct vm_operations_struct special_mapping_vmops = { 3222a62c34bdSAndy Lutomirski .close = special_mapping_close, 3223a62c34bdSAndy Lutomirski .fault = special_mapping_fault, 3224b059a453SDmitry Safonov .mremap = special_mapping_mremap, 3225a62c34bdSAndy Lutomirski .name = special_mapping_name, 3226a62c34bdSAndy Lutomirski }; 3227a62c34bdSAndy Lutomirski 3228a62c34bdSAndy Lutomirski static const struct vm_operations_struct legacy_special_mapping_vmops = { 3229a62c34bdSAndy Lutomirski .close = special_mapping_close, 3230a62c34bdSAndy Lutomirski .fault = special_mapping_fault, 3231a62c34bdSAndy Lutomirski }; 3232fa5dc22fSRoland McGrath 323311bac800SDave Jiang static int special_mapping_fault(struct vm_fault *vmf) 3234fa5dc22fSRoland McGrath { 323511bac800SDave Jiang struct vm_area_struct *vma = vmf->vma; 3236b1d0e4f5SNick Piggin pgoff_t pgoff; 3237fa5dc22fSRoland McGrath struct page **pages; 3238fa5dc22fSRoland McGrath 3239f872f540SAndy Lutomirski if (vma->vm_ops == &legacy_special_mapping_vmops) { 3240a62c34bdSAndy Lutomirski pages = vma->vm_private_data; 3241f872f540SAndy Lutomirski } else { 3242f872f540SAndy Lutomirski struct vm_special_mapping *sm = vma->vm_private_data; 3243f872f540SAndy Lutomirski 3244f872f540SAndy Lutomirski if (sm->fault) 324511bac800SDave Jiang return sm->fault(sm, vmf->vma, vmf); 3246f872f540SAndy Lutomirski 3247f872f540SAndy Lutomirski pages = sm->pages; 3248f872f540SAndy Lutomirski } 3249a62c34bdSAndy Lutomirski 32508a9cc3b5SOleg Nesterov for (pgoff = vmf->pgoff; pgoff && *pages; ++pages) 3251b1d0e4f5SNick Piggin pgoff--; 3252fa5dc22fSRoland McGrath 3253fa5dc22fSRoland McGrath if (*pages) { 3254fa5dc22fSRoland McGrath struct page *page = *pages; 3255fa5dc22fSRoland McGrath get_page(page); 3256b1d0e4f5SNick Piggin vmf->page = page; 3257b1d0e4f5SNick Piggin return 0; 3258fa5dc22fSRoland McGrath } 3259fa5dc22fSRoland McGrath 3260b1d0e4f5SNick Piggin return VM_FAULT_SIGBUS; 3261fa5dc22fSRoland McGrath } 3262fa5dc22fSRoland McGrath 3263a62c34bdSAndy Lutomirski static struct vm_area_struct *__install_special_mapping( 3264a62c34bdSAndy Lutomirski struct mm_struct *mm, 3265fa5dc22fSRoland McGrath unsigned long addr, unsigned long len, 326627f28b97SChen Gang unsigned long vm_flags, void *priv, 326727f28b97SChen Gang const struct vm_operations_struct *ops) 3268fa5dc22fSRoland McGrath { 3269462e635eSTavis Ormandy int ret; 3270fa5dc22fSRoland McGrath struct vm_area_struct *vma; 3271fa5dc22fSRoland McGrath 3272fa5dc22fSRoland McGrath vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL); 3273fa5dc22fSRoland McGrath if (unlikely(vma == NULL)) 32743935ed6aSStefani Seibold return ERR_PTR(-ENOMEM); 3275fa5dc22fSRoland McGrath 32765beb4930SRik van Riel INIT_LIST_HEAD(&vma->anon_vma_chain); 3277fa5dc22fSRoland McGrath vma->vm_mm = mm; 3278fa5dc22fSRoland McGrath vma->vm_start = addr; 3279fa5dc22fSRoland McGrath vma->vm_end = addr + len; 3280fa5dc22fSRoland McGrath 3281d9104d1cSCyrill Gorcunov vma->vm_flags = vm_flags | mm->def_flags | VM_DONTEXPAND | VM_SOFTDIRTY; 32823ed75eb8SColy Li vma->vm_page_prot = vm_get_page_prot(vma->vm_flags); 3283fa5dc22fSRoland McGrath 3284a62c34bdSAndy Lutomirski vma->vm_ops = ops; 3285a62c34bdSAndy Lutomirski vma->vm_private_data = priv; 3286fa5dc22fSRoland McGrath 3287462e635eSTavis Ormandy ret = insert_vm_struct(mm, vma); 3288462e635eSTavis Ormandy if (ret) 3289462e635eSTavis Ormandy goto out; 3290fa5dc22fSRoland McGrath 329184638335SKonstantin Khlebnikov vm_stat_account(mm, vma->vm_flags, len >> PAGE_SHIFT); 3292fa5dc22fSRoland McGrath 3293cdd6c482SIngo Molnar perf_event_mmap(vma); 3294089dd79dSPeter Zijlstra 32953935ed6aSStefani Seibold return vma; 3296462e635eSTavis Ormandy 3297462e635eSTavis Ormandy out: 3298462e635eSTavis Ormandy kmem_cache_free(vm_area_cachep, vma); 32993935ed6aSStefani Seibold return ERR_PTR(ret); 33003935ed6aSStefani Seibold } 33013935ed6aSStefani Seibold 33022eefd878SDmitry Safonov bool vma_is_special_mapping(const struct vm_area_struct *vma, 33032eefd878SDmitry Safonov const struct vm_special_mapping *sm) 33042eefd878SDmitry Safonov { 33052eefd878SDmitry Safonov return vma->vm_private_data == sm && 33062eefd878SDmitry Safonov (vma->vm_ops == &special_mapping_vmops || 33072eefd878SDmitry Safonov vma->vm_ops == &legacy_special_mapping_vmops); 33082eefd878SDmitry Safonov } 33092eefd878SDmitry Safonov 3310a62c34bdSAndy Lutomirski /* 3311a62c34bdSAndy Lutomirski * Called with mm->mmap_sem held for writing. 3312a62c34bdSAndy Lutomirski * Insert a new vma covering the given region, with the given flags. 3313a62c34bdSAndy Lutomirski * Its pages are supplied by the given array of struct page *. 3314a62c34bdSAndy Lutomirski * The array can be shorter than len >> PAGE_SHIFT if it's null-terminated. 3315a62c34bdSAndy Lutomirski * The region past the last page supplied will always produce SIGBUS. 3316a62c34bdSAndy Lutomirski * The array pointer and the pages it points to are assumed to stay alive 3317a62c34bdSAndy Lutomirski * for as long as this mapping might exist. 3318a62c34bdSAndy Lutomirski */ 3319a62c34bdSAndy Lutomirski struct vm_area_struct *_install_special_mapping( 3320a62c34bdSAndy Lutomirski struct mm_struct *mm, 3321a62c34bdSAndy Lutomirski unsigned long addr, unsigned long len, 3322a62c34bdSAndy Lutomirski unsigned long vm_flags, const struct vm_special_mapping *spec) 3323a62c34bdSAndy Lutomirski { 332427f28b97SChen Gang return __install_special_mapping(mm, addr, len, vm_flags, (void *)spec, 332527f28b97SChen Gang &special_mapping_vmops); 3326a62c34bdSAndy Lutomirski } 3327a62c34bdSAndy Lutomirski 33283935ed6aSStefani Seibold int install_special_mapping(struct mm_struct *mm, 33293935ed6aSStefani Seibold unsigned long addr, unsigned long len, 33303935ed6aSStefani Seibold unsigned long vm_flags, struct page **pages) 33313935ed6aSStefani Seibold { 3332a62c34bdSAndy Lutomirski struct vm_area_struct *vma = __install_special_mapping( 333327f28b97SChen Gang mm, addr, len, vm_flags, (void *)pages, 333427f28b97SChen Gang &legacy_special_mapping_vmops); 33353935ed6aSStefani Seibold 333614bd5b45SDuan Jiong return PTR_ERR_OR_ZERO(vma); 3337fa5dc22fSRoland McGrath } 33387906d00cSAndrea Arcangeli 33397906d00cSAndrea Arcangeli static DEFINE_MUTEX(mm_all_locks_mutex); 33407906d00cSAndrea Arcangeli 3341454ed842SPeter Zijlstra static void vm_lock_anon_vma(struct mm_struct *mm, struct anon_vma *anon_vma) 33427906d00cSAndrea Arcangeli { 3343bf181b9fSMichel Lespinasse if (!test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_node)) { 33447906d00cSAndrea Arcangeli /* 33457906d00cSAndrea Arcangeli * The LSB of head.next can't change from under us 33467906d00cSAndrea Arcangeli * because we hold the mm_all_locks_mutex. 33477906d00cSAndrea Arcangeli */ 3348572043c9SJiri Kosina down_write_nest_lock(&anon_vma->root->rwsem, &mm->mmap_sem); 33497906d00cSAndrea Arcangeli /* 33507906d00cSAndrea Arcangeli * We can safely modify head.next after taking the 33515a505085SIngo Molnar * anon_vma->root->rwsem. If some other vma in this mm shares 33527906d00cSAndrea Arcangeli * the same anon_vma we won't take it again. 33537906d00cSAndrea Arcangeli * 33547906d00cSAndrea Arcangeli * No need of atomic instructions here, head.next 33557906d00cSAndrea Arcangeli * can't change from under us thanks to the 33565a505085SIngo Molnar * anon_vma->root->rwsem. 33577906d00cSAndrea Arcangeli */ 33587906d00cSAndrea Arcangeli if (__test_and_set_bit(0, (unsigned long *) 3359bf181b9fSMichel Lespinasse &anon_vma->root->rb_root.rb_node)) 33607906d00cSAndrea Arcangeli BUG(); 33617906d00cSAndrea Arcangeli } 33627906d00cSAndrea Arcangeli } 33637906d00cSAndrea Arcangeli 3364454ed842SPeter Zijlstra static void vm_lock_mapping(struct mm_struct *mm, struct address_space *mapping) 33657906d00cSAndrea Arcangeli { 33667906d00cSAndrea Arcangeli if (!test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) { 33677906d00cSAndrea Arcangeli /* 33687906d00cSAndrea Arcangeli * AS_MM_ALL_LOCKS can't change from under us because 33697906d00cSAndrea Arcangeli * we hold the mm_all_locks_mutex. 33707906d00cSAndrea Arcangeli * 33717906d00cSAndrea Arcangeli * Operations on ->flags have to be atomic because 33727906d00cSAndrea Arcangeli * even if AS_MM_ALL_LOCKS is stable thanks to the 33737906d00cSAndrea Arcangeli * mm_all_locks_mutex, there may be other cpus 33747906d00cSAndrea Arcangeli * changing other bitflags in parallel to us. 33757906d00cSAndrea Arcangeli */ 33767906d00cSAndrea Arcangeli if (test_and_set_bit(AS_MM_ALL_LOCKS, &mapping->flags)) 33777906d00cSAndrea Arcangeli BUG(); 3378c8c06efaSDavidlohr Bueso down_write_nest_lock(&mapping->i_mmap_rwsem, &mm->mmap_sem); 33797906d00cSAndrea Arcangeli } 33807906d00cSAndrea Arcangeli } 33817906d00cSAndrea Arcangeli 33827906d00cSAndrea Arcangeli /* 33837906d00cSAndrea Arcangeli * This operation locks against the VM for all pte/vma/mm related 33847906d00cSAndrea Arcangeli * operations that could ever happen on a certain mm. This includes 33857906d00cSAndrea Arcangeli * vmtruncate, try_to_unmap, and all page faults. 33867906d00cSAndrea Arcangeli * 33877906d00cSAndrea Arcangeli * The caller must take the mmap_sem in write mode before calling 33887906d00cSAndrea Arcangeli * mm_take_all_locks(). The caller isn't allowed to release the 33897906d00cSAndrea Arcangeli * mmap_sem until mm_drop_all_locks() returns. 33907906d00cSAndrea Arcangeli * 33917906d00cSAndrea Arcangeli * mmap_sem in write mode is required in order to block all operations 33927906d00cSAndrea Arcangeli * that could modify pagetables and free pages without need of 339327ba0644SKirill A. Shutemov * altering the vma layout. It's also needed in write mode to avoid new 33947906d00cSAndrea Arcangeli * anon_vmas to be associated with existing vmas. 33957906d00cSAndrea Arcangeli * 33967906d00cSAndrea Arcangeli * A single task can't take more than one mm_take_all_locks() in a row 33977906d00cSAndrea Arcangeli * or it would deadlock. 33987906d00cSAndrea Arcangeli * 3399bf181b9fSMichel Lespinasse * The LSB in anon_vma->rb_root.rb_node and the AS_MM_ALL_LOCKS bitflag in 34007906d00cSAndrea Arcangeli * mapping->flags avoid to take the same lock twice, if more than one 34017906d00cSAndrea Arcangeli * vma in this mm is backed by the same anon_vma or address_space. 34027906d00cSAndrea Arcangeli * 340388f306b6SKirill A. Shutemov * We take locks in following order, accordingly to comment at beginning 340488f306b6SKirill A. Shutemov * of mm/rmap.c: 340588f306b6SKirill A. Shutemov * - all hugetlbfs_i_mmap_rwsem_key locks (aka mapping->i_mmap_rwsem for 340688f306b6SKirill A. Shutemov * hugetlb mapping); 340788f306b6SKirill A. Shutemov * - all i_mmap_rwsem locks; 340888f306b6SKirill A. Shutemov * - all anon_vma->rwseml 340988f306b6SKirill A. Shutemov * 341088f306b6SKirill A. Shutemov * We can take all locks within these types randomly because the VM code 341188f306b6SKirill A. Shutemov * doesn't nest them and we protected from parallel mm_take_all_locks() by 341288f306b6SKirill A. Shutemov * mm_all_locks_mutex. 34137906d00cSAndrea Arcangeli * 34147906d00cSAndrea Arcangeli * mm_take_all_locks() and mm_drop_all_locks are expensive operations 34157906d00cSAndrea Arcangeli * that may have to take thousand of locks. 34167906d00cSAndrea Arcangeli * 34177906d00cSAndrea Arcangeli * mm_take_all_locks() can fail if it's interrupted by signals. 34187906d00cSAndrea Arcangeli */ 34197906d00cSAndrea Arcangeli int mm_take_all_locks(struct mm_struct *mm) 34207906d00cSAndrea Arcangeli { 34217906d00cSAndrea Arcangeli struct vm_area_struct *vma; 34225beb4930SRik van Riel struct anon_vma_chain *avc; 34237906d00cSAndrea Arcangeli 34247906d00cSAndrea Arcangeli BUG_ON(down_read_trylock(&mm->mmap_sem)); 34257906d00cSAndrea Arcangeli 34267906d00cSAndrea Arcangeli mutex_lock(&mm_all_locks_mutex); 34277906d00cSAndrea Arcangeli 34287906d00cSAndrea Arcangeli for (vma = mm->mmap; vma; vma = vma->vm_next) { 34297906d00cSAndrea Arcangeli if (signal_pending(current)) 34307906d00cSAndrea Arcangeli goto out_unlock; 343188f306b6SKirill A. Shutemov if (vma->vm_file && vma->vm_file->f_mapping && 343288f306b6SKirill A. Shutemov is_vm_hugetlb_page(vma)) 343388f306b6SKirill A. Shutemov vm_lock_mapping(mm, vma->vm_file->f_mapping); 343488f306b6SKirill A. Shutemov } 343588f306b6SKirill A. Shutemov 343688f306b6SKirill A. Shutemov for (vma = mm->mmap; vma; vma = vma->vm_next) { 343788f306b6SKirill A. Shutemov if (signal_pending(current)) 343888f306b6SKirill A. Shutemov goto out_unlock; 343988f306b6SKirill A. Shutemov if (vma->vm_file && vma->vm_file->f_mapping && 344088f306b6SKirill A. Shutemov !is_vm_hugetlb_page(vma)) 3441454ed842SPeter Zijlstra vm_lock_mapping(mm, vma->vm_file->f_mapping); 34427906d00cSAndrea Arcangeli } 34437cd5a02fSPeter Zijlstra 34447cd5a02fSPeter Zijlstra for (vma = mm->mmap; vma; vma = vma->vm_next) { 34457cd5a02fSPeter Zijlstra if (signal_pending(current)) 34467cd5a02fSPeter Zijlstra goto out_unlock; 34477cd5a02fSPeter Zijlstra if (vma->anon_vma) 34485beb4930SRik van Riel list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) 34495beb4930SRik van Riel vm_lock_anon_vma(mm, avc->anon_vma); 34507cd5a02fSPeter Zijlstra } 34517cd5a02fSPeter Zijlstra 3452584cff54SKautuk Consul return 0; 34537906d00cSAndrea Arcangeli 34547906d00cSAndrea Arcangeli out_unlock: 34557906d00cSAndrea Arcangeli mm_drop_all_locks(mm); 3456584cff54SKautuk Consul return -EINTR; 34577906d00cSAndrea Arcangeli } 34587906d00cSAndrea Arcangeli 34597906d00cSAndrea Arcangeli static void vm_unlock_anon_vma(struct anon_vma *anon_vma) 34607906d00cSAndrea Arcangeli { 3461bf181b9fSMichel Lespinasse if (test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_node)) { 34627906d00cSAndrea Arcangeli /* 34637906d00cSAndrea Arcangeli * The LSB of head.next can't change to 0 from under 34647906d00cSAndrea Arcangeli * us because we hold the mm_all_locks_mutex. 34657906d00cSAndrea Arcangeli * 34667906d00cSAndrea Arcangeli * We must however clear the bitflag before unlocking 3467bf181b9fSMichel Lespinasse * the vma so the users using the anon_vma->rb_root will 34687906d00cSAndrea Arcangeli * never see our bitflag. 34697906d00cSAndrea Arcangeli * 34707906d00cSAndrea Arcangeli * No need of atomic instructions here, head.next 34717906d00cSAndrea Arcangeli * can't change from under us until we release the 34725a505085SIngo Molnar * anon_vma->root->rwsem. 34737906d00cSAndrea Arcangeli */ 34747906d00cSAndrea Arcangeli if (!__test_and_clear_bit(0, (unsigned long *) 3475bf181b9fSMichel Lespinasse &anon_vma->root->rb_root.rb_node)) 34767906d00cSAndrea Arcangeli BUG(); 347708b52706SKonstantin Khlebnikov anon_vma_unlock_write(anon_vma); 34787906d00cSAndrea Arcangeli } 34797906d00cSAndrea Arcangeli } 34807906d00cSAndrea Arcangeli 34817906d00cSAndrea Arcangeli static void vm_unlock_mapping(struct address_space *mapping) 34827906d00cSAndrea Arcangeli { 34837906d00cSAndrea Arcangeli if (test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) { 34847906d00cSAndrea Arcangeli /* 34857906d00cSAndrea Arcangeli * AS_MM_ALL_LOCKS can't change to 0 from under us 34867906d00cSAndrea Arcangeli * because we hold the mm_all_locks_mutex. 34877906d00cSAndrea Arcangeli */ 348883cde9e8SDavidlohr Bueso i_mmap_unlock_write(mapping); 34897906d00cSAndrea Arcangeli if (!test_and_clear_bit(AS_MM_ALL_LOCKS, 34907906d00cSAndrea Arcangeli &mapping->flags)) 34917906d00cSAndrea Arcangeli BUG(); 34927906d00cSAndrea Arcangeli } 34937906d00cSAndrea Arcangeli } 34947906d00cSAndrea Arcangeli 34957906d00cSAndrea Arcangeli /* 34967906d00cSAndrea Arcangeli * The mmap_sem cannot be released by the caller until 34977906d00cSAndrea Arcangeli * mm_drop_all_locks() returns. 34987906d00cSAndrea Arcangeli */ 34997906d00cSAndrea Arcangeli void mm_drop_all_locks(struct mm_struct *mm) 35007906d00cSAndrea Arcangeli { 35017906d00cSAndrea Arcangeli struct vm_area_struct *vma; 35025beb4930SRik van Riel struct anon_vma_chain *avc; 35037906d00cSAndrea Arcangeli 35047906d00cSAndrea Arcangeli BUG_ON(down_read_trylock(&mm->mmap_sem)); 35057906d00cSAndrea Arcangeli BUG_ON(!mutex_is_locked(&mm_all_locks_mutex)); 35067906d00cSAndrea Arcangeli 35077906d00cSAndrea Arcangeli for (vma = mm->mmap; vma; vma = vma->vm_next) { 35087906d00cSAndrea Arcangeli if (vma->anon_vma) 35095beb4930SRik van Riel list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) 35105beb4930SRik van Riel vm_unlock_anon_vma(avc->anon_vma); 35117906d00cSAndrea Arcangeli if (vma->vm_file && vma->vm_file->f_mapping) 35127906d00cSAndrea Arcangeli vm_unlock_mapping(vma->vm_file->f_mapping); 35137906d00cSAndrea Arcangeli } 35147906d00cSAndrea Arcangeli 35157906d00cSAndrea Arcangeli mutex_unlock(&mm_all_locks_mutex); 35167906d00cSAndrea Arcangeli } 35178feae131SDavid Howells 35188feae131SDavid Howells /* 35193edf41d8Sseokhoon.yoon * initialise the percpu counter for VM 35208feae131SDavid Howells */ 35218feae131SDavid Howells void __init mmap_init(void) 35228feae131SDavid Howells { 352300a62ce9SKOSAKI Motohiro int ret; 352400a62ce9SKOSAKI Motohiro 3525908c7f19STejun Heo ret = percpu_counter_init(&vm_committed_as, 0, GFP_KERNEL); 352600a62ce9SKOSAKI Motohiro VM_BUG_ON(ret); 35278feae131SDavid Howells } 3528c9b1d098SAndrew Shewmaker 3529c9b1d098SAndrew Shewmaker /* 3530c9b1d098SAndrew Shewmaker * Initialise sysctl_user_reserve_kbytes. 3531c9b1d098SAndrew Shewmaker * 3532c9b1d098SAndrew Shewmaker * This is intended to prevent a user from starting a single memory hogging 3533c9b1d098SAndrew Shewmaker * process, such that they cannot recover (kill the hog) in OVERCOMMIT_NEVER 3534c9b1d098SAndrew Shewmaker * mode. 3535c9b1d098SAndrew Shewmaker * 3536c9b1d098SAndrew Shewmaker * The default value is min(3% of free memory, 128MB) 3537c9b1d098SAndrew Shewmaker * 128MB is enough to recover with sshd/login, bash, and top/kill. 3538c9b1d098SAndrew Shewmaker */ 35391640879aSAndrew Shewmaker static int init_user_reserve(void) 3540c9b1d098SAndrew Shewmaker { 3541c9b1d098SAndrew Shewmaker unsigned long free_kbytes; 3542c9b1d098SAndrew Shewmaker 3543c41f012aSMichal Hocko free_kbytes = global_zone_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10); 3544c9b1d098SAndrew Shewmaker 3545c9b1d098SAndrew Shewmaker sysctl_user_reserve_kbytes = min(free_kbytes / 32, 1UL << 17); 3546c9b1d098SAndrew Shewmaker return 0; 3547c9b1d098SAndrew Shewmaker } 3548a64fb3cdSPaul Gortmaker subsys_initcall(init_user_reserve); 35494eeab4f5SAndrew Shewmaker 35504eeab4f5SAndrew Shewmaker /* 35514eeab4f5SAndrew Shewmaker * Initialise sysctl_admin_reserve_kbytes. 35524eeab4f5SAndrew Shewmaker * 35534eeab4f5SAndrew Shewmaker * The purpose of sysctl_admin_reserve_kbytes is to allow the sys admin 35544eeab4f5SAndrew Shewmaker * to log in and kill a memory hogging process. 35554eeab4f5SAndrew Shewmaker * 35564eeab4f5SAndrew Shewmaker * Systems with more than 256MB will reserve 8MB, enough to recover 35574eeab4f5SAndrew Shewmaker * with sshd, bash, and top in OVERCOMMIT_GUESS. Smaller systems will 35584eeab4f5SAndrew Shewmaker * only reserve 3% of free pages by default. 35594eeab4f5SAndrew Shewmaker */ 35601640879aSAndrew Shewmaker static int init_admin_reserve(void) 35614eeab4f5SAndrew Shewmaker { 35624eeab4f5SAndrew Shewmaker unsigned long free_kbytes; 35634eeab4f5SAndrew Shewmaker 3564c41f012aSMichal Hocko free_kbytes = global_zone_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10); 35654eeab4f5SAndrew Shewmaker 35664eeab4f5SAndrew Shewmaker sysctl_admin_reserve_kbytes = min(free_kbytes / 32, 1UL << 13); 35674eeab4f5SAndrew Shewmaker return 0; 35684eeab4f5SAndrew Shewmaker } 3569a64fb3cdSPaul Gortmaker subsys_initcall(init_admin_reserve); 35701640879aSAndrew Shewmaker 35711640879aSAndrew Shewmaker /* 35721640879aSAndrew Shewmaker * Reinititalise user and admin reserves if memory is added or removed. 35731640879aSAndrew Shewmaker * 35741640879aSAndrew Shewmaker * The default user reserve max is 128MB, and the default max for the 35751640879aSAndrew Shewmaker * admin reserve is 8MB. These are usually, but not always, enough to 35761640879aSAndrew Shewmaker * enable recovery from a memory hogging process using login/sshd, a shell, 35771640879aSAndrew Shewmaker * and tools like top. It may make sense to increase or even disable the 35781640879aSAndrew Shewmaker * reserve depending on the existence of swap or variations in the recovery 35791640879aSAndrew Shewmaker * tools. So, the admin may have changed them. 35801640879aSAndrew Shewmaker * 35811640879aSAndrew Shewmaker * If memory is added and the reserves have been eliminated or increased above 35821640879aSAndrew Shewmaker * the default max, then we'll trust the admin. 35831640879aSAndrew Shewmaker * 35841640879aSAndrew Shewmaker * If memory is removed and there isn't enough free memory, then we 35851640879aSAndrew Shewmaker * need to reset the reserves. 35861640879aSAndrew Shewmaker * 35871640879aSAndrew Shewmaker * Otherwise keep the reserve set by the admin. 35881640879aSAndrew Shewmaker */ 35891640879aSAndrew Shewmaker static int reserve_mem_notifier(struct notifier_block *nb, 35901640879aSAndrew Shewmaker unsigned long action, void *data) 35911640879aSAndrew Shewmaker { 35921640879aSAndrew Shewmaker unsigned long tmp, free_kbytes; 35931640879aSAndrew Shewmaker 35941640879aSAndrew Shewmaker switch (action) { 35951640879aSAndrew Shewmaker case MEM_ONLINE: 35961640879aSAndrew Shewmaker /* Default max is 128MB. Leave alone if modified by operator. */ 35971640879aSAndrew Shewmaker tmp = sysctl_user_reserve_kbytes; 35981640879aSAndrew Shewmaker if (0 < tmp && tmp < (1UL << 17)) 35991640879aSAndrew Shewmaker init_user_reserve(); 36001640879aSAndrew Shewmaker 36011640879aSAndrew Shewmaker /* Default max is 8MB. Leave alone if modified by operator. */ 36021640879aSAndrew Shewmaker tmp = sysctl_admin_reserve_kbytes; 36031640879aSAndrew Shewmaker if (0 < tmp && tmp < (1UL << 13)) 36041640879aSAndrew Shewmaker init_admin_reserve(); 36051640879aSAndrew Shewmaker 36061640879aSAndrew Shewmaker break; 36071640879aSAndrew Shewmaker case MEM_OFFLINE: 3608c41f012aSMichal Hocko free_kbytes = global_zone_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10); 36091640879aSAndrew Shewmaker 36101640879aSAndrew Shewmaker if (sysctl_user_reserve_kbytes > free_kbytes) { 36111640879aSAndrew Shewmaker init_user_reserve(); 36121640879aSAndrew Shewmaker pr_info("vm.user_reserve_kbytes reset to %lu\n", 36131640879aSAndrew Shewmaker sysctl_user_reserve_kbytes); 36141640879aSAndrew Shewmaker } 36151640879aSAndrew Shewmaker 36161640879aSAndrew Shewmaker if (sysctl_admin_reserve_kbytes > free_kbytes) { 36171640879aSAndrew Shewmaker init_admin_reserve(); 36181640879aSAndrew Shewmaker pr_info("vm.admin_reserve_kbytes reset to %lu\n", 36191640879aSAndrew Shewmaker sysctl_admin_reserve_kbytes); 36201640879aSAndrew Shewmaker } 36211640879aSAndrew Shewmaker break; 36221640879aSAndrew Shewmaker default: 36231640879aSAndrew Shewmaker break; 36241640879aSAndrew Shewmaker } 36251640879aSAndrew Shewmaker return NOTIFY_OK; 36261640879aSAndrew Shewmaker } 36271640879aSAndrew Shewmaker 36281640879aSAndrew Shewmaker static struct notifier_block reserve_mem_nb = { 36291640879aSAndrew Shewmaker .notifier_call = reserve_mem_notifier, 36301640879aSAndrew Shewmaker }; 36311640879aSAndrew Shewmaker 36321640879aSAndrew Shewmaker static int __meminit init_reserve_notifier(void) 36331640879aSAndrew Shewmaker { 36341640879aSAndrew Shewmaker if (register_hotmemory_notifier(&reserve_mem_nb)) 3635b1de0d13SMitchel Humpherys pr_err("Failed registering memory add/remove notifier for admin reserve\n"); 36361640879aSAndrew Shewmaker 36371640879aSAndrew Shewmaker return 0; 36381640879aSAndrew Shewmaker } 3639a64fb3cdSPaul Gortmaker subsys_initcall(init_reserve_notifier); 3640