1*457c8996SThomas Gleixner // SPDX-License-Identifier: GPL-2.0-only 21da177e4SLinus Torvalds /* 31da177e4SLinus Torvalds * mm/mmap.c 41da177e4SLinus Torvalds * 51da177e4SLinus Torvalds * Written by obz. 61da177e4SLinus Torvalds * 7046c6884SAlan Cox * Address space accounting code <alan@lxorguk.ukuu.org.uk> 81da177e4SLinus Torvalds */ 91da177e4SLinus Torvalds 10b1de0d13SMitchel Humpherys #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt 11b1de0d13SMitchel Humpherys 12e8420a8eSCyril Hrubis #include <linux/kernel.h> 131da177e4SLinus Torvalds #include <linux/slab.h> 144af3c9ccSAlexey Dobriyan #include <linux/backing-dev.h> 151da177e4SLinus Torvalds #include <linux/mm.h> 16615d6e87SDavidlohr Bueso #include <linux/vmacache.h> 171da177e4SLinus Torvalds #include <linux/shm.h> 181da177e4SLinus Torvalds #include <linux/mman.h> 191da177e4SLinus Torvalds #include <linux/pagemap.h> 201da177e4SLinus Torvalds #include <linux/swap.h> 211da177e4SLinus Torvalds #include <linux/syscalls.h> 22c59ede7bSRandy.Dunlap #include <linux/capability.h> 231da177e4SLinus Torvalds #include <linux/init.h> 241da177e4SLinus Torvalds #include <linux/file.h> 251da177e4SLinus Torvalds #include <linux/fs.h> 261da177e4SLinus Torvalds #include <linux/personality.h> 271da177e4SLinus Torvalds #include <linux/security.h> 281da177e4SLinus Torvalds #include <linux/hugetlb.h> 29c01d5b30SHugh Dickins #include <linux/shmem_fs.h> 301da177e4SLinus Torvalds #include <linux/profile.h> 31b95f1b31SPaul Gortmaker #include <linux/export.h> 321da177e4SLinus Torvalds #include <linux/mount.h> 331da177e4SLinus Torvalds #include <linux/mempolicy.h> 341da177e4SLinus Torvalds #include <linux/rmap.h> 35cddb8a5cSAndrea Arcangeli #include <linux/mmu_notifier.h> 3682f71ae4SKonstantin Khlebnikov #include <linux/mmdebug.h> 37cdd6c482SIngo Molnar #include <linux/perf_event.h> 38120a795dSAl Viro #include <linux/audit.h> 39b15d00b6SAndrea Arcangeli #include <linux/khugepaged.h> 402b144498SSrikar Dronamraju #include <linux/uprobes.h> 41d3737187SMichel Lespinasse #include <linux/rbtree_augmented.h> 421640879aSAndrew Shewmaker #include <linux/notifier.h> 431640879aSAndrew Shewmaker #include <linux/memory.h> 44b1de0d13SMitchel Humpherys #include <linux/printk.h> 4519a809afSAndrea Arcangeli #include <linux/userfaultfd_k.h> 46d977d56cSKonstantin Khlebnikov #include <linux/moduleparam.h> 4762b5f7d0SDave Hansen #include <linux/pkeys.h> 4821292580SAndrea Arcangeli #include <linux/oom.h> 4904f5866eSAndrea Arcangeli #include <linux/sched/mm.h> 501da177e4SLinus Torvalds 517c0f6ba6SLinus Torvalds #include <linux/uaccess.h> 521da177e4SLinus Torvalds #include <asm/cacheflush.h> 531da177e4SLinus Torvalds #include <asm/tlb.h> 54d6dd61c8SJeremy Fitzhardinge #include <asm/mmu_context.h> 551da177e4SLinus Torvalds 5642b77728SJan Beulich #include "internal.h" 5742b77728SJan Beulich 583a459756SKirill Korotaev #ifndef arch_mmap_check 593a459756SKirill Korotaev #define arch_mmap_check(addr, len, flags) (0) 603a459756SKirill Korotaev #endif 613a459756SKirill Korotaev 62d07e2259SDaniel Cashman #ifdef CONFIG_HAVE_ARCH_MMAP_RND_BITS 63d07e2259SDaniel Cashman const int mmap_rnd_bits_min = CONFIG_ARCH_MMAP_RND_BITS_MIN; 64d07e2259SDaniel Cashman const int mmap_rnd_bits_max = CONFIG_ARCH_MMAP_RND_BITS_MAX; 65d07e2259SDaniel Cashman int mmap_rnd_bits __read_mostly = CONFIG_ARCH_MMAP_RND_BITS; 66d07e2259SDaniel Cashman #endif 67d07e2259SDaniel Cashman #ifdef CONFIG_HAVE_ARCH_MMAP_RND_COMPAT_BITS 68d07e2259SDaniel Cashman const int mmap_rnd_compat_bits_min = CONFIG_ARCH_MMAP_RND_COMPAT_BITS_MIN; 69d07e2259SDaniel Cashman const int mmap_rnd_compat_bits_max = CONFIG_ARCH_MMAP_RND_COMPAT_BITS_MAX; 70d07e2259SDaniel Cashman int mmap_rnd_compat_bits __read_mostly = CONFIG_ARCH_MMAP_RND_COMPAT_BITS; 71d07e2259SDaniel Cashman #endif 72d07e2259SDaniel Cashman 73f4fcd558SKonstantin Khlebnikov static bool ignore_rlimit_data; 74d977d56cSKonstantin Khlebnikov core_param(ignore_rlimit_data, ignore_rlimit_data, bool, 0644); 75d07e2259SDaniel Cashman 76e0da382cSHugh Dickins static void unmap_region(struct mm_struct *mm, 77e0da382cSHugh Dickins struct vm_area_struct *vma, struct vm_area_struct *prev, 78e0da382cSHugh Dickins unsigned long start, unsigned long end); 79e0da382cSHugh Dickins 801da177e4SLinus Torvalds /* description of effects of mapping type and prot in current implementation. 811da177e4SLinus Torvalds * this is due to the limited x86 page protection hardware. The expected 821da177e4SLinus Torvalds * behavior is in parens: 831da177e4SLinus Torvalds * 841da177e4SLinus Torvalds * map_type prot 851da177e4SLinus Torvalds * PROT_NONE PROT_READ PROT_WRITE PROT_EXEC 861da177e4SLinus Torvalds * MAP_SHARED r: (no) no r: (yes) yes r: (no) yes r: (no) yes 871da177e4SLinus Torvalds * w: (no) no w: (no) no w: (yes) yes w: (no) no 881da177e4SLinus Torvalds * x: (no) no x: (no) yes x: (no) yes x: (yes) yes 891da177e4SLinus Torvalds * 901da177e4SLinus Torvalds * MAP_PRIVATE r: (no) no r: (yes) yes r: (no) yes r: (no) yes 911da177e4SLinus Torvalds * w: (no) no w: (no) no w: (copy) copy w: (no) no 921da177e4SLinus Torvalds * x: (no) no x: (no) yes x: (no) yes x: (yes) yes 931da177e4SLinus Torvalds * 94cab15ce6SCatalin Marinas * On arm64, PROT_EXEC has the following behaviour for both MAP_SHARED and 95cab15ce6SCatalin Marinas * MAP_PRIVATE: 96cab15ce6SCatalin Marinas * r: (no) no 97cab15ce6SCatalin Marinas * w: (no) no 98cab15ce6SCatalin Marinas * x: (yes) yes 991da177e4SLinus Torvalds */ 100ac34ceafSDaniel Micay pgprot_t protection_map[16] __ro_after_init = { 1011da177e4SLinus Torvalds __P000, __P001, __P010, __P011, __P100, __P101, __P110, __P111, 1021da177e4SLinus Torvalds __S000, __S001, __S010, __S011, __S100, __S101, __S110, __S111 1031da177e4SLinus Torvalds }; 1041da177e4SLinus Torvalds 105316d097cSDave Hansen #ifndef CONFIG_ARCH_HAS_FILTER_PGPROT 106316d097cSDave Hansen static inline pgprot_t arch_filter_pgprot(pgprot_t prot) 107316d097cSDave Hansen { 108316d097cSDave Hansen return prot; 109316d097cSDave Hansen } 110316d097cSDave Hansen #endif 111316d097cSDave Hansen 112804af2cfSHugh Dickins pgprot_t vm_get_page_prot(unsigned long vm_flags) 113804af2cfSHugh Dickins { 114316d097cSDave Hansen pgprot_t ret = __pgprot(pgprot_val(protection_map[vm_flags & 115b845f313SDave Kleikamp (VM_READ|VM_WRITE|VM_EXEC|VM_SHARED)]) | 116b845f313SDave Kleikamp pgprot_val(arch_vm_get_page_prot(vm_flags))); 117316d097cSDave Hansen 118316d097cSDave Hansen return arch_filter_pgprot(ret); 119804af2cfSHugh Dickins } 120804af2cfSHugh Dickins EXPORT_SYMBOL(vm_get_page_prot); 121804af2cfSHugh Dickins 12264e45507SPeter Feiner static pgprot_t vm_pgprot_modify(pgprot_t oldprot, unsigned long vm_flags) 12364e45507SPeter Feiner { 12464e45507SPeter Feiner return pgprot_modify(oldprot, vm_get_page_prot(vm_flags)); 12564e45507SPeter Feiner } 12664e45507SPeter Feiner 12764e45507SPeter Feiner /* Update vma->vm_page_prot to reflect vma->vm_flags. */ 12864e45507SPeter Feiner void vma_set_page_prot(struct vm_area_struct *vma) 12964e45507SPeter Feiner { 13064e45507SPeter Feiner unsigned long vm_flags = vma->vm_flags; 1316d2329f8SAndrea Arcangeli pgprot_t vm_page_prot; 13264e45507SPeter Feiner 1336d2329f8SAndrea Arcangeli vm_page_prot = vm_pgprot_modify(vma->vm_page_prot, vm_flags); 1346d2329f8SAndrea Arcangeli if (vma_wants_writenotify(vma, vm_page_prot)) { 13564e45507SPeter Feiner vm_flags &= ~VM_SHARED; 1366d2329f8SAndrea Arcangeli vm_page_prot = vm_pgprot_modify(vm_page_prot, vm_flags); 13764e45507SPeter Feiner } 1386d2329f8SAndrea Arcangeli /* remove_protection_ptes reads vma->vm_page_prot without mmap_sem */ 1396d2329f8SAndrea Arcangeli WRITE_ONCE(vma->vm_page_prot, vm_page_prot); 14064e45507SPeter Feiner } 14164e45507SPeter Feiner 1421da177e4SLinus Torvalds /* 143c8c06efaSDavidlohr Bueso * Requires inode->i_mapping->i_mmap_rwsem 1441da177e4SLinus Torvalds */ 1451da177e4SLinus Torvalds static void __remove_shared_vm_struct(struct vm_area_struct *vma, 1461da177e4SLinus Torvalds struct file *file, struct address_space *mapping) 1471da177e4SLinus Torvalds { 1481da177e4SLinus Torvalds if (vma->vm_flags & VM_DENYWRITE) 149496ad9aaSAl Viro atomic_inc(&file_inode(file)->i_writecount); 1501da177e4SLinus Torvalds if (vma->vm_flags & VM_SHARED) 1514bb5f5d9SDavid Herrmann mapping_unmap_writable(mapping); 1521da177e4SLinus Torvalds 1531da177e4SLinus Torvalds flush_dcache_mmap_lock(mapping); 1546b2dbba8SMichel Lespinasse vma_interval_tree_remove(vma, &mapping->i_mmap); 1551da177e4SLinus Torvalds flush_dcache_mmap_unlock(mapping); 1561da177e4SLinus Torvalds } 1571da177e4SLinus Torvalds 1581da177e4SLinus Torvalds /* 1596b2dbba8SMichel Lespinasse * Unlink a file-based vm structure from its interval tree, to hide 160a8fb5618SHugh Dickins * vma from rmap and vmtruncate before freeing its page tables. 1611da177e4SLinus Torvalds */ 162a8fb5618SHugh Dickins void unlink_file_vma(struct vm_area_struct *vma) 1631da177e4SLinus Torvalds { 1641da177e4SLinus Torvalds struct file *file = vma->vm_file; 1651da177e4SLinus Torvalds 1661da177e4SLinus Torvalds if (file) { 1671da177e4SLinus Torvalds struct address_space *mapping = file->f_mapping; 16883cde9e8SDavidlohr Bueso i_mmap_lock_write(mapping); 1691da177e4SLinus Torvalds __remove_shared_vm_struct(vma, file, mapping); 17083cde9e8SDavidlohr Bueso i_mmap_unlock_write(mapping); 1711da177e4SLinus Torvalds } 172a8fb5618SHugh Dickins } 173a8fb5618SHugh Dickins 174a8fb5618SHugh Dickins /* 175a8fb5618SHugh Dickins * Close a vm structure and free it, returning the next. 176a8fb5618SHugh Dickins */ 177a8fb5618SHugh Dickins static struct vm_area_struct *remove_vma(struct vm_area_struct *vma) 178a8fb5618SHugh Dickins { 179a8fb5618SHugh Dickins struct vm_area_struct *next = vma->vm_next; 180a8fb5618SHugh Dickins 181a8fb5618SHugh Dickins might_sleep(); 1821da177e4SLinus Torvalds if (vma->vm_ops && vma->vm_ops->close) 1831da177e4SLinus Torvalds vma->vm_ops->close(vma); 184e9714acfSKonstantin Khlebnikov if (vma->vm_file) 185a8fb5618SHugh Dickins fput(vma->vm_file); 186f0be3d32SLee Schermerhorn mpol_put(vma_policy(vma)); 1873928d4f5SLinus Torvalds vm_area_free(vma); 188a8fb5618SHugh Dickins return next; 1891da177e4SLinus Torvalds } 1901da177e4SLinus Torvalds 191bb177a73SMichal Hocko static int do_brk_flags(unsigned long addr, unsigned long request, unsigned long flags, 192bb177a73SMichal Hocko struct list_head *uf); 1936a6160a7SHeiko Carstens SYSCALL_DEFINE1(brk, unsigned long, brk) 1941da177e4SLinus Torvalds { 1958764b338SCyrill Gorcunov unsigned long retval; 1969bc8039eSYang Shi unsigned long newbrk, oldbrk, origbrk; 1971da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 1981be7107fSHugh Dickins struct vm_area_struct *next; 199a5b4592cSJiri Kosina unsigned long min_brk; 200128557ffSMichel Lespinasse bool populate; 2019bc8039eSYang Shi bool downgraded = false; 202897ab3e0SMike Rapoport LIST_HEAD(uf); 2031da177e4SLinus Torvalds 204dc0ef0dfSMichal Hocko if (down_write_killable(&mm->mmap_sem)) 205dc0ef0dfSMichal Hocko return -EINTR; 2061da177e4SLinus Torvalds 2079bc8039eSYang Shi origbrk = mm->brk; 2089bc8039eSYang Shi 209a5b4592cSJiri Kosina #ifdef CONFIG_COMPAT_BRK 2105520e894SJiri Kosina /* 2115520e894SJiri Kosina * CONFIG_COMPAT_BRK can still be overridden by setting 2125520e894SJiri Kosina * randomize_va_space to 2, which will still cause mm->start_brk 2135520e894SJiri Kosina * to be arbitrarily shifted 2145520e894SJiri Kosina */ 2154471a675SJiri Kosina if (current->brk_randomized) 2165520e894SJiri Kosina min_brk = mm->start_brk; 2175520e894SJiri Kosina else 2185520e894SJiri Kosina min_brk = mm->end_data; 219a5b4592cSJiri Kosina #else 220a5b4592cSJiri Kosina min_brk = mm->start_brk; 221a5b4592cSJiri Kosina #endif 222a5b4592cSJiri Kosina if (brk < min_brk) 2231da177e4SLinus Torvalds goto out; 2241e624196SRam Gupta 2251e624196SRam Gupta /* 2261e624196SRam Gupta * Check against rlimit here. If this check is done later after the test 2271e624196SRam Gupta * of oldbrk with newbrk then it can escape the test and let the data 2281e624196SRam Gupta * segment grow beyond its set limit the in case where the limit is 2291e624196SRam Gupta * not page aligned -Ram Gupta 2301e624196SRam Gupta */ 2318764b338SCyrill Gorcunov if (check_data_rlimit(rlimit(RLIMIT_DATA), brk, mm->start_brk, 2328764b338SCyrill Gorcunov mm->end_data, mm->start_data)) 2331e624196SRam Gupta goto out; 2341e624196SRam Gupta 2351da177e4SLinus Torvalds newbrk = PAGE_ALIGN(brk); 2361da177e4SLinus Torvalds oldbrk = PAGE_ALIGN(mm->brk); 2379bc8039eSYang Shi if (oldbrk == newbrk) { 2389bc8039eSYang Shi mm->brk = brk; 2399bc8039eSYang Shi goto success; 2409bc8039eSYang Shi } 2411da177e4SLinus Torvalds 2429bc8039eSYang Shi /* 2439bc8039eSYang Shi * Always allow shrinking brk. 2449bc8039eSYang Shi * __do_munmap() may downgrade mmap_sem to read. 2459bc8039eSYang Shi */ 2461da177e4SLinus Torvalds if (brk <= mm->brk) { 2479bc8039eSYang Shi int ret; 2489bc8039eSYang Shi 2499bc8039eSYang Shi /* 2509bc8039eSYang Shi * mm->brk must to be protected by write mmap_sem so update it 2519bc8039eSYang Shi * before downgrading mmap_sem. When __do_munmap() fails, 2529bc8039eSYang Shi * mm->brk will be restored from origbrk. 2539bc8039eSYang Shi */ 2549bc8039eSYang Shi mm->brk = brk; 2559bc8039eSYang Shi ret = __do_munmap(mm, newbrk, oldbrk-newbrk, &uf, true); 2569bc8039eSYang Shi if (ret < 0) { 2579bc8039eSYang Shi mm->brk = origbrk; 2581da177e4SLinus Torvalds goto out; 2599bc8039eSYang Shi } else if (ret == 1) { 2609bc8039eSYang Shi downgraded = true; 2619bc8039eSYang Shi } 2629bc8039eSYang Shi goto success; 2631da177e4SLinus Torvalds } 2641da177e4SLinus Torvalds 2651da177e4SLinus Torvalds /* Check against existing mmap mappings. */ 2661be7107fSHugh Dickins next = find_vma(mm, oldbrk); 2671be7107fSHugh Dickins if (next && newbrk + PAGE_SIZE > vm_start_gap(next)) 2681da177e4SLinus Torvalds goto out; 2691da177e4SLinus Torvalds 2701da177e4SLinus Torvalds /* Ok, looks good - let it rip. */ 271bb177a73SMichal Hocko if (do_brk_flags(oldbrk, newbrk-oldbrk, 0, &uf) < 0) 2721da177e4SLinus Torvalds goto out; 2731da177e4SLinus Torvalds mm->brk = brk; 2749bc8039eSYang Shi 2759bc8039eSYang Shi success: 276128557ffSMichel Lespinasse populate = newbrk > oldbrk && (mm->def_flags & VM_LOCKED) != 0; 2779bc8039eSYang Shi if (downgraded) 2789bc8039eSYang Shi up_read(&mm->mmap_sem); 2799bc8039eSYang Shi else 280128557ffSMichel Lespinasse up_write(&mm->mmap_sem); 281897ab3e0SMike Rapoport userfaultfd_unmap_complete(mm, &uf); 282128557ffSMichel Lespinasse if (populate) 283128557ffSMichel Lespinasse mm_populate(oldbrk, newbrk - oldbrk); 284128557ffSMichel Lespinasse return brk; 285128557ffSMichel Lespinasse 2861da177e4SLinus Torvalds out: 2879bc8039eSYang Shi retval = origbrk; 2881da177e4SLinus Torvalds up_write(&mm->mmap_sem); 2891da177e4SLinus Torvalds return retval; 2901da177e4SLinus Torvalds } 2911da177e4SLinus Torvalds 292d3737187SMichel Lespinasse static long vma_compute_subtree_gap(struct vm_area_struct *vma) 293d3737187SMichel Lespinasse { 2941be7107fSHugh Dickins unsigned long max, prev_end, subtree_gap; 2951be7107fSHugh Dickins 2961be7107fSHugh Dickins /* 2971be7107fSHugh Dickins * Note: in the rare case of a VM_GROWSDOWN above a VM_GROWSUP, we 2981be7107fSHugh Dickins * allow two stack_guard_gaps between them here, and when choosing 2991be7107fSHugh Dickins * an unmapped area; whereas when expanding we only require one. 3001be7107fSHugh Dickins * That's a little inconsistent, but keeps the code here simpler. 3011be7107fSHugh Dickins */ 3021be7107fSHugh Dickins max = vm_start_gap(vma); 3031be7107fSHugh Dickins if (vma->vm_prev) { 3041be7107fSHugh Dickins prev_end = vm_end_gap(vma->vm_prev); 3051be7107fSHugh Dickins if (max > prev_end) 3061be7107fSHugh Dickins max -= prev_end; 3071be7107fSHugh Dickins else 3081be7107fSHugh Dickins max = 0; 3091be7107fSHugh Dickins } 310d3737187SMichel Lespinasse if (vma->vm_rb.rb_left) { 311d3737187SMichel Lespinasse subtree_gap = rb_entry(vma->vm_rb.rb_left, 312d3737187SMichel Lespinasse struct vm_area_struct, vm_rb)->rb_subtree_gap; 313d3737187SMichel Lespinasse if (subtree_gap > max) 314d3737187SMichel Lespinasse max = subtree_gap; 315d3737187SMichel Lespinasse } 316d3737187SMichel Lespinasse if (vma->vm_rb.rb_right) { 317d3737187SMichel Lespinasse subtree_gap = rb_entry(vma->vm_rb.rb_right, 318d3737187SMichel Lespinasse struct vm_area_struct, vm_rb)->rb_subtree_gap; 319d3737187SMichel Lespinasse if (subtree_gap > max) 320d3737187SMichel Lespinasse max = subtree_gap; 321d3737187SMichel Lespinasse } 322d3737187SMichel Lespinasse return max; 323d3737187SMichel Lespinasse } 324d3737187SMichel Lespinasse 325ed8ea815SMichel Lespinasse #ifdef CONFIG_DEBUG_VM_RB 326acf128d0SAndrea Arcangeli static int browse_rb(struct mm_struct *mm) 3271da177e4SLinus Torvalds { 328acf128d0SAndrea Arcangeli struct rb_root *root = &mm->mm_rb; 3295a0768f6SMichel Lespinasse int i = 0, j, bug = 0; 3301da177e4SLinus Torvalds struct rb_node *nd, *pn = NULL; 3311da177e4SLinus Torvalds unsigned long prev = 0, pend = 0; 3321da177e4SLinus Torvalds 3331da177e4SLinus Torvalds for (nd = rb_first(root); nd; nd = rb_next(nd)) { 3341da177e4SLinus Torvalds struct vm_area_struct *vma; 3351da177e4SLinus Torvalds vma = rb_entry(nd, struct vm_area_struct, vm_rb); 3365a0768f6SMichel Lespinasse if (vma->vm_start < prev) { 337ff26f70fSAndrew Morton pr_emerg("vm_start %lx < prev %lx\n", 338ff26f70fSAndrew Morton vma->vm_start, prev); 3395a0768f6SMichel Lespinasse bug = 1; 3405a0768f6SMichel Lespinasse } 3415a0768f6SMichel Lespinasse if (vma->vm_start < pend) { 342ff26f70fSAndrew Morton pr_emerg("vm_start %lx < pend %lx\n", 343ff26f70fSAndrew Morton vma->vm_start, pend); 3445a0768f6SMichel Lespinasse bug = 1; 3455a0768f6SMichel Lespinasse } 3465a0768f6SMichel Lespinasse if (vma->vm_start > vma->vm_end) { 347ff26f70fSAndrew Morton pr_emerg("vm_start %lx > vm_end %lx\n", 348ff26f70fSAndrew Morton vma->vm_start, vma->vm_end); 3495a0768f6SMichel Lespinasse bug = 1; 3505a0768f6SMichel Lespinasse } 351acf128d0SAndrea Arcangeli spin_lock(&mm->page_table_lock); 3525a0768f6SMichel Lespinasse if (vma->rb_subtree_gap != vma_compute_subtree_gap(vma)) { 3538542bdfcSSasha Levin pr_emerg("free gap %lx, correct %lx\n", 3545a0768f6SMichel Lespinasse vma->rb_subtree_gap, 3555a0768f6SMichel Lespinasse vma_compute_subtree_gap(vma)); 3565a0768f6SMichel Lespinasse bug = 1; 3575a0768f6SMichel Lespinasse } 358acf128d0SAndrea Arcangeli spin_unlock(&mm->page_table_lock); 3591da177e4SLinus Torvalds i++; 3601da177e4SLinus Torvalds pn = nd; 361d1af65d1SDavid Miller prev = vma->vm_start; 362d1af65d1SDavid Miller pend = vma->vm_end; 3631da177e4SLinus Torvalds } 3641da177e4SLinus Torvalds j = 0; 3655a0768f6SMichel Lespinasse for (nd = pn; nd; nd = rb_prev(nd)) 3661da177e4SLinus Torvalds j++; 3675a0768f6SMichel Lespinasse if (i != j) { 3688542bdfcSSasha Levin pr_emerg("backwards %d, forwards %d\n", j, i); 3695a0768f6SMichel Lespinasse bug = 1; 3701da177e4SLinus Torvalds } 3715a0768f6SMichel Lespinasse return bug ? -1 : i; 3721da177e4SLinus Torvalds } 3731da177e4SLinus Torvalds 374d3737187SMichel Lespinasse static void validate_mm_rb(struct rb_root *root, struct vm_area_struct *ignore) 375d3737187SMichel Lespinasse { 376d3737187SMichel Lespinasse struct rb_node *nd; 377d3737187SMichel Lespinasse 378d3737187SMichel Lespinasse for (nd = rb_first(root); nd; nd = rb_next(nd)) { 379d3737187SMichel Lespinasse struct vm_area_struct *vma; 380d3737187SMichel Lespinasse vma = rb_entry(nd, struct vm_area_struct, vm_rb); 38196dad67fSSasha Levin VM_BUG_ON_VMA(vma != ignore && 38296dad67fSSasha Levin vma->rb_subtree_gap != vma_compute_subtree_gap(vma), 38396dad67fSSasha Levin vma); 384d3737187SMichel Lespinasse } 3851da177e4SLinus Torvalds } 3861da177e4SLinus Torvalds 387eafd4dc4SRashika Kheria static void validate_mm(struct mm_struct *mm) 3881da177e4SLinus Torvalds { 3891da177e4SLinus Torvalds int bug = 0; 3901da177e4SLinus Torvalds int i = 0; 3915a0768f6SMichel Lespinasse unsigned long highest_address = 0; 392ed8ea815SMichel Lespinasse struct vm_area_struct *vma = mm->mmap; 393ff26f70fSAndrew Morton 394ed8ea815SMichel Lespinasse while (vma) { 39512352d3cSKonstantin Khlebnikov struct anon_vma *anon_vma = vma->anon_vma; 396ed8ea815SMichel Lespinasse struct anon_vma_chain *avc; 397ff26f70fSAndrew Morton 39812352d3cSKonstantin Khlebnikov if (anon_vma) { 39912352d3cSKonstantin Khlebnikov anon_vma_lock_read(anon_vma); 400ed8ea815SMichel Lespinasse list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) 401ed8ea815SMichel Lespinasse anon_vma_interval_tree_verify(avc); 40212352d3cSKonstantin Khlebnikov anon_vma_unlock_read(anon_vma); 40312352d3cSKonstantin Khlebnikov } 40412352d3cSKonstantin Khlebnikov 4051be7107fSHugh Dickins highest_address = vm_end_gap(vma); 406ed8ea815SMichel Lespinasse vma = vma->vm_next; 4071da177e4SLinus Torvalds i++; 4081da177e4SLinus Torvalds } 4095a0768f6SMichel Lespinasse if (i != mm->map_count) { 4108542bdfcSSasha Levin pr_emerg("map_count %d vm_next %d\n", mm->map_count, i); 4115a0768f6SMichel Lespinasse bug = 1; 4125a0768f6SMichel Lespinasse } 4135a0768f6SMichel Lespinasse if (highest_address != mm->highest_vm_end) { 4148542bdfcSSasha Levin pr_emerg("mm->highest_vm_end %lx, found %lx\n", 4155a0768f6SMichel Lespinasse mm->highest_vm_end, highest_address); 4165a0768f6SMichel Lespinasse bug = 1; 4175a0768f6SMichel Lespinasse } 418acf128d0SAndrea Arcangeli i = browse_rb(mm); 4195a0768f6SMichel Lespinasse if (i != mm->map_count) { 420ff26f70fSAndrew Morton if (i != -1) 4218542bdfcSSasha Levin pr_emerg("map_count %d rb %d\n", mm->map_count, i); 4225a0768f6SMichel Lespinasse bug = 1; 4235a0768f6SMichel Lespinasse } 42496dad67fSSasha Levin VM_BUG_ON_MM(bug, mm); 4251da177e4SLinus Torvalds } 4261da177e4SLinus Torvalds #else 427d3737187SMichel Lespinasse #define validate_mm_rb(root, ignore) do { } while (0) 4281da177e4SLinus Torvalds #define validate_mm(mm) do { } while (0) 4291da177e4SLinus Torvalds #endif 4301da177e4SLinus Torvalds 431d3737187SMichel Lespinasse RB_DECLARE_CALLBACKS(static, vma_gap_callbacks, struct vm_area_struct, vm_rb, 432d3737187SMichel Lespinasse unsigned long, rb_subtree_gap, vma_compute_subtree_gap) 433d3737187SMichel Lespinasse 434d3737187SMichel Lespinasse /* 435d3737187SMichel Lespinasse * Update augmented rbtree rb_subtree_gap values after vma->vm_start or 436d3737187SMichel Lespinasse * vma->vm_prev->vm_end values changed, without modifying the vma's position 437d3737187SMichel Lespinasse * in the rbtree. 438d3737187SMichel Lespinasse */ 439d3737187SMichel Lespinasse static void vma_gap_update(struct vm_area_struct *vma) 440d3737187SMichel Lespinasse { 441d3737187SMichel Lespinasse /* 442d3737187SMichel Lespinasse * As it turns out, RB_DECLARE_CALLBACKS() already created a callback 4438bb4e7a2SWei Yang * function that does exactly what we want. 444d3737187SMichel Lespinasse */ 445d3737187SMichel Lespinasse vma_gap_callbacks_propagate(&vma->vm_rb, NULL); 446d3737187SMichel Lespinasse } 447d3737187SMichel Lespinasse 448d3737187SMichel Lespinasse static inline void vma_rb_insert(struct vm_area_struct *vma, 449d3737187SMichel Lespinasse struct rb_root *root) 450d3737187SMichel Lespinasse { 451d3737187SMichel Lespinasse /* All rb_subtree_gap values must be consistent prior to insertion */ 452d3737187SMichel Lespinasse validate_mm_rb(root, NULL); 453d3737187SMichel Lespinasse 454d3737187SMichel Lespinasse rb_insert_augmented(&vma->vm_rb, root, &vma_gap_callbacks); 455d3737187SMichel Lespinasse } 456d3737187SMichel Lespinasse 4578f26e0b1SAndrea Arcangeli static void __vma_rb_erase(struct vm_area_struct *vma, struct rb_root *root) 4588f26e0b1SAndrea Arcangeli { 4598f26e0b1SAndrea Arcangeli /* 4608f26e0b1SAndrea Arcangeli * Note rb_erase_augmented is a fairly large inline function, 4618f26e0b1SAndrea Arcangeli * so make sure we instantiate it only once with our desired 4628f26e0b1SAndrea Arcangeli * augmented rbtree callbacks. 4638f26e0b1SAndrea Arcangeli */ 4648f26e0b1SAndrea Arcangeli rb_erase_augmented(&vma->vm_rb, root, &vma_gap_callbacks); 4658f26e0b1SAndrea Arcangeli } 4668f26e0b1SAndrea Arcangeli 4678f26e0b1SAndrea Arcangeli static __always_inline void vma_rb_erase_ignore(struct vm_area_struct *vma, 4688f26e0b1SAndrea Arcangeli struct rb_root *root, 4698f26e0b1SAndrea Arcangeli struct vm_area_struct *ignore) 4708f26e0b1SAndrea Arcangeli { 4718f26e0b1SAndrea Arcangeli /* 4728f26e0b1SAndrea Arcangeli * All rb_subtree_gap values must be consistent prior to erase, 4738f26e0b1SAndrea Arcangeli * with the possible exception of the "next" vma being erased if 4748f26e0b1SAndrea Arcangeli * next->vm_start was reduced. 4758f26e0b1SAndrea Arcangeli */ 4768f26e0b1SAndrea Arcangeli validate_mm_rb(root, ignore); 4778f26e0b1SAndrea Arcangeli 4788f26e0b1SAndrea Arcangeli __vma_rb_erase(vma, root); 4798f26e0b1SAndrea Arcangeli } 4808f26e0b1SAndrea Arcangeli 4818f26e0b1SAndrea Arcangeli static __always_inline void vma_rb_erase(struct vm_area_struct *vma, 4828f26e0b1SAndrea Arcangeli struct rb_root *root) 483d3737187SMichel Lespinasse { 484d3737187SMichel Lespinasse /* 485d3737187SMichel Lespinasse * All rb_subtree_gap values must be consistent prior to erase, 486d3737187SMichel Lespinasse * with the possible exception of the vma being erased. 487d3737187SMichel Lespinasse */ 488d3737187SMichel Lespinasse validate_mm_rb(root, vma); 489d3737187SMichel Lespinasse 4908f26e0b1SAndrea Arcangeli __vma_rb_erase(vma, root); 491d3737187SMichel Lespinasse } 492d3737187SMichel Lespinasse 493bf181b9fSMichel Lespinasse /* 494bf181b9fSMichel Lespinasse * vma has some anon_vma assigned, and is already inserted on that 495bf181b9fSMichel Lespinasse * anon_vma's interval trees. 496bf181b9fSMichel Lespinasse * 497bf181b9fSMichel Lespinasse * Before updating the vma's vm_start / vm_end / vm_pgoff fields, the 498bf181b9fSMichel Lespinasse * vma must be removed from the anon_vma's interval trees using 499bf181b9fSMichel Lespinasse * anon_vma_interval_tree_pre_update_vma(). 500bf181b9fSMichel Lespinasse * 501bf181b9fSMichel Lespinasse * After the update, the vma will be reinserted using 502bf181b9fSMichel Lespinasse * anon_vma_interval_tree_post_update_vma(). 503bf181b9fSMichel Lespinasse * 504bf181b9fSMichel Lespinasse * The entire update must be protected by exclusive mmap_sem and by 505bf181b9fSMichel Lespinasse * the root anon_vma's mutex. 506bf181b9fSMichel Lespinasse */ 507bf181b9fSMichel Lespinasse static inline void 508bf181b9fSMichel Lespinasse anon_vma_interval_tree_pre_update_vma(struct vm_area_struct *vma) 509bf181b9fSMichel Lespinasse { 510bf181b9fSMichel Lespinasse struct anon_vma_chain *avc; 511bf181b9fSMichel Lespinasse 512bf181b9fSMichel Lespinasse list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) 513bf181b9fSMichel Lespinasse anon_vma_interval_tree_remove(avc, &avc->anon_vma->rb_root); 514bf181b9fSMichel Lespinasse } 515bf181b9fSMichel Lespinasse 516bf181b9fSMichel Lespinasse static inline void 517bf181b9fSMichel Lespinasse anon_vma_interval_tree_post_update_vma(struct vm_area_struct *vma) 518bf181b9fSMichel Lespinasse { 519bf181b9fSMichel Lespinasse struct anon_vma_chain *avc; 520bf181b9fSMichel Lespinasse 521bf181b9fSMichel Lespinasse list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) 522bf181b9fSMichel Lespinasse anon_vma_interval_tree_insert(avc, &avc->anon_vma->rb_root); 523bf181b9fSMichel Lespinasse } 524bf181b9fSMichel Lespinasse 5256597d783SHugh Dickins static int find_vma_links(struct mm_struct *mm, unsigned long addr, 5266597d783SHugh Dickins unsigned long end, struct vm_area_struct **pprev, 5276597d783SHugh Dickins struct rb_node ***rb_link, struct rb_node **rb_parent) 5281da177e4SLinus Torvalds { 5291da177e4SLinus Torvalds struct rb_node **__rb_link, *__rb_parent, *rb_prev; 5301da177e4SLinus Torvalds 5311da177e4SLinus Torvalds __rb_link = &mm->mm_rb.rb_node; 5321da177e4SLinus Torvalds rb_prev = __rb_parent = NULL; 5331da177e4SLinus Torvalds 5341da177e4SLinus Torvalds while (*__rb_link) { 5351da177e4SLinus Torvalds struct vm_area_struct *vma_tmp; 5361da177e4SLinus Torvalds 5371da177e4SLinus Torvalds __rb_parent = *__rb_link; 5381da177e4SLinus Torvalds vma_tmp = rb_entry(__rb_parent, struct vm_area_struct, vm_rb); 5391da177e4SLinus Torvalds 5401da177e4SLinus Torvalds if (vma_tmp->vm_end > addr) { 5416597d783SHugh Dickins /* Fail if an existing vma overlaps the area */ 5426597d783SHugh Dickins if (vma_tmp->vm_start < end) 5436597d783SHugh Dickins return -ENOMEM; 5441da177e4SLinus Torvalds __rb_link = &__rb_parent->rb_left; 5451da177e4SLinus Torvalds } else { 5461da177e4SLinus Torvalds rb_prev = __rb_parent; 5471da177e4SLinus Torvalds __rb_link = &__rb_parent->rb_right; 5481da177e4SLinus Torvalds } 5491da177e4SLinus Torvalds } 5501da177e4SLinus Torvalds 5511da177e4SLinus Torvalds *pprev = NULL; 5521da177e4SLinus Torvalds if (rb_prev) 5531da177e4SLinus Torvalds *pprev = rb_entry(rb_prev, struct vm_area_struct, vm_rb); 5541da177e4SLinus Torvalds *rb_link = __rb_link; 5551da177e4SLinus Torvalds *rb_parent = __rb_parent; 5566597d783SHugh Dickins return 0; 5571da177e4SLinus Torvalds } 5581da177e4SLinus Torvalds 559e8420a8eSCyril Hrubis static unsigned long count_vma_pages_range(struct mm_struct *mm, 560e8420a8eSCyril Hrubis unsigned long addr, unsigned long end) 561e8420a8eSCyril Hrubis { 562e8420a8eSCyril Hrubis unsigned long nr_pages = 0; 563e8420a8eSCyril Hrubis struct vm_area_struct *vma; 564e8420a8eSCyril Hrubis 565e8420a8eSCyril Hrubis /* Find first overlaping mapping */ 566e8420a8eSCyril Hrubis vma = find_vma_intersection(mm, addr, end); 567e8420a8eSCyril Hrubis if (!vma) 568e8420a8eSCyril Hrubis return 0; 569e8420a8eSCyril Hrubis 570e8420a8eSCyril Hrubis nr_pages = (min(end, vma->vm_end) - 571e8420a8eSCyril Hrubis max(addr, vma->vm_start)) >> PAGE_SHIFT; 572e8420a8eSCyril Hrubis 573e8420a8eSCyril Hrubis /* Iterate over the rest of the overlaps */ 574e8420a8eSCyril Hrubis for (vma = vma->vm_next; vma; vma = vma->vm_next) { 575e8420a8eSCyril Hrubis unsigned long overlap_len; 576e8420a8eSCyril Hrubis 577e8420a8eSCyril Hrubis if (vma->vm_start > end) 578e8420a8eSCyril Hrubis break; 579e8420a8eSCyril Hrubis 580e8420a8eSCyril Hrubis overlap_len = min(end, vma->vm_end) - vma->vm_start; 581e8420a8eSCyril Hrubis nr_pages += overlap_len >> PAGE_SHIFT; 582e8420a8eSCyril Hrubis } 583e8420a8eSCyril Hrubis 584e8420a8eSCyril Hrubis return nr_pages; 585e8420a8eSCyril Hrubis } 586e8420a8eSCyril Hrubis 5871da177e4SLinus Torvalds void __vma_link_rb(struct mm_struct *mm, struct vm_area_struct *vma, 5881da177e4SLinus Torvalds struct rb_node **rb_link, struct rb_node *rb_parent) 5891da177e4SLinus Torvalds { 590d3737187SMichel Lespinasse /* Update tracking information for the gap following the new vma. */ 591d3737187SMichel Lespinasse if (vma->vm_next) 592d3737187SMichel Lespinasse vma_gap_update(vma->vm_next); 593d3737187SMichel Lespinasse else 5941be7107fSHugh Dickins mm->highest_vm_end = vm_end_gap(vma); 595d3737187SMichel Lespinasse 596d3737187SMichel Lespinasse /* 597d3737187SMichel Lespinasse * vma->vm_prev wasn't known when we followed the rbtree to find the 598d3737187SMichel Lespinasse * correct insertion point for that vma. As a result, we could not 599d3737187SMichel Lespinasse * update the vma vm_rb parents rb_subtree_gap values on the way down. 600d3737187SMichel Lespinasse * So, we first insert the vma with a zero rb_subtree_gap value 601d3737187SMichel Lespinasse * (to be consistent with what we did on the way down), and then 602d3737187SMichel Lespinasse * immediately update the gap to the correct value. Finally we 603d3737187SMichel Lespinasse * rebalance the rbtree after all augmented values have been set. 604d3737187SMichel Lespinasse */ 6051da177e4SLinus Torvalds rb_link_node(&vma->vm_rb, rb_parent, rb_link); 606d3737187SMichel Lespinasse vma->rb_subtree_gap = 0; 607d3737187SMichel Lespinasse vma_gap_update(vma); 608d3737187SMichel Lespinasse vma_rb_insert(vma, &mm->mm_rb); 6091da177e4SLinus Torvalds } 6101da177e4SLinus Torvalds 611cb8f488cSDenys Vlasenko static void __vma_link_file(struct vm_area_struct *vma) 6121da177e4SLinus Torvalds { 6131da177e4SLinus Torvalds struct file *file; 6141da177e4SLinus Torvalds 6151da177e4SLinus Torvalds file = vma->vm_file; 6161da177e4SLinus Torvalds if (file) { 6171da177e4SLinus Torvalds struct address_space *mapping = file->f_mapping; 6181da177e4SLinus Torvalds 6191da177e4SLinus Torvalds if (vma->vm_flags & VM_DENYWRITE) 620496ad9aaSAl Viro atomic_dec(&file_inode(file)->i_writecount); 6211da177e4SLinus Torvalds if (vma->vm_flags & VM_SHARED) 6224bb5f5d9SDavid Herrmann atomic_inc(&mapping->i_mmap_writable); 6231da177e4SLinus Torvalds 6241da177e4SLinus Torvalds flush_dcache_mmap_lock(mapping); 6256b2dbba8SMichel Lespinasse vma_interval_tree_insert(vma, &mapping->i_mmap); 6261da177e4SLinus Torvalds flush_dcache_mmap_unlock(mapping); 6271da177e4SLinus Torvalds } 6281da177e4SLinus Torvalds } 6291da177e4SLinus Torvalds 6301da177e4SLinus Torvalds static void 6311da177e4SLinus Torvalds __vma_link(struct mm_struct *mm, struct vm_area_struct *vma, 6321da177e4SLinus Torvalds struct vm_area_struct *prev, struct rb_node **rb_link, 6331da177e4SLinus Torvalds struct rb_node *rb_parent) 6341da177e4SLinus Torvalds { 6351da177e4SLinus Torvalds __vma_link_list(mm, vma, prev, rb_parent); 6361da177e4SLinus Torvalds __vma_link_rb(mm, vma, rb_link, rb_parent); 6371da177e4SLinus Torvalds } 6381da177e4SLinus Torvalds 6391da177e4SLinus Torvalds static void vma_link(struct mm_struct *mm, struct vm_area_struct *vma, 6401da177e4SLinus Torvalds struct vm_area_struct *prev, struct rb_node **rb_link, 6411da177e4SLinus Torvalds struct rb_node *rb_parent) 6421da177e4SLinus Torvalds { 6431da177e4SLinus Torvalds struct address_space *mapping = NULL; 6441da177e4SLinus Torvalds 64564ac4940SHuang Shijie if (vma->vm_file) { 6461da177e4SLinus Torvalds mapping = vma->vm_file->f_mapping; 64783cde9e8SDavidlohr Bueso i_mmap_lock_write(mapping); 64864ac4940SHuang Shijie } 6491da177e4SLinus Torvalds 6501da177e4SLinus Torvalds __vma_link(mm, vma, prev, rb_link, rb_parent); 6511da177e4SLinus Torvalds __vma_link_file(vma); 6521da177e4SLinus Torvalds 6531da177e4SLinus Torvalds if (mapping) 65483cde9e8SDavidlohr Bueso i_mmap_unlock_write(mapping); 6551da177e4SLinus Torvalds 6561da177e4SLinus Torvalds mm->map_count++; 6571da177e4SLinus Torvalds validate_mm(mm); 6581da177e4SLinus Torvalds } 6591da177e4SLinus Torvalds 6601da177e4SLinus Torvalds /* 66188f6b4c3SKautuk Consul * Helper for vma_adjust() in the split_vma insert case: insert a vma into the 6626b2dbba8SMichel Lespinasse * mm's list and rbtree. It has already been inserted into the interval tree. 6631da177e4SLinus Torvalds */ 66448aae425SZhenwenXu static void __insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma) 6651da177e4SLinus Torvalds { 6666597d783SHugh Dickins struct vm_area_struct *prev; 6671da177e4SLinus Torvalds struct rb_node **rb_link, *rb_parent; 6681da177e4SLinus Torvalds 6696597d783SHugh Dickins if (find_vma_links(mm, vma->vm_start, vma->vm_end, 6706597d783SHugh Dickins &prev, &rb_link, &rb_parent)) 6716597d783SHugh Dickins BUG(); 6721da177e4SLinus Torvalds __vma_link(mm, vma, prev, rb_link, rb_parent); 6731da177e4SLinus Torvalds mm->map_count++; 6741da177e4SLinus Torvalds } 6751da177e4SLinus Torvalds 676e86f15eeSAndrea Arcangeli static __always_inline void __vma_unlink_common(struct mm_struct *mm, 677e86f15eeSAndrea Arcangeli struct vm_area_struct *vma, 678e86f15eeSAndrea Arcangeli struct vm_area_struct *prev, 6798f26e0b1SAndrea Arcangeli bool has_prev, 6808f26e0b1SAndrea Arcangeli struct vm_area_struct *ignore) 6811da177e4SLinus Torvalds { 682d3737187SMichel Lespinasse struct vm_area_struct *next; 683297c5eeeSLinus Torvalds 6848f26e0b1SAndrea Arcangeli vma_rb_erase_ignore(vma, &mm->mm_rb, ignore); 685e86f15eeSAndrea Arcangeli next = vma->vm_next; 686e86f15eeSAndrea Arcangeli if (has_prev) 687e86f15eeSAndrea Arcangeli prev->vm_next = next; 688e86f15eeSAndrea Arcangeli else { 689e86f15eeSAndrea Arcangeli prev = vma->vm_prev; 690e86f15eeSAndrea Arcangeli if (prev) 691e86f15eeSAndrea Arcangeli prev->vm_next = next; 692e86f15eeSAndrea Arcangeli else 693e86f15eeSAndrea Arcangeli mm->mmap = next; 694e86f15eeSAndrea Arcangeli } 695297c5eeeSLinus Torvalds if (next) 696297c5eeeSLinus Torvalds next->vm_prev = prev; 697615d6e87SDavidlohr Bueso 698615d6e87SDavidlohr Bueso /* Kill the cache */ 699615d6e87SDavidlohr Bueso vmacache_invalidate(mm); 7001da177e4SLinus Torvalds } 7011da177e4SLinus Torvalds 702e86f15eeSAndrea Arcangeli static inline void __vma_unlink_prev(struct mm_struct *mm, 703e86f15eeSAndrea Arcangeli struct vm_area_struct *vma, 704e86f15eeSAndrea Arcangeli struct vm_area_struct *prev) 705e86f15eeSAndrea Arcangeli { 7068f26e0b1SAndrea Arcangeli __vma_unlink_common(mm, vma, prev, true, vma); 707e86f15eeSAndrea Arcangeli } 708e86f15eeSAndrea Arcangeli 7091da177e4SLinus Torvalds /* 7101da177e4SLinus Torvalds * We cannot adjust vm_start, vm_end, vm_pgoff fields of a vma that 7111da177e4SLinus Torvalds * is already present in an i_mmap tree without adjusting the tree. 7121da177e4SLinus Torvalds * The following helper function should be used when such adjustments 7131da177e4SLinus Torvalds * are necessary. The "insert" vma (if any) is to be inserted 7141da177e4SLinus Torvalds * before we drop the necessary locks. 7151da177e4SLinus Torvalds */ 716e86f15eeSAndrea Arcangeli int __vma_adjust(struct vm_area_struct *vma, unsigned long start, 717e86f15eeSAndrea Arcangeli unsigned long end, pgoff_t pgoff, struct vm_area_struct *insert, 718e86f15eeSAndrea Arcangeli struct vm_area_struct *expand) 7191da177e4SLinus Torvalds { 7201da177e4SLinus Torvalds struct mm_struct *mm = vma->vm_mm; 721e86f15eeSAndrea Arcangeli struct vm_area_struct *next = vma->vm_next, *orig_vma = vma; 7221da177e4SLinus Torvalds struct address_space *mapping = NULL; 723f808c13fSDavidlohr Bueso struct rb_root_cached *root = NULL; 724012f1800SRik van Riel struct anon_vma *anon_vma = NULL; 7251da177e4SLinus Torvalds struct file *file = vma->vm_file; 726d3737187SMichel Lespinasse bool start_changed = false, end_changed = false; 7271da177e4SLinus Torvalds long adjust_next = 0; 7281da177e4SLinus Torvalds int remove_next = 0; 7291da177e4SLinus Torvalds 7301da177e4SLinus Torvalds if (next && !insert) { 731734537c9SKirill A. Shutemov struct vm_area_struct *exporter = NULL, *importer = NULL; 732287d97acSLinus Torvalds 7331da177e4SLinus Torvalds if (end >= next->vm_end) { 7341da177e4SLinus Torvalds /* 7351da177e4SLinus Torvalds * vma expands, overlapping all the next, and 7361da177e4SLinus Torvalds * perhaps the one after too (mprotect case 6). 73786d12e47SAndrea Arcangeli * The only other cases that gets here are 738e86f15eeSAndrea Arcangeli * case 1, case 7 and case 8. 739e86f15eeSAndrea Arcangeli */ 740e86f15eeSAndrea Arcangeli if (next == expand) { 741e86f15eeSAndrea Arcangeli /* 742e86f15eeSAndrea Arcangeli * The only case where we don't expand "vma" 743e86f15eeSAndrea Arcangeli * and we expand "next" instead is case 8. 744e86f15eeSAndrea Arcangeli */ 745e86f15eeSAndrea Arcangeli VM_WARN_ON(end != next->vm_end); 746e86f15eeSAndrea Arcangeli /* 747e86f15eeSAndrea Arcangeli * remove_next == 3 means we're 748e86f15eeSAndrea Arcangeli * removing "vma" and that to do so we 749e86f15eeSAndrea Arcangeli * swapped "vma" and "next". 750e86f15eeSAndrea Arcangeli */ 751e86f15eeSAndrea Arcangeli remove_next = 3; 752e86f15eeSAndrea Arcangeli VM_WARN_ON(file != next->vm_file); 753e86f15eeSAndrea Arcangeli swap(vma, next); 754e86f15eeSAndrea Arcangeli } else { 755e86f15eeSAndrea Arcangeli VM_WARN_ON(expand != vma); 756e86f15eeSAndrea Arcangeli /* 757e86f15eeSAndrea Arcangeli * case 1, 6, 7, remove_next == 2 is case 6, 758e86f15eeSAndrea Arcangeli * remove_next == 1 is case 1 or 7. 7591da177e4SLinus Torvalds */ 760734537c9SKirill A. Shutemov remove_next = 1 + (end > next->vm_end); 761e86f15eeSAndrea Arcangeli VM_WARN_ON(remove_next == 2 && 762e86f15eeSAndrea Arcangeli end != next->vm_next->vm_end); 763e86f15eeSAndrea Arcangeli VM_WARN_ON(remove_next == 1 && 764e86f15eeSAndrea Arcangeli end != next->vm_end); 765e86f15eeSAndrea Arcangeli /* trim end to next, for case 6 first pass */ 7661da177e4SLinus Torvalds end = next->vm_end; 767e86f15eeSAndrea Arcangeli } 768e86f15eeSAndrea Arcangeli 769287d97acSLinus Torvalds exporter = next; 7701da177e4SLinus Torvalds importer = vma; 771734537c9SKirill A. Shutemov 772734537c9SKirill A. Shutemov /* 773734537c9SKirill A. Shutemov * If next doesn't have anon_vma, import from vma after 774734537c9SKirill A. Shutemov * next, if the vma overlaps with it. 775734537c9SKirill A. Shutemov */ 77697a42cd4SAndrea Arcangeli if (remove_next == 2 && !next->anon_vma) 777734537c9SKirill A. Shutemov exporter = next->vm_next; 778734537c9SKirill A. Shutemov 7791da177e4SLinus Torvalds } else if (end > next->vm_start) { 7801da177e4SLinus Torvalds /* 7811da177e4SLinus Torvalds * vma expands, overlapping part of the next: 7821da177e4SLinus Torvalds * mprotect case 5 shifting the boundary up. 7831da177e4SLinus Torvalds */ 7841da177e4SLinus Torvalds adjust_next = (end - next->vm_start) >> PAGE_SHIFT; 785287d97acSLinus Torvalds exporter = next; 7861da177e4SLinus Torvalds importer = vma; 787e86f15eeSAndrea Arcangeli VM_WARN_ON(expand != importer); 7881da177e4SLinus Torvalds } else if (end < vma->vm_end) { 7891da177e4SLinus Torvalds /* 7901da177e4SLinus Torvalds * vma shrinks, and !insert tells it's not 7911da177e4SLinus Torvalds * split_vma inserting another: so it must be 7921da177e4SLinus Torvalds * mprotect case 4 shifting the boundary down. 7931da177e4SLinus Torvalds */ 7941da177e4SLinus Torvalds adjust_next = -((vma->vm_end - end) >> PAGE_SHIFT); 795287d97acSLinus Torvalds exporter = vma; 7961da177e4SLinus Torvalds importer = next; 797e86f15eeSAndrea Arcangeli VM_WARN_ON(expand != importer); 7981da177e4SLinus Torvalds } 7991da177e4SLinus Torvalds 8005beb4930SRik van Riel /* 8015beb4930SRik van Riel * Easily overlooked: when mprotect shifts the boundary, 8025beb4930SRik van Riel * make sure the expanding vma has anon_vma set if the 8035beb4930SRik van Riel * shrinking vma had, to cover any anon pages imported. 8045beb4930SRik van Riel */ 805287d97acSLinus Torvalds if (exporter && exporter->anon_vma && !importer->anon_vma) { 806c4ea95d7SDaniel Forrest int error; 807c4ea95d7SDaniel Forrest 808287d97acSLinus Torvalds importer->anon_vma = exporter->anon_vma; 809b800c91aSKonstantin Khlebnikov error = anon_vma_clone(importer, exporter); 8103fe89b3eSLeon Yu if (error) 811b800c91aSKonstantin Khlebnikov return error; 812b800c91aSKonstantin Khlebnikov } 8135beb4930SRik van Riel } 814734537c9SKirill A. Shutemov again: 815e86f15eeSAndrea Arcangeli vma_adjust_trans_huge(orig_vma, start, end, adjust_next); 81637f9f559SKirill A. Shutemov 8171da177e4SLinus Torvalds if (file) { 8181da177e4SLinus Torvalds mapping = file->f_mapping; 8191da177e4SLinus Torvalds root = &mapping->i_mmap; 820cbc91f71SSrikar Dronamraju uprobe_munmap(vma, vma->vm_start, vma->vm_end); 821682968e0SSrikar Dronamraju 822682968e0SSrikar Dronamraju if (adjust_next) 82327ba0644SKirill A. Shutemov uprobe_munmap(next, next->vm_start, next->vm_end); 824682968e0SSrikar Dronamraju 82583cde9e8SDavidlohr Bueso i_mmap_lock_write(mapping); 8261da177e4SLinus Torvalds if (insert) { 8271da177e4SLinus Torvalds /* 8286b2dbba8SMichel Lespinasse * Put into interval tree now, so instantiated pages 8291da177e4SLinus Torvalds * are visible to arm/parisc __flush_dcache_page 8301da177e4SLinus Torvalds * throughout; but we cannot insert into address 8311da177e4SLinus Torvalds * space until vma start or end is updated. 8321da177e4SLinus Torvalds */ 8331da177e4SLinus Torvalds __vma_link_file(insert); 8341da177e4SLinus Torvalds } 8351da177e4SLinus Torvalds } 8361da177e4SLinus Torvalds 837012f1800SRik van Riel anon_vma = vma->anon_vma; 838bf181b9fSMichel Lespinasse if (!anon_vma && adjust_next) 839bf181b9fSMichel Lespinasse anon_vma = next->anon_vma; 840bf181b9fSMichel Lespinasse if (anon_vma) { 841e86f15eeSAndrea Arcangeli VM_WARN_ON(adjust_next && next->anon_vma && 842e86f15eeSAndrea Arcangeli anon_vma != next->anon_vma); 8434fc3f1d6SIngo Molnar anon_vma_lock_write(anon_vma); 844bf181b9fSMichel Lespinasse anon_vma_interval_tree_pre_update_vma(vma); 845bf181b9fSMichel Lespinasse if (adjust_next) 846bf181b9fSMichel Lespinasse anon_vma_interval_tree_pre_update_vma(next); 847bf181b9fSMichel Lespinasse } 848012f1800SRik van Riel 8491da177e4SLinus Torvalds if (root) { 8501da177e4SLinus Torvalds flush_dcache_mmap_lock(mapping); 8516b2dbba8SMichel Lespinasse vma_interval_tree_remove(vma, root); 8521da177e4SLinus Torvalds if (adjust_next) 8536b2dbba8SMichel Lespinasse vma_interval_tree_remove(next, root); 8541da177e4SLinus Torvalds } 8551da177e4SLinus Torvalds 856d3737187SMichel Lespinasse if (start != vma->vm_start) { 8571da177e4SLinus Torvalds vma->vm_start = start; 858d3737187SMichel Lespinasse start_changed = true; 859d3737187SMichel Lespinasse } 860d3737187SMichel Lespinasse if (end != vma->vm_end) { 8611da177e4SLinus Torvalds vma->vm_end = end; 862d3737187SMichel Lespinasse end_changed = true; 863d3737187SMichel Lespinasse } 8641da177e4SLinus Torvalds vma->vm_pgoff = pgoff; 8651da177e4SLinus Torvalds if (adjust_next) { 8661da177e4SLinus Torvalds next->vm_start += adjust_next << PAGE_SHIFT; 8671da177e4SLinus Torvalds next->vm_pgoff += adjust_next; 8681da177e4SLinus Torvalds } 8691da177e4SLinus Torvalds 8701da177e4SLinus Torvalds if (root) { 8711da177e4SLinus Torvalds if (adjust_next) 8726b2dbba8SMichel Lespinasse vma_interval_tree_insert(next, root); 8736b2dbba8SMichel Lespinasse vma_interval_tree_insert(vma, root); 8741da177e4SLinus Torvalds flush_dcache_mmap_unlock(mapping); 8751da177e4SLinus Torvalds } 8761da177e4SLinus Torvalds 8771da177e4SLinus Torvalds if (remove_next) { 8781da177e4SLinus Torvalds /* 8791da177e4SLinus Torvalds * vma_merge has merged next into vma, and needs 8801da177e4SLinus Torvalds * us to remove next before dropping the locks. 8811da177e4SLinus Torvalds */ 882e86f15eeSAndrea Arcangeli if (remove_next != 3) 883e86f15eeSAndrea Arcangeli __vma_unlink_prev(mm, next, vma); 884e86f15eeSAndrea Arcangeli else 8858f26e0b1SAndrea Arcangeli /* 8868f26e0b1SAndrea Arcangeli * vma is not before next if they've been 8878f26e0b1SAndrea Arcangeli * swapped. 8888f26e0b1SAndrea Arcangeli * 8898f26e0b1SAndrea Arcangeli * pre-swap() next->vm_start was reduced so 8908f26e0b1SAndrea Arcangeli * tell validate_mm_rb to ignore pre-swap() 8918f26e0b1SAndrea Arcangeli * "next" (which is stored in post-swap() 8928f26e0b1SAndrea Arcangeli * "vma"). 8938f26e0b1SAndrea Arcangeli */ 8948f26e0b1SAndrea Arcangeli __vma_unlink_common(mm, next, NULL, false, vma); 8951da177e4SLinus Torvalds if (file) 8961da177e4SLinus Torvalds __remove_shared_vm_struct(next, file, mapping); 8971da177e4SLinus Torvalds } else if (insert) { 8981da177e4SLinus Torvalds /* 8991da177e4SLinus Torvalds * split_vma has split insert from vma, and needs 9001da177e4SLinus Torvalds * us to insert it before dropping the locks 9011da177e4SLinus Torvalds * (it may either follow vma or precede it). 9021da177e4SLinus Torvalds */ 9031da177e4SLinus Torvalds __insert_vm_struct(mm, insert); 904d3737187SMichel Lespinasse } else { 905d3737187SMichel Lespinasse if (start_changed) 906d3737187SMichel Lespinasse vma_gap_update(vma); 907d3737187SMichel Lespinasse if (end_changed) { 908d3737187SMichel Lespinasse if (!next) 9091be7107fSHugh Dickins mm->highest_vm_end = vm_end_gap(vma); 910d3737187SMichel Lespinasse else if (!adjust_next) 911d3737187SMichel Lespinasse vma_gap_update(next); 912d3737187SMichel Lespinasse } 9131da177e4SLinus Torvalds } 9141da177e4SLinus Torvalds 915bf181b9fSMichel Lespinasse if (anon_vma) { 916bf181b9fSMichel Lespinasse anon_vma_interval_tree_post_update_vma(vma); 917bf181b9fSMichel Lespinasse if (adjust_next) 918bf181b9fSMichel Lespinasse anon_vma_interval_tree_post_update_vma(next); 91908b52706SKonstantin Khlebnikov anon_vma_unlock_write(anon_vma); 920bf181b9fSMichel Lespinasse } 9211da177e4SLinus Torvalds if (mapping) 92283cde9e8SDavidlohr Bueso i_mmap_unlock_write(mapping); 9231da177e4SLinus Torvalds 9242b144498SSrikar Dronamraju if (root) { 9257b2d81d4SIngo Molnar uprobe_mmap(vma); 9262b144498SSrikar Dronamraju 9272b144498SSrikar Dronamraju if (adjust_next) 9287b2d81d4SIngo Molnar uprobe_mmap(next); 9292b144498SSrikar Dronamraju } 9302b144498SSrikar Dronamraju 9311da177e4SLinus Torvalds if (remove_next) { 932925d1c40SMatt Helsley if (file) { 933cbc91f71SSrikar Dronamraju uprobe_munmap(next, next->vm_start, next->vm_end); 9341da177e4SLinus Torvalds fput(file); 935925d1c40SMatt Helsley } 9365beb4930SRik van Riel if (next->anon_vma) 9375beb4930SRik van Riel anon_vma_merge(vma, next); 9381da177e4SLinus Torvalds mm->map_count--; 9393964acd0SOleg Nesterov mpol_put(vma_policy(next)); 9403928d4f5SLinus Torvalds vm_area_free(next); 9411da177e4SLinus Torvalds /* 9421da177e4SLinus Torvalds * In mprotect's case 6 (see comments on vma_merge), 9431da177e4SLinus Torvalds * we must remove another next too. It would clutter 9441da177e4SLinus Torvalds * up the code too much to do both in one go. 9451da177e4SLinus Torvalds */ 946e86f15eeSAndrea Arcangeli if (remove_next != 3) { 947e86f15eeSAndrea Arcangeli /* 948e86f15eeSAndrea Arcangeli * If "next" was removed and vma->vm_end was 949e86f15eeSAndrea Arcangeli * expanded (up) over it, in turn 950e86f15eeSAndrea Arcangeli * "next->vm_prev->vm_end" changed and the 951e86f15eeSAndrea Arcangeli * "vma->vm_next" gap must be updated. 952e86f15eeSAndrea Arcangeli */ 9531da177e4SLinus Torvalds next = vma->vm_next; 954e86f15eeSAndrea Arcangeli } else { 955e86f15eeSAndrea Arcangeli /* 956e86f15eeSAndrea Arcangeli * For the scope of the comment "next" and 957e86f15eeSAndrea Arcangeli * "vma" considered pre-swap(): if "vma" was 958e86f15eeSAndrea Arcangeli * removed, next->vm_start was expanded (down) 959e86f15eeSAndrea Arcangeli * over it and the "next" gap must be updated. 960e86f15eeSAndrea Arcangeli * Because of the swap() the post-swap() "vma" 961e86f15eeSAndrea Arcangeli * actually points to pre-swap() "next" 962e86f15eeSAndrea Arcangeli * (post-swap() "next" as opposed is now a 963e86f15eeSAndrea Arcangeli * dangling pointer). 964e86f15eeSAndrea Arcangeli */ 965e86f15eeSAndrea Arcangeli next = vma; 966e86f15eeSAndrea Arcangeli } 967734537c9SKirill A. Shutemov if (remove_next == 2) { 968734537c9SKirill A. Shutemov remove_next = 1; 969734537c9SKirill A. Shutemov end = next->vm_end; 9701da177e4SLinus Torvalds goto again; 971734537c9SKirill A. Shutemov } 972d3737187SMichel Lespinasse else if (next) 973d3737187SMichel Lespinasse vma_gap_update(next); 974fb8c41e9SAndrea Arcangeli else { 975fb8c41e9SAndrea Arcangeli /* 976fb8c41e9SAndrea Arcangeli * If remove_next == 2 we obviously can't 977fb8c41e9SAndrea Arcangeli * reach this path. 978fb8c41e9SAndrea Arcangeli * 979fb8c41e9SAndrea Arcangeli * If remove_next == 3 we can't reach this 980fb8c41e9SAndrea Arcangeli * path because pre-swap() next is always not 981fb8c41e9SAndrea Arcangeli * NULL. pre-swap() "next" is not being 982fb8c41e9SAndrea Arcangeli * removed and its next->vm_end is not altered 983fb8c41e9SAndrea Arcangeli * (and furthermore "end" already matches 984fb8c41e9SAndrea Arcangeli * next->vm_end in remove_next == 3). 985fb8c41e9SAndrea Arcangeli * 986fb8c41e9SAndrea Arcangeli * We reach this only in the remove_next == 1 987fb8c41e9SAndrea Arcangeli * case if the "next" vma that was removed was 988fb8c41e9SAndrea Arcangeli * the highest vma of the mm. However in such 989fb8c41e9SAndrea Arcangeli * case next->vm_end == "end" and the extended 990fb8c41e9SAndrea Arcangeli * "vma" has vma->vm_end == next->vm_end so 991fb8c41e9SAndrea Arcangeli * mm->highest_vm_end doesn't need any update 992fb8c41e9SAndrea Arcangeli * in remove_next == 1 case. 993fb8c41e9SAndrea Arcangeli */ 9941be7107fSHugh Dickins VM_WARN_ON(mm->highest_vm_end != vm_end_gap(vma)); 995fb8c41e9SAndrea Arcangeli } 9961da177e4SLinus Torvalds } 9972b144498SSrikar Dronamraju if (insert && file) 9987b2d81d4SIngo Molnar uprobe_mmap(insert); 9991da177e4SLinus Torvalds 10001da177e4SLinus Torvalds validate_mm(mm); 10015beb4930SRik van Riel 10025beb4930SRik van Riel return 0; 10031da177e4SLinus Torvalds } 10041da177e4SLinus Torvalds 10051da177e4SLinus Torvalds /* 10061da177e4SLinus Torvalds * If the vma has a ->close operation then the driver probably needs to release 10071da177e4SLinus Torvalds * per-vma resources, so we don't attempt to merge those. 10081da177e4SLinus Torvalds */ 10091da177e4SLinus Torvalds static inline int is_mergeable_vma(struct vm_area_struct *vma, 101019a809afSAndrea Arcangeli struct file *file, unsigned long vm_flags, 101119a809afSAndrea Arcangeli struct vm_userfaultfd_ctx vm_userfaultfd_ctx) 10121da177e4SLinus Torvalds { 101334228d47SCyrill Gorcunov /* 101434228d47SCyrill Gorcunov * VM_SOFTDIRTY should not prevent from VMA merging, if we 101534228d47SCyrill Gorcunov * match the flags but dirty bit -- the caller should mark 101634228d47SCyrill Gorcunov * merged VMA as dirty. If dirty bit won't be excluded from 10178bb4e7a2SWei Yang * comparison, we increase pressure on the memory system forcing 101834228d47SCyrill Gorcunov * the kernel to generate new VMAs when old one could be 101934228d47SCyrill Gorcunov * extended instead. 102034228d47SCyrill Gorcunov */ 102134228d47SCyrill Gorcunov if ((vma->vm_flags ^ vm_flags) & ~VM_SOFTDIRTY) 10221da177e4SLinus Torvalds return 0; 10231da177e4SLinus Torvalds if (vma->vm_file != file) 10241da177e4SLinus Torvalds return 0; 10251da177e4SLinus Torvalds if (vma->vm_ops && vma->vm_ops->close) 10261da177e4SLinus Torvalds return 0; 102719a809afSAndrea Arcangeli if (!is_mergeable_vm_userfaultfd_ctx(vma, vm_userfaultfd_ctx)) 102819a809afSAndrea Arcangeli return 0; 10291da177e4SLinus Torvalds return 1; 10301da177e4SLinus Torvalds } 10311da177e4SLinus Torvalds 10321da177e4SLinus Torvalds static inline int is_mergeable_anon_vma(struct anon_vma *anon_vma1, 1033965f55deSShaohua Li struct anon_vma *anon_vma2, 1034965f55deSShaohua Li struct vm_area_struct *vma) 10351da177e4SLinus Torvalds { 1036965f55deSShaohua Li /* 1037965f55deSShaohua Li * The list_is_singular() test is to avoid merging VMA cloned from 1038965f55deSShaohua Li * parents. This can improve scalability caused by anon_vma lock. 1039965f55deSShaohua Li */ 1040965f55deSShaohua Li if ((!anon_vma1 || !anon_vma2) && (!vma || 1041965f55deSShaohua Li list_is_singular(&vma->anon_vma_chain))) 1042965f55deSShaohua Li return 1; 1043965f55deSShaohua Li return anon_vma1 == anon_vma2; 10441da177e4SLinus Torvalds } 10451da177e4SLinus Torvalds 10461da177e4SLinus Torvalds /* 10471da177e4SLinus Torvalds * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff) 10481da177e4SLinus Torvalds * in front of (at a lower virtual address and file offset than) the vma. 10491da177e4SLinus Torvalds * 10501da177e4SLinus Torvalds * We cannot merge two vmas if they have differently assigned (non-NULL) 10511da177e4SLinus Torvalds * anon_vmas, nor if same anon_vma is assigned but offsets incompatible. 10521da177e4SLinus Torvalds * 10531da177e4SLinus Torvalds * We don't check here for the merged mmap wrapping around the end of pagecache 10541da177e4SLinus Torvalds * indices (16TB on ia32) because do_mmap_pgoff() does not permit mmap's which 10551da177e4SLinus Torvalds * wrap, nor mmaps which cover the final page at index -1UL. 10561da177e4SLinus Torvalds */ 10571da177e4SLinus Torvalds static int 10581da177e4SLinus Torvalds can_vma_merge_before(struct vm_area_struct *vma, unsigned long vm_flags, 105919a809afSAndrea Arcangeli struct anon_vma *anon_vma, struct file *file, 106019a809afSAndrea Arcangeli pgoff_t vm_pgoff, 106119a809afSAndrea Arcangeli struct vm_userfaultfd_ctx vm_userfaultfd_ctx) 10621da177e4SLinus Torvalds { 106319a809afSAndrea Arcangeli if (is_mergeable_vma(vma, file, vm_flags, vm_userfaultfd_ctx) && 1064965f55deSShaohua Li is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) { 10651da177e4SLinus Torvalds if (vma->vm_pgoff == vm_pgoff) 10661da177e4SLinus Torvalds return 1; 10671da177e4SLinus Torvalds } 10681da177e4SLinus Torvalds return 0; 10691da177e4SLinus Torvalds } 10701da177e4SLinus Torvalds 10711da177e4SLinus Torvalds /* 10721da177e4SLinus Torvalds * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff) 10731da177e4SLinus Torvalds * beyond (at a higher virtual address and file offset than) the vma. 10741da177e4SLinus Torvalds * 10751da177e4SLinus Torvalds * We cannot merge two vmas if they have differently assigned (non-NULL) 10761da177e4SLinus Torvalds * anon_vmas, nor if same anon_vma is assigned but offsets incompatible. 10771da177e4SLinus Torvalds */ 10781da177e4SLinus Torvalds static int 10791da177e4SLinus Torvalds can_vma_merge_after(struct vm_area_struct *vma, unsigned long vm_flags, 108019a809afSAndrea Arcangeli struct anon_vma *anon_vma, struct file *file, 108119a809afSAndrea Arcangeli pgoff_t vm_pgoff, 108219a809afSAndrea Arcangeli struct vm_userfaultfd_ctx vm_userfaultfd_ctx) 10831da177e4SLinus Torvalds { 108419a809afSAndrea Arcangeli if (is_mergeable_vma(vma, file, vm_flags, vm_userfaultfd_ctx) && 1085965f55deSShaohua Li is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) { 10861da177e4SLinus Torvalds pgoff_t vm_pglen; 1087d6e93217SLibin vm_pglen = vma_pages(vma); 10881da177e4SLinus Torvalds if (vma->vm_pgoff + vm_pglen == vm_pgoff) 10891da177e4SLinus Torvalds return 1; 10901da177e4SLinus Torvalds } 10911da177e4SLinus Torvalds return 0; 10921da177e4SLinus Torvalds } 10931da177e4SLinus Torvalds 10941da177e4SLinus Torvalds /* 10951da177e4SLinus Torvalds * Given a mapping request (addr,end,vm_flags,file,pgoff), figure out 10961da177e4SLinus Torvalds * whether that can be merged with its predecessor or its successor. 10971da177e4SLinus Torvalds * Or both (it neatly fills a hole). 10981da177e4SLinus Torvalds * 10991da177e4SLinus Torvalds * In most cases - when called for mmap, brk or mremap - [addr,end) is 11001da177e4SLinus Torvalds * certain not to be mapped by the time vma_merge is called; but when 11011da177e4SLinus Torvalds * called for mprotect, it is certain to be already mapped (either at 11021da177e4SLinus Torvalds * an offset within prev, or at the start of next), and the flags of 11031da177e4SLinus Torvalds * this area are about to be changed to vm_flags - and the no-change 11041da177e4SLinus Torvalds * case has already been eliminated. 11051da177e4SLinus Torvalds * 11061da177e4SLinus Torvalds * The following mprotect cases have to be considered, where AAAA is 11071da177e4SLinus Torvalds * the area passed down from mprotect_fixup, never extending beyond one 11081da177e4SLinus Torvalds * vma, PPPPPP is the prev vma specified, and NNNNNN the next vma after: 11091da177e4SLinus Torvalds * 11101da177e4SLinus Torvalds * AAAA AAAA AAAA AAAA 11111da177e4SLinus Torvalds * PPPPPPNNNNNN PPPPPPNNNNNN PPPPPPNNNNNN PPPPNNNNXXXX 11121da177e4SLinus Torvalds * cannot merge might become might become might become 11131da177e4SLinus Torvalds * PPNNNNNNNNNN PPPPPPPPPPNN PPPPPPPPPPPP 6 or 11141da177e4SLinus Torvalds * mmap, brk or case 4 below case 5 below PPPPPPPPXXXX 7 or 1115e86f15eeSAndrea Arcangeli * mremap move: PPPPXXXXXXXX 8 11161da177e4SLinus Torvalds * AAAA 11171da177e4SLinus Torvalds * PPPP NNNN PPPPPPPPPPPP PPPPPPPPNNNN PPPPNNNNNNNN 11181da177e4SLinus Torvalds * might become case 1 below case 2 below case 3 below 11191da177e4SLinus Torvalds * 11208bb4e7a2SWei Yang * It is important for case 8 that the vma NNNN overlapping the 1121e86f15eeSAndrea Arcangeli * region AAAA is never going to extended over XXXX. Instead XXXX must 1122e86f15eeSAndrea Arcangeli * be extended in region AAAA and NNNN must be removed. This way in 1123e86f15eeSAndrea Arcangeli * all cases where vma_merge succeeds, the moment vma_adjust drops the 1124e86f15eeSAndrea Arcangeli * rmap_locks, the properties of the merged vma will be already 1125e86f15eeSAndrea Arcangeli * correct for the whole merged range. Some of those properties like 1126e86f15eeSAndrea Arcangeli * vm_page_prot/vm_flags may be accessed by rmap_walks and they must 1127e86f15eeSAndrea Arcangeli * be correct for the whole merged range immediately after the 1128e86f15eeSAndrea Arcangeli * rmap_locks are released. Otherwise if XXXX would be removed and 1129e86f15eeSAndrea Arcangeli * NNNN would be extended over the XXXX range, remove_migration_ptes 1130e86f15eeSAndrea Arcangeli * or other rmap walkers (if working on addresses beyond the "end" 1131e86f15eeSAndrea Arcangeli * parameter) may establish ptes with the wrong permissions of NNNN 1132e86f15eeSAndrea Arcangeli * instead of the right permissions of XXXX. 11331da177e4SLinus Torvalds */ 11341da177e4SLinus Torvalds struct vm_area_struct *vma_merge(struct mm_struct *mm, 11351da177e4SLinus Torvalds struct vm_area_struct *prev, unsigned long addr, 11361da177e4SLinus Torvalds unsigned long end, unsigned long vm_flags, 11371da177e4SLinus Torvalds struct anon_vma *anon_vma, struct file *file, 113819a809afSAndrea Arcangeli pgoff_t pgoff, struct mempolicy *policy, 113919a809afSAndrea Arcangeli struct vm_userfaultfd_ctx vm_userfaultfd_ctx) 11401da177e4SLinus Torvalds { 11411da177e4SLinus Torvalds pgoff_t pglen = (end - addr) >> PAGE_SHIFT; 11421da177e4SLinus Torvalds struct vm_area_struct *area, *next; 11435beb4930SRik van Riel int err; 11441da177e4SLinus Torvalds 11451da177e4SLinus Torvalds /* 11461da177e4SLinus Torvalds * We later require that vma->vm_flags == vm_flags, 11471da177e4SLinus Torvalds * so this tests vma->vm_flags & VM_SPECIAL, too. 11481da177e4SLinus Torvalds */ 11491da177e4SLinus Torvalds if (vm_flags & VM_SPECIAL) 11501da177e4SLinus Torvalds return NULL; 11511da177e4SLinus Torvalds 11521da177e4SLinus Torvalds if (prev) 11531da177e4SLinus Torvalds next = prev->vm_next; 11541da177e4SLinus Torvalds else 11551da177e4SLinus Torvalds next = mm->mmap; 11561da177e4SLinus Torvalds area = next; 1157e86f15eeSAndrea Arcangeli if (area && area->vm_end == end) /* cases 6, 7, 8 */ 11581da177e4SLinus Torvalds next = next->vm_next; 11591da177e4SLinus Torvalds 1160e86f15eeSAndrea Arcangeli /* verify some invariant that must be enforced by the caller */ 1161e86f15eeSAndrea Arcangeli VM_WARN_ON(prev && addr <= prev->vm_start); 1162e86f15eeSAndrea Arcangeli VM_WARN_ON(area && end > area->vm_end); 1163e86f15eeSAndrea Arcangeli VM_WARN_ON(addr >= end); 1164e86f15eeSAndrea Arcangeli 11651da177e4SLinus Torvalds /* 11661da177e4SLinus Torvalds * Can it merge with the predecessor? 11671da177e4SLinus Torvalds */ 11681da177e4SLinus Torvalds if (prev && prev->vm_end == addr && 11691da177e4SLinus Torvalds mpol_equal(vma_policy(prev), policy) && 11701da177e4SLinus Torvalds can_vma_merge_after(prev, vm_flags, 117119a809afSAndrea Arcangeli anon_vma, file, pgoff, 117219a809afSAndrea Arcangeli vm_userfaultfd_ctx)) { 11731da177e4SLinus Torvalds /* 11741da177e4SLinus Torvalds * OK, it can. Can we now merge in the successor as well? 11751da177e4SLinus Torvalds */ 11761da177e4SLinus Torvalds if (next && end == next->vm_start && 11771da177e4SLinus Torvalds mpol_equal(policy, vma_policy(next)) && 11781da177e4SLinus Torvalds can_vma_merge_before(next, vm_flags, 117919a809afSAndrea Arcangeli anon_vma, file, 118019a809afSAndrea Arcangeli pgoff+pglen, 118119a809afSAndrea Arcangeli vm_userfaultfd_ctx) && 11821da177e4SLinus Torvalds is_mergeable_anon_vma(prev->anon_vma, 1183965f55deSShaohua Li next->anon_vma, NULL)) { 11841da177e4SLinus Torvalds /* cases 1, 6 */ 1185e86f15eeSAndrea Arcangeli err = __vma_adjust(prev, prev->vm_start, 1186e86f15eeSAndrea Arcangeli next->vm_end, prev->vm_pgoff, NULL, 1187e86f15eeSAndrea Arcangeli prev); 11881da177e4SLinus Torvalds } else /* cases 2, 5, 7 */ 1189e86f15eeSAndrea Arcangeli err = __vma_adjust(prev, prev->vm_start, 1190e86f15eeSAndrea Arcangeli end, prev->vm_pgoff, NULL, prev); 11915beb4930SRik van Riel if (err) 11925beb4930SRik van Riel return NULL; 11936d50e60cSDavid Rientjes khugepaged_enter_vma_merge(prev, vm_flags); 11941da177e4SLinus Torvalds return prev; 11951da177e4SLinus Torvalds } 11961da177e4SLinus Torvalds 11971da177e4SLinus Torvalds /* 11981da177e4SLinus Torvalds * Can this new request be merged in front of next? 11991da177e4SLinus Torvalds */ 12001da177e4SLinus Torvalds if (next && end == next->vm_start && 12011da177e4SLinus Torvalds mpol_equal(policy, vma_policy(next)) && 12021da177e4SLinus Torvalds can_vma_merge_before(next, vm_flags, 120319a809afSAndrea Arcangeli anon_vma, file, pgoff+pglen, 120419a809afSAndrea Arcangeli vm_userfaultfd_ctx)) { 12051da177e4SLinus Torvalds if (prev && addr < prev->vm_end) /* case 4 */ 1206e86f15eeSAndrea Arcangeli err = __vma_adjust(prev, prev->vm_start, 1207e86f15eeSAndrea Arcangeli addr, prev->vm_pgoff, NULL, next); 1208e86f15eeSAndrea Arcangeli else { /* cases 3, 8 */ 1209e86f15eeSAndrea Arcangeli err = __vma_adjust(area, addr, next->vm_end, 1210e86f15eeSAndrea Arcangeli next->vm_pgoff - pglen, NULL, next); 1211e86f15eeSAndrea Arcangeli /* 1212e86f15eeSAndrea Arcangeli * In case 3 area is already equal to next and 1213e86f15eeSAndrea Arcangeli * this is a noop, but in case 8 "area" has 1214e86f15eeSAndrea Arcangeli * been removed and next was expanded over it. 1215e86f15eeSAndrea Arcangeli */ 1216e86f15eeSAndrea Arcangeli area = next; 1217e86f15eeSAndrea Arcangeli } 12185beb4930SRik van Riel if (err) 12195beb4930SRik van Riel return NULL; 12206d50e60cSDavid Rientjes khugepaged_enter_vma_merge(area, vm_flags); 12211da177e4SLinus Torvalds return area; 12221da177e4SLinus Torvalds } 12231da177e4SLinus Torvalds 12241da177e4SLinus Torvalds return NULL; 12251da177e4SLinus Torvalds } 12261da177e4SLinus Torvalds 12271da177e4SLinus Torvalds /* 1228d0e9fe17SLinus Torvalds * Rough compatbility check to quickly see if it's even worth looking 1229d0e9fe17SLinus Torvalds * at sharing an anon_vma. 1230d0e9fe17SLinus Torvalds * 1231d0e9fe17SLinus Torvalds * They need to have the same vm_file, and the flags can only differ 1232d0e9fe17SLinus Torvalds * in things that mprotect may change. 1233d0e9fe17SLinus Torvalds * 1234d0e9fe17SLinus Torvalds * NOTE! The fact that we share an anon_vma doesn't _have_ to mean that 1235d0e9fe17SLinus Torvalds * we can merge the two vma's. For example, we refuse to merge a vma if 1236d0e9fe17SLinus Torvalds * there is a vm_ops->close() function, because that indicates that the 1237d0e9fe17SLinus Torvalds * driver is doing some kind of reference counting. But that doesn't 1238d0e9fe17SLinus Torvalds * really matter for the anon_vma sharing case. 1239d0e9fe17SLinus Torvalds */ 1240d0e9fe17SLinus Torvalds static int anon_vma_compatible(struct vm_area_struct *a, struct vm_area_struct *b) 1241d0e9fe17SLinus Torvalds { 1242d0e9fe17SLinus Torvalds return a->vm_end == b->vm_start && 1243d0e9fe17SLinus Torvalds mpol_equal(vma_policy(a), vma_policy(b)) && 1244d0e9fe17SLinus Torvalds a->vm_file == b->vm_file && 124534228d47SCyrill Gorcunov !((a->vm_flags ^ b->vm_flags) & ~(VM_READ|VM_WRITE|VM_EXEC|VM_SOFTDIRTY)) && 1246d0e9fe17SLinus Torvalds b->vm_pgoff == a->vm_pgoff + ((b->vm_start - a->vm_start) >> PAGE_SHIFT); 1247d0e9fe17SLinus Torvalds } 1248d0e9fe17SLinus Torvalds 1249d0e9fe17SLinus Torvalds /* 1250d0e9fe17SLinus Torvalds * Do some basic sanity checking to see if we can re-use the anon_vma 1251d0e9fe17SLinus Torvalds * from 'old'. The 'a'/'b' vma's are in VM order - one of them will be 1252d0e9fe17SLinus Torvalds * the same as 'old', the other will be the new one that is trying 1253d0e9fe17SLinus Torvalds * to share the anon_vma. 1254d0e9fe17SLinus Torvalds * 1255d0e9fe17SLinus Torvalds * NOTE! This runs with mm_sem held for reading, so it is possible that 1256d0e9fe17SLinus Torvalds * the anon_vma of 'old' is concurrently in the process of being set up 1257d0e9fe17SLinus Torvalds * by another page fault trying to merge _that_. But that's ok: if it 1258d0e9fe17SLinus Torvalds * is being set up, that automatically means that it will be a singleton 1259d0e9fe17SLinus Torvalds * acceptable for merging, so we can do all of this optimistically. But 12604db0c3c2SJason Low * we do that READ_ONCE() to make sure that we never re-load the pointer. 1261d0e9fe17SLinus Torvalds * 1262d0e9fe17SLinus Torvalds * IOW: that the "list_is_singular()" test on the anon_vma_chain only 1263d0e9fe17SLinus Torvalds * matters for the 'stable anon_vma' case (ie the thing we want to avoid 1264d0e9fe17SLinus Torvalds * is to return an anon_vma that is "complex" due to having gone through 1265d0e9fe17SLinus Torvalds * a fork). 1266d0e9fe17SLinus Torvalds * 1267d0e9fe17SLinus Torvalds * We also make sure that the two vma's are compatible (adjacent, 1268d0e9fe17SLinus Torvalds * and with the same memory policies). That's all stable, even with just 1269d0e9fe17SLinus Torvalds * a read lock on the mm_sem. 1270d0e9fe17SLinus Torvalds */ 1271d0e9fe17SLinus Torvalds static struct anon_vma *reusable_anon_vma(struct vm_area_struct *old, struct vm_area_struct *a, struct vm_area_struct *b) 1272d0e9fe17SLinus Torvalds { 1273d0e9fe17SLinus Torvalds if (anon_vma_compatible(a, b)) { 12744db0c3c2SJason Low struct anon_vma *anon_vma = READ_ONCE(old->anon_vma); 1275d0e9fe17SLinus Torvalds 1276d0e9fe17SLinus Torvalds if (anon_vma && list_is_singular(&old->anon_vma_chain)) 1277d0e9fe17SLinus Torvalds return anon_vma; 1278d0e9fe17SLinus Torvalds } 1279d0e9fe17SLinus Torvalds return NULL; 1280d0e9fe17SLinus Torvalds } 1281d0e9fe17SLinus Torvalds 1282d0e9fe17SLinus Torvalds /* 12831da177e4SLinus Torvalds * find_mergeable_anon_vma is used by anon_vma_prepare, to check 12841da177e4SLinus Torvalds * neighbouring vmas for a suitable anon_vma, before it goes off 12851da177e4SLinus Torvalds * to allocate a new anon_vma. It checks because a repetitive 12861da177e4SLinus Torvalds * sequence of mprotects and faults may otherwise lead to distinct 12871da177e4SLinus Torvalds * anon_vmas being allocated, preventing vma merge in subsequent 12881da177e4SLinus Torvalds * mprotect. 12891da177e4SLinus Torvalds */ 12901da177e4SLinus Torvalds struct anon_vma *find_mergeable_anon_vma(struct vm_area_struct *vma) 12911da177e4SLinus Torvalds { 1292d0e9fe17SLinus Torvalds struct anon_vma *anon_vma; 12931da177e4SLinus Torvalds struct vm_area_struct *near; 12941da177e4SLinus Torvalds 12951da177e4SLinus Torvalds near = vma->vm_next; 12961da177e4SLinus Torvalds if (!near) 12971da177e4SLinus Torvalds goto try_prev; 12981da177e4SLinus Torvalds 1299d0e9fe17SLinus Torvalds anon_vma = reusable_anon_vma(near, vma, near); 1300d0e9fe17SLinus Torvalds if (anon_vma) 1301d0e9fe17SLinus Torvalds return anon_vma; 13021da177e4SLinus Torvalds try_prev: 13039be34c9dSLinus Torvalds near = vma->vm_prev; 13041da177e4SLinus Torvalds if (!near) 13051da177e4SLinus Torvalds goto none; 13061da177e4SLinus Torvalds 1307d0e9fe17SLinus Torvalds anon_vma = reusable_anon_vma(near, near, vma); 1308d0e9fe17SLinus Torvalds if (anon_vma) 1309d0e9fe17SLinus Torvalds return anon_vma; 13101da177e4SLinus Torvalds none: 13111da177e4SLinus Torvalds /* 13121da177e4SLinus Torvalds * There's no absolute need to look only at touching neighbours: 13131da177e4SLinus Torvalds * we could search further afield for "compatible" anon_vmas. 13141da177e4SLinus Torvalds * But it would probably just be a waste of time searching, 13151da177e4SLinus Torvalds * or lead to too many vmas hanging off the same anon_vma. 13161da177e4SLinus Torvalds * We're trying to allow mprotect remerging later on, 13171da177e4SLinus Torvalds * not trying to minimize memory used for anon_vmas. 13181da177e4SLinus Torvalds */ 13191da177e4SLinus Torvalds return NULL; 13201da177e4SLinus Torvalds } 13211da177e4SLinus Torvalds 13221da177e4SLinus Torvalds /* 132340401530SAl Viro * If a hint addr is less than mmap_min_addr change hint to be as 132440401530SAl Viro * low as possible but still greater than mmap_min_addr 132540401530SAl Viro */ 132640401530SAl Viro static inline unsigned long round_hint_to_min(unsigned long hint) 132740401530SAl Viro { 132840401530SAl Viro hint &= PAGE_MASK; 132940401530SAl Viro if (((void *)hint != NULL) && 133040401530SAl Viro (hint < mmap_min_addr)) 133140401530SAl Viro return PAGE_ALIGN(mmap_min_addr); 133240401530SAl Viro return hint; 133340401530SAl Viro } 133440401530SAl Viro 1335363ee17fSDavidlohr Bueso static inline int mlock_future_check(struct mm_struct *mm, 1336363ee17fSDavidlohr Bueso unsigned long flags, 1337363ee17fSDavidlohr Bueso unsigned long len) 1338363ee17fSDavidlohr Bueso { 1339363ee17fSDavidlohr Bueso unsigned long locked, lock_limit; 1340363ee17fSDavidlohr Bueso 1341363ee17fSDavidlohr Bueso /* mlock MCL_FUTURE? */ 1342363ee17fSDavidlohr Bueso if (flags & VM_LOCKED) { 1343363ee17fSDavidlohr Bueso locked = len >> PAGE_SHIFT; 1344363ee17fSDavidlohr Bueso locked += mm->locked_vm; 1345363ee17fSDavidlohr Bueso lock_limit = rlimit(RLIMIT_MEMLOCK); 1346363ee17fSDavidlohr Bueso lock_limit >>= PAGE_SHIFT; 1347363ee17fSDavidlohr Bueso if (locked > lock_limit && !capable(CAP_IPC_LOCK)) 1348363ee17fSDavidlohr Bueso return -EAGAIN; 1349363ee17fSDavidlohr Bueso } 1350363ee17fSDavidlohr Bueso return 0; 1351363ee17fSDavidlohr Bueso } 1352363ee17fSDavidlohr Bueso 1353be83bbf8SLinus Torvalds static inline u64 file_mmap_size_max(struct file *file, struct inode *inode) 1354be83bbf8SLinus Torvalds { 1355be83bbf8SLinus Torvalds if (S_ISREG(inode->i_mode)) 1356423913adSLinus Torvalds return MAX_LFS_FILESIZE; 1357be83bbf8SLinus Torvalds 1358be83bbf8SLinus Torvalds if (S_ISBLK(inode->i_mode)) 1359be83bbf8SLinus Torvalds return MAX_LFS_FILESIZE; 1360be83bbf8SLinus Torvalds 1361be83bbf8SLinus Torvalds /* Special "we do even unsigned file positions" case */ 1362be83bbf8SLinus Torvalds if (file->f_mode & FMODE_UNSIGNED_OFFSET) 1363be83bbf8SLinus Torvalds return 0; 1364be83bbf8SLinus Torvalds 1365be83bbf8SLinus Torvalds /* Yes, random drivers might want more. But I'm tired of buggy drivers */ 1366be83bbf8SLinus Torvalds return ULONG_MAX; 1367be83bbf8SLinus Torvalds } 1368be83bbf8SLinus Torvalds 1369be83bbf8SLinus Torvalds static inline bool file_mmap_ok(struct file *file, struct inode *inode, 1370be83bbf8SLinus Torvalds unsigned long pgoff, unsigned long len) 1371be83bbf8SLinus Torvalds { 1372be83bbf8SLinus Torvalds u64 maxsize = file_mmap_size_max(file, inode); 1373be83bbf8SLinus Torvalds 1374be83bbf8SLinus Torvalds if (maxsize && len > maxsize) 1375be83bbf8SLinus Torvalds return false; 1376be83bbf8SLinus Torvalds maxsize -= len; 1377be83bbf8SLinus Torvalds if (pgoff > maxsize >> PAGE_SHIFT) 1378be83bbf8SLinus Torvalds return false; 1379be83bbf8SLinus Torvalds return true; 1380be83bbf8SLinus Torvalds } 1381be83bbf8SLinus Torvalds 138240401530SAl Viro /* 138327f5de79SJianjun Kong * The caller must hold down_write(¤t->mm->mmap_sem). 13841da177e4SLinus Torvalds */ 13851fcfd8dbSOleg Nesterov unsigned long do_mmap(struct file *file, unsigned long addr, 13861da177e4SLinus Torvalds unsigned long len, unsigned long prot, 13871fcfd8dbSOleg Nesterov unsigned long flags, vm_flags_t vm_flags, 1388897ab3e0SMike Rapoport unsigned long pgoff, unsigned long *populate, 1389897ab3e0SMike Rapoport struct list_head *uf) 13901da177e4SLinus Torvalds { 13911da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 139262b5f7d0SDave Hansen int pkey = 0; 13931da177e4SLinus Torvalds 139441badc15SMichel Lespinasse *populate = 0; 1395bebeb3d6SMichel Lespinasse 1396e37609bbSPiotr Kwapulinski if (!len) 1397e37609bbSPiotr Kwapulinski return -EINVAL; 1398e37609bbSPiotr Kwapulinski 13991da177e4SLinus Torvalds /* 14001da177e4SLinus Torvalds * Does the application expect PROT_READ to imply PROT_EXEC? 14011da177e4SLinus Torvalds * 14021da177e4SLinus Torvalds * (the exception is when the underlying filesystem is noexec 14031da177e4SLinus Torvalds * mounted, in which case we dont add PROT_EXEC.) 14041da177e4SLinus Torvalds */ 14051da177e4SLinus Torvalds if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC)) 140690f8572bSEric W. Biederman if (!(file && path_noexec(&file->f_path))) 14071da177e4SLinus Torvalds prot |= PROT_EXEC; 14081da177e4SLinus Torvalds 1409a4ff8e86SMichal Hocko /* force arch specific MAP_FIXED handling in get_unmapped_area */ 1410a4ff8e86SMichal Hocko if (flags & MAP_FIXED_NOREPLACE) 1411a4ff8e86SMichal Hocko flags |= MAP_FIXED; 1412a4ff8e86SMichal Hocko 14137cd94146SEric Paris if (!(flags & MAP_FIXED)) 14147cd94146SEric Paris addr = round_hint_to_min(addr); 14157cd94146SEric Paris 14161da177e4SLinus Torvalds /* Careful about overflows.. */ 14171da177e4SLinus Torvalds len = PAGE_ALIGN(len); 14189206de95SAl Viro if (!len) 14191da177e4SLinus Torvalds return -ENOMEM; 14201da177e4SLinus Torvalds 14211da177e4SLinus Torvalds /* offset overflow? */ 14221da177e4SLinus Torvalds if ((pgoff + (len >> PAGE_SHIFT)) < pgoff) 14231da177e4SLinus Torvalds return -EOVERFLOW; 14241da177e4SLinus Torvalds 14251da177e4SLinus Torvalds /* Too many mappings? */ 14261da177e4SLinus Torvalds if (mm->map_count > sysctl_max_map_count) 14271da177e4SLinus Torvalds return -ENOMEM; 14281da177e4SLinus Torvalds 14291da177e4SLinus Torvalds /* Obtain the address to map to. we verify (or select) it and ensure 14301da177e4SLinus Torvalds * that it represents a valid section of the address space. 14311da177e4SLinus Torvalds */ 14321da177e4SLinus Torvalds addr = get_unmapped_area(file, addr, len, pgoff, flags); 1433de1741a1SAlexander Kuleshov if (offset_in_page(addr)) 14341da177e4SLinus Torvalds return addr; 14351da177e4SLinus Torvalds 1436a4ff8e86SMichal Hocko if (flags & MAP_FIXED_NOREPLACE) { 1437a4ff8e86SMichal Hocko struct vm_area_struct *vma = find_vma(mm, addr); 1438a4ff8e86SMichal Hocko 14397aa867ddSJann Horn if (vma && vma->vm_start < addr + len) 1440a4ff8e86SMichal Hocko return -EEXIST; 1441a4ff8e86SMichal Hocko } 1442a4ff8e86SMichal Hocko 144362b5f7d0SDave Hansen if (prot == PROT_EXEC) { 144462b5f7d0SDave Hansen pkey = execute_only_pkey(mm); 144562b5f7d0SDave Hansen if (pkey < 0) 144662b5f7d0SDave Hansen pkey = 0; 144762b5f7d0SDave Hansen } 144862b5f7d0SDave Hansen 14491da177e4SLinus Torvalds /* Do simple checking here so the lower-level routines won't have 14501da177e4SLinus Torvalds * to. we assume access permissions have been handled by the open 14511da177e4SLinus Torvalds * of the memory object, so we don't do any here. 14521da177e4SLinus Torvalds */ 145362b5f7d0SDave Hansen vm_flags |= calc_vm_prot_bits(prot, pkey) | calc_vm_flag_bits(flags) | 14541da177e4SLinus Torvalds mm->def_flags | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC; 14551da177e4SLinus Torvalds 1456cdf7b341SHuang Shijie if (flags & MAP_LOCKED) 14571da177e4SLinus Torvalds if (!can_do_mlock()) 14581da177e4SLinus Torvalds return -EPERM; 1459ba470de4SRik van Riel 1460363ee17fSDavidlohr Bueso if (mlock_future_check(mm, vm_flags, len)) 14611da177e4SLinus Torvalds return -EAGAIN; 14621da177e4SLinus Torvalds 14631da177e4SLinus Torvalds if (file) { 1464077bf22bSOleg Nesterov struct inode *inode = file_inode(file); 14651c972597SDan Williams unsigned long flags_mask; 14661c972597SDan Williams 1467be83bbf8SLinus Torvalds if (!file_mmap_ok(file, inode, pgoff, len)) 1468be83bbf8SLinus Torvalds return -EOVERFLOW; 1469be83bbf8SLinus Torvalds 14701c972597SDan Williams flags_mask = LEGACY_MAP_MASK | file->f_op->mmap_supported_flags; 1471077bf22bSOleg Nesterov 14721da177e4SLinus Torvalds switch (flags & MAP_TYPE) { 14731da177e4SLinus Torvalds case MAP_SHARED: 14741c972597SDan Williams /* 14751c972597SDan Williams * Force use of MAP_SHARED_VALIDATE with non-legacy 14761c972597SDan Williams * flags. E.g. MAP_SYNC is dangerous to use with 14771c972597SDan Williams * MAP_SHARED as you don't know which consistency model 14781c972597SDan Williams * you will get. We silently ignore unsupported flags 14791c972597SDan Williams * with MAP_SHARED to preserve backward compatibility. 14801c972597SDan Williams */ 14811c972597SDan Williams flags &= LEGACY_MAP_MASK; 14821c972597SDan Williams /* fall through */ 14831c972597SDan Williams case MAP_SHARED_VALIDATE: 14841c972597SDan Williams if (flags & ~flags_mask) 14851c972597SDan Williams return -EOPNOTSUPP; 14861da177e4SLinus Torvalds if ((prot&PROT_WRITE) && !(file->f_mode&FMODE_WRITE)) 14871da177e4SLinus Torvalds return -EACCES; 14881da177e4SLinus Torvalds 14891da177e4SLinus Torvalds /* 14901da177e4SLinus Torvalds * Make sure we don't allow writing to an append-only 14911da177e4SLinus Torvalds * file.. 14921da177e4SLinus Torvalds */ 14931da177e4SLinus Torvalds if (IS_APPEND(inode) && (file->f_mode & FMODE_WRITE)) 14941da177e4SLinus Torvalds return -EACCES; 14951da177e4SLinus Torvalds 14961da177e4SLinus Torvalds /* 14971da177e4SLinus Torvalds * Make sure there are no mandatory locks on the file. 14981da177e4SLinus Torvalds */ 1499d7a06983SJeff Layton if (locks_verify_locked(file)) 15001da177e4SLinus Torvalds return -EAGAIN; 15011da177e4SLinus Torvalds 15021da177e4SLinus Torvalds vm_flags |= VM_SHARED | VM_MAYSHARE; 15031da177e4SLinus Torvalds if (!(file->f_mode & FMODE_WRITE)) 15041da177e4SLinus Torvalds vm_flags &= ~(VM_MAYWRITE | VM_SHARED); 15051da177e4SLinus Torvalds 15061da177e4SLinus Torvalds /* fall through */ 15071da177e4SLinus Torvalds case MAP_PRIVATE: 15081da177e4SLinus Torvalds if (!(file->f_mode & FMODE_READ)) 15091da177e4SLinus Torvalds return -EACCES; 151090f8572bSEric W. Biederman if (path_noexec(&file->f_path)) { 151180c5606cSLinus Torvalds if (vm_flags & VM_EXEC) 151280c5606cSLinus Torvalds return -EPERM; 151380c5606cSLinus Torvalds vm_flags &= ~VM_MAYEXEC; 151480c5606cSLinus Torvalds } 151580c5606cSLinus Torvalds 151672c2d531SAl Viro if (!file->f_op->mmap) 151780c5606cSLinus Torvalds return -ENODEV; 1518b2c56e4fSOleg Nesterov if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP)) 1519b2c56e4fSOleg Nesterov return -EINVAL; 15201da177e4SLinus Torvalds break; 15211da177e4SLinus Torvalds 15221da177e4SLinus Torvalds default: 15231da177e4SLinus Torvalds return -EINVAL; 15241da177e4SLinus Torvalds } 15251da177e4SLinus Torvalds } else { 15261da177e4SLinus Torvalds switch (flags & MAP_TYPE) { 15271da177e4SLinus Torvalds case MAP_SHARED: 1528b2c56e4fSOleg Nesterov if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP)) 1529b2c56e4fSOleg Nesterov return -EINVAL; 1530ce363942STejun Heo /* 1531ce363942STejun Heo * Ignore pgoff. 1532ce363942STejun Heo */ 1533ce363942STejun Heo pgoff = 0; 15341da177e4SLinus Torvalds vm_flags |= VM_SHARED | VM_MAYSHARE; 15351da177e4SLinus Torvalds break; 15361da177e4SLinus Torvalds case MAP_PRIVATE: 15371da177e4SLinus Torvalds /* 15381da177e4SLinus Torvalds * Set pgoff according to addr for anon_vma. 15391da177e4SLinus Torvalds */ 15401da177e4SLinus Torvalds pgoff = addr >> PAGE_SHIFT; 15411da177e4SLinus Torvalds break; 15421da177e4SLinus Torvalds default: 15431da177e4SLinus Torvalds return -EINVAL; 15441da177e4SLinus Torvalds } 15451da177e4SLinus Torvalds } 15461da177e4SLinus Torvalds 1547c22c0d63SMichel Lespinasse /* 1548c22c0d63SMichel Lespinasse * Set 'VM_NORESERVE' if we should not account for the 1549c22c0d63SMichel Lespinasse * memory use of this mapping. 1550c22c0d63SMichel Lespinasse */ 1551c22c0d63SMichel Lespinasse if (flags & MAP_NORESERVE) { 1552c22c0d63SMichel Lespinasse /* We honor MAP_NORESERVE if allowed to overcommit */ 1553c22c0d63SMichel Lespinasse if (sysctl_overcommit_memory != OVERCOMMIT_NEVER) 1554c22c0d63SMichel Lespinasse vm_flags |= VM_NORESERVE; 1555c22c0d63SMichel Lespinasse 1556c22c0d63SMichel Lespinasse /* hugetlb applies strict overcommit unless MAP_NORESERVE */ 1557c22c0d63SMichel Lespinasse if (file && is_file_hugepages(file)) 1558c22c0d63SMichel Lespinasse vm_flags |= VM_NORESERVE; 1559c22c0d63SMichel Lespinasse } 1560c22c0d63SMichel Lespinasse 1561897ab3e0SMike Rapoport addr = mmap_region(file, addr, len, vm_flags, pgoff, uf); 156209a9f1d2SMichel Lespinasse if (!IS_ERR_VALUE(addr) && 156309a9f1d2SMichel Lespinasse ((vm_flags & VM_LOCKED) || 156409a9f1d2SMichel Lespinasse (flags & (MAP_POPULATE | MAP_NONBLOCK)) == MAP_POPULATE)) 156541badc15SMichel Lespinasse *populate = len; 1566bebeb3d6SMichel Lespinasse return addr; 15670165ab44SMiklos Szeredi } 15686be5ceb0SLinus Torvalds 1569a90f590aSDominik Brodowski unsigned long ksys_mmap_pgoff(unsigned long addr, unsigned long len, 1570a90f590aSDominik Brodowski unsigned long prot, unsigned long flags, 1571a90f590aSDominik Brodowski unsigned long fd, unsigned long pgoff) 157266f0dc48SHugh Dickins { 157366f0dc48SHugh Dickins struct file *file = NULL; 15741e3ee14bSChen Gang unsigned long retval; 157566f0dc48SHugh Dickins 157666f0dc48SHugh Dickins if (!(flags & MAP_ANONYMOUS)) { 1577120a795dSAl Viro audit_mmap_fd(fd, flags); 157866f0dc48SHugh Dickins file = fget(fd); 157966f0dc48SHugh Dickins if (!file) 15801e3ee14bSChen Gang return -EBADF; 1581af73e4d9SNaoya Horiguchi if (is_file_hugepages(file)) 1582af73e4d9SNaoya Horiguchi len = ALIGN(len, huge_page_size(hstate_file(file))); 1583493af578SJörn Engel retval = -EINVAL; 1584493af578SJörn Engel if (unlikely(flags & MAP_HUGETLB && !is_file_hugepages(file))) 1585493af578SJörn Engel goto out_fput; 158666f0dc48SHugh Dickins } else if (flags & MAP_HUGETLB) { 158766f0dc48SHugh Dickins struct user_struct *user = NULL; 1588c103a4dcSAndrew Morton struct hstate *hs; 1589af73e4d9SNaoya Horiguchi 159020ac2893SAnshuman Khandual hs = hstate_sizelog((flags >> MAP_HUGE_SHIFT) & MAP_HUGE_MASK); 1591091d0d55SLi Zefan if (!hs) 1592091d0d55SLi Zefan return -EINVAL; 1593091d0d55SLi Zefan 1594091d0d55SLi Zefan len = ALIGN(len, huge_page_size(hs)); 159566f0dc48SHugh Dickins /* 159666f0dc48SHugh Dickins * VM_NORESERVE is used because the reservations will be 159766f0dc48SHugh Dickins * taken when vm_ops->mmap() is called 159866f0dc48SHugh Dickins * A dummy user value is used because we are not locking 159966f0dc48SHugh Dickins * memory so no accounting is necessary 160066f0dc48SHugh Dickins */ 1601af73e4d9SNaoya Horiguchi file = hugetlb_file_setup(HUGETLB_ANON_FILE, len, 160242d7395fSAndi Kleen VM_NORESERVE, 160342d7395fSAndi Kleen &user, HUGETLB_ANONHUGE_INODE, 160442d7395fSAndi Kleen (flags >> MAP_HUGE_SHIFT) & MAP_HUGE_MASK); 160566f0dc48SHugh Dickins if (IS_ERR(file)) 160666f0dc48SHugh Dickins return PTR_ERR(file); 160766f0dc48SHugh Dickins } 160866f0dc48SHugh Dickins 160966f0dc48SHugh Dickins flags &= ~(MAP_EXECUTABLE | MAP_DENYWRITE); 161066f0dc48SHugh Dickins 16119fbeb5abSMichal Hocko retval = vm_mmap_pgoff(file, addr, len, prot, flags, pgoff); 1612493af578SJörn Engel out_fput: 161366f0dc48SHugh Dickins if (file) 161466f0dc48SHugh Dickins fput(file); 161566f0dc48SHugh Dickins return retval; 161666f0dc48SHugh Dickins } 161766f0dc48SHugh Dickins 1618a90f590aSDominik Brodowski SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len, 1619a90f590aSDominik Brodowski unsigned long, prot, unsigned long, flags, 1620a90f590aSDominik Brodowski unsigned long, fd, unsigned long, pgoff) 1621a90f590aSDominik Brodowski { 1622a90f590aSDominik Brodowski return ksys_mmap_pgoff(addr, len, prot, flags, fd, pgoff); 1623a90f590aSDominik Brodowski } 1624a90f590aSDominik Brodowski 1625a4679373SChristoph Hellwig #ifdef __ARCH_WANT_SYS_OLD_MMAP 1626a4679373SChristoph Hellwig struct mmap_arg_struct { 1627a4679373SChristoph Hellwig unsigned long addr; 1628a4679373SChristoph Hellwig unsigned long len; 1629a4679373SChristoph Hellwig unsigned long prot; 1630a4679373SChristoph Hellwig unsigned long flags; 1631a4679373SChristoph Hellwig unsigned long fd; 1632a4679373SChristoph Hellwig unsigned long offset; 1633a4679373SChristoph Hellwig }; 1634a4679373SChristoph Hellwig 1635a4679373SChristoph Hellwig SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg) 1636a4679373SChristoph Hellwig { 1637a4679373SChristoph Hellwig struct mmap_arg_struct a; 1638a4679373SChristoph Hellwig 1639a4679373SChristoph Hellwig if (copy_from_user(&a, arg, sizeof(a))) 1640a4679373SChristoph Hellwig return -EFAULT; 1641de1741a1SAlexander Kuleshov if (offset_in_page(a.offset)) 1642a4679373SChristoph Hellwig return -EINVAL; 1643a4679373SChristoph Hellwig 1644a90f590aSDominik Brodowski return ksys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd, 1645a4679373SChristoph Hellwig a.offset >> PAGE_SHIFT); 1646a4679373SChristoph Hellwig } 1647a4679373SChristoph Hellwig #endif /* __ARCH_WANT_SYS_OLD_MMAP */ 1648a4679373SChristoph Hellwig 16494e950f6fSAlexey Dobriyan /* 16508bb4e7a2SWei Yang * Some shared mappings will want the pages marked read-only 16514e950f6fSAlexey Dobriyan * to track write events. If so, we'll downgrade vm_page_prot 16524e950f6fSAlexey Dobriyan * to the private version (using protection_map[] without the 16534e950f6fSAlexey Dobriyan * VM_SHARED bit). 16544e950f6fSAlexey Dobriyan */ 16556d2329f8SAndrea Arcangeli int vma_wants_writenotify(struct vm_area_struct *vma, pgprot_t vm_page_prot) 16564e950f6fSAlexey Dobriyan { 1657ca16d140SKOSAKI Motohiro vm_flags_t vm_flags = vma->vm_flags; 16588a04446aSKirill A. Shutemov const struct vm_operations_struct *vm_ops = vma->vm_ops; 16594e950f6fSAlexey Dobriyan 16604e950f6fSAlexey Dobriyan /* If it was private or non-writable, the write bit is already clear */ 16614e950f6fSAlexey Dobriyan if ((vm_flags & (VM_WRITE|VM_SHARED)) != ((VM_WRITE|VM_SHARED))) 16624e950f6fSAlexey Dobriyan return 0; 16634e950f6fSAlexey Dobriyan 16644e950f6fSAlexey Dobriyan /* The backer wishes to know when pages are first written to? */ 16658a04446aSKirill A. Shutemov if (vm_ops && (vm_ops->page_mkwrite || vm_ops->pfn_mkwrite)) 16664e950f6fSAlexey Dobriyan return 1; 16674e950f6fSAlexey Dobriyan 166864e45507SPeter Feiner /* The open routine did something to the protections that pgprot_modify 166964e45507SPeter Feiner * won't preserve? */ 16706d2329f8SAndrea Arcangeli if (pgprot_val(vm_page_prot) != 16716d2329f8SAndrea Arcangeli pgprot_val(vm_pgprot_modify(vm_page_prot, vm_flags))) 16724e950f6fSAlexey Dobriyan return 0; 16734e950f6fSAlexey Dobriyan 167464e45507SPeter Feiner /* Do we need to track softdirty? */ 167564e45507SPeter Feiner if (IS_ENABLED(CONFIG_MEM_SOFT_DIRTY) && !(vm_flags & VM_SOFTDIRTY)) 167664e45507SPeter Feiner return 1; 167764e45507SPeter Feiner 16784e950f6fSAlexey Dobriyan /* Specialty mapping? */ 16794b6e1e37SKonstantin Khlebnikov if (vm_flags & VM_PFNMAP) 16804e950f6fSAlexey Dobriyan return 0; 16814e950f6fSAlexey Dobriyan 16824e950f6fSAlexey Dobriyan /* Can the mapping track the dirty pages? */ 16834e950f6fSAlexey Dobriyan return vma->vm_file && vma->vm_file->f_mapping && 16844e950f6fSAlexey Dobriyan mapping_cap_account_dirty(vma->vm_file->f_mapping); 16854e950f6fSAlexey Dobriyan } 16864e950f6fSAlexey Dobriyan 1687fc8744adSLinus Torvalds /* 1688fc8744adSLinus Torvalds * We account for memory if it's a private writeable mapping, 16895a6fe125SMel Gorman * not hugepages and VM_NORESERVE wasn't set. 1690fc8744adSLinus Torvalds */ 1691ca16d140SKOSAKI Motohiro static inline int accountable_mapping(struct file *file, vm_flags_t vm_flags) 1692fc8744adSLinus Torvalds { 16935a6fe125SMel Gorman /* 16945a6fe125SMel Gorman * hugetlb has its own accounting separate from the core VM 16955a6fe125SMel Gorman * VM_HUGETLB may not be set yet so we cannot check for that flag. 16965a6fe125SMel Gorman */ 16975a6fe125SMel Gorman if (file && is_file_hugepages(file)) 16985a6fe125SMel Gorman return 0; 16995a6fe125SMel Gorman 1700fc8744adSLinus Torvalds return (vm_flags & (VM_NORESERVE | VM_SHARED | VM_WRITE)) == VM_WRITE; 1701fc8744adSLinus Torvalds } 1702fc8744adSLinus Torvalds 17030165ab44SMiklos Szeredi unsigned long mmap_region(struct file *file, unsigned long addr, 1704897ab3e0SMike Rapoport unsigned long len, vm_flags_t vm_flags, unsigned long pgoff, 1705897ab3e0SMike Rapoport struct list_head *uf) 17060165ab44SMiklos Szeredi { 17070165ab44SMiklos Szeredi struct mm_struct *mm = current->mm; 17080165ab44SMiklos Szeredi struct vm_area_struct *vma, *prev; 17090165ab44SMiklos Szeredi int error; 17100165ab44SMiklos Szeredi struct rb_node **rb_link, *rb_parent; 17110165ab44SMiklos Szeredi unsigned long charged = 0; 17120165ab44SMiklos Szeredi 1713e8420a8eSCyril Hrubis /* Check against address space limit. */ 171484638335SKonstantin Khlebnikov if (!may_expand_vm(mm, vm_flags, len >> PAGE_SHIFT)) { 1715e8420a8eSCyril Hrubis unsigned long nr_pages; 1716e8420a8eSCyril Hrubis 1717e8420a8eSCyril Hrubis /* 1718e8420a8eSCyril Hrubis * MAP_FIXED may remove pages of mappings that intersects with 1719e8420a8eSCyril Hrubis * requested mapping. Account for the pages it would unmap. 1720e8420a8eSCyril Hrubis */ 1721e8420a8eSCyril Hrubis nr_pages = count_vma_pages_range(mm, addr, addr + len); 1722e8420a8eSCyril Hrubis 172384638335SKonstantin Khlebnikov if (!may_expand_vm(mm, vm_flags, 172484638335SKonstantin Khlebnikov (len >> PAGE_SHIFT) - nr_pages)) 1725e8420a8eSCyril Hrubis return -ENOMEM; 1726e8420a8eSCyril Hrubis } 1727e8420a8eSCyril Hrubis 17281da177e4SLinus Torvalds /* Clear old maps */ 17299fcd1457SRasmus Villemoes while (find_vma_links(mm, addr, addr + len, &prev, &rb_link, 17309fcd1457SRasmus Villemoes &rb_parent)) { 1731897ab3e0SMike Rapoport if (do_munmap(mm, addr, len, uf)) 17321da177e4SLinus Torvalds return -ENOMEM; 17331da177e4SLinus Torvalds } 17341da177e4SLinus Torvalds 1735fc8744adSLinus Torvalds /* 17361da177e4SLinus Torvalds * Private writable mapping: check memory availability 17371da177e4SLinus Torvalds */ 17385a6fe125SMel Gorman if (accountable_mapping(file, vm_flags)) { 17391da177e4SLinus Torvalds charged = len >> PAGE_SHIFT; 1740191c5424SAl Viro if (security_vm_enough_memory_mm(mm, charged)) 17411da177e4SLinus Torvalds return -ENOMEM; 17421da177e4SLinus Torvalds vm_flags |= VM_ACCOUNT; 17431da177e4SLinus Torvalds } 17441da177e4SLinus Torvalds 17451da177e4SLinus Torvalds /* 1746de33c8dbSLinus Torvalds * Can we just expand an old mapping? 17471da177e4SLinus Torvalds */ 174819a809afSAndrea Arcangeli vma = vma_merge(mm, prev, addr, addr + len, vm_flags, 174919a809afSAndrea Arcangeli NULL, file, pgoff, NULL, NULL_VM_UFFD_CTX); 1750ba470de4SRik van Riel if (vma) 17511da177e4SLinus Torvalds goto out; 17521da177e4SLinus Torvalds 17531da177e4SLinus Torvalds /* 17541da177e4SLinus Torvalds * Determine the object being mapped and call the appropriate 17551da177e4SLinus Torvalds * specific mapper. the address has already been validated, but 17561da177e4SLinus Torvalds * not unmapped, but the maps are removed from the list. 17571da177e4SLinus Torvalds */ 1758490fc053SLinus Torvalds vma = vm_area_alloc(mm); 17591da177e4SLinus Torvalds if (!vma) { 17601da177e4SLinus Torvalds error = -ENOMEM; 17611da177e4SLinus Torvalds goto unacct_error; 17621da177e4SLinus Torvalds } 17631da177e4SLinus Torvalds 17641da177e4SLinus Torvalds vma->vm_start = addr; 17651da177e4SLinus Torvalds vma->vm_end = addr + len; 17661da177e4SLinus Torvalds vma->vm_flags = vm_flags; 17673ed75eb8SColy Li vma->vm_page_prot = vm_get_page_prot(vm_flags); 17681da177e4SLinus Torvalds vma->vm_pgoff = pgoff; 17691da177e4SLinus Torvalds 17701da177e4SLinus Torvalds if (file) { 17711da177e4SLinus Torvalds if (vm_flags & VM_DENYWRITE) { 17721da177e4SLinus Torvalds error = deny_write_access(file); 17731da177e4SLinus Torvalds if (error) 17741da177e4SLinus Torvalds goto free_vma; 17751da177e4SLinus Torvalds } 17764bb5f5d9SDavid Herrmann if (vm_flags & VM_SHARED) { 17774bb5f5d9SDavid Herrmann error = mapping_map_writable(file->f_mapping); 17784bb5f5d9SDavid Herrmann if (error) 17794bb5f5d9SDavid Herrmann goto allow_write_and_free_vma; 17804bb5f5d9SDavid Herrmann } 17814bb5f5d9SDavid Herrmann 17824bb5f5d9SDavid Herrmann /* ->mmap() can change vma->vm_file, but must guarantee that 17834bb5f5d9SDavid Herrmann * vma_link() below can deny write-access if VM_DENYWRITE is set 17844bb5f5d9SDavid Herrmann * and map writably if VM_SHARED is set. This usually means the 17854bb5f5d9SDavid Herrmann * new file must not have been exposed to user-space, yet. 17864bb5f5d9SDavid Herrmann */ 1787cb0942b8SAl Viro vma->vm_file = get_file(file); 1788f74ac015SMiklos Szeredi error = call_mmap(file, vma); 17891da177e4SLinus Torvalds if (error) 17901da177e4SLinus Torvalds goto unmap_and_free_vma; 17911da177e4SLinus Torvalds 17921da177e4SLinus Torvalds /* Can addr have changed?? 17931da177e4SLinus Torvalds * 17941da177e4SLinus Torvalds * Answer: Yes, several device drivers can do it in their 17951da177e4SLinus Torvalds * f_op->mmap method. -DaveM 17962897b4d2SJoonsoo Kim * Bug: If addr is changed, prev, rb_link, rb_parent should 17972897b4d2SJoonsoo Kim * be updated for vma_link() 17981da177e4SLinus Torvalds */ 17992897b4d2SJoonsoo Kim WARN_ON_ONCE(addr != vma->vm_start); 18002897b4d2SJoonsoo Kim 18011da177e4SLinus Torvalds addr = vma->vm_start; 18021da177e4SLinus Torvalds vm_flags = vma->vm_flags; 1803f8dbf0a7SHuang Shijie } else if (vm_flags & VM_SHARED) { 1804f8dbf0a7SHuang Shijie error = shmem_zero_setup(vma); 1805f8dbf0a7SHuang Shijie if (error) 1806f8dbf0a7SHuang Shijie goto free_vma; 1807bfd40eafSKirill A. Shutemov } else { 1808bfd40eafSKirill A. Shutemov vma_set_anonymous(vma); 1809f8dbf0a7SHuang Shijie } 18101da177e4SLinus Torvalds 18114d3d5b41SOleg Nesterov vma_link(mm, vma, prev, rb_link, rb_parent); 18124d3d5b41SOleg Nesterov /* Once vma denies write, undo our temporary denial count */ 18134bb5f5d9SDavid Herrmann if (file) { 18144bb5f5d9SDavid Herrmann if (vm_flags & VM_SHARED) 18154bb5f5d9SDavid Herrmann mapping_unmap_writable(file->f_mapping); 1816e8686772SOleg Nesterov if (vm_flags & VM_DENYWRITE) 1817e8686772SOleg Nesterov allow_write_access(file); 18184bb5f5d9SDavid Herrmann } 1819e8686772SOleg Nesterov file = vma->vm_file; 18201da177e4SLinus Torvalds out: 1821cdd6c482SIngo Molnar perf_event_mmap(vma); 18220a4a9391SPeter Zijlstra 182384638335SKonstantin Khlebnikov vm_stat_account(mm, vm_flags, len >> PAGE_SHIFT); 18241da177e4SLinus Torvalds if (vm_flags & VM_LOCKED) { 1825e1fb4a08SDave Jiang if ((vm_flags & VM_SPECIAL) || vma_is_dax(vma) || 1826e1fb4a08SDave Jiang is_vm_hugetlb_page(vma) || 1827e1fb4a08SDave Jiang vma == get_gate_vma(current->mm)) 1828de60f5f1SEric B Munson vma->vm_flags &= VM_LOCKED_CLEAR_MASK; 1829e1fb4a08SDave Jiang else 1830e1fb4a08SDave Jiang mm->locked_vm += (len >> PAGE_SHIFT); 1831bebeb3d6SMichel Lespinasse } 18322b144498SSrikar Dronamraju 1833c7a3a88cSOleg Nesterov if (file) 1834c7a3a88cSOleg Nesterov uprobe_mmap(vma); 18352b144498SSrikar Dronamraju 1836d9104d1cSCyrill Gorcunov /* 1837d9104d1cSCyrill Gorcunov * New (or expanded) vma always get soft dirty status. 1838d9104d1cSCyrill Gorcunov * Otherwise user-space soft-dirty page tracker won't 1839d9104d1cSCyrill Gorcunov * be able to distinguish situation when vma area unmapped, 1840d9104d1cSCyrill Gorcunov * then new mapped in-place (which must be aimed as 1841d9104d1cSCyrill Gorcunov * a completely new data area). 1842d9104d1cSCyrill Gorcunov */ 1843d9104d1cSCyrill Gorcunov vma->vm_flags |= VM_SOFTDIRTY; 1844d9104d1cSCyrill Gorcunov 184564e45507SPeter Feiner vma_set_page_prot(vma); 184664e45507SPeter Feiner 18471da177e4SLinus Torvalds return addr; 18481da177e4SLinus Torvalds 18491da177e4SLinus Torvalds unmap_and_free_vma: 18501da177e4SLinus Torvalds vma->vm_file = NULL; 18511da177e4SLinus Torvalds fput(file); 18521da177e4SLinus Torvalds 18531da177e4SLinus Torvalds /* Undo any partial mapping done by a device driver. */ 1854e0da382cSHugh Dickins unmap_region(mm, vma, prev, vma->vm_start, vma->vm_end); 1855e0da382cSHugh Dickins charged = 0; 18564bb5f5d9SDavid Herrmann if (vm_flags & VM_SHARED) 18574bb5f5d9SDavid Herrmann mapping_unmap_writable(file->f_mapping); 18584bb5f5d9SDavid Herrmann allow_write_and_free_vma: 18594bb5f5d9SDavid Herrmann if (vm_flags & VM_DENYWRITE) 18604bb5f5d9SDavid Herrmann allow_write_access(file); 18611da177e4SLinus Torvalds free_vma: 18623928d4f5SLinus Torvalds vm_area_free(vma); 18631da177e4SLinus Torvalds unacct_error: 18641da177e4SLinus Torvalds if (charged) 18651da177e4SLinus Torvalds vm_unacct_memory(charged); 18661da177e4SLinus Torvalds return error; 18671da177e4SLinus Torvalds } 18681da177e4SLinus Torvalds 1869db4fbfb9SMichel Lespinasse unsigned long unmapped_area(struct vm_unmapped_area_info *info) 1870db4fbfb9SMichel Lespinasse { 1871db4fbfb9SMichel Lespinasse /* 1872db4fbfb9SMichel Lespinasse * We implement the search by looking for an rbtree node that 1873db4fbfb9SMichel Lespinasse * immediately follows a suitable gap. That is, 1874db4fbfb9SMichel Lespinasse * - gap_start = vma->vm_prev->vm_end <= info->high_limit - length; 1875db4fbfb9SMichel Lespinasse * - gap_end = vma->vm_start >= info->low_limit + length; 1876db4fbfb9SMichel Lespinasse * - gap_end - gap_start >= length 1877db4fbfb9SMichel Lespinasse */ 1878db4fbfb9SMichel Lespinasse 1879db4fbfb9SMichel Lespinasse struct mm_struct *mm = current->mm; 1880db4fbfb9SMichel Lespinasse struct vm_area_struct *vma; 1881db4fbfb9SMichel Lespinasse unsigned long length, low_limit, high_limit, gap_start, gap_end; 1882db4fbfb9SMichel Lespinasse 1883db4fbfb9SMichel Lespinasse /* Adjust search length to account for worst case alignment overhead */ 1884db4fbfb9SMichel Lespinasse length = info->length + info->align_mask; 1885db4fbfb9SMichel Lespinasse if (length < info->length) 1886db4fbfb9SMichel Lespinasse return -ENOMEM; 1887db4fbfb9SMichel Lespinasse 1888db4fbfb9SMichel Lespinasse /* Adjust search limits by the desired length */ 1889db4fbfb9SMichel Lespinasse if (info->high_limit < length) 1890db4fbfb9SMichel Lespinasse return -ENOMEM; 1891db4fbfb9SMichel Lespinasse high_limit = info->high_limit - length; 1892db4fbfb9SMichel Lespinasse 1893db4fbfb9SMichel Lespinasse if (info->low_limit > high_limit) 1894db4fbfb9SMichel Lespinasse return -ENOMEM; 1895db4fbfb9SMichel Lespinasse low_limit = info->low_limit + length; 1896db4fbfb9SMichel Lespinasse 1897db4fbfb9SMichel Lespinasse /* Check if rbtree root looks promising */ 1898db4fbfb9SMichel Lespinasse if (RB_EMPTY_ROOT(&mm->mm_rb)) 1899db4fbfb9SMichel Lespinasse goto check_highest; 1900db4fbfb9SMichel Lespinasse vma = rb_entry(mm->mm_rb.rb_node, struct vm_area_struct, vm_rb); 1901db4fbfb9SMichel Lespinasse if (vma->rb_subtree_gap < length) 1902db4fbfb9SMichel Lespinasse goto check_highest; 1903db4fbfb9SMichel Lespinasse 1904db4fbfb9SMichel Lespinasse while (true) { 1905db4fbfb9SMichel Lespinasse /* Visit left subtree if it looks promising */ 19061be7107fSHugh Dickins gap_end = vm_start_gap(vma); 1907db4fbfb9SMichel Lespinasse if (gap_end >= low_limit && vma->vm_rb.rb_left) { 1908db4fbfb9SMichel Lespinasse struct vm_area_struct *left = 1909db4fbfb9SMichel Lespinasse rb_entry(vma->vm_rb.rb_left, 1910db4fbfb9SMichel Lespinasse struct vm_area_struct, vm_rb); 1911db4fbfb9SMichel Lespinasse if (left->rb_subtree_gap >= length) { 1912db4fbfb9SMichel Lespinasse vma = left; 1913db4fbfb9SMichel Lespinasse continue; 1914db4fbfb9SMichel Lespinasse } 1915db4fbfb9SMichel Lespinasse } 1916db4fbfb9SMichel Lespinasse 19171be7107fSHugh Dickins gap_start = vma->vm_prev ? vm_end_gap(vma->vm_prev) : 0; 1918db4fbfb9SMichel Lespinasse check_current: 1919db4fbfb9SMichel Lespinasse /* Check if current node has a suitable gap */ 1920db4fbfb9SMichel Lespinasse if (gap_start > high_limit) 1921db4fbfb9SMichel Lespinasse return -ENOMEM; 1922f4cb767dSHugh Dickins if (gap_end >= low_limit && 1923f4cb767dSHugh Dickins gap_end > gap_start && gap_end - gap_start >= length) 1924db4fbfb9SMichel Lespinasse goto found; 1925db4fbfb9SMichel Lespinasse 1926db4fbfb9SMichel Lespinasse /* Visit right subtree if it looks promising */ 1927db4fbfb9SMichel Lespinasse if (vma->vm_rb.rb_right) { 1928db4fbfb9SMichel Lespinasse struct vm_area_struct *right = 1929db4fbfb9SMichel Lespinasse rb_entry(vma->vm_rb.rb_right, 1930db4fbfb9SMichel Lespinasse struct vm_area_struct, vm_rb); 1931db4fbfb9SMichel Lespinasse if (right->rb_subtree_gap >= length) { 1932db4fbfb9SMichel Lespinasse vma = right; 1933db4fbfb9SMichel Lespinasse continue; 1934db4fbfb9SMichel Lespinasse } 1935db4fbfb9SMichel Lespinasse } 1936db4fbfb9SMichel Lespinasse 1937db4fbfb9SMichel Lespinasse /* Go back up the rbtree to find next candidate node */ 1938db4fbfb9SMichel Lespinasse while (true) { 1939db4fbfb9SMichel Lespinasse struct rb_node *prev = &vma->vm_rb; 1940db4fbfb9SMichel Lespinasse if (!rb_parent(prev)) 1941db4fbfb9SMichel Lespinasse goto check_highest; 1942db4fbfb9SMichel Lespinasse vma = rb_entry(rb_parent(prev), 1943db4fbfb9SMichel Lespinasse struct vm_area_struct, vm_rb); 1944db4fbfb9SMichel Lespinasse if (prev == vma->vm_rb.rb_left) { 19451be7107fSHugh Dickins gap_start = vm_end_gap(vma->vm_prev); 19461be7107fSHugh Dickins gap_end = vm_start_gap(vma); 1947db4fbfb9SMichel Lespinasse goto check_current; 1948db4fbfb9SMichel Lespinasse } 1949db4fbfb9SMichel Lespinasse } 1950db4fbfb9SMichel Lespinasse } 1951db4fbfb9SMichel Lespinasse 1952db4fbfb9SMichel Lespinasse check_highest: 1953db4fbfb9SMichel Lespinasse /* Check highest gap, which does not precede any rbtree node */ 1954db4fbfb9SMichel Lespinasse gap_start = mm->highest_vm_end; 1955db4fbfb9SMichel Lespinasse gap_end = ULONG_MAX; /* Only for VM_BUG_ON below */ 1956db4fbfb9SMichel Lespinasse if (gap_start > high_limit) 1957db4fbfb9SMichel Lespinasse return -ENOMEM; 1958db4fbfb9SMichel Lespinasse 1959db4fbfb9SMichel Lespinasse found: 1960db4fbfb9SMichel Lespinasse /* We found a suitable gap. Clip it with the original low_limit. */ 1961db4fbfb9SMichel Lespinasse if (gap_start < info->low_limit) 1962db4fbfb9SMichel Lespinasse gap_start = info->low_limit; 1963db4fbfb9SMichel Lespinasse 1964db4fbfb9SMichel Lespinasse /* Adjust gap address to the desired alignment */ 1965db4fbfb9SMichel Lespinasse gap_start += (info->align_offset - gap_start) & info->align_mask; 1966db4fbfb9SMichel Lespinasse 1967db4fbfb9SMichel Lespinasse VM_BUG_ON(gap_start + info->length > info->high_limit); 1968db4fbfb9SMichel Lespinasse VM_BUG_ON(gap_start + info->length > gap_end); 1969db4fbfb9SMichel Lespinasse return gap_start; 1970db4fbfb9SMichel Lespinasse } 1971db4fbfb9SMichel Lespinasse 1972db4fbfb9SMichel Lespinasse unsigned long unmapped_area_topdown(struct vm_unmapped_area_info *info) 1973db4fbfb9SMichel Lespinasse { 1974db4fbfb9SMichel Lespinasse struct mm_struct *mm = current->mm; 1975db4fbfb9SMichel Lespinasse struct vm_area_struct *vma; 1976db4fbfb9SMichel Lespinasse unsigned long length, low_limit, high_limit, gap_start, gap_end; 1977db4fbfb9SMichel Lespinasse 1978db4fbfb9SMichel Lespinasse /* Adjust search length to account for worst case alignment overhead */ 1979db4fbfb9SMichel Lespinasse length = info->length + info->align_mask; 1980db4fbfb9SMichel Lespinasse if (length < info->length) 1981db4fbfb9SMichel Lespinasse return -ENOMEM; 1982db4fbfb9SMichel Lespinasse 1983db4fbfb9SMichel Lespinasse /* 1984db4fbfb9SMichel Lespinasse * Adjust search limits by the desired length. 1985db4fbfb9SMichel Lespinasse * See implementation comment at top of unmapped_area(). 1986db4fbfb9SMichel Lespinasse */ 1987db4fbfb9SMichel Lespinasse gap_end = info->high_limit; 1988db4fbfb9SMichel Lespinasse if (gap_end < length) 1989db4fbfb9SMichel Lespinasse return -ENOMEM; 1990db4fbfb9SMichel Lespinasse high_limit = gap_end - length; 1991db4fbfb9SMichel Lespinasse 1992db4fbfb9SMichel Lespinasse if (info->low_limit > high_limit) 1993db4fbfb9SMichel Lespinasse return -ENOMEM; 1994db4fbfb9SMichel Lespinasse low_limit = info->low_limit + length; 1995db4fbfb9SMichel Lespinasse 1996db4fbfb9SMichel Lespinasse /* Check highest gap, which does not precede any rbtree node */ 1997db4fbfb9SMichel Lespinasse gap_start = mm->highest_vm_end; 1998db4fbfb9SMichel Lespinasse if (gap_start <= high_limit) 1999db4fbfb9SMichel Lespinasse goto found_highest; 2000db4fbfb9SMichel Lespinasse 2001db4fbfb9SMichel Lespinasse /* Check if rbtree root looks promising */ 2002db4fbfb9SMichel Lespinasse if (RB_EMPTY_ROOT(&mm->mm_rb)) 2003db4fbfb9SMichel Lespinasse return -ENOMEM; 2004db4fbfb9SMichel Lespinasse vma = rb_entry(mm->mm_rb.rb_node, struct vm_area_struct, vm_rb); 2005db4fbfb9SMichel Lespinasse if (vma->rb_subtree_gap < length) 2006db4fbfb9SMichel Lespinasse return -ENOMEM; 2007db4fbfb9SMichel Lespinasse 2008db4fbfb9SMichel Lespinasse while (true) { 2009db4fbfb9SMichel Lespinasse /* Visit right subtree if it looks promising */ 20101be7107fSHugh Dickins gap_start = vma->vm_prev ? vm_end_gap(vma->vm_prev) : 0; 2011db4fbfb9SMichel Lespinasse if (gap_start <= high_limit && vma->vm_rb.rb_right) { 2012db4fbfb9SMichel Lespinasse struct vm_area_struct *right = 2013db4fbfb9SMichel Lespinasse rb_entry(vma->vm_rb.rb_right, 2014db4fbfb9SMichel Lespinasse struct vm_area_struct, vm_rb); 2015db4fbfb9SMichel Lespinasse if (right->rb_subtree_gap >= length) { 2016db4fbfb9SMichel Lespinasse vma = right; 2017db4fbfb9SMichel Lespinasse continue; 2018db4fbfb9SMichel Lespinasse } 2019db4fbfb9SMichel Lespinasse } 2020db4fbfb9SMichel Lespinasse 2021db4fbfb9SMichel Lespinasse check_current: 2022db4fbfb9SMichel Lespinasse /* Check if current node has a suitable gap */ 20231be7107fSHugh Dickins gap_end = vm_start_gap(vma); 2024db4fbfb9SMichel Lespinasse if (gap_end < low_limit) 2025db4fbfb9SMichel Lespinasse return -ENOMEM; 2026f4cb767dSHugh Dickins if (gap_start <= high_limit && 2027f4cb767dSHugh Dickins gap_end > gap_start && gap_end - gap_start >= length) 2028db4fbfb9SMichel Lespinasse goto found; 2029db4fbfb9SMichel Lespinasse 2030db4fbfb9SMichel Lespinasse /* Visit left subtree if it looks promising */ 2031db4fbfb9SMichel Lespinasse if (vma->vm_rb.rb_left) { 2032db4fbfb9SMichel Lespinasse struct vm_area_struct *left = 2033db4fbfb9SMichel Lespinasse rb_entry(vma->vm_rb.rb_left, 2034db4fbfb9SMichel Lespinasse struct vm_area_struct, vm_rb); 2035db4fbfb9SMichel Lespinasse if (left->rb_subtree_gap >= length) { 2036db4fbfb9SMichel Lespinasse vma = left; 2037db4fbfb9SMichel Lespinasse continue; 2038db4fbfb9SMichel Lespinasse } 2039db4fbfb9SMichel Lespinasse } 2040db4fbfb9SMichel Lespinasse 2041db4fbfb9SMichel Lespinasse /* Go back up the rbtree to find next candidate node */ 2042db4fbfb9SMichel Lespinasse while (true) { 2043db4fbfb9SMichel Lespinasse struct rb_node *prev = &vma->vm_rb; 2044db4fbfb9SMichel Lespinasse if (!rb_parent(prev)) 2045db4fbfb9SMichel Lespinasse return -ENOMEM; 2046db4fbfb9SMichel Lespinasse vma = rb_entry(rb_parent(prev), 2047db4fbfb9SMichel Lespinasse struct vm_area_struct, vm_rb); 2048db4fbfb9SMichel Lespinasse if (prev == vma->vm_rb.rb_right) { 2049db4fbfb9SMichel Lespinasse gap_start = vma->vm_prev ? 20501be7107fSHugh Dickins vm_end_gap(vma->vm_prev) : 0; 2051db4fbfb9SMichel Lespinasse goto check_current; 2052db4fbfb9SMichel Lespinasse } 2053db4fbfb9SMichel Lespinasse } 2054db4fbfb9SMichel Lespinasse } 2055db4fbfb9SMichel Lespinasse 2056db4fbfb9SMichel Lespinasse found: 2057db4fbfb9SMichel Lespinasse /* We found a suitable gap. Clip it with the original high_limit. */ 2058db4fbfb9SMichel Lespinasse if (gap_end > info->high_limit) 2059db4fbfb9SMichel Lespinasse gap_end = info->high_limit; 2060db4fbfb9SMichel Lespinasse 2061db4fbfb9SMichel Lespinasse found_highest: 2062db4fbfb9SMichel Lespinasse /* Compute highest gap address at the desired alignment */ 2063db4fbfb9SMichel Lespinasse gap_end -= info->length; 2064db4fbfb9SMichel Lespinasse gap_end -= (gap_end - info->align_offset) & info->align_mask; 2065db4fbfb9SMichel Lespinasse 2066db4fbfb9SMichel Lespinasse VM_BUG_ON(gap_end < info->low_limit); 2067db4fbfb9SMichel Lespinasse VM_BUG_ON(gap_end < gap_start); 2068db4fbfb9SMichel Lespinasse return gap_end; 2069db4fbfb9SMichel Lespinasse } 2070db4fbfb9SMichel Lespinasse 2071f6795053SSteve Capper 2072f6795053SSteve Capper #ifndef arch_get_mmap_end 2073f6795053SSteve Capper #define arch_get_mmap_end(addr) (TASK_SIZE) 2074f6795053SSteve Capper #endif 2075f6795053SSteve Capper 2076f6795053SSteve Capper #ifndef arch_get_mmap_base 2077f6795053SSteve Capper #define arch_get_mmap_base(addr, base) (base) 2078f6795053SSteve Capper #endif 2079f6795053SSteve Capper 20801da177e4SLinus Torvalds /* Get an address range which is currently unmapped. 20811da177e4SLinus Torvalds * For shmat() with addr=0. 20821da177e4SLinus Torvalds * 20831da177e4SLinus Torvalds * Ugly calling convention alert: 20841da177e4SLinus Torvalds * Return value with the low bits set means error value, 20851da177e4SLinus Torvalds * ie 20861da177e4SLinus Torvalds * if (ret & ~PAGE_MASK) 20871da177e4SLinus Torvalds * error = ret; 20881da177e4SLinus Torvalds * 20891da177e4SLinus Torvalds * This function "knows" that -ENOMEM has the bits set. 20901da177e4SLinus Torvalds */ 20911da177e4SLinus Torvalds #ifndef HAVE_ARCH_UNMAPPED_AREA 20921da177e4SLinus Torvalds unsigned long 20931da177e4SLinus Torvalds arch_get_unmapped_area(struct file *filp, unsigned long addr, 20941da177e4SLinus Torvalds unsigned long len, unsigned long pgoff, unsigned long flags) 20951da177e4SLinus Torvalds { 20961da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 20971be7107fSHugh Dickins struct vm_area_struct *vma, *prev; 2098db4fbfb9SMichel Lespinasse struct vm_unmapped_area_info info; 2099f6795053SSteve Capper const unsigned long mmap_end = arch_get_mmap_end(addr); 21001da177e4SLinus Torvalds 2101f6795053SSteve Capper if (len > mmap_end - mmap_min_addr) 21021da177e4SLinus Torvalds return -ENOMEM; 21031da177e4SLinus Torvalds 210406abdfb4SBenjamin Herrenschmidt if (flags & MAP_FIXED) 210506abdfb4SBenjamin Herrenschmidt return addr; 210606abdfb4SBenjamin Herrenschmidt 21071da177e4SLinus Torvalds if (addr) { 21081da177e4SLinus Torvalds addr = PAGE_ALIGN(addr); 21091be7107fSHugh Dickins vma = find_vma_prev(mm, addr, &prev); 2110f6795053SSteve Capper if (mmap_end - len >= addr && addr >= mmap_min_addr && 21111be7107fSHugh Dickins (!vma || addr + len <= vm_start_gap(vma)) && 21121be7107fSHugh Dickins (!prev || addr >= vm_end_gap(prev))) 21131da177e4SLinus Torvalds return addr; 21141da177e4SLinus Torvalds } 21151da177e4SLinus Torvalds 2116db4fbfb9SMichel Lespinasse info.flags = 0; 2117db4fbfb9SMichel Lespinasse info.length = len; 21184e99b021SHeiko Carstens info.low_limit = mm->mmap_base; 2119f6795053SSteve Capper info.high_limit = mmap_end; 2120db4fbfb9SMichel Lespinasse info.align_mask = 0; 2121db4fbfb9SMichel Lespinasse return vm_unmapped_area(&info); 21221da177e4SLinus Torvalds } 21231da177e4SLinus Torvalds #endif 21241da177e4SLinus Torvalds 21251da177e4SLinus Torvalds /* 21261da177e4SLinus Torvalds * This mmap-allocator allocates new areas top-down from below the 21271da177e4SLinus Torvalds * stack's low limit (the base): 21281da177e4SLinus Torvalds */ 21291da177e4SLinus Torvalds #ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN 21301da177e4SLinus Torvalds unsigned long 213143cca0b1SYang Fan arch_get_unmapped_area_topdown(struct file *filp, unsigned long addr, 213243cca0b1SYang Fan unsigned long len, unsigned long pgoff, 213343cca0b1SYang Fan unsigned long flags) 21341da177e4SLinus Torvalds { 21351be7107fSHugh Dickins struct vm_area_struct *vma, *prev; 21361da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 2137db4fbfb9SMichel Lespinasse struct vm_unmapped_area_info info; 2138f6795053SSteve Capper const unsigned long mmap_end = arch_get_mmap_end(addr); 21391da177e4SLinus Torvalds 21401da177e4SLinus Torvalds /* requested length too big for entire address space */ 2141f6795053SSteve Capper if (len > mmap_end - mmap_min_addr) 21421da177e4SLinus Torvalds return -ENOMEM; 21431da177e4SLinus Torvalds 214406abdfb4SBenjamin Herrenschmidt if (flags & MAP_FIXED) 214506abdfb4SBenjamin Herrenschmidt return addr; 214606abdfb4SBenjamin Herrenschmidt 21471da177e4SLinus Torvalds /* requesting a specific address */ 21481da177e4SLinus Torvalds if (addr) { 21491da177e4SLinus Torvalds addr = PAGE_ALIGN(addr); 21501be7107fSHugh Dickins vma = find_vma_prev(mm, addr, &prev); 2151f6795053SSteve Capper if (mmap_end - len >= addr && addr >= mmap_min_addr && 21521be7107fSHugh Dickins (!vma || addr + len <= vm_start_gap(vma)) && 21531be7107fSHugh Dickins (!prev || addr >= vm_end_gap(prev))) 21541da177e4SLinus Torvalds return addr; 21551da177e4SLinus Torvalds } 21561da177e4SLinus Torvalds 2157db4fbfb9SMichel Lespinasse info.flags = VM_UNMAPPED_AREA_TOPDOWN; 2158db4fbfb9SMichel Lespinasse info.length = len; 21592afc745fSAkira Takeuchi info.low_limit = max(PAGE_SIZE, mmap_min_addr); 2160f6795053SSteve Capper info.high_limit = arch_get_mmap_base(addr, mm->mmap_base); 2161db4fbfb9SMichel Lespinasse info.align_mask = 0; 2162db4fbfb9SMichel Lespinasse addr = vm_unmapped_area(&info); 2163b716ad95SXiao Guangrong 21641da177e4SLinus Torvalds /* 21651da177e4SLinus Torvalds * A failed mmap() very likely causes application failure, 21661da177e4SLinus Torvalds * so fall back to the bottom-up function here. This scenario 21671da177e4SLinus Torvalds * can happen with large stack limits and large mmap() 21681da177e4SLinus Torvalds * allocations. 21691da177e4SLinus Torvalds */ 2170de1741a1SAlexander Kuleshov if (offset_in_page(addr)) { 2171db4fbfb9SMichel Lespinasse VM_BUG_ON(addr != -ENOMEM); 2172db4fbfb9SMichel Lespinasse info.flags = 0; 2173db4fbfb9SMichel Lespinasse info.low_limit = TASK_UNMAPPED_BASE; 2174f6795053SSteve Capper info.high_limit = mmap_end; 2175db4fbfb9SMichel Lespinasse addr = vm_unmapped_area(&info); 2176db4fbfb9SMichel Lespinasse } 21771da177e4SLinus Torvalds 21781da177e4SLinus Torvalds return addr; 21791da177e4SLinus Torvalds } 21801da177e4SLinus Torvalds #endif 21811da177e4SLinus Torvalds 21821da177e4SLinus Torvalds unsigned long 21831da177e4SLinus Torvalds get_unmapped_area(struct file *file, unsigned long addr, unsigned long len, 21841da177e4SLinus Torvalds unsigned long pgoff, unsigned long flags) 21851da177e4SLinus Torvalds { 218606abdfb4SBenjamin Herrenschmidt unsigned long (*get_area)(struct file *, unsigned long, 218706abdfb4SBenjamin Herrenschmidt unsigned long, unsigned long, unsigned long); 218807ab67c8SLinus Torvalds 21899206de95SAl Viro unsigned long error = arch_mmap_check(addr, len, flags); 21909206de95SAl Viro if (error) 21919206de95SAl Viro return error; 21929206de95SAl Viro 21939206de95SAl Viro /* Careful about overflows.. */ 21949206de95SAl Viro if (len > TASK_SIZE) 21959206de95SAl Viro return -ENOMEM; 21969206de95SAl Viro 219707ab67c8SLinus Torvalds get_area = current->mm->get_unmapped_area; 2198c01d5b30SHugh Dickins if (file) { 2199c01d5b30SHugh Dickins if (file->f_op->get_unmapped_area) 220007ab67c8SLinus Torvalds get_area = file->f_op->get_unmapped_area; 2201c01d5b30SHugh Dickins } else if (flags & MAP_SHARED) { 2202c01d5b30SHugh Dickins /* 2203c01d5b30SHugh Dickins * mmap_region() will call shmem_zero_setup() to create a file, 2204c01d5b30SHugh Dickins * so use shmem's get_unmapped_area in case it can be huge. 2205c01d5b30SHugh Dickins * do_mmap_pgoff() will clear pgoff, so match alignment. 2206c01d5b30SHugh Dickins */ 2207c01d5b30SHugh Dickins pgoff = 0; 2208c01d5b30SHugh Dickins get_area = shmem_get_unmapped_area; 2209c01d5b30SHugh Dickins } 2210c01d5b30SHugh Dickins 221107ab67c8SLinus Torvalds addr = get_area(file, addr, len, pgoff, flags); 221207ab67c8SLinus Torvalds if (IS_ERR_VALUE(addr)) 221307ab67c8SLinus Torvalds return addr; 221407ab67c8SLinus Torvalds 22151da177e4SLinus Torvalds if (addr > TASK_SIZE - len) 22161da177e4SLinus Torvalds return -ENOMEM; 2217de1741a1SAlexander Kuleshov if (offset_in_page(addr)) 22181da177e4SLinus Torvalds return -EINVAL; 221906abdfb4SBenjamin Herrenschmidt 22209ac4ed4bSAl Viro error = security_mmap_addr(addr); 22219ac4ed4bSAl Viro return error ? error : addr; 22221da177e4SLinus Torvalds } 22231da177e4SLinus Torvalds 22241da177e4SLinus Torvalds EXPORT_SYMBOL(get_unmapped_area); 22251da177e4SLinus Torvalds 22261da177e4SLinus Torvalds /* Look up the first VMA which satisfies addr < vm_end, NULL if none. */ 22271da177e4SLinus Torvalds struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr) 22281da177e4SLinus Torvalds { 2229615d6e87SDavidlohr Bueso struct rb_node *rb_node; 2230615d6e87SDavidlohr Bueso struct vm_area_struct *vma; 22311da177e4SLinus Torvalds 22321da177e4SLinus Torvalds /* Check the cache first. */ 2233615d6e87SDavidlohr Bueso vma = vmacache_find(mm, addr); 2234615d6e87SDavidlohr Bueso if (likely(vma)) 2235615d6e87SDavidlohr Bueso return vma; 22361da177e4SLinus Torvalds 22371da177e4SLinus Torvalds rb_node = mm->mm_rb.rb_node; 22381da177e4SLinus Torvalds 22391da177e4SLinus Torvalds while (rb_node) { 2240615d6e87SDavidlohr Bueso struct vm_area_struct *tmp; 22411da177e4SLinus Torvalds 2242615d6e87SDavidlohr Bueso tmp = rb_entry(rb_node, struct vm_area_struct, vm_rb); 22431da177e4SLinus Torvalds 2244615d6e87SDavidlohr Bueso if (tmp->vm_end > addr) { 2245615d6e87SDavidlohr Bueso vma = tmp; 2246615d6e87SDavidlohr Bueso if (tmp->vm_start <= addr) 22471da177e4SLinus Torvalds break; 22481da177e4SLinus Torvalds rb_node = rb_node->rb_left; 22491da177e4SLinus Torvalds } else 22501da177e4SLinus Torvalds rb_node = rb_node->rb_right; 22511da177e4SLinus Torvalds } 2252615d6e87SDavidlohr Bueso 22531da177e4SLinus Torvalds if (vma) 2254615d6e87SDavidlohr Bueso vmacache_update(addr, vma); 22551da177e4SLinus Torvalds return vma; 22561da177e4SLinus Torvalds } 22571da177e4SLinus Torvalds 22581da177e4SLinus Torvalds EXPORT_SYMBOL(find_vma); 22591da177e4SLinus Torvalds 22606bd4837dSKOSAKI Motohiro /* 22616bd4837dSKOSAKI Motohiro * Same as find_vma, but also return a pointer to the previous VMA in *pprev. 22626bd4837dSKOSAKI Motohiro */ 22631da177e4SLinus Torvalds struct vm_area_struct * 22641da177e4SLinus Torvalds find_vma_prev(struct mm_struct *mm, unsigned long addr, 22651da177e4SLinus Torvalds struct vm_area_struct **pprev) 22661da177e4SLinus Torvalds { 22676bd4837dSKOSAKI Motohiro struct vm_area_struct *vma; 22681da177e4SLinus Torvalds 22696bd4837dSKOSAKI Motohiro vma = find_vma(mm, addr); 227083cd904dSMikulas Patocka if (vma) { 227183cd904dSMikulas Patocka *pprev = vma->vm_prev; 227283cd904dSMikulas Patocka } else { 227383cd904dSMikulas Patocka struct rb_node *rb_node = mm->mm_rb.rb_node; 227483cd904dSMikulas Patocka *pprev = NULL; 227583cd904dSMikulas Patocka while (rb_node) { 227683cd904dSMikulas Patocka *pprev = rb_entry(rb_node, struct vm_area_struct, vm_rb); 227783cd904dSMikulas Patocka rb_node = rb_node->rb_right; 227883cd904dSMikulas Patocka } 227983cd904dSMikulas Patocka } 22806bd4837dSKOSAKI Motohiro return vma; 22811da177e4SLinus Torvalds } 22821da177e4SLinus Torvalds 22831da177e4SLinus Torvalds /* 22841da177e4SLinus Torvalds * Verify that the stack growth is acceptable and 22851da177e4SLinus Torvalds * update accounting. This is shared with both the 22861da177e4SLinus Torvalds * grow-up and grow-down cases. 22871da177e4SLinus Torvalds */ 22881be7107fSHugh Dickins static int acct_stack_growth(struct vm_area_struct *vma, 22891be7107fSHugh Dickins unsigned long size, unsigned long grow) 22901da177e4SLinus Torvalds { 22911da177e4SLinus Torvalds struct mm_struct *mm = vma->vm_mm; 22921be7107fSHugh Dickins unsigned long new_start; 22931da177e4SLinus Torvalds 22941da177e4SLinus Torvalds /* address space limit tests */ 229584638335SKonstantin Khlebnikov if (!may_expand_vm(mm, vma->vm_flags, grow)) 22961da177e4SLinus Torvalds return -ENOMEM; 22971da177e4SLinus Torvalds 22981da177e4SLinus Torvalds /* Stack limit test */ 229924c79d8eSKrzysztof Opasiak if (size > rlimit(RLIMIT_STACK)) 23001da177e4SLinus Torvalds return -ENOMEM; 23011da177e4SLinus Torvalds 23021da177e4SLinus Torvalds /* mlock limit tests */ 23031da177e4SLinus Torvalds if (vma->vm_flags & VM_LOCKED) { 23041da177e4SLinus Torvalds unsigned long locked; 23051da177e4SLinus Torvalds unsigned long limit; 23061da177e4SLinus Torvalds locked = mm->locked_vm + grow; 230724c79d8eSKrzysztof Opasiak limit = rlimit(RLIMIT_MEMLOCK); 230859e99e5bSJiri Slaby limit >>= PAGE_SHIFT; 23091da177e4SLinus Torvalds if (locked > limit && !capable(CAP_IPC_LOCK)) 23101da177e4SLinus Torvalds return -ENOMEM; 23111da177e4SLinus Torvalds } 23121da177e4SLinus Torvalds 23130d59a01bSAdam Litke /* Check to ensure the stack will not grow into a hugetlb-only region */ 23140d59a01bSAdam Litke new_start = (vma->vm_flags & VM_GROWSUP) ? vma->vm_start : 23150d59a01bSAdam Litke vma->vm_end - size; 23160d59a01bSAdam Litke if (is_hugepage_only_range(vma->vm_mm, new_start, size)) 23170d59a01bSAdam Litke return -EFAULT; 23180d59a01bSAdam Litke 23191da177e4SLinus Torvalds /* 23201da177e4SLinus Torvalds * Overcommit.. This must be the final test, as it will 23211da177e4SLinus Torvalds * update security statistics. 23221da177e4SLinus Torvalds */ 232305fa199dSHugh Dickins if (security_vm_enough_memory_mm(mm, grow)) 23241da177e4SLinus Torvalds return -ENOMEM; 23251da177e4SLinus Torvalds 23261da177e4SLinus Torvalds return 0; 23271da177e4SLinus Torvalds } 23281da177e4SLinus Torvalds 232946dea3d0SHugh Dickins #if defined(CONFIG_STACK_GROWSUP) || defined(CONFIG_IA64) 23301da177e4SLinus Torvalds /* 233146dea3d0SHugh Dickins * PA-RISC uses this for its stack; IA64 for its Register Backing Store. 233246dea3d0SHugh Dickins * vma is the last one with address > vma->vm_end. Have to extend vma. 23331da177e4SLinus Torvalds */ 233446dea3d0SHugh Dickins int expand_upwards(struct vm_area_struct *vma, unsigned long address) 23351da177e4SLinus Torvalds { 233609357814SOleg Nesterov struct mm_struct *mm = vma->vm_mm; 23371be7107fSHugh Dickins struct vm_area_struct *next; 23381be7107fSHugh Dickins unsigned long gap_addr; 233912352d3cSKonstantin Khlebnikov int error = 0; 23401da177e4SLinus Torvalds 23411da177e4SLinus Torvalds if (!(vma->vm_flags & VM_GROWSUP)) 23421da177e4SLinus Torvalds return -EFAULT; 23431da177e4SLinus Torvalds 2344bd726c90SHelge Deller /* Guard against exceeding limits of the address space. */ 23451be7107fSHugh Dickins address &= PAGE_MASK; 234637511fb5SHelge Deller if (address >= (TASK_SIZE & PAGE_MASK)) 234712352d3cSKonstantin Khlebnikov return -ENOMEM; 2348bd726c90SHelge Deller address += PAGE_SIZE; 234912352d3cSKonstantin Khlebnikov 23501be7107fSHugh Dickins /* Enforce stack_guard_gap */ 23511be7107fSHugh Dickins gap_addr = address + stack_guard_gap; 2352bd726c90SHelge Deller 2353bd726c90SHelge Deller /* Guard against overflow */ 2354bd726c90SHelge Deller if (gap_addr < address || gap_addr > TASK_SIZE) 2355bd726c90SHelge Deller gap_addr = TASK_SIZE; 2356bd726c90SHelge Deller 23571be7107fSHugh Dickins next = vma->vm_next; 2358561b5e07SMichal Hocko if (next && next->vm_start < gap_addr && 2359561b5e07SMichal Hocko (next->vm_flags & (VM_WRITE|VM_READ|VM_EXEC))) { 23601be7107fSHugh Dickins if (!(next->vm_flags & VM_GROWSUP)) 23611be7107fSHugh Dickins return -ENOMEM; 23621be7107fSHugh Dickins /* Check that both stack segments have the same anon_vma? */ 23631be7107fSHugh Dickins } 23641be7107fSHugh Dickins 236512352d3cSKonstantin Khlebnikov /* We must make sure the anon_vma is allocated. */ 23661da177e4SLinus Torvalds if (unlikely(anon_vma_prepare(vma))) 23671da177e4SLinus Torvalds return -ENOMEM; 23681da177e4SLinus Torvalds 23691da177e4SLinus Torvalds /* 23701da177e4SLinus Torvalds * vma->vm_start/vm_end cannot change under us because the caller 23711da177e4SLinus Torvalds * is required to hold the mmap_sem in read mode. We need the 23721da177e4SLinus Torvalds * anon_vma lock to serialize against concurrent expand_stacks. 23731da177e4SLinus Torvalds */ 237412352d3cSKonstantin Khlebnikov anon_vma_lock_write(vma->anon_vma); 23751da177e4SLinus Torvalds 23761da177e4SLinus Torvalds /* Somebody else might have raced and expanded it already */ 23771da177e4SLinus Torvalds if (address > vma->vm_end) { 23781da177e4SLinus Torvalds unsigned long size, grow; 23791da177e4SLinus Torvalds 23801da177e4SLinus Torvalds size = address - vma->vm_start; 23811da177e4SLinus Torvalds grow = (address - vma->vm_end) >> PAGE_SHIFT; 23821da177e4SLinus Torvalds 238342c36f63SHugh Dickins error = -ENOMEM; 238442c36f63SHugh Dickins if (vma->vm_pgoff + (size >> PAGE_SHIFT) >= vma->vm_pgoff) { 23851da177e4SLinus Torvalds error = acct_stack_growth(vma, size, grow); 23863af9e859SEric B Munson if (!error) { 23874128997bSMichel Lespinasse /* 23884128997bSMichel Lespinasse * vma_gap_update() doesn't support concurrent 23894128997bSMichel Lespinasse * updates, but we only hold a shared mmap_sem 23904128997bSMichel Lespinasse * lock here, so we need to protect against 23914128997bSMichel Lespinasse * concurrent vma expansions. 239212352d3cSKonstantin Khlebnikov * anon_vma_lock_write() doesn't help here, as 23934128997bSMichel Lespinasse * we don't guarantee that all growable vmas 23944128997bSMichel Lespinasse * in a mm share the same root anon vma. 23954128997bSMichel Lespinasse * So, we reuse mm->page_table_lock to guard 23964128997bSMichel Lespinasse * against concurrent vma expansions. 23974128997bSMichel Lespinasse */ 239809357814SOleg Nesterov spin_lock(&mm->page_table_lock); 239987e8827bSOleg Nesterov if (vma->vm_flags & VM_LOCKED) 240009357814SOleg Nesterov mm->locked_vm += grow; 240184638335SKonstantin Khlebnikov vm_stat_account(mm, vma->vm_flags, grow); 2402bf181b9fSMichel Lespinasse anon_vma_interval_tree_pre_update_vma(vma); 24031da177e4SLinus Torvalds vma->vm_end = address; 2404bf181b9fSMichel Lespinasse anon_vma_interval_tree_post_update_vma(vma); 2405d3737187SMichel Lespinasse if (vma->vm_next) 2406d3737187SMichel Lespinasse vma_gap_update(vma->vm_next); 2407d3737187SMichel Lespinasse else 24081be7107fSHugh Dickins mm->highest_vm_end = vm_end_gap(vma); 240909357814SOleg Nesterov spin_unlock(&mm->page_table_lock); 24104128997bSMichel Lespinasse 24113af9e859SEric B Munson perf_event_mmap(vma); 24123af9e859SEric B Munson } 24131da177e4SLinus Torvalds } 241442c36f63SHugh Dickins } 241512352d3cSKonstantin Khlebnikov anon_vma_unlock_write(vma->anon_vma); 24166d50e60cSDavid Rientjes khugepaged_enter_vma_merge(vma, vma->vm_flags); 241709357814SOleg Nesterov validate_mm(mm); 24181da177e4SLinus Torvalds return error; 24191da177e4SLinus Torvalds } 242046dea3d0SHugh Dickins #endif /* CONFIG_STACK_GROWSUP || CONFIG_IA64 */ 242146dea3d0SHugh Dickins 24221da177e4SLinus Torvalds /* 24231da177e4SLinus Torvalds * vma is the first one with address < vma->vm_start. Have to extend vma. 24241da177e4SLinus Torvalds */ 2425d05f3169SMichal Hocko int expand_downwards(struct vm_area_struct *vma, 2426b6a2fea3SOllie Wild unsigned long address) 24271da177e4SLinus Torvalds { 242809357814SOleg Nesterov struct mm_struct *mm = vma->vm_mm; 24291be7107fSHugh Dickins struct vm_area_struct *prev; 24300a1d5299SJann Horn int error = 0; 24311da177e4SLinus Torvalds 24328869477aSEric Paris address &= PAGE_MASK; 24330a1d5299SJann Horn if (address < mmap_min_addr) 24340a1d5299SJann Horn return -EPERM; 24358869477aSEric Paris 24361be7107fSHugh Dickins /* Enforce stack_guard_gap */ 24371be7107fSHugh Dickins prev = vma->vm_prev; 24381be7107fSHugh Dickins /* Check that both stack segments have the same anon_vma? */ 243932e4e6d5SOleg Nesterov if (prev && !(prev->vm_flags & VM_GROWSDOWN) && 244032e4e6d5SOleg Nesterov (prev->vm_flags & (VM_WRITE|VM_READ|VM_EXEC))) { 244132e4e6d5SOleg Nesterov if (address - prev->vm_end < stack_guard_gap) 244232e4e6d5SOleg Nesterov return -ENOMEM; 24431be7107fSHugh Dickins } 24441be7107fSHugh Dickins 244512352d3cSKonstantin Khlebnikov /* We must make sure the anon_vma is allocated. */ 244612352d3cSKonstantin Khlebnikov if (unlikely(anon_vma_prepare(vma))) 244712352d3cSKonstantin Khlebnikov return -ENOMEM; 24481da177e4SLinus Torvalds 24491da177e4SLinus Torvalds /* 24501da177e4SLinus Torvalds * vma->vm_start/vm_end cannot change under us because the caller 24511da177e4SLinus Torvalds * is required to hold the mmap_sem in read mode. We need the 24521da177e4SLinus Torvalds * anon_vma lock to serialize against concurrent expand_stacks. 24531da177e4SLinus Torvalds */ 245412352d3cSKonstantin Khlebnikov anon_vma_lock_write(vma->anon_vma); 24551da177e4SLinus Torvalds 24561da177e4SLinus Torvalds /* Somebody else might have raced and expanded it already */ 24571da177e4SLinus Torvalds if (address < vma->vm_start) { 24581da177e4SLinus Torvalds unsigned long size, grow; 24591da177e4SLinus Torvalds 24601da177e4SLinus Torvalds size = vma->vm_end - address; 24611da177e4SLinus Torvalds grow = (vma->vm_start - address) >> PAGE_SHIFT; 24621da177e4SLinus Torvalds 2463a626ca6aSLinus Torvalds error = -ENOMEM; 2464a626ca6aSLinus Torvalds if (grow <= vma->vm_pgoff) { 24651da177e4SLinus Torvalds error = acct_stack_growth(vma, size, grow); 24661da177e4SLinus Torvalds if (!error) { 24674128997bSMichel Lespinasse /* 24684128997bSMichel Lespinasse * vma_gap_update() doesn't support concurrent 24694128997bSMichel Lespinasse * updates, but we only hold a shared mmap_sem 24704128997bSMichel Lespinasse * lock here, so we need to protect against 24714128997bSMichel Lespinasse * concurrent vma expansions. 247212352d3cSKonstantin Khlebnikov * anon_vma_lock_write() doesn't help here, as 24734128997bSMichel Lespinasse * we don't guarantee that all growable vmas 24744128997bSMichel Lespinasse * in a mm share the same root anon vma. 24754128997bSMichel Lespinasse * So, we reuse mm->page_table_lock to guard 24764128997bSMichel Lespinasse * against concurrent vma expansions. 24774128997bSMichel Lespinasse */ 247809357814SOleg Nesterov spin_lock(&mm->page_table_lock); 247987e8827bSOleg Nesterov if (vma->vm_flags & VM_LOCKED) 248009357814SOleg Nesterov mm->locked_vm += grow; 248184638335SKonstantin Khlebnikov vm_stat_account(mm, vma->vm_flags, grow); 2482bf181b9fSMichel Lespinasse anon_vma_interval_tree_pre_update_vma(vma); 24831da177e4SLinus Torvalds vma->vm_start = address; 24841da177e4SLinus Torvalds vma->vm_pgoff -= grow; 2485bf181b9fSMichel Lespinasse anon_vma_interval_tree_post_update_vma(vma); 2486d3737187SMichel Lespinasse vma_gap_update(vma); 248709357814SOleg Nesterov spin_unlock(&mm->page_table_lock); 24884128997bSMichel Lespinasse 24893af9e859SEric B Munson perf_event_mmap(vma); 24901da177e4SLinus Torvalds } 24911da177e4SLinus Torvalds } 2492a626ca6aSLinus Torvalds } 249312352d3cSKonstantin Khlebnikov anon_vma_unlock_write(vma->anon_vma); 24946d50e60cSDavid Rientjes khugepaged_enter_vma_merge(vma, vma->vm_flags); 249509357814SOleg Nesterov validate_mm(mm); 24961da177e4SLinus Torvalds return error; 24971da177e4SLinus Torvalds } 24981da177e4SLinus Torvalds 24991be7107fSHugh Dickins /* enforced gap between the expanding stack and other mappings. */ 25001be7107fSHugh Dickins unsigned long stack_guard_gap = 256UL<<PAGE_SHIFT; 25011be7107fSHugh Dickins 25021be7107fSHugh Dickins static int __init cmdline_parse_stack_guard_gap(char *p) 25031be7107fSHugh Dickins { 25041be7107fSHugh Dickins unsigned long val; 25051be7107fSHugh Dickins char *endptr; 25061be7107fSHugh Dickins 25071be7107fSHugh Dickins val = simple_strtoul(p, &endptr, 10); 25081be7107fSHugh Dickins if (!*endptr) 25091be7107fSHugh Dickins stack_guard_gap = val << PAGE_SHIFT; 25101be7107fSHugh Dickins 25111be7107fSHugh Dickins return 0; 25121be7107fSHugh Dickins } 25131be7107fSHugh Dickins __setup("stack_guard_gap=", cmdline_parse_stack_guard_gap); 25141be7107fSHugh Dickins 2515b6a2fea3SOllie Wild #ifdef CONFIG_STACK_GROWSUP 2516b6a2fea3SOllie Wild int expand_stack(struct vm_area_struct *vma, unsigned long address) 2517b6a2fea3SOllie Wild { 2518b6a2fea3SOllie Wild return expand_upwards(vma, address); 2519b6a2fea3SOllie Wild } 2520b6a2fea3SOllie Wild 2521b6a2fea3SOllie Wild struct vm_area_struct * 2522b6a2fea3SOllie Wild find_extend_vma(struct mm_struct *mm, unsigned long addr) 2523b6a2fea3SOllie Wild { 2524b6a2fea3SOllie Wild struct vm_area_struct *vma, *prev; 2525b6a2fea3SOllie Wild 2526b6a2fea3SOllie Wild addr &= PAGE_MASK; 2527b6a2fea3SOllie Wild vma = find_vma_prev(mm, addr, &prev); 2528b6a2fea3SOllie Wild if (vma && (vma->vm_start <= addr)) 2529b6a2fea3SOllie Wild return vma; 253004f5866eSAndrea Arcangeli /* don't alter vm_end if the coredump is running */ 253104f5866eSAndrea Arcangeli if (!prev || !mmget_still_valid(mm) || expand_stack(prev, addr)) 2532b6a2fea3SOllie Wild return NULL; 2533cea10a19SMichel Lespinasse if (prev->vm_flags & VM_LOCKED) 2534fc05f566SKirill A. Shutemov populate_vma_page_range(prev, addr, prev->vm_end, NULL); 2535b6a2fea3SOllie Wild return prev; 2536b6a2fea3SOllie Wild } 2537b6a2fea3SOllie Wild #else 2538b6a2fea3SOllie Wild int expand_stack(struct vm_area_struct *vma, unsigned long address) 2539b6a2fea3SOllie Wild { 2540b6a2fea3SOllie Wild return expand_downwards(vma, address); 2541b6a2fea3SOllie Wild } 2542b6a2fea3SOllie Wild 25431da177e4SLinus Torvalds struct vm_area_struct * 25441da177e4SLinus Torvalds find_extend_vma(struct mm_struct *mm, unsigned long addr) 25451da177e4SLinus Torvalds { 25461da177e4SLinus Torvalds struct vm_area_struct *vma; 25471da177e4SLinus Torvalds unsigned long start; 25481da177e4SLinus Torvalds 25491da177e4SLinus Torvalds addr &= PAGE_MASK; 25501da177e4SLinus Torvalds vma = find_vma(mm, addr); 25511da177e4SLinus Torvalds if (!vma) 25521da177e4SLinus Torvalds return NULL; 25531da177e4SLinus Torvalds if (vma->vm_start <= addr) 25541da177e4SLinus Torvalds return vma; 25551da177e4SLinus Torvalds if (!(vma->vm_flags & VM_GROWSDOWN)) 25561da177e4SLinus Torvalds return NULL; 255704f5866eSAndrea Arcangeli /* don't alter vm_start if the coredump is running */ 255804f5866eSAndrea Arcangeli if (!mmget_still_valid(mm)) 255904f5866eSAndrea Arcangeli return NULL; 25601da177e4SLinus Torvalds start = vma->vm_start; 25611da177e4SLinus Torvalds if (expand_stack(vma, addr)) 25621da177e4SLinus Torvalds return NULL; 2563cea10a19SMichel Lespinasse if (vma->vm_flags & VM_LOCKED) 2564fc05f566SKirill A. Shutemov populate_vma_page_range(vma, addr, start, NULL); 25651da177e4SLinus Torvalds return vma; 25661da177e4SLinus Torvalds } 25671da177e4SLinus Torvalds #endif 25681da177e4SLinus Torvalds 2569e1d6d01aSJesse Barnes EXPORT_SYMBOL_GPL(find_extend_vma); 2570e1d6d01aSJesse Barnes 25712c0b3814SHugh Dickins /* 25722c0b3814SHugh Dickins * Ok - we have the memory areas we should free on the vma list, 25732c0b3814SHugh Dickins * so release them, and do the vma updates. 25741da177e4SLinus Torvalds * 25752c0b3814SHugh Dickins * Called with the mm semaphore held. 25761da177e4SLinus Torvalds */ 25772c0b3814SHugh Dickins static void remove_vma_list(struct mm_struct *mm, struct vm_area_struct *vma) 25781da177e4SLinus Torvalds { 25794f74d2c8SLinus Torvalds unsigned long nr_accounted = 0; 25804f74d2c8SLinus Torvalds 2581365e9c87SHugh Dickins /* Update high watermark before we lower total_vm */ 2582365e9c87SHugh Dickins update_hiwater_vm(mm); 25832c0b3814SHugh Dickins do { 2584ab50b8edSHugh Dickins long nrpages = vma_pages(vma); 25851da177e4SLinus Torvalds 25864f74d2c8SLinus Torvalds if (vma->vm_flags & VM_ACCOUNT) 25874f74d2c8SLinus Torvalds nr_accounted += nrpages; 258884638335SKonstantin Khlebnikov vm_stat_account(mm, vma->vm_flags, -nrpages); 2589a8fb5618SHugh Dickins vma = remove_vma(vma); 2590146425a3SHugh Dickins } while (vma); 25914f74d2c8SLinus Torvalds vm_unacct_memory(nr_accounted); 25921da177e4SLinus Torvalds validate_mm(mm); 25931da177e4SLinus Torvalds } 25941da177e4SLinus Torvalds 25951da177e4SLinus Torvalds /* 25961da177e4SLinus Torvalds * Get rid of page table information in the indicated region. 25971da177e4SLinus Torvalds * 2598f10df686SPaolo 'Blaisorblade' Giarrusso * Called with the mm semaphore held. 25991da177e4SLinus Torvalds */ 26001da177e4SLinus Torvalds static void unmap_region(struct mm_struct *mm, 2601e0da382cSHugh Dickins struct vm_area_struct *vma, struct vm_area_struct *prev, 2602e0da382cSHugh Dickins unsigned long start, unsigned long end) 26031da177e4SLinus Torvalds { 2604e0da382cSHugh Dickins struct vm_area_struct *next = prev ? prev->vm_next : mm->mmap; 2605d16dfc55SPeter Zijlstra struct mmu_gather tlb; 26061da177e4SLinus Torvalds 26071da177e4SLinus Torvalds lru_add_drain(); 26082b047252SLinus Torvalds tlb_gather_mmu(&tlb, mm, start, end); 2609365e9c87SHugh Dickins update_hiwater_rss(mm); 26104f74d2c8SLinus Torvalds unmap_vmas(&tlb, vma, start, end); 2611d16dfc55SPeter Zijlstra free_pgtables(&tlb, vma, prev ? prev->vm_end : FIRST_USER_ADDRESS, 26126ee8630eSHugh Dickins next ? next->vm_start : USER_PGTABLES_CEILING); 2613d16dfc55SPeter Zijlstra tlb_finish_mmu(&tlb, start, end); 26141da177e4SLinus Torvalds } 26151da177e4SLinus Torvalds 26161da177e4SLinus Torvalds /* 26171da177e4SLinus Torvalds * Create a list of vma's touched by the unmap, removing them from the mm's 26181da177e4SLinus Torvalds * vma list as we go.. 26191da177e4SLinus Torvalds */ 26201da177e4SLinus Torvalds static void 26211da177e4SLinus Torvalds detach_vmas_to_be_unmapped(struct mm_struct *mm, struct vm_area_struct *vma, 26221da177e4SLinus Torvalds struct vm_area_struct *prev, unsigned long end) 26231da177e4SLinus Torvalds { 26241da177e4SLinus Torvalds struct vm_area_struct **insertion_point; 26251da177e4SLinus Torvalds struct vm_area_struct *tail_vma = NULL; 26261da177e4SLinus Torvalds 26271da177e4SLinus Torvalds insertion_point = (prev ? &prev->vm_next : &mm->mmap); 2628297c5eeeSLinus Torvalds vma->vm_prev = NULL; 26291da177e4SLinus Torvalds do { 2630d3737187SMichel Lespinasse vma_rb_erase(vma, &mm->mm_rb); 26311da177e4SLinus Torvalds mm->map_count--; 26321da177e4SLinus Torvalds tail_vma = vma; 26331da177e4SLinus Torvalds vma = vma->vm_next; 26341da177e4SLinus Torvalds } while (vma && vma->vm_start < end); 26351da177e4SLinus Torvalds *insertion_point = vma; 2636d3737187SMichel Lespinasse if (vma) { 2637297c5eeeSLinus Torvalds vma->vm_prev = prev; 2638d3737187SMichel Lespinasse vma_gap_update(vma); 2639d3737187SMichel Lespinasse } else 26401be7107fSHugh Dickins mm->highest_vm_end = prev ? vm_end_gap(prev) : 0; 26411da177e4SLinus Torvalds tail_vma->vm_next = NULL; 2642615d6e87SDavidlohr Bueso 2643615d6e87SDavidlohr Bueso /* Kill the cache */ 2644615d6e87SDavidlohr Bueso vmacache_invalidate(mm); 26451da177e4SLinus Torvalds } 26461da177e4SLinus Torvalds 26471da177e4SLinus Torvalds /* 2648def5efe0SDavid Rientjes * __split_vma() bypasses sysctl_max_map_count checking. We use this where it 2649def5efe0SDavid Rientjes * has already been checked or doesn't make sense to fail. 26501da177e4SLinus Torvalds */ 2651def5efe0SDavid Rientjes int __split_vma(struct mm_struct *mm, struct vm_area_struct *vma, 26521da177e4SLinus Torvalds unsigned long addr, int new_below) 26531da177e4SLinus Torvalds { 26541da177e4SLinus Torvalds struct vm_area_struct *new; 2655e3975891SChen Gang int err; 26561da177e4SLinus Torvalds 265731383c68SDan Williams if (vma->vm_ops && vma->vm_ops->split) { 265831383c68SDan Williams err = vma->vm_ops->split(vma, addr); 265931383c68SDan Williams if (err) 266031383c68SDan Williams return err; 266131383c68SDan Williams } 26621da177e4SLinus Torvalds 26633928d4f5SLinus Torvalds new = vm_area_dup(vma); 26641da177e4SLinus Torvalds if (!new) 2665e3975891SChen Gang return -ENOMEM; 26661da177e4SLinus Torvalds 26671da177e4SLinus Torvalds if (new_below) 26681da177e4SLinus Torvalds new->vm_end = addr; 26691da177e4SLinus Torvalds else { 26701da177e4SLinus Torvalds new->vm_start = addr; 26711da177e4SLinus Torvalds new->vm_pgoff += ((addr - vma->vm_start) >> PAGE_SHIFT); 26721da177e4SLinus Torvalds } 26731da177e4SLinus Torvalds 2674ef0855d3SOleg Nesterov err = vma_dup_policy(vma, new); 2675ef0855d3SOleg Nesterov if (err) 26765beb4930SRik van Riel goto out_free_vma; 26771da177e4SLinus Torvalds 2678c4ea95d7SDaniel Forrest err = anon_vma_clone(new, vma); 2679c4ea95d7SDaniel Forrest if (err) 26805beb4930SRik van Riel goto out_free_mpol; 26815beb4930SRik van Riel 2682e9714acfSKonstantin Khlebnikov if (new->vm_file) 26831da177e4SLinus Torvalds get_file(new->vm_file); 26841da177e4SLinus Torvalds 26851da177e4SLinus Torvalds if (new->vm_ops && new->vm_ops->open) 26861da177e4SLinus Torvalds new->vm_ops->open(new); 26871da177e4SLinus Torvalds 26881da177e4SLinus Torvalds if (new_below) 26895beb4930SRik van Riel err = vma_adjust(vma, addr, vma->vm_end, vma->vm_pgoff + 26901da177e4SLinus Torvalds ((addr - new->vm_start) >> PAGE_SHIFT), new); 26911da177e4SLinus Torvalds else 26925beb4930SRik van Riel err = vma_adjust(vma, vma->vm_start, addr, vma->vm_pgoff, new); 26931da177e4SLinus Torvalds 26945beb4930SRik van Riel /* Success. */ 26955beb4930SRik van Riel if (!err) 26961da177e4SLinus Torvalds return 0; 26975beb4930SRik van Riel 26985beb4930SRik van Riel /* Clean everything up if vma_adjust failed. */ 269958927533SRik van Riel if (new->vm_ops && new->vm_ops->close) 27005beb4930SRik van Riel new->vm_ops->close(new); 2701e9714acfSKonstantin Khlebnikov if (new->vm_file) 27025beb4930SRik van Riel fput(new->vm_file); 27032aeadc30SAndrea Arcangeli unlink_anon_vmas(new); 27045beb4930SRik van Riel out_free_mpol: 2705ef0855d3SOleg Nesterov mpol_put(vma_policy(new)); 27065beb4930SRik van Riel out_free_vma: 27073928d4f5SLinus Torvalds vm_area_free(new); 27085beb4930SRik van Riel return err; 27091da177e4SLinus Torvalds } 27101da177e4SLinus Torvalds 2711659ace58SKOSAKI Motohiro /* 2712659ace58SKOSAKI Motohiro * Split a vma into two pieces at address 'addr', a new vma is allocated 2713659ace58SKOSAKI Motohiro * either for the first part or the tail. 2714659ace58SKOSAKI Motohiro */ 2715659ace58SKOSAKI Motohiro int split_vma(struct mm_struct *mm, struct vm_area_struct *vma, 2716659ace58SKOSAKI Motohiro unsigned long addr, int new_below) 2717659ace58SKOSAKI Motohiro { 2718659ace58SKOSAKI Motohiro if (mm->map_count >= sysctl_max_map_count) 2719659ace58SKOSAKI Motohiro return -ENOMEM; 2720659ace58SKOSAKI Motohiro 2721659ace58SKOSAKI Motohiro return __split_vma(mm, vma, addr, new_below); 2722659ace58SKOSAKI Motohiro } 2723659ace58SKOSAKI Motohiro 27241da177e4SLinus Torvalds /* Munmap is split into 2 main parts -- this part which finds 27251da177e4SLinus Torvalds * what needs doing, and the areas themselves, which do the 27261da177e4SLinus Torvalds * work. This now handles partial unmappings. 27271da177e4SLinus Torvalds * Jeremy Fitzhardinge <jeremy@goop.org> 27281da177e4SLinus Torvalds */ 272985a06835SYang Shi int __do_munmap(struct mm_struct *mm, unsigned long start, size_t len, 2730dd2283f2SYang Shi struct list_head *uf, bool downgrade) 27311da177e4SLinus Torvalds { 27321da177e4SLinus Torvalds unsigned long end; 2733146425a3SHugh Dickins struct vm_area_struct *vma, *prev, *last; 27341da177e4SLinus Torvalds 2735de1741a1SAlexander Kuleshov if ((offset_in_page(start)) || start > TASK_SIZE || len > TASK_SIZE-start) 27361da177e4SLinus Torvalds return -EINVAL; 27371da177e4SLinus Torvalds 2738cc71aba3Svishnu.ps len = PAGE_ALIGN(len); 27395a28fc94SDave Hansen end = start + len; 2740cc71aba3Svishnu.ps if (len == 0) 27411da177e4SLinus Torvalds return -EINVAL; 27421da177e4SLinus Torvalds 27435a28fc94SDave Hansen /* 27445a28fc94SDave Hansen * arch_unmap() might do unmaps itself. It must be called 27455a28fc94SDave Hansen * and finish any rbtree manipulation before this code 27465a28fc94SDave Hansen * runs and also starts to manipulate the rbtree. 27475a28fc94SDave Hansen */ 27485a28fc94SDave Hansen arch_unmap(mm, start, end); 27495a28fc94SDave Hansen 27501da177e4SLinus Torvalds /* Find the first overlapping VMA */ 27519be34c9dSLinus Torvalds vma = find_vma(mm, start); 2752146425a3SHugh Dickins if (!vma) 27531da177e4SLinus Torvalds return 0; 27549be34c9dSLinus Torvalds prev = vma->vm_prev; 2755146425a3SHugh Dickins /* we have start < vma->vm_end */ 27561da177e4SLinus Torvalds 27571da177e4SLinus Torvalds /* if it doesn't overlap, we have nothing.. */ 2758146425a3SHugh Dickins if (vma->vm_start >= end) 27591da177e4SLinus Torvalds return 0; 27601da177e4SLinus Torvalds 27611da177e4SLinus Torvalds /* 27621da177e4SLinus Torvalds * If we need to split any vma, do it now to save pain later. 27631da177e4SLinus Torvalds * 27641da177e4SLinus Torvalds * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially 27651da177e4SLinus Torvalds * unmapped vm_area_struct will remain in use: so lower split_vma 27661da177e4SLinus Torvalds * places tmp vma above, and higher split_vma places tmp vma below. 27671da177e4SLinus Torvalds */ 2768146425a3SHugh Dickins if (start > vma->vm_start) { 2769659ace58SKOSAKI Motohiro int error; 2770659ace58SKOSAKI Motohiro 2771659ace58SKOSAKI Motohiro /* 2772659ace58SKOSAKI Motohiro * Make sure that map_count on return from munmap() will 2773659ace58SKOSAKI Motohiro * not exceed its limit; but let map_count go just above 2774659ace58SKOSAKI Motohiro * its limit temporarily, to help free resources as expected. 2775659ace58SKOSAKI Motohiro */ 2776659ace58SKOSAKI Motohiro if (end < vma->vm_end && mm->map_count >= sysctl_max_map_count) 2777659ace58SKOSAKI Motohiro return -ENOMEM; 2778659ace58SKOSAKI Motohiro 2779659ace58SKOSAKI Motohiro error = __split_vma(mm, vma, start, 0); 27801da177e4SLinus Torvalds if (error) 27811da177e4SLinus Torvalds return error; 2782146425a3SHugh Dickins prev = vma; 27831da177e4SLinus Torvalds } 27841da177e4SLinus Torvalds 27851da177e4SLinus Torvalds /* Does it split the last one? */ 27861da177e4SLinus Torvalds last = find_vma(mm, end); 27871da177e4SLinus Torvalds if (last && end > last->vm_start) { 2788659ace58SKOSAKI Motohiro int error = __split_vma(mm, last, end, 1); 27891da177e4SLinus Torvalds if (error) 27901da177e4SLinus Torvalds return error; 27911da177e4SLinus Torvalds } 2792146425a3SHugh Dickins vma = prev ? prev->vm_next : mm->mmap; 27931da177e4SLinus Torvalds 27942376dd7cSAndrea Arcangeli if (unlikely(uf)) { 27952376dd7cSAndrea Arcangeli /* 27962376dd7cSAndrea Arcangeli * If userfaultfd_unmap_prep returns an error the vmas 27972376dd7cSAndrea Arcangeli * will remain splitted, but userland will get a 27982376dd7cSAndrea Arcangeli * highly unexpected error anyway. This is no 27992376dd7cSAndrea Arcangeli * different than the case where the first of the two 28002376dd7cSAndrea Arcangeli * __split_vma fails, but we don't undo the first 28012376dd7cSAndrea Arcangeli * split, despite we could. This is unlikely enough 28022376dd7cSAndrea Arcangeli * failure that it's not worth optimizing it for. 28032376dd7cSAndrea Arcangeli */ 28042376dd7cSAndrea Arcangeli int error = userfaultfd_unmap_prep(vma, start, end, uf); 28052376dd7cSAndrea Arcangeli if (error) 28062376dd7cSAndrea Arcangeli return error; 28072376dd7cSAndrea Arcangeli } 28082376dd7cSAndrea Arcangeli 28091da177e4SLinus Torvalds /* 2810ba470de4SRik van Riel * unlock any mlock()ed ranges before detaching vmas 2811ba470de4SRik van Riel */ 2812ba470de4SRik van Riel if (mm->locked_vm) { 2813ba470de4SRik van Riel struct vm_area_struct *tmp = vma; 2814ba470de4SRik van Riel while (tmp && tmp->vm_start < end) { 2815ba470de4SRik van Riel if (tmp->vm_flags & VM_LOCKED) { 2816ba470de4SRik van Riel mm->locked_vm -= vma_pages(tmp); 2817ba470de4SRik van Riel munlock_vma_pages_all(tmp); 2818ba470de4SRik van Riel } 2819dd2283f2SYang Shi 2820ba470de4SRik van Riel tmp = tmp->vm_next; 2821ba470de4SRik van Riel } 2822ba470de4SRik van Riel } 2823ba470de4SRik van Riel 2824dd2283f2SYang Shi /* Detach vmas from rbtree */ 2825146425a3SHugh Dickins detach_vmas_to_be_unmapped(mm, vma, prev, end); 28261da177e4SLinus Torvalds 2827dd2283f2SYang Shi if (downgrade) 2828dd2283f2SYang Shi downgrade_write(&mm->mmap_sem); 2829dd2283f2SYang Shi 2830dd2283f2SYang Shi unmap_region(mm, vma, prev, start, end); 2831dd2283f2SYang Shi 28321da177e4SLinus Torvalds /* Fix up all other VM information */ 28332c0b3814SHugh Dickins remove_vma_list(mm, vma); 28341da177e4SLinus Torvalds 2835dd2283f2SYang Shi return downgrade ? 1 : 0; 28361da177e4SLinus Torvalds } 28371da177e4SLinus Torvalds 2838dd2283f2SYang Shi int do_munmap(struct mm_struct *mm, unsigned long start, size_t len, 2839dd2283f2SYang Shi struct list_head *uf) 2840dd2283f2SYang Shi { 2841dd2283f2SYang Shi return __do_munmap(mm, start, len, uf, false); 2842dd2283f2SYang Shi } 2843dd2283f2SYang Shi 2844dd2283f2SYang Shi static int __vm_munmap(unsigned long start, size_t len, bool downgrade) 2845a46ef99dSLinus Torvalds { 2846a46ef99dSLinus Torvalds int ret; 2847bfce281cSAl Viro struct mm_struct *mm = current->mm; 2848897ab3e0SMike Rapoport LIST_HEAD(uf); 2849a46ef99dSLinus Torvalds 2850ae798783SMichal Hocko if (down_write_killable(&mm->mmap_sem)) 2851ae798783SMichal Hocko return -EINTR; 2852ae798783SMichal Hocko 2853dd2283f2SYang Shi ret = __do_munmap(mm, start, len, &uf, downgrade); 2854dd2283f2SYang Shi /* 2855dd2283f2SYang Shi * Returning 1 indicates mmap_sem is downgraded. 2856dd2283f2SYang Shi * But 1 is not legal return value of vm_munmap() and munmap(), reset 2857dd2283f2SYang Shi * it to 0 before return. 2858dd2283f2SYang Shi */ 2859dd2283f2SYang Shi if (ret == 1) { 2860dd2283f2SYang Shi up_read(&mm->mmap_sem); 2861dd2283f2SYang Shi ret = 0; 2862dd2283f2SYang Shi } else 2863a46ef99dSLinus Torvalds up_write(&mm->mmap_sem); 2864dd2283f2SYang Shi 2865897ab3e0SMike Rapoport userfaultfd_unmap_complete(mm, &uf); 2866a46ef99dSLinus Torvalds return ret; 2867a46ef99dSLinus Torvalds } 2868dd2283f2SYang Shi 2869dd2283f2SYang Shi int vm_munmap(unsigned long start, size_t len) 2870dd2283f2SYang Shi { 2871dd2283f2SYang Shi return __vm_munmap(start, len, false); 2872dd2283f2SYang Shi } 2873a46ef99dSLinus Torvalds EXPORT_SYMBOL(vm_munmap); 2874a46ef99dSLinus Torvalds 28756a6160a7SHeiko Carstens SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len) 28761da177e4SLinus Torvalds { 28771da177e4SLinus Torvalds profile_munmap(addr); 2878dd2283f2SYang Shi return __vm_munmap(addr, len, true); 28791da177e4SLinus Torvalds } 28801da177e4SLinus Torvalds 2881c8d78c18SKirill A. Shutemov 2882c8d78c18SKirill A. Shutemov /* 2883c8d78c18SKirill A. Shutemov * Emulation of deprecated remap_file_pages() syscall. 2884c8d78c18SKirill A. Shutemov */ 2885c8d78c18SKirill A. Shutemov SYSCALL_DEFINE5(remap_file_pages, unsigned long, start, unsigned long, size, 2886c8d78c18SKirill A. Shutemov unsigned long, prot, unsigned long, pgoff, unsigned long, flags) 2887c8d78c18SKirill A. Shutemov { 2888c8d78c18SKirill A. Shutemov 2889c8d78c18SKirill A. Shutemov struct mm_struct *mm = current->mm; 2890c8d78c18SKirill A. Shutemov struct vm_area_struct *vma; 2891c8d78c18SKirill A. Shutemov unsigned long populate = 0; 2892c8d78c18SKirill A. Shutemov unsigned long ret = -EINVAL; 2893c8d78c18SKirill A. Shutemov struct file *file; 2894c8d78c18SKirill A. Shutemov 2895ad56b738SMike Rapoport pr_warn_once("%s (%d) uses deprecated remap_file_pages() syscall. See Documentation/vm/remap_file_pages.rst.\n", 2896c8d78c18SKirill A. Shutemov current->comm, current->pid); 2897c8d78c18SKirill A. Shutemov 2898c8d78c18SKirill A. Shutemov if (prot) 2899c8d78c18SKirill A. Shutemov return ret; 2900c8d78c18SKirill A. Shutemov start = start & PAGE_MASK; 2901c8d78c18SKirill A. Shutemov size = size & PAGE_MASK; 2902c8d78c18SKirill A. Shutemov 2903c8d78c18SKirill A. Shutemov if (start + size <= start) 2904c8d78c18SKirill A. Shutemov return ret; 2905c8d78c18SKirill A. Shutemov 2906c8d78c18SKirill A. Shutemov /* Does pgoff wrap? */ 2907c8d78c18SKirill A. Shutemov if (pgoff + (size >> PAGE_SHIFT) < pgoff) 2908c8d78c18SKirill A. Shutemov return ret; 2909c8d78c18SKirill A. Shutemov 2910dc0ef0dfSMichal Hocko if (down_write_killable(&mm->mmap_sem)) 2911dc0ef0dfSMichal Hocko return -EINTR; 2912dc0ef0dfSMichal Hocko 2913c8d78c18SKirill A. Shutemov vma = find_vma(mm, start); 2914c8d78c18SKirill A. Shutemov 2915c8d78c18SKirill A. Shutemov if (!vma || !(vma->vm_flags & VM_SHARED)) 2916c8d78c18SKirill A. Shutemov goto out; 2917c8d78c18SKirill A. Shutemov 291848f7df32SKirill A. Shutemov if (start < vma->vm_start) 2919c8d78c18SKirill A. Shutemov goto out; 2920c8d78c18SKirill A. Shutemov 292148f7df32SKirill A. Shutemov if (start + size > vma->vm_end) { 292248f7df32SKirill A. Shutemov struct vm_area_struct *next; 292348f7df32SKirill A. Shutemov 292448f7df32SKirill A. Shutemov for (next = vma->vm_next; next; next = next->vm_next) { 292548f7df32SKirill A. Shutemov /* hole between vmas ? */ 292648f7df32SKirill A. Shutemov if (next->vm_start != next->vm_prev->vm_end) 292748f7df32SKirill A. Shutemov goto out; 292848f7df32SKirill A. Shutemov 292948f7df32SKirill A. Shutemov if (next->vm_file != vma->vm_file) 293048f7df32SKirill A. Shutemov goto out; 293148f7df32SKirill A. Shutemov 293248f7df32SKirill A. Shutemov if (next->vm_flags != vma->vm_flags) 293348f7df32SKirill A. Shutemov goto out; 293448f7df32SKirill A. Shutemov 293548f7df32SKirill A. Shutemov if (start + size <= next->vm_end) 293648f7df32SKirill A. Shutemov break; 293748f7df32SKirill A. Shutemov } 293848f7df32SKirill A. Shutemov 293948f7df32SKirill A. Shutemov if (!next) 2940c8d78c18SKirill A. Shutemov goto out; 2941c8d78c18SKirill A. Shutemov } 2942c8d78c18SKirill A. Shutemov 2943c8d78c18SKirill A. Shutemov prot |= vma->vm_flags & VM_READ ? PROT_READ : 0; 2944c8d78c18SKirill A. Shutemov prot |= vma->vm_flags & VM_WRITE ? PROT_WRITE : 0; 2945c8d78c18SKirill A. Shutemov prot |= vma->vm_flags & VM_EXEC ? PROT_EXEC : 0; 2946c8d78c18SKirill A. Shutemov 2947c8d78c18SKirill A. Shutemov flags &= MAP_NONBLOCK; 2948c8d78c18SKirill A. Shutemov flags |= MAP_SHARED | MAP_FIXED | MAP_POPULATE; 2949c8d78c18SKirill A. Shutemov if (vma->vm_flags & VM_LOCKED) { 295048f7df32SKirill A. Shutemov struct vm_area_struct *tmp; 2951c8d78c18SKirill A. Shutemov flags |= MAP_LOCKED; 295248f7df32SKirill A. Shutemov 2953c8d78c18SKirill A. Shutemov /* drop PG_Mlocked flag for over-mapped range */ 295448f7df32SKirill A. Shutemov for (tmp = vma; tmp->vm_start >= start + size; 295548f7df32SKirill A. Shutemov tmp = tmp->vm_next) { 29569a73f61bSKirill A. Shutemov /* 29579a73f61bSKirill A. Shutemov * Split pmd and munlock page on the border 29589a73f61bSKirill A. Shutemov * of the range. 29599a73f61bSKirill A. Shutemov */ 29609a73f61bSKirill A. Shutemov vma_adjust_trans_huge(tmp, start, start + size, 0); 29619a73f61bSKirill A. Shutemov 296248f7df32SKirill A. Shutemov munlock_vma_pages_range(tmp, 296348f7df32SKirill A. Shutemov max(tmp->vm_start, start), 296448f7df32SKirill A. Shutemov min(tmp->vm_end, start + size)); 296548f7df32SKirill A. Shutemov } 2966c8d78c18SKirill A. Shutemov } 2967c8d78c18SKirill A. Shutemov 2968c8d78c18SKirill A. Shutemov file = get_file(vma->vm_file); 2969c8d78c18SKirill A. Shutemov ret = do_mmap_pgoff(vma->vm_file, start, size, 2970897ab3e0SMike Rapoport prot, flags, pgoff, &populate, NULL); 2971c8d78c18SKirill A. Shutemov fput(file); 2972c8d78c18SKirill A. Shutemov out: 2973c8d78c18SKirill A. Shutemov up_write(&mm->mmap_sem); 2974c8d78c18SKirill A. Shutemov if (populate) 2975c8d78c18SKirill A. Shutemov mm_populate(ret, populate); 2976c8d78c18SKirill A. Shutemov if (!IS_ERR_VALUE(ret)) 2977c8d78c18SKirill A. Shutemov ret = 0; 2978c8d78c18SKirill A. Shutemov return ret; 2979c8d78c18SKirill A. Shutemov } 2980c8d78c18SKirill A. Shutemov 29811da177e4SLinus Torvalds /* 29821da177e4SLinus Torvalds * this is really a simplified "do_mmap". it only handles 29831da177e4SLinus Torvalds * anonymous maps. eventually we may be able to do some 29841da177e4SLinus Torvalds * brk-specific accounting here. 29851da177e4SLinus Torvalds */ 2986bb177a73SMichal Hocko static int do_brk_flags(unsigned long addr, unsigned long len, unsigned long flags, struct list_head *uf) 29871da177e4SLinus Torvalds { 29881da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 29891da177e4SLinus Torvalds struct vm_area_struct *vma, *prev; 29901da177e4SLinus Torvalds struct rb_node **rb_link, *rb_parent; 29911da177e4SLinus Torvalds pgoff_t pgoff = addr >> PAGE_SHIFT; 29923a459756SKirill Korotaev int error; 29931da177e4SLinus Torvalds 299416e72e9bSDenys Vlasenko /* Until we need other flags, refuse anything except VM_EXEC. */ 299516e72e9bSDenys Vlasenko if ((flags & (~VM_EXEC)) != 0) 299616e72e9bSDenys Vlasenko return -EINVAL; 299716e72e9bSDenys Vlasenko flags |= VM_DATA_DEFAULT_FLAGS | VM_ACCOUNT | mm->def_flags; 29983a459756SKirill Korotaev 29992c6a1016SAl Viro error = get_unmapped_area(NULL, addr, len, 0, MAP_FIXED); 3000de1741a1SAlexander Kuleshov if (offset_in_page(error)) 30013a459756SKirill Korotaev return error; 30023a459756SKirill Korotaev 3003363ee17fSDavidlohr Bueso error = mlock_future_check(mm, mm->def_flags, len); 3004363ee17fSDavidlohr Bueso if (error) 3005363ee17fSDavidlohr Bueso return error; 30061da177e4SLinus Torvalds 30071da177e4SLinus Torvalds /* 30081da177e4SLinus Torvalds * Clear old maps. this also does some error checking for us 30091da177e4SLinus Torvalds */ 30109fcd1457SRasmus Villemoes while (find_vma_links(mm, addr, addr + len, &prev, &rb_link, 30119fcd1457SRasmus Villemoes &rb_parent)) { 3012897ab3e0SMike Rapoport if (do_munmap(mm, addr, len, uf)) 30131da177e4SLinus Torvalds return -ENOMEM; 30141da177e4SLinus Torvalds } 30151da177e4SLinus Torvalds 30161da177e4SLinus Torvalds /* Check against address space limits *after* clearing old maps... */ 301784638335SKonstantin Khlebnikov if (!may_expand_vm(mm, flags, len >> PAGE_SHIFT)) 30181da177e4SLinus Torvalds return -ENOMEM; 30191da177e4SLinus Torvalds 30201da177e4SLinus Torvalds if (mm->map_count > sysctl_max_map_count) 30211da177e4SLinus Torvalds return -ENOMEM; 30221da177e4SLinus Torvalds 3023191c5424SAl Viro if (security_vm_enough_memory_mm(mm, len >> PAGE_SHIFT)) 30241da177e4SLinus Torvalds return -ENOMEM; 30251da177e4SLinus Torvalds 30261da177e4SLinus Torvalds /* Can we just expand an old private anonymous mapping? */ 3027ba470de4SRik van Riel vma = vma_merge(mm, prev, addr, addr + len, flags, 302819a809afSAndrea Arcangeli NULL, NULL, pgoff, NULL, NULL_VM_UFFD_CTX); 3029ba470de4SRik van Riel if (vma) 30301da177e4SLinus Torvalds goto out; 30311da177e4SLinus Torvalds 30321da177e4SLinus Torvalds /* 30331da177e4SLinus Torvalds * create a vma struct for an anonymous mapping 30341da177e4SLinus Torvalds */ 3035490fc053SLinus Torvalds vma = vm_area_alloc(mm); 30361da177e4SLinus Torvalds if (!vma) { 30371da177e4SLinus Torvalds vm_unacct_memory(len >> PAGE_SHIFT); 30381da177e4SLinus Torvalds return -ENOMEM; 30391da177e4SLinus Torvalds } 30401da177e4SLinus Torvalds 3041bfd40eafSKirill A. Shutemov vma_set_anonymous(vma); 30421da177e4SLinus Torvalds vma->vm_start = addr; 30431da177e4SLinus Torvalds vma->vm_end = addr + len; 30441da177e4SLinus Torvalds vma->vm_pgoff = pgoff; 30451da177e4SLinus Torvalds vma->vm_flags = flags; 30463ed75eb8SColy Li vma->vm_page_prot = vm_get_page_prot(flags); 30471da177e4SLinus Torvalds vma_link(mm, vma, prev, rb_link, rb_parent); 30481da177e4SLinus Torvalds out: 30493af9e859SEric B Munson perf_event_mmap(vma); 30501da177e4SLinus Torvalds mm->total_vm += len >> PAGE_SHIFT; 305184638335SKonstantin Khlebnikov mm->data_vm += len >> PAGE_SHIFT; 3052128557ffSMichel Lespinasse if (flags & VM_LOCKED) 3053ba470de4SRik van Riel mm->locked_vm += (len >> PAGE_SHIFT); 3054d9104d1cSCyrill Gorcunov vma->vm_flags |= VM_SOFTDIRTY; 30555d22fc25SLinus Torvalds return 0; 30561da177e4SLinus Torvalds } 30571da177e4SLinus Torvalds 3058bb177a73SMichal Hocko int vm_brk_flags(unsigned long addr, unsigned long request, unsigned long flags) 3059e4eb1ff6SLinus Torvalds { 3060e4eb1ff6SLinus Torvalds struct mm_struct *mm = current->mm; 3061bb177a73SMichal Hocko unsigned long len; 30625d22fc25SLinus Torvalds int ret; 3063128557ffSMichel Lespinasse bool populate; 3064897ab3e0SMike Rapoport LIST_HEAD(uf); 3065e4eb1ff6SLinus Torvalds 3066bb177a73SMichal Hocko len = PAGE_ALIGN(request); 3067bb177a73SMichal Hocko if (len < request) 3068bb177a73SMichal Hocko return -ENOMEM; 3069bb177a73SMichal Hocko if (!len) 3070bb177a73SMichal Hocko return 0; 3071bb177a73SMichal Hocko 30722d6c9282SMichal Hocko if (down_write_killable(&mm->mmap_sem)) 30732d6c9282SMichal Hocko return -EINTR; 30742d6c9282SMichal Hocko 3075897ab3e0SMike Rapoport ret = do_brk_flags(addr, len, flags, &uf); 3076128557ffSMichel Lespinasse populate = ((mm->def_flags & VM_LOCKED) != 0); 3077e4eb1ff6SLinus Torvalds up_write(&mm->mmap_sem); 3078897ab3e0SMike Rapoport userfaultfd_unmap_complete(mm, &uf); 30795d22fc25SLinus Torvalds if (populate && !ret) 3080128557ffSMichel Lespinasse mm_populate(addr, len); 3081e4eb1ff6SLinus Torvalds return ret; 3082e4eb1ff6SLinus Torvalds } 308316e72e9bSDenys Vlasenko EXPORT_SYMBOL(vm_brk_flags); 308416e72e9bSDenys Vlasenko 308516e72e9bSDenys Vlasenko int vm_brk(unsigned long addr, unsigned long len) 308616e72e9bSDenys Vlasenko { 308716e72e9bSDenys Vlasenko return vm_brk_flags(addr, len, 0); 308816e72e9bSDenys Vlasenko } 3089e4eb1ff6SLinus Torvalds EXPORT_SYMBOL(vm_brk); 30901da177e4SLinus Torvalds 30911da177e4SLinus Torvalds /* Release all mmaps. */ 30921da177e4SLinus Torvalds void exit_mmap(struct mm_struct *mm) 30931da177e4SLinus Torvalds { 3094d16dfc55SPeter Zijlstra struct mmu_gather tlb; 3095ba470de4SRik van Riel struct vm_area_struct *vma; 30961da177e4SLinus Torvalds unsigned long nr_accounted = 0; 30971da177e4SLinus Torvalds 3098d6dd61c8SJeremy Fitzhardinge /* mm's last user has gone, and its about to be pulled down */ 3099cddb8a5cSAndrea Arcangeli mmu_notifier_release(mm); 3100d6dd61c8SJeremy Fitzhardinge 310127ae357fSDavid Rientjes if (unlikely(mm_is_oom_victim(mm))) { 310227ae357fSDavid Rientjes /* 310327ae357fSDavid Rientjes * Manually reap the mm to free as much memory as possible. 310427ae357fSDavid Rientjes * Then, as the oom reaper does, set MMF_OOM_SKIP to disregard 310527ae357fSDavid Rientjes * this mm from further consideration. Taking mm->mmap_sem for 310627ae357fSDavid Rientjes * write after setting MMF_OOM_SKIP will guarantee that the oom 310727ae357fSDavid Rientjes * reaper will not run on this mm again after mmap_sem is 310827ae357fSDavid Rientjes * dropped. 310927ae357fSDavid Rientjes * 311027ae357fSDavid Rientjes * Nothing can be holding mm->mmap_sem here and the above call 311127ae357fSDavid Rientjes * to mmu_notifier_release(mm) ensures mmu notifier callbacks in 311227ae357fSDavid Rientjes * __oom_reap_task_mm() will not block. 311327ae357fSDavid Rientjes * 311427ae357fSDavid Rientjes * This needs to be done before calling munlock_vma_pages_all(), 311527ae357fSDavid Rientjes * which clears VM_LOCKED, otherwise the oom reaper cannot 311627ae357fSDavid Rientjes * reliably test it. 311727ae357fSDavid Rientjes */ 311893065ac7SMichal Hocko (void)__oom_reap_task_mm(mm); 311927ae357fSDavid Rientjes 312027ae357fSDavid Rientjes set_bit(MMF_OOM_SKIP, &mm->flags); 312127ae357fSDavid Rientjes down_write(&mm->mmap_sem); 312227ae357fSDavid Rientjes up_write(&mm->mmap_sem); 312327ae357fSDavid Rientjes } 312427ae357fSDavid Rientjes 3125ba470de4SRik van Riel if (mm->locked_vm) { 3126ba470de4SRik van Riel vma = mm->mmap; 3127ba470de4SRik van Riel while (vma) { 3128ba470de4SRik van Riel if (vma->vm_flags & VM_LOCKED) 3129ba470de4SRik van Riel munlock_vma_pages_all(vma); 3130ba470de4SRik van Riel vma = vma->vm_next; 3131ba470de4SRik van Riel } 3132ba470de4SRik van Riel } 31339480c53eSJeremy Fitzhardinge 31349480c53eSJeremy Fitzhardinge arch_exit_mmap(mm); 31359480c53eSJeremy Fitzhardinge 3136ba470de4SRik van Riel vma = mm->mmap; 31379480c53eSJeremy Fitzhardinge if (!vma) /* Can happen if dup_mmap() received an OOM */ 31389480c53eSJeremy Fitzhardinge return; 31399480c53eSJeremy Fitzhardinge 31401da177e4SLinus Torvalds lru_add_drain(); 31411da177e4SLinus Torvalds flush_cache_mm(mm); 31422b047252SLinus Torvalds tlb_gather_mmu(&tlb, mm, 0, -1); 3143901608d9SOleg Nesterov /* update_hiwater_rss(mm) here? but nobody should be looking */ 3144e0da382cSHugh Dickins /* Use -1 here to ensure all VMAs in the mm are unmapped */ 31454f74d2c8SLinus Torvalds unmap_vmas(&tlb, vma, 0, -1); 31466ee8630eSHugh Dickins free_pgtables(&tlb, vma, FIRST_USER_ADDRESS, USER_PGTABLES_CEILING); 3147853f5e26SAl Viro tlb_finish_mmu(&tlb, 0, -1); 31481da177e4SLinus Torvalds 31491da177e4SLinus Torvalds /* 31508f4f8c16SHugh Dickins * Walk the list again, actually closing and freeing it, 31518f4f8c16SHugh Dickins * with preemption enabled, without holding any MM locks. 31521da177e4SLinus Torvalds */ 31534f74d2c8SLinus Torvalds while (vma) { 31544f74d2c8SLinus Torvalds if (vma->vm_flags & VM_ACCOUNT) 31554f74d2c8SLinus Torvalds nr_accounted += vma_pages(vma); 3156a8fb5618SHugh Dickins vma = remove_vma(vma); 31574f74d2c8SLinus Torvalds } 31584f74d2c8SLinus Torvalds vm_unacct_memory(nr_accounted); 31591da177e4SLinus Torvalds } 31601da177e4SLinus Torvalds 31611da177e4SLinus Torvalds /* Insert vm structure into process list sorted by address 31621da177e4SLinus Torvalds * and into the inode's i_mmap tree. If vm_file is non-NULL 3163c8c06efaSDavidlohr Bueso * then i_mmap_rwsem is taken here. 31641da177e4SLinus Torvalds */ 31651da177e4SLinus Torvalds int insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma) 31661da177e4SLinus Torvalds { 31676597d783SHugh Dickins struct vm_area_struct *prev; 31681da177e4SLinus Torvalds struct rb_node **rb_link, *rb_parent; 31691da177e4SLinus Torvalds 3170c9d13f5fSChen Gang if (find_vma_links(mm, vma->vm_start, vma->vm_end, 3171c9d13f5fSChen Gang &prev, &rb_link, &rb_parent)) 3172c9d13f5fSChen Gang return -ENOMEM; 3173c9d13f5fSChen Gang if ((vma->vm_flags & VM_ACCOUNT) && 3174c9d13f5fSChen Gang security_vm_enough_memory_mm(mm, vma_pages(vma))) 3175c9d13f5fSChen Gang return -ENOMEM; 3176c9d13f5fSChen Gang 31771da177e4SLinus Torvalds /* 31781da177e4SLinus Torvalds * The vm_pgoff of a purely anonymous vma should be irrelevant 31791da177e4SLinus Torvalds * until its first write fault, when page's anon_vma and index 31801da177e4SLinus Torvalds * are set. But now set the vm_pgoff it will almost certainly 31811da177e4SLinus Torvalds * end up with (unless mremap moves it elsewhere before that 31821da177e4SLinus Torvalds * first wfault), so /proc/pid/maps tells a consistent story. 31831da177e4SLinus Torvalds * 31841da177e4SLinus Torvalds * By setting it to reflect the virtual start address of the 31851da177e4SLinus Torvalds * vma, merges and splits can happen in a seamless way, just 31861da177e4SLinus Torvalds * using the existing file pgoff checks and manipulations. 31871da177e4SLinus Torvalds * Similarly in do_mmap_pgoff and in do_brk. 31881da177e4SLinus Torvalds */ 31898a9cc3b5SOleg Nesterov if (vma_is_anonymous(vma)) { 31901da177e4SLinus Torvalds BUG_ON(vma->anon_vma); 31911da177e4SLinus Torvalds vma->vm_pgoff = vma->vm_start >> PAGE_SHIFT; 31921da177e4SLinus Torvalds } 31932b144498SSrikar Dronamraju 31941da177e4SLinus Torvalds vma_link(mm, vma, prev, rb_link, rb_parent); 31951da177e4SLinus Torvalds return 0; 31961da177e4SLinus Torvalds } 31971da177e4SLinus Torvalds 31981da177e4SLinus Torvalds /* 31991da177e4SLinus Torvalds * Copy the vma structure to a new location in the same mm, 32001da177e4SLinus Torvalds * prior to moving page table entries, to effect an mremap move. 32011da177e4SLinus Torvalds */ 32021da177e4SLinus Torvalds struct vm_area_struct *copy_vma(struct vm_area_struct **vmap, 320338a76013SMichel Lespinasse unsigned long addr, unsigned long len, pgoff_t pgoff, 320438a76013SMichel Lespinasse bool *need_rmap_locks) 32051da177e4SLinus Torvalds { 32061da177e4SLinus Torvalds struct vm_area_struct *vma = *vmap; 32071da177e4SLinus Torvalds unsigned long vma_start = vma->vm_start; 32081da177e4SLinus Torvalds struct mm_struct *mm = vma->vm_mm; 32091da177e4SLinus Torvalds struct vm_area_struct *new_vma, *prev; 32101da177e4SLinus Torvalds struct rb_node **rb_link, *rb_parent; 3211948f017bSAndrea Arcangeli bool faulted_in_anon_vma = true; 32121da177e4SLinus Torvalds 32131da177e4SLinus Torvalds /* 32141da177e4SLinus Torvalds * If anonymous vma has not yet been faulted, update new pgoff 32151da177e4SLinus Torvalds * to match new location, to increase its chance of merging. 32161da177e4SLinus Torvalds */ 3217ce75799bSOleg Nesterov if (unlikely(vma_is_anonymous(vma) && !vma->anon_vma)) { 32181da177e4SLinus Torvalds pgoff = addr >> PAGE_SHIFT; 3219948f017bSAndrea Arcangeli faulted_in_anon_vma = false; 3220948f017bSAndrea Arcangeli } 32211da177e4SLinus Torvalds 32226597d783SHugh Dickins if (find_vma_links(mm, addr, addr + len, &prev, &rb_link, &rb_parent)) 32236597d783SHugh Dickins return NULL; /* should never get here */ 32241da177e4SLinus Torvalds new_vma = vma_merge(mm, prev, addr, addr + len, vma->vm_flags, 322519a809afSAndrea Arcangeli vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma), 322619a809afSAndrea Arcangeli vma->vm_userfaultfd_ctx); 32271da177e4SLinus Torvalds if (new_vma) { 32281da177e4SLinus Torvalds /* 32291da177e4SLinus Torvalds * Source vma may have been merged into new_vma 32301da177e4SLinus Torvalds */ 3231948f017bSAndrea Arcangeli if (unlikely(vma_start >= new_vma->vm_start && 3232948f017bSAndrea Arcangeli vma_start < new_vma->vm_end)) { 3233948f017bSAndrea Arcangeli /* 3234948f017bSAndrea Arcangeli * The only way we can get a vma_merge with 3235948f017bSAndrea Arcangeli * self during an mremap is if the vma hasn't 3236948f017bSAndrea Arcangeli * been faulted in yet and we were allowed to 3237948f017bSAndrea Arcangeli * reset the dst vma->vm_pgoff to the 3238948f017bSAndrea Arcangeli * destination address of the mremap to allow 3239948f017bSAndrea Arcangeli * the merge to happen. mremap must change the 3240948f017bSAndrea Arcangeli * vm_pgoff linearity between src and dst vmas 3241948f017bSAndrea Arcangeli * (in turn preventing a vma_merge) to be 3242948f017bSAndrea Arcangeli * safe. It is only safe to keep the vm_pgoff 3243948f017bSAndrea Arcangeli * linear if there are no pages mapped yet. 3244948f017bSAndrea Arcangeli */ 324581d1b09cSSasha Levin VM_BUG_ON_VMA(faulted_in_anon_vma, new_vma); 324638a76013SMichel Lespinasse *vmap = vma = new_vma; 3247108d6642SMichel Lespinasse } 324838a76013SMichel Lespinasse *need_rmap_locks = (new_vma->vm_pgoff <= vma->vm_pgoff); 32491da177e4SLinus Torvalds } else { 32503928d4f5SLinus Torvalds new_vma = vm_area_dup(vma); 3251e3975891SChen Gang if (!new_vma) 3252e3975891SChen Gang goto out; 32531da177e4SLinus Torvalds new_vma->vm_start = addr; 32541da177e4SLinus Torvalds new_vma->vm_end = addr + len; 32551da177e4SLinus Torvalds new_vma->vm_pgoff = pgoff; 3256ef0855d3SOleg Nesterov if (vma_dup_policy(vma, new_vma)) 3257523d4e20SMichel Lespinasse goto out_free_vma; 3258523d4e20SMichel Lespinasse if (anon_vma_clone(new_vma, vma)) 3259523d4e20SMichel Lespinasse goto out_free_mempol; 3260e9714acfSKonstantin Khlebnikov if (new_vma->vm_file) 32611da177e4SLinus Torvalds get_file(new_vma->vm_file); 32621da177e4SLinus Torvalds if (new_vma->vm_ops && new_vma->vm_ops->open) 32631da177e4SLinus Torvalds new_vma->vm_ops->open(new_vma); 32641da177e4SLinus Torvalds vma_link(mm, new_vma, prev, rb_link, rb_parent); 326538a76013SMichel Lespinasse *need_rmap_locks = false; 32661da177e4SLinus Torvalds } 32671da177e4SLinus Torvalds return new_vma; 32685beb4930SRik van Riel 32695beb4930SRik van Riel out_free_mempol: 3270ef0855d3SOleg Nesterov mpol_put(vma_policy(new_vma)); 32715beb4930SRik van Riel out_free_vma: 32723928d4f5SLinus Torvalds vm_area_free(new_vma); 3273e3975891SChen Gang out: 32745beb4930SRik van Riel return NULL; 32751da177e4SLinus Torvalds } 3276119f657cSakpm@osdl.org 3277119f657cSakpm@osdl.org /* 3278119f657cSakpm@osdl.org * Return true if the calling process may expand its vm space by the passed 3279119f657cSakpm@osdl.org * number of pages 3280119f657cSakpm@osdl.org */ 328184638335SKonstantin Khlebnikov bool may_expand_vm(struct mm_struct *mm, vm_flags_t flags, unsigned long npages) 3282119f657cSakpm@osdl.org { 328384638335SKonstantin Khlebnikov if (mm->total_vm + npages > rlimit(RLIMIT_AS) >> PAGE_SHIFT) 328484638335SKonstantin Khlebnikov return false; 3285119f657cSakpm@osdl.org 3286d977d56cSKonstantin Khlebnikov if (is_data_mapping(flags) && 3287d977d56cSKonstantin Khlebnikov mm->data_vm + npages > rlimit(RLIMIT_DATA) >> PAGE_SHIFT) { 3288f4fcd558SKonstantin Khlebnikov /* Workaround for Valgrind */ 3289f4fcd558SKonstantin Khlebnikov if (rlimit(RLIMIT_DATA) == 0 && 3290f4fcd558SKonstantin Khlebnikov mm->data_vm + npages <= rlimit_max(RLIMIT_DATA) >> PAGE_SHIFT) 3291f4fcd558SKonstantin Khlebnikov return true; 329257a7702bSDavid Woodhouse 329357a7702bSDavid Woodhouse pr_warn_once("%s (%d): VmData %lu exceed data ulimit %lu. Update limits%s.\n", 3294d977d56cSKonstantin Khlebnikov current->comm, current->pid, 3295d977d56cSKonstantin Khlebnikov (mm->data_vm + npages) << PAGE_SHIFT, 329657a7702bSDavid Woodhouse rlimit(RLIMIT_DATA), 329757a7702bSDavid Woodhouse ignore_rlimit_data ? "" : " or use boot option ignore_rlimit_data"); 329857a7702bSDavid Woodhouse 329957a7702bSDavid Woodhouse if (!ignore_rlimit_data) 3300d977d56cSKonstantin Khlebnikov return false; 3301d977d56cSKonstantin Khlebnikov } 3302119f657cSakpm@osdl.org 330384638335SKonstantin Khlebnikov return true; 330484638335SKonstantin Khlebnikov } 330584638335SKonstantin Khlebnikov 330684638335SKonstantin Khlebnikov void vm_stat_account(struct mm_struct *mm, vm_flags_t flags, long npages) 330784638335SKonstantin Khlebnikov { 330884638335SKonstantin Khlebnikov mm->total_vm += npages; 330984638335SKonstantin Khlebnikov 3310d977d56cSKonstantin Khlebnikov if (is_exec_mapping(flags)) 331184638335SKonstantin Khlebnikov mm->exec_vm += npages; 3312d977d56cSKonstantin Khlebnikov else if (is_stack_mapping(flags)) 331384638335SKonstantin Khlebnikov mm->stack_vm += npages; 3314d977d56cSKonstantin Khlebnikov else if (is_data_mapping(flags)) 331584638335SKonstantin Khlebnikov mm->data_vm += npages; 3316119f657cSakpm@osdl.org } 3317fa5dc22fSRoland McGrath 3318b3ec9f33SSouptick Joarder static vm_fault_t special_mapping_fault(struct vm_fault *vmf); 3319a62c34bdSAndy Lutomirski 3320a62c34bdSAndy Lutomirski /* 3321a62c34bdSAndy Lutomirski * Having a close hook prevents vma merging regardless of flags. 3322a62c34bdSAndy Lutomirski */ 3323a62c34bdSAndy Lutomirski static void special_mapping_close(struct vm_area_struct *vma) 3324a62c34bdSAndy Lutomirski { 3325a62c34bdSAndy Lutomirski } 3326a62c34bdSAndy Lutomirski 3327a62c34bdSAndy Lutomirski static const char *special_mapping_name(struct vm_area_struct *vma) 3328a62c34bdSAndy Lutomirski { 3329a62c34bdSAndy Lutomirski return ((struct vm_special_mapping *)vma->vm_private_data)->name; 3330a62c34bdSAndy Lutomirski } 3331a62c34bdSAndy Lutomirski 3332b059a453SDmitry Safonov static int special_mapping_mremap(struct vm_area_struct *new_vma) 3333b059a453SDmitry Safonov { 3334b059a453SDmitry Safonov struct vm_special_mapping *sm = new_vma->vm_private_data; 3335b059a453SDmitry Safonov 3336280e87e9SDmitry Safonov if (WARN_ON_ONCE(current->mm != new_vma->vm_mm)) 3337280e87e9SDmitry Safonov return -EFAULT; 3338280e87e9SDmitry Safonov 3339b059a453SDmitry Safonov if (sm->mremap) 3340b059a453SDmitry Safonov return sm->mremap(sm, new_vma); 3341280e87e9SDmitry Safonov 3342b059a453SDmitry Safonov return 0; 3343b059a453SDmitry Safonov } 3344b059a453SDmitry Safonov 3345a62c34bdSAndy Lutomirski static const struct vm_operations_struct special_mapping_vmops = { 3346a62c34bdSAndy Lutomirski .close = special_mapping_close, 3347a62c34bdSAndy Lutomirski .fault = special_mapping_fault, 3348b059a453SDmitry Safonov .mremap = special_mapping_mremap, 3349a62c34bdSAndy Lutomirski .name = special_mapping_name, 3350a62c34bdSAndy Lutomirski }; 3351a62c34bdSAndy Lutomirski 3352a62c34bdSAndy Lutomirski static const struct vm_operations_struct legacy_special_mapping_vmops = { 3353a62c34bdSAndy Lutomirski .close = special_mapping_close, 3354a62c34bdSAndy Lutomirski .fault = special_mapping_fault, 3355a62c34bdSAndy Lutomirski }; 3356fa5dc22fSRoland McGrath 3357b3ec9f33SSouptick Joarder static vm_fault_t special_mapping_fault(struct vm_fault *vmf) 3358fa5dc22fSRoland McGrath { 335911bac800SDave Jiang struct vm_area_struct *vma = vmf->vma; 3360b1d0e4f5SNick Piggin pgoff_t pgoff; 3361fa5dc22fSRoland McGrath struct page **pages; 3362fa5dc22fSRoland McGrath 3363f872f540SAndy Lutomirski if (vma->vm_ops == &legacy_special_mapping_vmops) { 3364a62c34bdSAndy Lutomirski pages = vma->vm_private_data; 3365f872f540SAndy Lutomirski } else { 3366f872f540SAndy Lutomirski struct vm_special_mapping *sm = vma->vm_private_data; 3367f872f540SAndy Lutomirski 3368f872f540SAndy Lutomirski if (sm->fault) 336911bac800SDave Jiang return sm->fault(sm, vmf->vma, vmf); 3370f872f540SAndy Lutomirski 3371f872f540SAndy Lutomirski pages = sm->pages; 3372f872f540SAndy Lutomirski } 3373a62c34bdSAndy Lutomirski 33748a9cc3b5SOleg Nesterov for (pgoff = vmf->pgoff; pgoff && *pages; ++pages) 3375b1d0e4f5SNick Piggin pgoff--; 3376fa5dc22fSRoland McGrath 3377fa5dc22fSRoland McGrath if (*pages) { 3378fa5dc22fSRoland McGrath struct page *page = *pages; 3379fa5dc22fSRoland McGrath get_page(page); 3380b1d0e4f5SNick Piggin vmf->page = page; 3381b1d0e4f5SNick Piggin return 0; 3382fa5dc22fSRoland McGrath } 3383fa5dc22fSRoland McGrath 3384b1d0e4f5SNick Piggin return VM_FAULT_SIGBUS; 3385fa5dc22fSRoland McGrath } 3386fa5dc22fSRoland McGrath 3387a62c34bdSAndy Lutomirski static struct vm_area_struct *__install_special_mapping( 3388a62c34bdSAndy Lutomirski struct mm_struct *mm, 3389fa5dc22fSRoland McGrath unsigned long addr, unsigned long len, 339027f28b97SChen Gang unsigned long vm_flags, void *priv, 339127f28b97SChen Gang const struct vm_operations_struct *ops) 3392fa5dc22fSRoland McGrath { 3393462e635eSTavis Ormandy int ret; 3394fa5dc22fSRoland McGrath struct vm_area_struct *vma; 3395fa5dc22fSRoland McGrath 3396490fc053SLinus Torvalds vma = vm_area_alloc(mm); 3397fa5dc22fSRoland McGrath if (unlikely(vma == NULL)) 33983935ed6aSStefani Seibold return ERR_PTR(-ENOMEM); 3399fa5dc22fSRoland McGrath 3400fa5dc22fSRoland McGrath vma->vm_start = addr; 3401fa5dc22fSRoland McGrath vma->vm_end = addr + len; 3402fa5dc22fSRoland McGrath 3403d9104d1cSCyrill Gorcunov vma->vm_flags = vm_flags | mm->def_flags | VM_DONTEXPAND | VM_SOFTDIRTY; 34043ed75eb8SColy Li vma->vm_page_prot = vm_get_page_prot(vma->vm_flags); 3405fa5dc22fSRoland McGrath 3406a62c34bdSAndy Lutomirski vma->vm_ops = ops; 3407a62c34bdSAndy Lutomirski vma->vm_private_data = priv; 3408fa5dc22fSRoland McGrath 3409462e635eSTavis Ormandy ret = insert_vm_struct(mm, vma); 3410462e635eSTavis Ormandy if (ret) 3411462e635eSTavis Ormandy goto out; 3412fa5dc22fSRoland McGrath 341384638335SKonstantin Khlebnikov vm_stat_account(mm, vma->vm_flags, len >> PAGE_SHIFT); 3414fa5dc22fSRoland McGrath 3415cdd6c482SIngo Molnar perf_event_mmap(vma); 3416089dd79dSPeter Zijlstra 34173935ed6aSStefani Seibold return vma; 3418462e635eSTavis Ormandy 3419462e635eSTavis Ormandy out: 34203928d4f5SLinus Torvalds vm_area_free(vma); 34213935ed6aSStefani Seibold return ERR_PTR(ret); 34223935ed6aSStefani Seibold } 34233935ed6aSStefani Seibold 34242eefd878SDmitry Safonov bool vma_is_special_mapping(const struct vm_area_struct *vma, 34252eefd878SDmitry Safonov const struct vm_special_mapping *sm) 34262eefd878SDmitry Safonov { 34272eefd878SDmitry Safonov return vma->vm_private_data == sm && 34282eefd878SDmitry Safonov (vma->vm_ops == &special_mapping_vmops || 34292eefd878SDmitry Safonov vma->vm_ops == &legacy_special_mapping_vmops); 34302eefd878SDmitry Safonov } 34312eefd878SDmitry Safonov 3432a62c34bdSAndy Lutomirski /* 3433a62c34bdSAndy Lutomirski * Called with mm->mmap_sem held for writing. 3434a62c34bdSAndy Lutomirski * Insert a new vma covering the given region, with the given flags. 3435a62c34bdSAndy Lutomirski * Its pages are supplied by the given array of struct page *. 3436a62c34bdSAndy Lutomirski * The array can be shorter than len >> PAGE_SHIFT if it's null-terminated. 3437a62c34bdSAndy Lutomirski * The region past the last page supplied will always produce SIGBUS. 3438a62c34bdSAndy Lutomirski * The array pointer and the pages it points to are assumed to stay alive 3439a62c34bdSAndy Lutomirski * for as long as this mapping might exist. 3440a62c34bdSAndy Lutomirski */ 3441a62c34bdSAndy Lutomirski struct vm_area_struct *_install_special_mapping( 3442a62c34bdSAndy Lutomirski struct mm_struct *mm, 3443a62c34bdSAndy Lutomirski unsigned long addr, unsigned long len, 3444a62c34bdSAndy Lutomirski unsigned long vm_flags, const struct vm_special_mapping *spec) 3445a62c34bdSAndy Lutomirski { 344627f28b97SChen Gang return __install_special_mapping(mm, addr, len, vm_flags, (void *)spec, 344727f28b97SChen Gang &special_mapping_vmops); 3448a62c34bdSAndy Lutomirski } 3449a62c34bdSAndy Lutomirski 34503935ed6aSStefani Seibold int install_special_mapping(struct mm_struct *mm, 34513935ed6aSStefani Seibold unsigned long addr, unsigned long len, 34523935ed6aSStefani Seibold unsigned long vm_flags, struct page **pages) 34533935ed6aSStefani Seibold { 3454a62c34bdSAndy Lutomirski struct vm_area_struct *vma = __install_special_mapping( 345527f28b97SChen Gang mm, addr, len, vm_flags, (void *)pages, 345627f28b97SChen Gang &legacy_special_mapping_vmops); 34573935ed6aSStefani Seibold 345814bd5b45SDuan Jiong return PTR_ERR_OR_ZERO(vma); 3459fa5dc22fSRoland McGrath } 34607906d00cSAndrea Arcangeli 34617906d00cSAndrea Arcangeli static DEFINE_MUTEX(mm_all_locks_mutex); 34627906d00cSAndrea Arcangeli 3463454ed842SPeter Zijlstra static void vm_lock_anon_vma(struct mm_struct *mm, struct anon_vma *anon_vma) 34647906d00cSAndrea Arcangeli { 3465f808c13fSDavidlohr Bueso if (!test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_root.rb_node)) { 34667906d00cSAndrea Arcangeli /* 34677906d00cSAndrea Arcangeli * The LSB of head.next can't change from under us 34687906d00cSAndrea Arcangeli * because we hold the mm_all_locks_mutex. 34697906d00cSAndrea Arcangeli */ 3470572043c9SJiri Kosina down_write_nest_lock(&anon_vma->root->rwsem, &mm->mmap_sem); 34717906d00cSAndrea Arcangeli /* 34727906d00cSAndrea Arcangeli * We can safely modify head.next after taking the 34735a505085SIngo Molnar * anon_vma->root->rwsem. If some other vma in this mm shares 34747906d00cSAndrea Arcangeli * the same anon_vma we won't take it again. 34757906d00cSAndrea Arcangeli * 34767906d00cSAndrea Arcangeli * No need of atomic instructions here, head.next 34777906d00cSAndrea Arcangeli * can't change from under us thanks to the 34785a505085SIngo Molnar * anon_vma->root->rwsem. 34797906d00cSAndrea Arcangeli */ 34807906d00cSAndrea Arcangeli if (__test_and_set_bit(0, (unsigned long *) 3481f808c13fSDavidlohr Bueso &anon_vma->root->rb_root.rb_root.rb_node)) 34827906d00cSAndrea Arcangeli BUG(); 34837906d00cSAndrea Arcangeli } 34847906d00cSAndrea Arcangeli } 34857906d00cSAndrea Arcangeli 3486454ed842SPeter Zijlstra static void vm_lock_mapping(struct mm_struct *mm, struct address_space *mapping) 34877906d00cSAndrea Arcangeli { 34887906d00cSAndrea Arcangeli if (!test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) { 34897906d00cSAndrea Arcangeli /* 34907906d00cSAndrea Arcangeli * AS_MM_ALL_LOCKS can't change from under us because 34917906d00cSAndrea Arcangeli * we hold the mm_all_locks_mutex. 34927906d00cSAndrea Arcangeli * 34937906d00cSAndrea Arcangeli * Operations on ->flags have to be atomic because 34947906d00cSAndrea Arcangeli * even if AS_MM_ALL_LOCKS is stable thanks to the 34957906d00cSAndrea Arcangeli * mm_all_locks_mutex, there may be other cpus 34967906d00cSAndrea Arcangeli * changing other bitflags in parallel to us. 34977906d00cSAndrea Arcangeli */ 34987906d00cSAndrea Arcangeli if (test_and_set_bit(AS_MM_ALL_LOCKS, &mapping->flags)) 34997906d00cSAndrea Arcangeli BUG(); 3500c8c06efaSDavidlohr Bueso down_write_nest_lock(&mapping->i_mmap_rwsem, &mm->mmap_sem); 35017906d00cSAndrea Arcangeli } 35027906d00cSAndrea Arcangeli } 35037906d00cSAndrea Arcangeli 35047906d00cSAndrea Arcangeli /* 35057906d00cSAndrea Arcangeli * This operation locks against the VM for all pte/vma/mm related 35067906d00cSAndrea Arcangeli * operations that could ever happen on a certain mm. This includes 35077906d00cSAndrea Arcangeli * vmtruncate, try_to_unmap, and all page faults. 35087906d00cSAndrea Arcangeli * 35097906d00cSAndrea Arcangeli * The caller must take the mmap_sem in write mode before calling 35107906d00cSAndrea Arcangeli * mm_take_all_locks(). The caller isn't allowed to release the 35117906d00cSAndrea Arcangeli * mmap_sem until mm_drop_all_locks() returns. 35127906d00cSAndrea Arcangeli * 35137906d00cSAndrea Arcangeli * mmap_sem in write mode is required in order to block all operations 35147906d00cSAndrea Arcangeli * that could modify pagetables and free pages without need of 351527ba0644SKirill A. Shutemov * altering the vma layout. It's also needed in write mode to avoid new 35167906d00cSAndrea Arcangeli * anon_vmas to be associated with existing vmas. 35177906d00cSAndrea Arcangeli * 35187906d00cSAndrea Arcangeli * A single task can't take more than one mm_take_all_locks() in a row 35197906d00cSAndrea Arcangeli * or it would deadlock. 35207906d00cSAndrea Arcangeli * 3521bf181b9fSMichel Lespinasse * The LSB in anon_vma->rb_root.rb_node and the AS_MM_ALL_LOCKS bitflag in 35227906d00cSAndrea Arcangeli * mapping->flags avoid to take the same lock twice, if more than one 35237906d00cSAndrea Arcangeli * vma in this mm is backed by the same anon_vma or address_space. 35247906d00cSAndrea Arcangeli * 352588f306b6SKirill A. Shutemov * We take locks in following order, accordingly to comment at beginning 352688f306b6SKirill A. Shutemov * of mm/rmap.c: 352788f306b6SKirill A. Shutemov * - all hugetlbfs_i_mmap_rwsem_key locks (aka mapping->i_mmap_rwsem for 352888f306b6SKirill A. Shutemov * hugetlb mapping); 352988f306b6SKirill A. Shutemov * - all i_mmap_rwsem locks; 353088f306b6SKirill A. Shutemov * - all anon_vma->rwseml 353188f306b6SKirill A. Shutemov * 353288f306b6SKirill A. Shutemov * We can take all locks within these types randomly because the VM code 353388f306b6SKirill A. Shutemov * doesn't nest them and we protected from parallel mm_take_all_locks() by 353488f306b6SKirill A. Shutemov * mm_all_locks_mutex. 35357906d00cSAndrea Arcangeli * 35367906d00cSAndrea Arcangeli * mm_take_all_locks() and mm_drop_all_locks are expensive operations 35377906d00cSAndrea Arcangeli * that may have to take thousand of locks. 35387906d00cSAndrea Arcangeli * 35397906d00cSAndrea Arcangeli * mm_take_all_locks() can fail if it's interrupted by signals. 35407906d00cSAndrea Arcangeli */ 35417906d00cSAndrea Arcangeli int mm_take_all_locks(struct mm_struct *mm) 35427906d00cSAndrea Arcangeli { 35437906d00cSAndrea Arcangeli struct vm_area_struct *vma; 35445beb4930SRik van Riel struct anon_vma_chain *avc; 35457906d00cSAndrea Arcangeli 35467906d00cSAndrea Arcangeli BUG_ON(down_read_trylock(&mm->mmap_sem)); 35477906d00cSAndrea Arcangeli 35487906d00cSAndrea Arcangeli mutex_lock(&mm_all_locks_mutex); 35497906d00cSAndrea Arcangeli 35507906d00cSAndrea Arcangeli for (vma = mm->mmap; vma; vma = vma->vm_next) { 35517906d00cSAndrea Arcangeli if (signal_pending(current)) 35527906d00cSAndrea Arcangeli goto out_unlock; 355388f306b6SKirill A. Shutemov if (vma->vm_file && vma->vm_file->f_mapping && 355488f306b6SKirill A. Shutemov is_vm_hugetlb_page(vma)) 355588f306b6SKirill A. Shutemov vm_lock_mapping(mm, vma->vm_file->f_mapping); 355688f306b6SKirill A. Shutemov } 355788f306b6SKirill A. Shutemov 355888f306b6SKirill A. Shutemov for (vma = mm->mmap; vma; vma = vma->vm_next) { 355988f306b6SKirill A. Shutemov if (signal_pending(current)) 356088f306b6SKirill A. Shutemov goto out_unlock; 356188f306b6SKirill A. Shutemov if (vma->vm_file && vma->vm_file->f_mapping && 356288f306b6SKirill A. Shutemov !is_vm_hugetlb_page(vma)) 3563454ed842SPeter Zijlstra vm_lock_mapping(mm, vma->vm_file->f_mapping); 35647906d00cSAndrea Arcangeli } 35657cd5a02fSPeter Zijlstra 35667cd5a02fSPeter Zijlstra for (vma = mm->mmap; vma; vma = vma->vm_next) { 35677cd5a02fSPeter Zijlstra if (signal_pending(current)) 35687cd5a02fSPeter Zijlstra goto out_unlock; 35697cd5a02fSPeter Zijlstra if (vma->anon_vma) 35705beb4930SRik van Riel list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) 35715beb4930SRik van Riel vm_lock_anon_vma(mm, avc->anon_vma); 35727cd5a02fSPeter Zijlstra } 35737cd5a02fSPeter Zijlstra 3574584cff54SKautuk Consul return 0; 35757906d00cSAndrea Arcangeli 35767906d00cSAndrea Arcangeli out_unlock: 35777906d00cSAndrea Arcangeli mm_drop_all_locks(mm); 3578584cff54SKautuk Consul return -EINTR; 35797906d00cSAndrea Arcangeli } 35807906d00cSAndrea Arcangeli 35817906d00cSAndrea Arcangeli static void vm_unlock_anon_vma(struct anon_vma *anon_vma) 35827906d00cSAndrea Arcangeli { 3583f808c13fSDavidlohr Bueso if (test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_root.rb_node)) { 35847906d00cSAndrea Arcangeli /* 35857906d00cSAndrea Arcangeli * The LSB of head.next can't change to 0 from under 35867906d00cSAndrea Arcangeli * us because we hold the mm_all_locks_mutex. 35877906d00cSAndrea Arcangeli * 35887906d00cSAndrea Arcangeli * We must however clear the bitflag before unlocking 3589bf181b9fSMichel Lespinasse * the vma so the users using the anon_vma->rb_root will 35907906d00cSAndrea Arcangeli * never see our bitflag. 35917906d00cSAndrea Arcangeli * 35927906d00cSAndrea Arcangeli * No need of atomic instructions here, head.next 35937906d00cSAndrea Arcangeli * can't change from under us until we release the 35945a505085SIngo Molnar * anon_vma->root->rwsem. 35957906d00cSAndrea Arcangeli */ 35967906d00cSAndrea Arcangeli if (!__test_and_clear_bit(0, (unsigned long *) 3597f808c13fSDavidlohr Bueso &anon_vma->root->rb_root.rb_root.rb_node)) 35987906d00cSAndrea Arcangeli BUG(); 359908b52706SKonstantin Khlebnikov anon_vma_unlock_write(anon_vma); 36007906d00cSAndrea Arcangeli } 36017906d00cSAndrea Arcangeli } 36027906d00cSAndrea Arcangeli 36037906d00cSAndrea Arcangeli static void vm_unlock_mapping(struct address_space *mapping) 36047906d00cSAndrea Arcangeli { 36057906d00cSAndrea Arcangeli if (test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) { 36067906d00cSAndrea Arcangeli /* 36077906d00cSAndrea Arcangeli * AS_MM_ALL_LOCKS can't change to 0 from under us 36087906d00cSAndrea Arcangeli * because we hold the mm_all_locks_mutex. 36097906d00cSAndrea Arcangeli */ 361083cde9e8SDavidlohr Bueso i_mmap_unlock_write(mapping); 36117906d00cSAndrea Arcangeli if (!test_and_clear_bit(AS_MM_ALL_LOCKS, 36127906d00cSAndrea Arcangeli &mapping->flags)) 36137906d00cSAndrea Arcangeli BUG(); 36147906d00cSAndrea Arcangeli } 36157906d00cSAndrea Arcangeli } 36167906d00cSAndrea Arcangeli 36177906d00cSAndrea Arcangeli /* 36187906d00cSAndrea Arcangeli * The mmap_sem cannot be released by the caller until 36197906d00cSAndrea Arcangeli * mm_drop_all_locks() returns. 36207906d00cSAndrea Arcangeli */ 36217906d00cSAndrea Arcangeli void mm_drop_all_locks(struct mm_struct *mm) 36227906d00cSAndrea Arcangeli { 36237906d00cSAndrea Arcangeli struct vm_area_struct *vma; 36245beb4930SRik van Riel struct anon_vma_chain *avc; 36257906d00cSAndrea Arcangeli 36267906d00cSAndrea Arcangeli BUG_ON(down_read_trylock(&mm->mmap_sem)); 36277906d00cSAndrea Arcangeli BUG_ON(!mutex_is_locked(&mm_all_locks_mutex)); 36287906d00cSAndrea Arcangeli 36297906d00cSAndrea Arcangeli for (vma = mm->mmap; vma; vma = vma->vm_next) { 36307906d00cSAndrea Arcangeli if (vma->anon_vma) 36315beb4930SRik van Riel list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) 36325beb4930SRik van Riel vm_unlock_anon_vma(avc->anon_vma); 36337906d00cSAndrea Arcangeli if (vma->vm_file && vma->vm_file->f_mapping) 36347906d00cSAndrea Arcangeli vm_unlock_mapping(vma->vm_file->f_mapping); 36357906d00cSAndrea Arcangeli } 36367906d00cSAndrea Arcangeli 36377906d00cSAndrea Arcangeli mutex_unlock(&mm_all_locks_mutex); 36387906d00cSAndrea Arcangeli } 36398feae131SDavid Howells 36408feae131SDavid Howells /* 36413edf41d8Sseokhoon.yoon * initialise the percpu counter for VM 36428feae131SDavid Howells */ 36438feae131SDavid Howells void __init mmap_init(void) 36448feae131SDavid Howells { 364500a62ce9SKOSAKI Motohiro int ret; 364600a62ce9SKOSAKI Motohiro 3647908c7f19STejun Heo ret = percpu_counter_init(&vm_committed_as, 0, GFP_KERNEL); 364800a62ce9SKOSAKI Motohiro VM_BUG_ON(ret); 36498feae131SDavid Howells } 3650c9b1d098SAndrew Shewmaker 3651c9b1d098SAndrew Shewmaker /* 3652c9b1d098SAndrew Shewmaker * Initialise sysctl_user_reserve_kbytes. 3653c9b1d098SAndrew Shewmaker * 3654c9b1d098SAndrew Shewmaker * This is intended to prevent a user from starting a single memory hogging 3655c9b1d098SAndrew Shewmaker * process, such that they cannot recover (kill the hog) in OVERCOMMIT_NEVER 3656c9b1d098SAndrew Shewmaker * mode. 3657c9b1d098SAndrew Shewmaker * 3658c9b1d098SAndrew Shewmaker * The default value is min(3% of free memory, 128MB) 3659c9b1d098SAndrew Shewmaker * 128MB is enough to recover with sshd/login, bash, and top/kill. 3660c9b1d098SAndrew Shewmaker */ 36611640879aSAndrew Shewmaker static int init_user_reserve(void) 3662c9b1d098SAndrew Shewmaker { 3663c9b1d098SAndrew Shewmaker unsigned long free_kbytes; 3664c9b1d098SAndrew Shewmaker 3665c41f012aSMichal Hocko free_kbytes = global_zone_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10); 3666c9b1d098SAndrew Shewmaker 3667c9b1d098SAndrew Shewmaker sysctl_user_reserve_kbytes = min(free_kbytes / 32, 1UL << 17); 3668c9b1d098SAndrew Shewmaker return 0; 3669c9b1d098SAndrew Shewmaker } 3670a64fb3cdSPaul Gortmaker subsys_initcall(init_user_reserve); 36714eeab4f5SAndrew Shewmaker 36724eeab4f5SAndrew Shewmaker /* 36734eeab4f5SAndrew Shewmaker * Initialise sysctl_admin_reserve_kbytes. 36744eeab4f5SAndrew Shewmaker * 36754eeab4f5SAndrew Shewmaker * The purpose of sysctl_admin_reserve_kbytes is to allow the sys admin 36764eeab4f5SAndrew Shewmaker * to log in and kill a memory hogging process. 36774eeab4f5SAndrew Shewmaker * 36784eeab4f5SAndrew Shewmaker * Systems with more than 256MB will reserve 8MB, enough to recover 36794eeab4f5SAndrew Shewmaker * with sshd, bash, and top in OVERCOMMIT_GUESS. Smaller systems will 36804eeab4f5SAndrew Shewmaker * only reserve 3% of free pages by default. 36814eeab4f5SAndrew Shewmaker */ 36821640879aSAndrew Shewmaker static int init_admin_reserve(void) 36834eeab4f5SAndrew Shewmaker { 36844eeab4f5SAndrew Shewmaker unsigned long free_kbytes; 36854eeab4f5SAndrew Shewmaker 3686c41f012aSMichal Hocko free_kbytes = global_zone_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10); 36874eeab4f5SAndrew Shewmaker 36884eeab4f5SAndrew Shewmaker sysctl_admin_reserve_kbytes = min(free_kbytes / 32, 1UL << 13); 36894eeab4f5SAndrew Shewmaker return 0; 36904eeab4f5SAndrew Shewmaker } 3691a64fb3cdSPaul Gortmaker subsys_initcall(init_admin_reserve); 36921640879aSAndrew Shewmaker 36931640879aSAndrew Shewmaker /* 36941640879aSAndrew Shewmaker * Reinititalise user and admin reserves if memory is added or removed. 36951640879aSAndrew Shewmaker * 36961640879aSAndrew Shewmaker * The default user reserve max is 128MB, and the default max for the 36971640879aSAndrew Shewmaker * admin reserve is 8MB. These are usually, but not always, enough to 36981640879aSAndrew Shewmaker * enable recovery from a memory hogging process using login/sshd, a shell, 36991640879aSAndrew Shewmaker * and tools like top. It may make sense to increase or even disable the 37001640879aSAndrew Shewmaker * reserve depending on the existence of swap or variations in the recovery 37011640879aSAndrew Shewmaker * tools. So, the admin may have changed them. 37021640879aSAndrew Shewmaker * 37031640879aSAndrew Shewmaker * If memory is added and the reserves have been eliminated or increased above 37041640879aSAndrew Shewmaker * the default max, then we'll trust the admin. 37051640879aSAndrew Shewmaker * 37061640879aSAndrew Shewmaker * If memory is removed and there isn't enough free memory, then we 37071640879aSAndrew Shewmaker * need to reset the reserves. 37081640879aSAndrew Shewmaker * 37091640879aSAndrew Shewmaker * Otherwise keep the reserve set by the admin. 37101640879aSAndrew Shewmaker */ 37111640879aSAndrew Shewmaker static int reserve_mem_notifier(struct notifier_block *nb, 37121640879aSAndrew Shewmaker unsigned long action, void *data) 37131640879aSAndrew Shewmaker { 37141640879aSAndrew Shewmaker unsigned long tmp, free_kbytes; 37151640879aSAndrew Shewmaker 37161640879aSAndrew Shewmaker switch (action) { 37171640879aSAndrew Shewmaker case MEM_ONLINE: 37181640879aSAndrew Shewmaker /* Default max is 128MB. Leave alone if modified by operator. */ 37191640879aSAndrew Shewmaker tmp = sysctl_user_reserve_kbytes; 37201640879aSAndrew Shewmaker if (0 < tmp && tmp < (1UL << 17)) 37211640879aSAndrew Shewmaker init_user_reserve(); 37221640879aSAndrew Shewmaker 37231640879aSAndrew Shewmaker /* Default max is 8MB. Leave alone if modified by operator. */ 37241640879aSAndrew Shewmaker tmp = sysctl_admin_reserve_kbytes; 37251640879aSAndrew Shewmaker if (0 < tmp && tmp < (1UL << 13)) 37261640879aSAndrew Shewmaker init_admin_reserve(); 37271640879aSAndrew Shewmaker 37281640879aSAndrew Shewmaker break; 37291640879aSAndrew Shewmaker case MEM_OFFLINE: 3730c41f012aSMichal Hocko free_kbytes = global_zone_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10); 37311640879aSAndrew Shewmaker 37321640879aSAndrew Shewmaker if (sysctl_user_reserve_kbytes > free_kbytes) { 37331640879aSAndrew Shewmaker init_user_reserve(); 37341640879aSAndrew Shewmaker pr_info("vm.user_reserve_kbytes reset to %lu\n", 37351640879aSAndrew Shewmaker sysctl_user_reserve_kbytes); 37361640879aSAndrew Shewmaker } 37371640879aSAndrew Shewmaker 37381640879aSAndrew Shewmaker if (sysctl_admin_reserve_kbytes > free_kbytes) { 37391640879aSAndrew Shewmaker init_admin_reserve(); 37401640879aSAndrew Shewmaker pr_info("vm.admin_reserve_kbytes reset to %lu\n", 37411640879aSAndrew Shewmaker sysctl_admin_reserve_kbytes); 37421640879aSAndrew Shewmaker } 37431640879aSAndrew Shewmaker break; 37441640879aSAndrew Shewmaker default: 37451640879aSAndrew Shewmaker break; 37461640879aSAndrew Shewmaker } 37471640879aSAndrew Shewmaker return NOTIFY_OK; 37481640879aSAndrew Shewmaker } 37491640879aSAndrew Shewmaker 37501640879aSAndrew Shewmaker static struct notifier_block reserve_mem_nb = { 37511640879aSAndrew Shewmaker .notifier_call = reserve_mem_notifier, 37521640879aSAndrew Shewmaker }; 37531640879aSAndrew Shewmaker 37541640879aSAndrew Shewmaker static int __meminit init_reserve_notifier(void) 37551640879aSAndrew Shewmaker { 37561640879aSAndrew Shewmaker if (register_hotmemory_notifier(&reserve_mem_nb)) 3757b1de0d13SMitchel Humpherys pr_err("Failed registering memory add/remove notifier for admin reserve\n"); 37581640879aSAndrew Shewmaker 37591640879aSAndrew Shewmaker return 0; 37601640879aSAndrew Shewmaker } 3761a64fb3cdSPaul Gortmaker subsys_initcall(init_reserve_notifier); 3762