1457c8996SThomas 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> 1617fca131SArnd Bergmann #include <linux/mm_inline.h> 17615d6e87SDavidlohr Bueso #include <linux/vmacache.h> 181da177e4SLinus Torvalds #include <linux/shm.h> 191da177e4SLinus Torvalds #include <linux/mman.h> 201da177e4SLinus Torvalds #include <linux/pagemap.h> 211da177e4SLinus Torvalds #include <linux/swap.h> 221da177e4SLinus Torvalds #include <linux/syscalls.h> 23c59ede7bSRandy.Dunlap #include <linux/capability.h> 241da177e4SLinus Torvalds #include <linux/init.h> 251da177e4SLinus Torvalds #include <linux/file.h> 261da177e4SLinus Torvalds #include <linux/fs.h> 271da177e4SLinus Torvalds #include <linux/personality.h> 281da177e4SLinus Torvalds #include <linux/security.h> 291da177e4SLinus Torvalds #include <linux/hugetlb.h> 30c01d5b30SHugh Dickins #include <linux/shmem_fs.h> 311da177e4SLinus Torvalds #include <linux/profile.h> 32b95f1b31SPaul Gortmaker #include <linux/export.h> 331da177e4SLinus Torvalds #include <linux/mount.h> 341da177e4SLinus Torvalds #include <linux/mempolicy.h> 351da177e4SLinus Torvalds #include <linux/rmap.h> 36cddb8a5cSAndrea Arcangeli #include <linux/mmu_notifier.h> 3782f71ae4SKonstantin Khlebnikov #include <linux/mmdebug.h> 38cdd6c482SIngo Molnar #include <linux/perf_event.h> 39120a795dSAl Viro #include <linux/audit.h> 40b15d00b6SAndrea Arcangeli #include <linux/khugepaged.h> 412b144498SSrikar Dronamraju #include <linux/uprobes.h> 42d3737187SMichel Lespinasse #include <linux/rbtree_augmented.h> 431640879aSAndrew Shewmaker #include <linux/notifier.h> 441640879aSAndrew Shewmaker #include <linux/memory.h> 45b1de0d13SMitchel Humpherys #include <linux/printk.h> 4619a809afSAndrea Arcangeli #include <linux/userfaultfd_k.h> 47d977d56cSKonstantin Khlebnikov #include <linux/moduleparam.h> 4862b5f7d0SDave Hansen #include <linux/pkeys.h> 4921292580SAndrea Arcangeli #include <linux/oom.h> 5004f5866eSAndrea Arcangeli #include <linux/sched/mm.h> 511da177e4SLinus Torvalds 527c0f6ba6SLinus Torvalds #include <linux/uaccess.h> 531da177e4SLinus Torvalds #include <asm/cacheflush.h> 541da177e4SLinus Torvalds #include <asm/tlb.h> 55d6dd61c8SJeremy Fitzhardinge #include <asm/mmu_context.h> 561da177e4SLinus Torvalds 57df529cabSJaewon Kim #define CREATE_TRACE_POINTS 58df529cabSJaewon Kim #include <trace/events/mmap.h> 59df529cabSJaewon Kim 6042b77728SJan Beulich #include "internal.h" 6142b77728SJan Beulich 623a459756SKirill Korotaev #ifndef arch_mmap_check 633a459756SKirill Korotaev #define arch_mmap_check(addr, len, flags) (0) 643a459756SKirill Korotaev #endif 653a459756SKirill Korotaev 66d07e2259SDaniel Cashman #ifdef CONFIG_HAVE_ARCH_MMAP_RND_BITS 67d07e2259SDaniel Cashman const int mmap_rnd_bits_min = CONFIG_ARCH_MMAP_RND_BITS_MIN; 68d07e2259SDaniel Cashman const int mmap_rnd_bits_max = CONFIG_ARCH_MMAP_RND_BITS_MAX; 69d07e2259SDaniel Cashman int mmap_rnd_bits __read_mostly = CONFIG_ARCH_MMAP_RND_BITS; 70d07e2259SDaniel Cashman #endif 71d07e2259SDaniel Cashman #ifdef CONFIG_HAVE_ARCH_MMAP_RND_COMPAT_BITS 72d07e2259SDaniel Cashman const int mmap_rnd_compat_bits_min = CONFIG_ARCH_MMAP_RND_COMPAT_BITS_MIN; 73d07e2259SDaniel Cashman const int mmap_rnd_compat_bits_max = CONFIG_ARCH_MMAP_RND_COMPAT_BITS_MAX; 74d07e2259SDaniel Cashman int mmap_rnd_compat_bits __read_mostly = CONFIG_ARCH_MMAP_RND_COMPAT_BITS; 75d07e2259SDaniel Cashman #endif 76d07e2259SDaniel Cashman 77f4fcd558SKonstantin Khlebnikov static bool ignore_rlimit_data; 78d977d56cSKonstantin Khlebnikov core_param(ignore_rlimit_data, ignore_rlimit_data, bool, 0644); 79d07e2259SDaniel Cashman 80e0da382cSHugh Dickins static void unmap_region(struct mm_struct *mm, 81e0da382cSHugh Dickins struct vm_area_struct *vma, struct vm_area_struct *prev, 82e0da382cSHugh Dickins unsigned long start, unsigned long end); 83e0da382cSHugh Dickins 841da177e4SLinus Torvalds /* description of effects of mapping type and prot in current implementation. 851da177e4SLinus Torvalds * this is due to the limited x86 page protection hardware. The expected 861da177e4SLinus Torvalds * behavior is in parens: 871da177e4SLinus Torvalds * 881da177e4SLinus Torvalds * map_type prot 891da177e4SLinus Torvalds * PROT_NONE PROT_READ PROT_WRITE PROT_EXEC 901da177e4SLinus Torvalds * MAP_SHARED r: (no) no r: (yes) yes r: (no) yes r: (no) yes 911da177e4SLinus Torvalds * w: (no) no w: (no) no w: (yes) yes w: (no) no 921da177e4SLinus Torvalds * x: (no) no x: (no) yes x: (no) yes x: (yes) yes 931da177e4SLinus Torvalds * 941da177e4SLinus Torvalds * MAP_PRIVATE r: (no) no r: (yes) yes r: (no) yes r: (no) yes 951da177e4SLinus Torvalds * w: (no) no w: (no) no w: (copy) copy w: (no) no 961da177e4SLinus Torvalds * x: (no) no x: (no) yes x: (no) yes x: (yes) yes 9718107f8aSVladimir Murzin * 9818107f8aSVladimir Murzin * On arm64, PROT_EXEC has the following behaviour for both MAP_SHARED and 9918107f8aSVladimir Murzin * MAP_PRIVATE (with Enhanced PAN supported): 10018107f8aSVladimir Murzin * r: (no) no 10118107f8aSVladimir Murzin * w: (no) no 10218107f8aSVladimir Murzin * x: (yes) yes 1031da177e4SLinus Torvalds */ 104ac34ceafSDaniel Micay pgprot_t protection_map[16] __ro_after_init = { 1056c862bd0SAnshuman Khandual [VM_NONE] = __P000, 1066c862bd0SAnshuman Khandual [VM_READ] = __P001, 1076c862bd0SAnshuman Khandual [VM_WRITE] = __P010, 1086c862bd0SAnshuman Khandual [VM_WRITE | VM_READ] = __P011, 1096c862bd0SAnshuman Khandual [VM_EXEC] = __P100, 1106c862bd0SAnshuman Khandual [VM_EXEC | VM_READ] = __P101, 1116c862bd0SAnshuman Khandual [VM_EXEC | VM_WRITE] = __P110, 1126c862bd0SAnshuman Khandual [VM_EXEC | VM_WRITE | VM_READ] = __P111, 1136c862bd0SAnshuman Khandual [VM_SHARED] = __S000, 1146c862bd0SAnshuman Khandual [VM_SHARED | VM_READ] = __S001, 1156c862bd0SAnshuman Khandual [VM_SHARED | VM_WRITE] = __S010, 1166c862bd0SAnshuman Khandual [VM_SHARED | VM_WRITE | VM_READ] = __S011, 1176c862bd0SAnshuman Khandual [VM_SHARED | VM_EXEC] = __S100, 1186c862bd0SAnshuman Khandual [VM_SHARED | VM_EXEC | VM_READ] = __S101, 1196c862bd0SAnshuman Khandual [VM_SHARED | VM_EXEC | VM_WRITE] = __S110, 1206c862bd0SAnshuman Khandual [VM_SHARED | VM_EXEC | VM_WRITE | VM_READ] = __S111 1211da177e4SLinus Torvalds }; 1221da177e4SLinus Torvalds 12367436193SAnshuman Khandual #ifndef CONFIG_ARCH_HAS_VM_GET_PAGE_PROT 124804af2cfSHugh Dickins pgprot_t vm_get_page_prot(unsigned long vm_flags) 125804af2cfSHugh Dickins { 1263afa7930SAnshuman Khandual return protection_map[vm_flags & (VM_READ|VM_WRITE|VM_EXEC|VM_SHARED)]; 127804af2cfSHugh Dickins } 128804af2cfSHugh Dickins EXPORT_SYMBOL(vm_get_page_prot); 12967436193SAnshuman Khandual #endif /* CONFIG_ARCH_HAS_VM_GET_PAGE_PROT */ 130804af2cfSHugh Dickins 13164e45507SPeter Feiner static pgprot_t vm_pgprot_modify(pgprot_t oldprot, unsigned long vm_flags) 13264e45507SPeter Feiner { 13364e45507SPeter Feiner return pgprot_modify(oldprot, vm_get_page_prot(vm_flags)); 13464e45507SPeter Feiner } 13564e45507SPeter Feiner 13664e45507SPeter Feiner /* Update vma->vm_page_prot to reflect vma->vm_flags. */ 13764e45507SPeter Feiner void vma_set_page_prot(struct vm_area_struct *vma) 13864e45507SPeter Feiner { 13964e45507SPeter Feiner unsigned long vm_flags = vma->vm_flags; 1406d2329f8SAndrea Arcangeli pgprot_t vm_page_prot; 14164e45507SPeter Feiner 1426d2329f8SAndrea Arcangeli vm_page_prot = vm_pgprot_modify(vma->vm_page_prot, vm_flags); 1436d2329f8SAndrea Arcangeli if (vma_wants_writenotify(vma, vm_page_prot)) { 14464e45507SPeter Feiner vm_flags &= ~VM_SHARED; 1456d2329f8SAndrea Arcangeli vm_page_prot = vm_pgprot_modify(vm_page_prot, vm_flags); 14664e45507SPeter Feiner } 147c1e8d7c6SMichel Lespinasse /* remove_protection_ptes reads vma->vm_page_prot without mmap_lock */ 1486d2329f8SAndrea Arcangeli WRITE_ONCE(vma->vm_page_prot, vm_page_prot); 14964e45507SPeter Feiner } 15064e45507SPeter Feiner 1511da177e4SLinus Torvalds /* 152c8c06efaSDavidlohr Bueso * Requires inode->i_mapping->i_mmap_rwsem 1531da177e4SLinus Torvalds */ 1541da177e4SLinus Torvalds static void __remove_shared_vm_struct(struct vm_area_struct *vma, 1551da177e4SLinus Torvalds struct file *file, struct address_space *mapping) 1561da177e4SLinus Torvalds { 1571da177e4SLinus Torvalds if (vma->vm_flags & VM_SHARED) 1584bb5f5d9SDavid Herrmann mapping_unmap_writable(mapping); 1591da177e4SLinus Torvalds 1601da177e4SLinus Torvalds flush_dcache_mmap_lock(mapping); 1616b2dbba8SMichel Lespinasse vma_interval_tree_remove(vma, &mapping->i_mmap); 1621da177e4SLinus Torvalds flush_dcache_mmap_unlock(mapping); 1631da177e4SLinus Torvalds } 1641da177e4SLinus Torvalds 1651da177e4SLinus Torvalds /* 1666b2dbba8SMichel Lespinasse * Unlink a file-based vm structure from its interval tree, to hide 167a8fb5618SHugh Dickins * vma from rmap and vmtruncate before freeing its page tables. 1681da177e4SLinus Torvalds */ 169a8fb5618SHugh Dickins void unlink_file_vma(struct vm_area_struct *vma) 1701da177e4SLinus Torvalds { 1711da177e4SLinus Torvalds struct file *file = vma->vm_file; 1721da177e4SLinus Torvalds 1731da177e4SLinus Torvalds if (file) { 1741da177e4SLinus Torvalds struct address_space *mapping = file->f_mapping; 17583cde9e8SDavidlohr Bueso i_mmap_lock_write(mapping); 1761da177e4SLinus Torvalds __remove_shared_vm_struct(vma, file, mapping); 17783cde9e8SDavidlohr Bueso i_mmap_unlock_write(mapping); 1781da177e4SLinus Torvalds } 179a8fb5618SHugh Dickins } 180a8fb5618SHugh Dickins 181a8fb5618SHugh Dickins /* 182a8fb5618SHugh Dickins * Close a vm structure and free it, returning the next. 183a8fb5618SHugh Dickins */ 184a8fb5618SHugh Dickins static struct vm_area_struct *remove_vma(struct vm_area_struct *vma) 185a8fb5618SHugh Dickins { 186a8fb5618SHugh Dickins struct vm_area_struct *next = vma->vm_next; 187a8fb5618SHugh Dickins 188a8fb5618SHugh Dickins might_sleep(); 1891da177e4SLinus Torvalds if (vma->vm_ops && vma->vm_ops->close) 1901da177e4SLinus Torvalds vma->vm_ops->close(vma); 191e9714acfSKonstantin Khlebnikov if (vma->vm_file) 192a8fb5618SHugh Dickins fput(vma->vm_file); 193f0be3d32SLee Schermerhorn mpol_put(vma_policy(vma)); 1943928d4f5SLinus Torvalds vm_area_free(vma); 195a8fb5618SHugh Dickins return next; 1961da177e4SLinus Torvalds } 1971da177e4SLinus Torvalds 198bb177a73SMichal Hocko static int do_brk_flags(unsigned long addr, unsigned long request, unsigned long flags, 199bb177a73SMichal Hocko struct list_head *uf); 2006a6160a7SHeiko Carstens SYSCALL_DEFINE1(brk, unsigned long, brk) 2011da177e4SLinus Torvalds { 2029bc8039eSYang Shi unsigned long newbrk, oldbrk, origbrk; 2031da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 2041be7107fSHugh Dickins struct vm_area_struct *next; 205a5b4592cSJiri Kosina unsigned long min_brk; 206128557ffSMichel Lespinasse bool populate; 2079bc8039eSYang Shi bool downgraded = false; 208897ab3e0SMike Rapoport LIST_HEAD(uf); 2091da177e4SLinus Torvalds 210d8ed45c5SMichel Lespinasse if (mmap_write_lock_killable(mm)) 211dc0ef0dfSMichal Hocko return -EINTR; 2121da177e4SLinus Torvalds 2139bc8039eSYang Shi origbrk = mm->brk; 2149bc8039eSYang Shi 215a5b4592cSJiri Kosina #ifdef CONFIG_COMPAT_BRK 2165520e894SJiri Kosina /* 2175520e894SJiri Kosina * CONFIG_COMPAT_BRK can still be overridden by setting 2185520e894SJiri Kosina * randomize_va_space to 2, which will still cause mm->start_brk 2195520e894SJiri Kosina * to be arbitrarily shifted 2205520e894SJiri Kosina */ 2214471a675SJiri Kosina if (current->brk_randomized) 2225520e894SJiri Kosina min_brk = mm->start_brk; 2235520e894SJiri Kosina else 2245520e894SJiri Kosina min_brk = mm->end_data; 225a5b4592cSJiri Kosina #else 226a5b4592cSJiri Kosina min_brk = mm->start_brk; 227a5b4592cSJiri Kosina #endif 228a5b4592cSJiri Kosina if (brk < min_brk) 2291da177e4SLinus Torvalds goto out; 2301e624196SRam Gupta 2311e624196SRam Gupta /* 2321e624196SRam Gupta * Check against rlimit here. If this check is done later after the test 2331e624196SRam Gupta * of oldbrk with newbrk then it can escape the test and let the data 2341e624196SRam Gupta * segment grow beyond its set limit the in case where the limit is 2351e624196SRam Gupta * not page aligned -Ram Gupta 2361e624196SRam Gupta */ 2378764b338SCyrill Gorcunov if (check_data_rlimit(rlimit(RLIMIT_DATA), brk, mm->start_brk, 2388764b338SCyrill Gorcunov mm->end_data, mm->start_data)) 2391e624196SRam Gupta goto out; 2401e624196SRam Gupta 2411da177e4SLinus Torvalds newbrk = PAGE_ALIGN(brk); 2421da177e4SLinus Torvalds oldbrk = PAGE_ALIGN(mm->brk); 2439bc8039eSYang Shi if (oldbrk == newbrk) { 2449bc8039eSYang Shi mm->brk = brk; 2459bc8039eSYang Shi goto success; 2469bc8039eSYang Shi } 2471da177e4SLinus Torvalds 2489bc8039eSYang Shi /* 2499bc8039eSYang Shi * Always allow shrinking brk. 250c1e8d7c6SMichel Lespinasse * __do_munmap() may downgrade mmap_lock to read. 2519bc8039eSYang Shi */ 2521da177e4SLinus Torvalds if (brk <= mm->brk) { 2539bc8039eSYang Shi int ret; 2549bc8039eSYang Shi 2559bc8039eSYang Shi /* 256c1e8d7c6SMichel Lespinasse * mm->brk must to be protected by write mmap_lock so update it 257c1e8d7c6SMichel Lespinasse * before downgrading mmap_lock. When __do_munmap() fails, 2589bc8039eSYang Shi * mm->brk will be restored from origbrk. 2599bc8039eSYang Shi */ 2609bc8039eSYang Shi mm->brk = brk; 2619bc8039eSYang Shi ret = __do_munmap(mm, newbrk, oldbrk-newbrk, &uf, true); 2629bc8039eSYang Shi if (ret < 0) { 2639bc8039eSYang Shi mm->brk = origbrk; 2641da177e4SLinus Torvalds goto out; 2659bc8039eSYang Shi } else if (ret == 1) { 2669bc8039eSYang Shi downgraded = true; 2679bc8039eSYang Shi } 2689bc8039eSYang Shi goto success; 2691da177e4SLinus Torvalds } 2701da177e4SLinus Torvalds 2711da177e4SLinus Torvalds /* Check against existing mmap mappings. */ 2721be7107fSHugh Dickins next = find_vma(mm, oldbrk); 2731be7107fSHugh Dickins if (next && newbrk + PAGE_SIZE > vm_start_gap(next)) 2741da177e4SLinus Torvalds goto out; 2751da177e4SLinus Torvalds 2761da177e4SLinus Torvalds /* Ok, looks good - let it rip. */ 277bb177a73SMichal Hocko if (do_brk_flags(oldbrk, newbrk-oldbrk, 0, &uf) < 0) 2781da177e4SLinus Torvalds goto out; 2791da177e4SLinus Torvalds mm->brk = brk; 2809bc8039eSYang Shi 2819bc8039eSYang Shi success: 282128557ffSMichel Lespinasse populate = newbrk > oldbrk && (mm->def_flags & VM_LOCKED) != 0; 2839bc8039eSYang Shi if (downgraded) 284d8ed45c5SMichel Lespinasse mmap_read_unlock(mm); 2859bc8039eSYang Shi else 286d8ed45c5SMichel Lespinasse mmap_write_unlock(mm); 287897ab3e0SMike Rapoport userfaultfd_unmap_complete(mm, &uf); 288128557ffSMichel Lespinasse if (populate) 289128557ffSMichel Lespinasse mm_populate(oldbrk, newbrk - oldbrk); 290128557ffSMichel Lespinasse return brk; 291128557ffSMichel Lespinasse 2921da177e4SLinus Torvalds out: 293d8ed45c5SMichel Lespinasse mmap_write_unlock(mm); 294b7204006SAdrian Huang return origbrk; 2951da177e4SLinus Torvalds } 2961da177e4SLinus Torvalds 297315cc066SMichel Lespinasse static inline unsigned long vma_compute_gap(struct vm_area_struct *vma) 298d3737187SMichel Lespinasse { 299315cc066SMichel Lespinasse unsigned long gap, prev_end; 3001be7107fSHugh Dickins 3011be7107fSHugh Dickins /* 3021be7107fSHugh Dickins * Note: in the rare case of a VM_GROWSDOWN above a VM_GROWSUP, we 3031be7107fSHugh Dickins * allow two stack_guard_gaps between them here, and when choosing 3041be7107fSHugh Dickins * an unmapped area; whereas when expanding we only require one. 3051be7107fSHugh Dickins * That's a little inconsistent, but keeps the code here simpler. 3061be7107fSHugh Dickins */ 307315cc066SMichel Lespinasse gap = vm_start_gap(vma); 3081be7107fSHugh Dickins if (vma->vm_prev) { 3091be7107fSHugh Dickins prev_end = vm_end_gap(vma->vm_prev); 310315cc066SMichel Lespinasse if (gap > prev_end) 311315cc066SMichel Lespinasse gap -= prev_end; 3121be7107fSHugh Dickins else 313315cc066SMichel Lespinasse gap = 0; 3141be7107fSHugh Dickins } 315315cc066SMichel Lespinasse return gap; 316315cc066SMichel Lespinasse } 317315cc066SMichel Lespinasse 318315cc066SMichel Lespinasse #ifdef CONFIG_DEBUG_VM_RB 319315cc066SMichel Lespinasse static unsigned long vma_compute_subtree_gap(struct vm_area_struct *vma) 320315cc066SMichel Lespinasse { 321315cc066SMichel Lespinasse unsigned long max = vma_compute_gap(vma), subtree_gap; 322d3737187SMichel Lespinasse if (vma->vm_rb.rb_left) { 323d3737187SMichel Lespinasse subtree_gap = rb_entry(vma->vm_rb.rb_left, 324d3737187SMichel Lespinasse struct vm_area_struct, vm_rb)->rb_subtree_gap; 325d3737187SMichel Lespinasse if (subtree_gap > max) 326d3737187SMichel Lespinasse max = subtree_gap; 327d3737187SMichel Lespinasse } 328d3737187SMichel Lespinasse if (vma->vm_rb.rb_right) { 329d3737187SMichel Lespinasse subtree_gap = rb_entry(vma->vm_rb.rb_right, 330d3737187SMichel Lespinasse struct vm_area_struct, vm_rb)->rb_subtree_gap; 331d3737187SMichel Lespinasse if (subtree_gap > max) 332d3737187SMichel Lespinasse max = subtree_gap; 333d3737187SMichel Lespinasse } 334d3737187SMichel Lespinasse return max; 335d3737187SMichel Lespinasse } 336d3737187SMichel Lespinasse 337acf128d0SAndrea Arcangeli static int browse_rb(struct mm_struct *mm) 3381da177e4SLinus Torvalds { 339acf128d0SAndrea Arcangeli struct rb_root *root = &mm->mm_rb; 3405a0768f6SMichel Lespinasse int i = 0, j, bug = 0; 3411da177e4SLinus Torvalds struct rb_node *nd, *pn = NULL; 3421da177e4SLinus Torvalds unsigned long prev = 0, pend = 0; 3431da177e4SLinus Torvalds 3441da177e4SLinus Torvalds for (nd = rb_first(root); nd; nd = rb_next(nd)) { 3451da177e4SLinus Torvalds struct vm_area_struct *vma; 3461da177e4SLinus Torvalds vma = rb_entry(nd, struct vm_area_struct, vm_rb); 3475a0768f6SMichel Lespinasse if (vma->vm_start < prev) { 348ff26f70fSAndrew Morton pr_emerg("vm_start %lx < prev %lx\n", 349ff26f70fSAndrew Morton vma->vm_start, prev); 3505a0768f6SMichel Lespinasse bug = 1; 3515a0768f6SMichel Lespinasse } 3525a0768f6SMichel Lespinasse if (vma->vm_start < pend) { 353ff26f70fSAndrew Morton pr_emerg("vm_start %lx < pend %lx\n", 354ff26f70fSAndrew Morton vma->vm_start, pend); 3555a0768f6SMichel Lespinasse bug = 1; 3565a0768f6SMichel Lespinasse } 3575a0768f6SMichel Lespinasse if (vma->vm_start > vma->vm_end) { 358ff26f70fSAndrew Morton pr_emerg("vm_start %lx > vm_end %lx\n", 359ff26f70fSAndrew Morton vma->vm_start, vma->vm_end); 3605a0768f6SMichel Lespinasse bug = 1; 3615a0768f6SMichel Lespinasse } 362acf128d0SAndrea Arcangeli spin_lock(&mm->page_table_lock); 3635a0768f6SMichel Lespinasse if (vma->rb_subtree_gap != vma_compute_subtree_gap(vma)) { 3648542bdfcSSasha Levin pr_emerg("free gap %lx, correct %lx\n", 3655a0768f6SMichel Lespinasse vma->rb_subtree_gap, 3665a0768f6SMichel Lespinasse vma_compute_subtree_gap(vma)); 3675a0768f6SMichel Lespinasse bug = 1; 3685a0768f6SMichel Lespinasse } 369acf128d0SAndrea Arcangeli spin_unlock(&mm->page_table_lock); 3701da177e4SLinus Torvalds i++; 3711da177e4SLinus Torvalds pn = nd; 372d1af65d1SDavid Miller prev = vma->vm_start; 373d1af65d1SDavid Miller pend = vma->vm_end; 3741da177e4SLinus Torvalds } 3751da177e4SLinus Torvalds j = 0; 3765a0768f6SMichel Lespinasse for (nd = pn; nd; nd = rb_prev(nd)) 3771da177e4SLinus Torvalds j++; 3785a0768f6SMichel Lespinasse if (i != j) { 3798542bdfcSSasha Levin pr_emerg("backwards %d, forwards %d\n", j, i); 3805a0768f6SMichel Lespinasse bug = 1; 3811da177e4SLinus Torvalds } 3825a0768f6SMichel Lespinasse return bug ? -1 : i; 3831da177e4SLinus Torvalds } 3841da177e4SLinus Torvalds 385d3737187SMichel Lespinasse static void validate_mm_rb(struct rb_root *root, struct vm_area_struct *ignore) 386d3737187SMichel Lespinasse { 387d3737187SMichel Lespinasse struct rb_node *nd; 388d3737187SMichel Lespinasse 389d3737187SMichel Lespinasse for (nd = rb_first(root); nd; nd = rb_next(nd)) { 390d3737187SMichel Lespinasse struct vm_area_struct *vma; 391d3737187SMichel Lespinasse vma = rb_entry(nd, struct vm_area_struct, vm_rb); 39296dad67fSSasha Levin VM_BUG_ON_VMA(vma != ignore && 39396dad67fSSasha Levin vma->rb_subtree_gap != vma_compute_subtree_gap(vma), 39496dad67fSSasha Levin vma); 395d3737187SMichel Lespinasse } 3961da177e4SLinus Torvalds } 3971da177e4SLinus Torvalds 398eafd4dc4SRashika Kheria static void validate_mm(struct mm_struct *mm) 3991da177e4SLinus Torvalds { 4001da177e4SLinus Torvalds int bug = 0; 4011da177e4SLinus Torvalds int i = 0; 4025a0768f6SMichel Lespinasse unsigned long highest_address = 0; 403ed8ea815SMichel Lespinasse struct vm_area_struct *vma = mm->mmap; 404ff26f70fSAndrew Morton 405ed8ea815SMichel Lespinasse while (vma) { 40612352d3cSKonstantin Khlebnikov struct anon_vma *anon_vma = vma->anon_vma; 407ed8ea815SMichel Lespinasse struct anon_vma_chain *avc; 408ff26f70fSAndrew Morton 40912352d3cSKonstantin Khlebnikov if (anon_vma) { 41012352d3cSKonstantin Khlebnikov anon_vma_lock_read(anon_vma); 411ed8ea815SMichel Lespinasse list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) 412ed8ea815SMichel Lespinasse anon_vma_interval_tree_verify(avc); 41312352d3cSKonstantin Khlebnikov anon_vma_unlock_read(anon_vma); 41412352d3cSKonstantin Khlebnikov } 41512352d3cSKonstantin Khlebnikov 4161be7107fSHugh Dickins highest_address = vm_end_gap(vma); 417ed8ea815SMichel Lespinasse vma = vma->vm_next; 4181da177e4SLinus Torvalds i++; 4191da177e4SLinus Torvalds } 4205a0768f6SMichel Lespinasse if (i != mm->map_count) { 4218542bdfcSSasha Levin pr_emerg("map_count %d vm_next %d\n", mm->map_count, i); 4225a0768f6SMichel Lespinasse bug = 1; 4235a0768f6SMichel Lespinasse } 4245a0768f6SMichel Lespinasse if (highest_address != mm->highest_vm_end) { 4258542bdfcSSasha Levin pr_emerg("mm->highest_vm_end %lx, found %lx\n", 4265a0768f6SMichel Lespinasse mm->highest_vm_end, highest_address); 4275a0768f6SMichel Lespinasse bug = 1; 4285a0768f6SMichel Lespinasse } 429acf128d0SAndrea Arcangeli i = browse_rb(mm); 4305a0768f6SMichel Lespinasse if (i != mm->map_count) { 431ff26f70fSAndrew Morton if (i != -1) 4328542bdfcSSasha Levin pr_emerg("map_count %d rb %d\n", mm->map_count, i); 4335a0768f6SMichel Lespinasse bug = 1; 4345a0768f6SMichel Lespinasse } 43596dad67fSSasha Levin VM_BUG_ON_MM(bug, mm); 4361da177e4SLinus Torvalds } 4371da177e4SLinus Torvalds #else 438d3737187SMichel Lespinasse #define validate_mm_rb(root, ignore) do { } while (0) 4391da177e4SLinus Torvalds #define validate_mm(mm) do { } while (0) 4401da177e4SLinus Torvalds #endif 4411da177e4SLinus Torvalds 442315cc066SMichel Lespinasse RB_DECLARE_CALLBACKS_MAX(static, vma_gap_callbacks, 443315cc066SMichel Lespinasse struct vm_area_struct, vm_rb, 444315cc066SMichel Lespinasse unsigned long, rb_subtree_gap, vma_compute_gap) 445d3737187SMichel Lespinasse 446d3737187SMichel Lespinasse /* 447d3737187SMichel Lespinasse * Update augmented rbtree rb_subtree_gap values after vma->vm_start or 448d3737187SMichel Lespinasse * vma->vm_prev->vm_end values changed, without modifying the vma's position 449d3737187SMichel Lespinasse * in the rbtree. 450d3737187SMichel Lespinasse */ 451d3737187SMichel Lespinasse static void vma_gap_update(struct vm_area_struct *vma) 452d3737187SMichel Lespinasse { 453d3737187SMichel Lespinasse /* 454315cc066SMichel Lespinasse * As it turns out, RB_DECLARE_CALLBACKS_MAX() already created 455315cc066SMichel Lespinasse * a callback function that does exactly what we want. 456d3737187SMichel Lespinasse */ 457d3737187SMichel Lespinasse vma_gap_callbacks_propagate(&vma->vm_rb, NULL); 458d3737187SMichel Lespinasse } 459d3737187SMichel Lespinasse 460d3737187SMichel Lespinasse static inline void vma_rb_insert(struct vm_area_struct *vma, 461d3737187SMichel Lespinasse struct rb_root *root) 462d3737187SMichel Lespinasse { 463d3737187SMichel Lespinasse /* All rb_subtree_gap values must be consistent prior to insertion */ 464d3737187SMichel Lespinasse validate_mm_rb(root, NULL); 465d3737187SMichel Lespinasse 466d3737187SMichel Lespinasse rb_insert_augmented(&vma->vm_rb, root, &vma_gap_callbacks); 467d3737187SMichel Lespinasse } 468d3737187SMichel Lespinasse 4698f26e0b1SAndrea Arcangeli static void __vma_rb_erase(struct vm_area_struct *vma, struct rb_root *root) 4708f26e0b1SAndrea Arcangeli { 4718f26e0b1SAndrea Arcangeli /* 4728f26e0b1SAndrea Arcangeli * Note rb_erase_augmented is a fairly large inline function, 4738f26e0b1SAndrea Arcangeli * so make sure we instantiate it only once with our desired 4748f26e0b1SAndrea Arcangeli * augmented rbtree callbacks. 4758f26e0b1SAndrea Arcangeli */ 4768f26e0b1SAndrea Arcangeli rb_erase_augmented(&vma->vm_rb, root, &vma_gap_callbacks); 4778f26e0b1SAndrea Arcangeli } 4788f26e0b1SAndrea Arcangeli 4798f26e0b1SAndrea Arcangeli static __always_inline void vma_rb_erase_ignore(struct vm_area_struct *vma, 4808f26e0b1SAndrea Arcangeli struct rb_root *root, 4818f26e0b1SAndrea Arcangeli struct vm_area_struct *ignore) 4828f26e0b1SAndrea Arcangeli { 4838f26e0b1SAndrea Arcangeli /* 4848f26e0b1SAndrea Arcangeli * All rb_subtree_gap values must be consistent prior to erase, 4854d1e7243SWei Yang * with the possible exception of 4864d1e7243SWei Yang * 4874d1e7243SWei Yang * a. the "next" vma being erased if next->vm_start was reduced in 4884d1e7243SWei Yang * __vma_adjust() -> __vma_unlink() 4894d1e7243SWei Yang * b. the vma being erased in detach_vmas_to_be_unmapped() -> 4904d1e7243SWei Yang * vma_rb_erase() 4918f26e0b1SAndrea Arcangeli */ 4928f26e0b1SAndrea Arcangeli validate_mm_rb(root, ignore); 4938f26e0b1SAndrea Arcangeli 4948f26e0b1SAndrea Arcangeli __vma_rb_erase(vma, root); 4958f26e0b1SAndrea Arcangeli } 4968f26e0b1SAndrea Arcangeli 4978f26e0b1SAndrea Arcangeli static __always_inline void vma_rb_erase(struct vm_area_struct *vma, 4988f26e0b1SAndrea Arcangeli struct rb_root *root) 499d3737187SMichel Lespinasse { 5004d1e7243SWei Yang vma_rb_erase_ignore(vma, root, vma); 501d3737187SMichel Lespinasse } 502d3737187SMichel Lespinasse 503bf181b9fSMichel Lespinasse /* 504bf181b9fSMichel Lespinasse * vma has some anon_vma assigned, and is already inserted on that 505bf181b9fSMichel Lespinasse * anon_vma's interval trees. 506bf181b9fSMichel Lespinasse * 507bf181b9fSMichel Lespinasse * Before updating the vma's vm_start / vm_end / vm_pgoff fields, the 508bf181b9fSMichel Lespinasse * vma must be removed from the anon_vma's interval trees using 509bf181b9fSMichel Lespinasse * anon_vma_interval_tree_pre_update_vma(). 510bf181b9fSMichel Lespinasse * 511bf181b9fSMichel Lespinasse * After the update, the vma will be reinserted using 512bf181b9fSMichel Lespinasse * anon_vma_interval_tree_post_update_vma(). 513bf181b9fSMichel Lespinasse * 514c1e8d7c6SMichel Lespinasse * The entire update must be protected by exclusive mmap_lock and by 515bf181b9fSMichel Lespinasse * the root anon_vma's mutex. 516bf181b9fSMichel Lespinasse */ 517bf181b9fSMichel Lespinasse static inline void 518bf181b9fSMichel Lespinasse anon_vma_interval_tree_pre_update_vma(struct vm_area_struct *vma) 519bf181b9fSMichel Lespinasse { 520bf181b9fSMichel Lespinasse struct anon_vma_chain *avc; 521bf181b9fSMichel Lespinasse 522bf181b9fSMichel Lespinasse list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) 523bf181b9fSMichel Lespinasse anon_vma_interval_tree_remove(avc, &avc->anon_vma->rb_root); 524bf181b9fSMichel Lespinasse } 525bf181b9fSMichel Lespinasse 526bf181b9fSMichel Lespinasse static inline void 527bf181b9fSMichel Lespinasse anon_vma_interval_tree_post_update_vma(struct vm_area_struct *vma) 528bf181b9fSMichel Lespinasse { 529bf181b9fSMichel Lespinasse struct anon_vma_chain *avc; 530bf181b9fSMichel Lespinasse 531bf181b9fSMichel Lespinasse list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) 532bf181b9fSMichel Lespinasse anon_vma_interval_tree_insert(avc, &avc->anon_vma->rb_root); 533bf181b9fSMichel Lespinasse } 534bf181b9fSMichel Lespinasse 5356597d783SHugh Dickins static int find_vma_links(struct mm_struct *mm, unsigned long addr, 5366597d783SHugh Dickins unsigned long end, struct vm_area_struct **pprev, 5376597d783SHugh Dickins struct rb_node ***rb_link, struct rb_node **rb_parent) 5381da177e4SLinus Torvalds { 5391da177e4SLinus Torvalds struct rb_node **__rb_link, *__rb_parent, *rb_prev; 5401da177e4SLinus Torvalds 5415b78ed24SLuigi Rizzo mmap_assert_locked(mm); 5421da177e4SLinus Torvalds __rb_link = &mm->mm_rb.rb_node; 5431da177e4SLinus Torvalds rb_prev = __rb_parent = NULL; 5441da177e4SLinus Torvalds 5451da177e4SLinus Torvalds while (*__rb_link) { 5461da177e4SLinus Torvalds struct vm_area_struct *vma_tmp; 5471da177e4SLinus Torvalds 5481da177e4SLinus Torvalds __rb_parent = *__rb_link; 5491da177e4SLinus Torvalds vma_tmp = rb_entry(__rb_parent, struct vm_area_struct, vm_rb); 5501da177e4SLinus Torvalds 5511da177e4SLinus Torvalds if (vma_tmp->vm_end > addr) { 5526597d783SHugh Dickins /* Fail if an existing vma overlaps the area */ 5536597d783SHugh Dickins if (vma_tmp->vm_start < end) 5546597d783SHugh Dickins return -ENOMEM; 5551da177e4SLinus Torvalds __rb_link = &__rb_parent->rb_left; 5561da177e4SLinus Torvalds } else { 5571da177e4SLinus Torvalds rb_prev = __rb_parent; 5581da177e4SLinus Torvalds __rb_link = &__rb_parent->rb_right; 5591da177e4SLinus Torvalds } 5601da177e4SLinus Torvalds } 5611da177e4SLinus Torvalds 5621da177e4SLinus Torvalds *pprev = NULL; 5631da177e4SLinus Torvalds if (rb_prev) 5641da177e4SLinus Torvalds *pprev = rb_entry(rb_prev, struct vm_area_struct, vm_rb); 5651da177e4SLinus Torvalds *rb_link = __rb_link; 5661da177e4SLinus Torvalds *rb_parent = __rb_parent; 5676597d783SHugh Dickins return 0; 5681da177e4SLinus Torvalds } 5691da177e4SLinus Torvalds 5703903b55aSLiam R. Howlett /* 5713903b55aSLiam R. Howlett * vma_next() - Get the next VMA. 5723903b55aSLiam R. Howlett * @mm: The mm_struct. 5733903b55aSLiam R. Howlett * @vma: The current vma. 5743903b55aSLiam R. Howlett * 5753903b55aSLiam R. Howlett * If @vma is NULL, return the first vma in the mm. 5763903b55aSLiam R. Howlett * 5773903b55aSLiam R. Howlett * Returns: The next VMA after @vma. 5783903b55aSLiam R. Howlett */ 5793903b55aSLiam R. Howlett static inline struct vm_area_struct *vma_next(struct mm_struct *mm, 5803903b55aSLiam R. Howlett struct vm_area_struct *vma) 5813903b55aSLiam R. Howlett { 5823903b55aSLiam R. Howlett if (!vma) 5833903b55aSLiam R. Howlett return mm->mmap; 5843903b55aSLiam R. Howlett 5853903b55aSLiam R. Howlett return vma->vm_next; 5863903b55aSLiam R. Howlett } 587fb8090b6SLiam R. Howlett 588fb8090b6SLiam R. Howlett /* 589fb8090b6SLiam R. Howlett * munmap_vma_range() - munmap VMAs that overlap a range. 590fb8090b6SLiam R. Howlett * @mm: The mm struct 591fb8090b6SLiam R. Howlett * @start: The start of the range. 592fb8090b6SLiam R. Howlett * @len: The length of the range. 593fb8090b6SLiam R. Howlett * @pprev: pointer to the pointer that will be set to previous vm_area_struct 594fb8090b6SLiam R. Howlett * @rb_link: the rb_node 595fb8090b6SLiam R. Howlett * @rb_parent: the parent rb_node 596fb8090b6SLiam R. Howlett * 597fb8090b6SLiam R. Howlett * Find all the vm_area_struct that overlap from @start to 598fb8090b6SLiam R. Howlett * @end and munmap them. Set @pprev to the previous vm_area_struct. 599fb8090b6SLiam R. Howlett * 600fb8090b6SLiam R. Howlett * Returns: -ENOMEM on munmap failure or 0 on success. 601fb8090b6SLiam R. Howlett */ 602fb8090b6SLiam R. Howlett static inline int 603fb8090b6SLiam R. Howlett munmap_vma_range(struct mm_struct *mm, unsigned long start, unsigned long len, 604fb8090b6SLiam R. Howlett struct vm_area_struct **pprev, struct rb_node ***link, 605fb8090b6SLiam R. Howlett struct rb_node **parent, struct list_head *uf) 606fb8090b6SLiam R. Howlett { 607fb8090b6SLiam R. Howlett 608fb8090b6SLiam R. Howlett while (find_vma_links(mm, start, start + len, pprev, link, parent)) 609fb8090b6SLiam R. Howlett if (do_munmap(mm, start, len, uf)) 610fb8090b6SLiam R. Howlett return -ENOMEM; 611fb8090b6SLiam R. Howlett 612fb8090b6SLiam R. Howlett return 0; 613fb8090b6SLiam R. Howlett } 614e8420a8eSCyril Hrubis static unsigned long count_vma_pages_range(struct mm_struct *mm, 615e8420a8eSCyril Hrubis unsigned long addr, unsigned long end) 616e8420a8eSCyril Hrubis { 617e8420a8eSCyril Hrubis unsigned long nr_pages = 0; 618e8420a8eSCyril Hrubis struct vm_area_struct *vma; 619e8420a8eSCyril Hrubis 620f0953a1bSIngo Molnar /* Find first overlapping mapping */ 621e8420a8eSCyril Hrubis vma = find_vma_intersection(mm, addr, end); 622e8420a8eSCyril Hrubis if (!vma) 623e8420a8eSCyril Hrubis return 0; 624e8420a8eSCyril Hrubis 625e8420a8eSCyril Hrubis nr_pages = (min(end, vma->vm_end) - 626e8420a8eSCyril Hrubis max(addr, vma->vm_start)) >> PAGE_SHIFT; 627e8420a8eSCyril Hrubis 628e8420a8eSCyril Hrubis /* Iterate over the rest of the overlaps */ 629e8420a8eSCyril Hrubis for (vma = vma->vm_next; vma; vma = vma->vm_next) { 630e8420a8eSCyril Hrubis unsigned long overlap_len; 631e8420a8eSCyril Hrubis 632e8420a8eSCyril Hrubis if (vma->vm_start > end) 633e8420a8eSCyril Hrubis break; 634e8420a8eSCyril Hrubis 635e8420a8eSCyril Hrubis overlap_len = min(end, vma->vm_end) - vma->vm_start; 636e8420a8eSCyril Hrubis nr_pages += overlap_len >> PAGE_SHIFT; 637e8420a8eSCyril Hrubis } 638e8420a8eSCyril Hrubis 639e8420a8eSCyril Hrubis return nr_pages; 640e8420a8eSCyril Hrubis } 641e8420a8eSCyril Hrubis 6421da177e4SLinus Torvalds void __vma_link_rb(struct mm_struct *mm, struct vm_area_struct *vma, 6431da177e4SLinus Torvalds struct rb_node **rb_link, struct rb_node *rb_parent) 6441da177e4SLinus Torvalds { 645d3737187SMichel Lespinasse /* Update tracking information for the gap following the new vma. */ 646d3737187SMichel Lespinasse if (vma->vm_next) 647d3737187SMichel Lespinasse vma_gap_update(vma->vm_next); 648d3737187SMichel Lespinasse else 6491be7107fSHugh Dickins mm->highest_vm_end = vm_end_gap(vma); 650d3737187SMichel Lespinasse 651d3737187SMichel Lespinasse /* 652d3737187SMichel Lespinasse * vma->vm_prev wasn't known when we followed the rbtree to find the 653d3737187SMichel Lespinasse * correct insertion point for that vma. As a result, we could not 654d3737187SMichel Lespinasse * update the vma vm_rb parents rb_subtree_gap values on the way down. 655d3737187SMichel Lespinasse * So, we first insert the vma with a zero rb_subtree_gap value 656d3737187SMichel Lespinasse * (to be consistent with what we did on the way down), and then 657d3737187SMichel Lespinasse * immediately update the gap to the correct value. Finally we 658d3737187SMichel Lespinasse * rebalance the rbtree after all augmented values have been set. 659d3737187SMichel Lespinasse */ 6601da177e4SLinus Torvalds rb_link_node(&vma->vm_rb, rb_parent, rb_link); 661d3737187SMichel Lespinasse vma->rb_subtree_gap = 0; 662d3737187SMichel Lespinasse vma_gap_update(vma); 663d3737187SMichel Lespinasse vma_rb_insert(vma, &mm->mm_rb); 6641da177e4SLinus Torvalds } 6651da177e4SLinus Torvalds 666cb8f488cSDenys Vlasenko static void __vma_link_file(struct vm_area_struct *vma) 6671da177e4SLinus Torvalds { 6681da177e4SLinus Torvalds struct file *file; 6691da177e4SLinus Torvalds 6701da177e4SLinus Torvalds file = vma->vm_file; 6711da177e4SLinus Torvalds if (file) { 6721da177e4SLinus Torvalds struct address_space *mapping = file->f_mapping; 6731da177e4SLinus Torvalds 6741da177e4SLinus Torvalds if (vma->vm_flags & VM_SHARED) 675cf508b58SMiaohe Lin mapping_allow_writable(mapping); 6761da177e4SLinus Torvalds 6771da177e4SLinus Torvalds flush_dcache_mmap_lock(mapping); 6786b2dbba8SMichel Lespinasse vma_interval_tree_insert(vma, &mapping->i_mmap); 6791da177e4SLinus Torvalds flush_dcache_mmap_unlock(mapping); 6801da177e4SLinus Torvalds } 6811da177e4SLinus Torvalds } 6821da177e4SLinus Torvalds 6831da177e4SLinus Torvalds static void 6841da177e4SLinus Torvalds __vma_link(struct mm_struct *mm, struct vm_area_struct *vma, 6851da177e4SLinus Torvalds struct vm_area_struct *prev, struct rb_node **rb_link, 6861da177e4SLinus Torvalds struct rb_node *rb_parent) 6871da177e4SLinus Torvalds { 688aba6dfb7SWei Yang __vma_link_list(mm, vma, prev); 6891da177e4SLinus Torvalds __vma_link_rb(mm, vma, rb_link, rb_parent); 6901da177e4SLinus Torvalds } 6911da177e4SLinus Torvalds 6921da177e4SLinus Torvalds static void vma_link(struct mm_struct *mm, struct vm_area_struct *vma, 6931da177e4SLinus Torvalds struct vm_area_struct *prev, struct rb_node **rb_link, 6941da177e4SLinus Torvalds struct rb_node *rb_parent) 6951da177e4SLinus Torvalds { 6961da177e4SLinus Torvalds struct address_space *mapping = NULL; 6971da177e4SLinus Torvalds 69864ac4940SHuang Shijie if (vma->vm_file) { 6991da177e4SLinus Torvalds mapping = vma->vm_file->f_mapping; 70083cde9e8SDavidlohr Bueso i_mmap_lock_write(mapping); 70164ac4940SHuang Shijie } 7021da177e4SLinus Torvalds 7031da177e4SLinus Torvalds __vma_link(mm, vma, prev, rb_link, rb_parent); 7041da177e4SLinus Torvalds __vma_link_file(vma); 7051da177e4SLinus Torvalds 7061da177e4SLinus Torvalds if (mapping) 70783cde9e8SDavidlohr Bueso i_mmap_unlock_write(mapping); 7081da177e4SLinus Torvalds 7091da177e4SLinus Torvalds mm->map_count++; 7101da177e4SLinus Torvalds validate_mm(mm); 7111da177e4SLinus Torvalds } 7121da177e4SLinus Torvalds 7131da177e4SLinus Torvalds /* 71488f6b4c3SKautuk Consul * Helper for vma_adjust() in the split_vma insert case: insert a vma into the 7156b2dbba8SMichel Lespinasse * mm's list and rbtree. It has already been inserted into the interval tree. 7161da177e4SLinus Torvalds */ 71748aae425SZhenwenXu static void __insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma) 7181da177e4SLinus Torvalds { 7196597d783SHugh Dickins struct vm_area_struct *prev; 7201da177e4SLinus Torvalds struct rb_node **rb_link, *rb_parent; 7211da177e4SLinus Torvalds 7226597d783SHugh Dickins if (find_vma_links(mm, vma->vm_start, vma->vm_end, 7236597d783SHugh Dickins &prev, &rb_link, &rb_parent)) 7246597d783SHugh Dickins BUG(); 7251da177e4SLinus Torvalds __vma_link(mm, vma, prev, rb_link, rb_parent); 7261da177e4SLinus Torvalds mm->map_count++; 7271da177e4SLinus Torvalds } 7281da177e4SLinus Torvalds 7297c61f917SWei Yang static __always_inline void __vma_unlink(struct mm_struct *mm, 730e86f15eeSAndrea Arcangeli struct vm_area_struct *vma, 7318f26e0b1SAndrea Arcangeli struct vm_area_struct *ignore) 7321da177e4SLinus Torvalds { 7338f26e0b1SAndrea Arcangeli vma_rb_erase_ignore(vma, &mm->mm_rb, ignore); 7341b9fc5b2SWei Yang __vma_unlink_list(mm, vma); 735615d6e87SDavidlohr Bueso /* Kill the cache */ 736615d6e87SDavidlohr Bueso vmacache_invalidate(mm); 7371da177e4SLinus Torvalds } 7381da177e4SLinus Torvalds 7391da177e4SLinus Torvalds /* 7401da177e4SLinus Torvalds * We cannot adjust vm_start, vm_end, vm_pgoff fields of a vma that 7411da177e4SLinus Torvalds * is already present in an i_mmap tree without adjusting the tree. 7421da177e4SLinus Torvalds * The following helper function should be used when such adjustments 7431da177e4SLinus Torvalds * are necessary. The "insert" vma (if any) is to be inserted 7441da177e4SLinus Torvalds * before we drop the necessary locks. 7451da177e4SLinus Torvalds */ 746e86f15eeSAndrea Arcangeli int __vma_adjust(struct vm_area_struct *vma, unsigned long start, 747e86f15eeSAndrea Arcangeli unsigned long end, pgoff_t pgoff, struct vm_area_struct *insert, 748e86f15eeSAndrea Arcangeli struct vm_area_struct *expand) 7491da177e4SLinus Torvalds { 7501da177e4SLinus Torvalds struct mm_struct *mm = vma->vm_mm; 751e86f15eeSAndrea Arcangeli struct vm_area_struct *next = vma->vm_next, *orig_vma = vma; 7521da177e4SLinus Torvalds struct address_space *mapping = NULL; 753f808c13fSDavidlohr Bueso struct rb_root_cached *root = NULL; 754012f1800SRik van Riel struct anon_vma *anon_vma = NULL; 7551da177e4SLinus Torvalds struct file *file = vma->vm_file; 756d3737187SMichel Lespinasse bool start_changed = false, end_changed = false; 7571da177e4SLinus Torvalds long adjust_next = 0; 7581da177e4SLinus Torvalds int remove_next = 0; 7591da177e4SLinus Torvalds 7601da177e4SLinus Torvalds if (next && !insert) { 761734537c9SKirill A. Shutemov struct vm_area_struct *exporter = NULL, *importer = NULL; 762287d97acSLinus Torvalds 7631da177e4SLinus Torvalds if (end >= next->vm_end) { 7641da177e4SLinus Torvalds /* 7651da177e4SLinus Torvalds * vma expands, overlapping all the next, and 7661da177e4SLinus Torvalds * perhaps the one after too (mprotect case 6). 76786d12e47SAndrea Arcangeli * The only other cases that gets here are 768e86f15eeSAndrea Arcangeli * case 1, case 7 and case 8. 769e86f15eeSAndrea Arcangeli */ 770e86f15eeSAndrea Arcangeli if (next == expand) { 771e86f15eeSAndrea Arcangeli /* 772e86f15eeSAndrea Arcangeli * The only case where we don't expand "vma" 773e86f15eeSAndrea Arcangeli * and we expand "next" instead is case 8. 774e86f15eeSAndrea Arcangeli */ 775e86f15eeSAndrea Arcangeli VM_WARN_ON(end != next->vm_end); 776e86f15eeSAndrea Arcangeli /* 777e86f15eeSAndrea Arcangeli * remove_next == 3 means we're 778e86f15eeSAndrea Arcangeli * removing "vma" and that to do so we 779e86f15eeSAndrea Arcangeli * swapped "vma" and "next". 780e86f15eeSAndrea Arcangeli */ 781e86f15eeSAndrea Arcangeli remove_next = 3; 782e86f15eeSAndrea Arcangeli VM_WARN_ON(file != next->vm_file); 783e86f15eeSAndrea Arcangeli swap(vma, next); 784e86f15eeSAndrea Arcangeli } else { 785e86f15eeSAndrea Arcangeli VM_WARN_ON(expand != vma); 786e86f15eeSAndrea Arcangeli /* 787e86f15eeSAndrea Arcangeli * case 1, 6, 7, remove_next == 2 is case 6, 788e86f15eeSAndrea Arcangeli * remove_next == 1 is case 1 or 7. 7891da177e4SLinus Torvalds */ 790734537c9SKirill A. Shutemov remove_next = 1 + (end > next->vm_end); 791e86f15eeSAndrea Arcangeli VM_WARN_ON(remove_next == 2 && 792e86f15eeSAndrea Arcangeli end != next->vm_next->vm_end); 793e86f15eeSAndrea Arcangeli /* trim end to next, for case 6 first pass */ 7941da177e4SLinus Torvalds end = next->vm_end; 795e86f15eeSAndrea Arcangeli } 796e86f15eeSAndrea Arcangeli 797287d97acSLinus Torvalds exporter = next; 7981da177e4SLinus Torvalds importer = vma; 799734537c9SKirill A. Shutemov 800734537c9SKirill A. Shutemov /* 801734537c9SKirill A. Shutemov * If next doesn't have anon_vma, import from vma after 802734537c9SKirill A. Shutemov * next, if the vma overlaps with it. 803734537c9SKirill A. Shutemov */ 80497a42cd4SAndrea Arcangeli if (remove_next == 2 && !next->anon_vma) 805734537c9SKirill A. Shutemov exporter = next->vm_next; 806734537c9SKirill A. Shutemov 8071da177e4SLinus Torvalds } else if (end > next->vm_start) { 8081da177e4SLinus Torvalds /* 8091da177e4SLinus Torvalds * vma expands, overlapping part of the next: 8101da177e4SLinus Torvalds * mprotect case 5 shifting the boundary up. 8111da177e4SLinus Torvalds */ 812f9d86a60SWei Yang adjust_next = (end - next->vm_start); 813287d97acSLinus Torvalds exporter = next; 8141da177e4SLinus Torvalds importer = vma; 815e86f15eeSAndrea Arcangeli VM_WARN_ON(expand != importer); 8161da177e4SLinus Torvalds } else if (end < vma->vm_end) { 8171da177e4SLinus Torvalds /* 8181da177e4SLinus Torvalds * vma shrinks, and !insert tells it's not 8191da177e4SLinus Torvalds * split_vma inserting another: so it must be 8201da177e4SLinus Torvalds * mprotect case 4 shifting the boundary down. 8211da177e4SLinus Torvalds */ 822f9d86a60SWei Yang adjust_next = -(vma->vm_end - end); 823287d97acSLinus Torvalds exporter = vma; 8241da177e4SLinus Torvalds importer = next; 825e86f15eeSAndrea Arcangeli VM_WARN_ON(expand != importer); 8261da177e4SLinus Torvalds } 8271da177e4SLinus Torvalds 8285beb4930SRik van Riel /* 8295beb4930SRik van Riel * Easily overlooked: when mprotect shifts the boundary, 8305beb4930SRik van Riel * make sure the expanding vma has anon_vma set if the 8315beb4930SRik van Riel * shrinking vma had, to cover any anon pages imported. 8325beb4930SRik van Riel */ 833287d97acSLinus Torvalds if (exporter && exporter->anon_vma && !importer->anon_vma) { 834c4ea95d7SDaniel Forrest int error; 835c4ea95d7SDaniel Forrest 836287d97acSLinus Torvalds importer->anon_vma = exporter->anon_vma; 837b800c91aSKonstantin Khlebnikov error = anon_vma_clone(importer, exporter); 8383fe89b3eSLeon Yu if (error) 839b800c91aSKonstantin Khlebnikov return error; 840b800c91aSKonstantin Khlebnikov } 8415beb4930SRik van Riel } 842734537c9SKirill A. Shutemov again: 843e86f15eeSAndrea Arcangeli vma_adjust_trans_huge(orig_vma, start, end, adjust_next); 84437f9f559SKirill A. Shutemov 8451da177e4SLinus Torvalds if (file) { 8461da177e4SLinus Torvalds mapping = file->f_mapping; 8471da177e4SLinus Torvalds root = &mapping->i_mmap; 848cbc91f71SSrikar Dronamraju uprobe_munmap(vma, vma->vm_start, vma->vm_end); 849682968e0SSrikar Dronamraju 850682968e0SSrikar Dronamraju if (adjust_next) 85127ba0644SKirill A. Shutemov uprobe_munmap(next, next->vm_start, next->vm_end); 852682968e0SSrikar Dronamraju 85383cde9e8SDavidlohr Bueso i_mmap_lock_write(mapping); 8541da177e4SLinus Torvalds if (insert) { 8551da177e4SLinus Torvalds /* 8566b2dbba8SMichel Lespinasse * Put into interval tree now, so instantiated pages 8571da177e4SLinus Torvalds * are visible to arm/parisc __flush_dcache_page 8581da177e4SLinus Torvalds * throughout; but we cannot insert into address 8591da177e4SLinus Torvalds * space until vma start or end is updated. 8601da177e4SLinus Torvalds */ 8611da177e4SLinus Torvalds __vma_link_file(insert); 8621da177e4SLinus Torvalds } 8631da177e4SLinus Torvalds } 8641da177e4SLinus Torvalds 865012f1800SRik van Riel anon_vma = vma->anon_vma; 866bf181b9fSMichel Lespinasse if (!anon_vma && adjust_next) 867bf181b9fSMichel Lespinasse anon_vma = next->anon_vma; 868bf181b9fSMichel Lespinasse if (anon_vma) { 869e86f15eeSAndrea Arcangeli VM_WARN_ON(adjust_next && next->anon_vma && 870e86f15eeSAndrea Arcangeli anon_vma != next->anon_vma); 8714fc3f1d6SIngo Molnar anon_vma_lock_write(anon_vma); 872bf181b9fSMichel Lespinasse anon_vma_interval_tree_pre_update_vma(vma); 873bf181b9fSMichel Lespinasse if (adjust_next) 874bf181b9fSMichel Lespinasse anon_vma_interval_tree_pre_update_vma(next); 875bf181b9fSMichel Lespinasse } 876012f1800SRik van Riel 8770fc48a6eSWei Yang if (file) { 8781da177e4SLinus Torvalds flush_dcache_mmap_lock(mapping); 8796b2dbba8SMichel Lespinasse vma_interval_tree_remove(vma, root); 8801da177e4SLinus Torvalds if (adjust_next) 8816b2dbba8SMichel Lespinasse vma_interval_tree_remove(next, root); 8821da177e4SLinus Torvalds } 8831da177e4SLinus Torvalds 884d3737187SMichel Lespinasse if (start != vma->vm_start) { 8851da177e4SLinus Torvalds vma->vm_start = start; 886d3737187SMichel Lespinasse start_changed = true; 887d3737187SMichel Lespinasse } 888d3737187SMichel Lespinasse if (end != vma->vm_end) { 8891da177e4SLinus Torvalds vma->vm_end = end; 890d3737187SMichel Lespinasse end_changed = true; 891d3737187SMichel Lespinasse } 8921da177e4SLinus Torvalds vma->vm_pgoff = pgoff; 8931da177e4SLinus Torvalds if (adjust_next) { 894f9d86a60SWei Yang next->vm_start += adjust_next; 895f9d86a60SWei Yang next->vm_pgoff += adjust_next >> PAGE_SHIFT; 8961da177e4SLinus Torvalds } 8971da177e4SLinus Torvalds 8980fc48a6eSWei Yang if (file) { 8991da177e4SLinus Torvalds if (adjust_next) 9006b2dbba8SMichel Lespinasse vma_interval_tree_insert(next, root); 9016b2dbba8SMichel Lespinasse vma_interval_tree_insert(vma, root); 9021da177e4SLinus Torvalds flush_dcache_mmap_unlock(mapping); 9031da177e4SLinus Torvalds } 9041da177e4SLinus Torvalds 9051da177e4SLinus Torvalds if (remove_next) { 9061da177e4SLinus Torvalds /* 9071da177e4SLinus Torvalds * vma_merge has merged next into vma, and needs 9081da177e4SLinus Torvalds * us to remove next before dropping the locks. 9091da177e4SLinus Torvalds */ 910e86f15eeSAndrea Arcangeli if (remove_next != 3) 9117c61f917SWei Yang __vma_unlink(mm, next, next); 912e86f15eeSAndrea Arcangeli else 9138f26e0b1SAndrea Arcangeli /* 9148f26e0b1SAndrea Arcangeli * vma is not before next if they've been 9158f26e0b1SAndrea Arcangeli * swapped. 9168f26e0b1SAndrea Arcangeli * 9178f26e0b1SAndrea Arcangeli * pre-swap() next->vm_start was reduced so 9188f26e0b1SAndrea Arcangeli * tell validate_mm_rb to ignore pre-swap() 9198f26e0b1SAndrea Arcangeli * "next" (which is stored in post-swap() 9208f26e0b1SAndrea Arcangeli * "vma"). 9218f26e0b1SAndrea Arcangeli */ 9227c61f917SWei Yang __vma_unlink(mm, next, vma); 9231da177e4SLinus Torvalds if (file) 9241da177e4SLinus Torvalds __remove_shared_vm_struct(next, file, mapping); 9251da177e4SLinus Torvalds } else if (insert) { 9261da177e4SLinus Torvalds /* 9271da177e4SLinus Torvalds * split_vma has split insert from vma, and needs 9281da177e4SLinus Torvalds * us to insert it before dropping the locks 9291da177e4SLinus Torvalds * (it may either follow vma or precede it). 9301da177e4SLinus Torvalds */ 9311da177e4SLinus Torvalds __insert_vm_struct(mm, insert); 932d3737187SMichel Lespinasse } else { 933d3737187SMichel Lespinasse if (start_changed) 934d3737187SMichel Lespinasse vma_gap_update(vma); 935d3737187SMichel Lespinasse if (end_changed) { 936d3737187SMichel Lespinasse if (!next) 9371be7107fSHugh Dickins mm->highest_vm_end = vm_end_gap(vma); 938d3737187SMichel Lespinasse else if (!adjust_next) 939d3737187SMichel Lespinasse vma_gap_update(next); 940d3737187SMichel Lespinasse } 9411da177e4SLinus Torvalds } 9421da177e4SLinus Torvalds 943bf181b9fSMichel Lespinasse if (anon_vma) { 944bf181b9fSMichel Lespinasse anon_vma_interval_tree_post_update_vma(vma); 945bf181b9fSMichel Lespinasse if (adjust_next) 946bf181b9fSMichel Lespinasse anon_vma_interval_tree_post_update_vma(next); 94708b52706SKonstantin Khlebnikov anon_vma_unlock_write(anon_vma); 948bf181b9fSMichel Lespinasse } 9491da177e4SLinus Torvalds 9500fc48a6eSWei Yang if (file) { 951808fbdbeSWei Yang i_mmap_unlock_write(mapping); 9527b2d81d4SIngo Molnar uprobe_mmap(vma); 9532b144498SSrikar Dronamraju 9542b144498SSrikar Dronamraju if (adjust_next) 9557b2d81d4SIngo Molnar uprobe_mmap(next); 9562b144498SSrikar Dronamraju } 9572b144498SSrikar Dronamraju 9581da177e4SLinus Torvalds if (remove_next) { 959925d1c40SMatt Helsley if (file) { 960cbc91f71SSrikar Dronamraju uprobe_munmap(next, next->vm_start, next->vm_end); 9611da177e4SLinus Torvalds fput(file); 962925d1c40SMatt Helsley } 9635beb4930SRik van Riel if (next->anon_vma) 9645beb4930SRik van Riel anon_vma_merge(vma, next); 9651da177e4SLinus Torvalds mm->map_count--; 9663964acd0SOleg Nesterov mpol_put(vma_policy(next)); 9673928d4f5SLinus Torvalds vm_area_free(next); 9681da177e4SLinus Torvalds /* 9691da177e4SLinus Torvalds * In mprotect's case 6 (see comments on vma_merge), 9701da177e4SLinus Torvalds * we must remove another next too. It would clutter 9711da177e4SLinus Torvalds * up the code too much to do both in one go. 9721da177e4SLinus Torvalds */ 973e86f15eeSAndrea Arcangeli if (remove_next != 3) { 974e86f15eeSAndrea Arcangeli /* 975e86f15eeSAndrea Arcangeli * If "next" was removed and vma->vm_end was 976e86f15eeSAndrea Arcangeli * expanded (up) over it, in turn 977e86f15eeSAndrea Arcangeli * "next->vm_prev->vm_end" changed and the 978e86f15eeSAndrea Arcangeli * "vma->vm_next" gap must be updated. 979e86f15eeSAndrea Arcangeli */ 9801da177e4SLinus Torvalds next = vma->vm_next; 981e86f15eeSAndrea Arcangeli } else { 982e86f15eeSAndrea Arcangeli /* 983e86f15eeSAndrea Arcangeli * For the scope of the comment "next" and 984e86f15eeSAndrea Arcangeli * "vma" considered pre-swap(): if "vma" was 985e86f15eeSAndrea Arcangeli * removed, next->vm_start was expanded (down) 986e86f15eeSAndrea Arcangeli * over it and the "next" gap must be updated. 987e86f15eeSAndrea Arcangeli * Because of the swap() the post-swap() "vma" 988e86f15eeSAndrea Arcangeli * actually points to pre-swap() "next" 989e86f15eeSAndrea Arcangeli * (post-swap() "next" as opposed is now a 990e86f15eeSAndrea Arcangeli * dangling pointer). 991e86f15eeSAndrea Arcangeli */ 992e86f15eeSAndrea Arcangeli next = vma; 993e86f15eeSAndrea Arcangeli } 994734537c9SKirill A. Shutemov if (remove_next == 2) { 995734537c9SKirill A. Shutemov remove_next = 1; 996734537c9SKirill A. Shutemov end = next->vm_end; 9971da177e4SLinus Torvalds goto again; 998734537c9SKirill A. Shutemov } 999d3737187SMichel Lespinasse else if (next) 1000d3737187SMichel Lespinasse vma_gap_update(next); 1001fb8c41e9SAndrea Arcangeli else { 1002fb8c41e9SAndrea Arcangeli /* 1003fb8c41e9SAndrea Arcangeli * If remove_next == 2 we obviously can't 1004fb8c41e9SAndrea Arcangeli * reach this path. 1005fb8c41e9SAndrea Arcangeli * 1006fb8c41e9SAndrea Arcangeli * If remove_next == 3 we can't reach this 1007fb8c41e9SAndrea Arcangeli * path because pre-swap() next is always not 1008fb8c41e9SAndrea Arcangeli * NULL. pre-swap() "next" is not being 1009fb8c41e9SAndrea Arcangeli * removed and its next->vm_end is not altered 1010fb8c41e9SAndrea Arcangeli * (and furthermore "end" already matches 1011fb8c41e9SAndrea Arcangeli * next->vm_end in remove_next == 3). 1012fb8c41e9SAndrea Arcangeli * 1013fb8c41e9SAndrea Arcangeli * We reach this only in the remove_next == 1 1014fb8c41e9SAndrea Arcangeli * case if the "next" vma that was removed was 1015fb8c41e9SAndrea Arcangeli * the highest vma of the mm. However in such 1016fb8c41e9SAndrea Arcangeli * case next->vm_end == "end" and the extended 1017fb8c41e9SAndrea Arcangeli * "vma" has vma->vm_end == next->vm_end so 1018fb8c41e9SAndrea Arcangeli * mm->highest_vm_end doesn't need any update 1019fb8c41e9SAndrea Arcangeli * in remove_next == 1 case. 1020fb8c41e9SAndrea Arcangeli */ 10211be7107fSHugh Dickins VM_WARN_ON(mm->highest_vm_end != vm_end_gap(vma)); 1022fb8c41e9SAndrea Arcangeli } 10231da177e4SLinus Torvalds } 10242b144498SSrikar Dronamraju if (insert && file) 10257b2d81d4SIngo Molnar uprobe_mmap(insert); 10261da177e4SLinus Torvalds 10271da177e4SLinus Torvalds validate_mm(mm); 10285beb4930SRik van Riel 10295beb4930SRik van Riel return 0; 10301da177e4SLinus Torvalds } 10311da177e4SLinus Torvalds 10321da177e4SLinus Torvalds /* 10331da177e4SLinus Torvalds * If the vma has a ->close operation then the driver probably needs to release 10341da177e4SLinus Torvalds * per-vma resources, so we don't attempt to merge those. 10351da177e4SLinus Torvalds */ 10361da177e4SLinus Torvalds static inline int is_mergeable_vma(struct vm_area_struct *vma, 103719a809afSAndrea Arcangeli struct file *file, unsigned long vm_flags, 10389a10064fSColin Cross struct vm_userfaultfd_ctx vm_userfaultfd_ctx, 10395c26f6acSSuren Baghdasaryan struct anon_vma_name *anon_name) 10401da177e4SLinus Torvalds { 104134228d47SCyrill Gorcunov /* 104234228d47SCyrill Gorcunov * VM_SOFTDIRTY should not prevent from VMA merging, if we 104334228d47SCyrill Gorcunov * match the flags but dirty bit -- the caller should mark 104434228d47SCyrill Gorcunov * merged VMA as dirty. If dirty bit won't be excluded from 10458bb4e7a2SWei Yang * comparison, we increase pressure on the memory system forcing 104634228d47SCyrill Gorcunov * the kernel to generate new VMAs when old one could be 104734228d47SCyrill Gorcunov * extended instead. 104834228d47SCyrill Gorcunov */ 104934228d47SCyrill Gorcunov if ((vma->vm_flags ^ vm_flags) & ~VM_SOFTDIRTY) 10501da177e4SLinus Torvalds return 0; 10511da177e4SLinus Torvalds if (vma->vm_file != file) 10521da177e4SLinus Torvalds return 0; 10531da177e4SLinus Torvalds if (vma->vm_ops && vma->vm_ops->close) 10541da177e4SLinus Torvalds return 0; 105519a809afSAndrea Arcangeli if (!is_mergeable_vm_userfaultfd_ctx(vma, vm_userfaultfd_ctx)) 105619a809afSAndrea Arcangeli return 0; 10575c26f6acSSuren Baghdasaryan if (!anon_vma_name_eq(anon_vma_name(vma), anon_name)) 10589a10064fSColin Cross return 0; 10591da177e4SLinus Torvalds return 1; 10601da177e4SLinus Torvalds } 10611da177e4SLinus Torvalds 10621da177e4SLinus Torvalds static inline int is_mergeable_anon_vma(struct anon_vma *anon_vma1, 1063965f55deSShaohua Li struct anon_vma *anon_vma2, 1064965f55deSShaohua Li struct vm_area_struct *vma) 10651da177e4SLinus Torvalds { 1066965f55deSShaohua Li /* 1067965f55deSShaohua Li * The list_is_singular() test is to avoid merging VMA cloned from 1068965f55deSShaohua Li * parents. This can improve scalability caused by anon_vma lock. 1069965f55deSShaohua Li */ 1070965f55deSShaohua Li if ((!anon_vma1 || !anon_vma2) && (!vma || 1071965f55deSShaohua Li list_is_singular(&vma->anon_vma_chain))) 1072965f55deSShaohua Li return 1; 1073965f55deSShaohua Li return anon_vma1 == anon_vma2; 10741da177e4SLinus Torvalds } 10751da177e4SLinus Torvalds 10761da177e4SLinus Torvalds /* 10771da177e4SLinus Torvalds * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff) 10781da177e4SLinus Torvalds * in front of (at a lower virtual address and file offset than) the vma. 10791da177e4SLinus Torvalds * 10801da177e4SLinus Torvalds * We cannot merge two vmas if they have differently assigned (non-NULL) 10811da177e4SLinus Torvalds * anon_vmas, nor if same anon_vma is assigned but offsets incompatible. 10821da177e4SLinus Torvalds * 10831da177e4SLinus Torvalds * We don't check here for the merged mmap wrapping around the end of pagecache 108445e55300SPeter Collingbourne * indices (16TB on ia32) because do_mmap() does not permit mmap's which 10851da177e4SLinus Torvalds * wrap, nor mmaps which cover the final page at index -1UL. 10861da177e4SLinus Torvalds */ 10871da177e4SLinus Torvalds static int 10881da177e4SLinus Torvalds can_vma_merge_before(struct vm_area_struct *vma, unsigned long vm_flags, 108919a809afSAndrea Arcangeli struct anon_vma *anon_vma, struct file *file, 109019a809afSAndrea Arcangeli pgoff_t vm_pgoff, 10919a10064fSColin Cross struct vm_userfaultfd_ctx vm_userfaultfd_ctx, 10925c26f6acSSuren Baghdasaryan struct anon_vma_name *anon_name) 10931da177e4SLinus Torvalds { 10949a10064fSColin Cross if (is_mergeable_vma(vma, file, vm_flags, vm_userfaultfd_ctx, anon_name) && 1095965f55deSShaohua Li is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) { 10961da177e4SLinus Torvalds if (vma->vm_pgoff == vm_pgoff) 10971da177e4SLinus Torvalds return 1; 10981da177e4SLinus Torvalds } 10991da177e4SLinus Torvalds return 0; 11001da177e4SLinus Torvalds } 11011da177e4SLinus Torvalds 11021da177e4SLinus Torvalds /* 11031da177e4SLinus Torvalds * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff) 11041da177e4SLinus Torvalds * beyond (at a higher virtual address and file offset than) the vma. 11051da177e4SLinus Torvalds * 11061da177e4SLinus Torvalds * We cannot merge two vmas if they have differently assigned (non-NULL) 11071da177e4SLinus Torvalds * anon_vmas, nor if same anon_vma is assigned but offsets incompatible. 11081da177e4SLinus Torvalds */ 11091da177e4SLinus Torvalds static int 11101da177e4SLinus Torvalds can_vma_merge_after(struct vm_area_struct *vma, unsigned long vm_flags, 111119a809afSAndrea Arcangeli struct anon_vma *anon_vma, struct file *file, 111219a809afSAndrea Arcangeli pgoff_t vm_pgoff, 11139a10064fSColin Cross struct vm_userfaultfd_ctx vm_userfaultfd_ctx, 11145c26f6acSSuren Baghdasaryan struct anon_vma_name *anon_name) 11151da177e4SLinus Torvalds { 11169a10064fSColin Cross if (is_mergeable_vma(vma, file, vm_flags, vm_userfaultfd_ctx, anon_name) && 1117965f55deSShaohua Li is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) { 11181da177e4SLinus Torvalds pgoff_t vm_pglen; 1119d6e93217SLibin vm_pglen = vma_pages(vma); 11201da177e4SLinus Torvalds if (vma->vm_pgoff + vm_pglen == vm_pgoff) 11211da177e4SLinus Torvalds return 1; 11221da177e4SLinus Torvalds } 11231da177e4SLinus Torvalds return 0; 11241da177e4SLinus Torvalds } 11251da177e4SLinus Torvalds 11261da177e4SLinus Torvalds /* 11279a10064fSColin Cross * Given a mapping request (addr,end,vm_flags,file,pgoff,anon_name), 11289a10064fSColin Cross * figure out whether that can be merged with its predecessor or its 11299a10064fSColin Cross * successor. Or both (it neatly fills a hole). 11301da177e4SLinus Torvalds * 11311da177e4SLinus Torvalds * In most cases - when called for mmap, brk or mremap - [addr,end) is 11321da177e4SLinus Torvalds * certain not to be mapped by the time vma_merge is called; but when 11331da177e4SLinus Torvalds * called for mprotect, it is certain to be already mapped (either at 11341da177e4SLinus Torvalds * an offset within prev, or at the start of next), and the flags of 11351da177e4SLinus Torvalds * this area are about to be changed to vm_flags - and the no-change 11361da177e4SLinus Torvalds * case has already been eliminated. 11371da177e4SLinus Torvalds * 11381da177e4SLinus Torvalds * The following mprotect cases have to be considered, where AAAA is 11391da177e4SLinus Torvalds * the area passed down from mprotect_fixup, never extending beyond one 11401da177e4SLinus Torvalds * vma, PPPPPP is the prev vma specified, and NNNNNN the next vma after: 11411da177e4SLinus Torvalds * 11425d42ab29SWei Yang * AAAA AAAA AAAA 11435d42ab29SWei Yang * PPPPPPNNNNNN PPPPPPNNNNNN PPPPPPNNNNNN 11445d42ab29SWei Yang * cannot merge might become might become 11455d42ab29SWei Yang * PPNNNNNNNNNN PPPPPPPPPPNN 11465d42ab29SWei Yang * mmap, brk or case 4 below case 5 below 11475d42ab29SWei Yang * mremap move: 11485d42ab29SWei Yang * AAAA AAAA 11495d42ab29SWei Yang * PPPP NNNN PPPPNNNNXXXX 11505d42ab29SWei Yang * might become might become 11515d42ab29SWei Yang * PPPPPPPPPPPP 1 or PPPPPPPPPPPP 6 or 11525d42ab29SWei Yang * PPPPPPPPNNNN 2 or PPPPPPPPXXXX 7 or 11535d42ab29SWei Yang * PPPPNNNNNNNN 3 PPPPXXXXXXXX 8 11541da177e4SLinus Torvalds * 11558bb4e7a2SWei Yang * It is important for case 8 that the vma NNNN overlapping the 1156e86f15eeSAndrea Arcangeli * region AAAA is never going to extended over XXXX. Instead XXXX must 1157e86f15eeSAndrea Arcangeli * be extended in region AAAA and NNNN must be removed. This way in 1158e86f15eeSAndrea Arcangeli * all cases where vma_merge succeeds, the moment vma_adjust drops the 1159e86f15eeSAndrea Arcangeli * rmap_locks, the properties of the merged vma will be already 1160e86f15eeSAndrea Arcangeli * correct for the whole merged range. Some of those properties like 1161e86f15eeSAndrea Arcangeli * vm_page_prot/vm_flags may be accessed by rmap_walks and they must 1162e86f15eeSAndrea Arcangeli * be correct for the whole merged range immediately after the 1163e86f15eeSAndrea Arcangeli * rmap_locks are released. Otherwise if XXXX would be removed and 1164e86f15eeSAndrea Arcangeli * NNNN would be extended over the XXXX range, remove_migration_ptes 1165e86f15eeSAndrea Arcangeli * or other rmap walkers (if working on addresses beyond the "end" 1166e86f15eeSAndrea Arcangeli * parameter) may establish ptes with the wrong permissions of NNNN 1167e86f15eeSAndrea Arcangeli * instead of the right permissions of XXXX. 11681da177e4SLinus Torvalds */ 11691da177e4SLinus Torvalds struct vm_area_struct *vma_merge(struct mm_struct *mm, 11701da177e4SLinus Torvalds struct vm_area_struct *prev, unsigned long addr, 11711da177e4SLinus Torvalds unsigned long end, unsigned long vm_flags, 11721da177e4SLinus Torvalds struct anon_vma *anon_vma, struct file *file, 117319a809afSAndrea Arcangeli pgoff_t pgoff, struct mempolicy *policy, 11749a10064fSColin Cross struct vm_userfaultfd_ctx vm_userfaultfd_ctx, 11755c26f6acSSuren Baghdasaryan struct anon_vma_name *anon_name) 11761da177e4SLinus Torvalds { 11771da177e4SLinus Torvalds pgoff_t pglen = (end - addr) >> PAGE_SHIFT; 11781da177e4SLinus Torvalds struct vm_area_struct *area, *next; 11795beb4930SRik van Riel int err; 11801da177e4SLinus Torvalds 11811da177e4SLinus Torvalds /* 11821da177e4SLinus Torvalds * We later require that vma->vm_flags == vm_flags, 11831da177e4SLinus Torvalds * so this tests vma->vm_flags & VM_SPECIAL, too. 11841da177e4SLinus Torvalds */ 11851da177e4SLinus Torvalds if (vm_flags & VM_SPECIAL) 11861da177e4SLinus Torvalds return NULL; 11871da177e4SLinus Torvalds 11883903b55aSLiam R. Howlett next = vma_next(mm, prev); 11891da177e4SLinus Torvalds area = next; 1190e86f15eeSAndrea Arcangeli if (area && area->vm_end == end) /* cases 6, 7, 8 */ 11911da177e4SLinus Torvalds next = next->vm_next; 11921da177e4SLinus Torvalds 1193e86f15eeSAndrea Arcangeli /* verify some invariant that must be enforced by the caller */ 1194e86f15eeSAndrea Arcangeli VM_WARN_ON(prev && addr <= prev->vm_start); 1195e86f15eeSAndrea Arcangeli VM_WARN_ON(area && end > area->vm_end); 1196e86f15eeSAndrea Arcangeli VM_WARN_ON(addr >= end); 1197e86f15eeSAndrea Arcangeli 11981da177e4SLinus Torvalds /* 11991da177e4SLinus Torvalds * Can it merge with the predecessor? 12001da177e4SLinus Torvalds */ 12011da177e4SLinus Torvalds if (prev && prev->vm_end == addr && 12021da177e4SLinus Torvalds mpol_equal(vma_policy(prev), policy) && 12031da177e4SLinus Torvalds can_vma_merge_after(prev, vm_flags, 120419a809afSAndrea Arcangeli anon_vma, file, pgoff, 12059a10064fSColin Cross vm_userfaultfd_ctx, anon_name)) { 12061da177e4SLinus Torvalds /* 12071da177e4SLinus Torvalds * OK, it can. Can we now merge in the successor as well? 12081da177e4SLinus Torvalds */ 12091da177e4SLinus Torvalds if (next && end == next->vm_start && 12101da177e4SLinus Torvalds mpol_equal(policy, vma_policy(next)) && 12111da177e4SLinus Torvalds can_vma_merge_before(next, vm_flags, 121219a809afSAndrea Arcangeli anon_vma, file, 121319a809afSAndrea Arcangeli pgoff+pglen, 12149a10064fSColin Cross vm_userfaultfd_ctx, anon_name) && 12151da177e4SLinus Torvalds is_mergeable_anon_vma(prev->anon_vma, 1216965f55deSShaohua Li next->anon_vma, NULL)) { 12171da177e4SLinus Torvalds /* cases 1, 6 */ 1218e86f15eeSAndrea Arcangeli err = __vma_adjust(prev, prev->vm_start, 1219e86f15eeSAndrea Arcangeli next->vm_end, prev->vm_pgoff, NULL, 1220e86f15eeSAndrea Arcangeli prev); 12211da177e4SLinus Torvalds } else /* cases 2, 5, 7 */ 1222e86f15eeSAndrea Arcangeli err = __vma_adjust(prev, prev->vm_start, 1223e86f15eeSAndrea Arcangeli end, prev->vm_pgoff, NULL, prev); 12245beb4930SRik van Riel if (err) 12255beb4930SRik van Riel return NULL; 1226c791576cSYang Shi khugepaged_enter_vma(prev, vm_flags); 12271da177e4SLinus Torvalds return prev; 12281da177e4SLinus Torvalds } 12291da177e4SLinus Torvalds 12301da177e4SLinus Torvalds /* 12311da177e4SLinus Torvalds * Can this new request be merged in front of next? 12321da177e4SLinus Torvalds */ 12331da177e4SLinus Torvalds if (next && end == next->vm_start && 12341da177e4SLinus Torvalds mpol_equal(policy, vma_policy(next)) && 12351da177e4SLinus Torvalds can_vma_merge_before(next, vm_flags, 123619a809afSAndrea Arcangeli anon_vma, file, pgoff+pglen, 12379a10064fSColin Cross vm_userfaultfd_ctx, anon_name)) { 12381da177e4SLinus Torvalds if (prev && addr < prev->vm_end) /* case 4 */ 1239e86f15eeSAndrea Arcangeli err = __vma_adjust(prev, prev->vm_start, 1240e86f15eeSAndrea Arcangeli addr, prev->vm_pgoff, NULL, next); 1241e86f15eeSAndrea Arcangeli else { /* cases 3, 8 */ 1242e86f15eeSAndrea Arcangeli err = __vma_adjust(area, addr, next->vm_end, 1243e86f15eeSAndrea Arcangeli next->vm_pgoff - pglen, NULL, next); 1244e86f15eeSAndrea Arcangeli /* 1245e86f15eeSAndrea Arcangeli * In case 3 area is already equal to next and 1246e86f15eeSAndrea Arcangeli * this is a noop, but in case 8 "area" has 1247e86f15eeSAndrea Arcangeli * been removed and next was expanded over it. 1248e86f15eeSAndrea Arcangeli */ 1249e86f15eeSAndrea Arcangeli area = next; 1250e86f15eeSAndrea Arcangeli } 12515beb4930SRik van Riel if (err) 12525beb4930SRik van Riel return NULL; 1253c791576cSYang Shi khugepaged_enter_vma(area, vm_flags); 12541da177e4SLinus Torvalds return area; 12551da177e4SLinus Torvalds } 12561da177e4SLinus Torvalds 12571da177e4SLinus Torvalds return NULL; 12581da177e4SLinus Torvalds } 12591da177e4SLinus Torvalds 12601da177e4SLinus Torvalds /* 1261b4f315b4SEthon Paul * Rough compatibility check to quickly see if it's even worth looking 1262d0e9fe17SLinus Torvalds * at sharing an anon_vma. 1263d0e9fe17SLinus Torvalds * 1264d0e9fe17SLinus Torvalds * They need to have the same vm_file, and the flags can only differ 1265d0e9fe17SLinus Torvalds * in things that mprotect may change. 1266d0e9fe17SLinus Torvalds * 1267d0e9fe17SLinus Torvalds * NOTE! The fact that we share an anon_vma doesn't _have_ to mean that 1268d0e9fe17SLinus Torvalds * we can merge the two vma's. For example, we refuse to merge a vma if 1269d0e9fe17SLinus Torvalds * there is a vm_ops->close() function, because that indicates that the 1270d0e9fe17SLinus Torvalds * driver is doing some kind of reference counting. But that doesn't 1271d0e9fe17SLinus Torvalds * really matter for the anon_vma sharing case. 1272d0e9fe17SLinus Torvalds */ 1273d0e9fe17SLinus Torvalds static int anon_vma_compatible(struct vm_area_struct *a, struct vm_area_struct *b) 1274d0e9fe17SLinus Torvalds { 1275d0e9fe17SLinus Torvalds return a->vm_end == b->vm_start && 1276d0e9fe17SLinus Torvalds mpol_equal(vma_policy(a), vma_policy(b)) && 1277d0e9fe17SLinus Torvalds a->vm_file == b->vm_file && 12786cb4d9a2SAnshuman Khandual !((a->vm_flags ^ b->vm_flags) & ~(VM_ACCESS_FLAGS | VM_SOFTDIRTY)) && 1279d0e9fe17SLinus Torvalds b->vm_pgoff == a->vm_pgoff + ((b->vm_start - a->vm_start) >> PAGE_SHIFT); 1280d0e9fe17SLinus Torvalds } 1281d0e9fe17SLinus Torvalds 1282d0e9fe17SLinus Torvalds /* 1283d0e9fe17SLinus Torvalds * Do some basic sanity checking to see if we can re-use the anon_vma 1284d0e9fe17SLinus Torvalds * from 'old'. The 'a'/'b' vma's are in VM order - one of them will be 1285d0e9fe17SLinus Torvalds * the same as 'old', the other will be the new one that is trying 1286d0e9fe17SLinus Torvalds * to share the anon_vma. 1287d0e9fe17SLinus Torvalds * 12885b449489SFlorian Rommel * NOTE! This runs with mmap_lock held for reading, so it is possible that 1289d0e9fe17SLinus Torvalds * the anon_vma of 'old' is concurrently in the process of being set up 1290d0e9fe17SLinus Torvalds * by another page fault trying to merge _that_. But that's ok: if it 1291d0e9fe17SLinus Torvalds * is being set up, that automatically means that it will be a singleton 1292d0e9fe17SLinus Torvalds * acceptable for merging, so we can do all of this optimistically. But 12934db0c3c2SJason Low * we do that READ_ONCE() to make sure that we never re-load the pointer. 1294d0e9fe17SLinus Torvalds * 1295d0e9fe17SLinus Torvalds * IOW: that the "list_is_singular()" test on the anon_vma_chain only 1296d0e9fe17SLinus Torvalds * matters for the 'stable anon_vma' case (ie the thing we want to avoid 1297d0e9fe17SLinus Torvalds * is to return an anon_vma that is "complex" due to having gone through 1298d0e9fe17SLinus Torvalds * a fork). 1299d0e9fe17SLinus Torvalds * 1300d0e9fe17SLinus Torvalds * We also make sure that the two vma's are compatible (adjacent, 1301d0e9fe17SLinus Torvalds * and with the same memory policies). That's all stable, even with just 13025b449489SFlorian Rommel * a read lock on the mmap_lock. 1303d0e9fe17SLinus Torvalds */ 1304d0e9fe17SLinus Torvalds static struct anon_vma *reusable_anon_vma(struct vm_area_struct *old, struct vm_area_struct *a, struct vm_area_struct *b) 1305d0e9fe17SLinus Torvalds { 1306d0e9fe17SLinus Torvalds if (anon_vma_compatible(a, b)) { 13074db0c3c2SJason Low struct anon_vma *anon_vma = READ_ONCE(old->anon_vma); 1308d0e9fe17SLinus Torvalds 1309d0e9fe17SLinus Torvalds if (anon_vma && list_is_singular(&old->anon_vma_chain)) 1310d0e9fe17SLinus Torvalds return anon_vma; 1311d0e9fe17SLinus Torvalds } 1312d0e9fe17SLinus Torvalds return NULL; 1313d0e9fe17SLinus Torvalds } 1314d0e9fe17SLinus Torvalds 1315d0e9fe17SLinus Torvalds /* 13161da177e4SLinus Torvalds * find_mergeable_anon_vma is used by anon_vma_prepare, to check 13171da177e4SLinus Torvalds * neighbouring vmas for a suitable anon_vma, before it goes off 13181da177e4SLinus Torvalds * to allocate a new anon_vma. It checks because a repetitive 13191da177e4SLinus Torvalds * sequence of mprotects and faults may otherwise lead to distinct 13201da177e4SLinus Torvalds * anon_vmas being allocated, preventing vma merge in subsequent 13211da177e4SLinus Torvalds * mprotect. 13221da177e4SLinus Torvalds */ 13231da177e4SLinus Torvalds struct anon_vma *find_mergeable_anon_vma(struct vm_area_struct *vma) 13241da177e4SLinus Torvalds { 1325a67c8caaSMiaohe Lin struct anon_vma *anon_vma = NULL; 13261da177e4SLinus Torvalds 1327a67c8caaSMiaohe Lin /* Try next first. */ 1328a67c8caaSMiaohe Lin if (vma->vm_next) { 1329a67c8caaSMiaohe Lin anon_vma = reusable_anon_vma(vma->vm_next, vma, vma->vm_next); 1330d0e9fe17SLinus Torvalds if (anon_vma) 1331d0e9fe17SLinus Torvalds return anon_vma; 1332a67c8caaSMiaohe Lin } 13331da177e4SLinus Torvalds 1334a67c8caaSMiaohe Lin /* Try prev next. */ 1335a67c8caaSMiaohe Lin if (vma->vm_prev) 1336a67c8caaSMiaohe Lin anon_vma = reusable_anon_vma(vma->vm_prev, vma->vm_prev, vma); 1337a67c8caaSMiaohe Lin 13381da177e4SLinus Torvalds /* 1339a67c8caaSMiaohe Lin * We might reach here with anon_vma == NULL if we can't find 1340a67c8caaSMiaohe Lin * any reusable anon_vma. 13411da177e4SLinus Torvalds * There's no absolute need to look only at touching neighbours: 13421da177e4SLinus Torvalds * we could search further afield for "compatible" anon_vmas. 13431da177e4SLinus Torvalds * But it would probably just be a waste of time searching, 13441da177e4SLinus Torvalds * or lead to too many vmas hanging off the same anon_vma. 13451da177e4SLinus Torvalds * We're trying to allow mprotect remerging later on, 13461da177e4SLinus Torvalds * not trying to minimize memory used for anon_vmas. 13471da177e4SLinus Torvalds */ 1348a67c8caaSMiaohe Lin return anon_vma; 13491da177e4SLinus Torvalds } 13501da177e4SLinus Torvalds 13511da177e4SLinus Torvalds /* 135240401530SAl Viro * If a hint addr is less than mmap_min_addr change hint to be as 135340401530SAl Viro * low as possible but still greater than mmap_min_addr 135440401530SAl Viro */ 135540401530SAl Viro static inline unsigned long round_hint_to_min(unsigned long hint) 135640401530SAl Viro { 135740401530SAl Viro hint &= PAGE_MASK; 135840401530SAl Viro if (((void *)hint != NULL) && 135940401530SAl Viro (hint < mmap_min_addr)) 136040401530SAl Viro return PAGE_ALIGN(mmap_min_addr); 136140401530SAl Viro return hint; 136240401530SAl Viro } 136340401530SAl Viro 13646aeb2542SMike Rapoport int mlock_future_check(struct mm_struct *mm, unsigned long flags, 1365363ee17fSDavidlohr Bueso unsigned long len) 1366363ee17fSDavidlohr Bueso { 1367363ee17fSDavidlohr Bueso unsigned long locked, lock_limit; 1368363ee17fSDavidlohr Bueso 1369363ee17fSDavidlohr Bueso /* mlock MCL_FUTURE? */ 1370363ee17fSDavidlohr Bueso if (flags & VM_LOCKED) { 1371363ee17fSDavidlohr Bueso locked = len >> PAGE_SHIFT; 1372363ee17fSDavidlohr Bueso locked += mm->locked_vm; 1373363ee17fSDavidlohr Bueso lock_limit = rlimit(RLIMIT_MEMLOCK); 1374363ee17fSDavidlohr Bueso lock_limit >>= PAGE_SHIFT; 1375363ee17fSDavidlohr Bueso if (locked > lock_limit && !capable(CAP_IPC_LOCK)) 1376363ee17fSDavidlohr Bueso return -EAGAIN; 1377363ee17fSDavidlohr Bueso } 1378363ee17fSDavidlohr Bueso return 0; 1379363ee17fSDavidlohr Bueso } 1380363ee17fSDavidlohr Bueso 1381be83bbf8SLinus Torvalds static inline u64 file_mmap_size_max(struct file *file, struct inode *inode) 1382be83bbf8SLinus Torvalds { 1383be83bbf8SLinus Torvalds if (S_ISREG(inode->i_mode)) 1384423913adSLinus Torvalds return MAX_LFS_FILESIZE; 1385be83bbf8SLinus Torvalds 1386be83bbf8SLinus Torvalds if (S_ISBLK(inode->i_mode)) 1387be83bbf8SLinus Torvalds return MAX_LFS_FILESIZE; 1388be83bbf8SLinus Torvalds 138976f34950SIvan Khoronzhuk if (S_ISSOCK(inode->i_mode)) 139076f34950SIvan Khoronzhuk return MAX_LFS_FILESIZE; 139176f34950SIvan Khoronzhuk 1392be83bbf8SLinus Torvalds /* Special "we do even unsigned file positions" case */ 1393be83bbf8SLinus Torvalds if (file->f_mode & FMODE_UNSIGNED_OFFSET) 1394be83bbf8SLinus Torvalds return 0; 1395be83bbf8SLinus Torvalds 1396be83bbf8SLinus Torvalds /* Yes, random drivers might want more. But I'm tired of buggy drivers */ 1397be83bbf8SLinus Torvalds return ULONG_MAX; 1398be83bbf8SLinus Torvalds } 1399be83bbf8SLinus Torvalds 1400be83bbf8SLinus Torvalds static inline bool file_mmap_ok(struct file *file, struct inode *inode, 1401be83bbf8SLinus Torvalds unsigned long pgoff, unsigned long len) 1402be83bbf8SLinus Torvalds { 1403be83bbf8SLinus Torvalds u64 maxsize = file_mmap_size_max(file, inode); 1404be83bbf8SLinus Torvalds 1405be83bbf8SLinus Torvalds if (maxsize && len > maxsize) 1406be83bbf8SLinus Torvalds return false; 1407be83bbf8SLinus Torvalds maxsize -= len; 1408be83bbf8SLinus Torvalds if (pgoff > maxsize >> PAGE_SHIFT) 1409be83bbf8SLinus Torvalds return false; 1410be83bbf8SLinus Torvalds return true; 1411be83bbf8SLinus Torvalds } 1412be83bbf8SLinus Torvalds 141340401530SAl Viro /* 14143e4e28c5SMichel Lespinasse * The caller must write-lock current->mm->mmap_lock. 14151da177e4SLinus Torvalds */ 14161fcfd8dbSOleg Nesterov unsigned long do_mmap(struct file *file, unsigned long addr, 14171da177e4SLinus Torvalds unsigned long len, unsigned long prot, 141845e55300SPeter Collingbourne unsigned long flags, unsigned long pgoff, 141945e55300SPeter Collingbourne unsigned long *populate, struct list_head *uf) 14201da177e4SLinus Torvalds { 14211da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 142245e55300SPeter Collingbourne vm_flags_t vm_flags; 142362b5f7d0SDave Hansen int pkey = 0; 14241da177e4SLinus Torvalds 142541badc15SMichel Lespinasse *populate = 0; 1426bebeb3d6SMichel Lespinasse 1427e37609bbSPiotr Kwapulinski if (!len) 1428e37609bbSPiotr Kwapulinski return -EINVAL; 1429e37609bbSPiotr Kwapulinski 14301da177e4SLinus Torvalds /* 14311da177e4SLinus Torvalds * Does the application expect PROT_READ to imply PROT_EXEC? 14321da177e4SLinus Torvalds * 14331da177e4SLinus Torvalds * (the exception is when the underlying filesystem is noexec 14341da177e4SLinus Torvalds * mounted, in which case we dont add PROT_EXEC.) 14351da177e4SLinus Torvalds */ 14361da177e4SLinus Torvalds if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC)) 143790f8572bSEric W. Biederman if (!(file && path_noexec(&file->f_path))) 14381da177e4SLinus Torvalds prot |= PROT_EXEC; 14391da177e4SLinus Torvalds 1440a4ff8e86SMichal Hocko /* force arch specific MAP_FIXED handling in get_unmapped_area */ 1441a4ff8e86SMichal Hocko if (flags & MAP_FIXED_NOREPLACE) 1442a4ff8e86SMichal Hocko flags |= MAP_FIXED; 1443a4ff8e86SMichal Hocko 14447cd94146SEric Paris if (!(flags & MAP_FIXED)) 14457cd94146SEric Paris addr = round_hint_to_min(addr); 14467cd94146SEric Paris 14471da177e4SLinus Torvalds /* Careful about overflows.. */ 14481da177e4SLinus Torvalds len = PAGE_ALIGN(len); 14499206de95SAl Viro if (!len) 14501da177e4SLinus Torvalds return -ENOMEM; 14511da177e4SLinus Torvalds 14521da177e4SLinus Torvalds /* offset overflow? */ 14531da177e4SLinus Torvalds if ((pgoff + (len >> PAGE_SHIFT)) < pgoff) 14541da177e4SLinus Torvalds return -EOVERFLOW; 14551da177e4SLinus Torvalds 14561da177e4SLinus Torvalds /* Too many mappings? */ 14571da177e4SLinus Torvalds if (mm->map_count > sysctl_max_map_count) 14581da177e4SLinus Torvalds return -ENOMEM; 14591da177e4SLinus Torvalds 14601da177e4SLinus Torvalds /* Obtain the address to map to. we verify (or select) it and ensure 14611da177e4SLinus Torvalds * that it represents a valid section of the address space. 14621da177e4SLinus Torvalds */ 14631da177e4SLinus Torvalds addr = get_unmapped_area(file, addr, len, pgoff, flags); 1464ff68dac6SGaowei Pu if (IS_ERR_VALUE(addr)) 14651da177e4SLinus Torvalds return addr; 14661da177e4SLinus Torvalds 1467a4ff8e86SMichal Hocko if (flags & MAP_FIXED_NOREPLACE) { 146835e43c5fSLiam Howlett if (find_vma_intersection(mm, addr, addr + len)) 1469a4ff8e86SMichal Hocko return -EEXIST; 1470a4ff8e86SMichal Hocko } 1471a4ff8e86SMichal Hocko 147262b5f7d0SDave Hansen if (prot == PROT_EXEC) { 147362b5f7d0SDave Hansen pkey = execute_only_pkey(mm); 147462b5f7d0SDave Hansen if (pkey < 0) 147562b5f7d0SDave Hansen pkey = 0; 147662b5f7d0SDave Hansen } 147762b5f7d0SDave Hansen 14781da177e4SLinus Torvalds /* Do simple checking here so the lower-level routines won't have 14791da177e4SLinus Torvalds * to. we assume access permissions have been handled by the open 14801da177e4SLinus Torvalds * of the memory object, so we don't do any here. 14811da177e4SLinus Torvalds */ 148245e55300SPeter Collingbourne vm_flags = calc_vm_prot_bits(prot, pkey) | calc_vm_flag_bits(flags) | 14831da177e4SLinus Torvalds mm->def_flags | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC; 14841da177e4SLinus Torvalds 1485cdf7b341SHuang Shijie if (flags & MAP_LOCKED) 14861da177e4SLinus Torvalds if (!can_do_mlock()) 14871da177e4SLinus Torvalds return -EPERM; 1488ba470de4SRik van Riel 1489363ee17fSDavidlohr Bueso if (mlock_future_check(mm, vm_flags, len)) 14901da177e4SLinus Torvalds return -EAGAIN; 14911da177e4SLinus Torvalds 14921da177e4SLinus Torvalds if (file) { 1493077bf22bSOleg Nesterov struct inode *inode = file_inode(file); 14941c972597SDan Williams unsigned long flags_mask; 14951c972597SDan Williams 1496be83bbf8SLinus Torvalds if (!file_mmap_ok(file, inode, pgoff, len)) 1497be83bbf8SLinus Torvalds return -EOVERFLOW; 1498be83bbf8SLinus Torvalds 14991c972597SDan Williams flags_mask = LEGACY_MAP_MASK | file->f_op->mmap_supported_flags; 1500077bf22bSOleg Nesterov 15011da177e4SLinus Torvalds switch (flags & MAP_TYPE) { 15021da177e4SLinus Torvalds case MAP_SHARED: 15031c972597SDan Williams /* 15041c972597SDan Williams * Force use of MAP_SHARED_VALIDATE with non-legacy 15051c972597SDan Williams * flags. E.g. MAP_SYNC is dangerous to use with 15061c972597SDan Williams * MAP_SHARED as you don't know which consistency model 15071c972597SDan Williams * you will get. We silently ignore unsupported flags 15081c972597SDan Williams * with MAP_SHARED to preserve backward compatibility. 15091c972597SDan Williams */ 15101c972597SDan Williams flags &= LEGACY_MAP_MASK; 1511e4a9bc58SJoe Perches fallthrough; 15121c972597SDan Williams case MAP_SHARED_VALIDATE: 15131c972597SDan Williams if (flags & ~flags_mask) 15141c972597SDan Williams return -EOPNOTSUPP; 1515dc617f29SDarrick J. Wong if (prot & PROT_WRITE) { 1516dc617f29SDarrick J. Wong if (!(file->f_mode & FMODE_WRITE)) 15171da177e4SLinus Torvalds return -EACCES; 1518dc617f29SDarrick J. Wong if (IS_SWAPFILE(file->f_mapping->host)) 1519dc617f29SDarrick J. Wong return -ETXTBSY; 1520dc617f29SDarrick J. Wong } 15211da177e4SLinus Torvalds 15221da177e4SLinus Torvalds /* 15231da177e4SLinus Torvalds * Make sure we don't allow writing to an append-only 15241da177e4SLinus Torvalds * file.. 15251da177e4SLinus Torvalds */ 15261da177e4SLinus Torvalds if (IS_APPEND(inode) && (file->f_mode & FMODE_WRITE)) 15271da177e4SLinus Torvalds return -EACCES; 15281da177e4SLinus Torvalds 15291da177e4SLinus Torvalds vm_flags |= VM_SHARED | VM_MAYSHARE; 15301da177e4SLinus Torvalds if (!(file->f_mode & FMODE_WRITE)) 15311da177e4SLinus Torvalds vm_flags &= ~(VM_MAYWRITE | VM_SHARED); 1532e4a9bc58SJoe Perches fallthrough; 15331da177e4SLinus Torvalds case MAP_PRIVATE: 15341da177e4SLinus Torvalds if (!(file->f_mode & FMODE_READ)) 15351da177e4SLinus Torvalds return -EACCES; 153690f8572bSEric W. Biederman if (path_noexec(&file->f_path)) { 153780c5606cSLinus Torvalds if (vm_flags & VM_EXEC) 153880c5606cSLinus Torvalds return -EPERM; 153980c5606cSLinus Torvalds vm_flags &= ~VM_MAYEXEC; 154080c5606cSLinus Torvalds } 154180c5606cSLinus Torvalds 154272c2d531SAl Viro if (!file->f_op->mmap) 154380c5606cSLinus Torvalds return -ENODEV; 1544b2c56e4fSOleg Nesterov if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP)) 1545b2c56e4fSOleg Nesterov return -EINVAL; 15461da177e4SLinus Torvalds break; 15471da177e4SLinus Torvalds 15481da177e4SLinus Torvalds default: 15491da177e4SLinus Torvalds return -EINVAL; 15501da177e4SLinus Torvalds } 15511da177e4SLinus Torvalds } else { 15521da177e4SLinus Torvalds switch (flags & MAP_TYPE) { 15531da177e4SLinus Torvalds case MAP_SHARED: 1554b2c56e4fSOleg Nesterov if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP)) 1555b2c56e4fSOleg Nesterov return -EINVAL; 1556ce363942STejun Heo /* 1557ce363942STejun Heo * Ignore pgoff. 1558ce363942STejun Heo */ 1559ce363942STejun Heo pgoff = 0; 15601da177e4SLinus Torvalds vm_flags |= VM_SHARED | VM_MAYSHARE; 15611da177e4SLinus Torvalds break; 15621da177e4SLinus Torvalds case MAP_PRIVATE: 15631da177e4SLinus Torvalds /* 15641da177e4SLinus Torvalds * Set pgoff according to addr for anon_vma. 15651da177e4SLinus Torvalds */ 15661da177e4SLinus Torvalds pgoff = addr >> PAGE_SHIFT; 15671da177e4SLinus Torvalds break; 15681da177e4SLinus Torvalds default: 15691da177e4SLinus Torvalds return -EINVAL; 15701da177e4SLinus Torvalds } 15711da177e4SLinus Torvalds } 15721da177e4SLinus Torvalds 1573c22c0d63SMichel Lespinasse /* 1574c22c0d63SMichel Lespinasse * Set 'VM_NORESERVE' if we should not account for the 1575c22c0d63SMichel Lespinasse * memory use of this mapping. 1576c22c0d63SMichel Lespinasse */ 1577c22c0d63SMichel Lespinasse if (flags & MAP_NORESERVE) { 1578c22c0d63SMichel Lespinasse /* We honor MAP_NORESERVE if allowed to overcommit */ 1579c22c0d63SMichel Lespinasse if (sysctl_overcommit_memory != OVERCOMMIT_NEVER) 1580c22c0d63SMichel Lespinasse vm_flags |= VM_NORESERVE; 1581c22c0d63SMichel Lespinasse 1582c22c0d63SMichel Lespinasse /* hugetlb applies strict overcommit unless MAP_NORESERVE */ 1583c22c0d63SMichel Lespinasse if (file && is_file_hugepages(file)) 1584c22c0d63SMichel Lespinasse vm_flags |= VM_NORESERVE; 1585c22c0d63SMichel Lespinasse } 1586c22c0d63SMichel Lespinasse 1587897ab3e0SMike Rapoport addr = mmap_region(file, addr, len, vm_flags, pgoff, uf); 158809a9f1d2SMichel Lespinasse if (!IS_ERR_VALUE(addr) && 158909a9f1d2SMichel Lespinasse ((vm_flags & VM_LOCKED) || 159009a9f1d2SMichel Lespinasse (flags & (MAP_POPULATE | MAP_NONBLOCK)) == MAP_POPULATE)) 159141badc15SMichel Lespinasse *populate = len; 1592bebeb3d6SMichel Lespinasse return addr; 15930165ab44SMiklos Szeredi } 15946be5ceb0SLinus Torvalds 1595a90f590aSDominik Brodowski unsigned long ksys_mmap_pgoff(unsigned long addr, unsigned long len, 1596a90f590aSDominik Brodowski unsigned long prot, unsigned long flags, 1597a90f590aSDominik Brodowski unsigned long fd, unsigned long pgoff) 159866f0dc48SHugh Dickins { 159966f0dc48SHugh Dickins struct file *file = NULL; 16001e3ee14bSChen Gang unsigned long retval; 160166f0dc48SHugh Dickins 160266f0dc48SHugh Dickins if (!(flags & MAP_ANONYMOUS)) { 1603120a795dSAl Viro audit_mmap_fd(fd, flags); 160466f0dc48SHugh Dickins file = fget(fd); 160566f0dc48SHugh Dickins if (!file) 16061e3ee14bSChen Gang return -EBADF; 16077bba8f0eSZhen Lei if (is_file_hugepages(file)) { 1608af73e4d9SNaoya Horiguchi len = ALIGN(len, huge_page_size(hstate_file(file))); 16097bba8f0eSZhen Lei } else if (unlikely(flags & MAP_HUGETLB)) { 1610493af578SJörn Engel retval = -EINVAL; 1611493af578SJörn Engel goto out_fput; 16127bba8f0eSZhen Lei } 161366f0dc48SHugh Dickins } else if (flags & MAP_HUGETLB) { 1614c103a4dcSAndrew Morton struct hstate *hs; 1615af73e4d9SNaoya Horiguchi 161620ac2893SAnshuman Khandual hs = hstate_sizelog((flags >> MAP_HUGE_SHIFT) & MAP_HUGE_MASK); 1617091d0d55SLi Zefan if (!hs) 1618091d0d55SLi Zefan return -EINVAL; 1619091d0d55SLi Zefan 1620091d0d55SLi Zefan len = ALIGN(len, huge_page_size(hs)); 162166f0dc48SHugh Dickins /* 162266f0dc48SHugh Dickins * VM_NORESERVE is used because the reservations will be 162366f0dc48SHugh Dickins * taken when vm_ops->mmap() is called 162466f0dc48SHugh Dickins */ 1625af73e4d9SNaoya Horiguchi file = hugetlb_file_setup(HUGETLB_ANON_FILE, len, 162642d7395fSAndi Kleen VM_NORESERVE, 162783c1fd76Szhangyiru HUGETLB_ANONHUGE_INODE, 162842d7395fSAndi Kleen (flags >> MAP_HUGE_SHIFT) & MAP_HUGE_MASK); 162966f0dc48SHugh Dickins if (IS_ERR(file)) 163066f0dc48SHugh Dickins return PTR_ERR(file); 163166f0dc48SHugh Dickins } 163266f0dc48SHugh Dickins 16339fbeb5abSMichal Hocko retval = vm_mmap_pgoff(file, addr, len, prot, flags, pgoff); 1634493af578SJörn Engel out_fput: 163566f0dc48SHugh Dickins if (file) 163666f0dc48SHugh Dickins fput(file); 163766f0dc48SHugh Dickins return retval; 163866f0dc48SHugh Dickins } 163966f0dc48SHugh Dickins 1640a90f590aSDominik Brodowski SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len, 1641a90f590aSDominik Brodowski unsigned long, prot, unsigned long, flags, 1642a90f590aSDominik Brodowski unsigned long, fd, unsigned long, pgoff) 1643a90f590aSDominik Brodowski { 1644a90f590aSDominik Brodowski return ksys_mmap_pgoff(addr, len, prot, flags, fd, pgoff); 1645a90f590aSDominik Brodowski } 1646a90f590aSDominik Brodowski 1647a4679373SChristoph Hellwig #ifdef __ARCH_WANT_SYS_OLD_MMAP 1648a4679373SChristoph Hellwig struct mmap_arg_struct { 1649a4679373SChristoph Hellwig unsigned long addr; 1650a4679373SChristoph Hellwig unsigned long len; 1651a4679373SChristoph Hellwig unsigned long prot; 1652a4679373SChristoph Hellwig unsigned long flags; 1653a4679373SChristoph Hellwig unsigned long fd; 1654a4679373SChristoph Hellwig unsigned long offset; 1655a4679373SChristoph Hellwig }; 1656a4679373SChristoph Hellwig 1657a4679373SChristoph Hellwig SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg) 1658a4679373SChristoph Hellwig { 1659a4679373SChristoph Hellwig struct mmap_arg_struct a; 1660a4679373SChristoph Hellwig 1661a4679373SChristoph Hellwig if (copy_from_user(&a, arg, sizeof(a))) 1662a4679373SChristoph Hellwig return -EFAULT; 1663de1741a1SAlexander Kuleshov if (offset_in_page(a.offset)) 1664a4679373SChristoph Hellwig return -EINVAL; 1665a4679373SChristoph Hellwig 1666a90f590aSDominik Brodowski return ksys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd, 1667a4679373SChristoph Hellwig a.offset >> PAGE_SHIFT); 1668a4679373SChristoph Hellwig } 1669a4679373SChristoph Hellwig #endif /* __ARCH_WANT_SYS_OLD_MMAP */ 1670a4679373SChristoph Hellwig 16714e950f6fSAlexey Dobriyan /* 16728bb4e7a2SWei Yang * Some shared mappings will want the pages marked read-only 16734e950f6fSAlexey Dobriyan * to track write events. If so, we'll downgrade vm_page_prot 16744e950f6fSAlexey Dobriyan * to the private version (using protection_map[] without the 16754e950f6fSAlexey Dobriyan * VM_SHARED bit). 16764e950f6fSAlexey Dobriyan */ 16776d2329f8SAndrea Arcangeli int vma_wants_writenotify(struct vm_area_struct *vma, pgprot_t vm_page_prot) 16784e950f6fSAlexey Dobriyan { 1679ca16d140SKOSAKI Motohiro vm_flags_t vm_flags = vma->vm_flags; 16808a04446aSKirill A. Shutemov const struct vm_operations_struct *vm_ops = vma->vm_ops; 16814e950f6fSAlexey Dobriyan 16824e950f6fSAlexey Dobriyan /* If it was private or non-writable, the write bit is already clear */ 16834e950f6fSAlexey Dobriyan if ((vm_flags & (VM_WRITE|VM_SHARED)) != ((VM_WRITE|VM_SHARED))) 16844e950f6fSAlexey Dobriyan return 0; 16854e950f6fSAlexey Dobriyan 16864e950f6fSAlexey Dobriyan /* The backer wishes to know when pages are first written to? */ 16878a04446aSKirill A. Shutemov if (vm_ops && (vm_ops->page_mkwrite || vm_ops->pfn_mkwrite)) 16884e950f6fSAlexey Dobriyan return 1; 16894e950f6fSAlexey Dobriyan 169064e45507SPeter Feiner /* The open routine did something to the protections that pgprot_modify 169164e45507SPeter Feiner * won't preserve? */ 16926d2329f8SAndrea Arcangeli if (pgprot_val(vm_page_prot) != 16936d2329f8SAndrea Arcangeli pgprot_val(vm_pgprot_modify(vm_page_prot, vm_flags))) 16944e950f6fSAlexey Dobriyan return 0; 16954e950f6fSAlexey Dobriyan 169664e45507SPeter Feiner /* Do we need to track softdirty? */ 169764e45507SPeter Feiner if (IS_ENABLED(CONFIG_MEM_SOFT_DIRTY) && !(vm_flags & VM_SOFTDIRTY)) 169864e45507SPeter Feiner return 1; 169964e45507SPeter Feiner 17004e950f6fSAlexey Dobriyan /* Specialty mapping? */ 17014b6e1e37SKonstantin Khlebnikov if (vm_flags & VM_PFNMAP) 17024e950f6fSAlexey Dobriyan return 0; 17034e950f6fSAlexey Dobriyan 17044e950f6fSAlexey Dobriyan /* Can the mapping track the dirty pages? */ 17054e950f6fSAlexey Dobriyan return vma->vm_file && vma->vm_file->f_mapping && 1706f56753acSChristoph Hellwig mapping_can_writeback(vma->vm_file->f_mapping); 17074e950f6fSAlexey Dobriyan } 17084e950f6fSAlexey Dobriyan 1709fc8744adSLinus Torvalds /* 1710fc8744adSLinus Torvalds * We account for memory if it's a private writeable mapping, 17115a6fe125SMel Gorman * not hugepages and VM_NORESERVE wasn't set. 1712fc8744adSLinus Torvalds */ 1713ca16d140SKOSAKI Motohiro static inline int accountable_mapping(struct file *file, vm_flags_t vm_flags) 1714fc8744adSLinus Torvalds { 17155a6fe125SMel Gorman /* 17165a6fe125SMel Gorman * hugetlb has its own accounting separate from the core VM 17175a6fe125SMel Gorman * VM_HUGETLB may not be set yet so we cannot check for that flag. 17185a6fe125SMel Gorman */ 17195a6fe125SMel Gorman if (file && is_file_hugepages(file)) 17205a6fe125SMel Gorman return 0; 17215a6fe125SMel Gorman 1722fc8744adSLinus Torvalds return (vm_flags & (VM_NORESERVE | VM_SHARED | VM_WRITE)) == VM_WRITE; 1723fc8744adSLinus Torvalds } 1724fc8744adSLinus Torvalds 17250165ab44SMiklos Szeredi unsigned long mmap_region(struct file *file, unsigned long addr, 1726897ab3e0SMike Rapoport unsigned long len, vm_flags_t vm_flags, unsigned long pgoff, 1727897ab3e0SMike Rapoport struct list_head *uf) 17280165ab44SMiklos Szeredi { 17290165ab44SMiklos Szeredi struct mm_struct *mm = current->mm; 1730d70cec89SMiaohe Lin struct vm_area_struct *vma, *prev, *merge; 17310165ab44SMiklos Szeredi int error; 17320165ab44SMiklos Szeredi struct rb_node **rb_link, *rb_parent; 17330165ab44SMiklos Szeredi unsigned long charged = 0; 17340165ab44SMiklos Szeredi 1735e8420a8eSCyril Hrubis /* Check against address space limit. */ 173684638335SKonstantin Khlebnikov if (!may_expand_vm(mm, vm_flags, len >> PAGE_SHIFT)) { 1737e8420a8eSCyril Hrubis unsigned long nr_pages; 1738e8420a8eSCyril Hrubis 1739e8420a8eSCyril Hrubis /* 1740e8420a8eSCyril Hrubis * MAP_FIXED may remove pages of mappings that intersects with 1741e8420a8eSCyril Hrubis * requested mapping. Account for the pages it would unmap. 1742e8420a8eSCyril Hrubis */ 1743e8420a8eSCyril Hrubis nr_pages = count_vma_pages_range(mm, addr, addr + len); 1744e8420a8eSCyril Hrubis 174584638335SKonstantin Khlebnikov if (!may_expand_vm(mm, vm_flags, 174684638335SKonstantin Khlebnikov (len >> PAGE_SHIFT) - nr_pages)) 1747e8420a8eSCyril Hrubis return -ENOMEM; 1748e8420a8eSCyril Hrubis } 1749e8420a8eSCyril Hrubis 1750fb8090b6SLiam R. Howlett /* Clear old maps, set up prev, rb_link, rb_parent, and uf */ 1751fb8090b6SLiam R. Howlett if (munmap_vma_range(mm, addr, len, &prev, &rb_link, &rb_parent, uf)) 17521da177e4SLinus Torvalds return -ENOMEM; 1753fc8744adSLinus Torvalds /* 17541da177e4SLinus Torvalds * Private writable mapping: check memory availability 17551da177e4SLinus Torvalds */ 17565a6fe125SMel Gorman if (accountable_mapping(file, vm_flags)) { 17571da177e4SLinus Torvalds charged = len >> PAGE_SHIFT; 1758191c5424SAl Viro if (security_vm_enough_memory_mm(mm, charged)) 17591da177e4SLinus Torvalds return -ENOMEM; 17601da177e4SLinus Torvalds vm_flags |= VM_ACCOUNT; 17611da177e4SLinus Torvalds } 17621da177e4SLinus Torvalds 17631da177e4SLinus Torvalds /* 1764de33c8dbSLinus Torvalds * Can we just expand an old mapping? 17651da177e4SLinus Torvalds */ 176619a809afSAndrea Arcangeli vma = vma_merge(mm, prev, addr, addr + len, vm_flags, 17679a10064fSColin Cross NULL, file, pgoff, NULL, NULL_VM_UFFD_CTX, NULL); 1768ba470de4SRik van Riel if (vma) 17691da177e4SLinus Torvalds goto out; 17701da177e4SLinus Torvalds 17711da177e4SLinus Torvalds /* 17721da177e4SLinus Torvalds * Determine the object being mapped and call the appropriate 17731da177e4SLinus Torvalds * specific mapper. the address has already been validated, but 17741da177e4SLinus Torvalds * not unmapped, but the maps are removed from the list. 17751da177e4SLinus Torvalds */ 1776490fc053SLinus Torvalds vma = vm_area_alloc(mm); 17771da177e4SLinus Torvalds if (!vma) { 17781da177e4SLinus Torvalds error = -ENOMEM; 17791da177e4SLinus Torvalds goto unacct_error; 17801da177e4SLinus Torvalds } 17811da177e4SLinus Torvalds 17821da177e4SLinus Torvalds vma->vm_start = addr; 17831da177e4SLinus Torvalds vma->vm_end = addr + len; 17841da177e4SLinus Torvalds vma->vm_flags = vm_flags; 17853ed75eb8SColy Li vma->vm_page_prot = vm_get_page_prot(vm_flags); 17861da177e4SLinus Torvalds vma->vm_pgoff = pgoff; 17871da177e4SLinus Torvalds 17881da177e4SLinus Torvalds if (file) { 17894bb5f5d9SDavid Herrmann if (vm_flags & VM_SHARED) { 17904bb5f5d9SDavid Herrmann error = mapping_map_writable(file->f_mapping); 17914bb5f5d9SDavid Herrmann if (error) 17928d0920bdSDavid Hildenbrand goto free_vma; 17934bb5f5d9SDavid Herrmann } 17944bb5f5d9SDavid Herrmann 1795cb0942b8SAl Viro vma->vm_file = get_file(file); 1796f74ac015SMiklos Szeredi error = call_mmap(file, vma); 17971da177e4SLinus Torvalds if (error) 17981da177e4SLinus Torvalds goto unmap_and_free_vma; 17991da177e4SLinus Torvalds 1800309d08d9SLiu Zixian /* Can addr have changed?? 1801309d08d9SLiu Zixian * 1802309d08d9SLiu Zixian * Answer: Yes, several device drivers can do it in their 1803309d08d9SLiu Zixian * f_op->mmap method. -DaveM 1804309d08d9SLiu Zixian * Bug: If addr is changed, prev, rb_link, rb_parent should 1805309d08d9SLiu Zixian * be updated for vma_link() 1806309d08d9SLiu Zixian */ 1807309d08d9SLiu Zixian WARN_ON_ONCE(addr != vma->vm_start); 1808309d08d9SLiu Zixian 1809309d08d9SLiu Zixian addr = vma->vm_start; 1810309d08d9SLiu Zixian 1811d70cec89SMiaohe Lin /* If vm_flags changed after call_mmap(), we should try merge vma again 1812d70cec89SMiaohe Lin * as we may succeed this time. 1813d70cec89SMiaohe Lin */ 1814d70cec89SMiaohe Lin if (unlikely(vm_flags != vma->vm_flags && prev)) { 1815d70cec89SMiaohe Lin merge = vma_merge(mm, prev, vma->vm_start, vma->vm_end, vma->vm_flags, 18169a10064fSColin Cross NULL, vma->vm_file, vma->vm_pgoff, NULL, NULL_VM_UFFD_CTX, NULL); 1817d70cec89SMiaohe Lin if (merge) { 1818bc4fe4cdSMiaohe Lin /* ->mmap() can change vma->vm_file and fput the original file. So 1819bc4fe4cdSMiaohe Lin * fput the vma->vm_file here or we would add an extra fput for file 1820bc4fe4cdSMiaohe Lin * and cause general protection fault ultimately. 1821bc4fe4cdSMiaohe Lin */ 1822bc4fe4cdSMiaohe Lin fput(vma->vm_file); 1823d70cec89SMiaohe Lin vm_area_free(vma); 1824d70cec89SMiaohe Lin vma = merge; 1825309d08d9SLiu Zixian /* Update vm_flags to pick up the change. */ 1826d70cec89SMiaohe Lin vm_flags = vma->vm_flags; 1827d70cec89SMiaohe Lin goto unmap_writable; 1828d70cec89SMiaohe Lin } 1829d70cec89SMiaohe Lin } 1830d70cec89SMiaohe Lin 18311da177e4SLinus Torvalds vm_flags = vma->vm_flags; 1832f8dbf0a7SHuang Shijie } else if (vm_flags & VM_SHARED) { 1833f8dbf0a7SHuang Shijie error = shmem_zero_setup(vma); 1834f8dbf0a7SHuang Shijie if (error) 1835f8dbf0a7SHuang Shijie goto free_vma; 1836bfd40eafSKirill A. Shutemov } else { 1837bfd40eafSKirill A. Shutemov vma_set_anonymous(vma); 1838f8dbf0a7SHuang Shijie } 18391da177e4SLinus Torvalds 1840c462ac28SCatalin Marinas /* Allow architectures to sanity-check the vm_flags */ 1841c462ac28SCatalin Marinas if (!arch_validate_flags(vma->vm_flags)) { 1842c462ac28SCatalin Marinas error = -EINVAL; 1843c462ac28SCatalin Marinas if (file) 1844c462ac28SCatalin Marinas goto unmap_and_free_vma; 1845c462ac28SCatalin Marinas else 1846c462ac28SCatalin Marinas goto free_vma; 1847c462ac28SCatalin Marinas } 1848c462ac28SCatalin Marinas 18494d3d5b41SOleg Nesterov vma_link(mm, vma, prev, rb_link, rb_parent); 1850*613bec09SYang Shi 1851*613bec09SYang Shi /* 1852*613bec09SYang Shi * vma_merge() calls khugepaged_enter_vma() either, the below 1853*613bec09SYang Shi * call covers the non-merge case. 1854*613bec09SYang Shi */ 1855*613bec09SYang Shi khugepaged_enter_vma(vma, vma->vm_flags); 1856*613bec09SYang Shi 18574d3d5b41SOleg Nesterov /* Once vma denies write, undo our temporary denial count */ 1858d70cec89SMiaohe Lin unmap_writable: 18598d0920bdSDavid Hildenbrand if (file && vm_flags & VM_SHARED) 18604bb5f5d9SDavid Herrmann mapping_unmap_writable(file->f_mapping); 1861e8686772SOleg Nesterov file = vma->vm_file; 18621da177e4SLinus Torvalds out: 1863cdd6c482SIngo Molnar perf_event_mmap(vma); 18640a4a9391SPeter Zijlstra 186584638335SKonstantin Khlebnikov vm_stat_account(mm, vm_flags, len >> PAGE_SHIFT); 18661da177e4SLinus Torvalds if (vm_flags & VM_LOCKED) { 1867e1fb4a08SDave Jiang if ((vm_flags & VM_SPECIAL) || vma_is_dax(vma) || 1868e1fb4a08SDave Jiang is_vm_hugetlb_page(vma) || 1869e1fb4a08SDave Jiang vma == get_gate_vma(current->mm)) 1870de60f5f1SEric B Munson vma->vm_flags &= VM_LOCKED_CLEAR_MASK; 1871e1fb4a08SDave Jiang else 1872e1fb4a08SDave Jiang mm->locked_vm += (len >> PAGE_SHIFT); 1873bebeb3d6SMichel Lespinasse } 18742b144498SSrikar Dronamraju 1875c7a3a88cSOleg Nesterov if (file) 1876c7a3a88cSOleg Nesterov uprobe_mmap(vma); 18772b144498SSrikar Dronamraju 1878d9104d1cSCyrill Gorcunov /* 1879d9104d1cSCyrill Gorcunov * New (or expanded) vma always get soft dirty status. 1880d9104d1cSCyrill Gorcunov * Otherwise user-space soft-dirty page tracker won't 1881d9104d1cSCyrill Gorcunov * be able to distinguish situation when vma area unmapped, 1882d9104d1cSCyrill Gorcunov * then new mapped in-place (which must be aimed as 1883d9104d1cSCyrill Gorcunov * a completely new data area). 1884d9104d1cSCyrill Gorcunov */ 1885d9104d1cSCyrill Gorcunov vma->vm_flags |= VM_SOFTDIRTY; 1886d9104d1cSCyrill Gorcunov 188764e45507SPeter Feiner vma_set_page_prot(vma); 188864e45507SPeter Feiner 18891da177e4SLinus Torvalds return addr; 18901da177e4SLinus Torvalds 18911da177e4SLinus Torvalds unmap_and_free_vma: 18921527f926SChristian König fput(vma->vm_file); 18931da177e4SLinus Torvalds vma->vm_file = NULL; 18941da177e4SLinus Torvalds 18951da177e4SLinus Torvalds /* Undo any partial mapping done by a device driver. */ 1896e0da382cSHugh Dickins unmap_region(mm, vma, prev, vma->vm_start, vma->vm_end); 1897e0da382cSHugh Dickins charged = 0; 18984bb5f5d9SDavid Herrmann if (vm_flags & VM_SHARED) 18994bb5f5d9SDavid Herrmann mapping_unmap_writable(file->f_mapping); 19001da177e4SLinus Torvalds free_vma: 19013928d4f5SLinus Torvalds vm_area_free(vma); 19021da177e4SLinus Torvalds unacct_error: 19031da177e4SLinus Torvalds if (charged) 19041da177e4SLinus Torvalds vm_unacct_memory(charged); 19051da177e4SLinus Torvalds return error; 19061da177e4SLinus Torvalds } 19071da177e4SLinus Torvalds 1908baceaf1cSJaewon Kim static unsigned long unmapped_area(struct vm_unmapped_area_info *info) 1909db4fbfb9SMichel Lespinasse { 1910db4fbfb9SMichel Lespinasse /* 1911db4fbfb9SMichel Lespinasse * We implement the search by looking for an rbtree node that 1912db4fbfb9SMichel Lespinasse * immediately follows a suitable gap. That is, 1913db4fbfb9SMichel Lespinasse * - gap_start = vma->vm_prev->vm_end <= info->high_limit - length; 1914db4fbfb9SMichel Lespinasse * - gap_end = vma->vm_start >= info->low_limit + length; 1915db4fbfb9SMichel Lespinasse * - gap_end - gap_start >= length 1916db4fbfb9SMichel Lespinasse */ 1917db4fbfb9SMichel Lespinasse 1918db4fbfb9SMichel Lespinasse struct mm_struct *mm = current->mm; 1919db4fbfb9SMichel Lespinasse struct vm_area_struct *vma; 1920db4fbfb9SMichel Lespinasse unsigned long length, low_limit, high_limit, gap_start, gap_end; 1921db4fbfb9SMichel Lespinasse 1922db4fbfb9SMichel Lespinasse /* Adjust search length to account for worst case alignment overhead */ 1923db4fbfb9SMichel Lespinasse length = info->length + info->align_mask; 1924db4fbfb9SMichel Lespinasse if (length < info->length) 1925db4fbfb9SMichel Lespinasse return -ENOMEM; 1926db4fbfb9SMichel Lespinasse 1927db4fbfb9SMichel Lespinasse /* Adjust search limits by the desired length */ 1928db4fbfb9SMichel Lespinasse if (info->high_limit < length) 1929db4fbfb9SMichel Lespinasse return -ENOMEM; 1930db4fbfb9SMichel Lespinasse high_limit = info->high_limit - length; 1931db4fbfb9SMichel Lespinasse 1932db4fbfb9SMichel Lespinasse if (info->low_limit > high_limit) 1933db4fbfb9SMichel Lespinasse return -ENOMEM; 1934db4fbfb9SMichel Lespinasse low_limit = info->low_limit + length; 1935db4fbfb9SMichel Lespinasse 1936db4fbfb9SMichel Lespinasse /* Check if rbtree root looks promising */ 1937db4fbfb9SMichel Lespinasse if (RB_EMPTY_ROOT(&mm->mm_rb)) 1938db4fbfb9SMichel Lespinasse goto check_highest; 1939db4fbfb9SMichel Lespinasse vma = rb_entry(mm->mm_rb.rb_node, struct vm_area_struct, vm_rb); 1940db4fbfb9SMichel Lespinasse if (vma->rb_subtree_gap < length) 1941db4fbfb9SMichel Lespinasse goto check_highest; 1942db4fbfb9SMichel Lespinasse 1943db4fbfb9SMichel Lespinasse while (true) { 1944db4fbfb9SMichel Lespinasse /* Visit left subtree if it looks promising */ 19451be7107fSHugh Dickins gap_end = vm_start_gap(vma); 1946db4fbfb9SMichel Lespinasse if (gap_end >= low_limit && vma->vm_rb.rb_left) { 1947db4fbfb9SMichel Lespinasse struct vm_area_struct *left = 1948db4fbfb9SMichel Lespinasse rb_entry(vma->vm_rb.rb_left, 1949db4fbfb9SMichel Lespinasse struct vm_area_struct, vm_rb); 1950db4fbfb9SMichel Lespinasse if (left->rb_subtree_gap >= length) { 1951db4fbfb9SMichel Lespinasse vma = left; 1952db4fbfb9SMichel Lespinasse continue; 1953db4fbfb9SMichel Lespinasse } 1954db4fbfb9SMichel Lespinasse } 1955db4fbfb9SMichel Lespinasse 19561be7107fSHugh Dickins gap_start = vma->vm_prev ? vm_end_gap(vma->vm_prev) : 0; 1957db4fbfb9SMichel Lespinasse check_current: 1958db4fbfb9SMichel Lespinasse /* Check if current node has a suitable gap */ 1959db4fbfb9SMichel Lespinasse if (gap_start > high_limit) 1960db4fbfb9SMichel Lespinasse return -ENOMEM; 1961f4cb767dSHugh Dickins if (gap_end >= low_limit && 1962f4cb767dSHugh Dickins gap_end > gap_start && gap_end - gap_start >= length) 1963db4fbfb9SMichel Lespinasse goto found; 1964db4fbfb9SMichel Lespinasse 1965db4fbfb9SMichel Lespinasse /* Visit right subtree if it looks promising */ 1966db4fbfb9SMichel Lespinasse if (vma->vm_rb.rb_right) { 1967db4fbfb9SMichel Lespinasse struct vm_area_struct *right = 1968db4fbfb9SMichel Lespinasse rb_entry(vma->vm_rb.rb_right, 1969db4fbfb9SMichel Lespinasse struct vm_area_struct, vm_rb); 1970db4fbfb9SMichel Lespinasse if (right->rb_subtree_gap >= length) { 1971db4fbfb9SMichel Lespinasse vma = right; 1972db4fbfb9SMichel Lespinasse continue; 1973db4fbfb9SMichel Lespinasse } 1974db4fbfb9SMichel Lespinasse } 1975db4fbfb9SMichel Lespinasse 1976db4fbfb9SMichel Lespinasse /* Go back up the rbtree to find next candidate node */ 1977db4fbfb9SMichel Lespinasse while (true) { 1978db4fbfb9SMichel Lespinasse struct rb_node *prev = &vma->vm_rb; 1979db4fbfb9SMichel Lespinasse if (!rb_parent(prev)) 1980db4fbfb9SMichel Lespinasse goto check_highest; 1981db4fbfb9SMichel Lespinasse vma = rb_entry(rb_parent(prev), 1982db4fbfb9SMichel Lespinasse struct vm_area_struct, vm_rb); 1983db4fbfb9SMichel Lespinasse if (prev == vma->vm_rb.rb_left) { 19841be7107fSHugh Dickins gap_start = vm_end_gap(vma->vm_prev); 19851be7107fSHugh Dickins gap_end = vm_start_gap(vma); 1986db4fbfb9SMichel Lespinasse goto check_current; 1987db4fbfb9SMichel Lespinasse } 1988db4fbfb9SMichel Lespinasse } 1989db4fbfb9SMichel Lespinasse } 1990db4fbfb9SMichel Lespinasse 1991db4fbfb9SMichel Lespinasse check_highest: 1992db4fbfb9SMichel Lespinasse /* Check highest gap, which does not precede any rbtree node */ 1993db4fbfb9SMichel Lespinasse gap_start = mm->highest_vm_end; 1994db4fbfb9SMichel Lespinasse gap_end = ULONG_MAX; /* Only for VM_BUG_ON below */ 1995db4fbfb9SMichel Lespinasse if (gap_start > high_limit) 1996db4fbfb9SMichel Lespinasse return -ENOMEM; 1997db4fbfb9SMichel Lespinasse 1998db4fbfb9SMichel Lespinasse found: 1999db4fbfb9SMichel Lespinasse /* We found a suitable gap. Clip it with the original low_limit. */ 2000db4fbfb9SMichel Lespinasse if (gap_start < info->low_limit) 2001db4fbfb9SMichel Lespinasse gap_start = info->low_limit; 2002db4fbfb9SMichel Lespinasse 2003db4fbfb9SMichel Lespinasse /* Adjust gap address to the desired alignment */ 2004db4fbfb9SMichel Lespinasse gap_start += (info->align_offset - gap_start) & info->align_mask; 2005db4fbfb9SMichel Lespinasse 2006db4fbfb9SMichel Lespinasse VM_BUG_ON(gap_start + info->length > info->high_limit); 2007db4fbfb9SMichel Lespinasse VM_BUG_ON(gap_start + info->length > gap_end); 2008db4fbfb9SMichel Lespinasse return gap_start; 2009db4fbfb9SMichel Lespinasse } 2010db4fbfb9SMichel Lespinasse 2011baceaf1cSJaewon Kim static unsigned long unmapped_area_topdown(struct vm_unmapped_area_info *info) 2012db4fbfb9SMichel Lespinasse { 2013db4fbfb9SMichel Lespinasse struct mm_struct *mm = current->mm; 2014db4fbfb9SMichel Lespinasse struct vm_area_struct *vma; 2015db4fbfb9SMichel Lespinasse unsigned long length, low_limit, high_limit, gap_start, gap_end; 2016db4fbfb9SMichel Lespinasse 2017db4fbfb9SMichel Lespinasse /* Adjust search length to account for worst case alignment overhead */ 2018db4fbfb9SMichel Lespinasse length = info->length + info->align_mask; 2019db4fbfb9SMichel Lespinasse if (length < info->length) 2020db4fbfb9SMichel Lespinasse return -ENOMEM; 2021db4fbfb9SMichel Lespinasse 2022db4fbfb9SMichel Lespinasse /* 2023db4fbfb9SMichel Lespinasse * Adjust search limits by the desired length. 2024db4fbfb9SMichel Lespinasse * See implementation comment at top of unmapped_area(). 2025db4fbfb9SMichel Lespinasse */ 2026db4fbfb9SMichel Lespinasse gap_end = info->high_limit; 2027db4fbfb9SMichel Lespinasse if (gap_end < length) 2028db4fbfb9SMichel Lespinasse return -ENOMEM; 2029db4fbfb9SMichel Lespinasse high_limit = gap_end - length; 2030db4fbfb9SMichel Lespinasse 2031db4fbfb9SMichel Lespinasse if (info->low_limit > high_limit) 2032db4fbfb9SMichel Lespinasse return -ENOMEM; 2033db4fbfb9SMichel Lespinasse low_limit = info->low_limit + length; 2034db4fbfb9SMichel Lespinasse 2035db4fbfb9SMichel Lespinasse /* Check highest gap, which does not precede any rbtree node */ 2036db4fbfb9SMichel Lespinasse gap_start = mm->highest_vm_end; 2037db4fbfb9SMichel Lespinasse if (gap_start <= high_limit) 2038db4fbfb9SMichel Lespinasse goto found_highest; 2039db4fbfb9SMichel Lespinasse 2040db4fbfb9SMichel Lespinasse /* Check if rbtree root looks promising */ 2041db4fbfb9SMichel Lespinasse if (RB_EMPTY_ROOT(&mm->mm_rb)) 2042db4fbfb9SMichel Lespinasse return -ENOMEM; 2043db4fbfb9SMichel Lespinasse vma = rb_entry(mm->mm_rb.rb_node, struct vm_area_struct, vm_rb); 2044db4fbfb9SMichel Lespinasse if (vma->rb_subtree_gap < length) 2045db4fbfb9SMichel Lespinasse return -ENOMEM; 2046db4fbfb9SMichel Lespinasse 2047db4fbfb9SMichel Lespinasse while (true) { 2048db4fbfb9SMichel Lespinasse /* Visit right subtree if it looks promising */ 20491be7107fSHugh Dickins gap_start = vma->vm_prev ? vm_end_gap(vma->vm_prev) : 0; 2050db4fbfb9SMichel Lespinasse if (gap_start <= high_limit && vma->vm_rb.rb_right) { 2051db4fbfb9SMichel Lespinasse struct vm_area_struct *right = 2052db4fbfb9SMichel Lespinasse rb_entry(vma->vm_rb.rb_right, 2053db4fbfb9SMichel Lespinasse struct vm_area_struct, vm_rb); 2054db4fbfb9SMichel Lespinasse if (right->rb_subtree_gap >= length) { 2055db4fbfb9SMichel Lespinasse vma = right; 2056db4fbfb9SMichel Lespinasse continue; 2057db4fbfb9SMichel Lespinasse } 2058db4fbfb9SMichel Lespinasse } 2059db4fbfb9SMichel Lespinasse 2060db4fbfb9SMichel Lespinasse check_current: 2061db4fbfb9SMichel Lespinasse /* Check if current node has a suitable gap */ 20621be7107fSHugh Dickins gap_end = vm_start_gap(vma); 2063db4fbfb9SMichel Lespinasse if (gap_end < low_limit) 2064db4fbfb9SMichel Lespinasse return -ENOMEM; 2065f4cb767dSHugh Dickins if (gap_start <= high_limit && 2066f4cb767dSHugh Dickins gap_end > gap_start && gap_end - gap_start >= length) 2067db4fbfb9SMichel Lespinasse goto found; 2068db4fbfb9SMichel Lespinasse 2069db4fbfb9SMichel Lespinasse /* Visit left subtree if it looks promising */ 2070db4fbfb9SMichel Lespinasse if (vma->vm_rb.rb_left) { 2071db4fbfb9SMichel Lespinasse struct vm_area_struct *left = 2072db4fbfb9SMichel Lespinasse rb_entry(vma->vm_rb.rb_left, 2073db4fbfb9SMichel Lespinasse struct vm_area_struct, vm_rb); 2074db4fbfb9SMichel Lespinasse if (left->rb_subtree_gap >= length) { 2075db4fbfb9SMichel Lespinasse vma = left; 2076db4fbfb9SMichel Lespinasse continue; 2077db4fbfb9SMichel Lespinasse } 2078db4fbfb9SMichel Lespinasse } 2079db4fbfb9SMichel Lespinasse 2080db4fbfb9SMichel Lespinasse /* Go back up the rbtree to find next candidate node */ 2081db4fbfb9SMichel Lespinasse while (true) { 2082db4fbfb9SMichel Lespinasse struct rb_node *prev = &vma->vm_rb; 2083db4fbfb9SMichel Lespinasse if (!rb_parent(prev)) 2084db4fbfb9SMichel Lespinasse return -ENOMEM; 2085db4fbfb9SMichel Lespinasse vma = rb_entry(rb_parent(prev), 2086db4fbfb9SMichel Lespinasse struct vm_area_struct, vm_rb); 2087db4fbfb9SMichel Lespinasse if (prev == vma->vm_rb.rb_right) { 2088db4fbfb9SMichel Lespinasse gap_start = vma->vm_prev ? 20891be7107fSHugh Dickins vm_end_gap(vma->vm_prev) : 0; 2090db4fbfb9SMichel Lespinasse goto check_current; 2091db4fbfb9SMichel Lespinasse } 2092db4fbfb9SMichel Lespinasse } 2093db4fbfb9SMichel Lespinasse } 2094db4fbfb9SMichel Lespinasse 2095db4fbfb9SMichel Lespinasse found: 2096db4fbfb9SMichel Lespinasse /* We found a suitable gap. Clip it with the original high_limit. */ 2097db4fbfb9SMichel Lespinasse if (gap_end > info->high_limit) 2098db4fbfb9SMichel Lespinasse gap_end = info->high_limit; 2099db4fbfb9SMichel Lespinasse 2100db4fbfb9SMichel Lespinasse found_highest: 2101db4fbfb9SMichel Lespinasse /* Compute highest gap address at the desired alignment */ 2102db4fbfb9SMichel Lespinasse gap_end -= info->length; 2103db4fbfb9SMichel Lespinasse gap_end -= (gap_end - info->align_offset) & info->align_mask; 2104db4fbfb9SMichel Lespinasse 2105db4fbfb9SMichel Lespinasse VM_BUG_ON(gap_end < info->low_limit); 2106db4fbfb9SMichel Lespinasse VM_BUG_ON(gap_end < gap_start); 2107db4fbfb9SMichel Lespinasse return gap_end; 2108db4fbfb9SMichel Lespinasse } 2109db4fbfb9SMichel Lespinasse 2110baceaf1cSJaewon Kim /* 2111baceaf1cSJaewon Kim * Search for an unmapped address range. 2112baceaf1cSJaewon Kim * 2113baceaf1cSJaewon Kim * We are looking for a range that: 2114baceaf1cSJaewon Kim * - does not intersect with any VMA; 2115baceaf1cSJaewon Kim * - is contained within the [low_limit, high_limit) interval; 2116baceaf1cSJaewon Kim * - is at least the desired size. 2117baceaf1cSJaewon Kim * - satisfies (begin_addr & align_mask) == (align_offset & align_mask) 2118baceaf1cSJaewon Kim */ 2119baceaf1cSJaewon Kim unsigned long vm_unmapped_area(struct vm_unmapped_area_info *info) 2120baceaf1cSJaewon Kim { 2121df529cabSJaewon Kim unsigned long addr; 2122df529cabSJaewon Kim 2123baceaf1cSJaewon Kim if (info->flags & VM_UNMAPPED_AREA_TOPDOWN) 2124df529cabSJaewon Kim addr = unmapped_area_topdown(info); 2125baceaf1cSJaewon Kim else 2126df529cabSJaewon Kim addr = unmapped_area(info); 2127df529cabSJaewon Kim 2128df529cabSJaewon Kim trace_vm_unmapped_area(addr, info); 2129df529cabSJaewon Kim return addr; 2130baceaf1cSJaewon Kim } 2131f6795053SSteve Capper 21321da177e4SLinus Torvalds /* Get an address range which is currently unmapped. 21331da177e4SLinus Torvalds * For shmat() with addr=0. 21341da177e4SLinus Torvalds * 21351da177e4SLinus Torvalds * Ugly calling convention alert: 21361da177e4SLinus Torvalds * Return value with the low bits set means error value, 21371da177e4SLinus Torvalds * ie 21381da177e4SLinus Torvalds * if (ret & ~PAGE_MASK) 21391da177e4SLinus Torvalds * error = ret; 21401da177e4SLinus Torvalds * 21411da177e4SLinus Torvalds * This function "knows" that -ENOMEM has the bits set. 21421da177e4SLinus Torvalds */ 21431da177e4SLinus Torvalds #ifndef HAVE_ARCH_UNMAPPED_AREA 21441da177e4SLinus Torvalds unsigned long 21451da177e4SLinus Torvalds arch_get_unmapped_area(struct file *filp, unsigned long addr, 21461da177e4SLinus Torvalds unsigned long len, unsigned long pgoff, unsigned long flags) 21471da177e4SLinus Torvalds { 21481da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 21491be7107fSHugh Dickins struct vm_area_struct *vma, *prev; 2150db4fbfb9SMichel Lespinasse struct vm_unmapped_area_info info; 2151f6795053SSteve Capper const unsigned long mmap_end = arch_get_mmap_end(addr); 21521da177e4SLinus Torvalds 2153f6795053SSteve Capper if (len > mmap_end - mmap_min_addr) 21541da177e4SLinus Torvalds return -ENOMEM; 21551da177e4SLinus Torvalds 215606abdfb4SBenjamin Herrenschmidt if (flags & MAP_FIXED) 215706abdfb4SBenjamin Herrenschmidt return addr; 215806abdfb4SBenjamin Herrenschmidt 21591da177e4SLinus Torvalds if (addr) { 21601da177e4SLinus Torvalds addr = PAGE_ALIGN(addr); 21611be7107fSHugh Dickins vma = find_vma_prev(mm, addr, &prev); 2162f6795053SSteve Capper if (mmap_end - len >= addr && addr >= mmap_min_addr && 21631be7107fSHugh Dickins (!vma || addr + len <= vm_start_gap(vma)) && 21641be7107fSHugh Dickins (!prev || addr >= vm_end_gap(prev))) 21651da177e4SLinus Torvalds return addr; 21661da177e4SLinus Torvalds } 21671da177e4SLinus Torvalds 2168db4fbfb9SMichel Lespinasse info.flags = 0; 2169db4fbfb9SMichel Lespinasse info.length = len; 21704e99b021SHeiko Carstens info.low_limit = mm->mmap_base; 2171f6795053SSteve Capper info.high_limit = mmap_end; 2172db4fbfb9SMichel Lespinasse info.align_mask = 0; 217309ef5283SJaewon Kim info.align_offset = 0; 2174db4fbfb9SMichel Lespinasse return vm_unmapped_area(&info); 21751da177e4SLinus Torvalds } 21761da177e4SLinus Torvalds #endif 21771da177e4SLinus Torvalds 21781da177e4SLinus Torvalds /* 21791da177e4SLinus Torvalds * This mmap-allocator allocates new areas top-down from below the 21801da177e4SLinus Torvalds * stack's low limit (the base): 21811da177e4SLinus Torvalds */ 21821da177e4SLinus Torvalds #ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN 21831da177e4SLinus Torvalds unsigned long 218443cca0b1SYang Fan arch_get_unmapped_area_topdown(struct file *filp, unsigned long addr, 218543cca0b1SYang Fan unsigned long len, unsigned long pgoff, 218643cca0b1SYang Fan unsigned long flags) 21871da177e4SLinus Torvalds { 21881be7107fSHugh Dickins struct vm_area_struct *vma, *prev; 21891da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 2190db4fbfb9SMichel Lespinasse struct vm_unmapped_area_info info; 2191f6795053SSteve Capper const unsigned long mmap_end = arch_get_mmap_end(addr); 21921da177e4SLinus Torvalds 21931da177e4SLinus Torvalds /* requested length too big for entire address space */ 2194f6795053SSteve Capper if (len > mmap_end - mmap_min_addr) 21951da177e4SLinus Torvalds return -ENOMEM; 21961da177e4SLinus Torvalds 219706abdfb4SBenjamin Herrenschmidt if (flags & MAP_FIXED) 219806abdfb4SBenjamin Herrenschmidt return addr; 219906abdfb4SBenjamin Herrenschmidt 22001da177e4SLinus Torvalds /* requesting a specific address */ 22011da177e4SLinus Torvalds if (addr) { 22021da177e4SLinus Torvalds addr = PAGE_ALIGN(addr); 22031be7107fSHugh Dickins vma = find_vma_prev(mm, addr, &prev); 2204f6795053SSteve Capper if (mmap_end - len >= addr && addr >= mmap_min_addr && 22051be7107fSHugh Dickins (!vma || addr + len <= vm_start_gap(vma)) && 22061be7107fSHugh Dickins (!prev || addr >= vm_end_gap(prev))) 22071da177e4SLinus Torvalds return addr; 22081da177e4SLinus Torvalds } 22091da177e4SLinus Torvalds 2210db4fbfb9SMichel Lespinasse info.flags = VM_UNMAPPED_AREA_TOPDOWN; 2211db4fbfb9SMichel Lespinasse info.length = len; 22122afc745fSAkira Takeuchi info.low_limit = max(PAGE_SIZE, mmap_min_addr); 2213f6795053SSteve Capper info.high_limit = arch_get_mmap_base(addr, mm->mmap_base); 2214db4fbfb9SMichel Lespinasse info.align_mask = 0; 221509ef5283SJaewon Kim info.align_offset = 0; 2216db4fbfb9SMichel Lespinasse addr = vm_unmapped_area(&info); 2217b716ad95SXiao Guangrong 22181da177e4SLinus Torvalds /* 22191da177e4SLinus Torvalds * A failed mmap() very likely causes application failure, 22201da177e4SLinus Torvalds * so fall back to the bottom-up function here. This scenario 22211da177e4SLinus Torvalds * can happen with large stack limits and large mmap() 22221da177e4SLinus Torvalds * allocations. 22231da177e4SLinus Torvalds */ 2224de1741a1SAlexander Kuleshov if (offset_in_page(addr)) { 2225db4fbfb9SMichel Lespinasse VM_BUG_ON(addr != -ENOMEM); 2226db4fbfb9SMichel Lespinasse info.flags = 0; 2227db4fbfb9SMichel Lespinasse info.low_limit = TASK_UNMAPPED_BASE; 2228f6795053SSteve Capper info.high_limit = mmap_end; 2229db4fbfb9SMichel Lespinasse addr = vm_unmapped_area(&info); 2230db4fbfb9SMichel Lespinasse } 22311da177e4SLinus Torvalds 22321da177e4SLinus Torvalds return addr; 22331da177e4SLinus Torvalds } 22341da177e4SLinus Torvalds #endif 22351da177e4SLinus Torvalds 22361da177e4SLinus Torvalds unsigned long 22371da177e4SLinus Torvalds get_unmapped_area(struct file *file, unsigned long addr, unsigned long len, 22381da177e4SLinus Torvalds unsigned long pgoff, unsigned long flags) 22391da177e4SLinus Torvalds { 224006abdfb4SBenjamin Herrenschmidt unsigned long (*get_area)(struct file *, unsigned long, 224106abdfb4SBenjamin Herrenschmidt unsigned long, unsigned long, unsigned long); 224207ab67c8SLinus Torvalds 22439206de95SAl Viro unsigned long error = arch_mmap_check(addr, len, flags); 22449206de95SAl Viro if (error) 22459206de95SAl Viro return error; 22469206de95SAl Viro 22479206de95SAl Viro /* Careful about overflows.. */ 22489206de95SAl Viro if (len > TASK_SIZE) 22499206de95SAl Viro return -ENOMEM; 22509206de95SAl Viro 225107ab67c8SLinus Torvalds get_area = current->mm->get_unmapped_area; 2252c01d5b30SHugh Dickins if (file) { 2253c01d5b30SHugh Dickins if (file->f_op->get_unmapped_area) 225407ab67c8SLinus Torvalds get_area = file->f_op->get_unmapped_area; 2255c01d5b30SHugh Dickins } else if (flags & MAP_SHARED) { 2256c01d5b30SHugh Dickins /* 2257c01d5b30SHugh Dickins * mmap_region() will call shmem_zero_setup() to create a file, 2258c01d5b30SHugh Dickins * so use shmem's get_unmapped_area in case it can be huge. 225945e55300SPeter Collingbourne * do_mmap() will clear pgoff, so match alignment. 2260c01d5b30SHugh Dickins */ 2261c01d5b30SHugh Dickins pgoff = 0; 2262c01d5b30SHugh Dickins get_area = shmem_get_unmapped_area; 2263c01d5b30SHugh Dickins } 2264c01d5b30SHugh Dickins 226507ab67c8SLinus Torvalds addr = get_area(file, addr, len, pgoff, flags); 226607ab67c8SLinus Torvalds if (IS_ERR_VALUE(addr)) 226707ab67c8SLinus Torvalds return addr; 226807ab67c8SLinus Torvalds 22691da177e4SLinus Torvalds if (addr > TASK_SIZE - len) 22701da177e4SLinus Torvalds return -ENOMEM; 2271de1741a1SAlexander Kuleshov if (offset_in_page(addr)) 22721da177e4SLinus Torvalds return -EINVAL; 227306abdfb4SBenjamin Herrenschmidt 22749ac4ed4bSAl Viro error = security_mmap_addr(addr); 22759ac4ed4bSAl Viro return error ? error : addr; 22761da177e4SLinus Torvalds } 22771da177e4SLinus Torvalds 22781da177e4SLinus Torvalds EXPORT_SYMBOL(get_unmapped_area); 22791da177e4SLinus Torvalds 22801da177e4SLinus Torvalds /* Look up the first VMA which satisfies addr < vm_end, NULL if none. */ 22811da177e4SLinus Torvalds struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr) 22821da177e4SLinus Torvalds { 2283615d6e87SDavidlohr Bueso struct rb_node *rb_node; 2284615d6e87SDavidlohr Bueso struct vm_area_struct *vma; 22851da177e4SLinus Torvalds 22865b78ed24SLuigi Rizzo mmap_assert_locked(mm); 22871da177e4SLinus Torvalds /* Check the cache first. */ 2288615d6e87SDavidlohr Bueso vma = vmacache_find(mm, addr); 2289615d6e87SDavidlohr Bueso if (likely(vma)) 2290615d6e87SDavidlohr Bueso return vma; 22911da177e4SLinus Torvalds 22921da177e4SLinus Torvalds rb_node = mm->mm_rb.rb_node; 22931da177e4SLinus Torvalds 22941da177e4SLinus Torvalds while (rb_node) { 2295615d6e87SDavidlohr Bueso struct vm_area_struct *tmp; 22961da177e4SLinus Torvalds 2297615d6e87SDavidlohr Bueso tmp = rb_entry(rb_node, struct vm_area_struct, vm_rb); 22981da177e4SLinus Torvalds 2299615d6e87SDavidlohr Bueso if (tmp->vm_end > addr) { 2300615d6e87SDavidlohr Bueso vma = tmp; 2301615d6e87SDavidlohr Bueso if (tmp->vm_start <= addr) 23021da177e4SLinus Torvalds break; 23031da177e4SLinus Torvalds rb_node = rb_node->rb_left; 23041da177e4SLinus Torvalds } else 23051da177e4SLinus Torvalds rb_node = rb_node->rb_right; 23061da177e4SLinus Torvalds } 2307615d6e87SDavidlohr Bueso 23081da177e4SLinus Torvalds if (vma) 2309615d6e87SDavidlohr Bueso vmacache_update(addr, vma); 23101da177e4SLinus Torvalds return vma; 23111da177e4SLinus Torvalds } 23121da177e4SLinus Torvalds 23131da177e4SLinus Torvalds EXPORT_SYMBOL(find_vma); 23141da177e4SLinus Torvalds 23156bd4837dSKOSAKI Motohiro /* 23166bd4837dSKOSAKI Motohiro * Same as find_vma, but also return a pointer to the previous VMA in *pprev. 23176bd4837dSKOSAKI Motohiro */ 23181da177e4SLinus Torvalds struct vm_area_struct * 23191da177e4SLinus Torvalds find_vma_prev(struct mm_struct *mm, unsigned long addr, 23201da177e4SLinus Torvalds struct vm_area_struct **pprev) 23211da177e4SLinus Torvalds { 23226bd4837dSKOSAKI Motohiro struct vm_area_struct *vma; 23231da177e4SLinus Torvalds 23246bd4837dSKOSAKI Motohiro vma = find_vma(mm, addr); 232583cd904dSMikulas Patocka if (vma) { 232683cd904dSMikulas Patocka *pprev = vma->vm_prev; 232783cd904dSMikulas Patocka } else { 232873848a97SWei Yang struct rb_node *rb_node = rb_last(&mm->mm_rb); 232973848a97SWei Yang 233073848a97SWei Yang *pprev = rb_node ? rb_entry(rb_node, struct vm_area_struct, vm_rb) : NULL; 233183cd904dSMikulas Patocka } 23326bd4837dSKOSAKI Motohiro return vma; 23331da177e4SLinus Torvalds } 23341da177e4SLinus Torvalds 23351da177e4SLinus Torvalds /* 23361da177e4SLinus Torvalds * Verify that the stack growth is acceptable and 23371da177e4SLinus Torvalds * update accounting. This is shared with both the 23381da177e4SLinus Torvalds * grow-up and grow-down cases. 23391da177e4SLinus Torvalds */ 23401be7107fSHugh Dickins static int acct_stack_growth(struct vm_area_struct *vma, 23411be7107fSHugh Dickins unsigned long size, unsigned long grow) 23421da177e4SLinus Torvalds { 23431da177e4SLinus Torvalds struct mm_struct *mm = vma->vm_mm; 23441be7107fSHugh Dickins unsigned long new_start; 23451da177e4SLinus Torvalds 23461da177e4SLinus Torvalds /* address space limit tests */ 234784638335SKonstantin Khlebnikov if (!may_expand_vm(mm, vma->vm_flags, grow)) 23481da177e4SLinus Torvalds return -ENOMEM; 23491da177e4SLinus Torvalds 23501da177e4SLinus Torvalds /* Stack limit test */ 235124c79d8eSKrzysztof Opasiak if (size > rlimit(RLIMIT_STACK)) 23521da177e4SLinus Torvalds return -ENOMEM; 23531da177e4SLinus Torvalds 23541da177e4SLinus Torvalds /* mlock limit tests */ 2355c5d8a364SMiaohe Lin if (mlock_future_check(mm, vma->vm_flags, grow << PAGE_SHIFT)) 23561da177e4SLinus Torvalds return -ENOMEM; 23571da177e4SLinus Torvalds 23580d59a01bSAdam Litke /* Check to ensure the stack will not grow into a hugetlb-only region */ 23590d59a01bSAdam Litke new_start = (vma->vm_flags & VM_GROWSUP) ? vma->vm_start : 23600d59a01bSAdam Litke vma->vm_end - size; 23610d59a01bSAdam Litke if (is_hugepage_only_range(vma->vm_mm, new_start, size)) 23620d59a01bSAdam Litke return -EFAULT; 23630d59a01bSAdam Litke 23641da177e4SLinus Torvalds /* 23651da177e4SLinus Torvalds * Overcommit.. This must be the final test, as it will 23661da177e4SLinus Torvalds * update security statistics. 23671da177e4SLinus Torvalds */ 236805fa199dSHugh Dickins if (security_vm_enough_memory_mm(mm, grow)) 23691da177e4SLinus Torvalds return -ENOMEM; 23701da177e4SLinus Torvalds 23711da177e4SLinus Torvalds return 0; 23721da177e4SLinus Torvalds } 23731da177e4SLinus Torvalds 237446dea3d0SHugh Dickins #if defined(CONFIG_STACK_GROWSUP) || defined(CONFIG_IA64) 23751da177e4SLinus Torvalds /* 237646dea3d0SHugh Dickins * PA-RISC uses this for its stack; IA64 for its Register Backing Store. 237746dea3d0SHugh Dickins * vma is the last one with address > vma->vm_end. Have to extend vma. 23781da177e4SLinus Torvalds */ 237946dea3d0SHugh Dickins int expand_upwards(struct vm_area_struct *vma, unsigned long address) 23801da177e4SLinus Torvalds { 238109357814SOleg Nesterov struct mm_struct *mm = vma->vm_mm; 23821be7107fSHugh Dickins struct vm_area_struct *next; 23831be7107fSHugh Dickins unsigned long gap_addr; 238412352d3cSKonstantin Khlebnikov int error = 0; 23851da177e4SLinus Torvalds 23861da177e4SLinus Torvalds if (!(vma->vm_flags & VM_GROWSUP)) 23871da177e4SLinus Torvalds return -EFAULT; 23881da177e4SLinus Torvalds 2389bd726c90SHelge Deller /* Guard against exceeding limits of the address space. */ 23901be7107fSHugh Dickins address &= PAGE_MASK; 239137511fb5SHelge Deller if (address >= (TASK_SIZE & PAGE_MASK)) 239212352d3cSKonstantin Khlebnikov return -ENOMEM; 2393bd726c90SHelge Deller address += PAGE_SIZE; 239412352d3cSKonstantin Khlebnikov 23951be7107fSHugh Dickins /* Enforce stack_guard_gap */ 23961be7107fSHugh Dickins gap_addr = address + stack_guard_gap; 2397bd726c90SHelge Deller 2398bd726c90SHelge Deller /* Guard against overflow */ 2399bd726c90SHelge Deller if (gap_addr < address || gap_addr > TASK_SIZE) 2400bd726c90SHelge Deller gap_addr = TASK_SIZE; 2401bd726c90SHelge Deller 24021be7107fSHugh Dickins next = vma->vm_next; 24033122e80eSAnshuman Khandual if (next && next->vm_start < gap_addr && vma_is_accessible(next)) { 24041be7107fSHugh Dickins if (!(next->vm_flags & VM_GROWSUP)) 24051be7107fSHugh Dickins return -ENOMEM; 24061be7107fSHugh Dickins /* Check that both stack segments have the same anon_vma? */ 24071be7107fSHugh Dickins } 24081be7107fSHugh Dickins 240912352d3cSKonstantin Khlebnikov /* We must make sure the anon_vma is allocated. */ 24101da177e4SLinus Torvalds if (unlikely(anon_vma_prepare(vma))) 24111da177e4SLinus Torvalds return -ENOMEM; 24121da177e4SLinus Torvalds 24131da177e4SLinus Torvalds /* 24141da177e4SLinus Torvalds * vma->vm_start/vm_end cannot change under us because the caller 2415c1e8d7c6SMichel Lespinasse * is required to hold the mmap_lock in read mode. We need the 24161da177e4SLinus Torvalds * anon_vma lock to serialize against concurrent expand_stacks. 24171da177e4SLinus Torvalds */ 241812352d3cSKonstantin Khlebnikov anon_vma_lock_write(vma->anon_vma); 24191da177e4SLinus Torvalds 24201da177e4SLinus Torvalds /* Somebody else might have raced and expanded it already */ 24211da177e4SLinus Torvalds if (address > vma->vm_end) { 24221da177e4SLinus Torvalds unsigned long size, grow; 24231da177e4SLinus Torvalds 24241da177e4SLinus Torvalds size = address - vma->vm_start; 24251da177e4SLinus Torvalds grow = (address - vma->vm_end) >> PAGE_SHIFT; 24261da177e4SLinus Torvalds 242742c36f63SHugh Dickins error = -ENOMEM; 242842c36f63SHugh Dickins if (vma->vm_pgoff + (size >> PAGE_SHIFT) >= vma->vm_pgoff) { 24291da177e4SLinus Torvalds error = acct_stack_growth(vma, size, grow); 24303af9e859SEric B Munson if (!error) { 24314128997bSMichel Lespinasse /* 24324128997bSMichel Lespinasse * vma_gap_update() doesn't support concurrent 2433c1e8d7c6SMichel Lespinasse * updates, but we only hold a shared mmap_lock 24344128997bSMichel Lespinasse * lock here, so we need to protect against 24354128997bSMichel Lespinasse * concurrent vma expansions. 243612352d3cSKonstantin Khlebnikov * anon_vma_lock_write() doesn't help here, as 24374128997bSMichel Lespinasse * we don't guarantee that all growable vmas 24384128997bSMichel Lespinasse * in a mm share the same root anon vma. 24394128997bSMichel Lespinasse * So, we reuse mm->page_table_lock to guard 24404128997bSMichel Lespinasse * against concurrent vma expansions. 24414128997bSMichel Lespinasse */ 244209357814SOleg Nesterov spin_lock(&mm->page_table_lock); 244387e8827bSOleg Nesterov if (vma->vm_flags & VM_LOCKED) 244409357814SOleg Nesterov mm->locked_vm += grow; 244584638335SKonstantin Khlebnikov vm_stat_account(mm, vma->vm_flags, grow); 2446bf181b9fSMichel Lespinasse anon_vma_interval_tree_pre_update_vma(vma); 24471da177e4SLinus Torvalds vma->vm_end = address; 2448bf181b9fSMichel Lespinasse anon_vma_interval_tree_post_update_vma(vma); 2449d3737187SMichel Lespinasse if (vma->vm_next) 2450d3737187SMichel Lespinasse vma_gap_update(vma->vm_next); 2451d3737187SMichel Lespinasse else 24521be7107fSHugh Dickins mm->highest_vm_end = vm_end_gap(vma); 245309357814SOleg Nesterov spin_unlock(&mm->page_table_lock); 24544128997bSMichel Lespinasse 24553af9e859SEric B Munson perf_event_mmap(vma); 24563af9e859SEric B Munson } 24571da177e4SLinus Torvalds } 245842c36f63SHugh Dickins } 245912352d3cSKonstantin Khlebnikov anon_vma_unlock_write(vma->anon_vma); 2460c791576cSYang Shi khugepaged_enter_vma(vma, vma->vm_flags); 246109357814SOleg Nesterov validate_mm(mm); 24621da177e4SLinus Torvalds return error; 24631da177e4SLinus Torvalds } 246446dea3d0SHugh Dickins #endif /* CONFIG_STACK_GROWSUP || CONFIG_IA64 */ 246546dea3d0SHugh Dickins 24661da177e4SLinus Torvalds /* 24671da177e4SLinus Torvalds * vma is the first one with address < vma->vm_start. Have to extend vma. 24681da177e4SLinus Torvalds */ 2469d05f3169SMichal Hocko int expand_downwards(struct vm_area_struct *vma, 2470b6a2fea3SOllie Wild unsigned long address) 24711da177e4SLinus Torvalds { 247209357814SOleg Nesterov struct mm_struct *mm = vma->vm_mm; 24731be7107fSHugh Dickins struct vm_area_struct *prev; 24740a1d5299SJann Horn int error = 0; 24751da177e4SLinus Torvalds 24768869477aSEric Paris address &= PAGE_MASK; 24770a1d5299SJann Horn if (address < mmap_min_addr) 24780a1d5299SJann Horn return -EPERM; 24798869477aSEric Paris 24801be7107fSHugh Dickins /* Enforce stack_guard_gap */ 24811be7107fSHugh Dickins prev = vma->vm_prev; 24821be7107fSHugh Dickins /* Check that both stack segments have the same anon_vma? */ 248332e4e6d5SOleg Nesterov if (prev && !(prev->vm_flags & VM_GROWSDOWN) && 24843122e80eSAnshuman Khandual vma_is_accessible(prev)) { 248532e4e6d5SOleg Nesterov if (address - prev->vm_end < stack_guard_gap) 248632e4e6d5SOleg Nesterov return -ENOMEM; 24871be7107fSHugh Dickins } 24881be7107fSHugh Dickins 248912352d3cSKonstantin Khlebnikov /* We must make sure the anon_vma is allocated. */ 249012352d3cSKonstantin Khlebnikov if (unlikely(anon_vma_prepare(vma))) 249112352d3cSKonstantin Khlebnikov return -ENOMEM; 24921da177e4SLinus Torvalds 24931da177e4SLinus Torvalds /* 24941da177e4SLinus Torvalds * vma->vm_start/vm_end cannot change under us because the caller 2495c1e8d7c6SMichel Lespinasse * is required to hold the mmap_lock in read mode. We need the 24961da177e4SLinus Torvalds * anon_vma lock to serialize against concurrent expand_stacks. 24971da177e4SLinus Torvalds */ 249812352d3cSKonstantin Khlebnikov anon_vma_lock_write(vma->anon_vma); 24991da177e4SLinus Torvalds 25001da177e4SLinus Torvalds /* Somebody else might have raced and expanded it already */ 25011da177e4SLinus Torvalds if (address < vma->vm_start) { 25021da177e4SLinus Torvalds unsigned long size, grow; 25031da177e4SLinus Torvalds 25041da177e4SLinus Torvalds size = vma->vm_end - address; 25051da177e4SLinus Torvalds grow = (vma->vm_start - address) >> PAGE_SHIFT; 25061da177e4SLinus Torvalds 2507a626ca6aSLinus Torvalds error = -ENOMEM; 2508a626ca6aSLinus Torvalds if (grow <= vma->vm_pgoff) { 25091da177e4SLinus Torvalds error = acct_stack_growth(vma, size, grow); 25101da177e4SLinus Torvalds if (!error) { 25114128997bSMichel Lespinasse /* 25124128997bSMichel Lespinasse * vma_gap_update() doesn't support concurrent 2513c1e8d7c6SMichel Lespinasse * updates, but we only hold a shared mmap_lock 25144128997bSMichel Lespinasse * lock here, so we need to protect against 25154128997bSMichel Lespinasse * concurrent vma expansions. 251612352d3cSKonstantin Khlebnikov * anon_vma_lock_write() doesn't help here, as 25174128997bSMichel Lespinasse * we don't guarantee that all growable vmas 25184128997bSMichel Lespinasse * in a mm share the same root anon vma. 25194128997bSMichel Lespinasse * So, we reuse mm->page_table_lock to guard 25204128997bSMichel Lespinasse * against concurrent vma expansions. 25214128997bSMichel Lespinasse */ 252209357814SOleg Nesterov spin_lock(&mm->page_table_lock); 252387e8827bSOleg Nesterov if (vma->vm_flags & VM_LOCKED) 252409357814SOleg Nesterov mm->locked_vm += grow; 252584638335SKonstantin Khlebnikov vm_stat_account(mm, vma->vm_flags, grow); 2526bf181b9fSMichel Lespinasse anon_vma_interval_tree_pre_update_vma(vma); 25271da177e4SLinus Torvalds vma->vm_start = address; 25281da177e4SLinus Torvalds vma->vm_pgoff -= grow; 2529bf181b9fSMichel Lespinasse anon_vma_interval_tree_post_update_vma(vma); 2530d3737187SMichel Lespinasse vma_gap_update(vma); 253109357814SOleg Nesterov spin_unlock(&mm->page_table_lock); 25324128997bSMichel Lespinasse 25333af9e859SEric B Munson perf_event_mmap(vma); 25341da177e4SLinus Torvalds } 25351da177e4SLinus Torvalds } 2536a626ca6aSLinus Torvalds } 253712352d3cSKonstantin Khlebnikov anon_vma_unlock_write(vma->anon_vma); 2538c791576cSYang Shi khugepaged_enter_vma(vma, vma->vm_flags); 253909357814SOleg Nesterov validate_mm(mm); 25401da177e4SLinus Torvalds return error; 25411da177e4SLinus Torvalds } 25421da177e4SLinus Torvalds 25431be7107fSHugh Dickins /* enforced gap between the expanding stack and other mappings. */ 25441be7107fSHugh Dickins unsigned long stack_guard_gap = 256UL<<PAGE_SHIFT; 25451be7107fSHugh Dickins 25461be7107fSHugh Dickins static int __init cmdline_parse_stack_guard_gap(char *p) 25471be7107fSHugh Dickins { 25481be7107fSHugh Dickins unsigned long val; 25491be7107fSHugh Dickins char *endptr; 25501be7107fSHugh Dickins 25511be7107fSHugh Dickins val = simple_strtoul(p, &endptr, 10); 25521be7107fSHugh Dickins if (!*endptr) 25531be7107fSHugh Dickins stack_guard_gap = val << PAGE_SHIFT; 25541be7107fSHugh Dickins 2555e6d09493SRandy Dunlap return 1; 25561be7107fSHugh Dickins } 25571be7107fSHugh Dickins __setup("stack_guard_gap=", cmdline_parse_stack_guard_gap); 25581be7107fSHugh Dickins 2559b6a2fea3SOllie Wild #ifdef CONFIG_STACK_GROWSUP 2560b6a2fea3SOllie Wild int expand_stack(struct vm_area_struct *vma, unsigned long address) 2561b6a2fea3SOllie Wild { 2562b6a2fea3SOllie Wild return expand_upwards(vma, address); 2563b6a2fea3SOllie Wild } 2564b6a2fea3SOllie Wild 2565b6a2fea3SOllie Wild struct vm_area_struct * 2566b6a2fea3SOllie Wild find_extend_vma(struct mm_struct *mm, unsigned long addr) 2567b6a2fea3SOllie Wild { 2568b6a2fea3SOllie Wild struct vm_area_struct *vma, *prev; 2569b6a2fea3SOllie Wild 2570b6a2fea3SOllie Wild addr &= PAGE_MASK; 2571b6a2fea3SOllie Wild vma = find_vma_prev(mm, addr, &prev); 2572b6a2fea3SOllie Wild if (vma && (vma->vm_start <= addr)) 2573b6a2fea3SOllie Wild return vma; 257404f5866eSAndrea Arcangeli /* don't alter vm_end if the coredump is running */ 25754d45e75aSJann Horn if (!prev || expand_stack(prev, addr)) 2576b6a2fea3SOllie Wild return NULL; 2577cea10a19SMichel Lespinasse if (prev->vm_flags & VM_LOCKED) 2578fc05f566SKirill A. Shutemov populate_vma_page_range(prev, addr, prev->vm_end, NULL); 2579b6a2fea3SOllie Wild return prev; 2580b6a2fea3SOllie Wild } 2581b6a2fea3SOllie Wild #else 2582b6a2fea3SOllie Wild int expand_stack(struct vm_area_struct *vma, unsigned long address) 2583b6a2fea3SOllie Wild { 2584b6a2fea3SOllie Wild return expand_downwards(vma, address); 2585b6a2fea3SOllie Wild } 2586b6a2fea3SOllie Wild 25871da177e4SLinus Torvalds struct vm_area_struct * 25881da177e4SLinus Torvalds find_extend_vma(struct mm_struct *mm, unsigned long addr) 25891da177e4SLinus Torvalds { 25901da177e4SLinus Torvalds struct vm_area_struct *vma; 25911da177e4SLinus Torvalds unsigned long start; 25921da177e4SLinus Torvalds 25931da177e4SLinus Torvalds addr &= PAGE_MASK; 25941da177e4SLinus Torvalds vma = find_vma(mm, addr); 25951da177e4SLinus Torvalds if (!vma) 25961da177e4SLinus Torvalds return NULL; 25971da177e4SLinus Torvalds if (vma->vm_start <= addr) 25981da177e4SLinus Torvalds return vma; 25991da177e4SLinus Torvalds if (!(vma->vm_flags & VM_GROWSDOWN)) 26001da177e4SLinus Torvalds return NULL; 26011da177e4SLinus Torvalds start = vma->vm_start; 26021da177e4SLinus Torvalds if (expand_stack(vma, addr)) 26031da177e4SLinus Torvalds return NULL; 2604cea10a19SMichel Lespinasse if (vma->vm_flags & VM_LOCKED) 2605fc05f566SKirill A. Shutemov populate_vma_page_range(vma, addr, start, NULL); 26061da177e4SLinus Torvalds return vma; 26071da177e4SLinus Torvalds } 26081da177e4SLinus Torvalds #endif 26091da177e4SLinus Torvalds 2610e1d6d01aSJesse Barnes EXPORT_SYMBOL_GPL(find_extend_vma); 2611e1d6d01aSJesse Barnes 26122c0b3814SHugh Dickins /* 26132c0b3814SHugh Dickins * Ok - we have the memory areas we should free on the vma list, 26142c0b3814SHugh Dickins * so release them, and do the vma updates. 26151da177e4SLinus Torvalds * 26162c0b3814SHugh Dickins * Called with the mm semaphore held. 26171da177e4SLinus Torvalds */ 26182c0b3814SHugh Dickins static void remove_vma_list(struct mm_struct *mm, struct vm_area_struct *vma) 26191da177e4SLinus Torvalds { 26204f74d2c8SLinus Torvalds unsigned long nr_accounted = 0; 26214f74d2c8SLinus Torvalds 2622365e9c87SHugh Dickins /* Update high watermark before we lower total_vm */ 2623365e9c87SHugh Dickins update_hiwater_vm(mm); 26242c0b3814SHugh Dickins do { 2625ab50b8edSHugh Dickins long nrpages = vma_pages(vma); 26261da177e4SLinus Torvalds 26274f74d2c8SLinus Torvalds if (vma->vm_flags & VM_ACCOUNT) 26284f74d2c8SLinus Torvalds nr_accounted += nrpages; 262984638335SKonstantin Khlebnikov vm_stat_account(mm, vma->vm_flags, -nrpages); 2630a8fb5618SHugh Dickins vma = remove_vma(vma); 2631146425a3SHugh Dickins } while (vma); 26324f74d2c8SLinus Torvalds vm_unacct_memory(nr_accounted); 26331da177e4SLinus Torvalds validate_mm(mm); 26341da177e4SLinus Torvalds } 26351da177e4SLinus Torvalds 26361da177e4SLinus Torvalds /* 26371da177e4SLinus Torvalds * Get rid of page table information in the indicated region. 26381da177e4SLinus Torvalds * 2639f10df686SPaolo 'Blaisorblade' Giarrusso * Called with the mm semaphore held. 26401da177e4SLinus Torvalds */ 26411da177e4SLinus Torvalds static void unmap_region(struct mm_struct *mm, 2642e0da382cSHugh Dickins struct vm_area_struct *vma, struct vm_area_struct *prev, 2643e0da382cSHugh Dickins unsigned long start, unsigned long end) 26441da177e4SLinus Torvalds { 26453903b55aSLiam R. Howlett struct vm_area_struct *next = vma_next(mm, prev); 2646d16dfc55SPeter Zijlstra struct mmu_gather tlb; 26471da177e4SLinus Torvalds 26481da177e4SLinus Torvalds lru_add_drain(); 2649a72afd87SWill Deacon tlb_gather_mmu(&tlb, mm); 2650365e9c87SHugh Dickins update_hiwater_rss(mm); 26514f74d2c8SLinus Torvalds unmap_vmas(&tlb, vma, start, end); 2652d16dfc55SPeter Zijlstra free_pgtables(&tlb, vma, prev ? prev->vm_end : FIRST_USER_ADDRESS, 26536ee8630eSHugh Dickins next ? next->vm_start : USER_PGTABLES_CEILING); 2654ae8eba8bSWill Deacon tlb_finish_mmu(&tlb); 26551da177e4SLinus Torvalds } 26561da177e4SLinus Torvalds 26571da177e4SLinus Torvalds /* 26581da177e4SLinus Torvalds * Create a list of vma's touched by the unmap, removing them from the mm's 26591da177e4SLinus Torvalds * vma list as we go.. 26601da177e4SLinus Torvalds */ 2661246c320aSKirill A. Shutemov static bool 26621da177e4SLinus Torvalds detach_vmas_to_be_unmapped(struct mm_struct *mm, struct vm_area_struct *vma, 26631da177e4SLinus Torvalds struct vm_area_struct *prev, unsigned long end) 26641da177e4SLinus Torvalds { 26651da177e4SLinus Torvalds struct vm_area_struct **insertion_point; 26661da177e4SLinus Torvalds struct vm_area_struct *tail_vma = NULL; 26671da177e4SLinus Torvalds 26681da177e4SLinus Torvalds insertion_point = (prev ? &prev->vm_next : &mm->mmap); 2669297c5eeeSLinus Torvalds vma->vm_prev = NULL; 26701da177e4SLinus Torvalds do { 2671d3737187SMichel Lespinasse vma_rb_erase(vma, &mm->mm_rb); 2672a213e5cfSHugh Dickins if (vma->vm_flags & VM_LOCKED) 2673a213e5cfSHugh Dickins mm->locked_vm -= vma_pages(vma); 26741da177e4SLinus Torvalds mm->map_count--; 26751da177e4SLinus Torvalds tail_vma = vma; 26761da177e4SLinus Torvalds vma = vma->vm_next; 26771da177e4SLinus Torvalds } while (vma && vma->vm_start < end); 26781da177e4SLinus Torvalds *insertion_point = vma; 2679d3737187SMichel Lespinasse if (vma) { 2680297c5eeeSLinus Torvalds vma->vm_prev = prev; 2681d3737187SMichel Lespinasse vma_gap_update(vma); 2682d3737187SMichel Lespinasse } else 26831be7107fSHugh Dickins mm->highest_vm_end = prev ? vm_end_gap(prev) : 0; 26841da177e4SLinus Torvalds tail_vma->vm_next = NULL; 2685615d6e87SDavidlohr Bueso 2686615d6e87SDavidlohr Bueso /* Kill the cache */ 2687615d6e87SDavidlohr Bueso vmacache_invalidate(mm); 2688246c320aSKirill A. Shutemov 2689246c320aSKirill A. Shutemov /* 2690246c320aSKirill A. Shutemov * Do not downgrade mmap_lock if we are next to VM_GROWSDOWN or 2691246c320aSKirill A. Shutemov * VM_GROWSUP VMA. Such VMAs can change their size under 2692246c320aSKirill A. Shutemov * down_read(mmap_lock) and collide with the VMA we are about to unmap. 2693246c320aSKirill A. Shutemov */ 2694246c320aSKirill A. Shutemov if (vma && (vma->vm_flags & VM_GROWSDOWN)) 2695246c320aSKirill A. Shutemov return false; 2696246c320aSKirill A. Shutemov if (prev && (prev->vm_flags & VM_GROWSUP)) 2697246c320aSKirill A. Shutemov return false; 2698246c320aSKirill A. Shutemov return true; 26991da177e4SLinus Torvalds } 27001da177e4SLinus Torvalds 27011da177e4SLinus Torvalds /* 2702def5efe0SDavid Rientjes * __split_vma() bypasses sysctl_max_map_count checking. We use this where it 2703def5efe0SDavid Rientjes * has already been checked or doesn't make sense to fail. 27041da177e4SLinus Torvalds */ 2705def5efe0SDavid Rientjes int __split_vma(struct mm_struct *mm, struct vm_area_struct *vma, 27061da177e4SLinus Torvalds unsigned long addr, int new_below) 27071da177e4SLinus Torvalds { 27081da177e4SLinus Torvalds struct vm_area_struct *new; 2709e3975891SChen Gang int err; 27101da177e4SLinus Torvalds 2711dd3b614fSDmitry Safonov if (vma->vm_ops && vma->vm_ops->may_split) { 2712dd3b614fSDmitry Safonov err = vma->vm_ops->may_split(vma, addr); 271331383c68SDan Williams if (err) 271431383c68SDan Williams return err; 271531383c68SDan Williams } 27161da177e4SLinus Torvalds 27173928d4f5SLinus Torvalds new = vm_area_dup(vma); 27181da177e4SLinus Torvalds if (!new) 2719e3975891SChen Gang return -ENOMEM; 27201da177e4SLinus Torvalds 27211da177e4SLinus Torvalds if (new_below) 27221da177e4SLinus Torvalds new->vm_end = addr; 27231da177e4SLinus Torvalds else { 27241da177e4SLinus Torvalds new->vm_start = addr; 27251da177e4SLinus Torvalds new->vm_pgoff += ((addr - vma->vm_start) >> PAGE_SHIFT); 27261da177e4SLinus Torvalds } 27271da177e4SLinus Torvalds 2728ef0855d3SOleg Nesterov err = vma_dup_policy(vma, new); 2729ef0855d3SOleg Nesterov if (err) 27305beb4930SRik van Riel goto out_free_vma; 27311da177e4SLinus Torvalds 2732c4ea95d7SDaniel Forrest err = anon_vma_clone(new, vma); 2733c4ea95d7SDaniel Forrest if (err) 27345beb4930SRik van Riel goto out_free_mpol; 27355beb4930SRik van Riel 2736e9714acfSKonstantin Khlebnikov if (new->vm_file) 27371da177e4SLinus Torvalds get_file(new->vm_file); 27381da177e4SLinus Torvalds 27391da177e4SLinus Torvalds if (new->vm_ops && new->vm_ops->open) 27401da177e4SLinus Torvalds new->vm_ops->open(new); 27411da177e4SLinus Torvalds 27421da177e4SLinus Torvalds if (new_below) 27435beb4930SRik van Riel err = vma_adjust(vma, addr, vma->vm_end, vma->vm_pgoff + 27441da177e4SLinus Torvalds ((addr - new->vm_start) >> PAGE_SHIFT), new); 27451da177e4SLinus Torvalds else 27465beb4930SRik van Riel err = vma_adjust(vma, vma->vm_start, addr, vma->vm_pgoff, new); 27471da177e4SLinus Torvalds 27485beb4930SRik van Riel /* Success. */ 27495beb4930SRik van Riel if (!err) 27501da177e4SLinus Torvalds return 0; 27515beb4930SRik van Riel 27525beb4930SRik van Riel /* Clean everything up if vma_adjust failed. */ 275358927533SRik van Riel if (new->vm_ops && new->vm_ops->close) 27545beb4930SRik van Riel new->vm_ops->close(new); 2755e9714acfSKonstantin Khlebnikov if (new->vm_file) 27565beb4930SRik van Riel fput(new->vm_file); 27572aeadc30SAndrea Arcangeli unlink_anon_vmas(new); 27585beb4930SRik van Riel out_free_mpol: 2759ef0855d3SOleg Nesterov mpol_put(vma_policy(new)); 27605beb4930SRik van Riel out_free_vma: 27613928d4f5SLinus Torvalds vm_area_free(new); 27625beb4930SRik van Riel return err; 27631da177e4SLinus Torvalds } 27641da177e4SLinus Torvalds 2765659ace58SKOSAKI Motohiro /* 2766659ace58SKOSAKI Motohiro * Split a vma into two pieces at address 'addr', a new vma is allocated 2767659ace58SKOSAKI Motohiro * either for the first part or the tail. 2768659ace58SKOSAKI Motohiro */ 2769659ace58SKOSAKI Motohiro int split_vma(struct mm_struct *mm, struct vm_area_struct *vma, 2770659ace58SKOSAKI Motohiro unsigned long addr, int new_below) 2771659ace58SKOSAKI Motohiro { 2772659ace58SKOSAKI Motohiro if (mm->map_count >= sysctl_max_map_count) 2773659ace58SKOSAKI Motohiro return -ENOMEM; 2774659ace58SKOSAKI Motohiro 2775659ace58SKOSAKI Motohiro return __split_vma(mm, vma, addr, new_below); 2776659ace58SKOSAKI Motohiro } 2777659ace58SKOSAKI Motohiro 27781da177e4SLinus Torvalds /* Munmap is split into 2 main parts -- this part which finds 27791da177e4SLinus Torvalds * what needs doing, and the areas themselves, which do the 27801da177e4SLinus Torvalds * work. This now handles partial unmappings. 27811da177e4SLinus Torvalds * Jeremy Fitzhardinge <jeremy@goop.org> 27821da177e4SLinus Torvalds */ 278385a06835SYang Shi int __do_munmap(struct mm_struct *mm, unsigned long start, size_t len, 2784dd2283f2SYang Shi struct list_head *uf, bool downgrade) 27851da177e4SLinus Torvalds { 27861da177e4SLinus Torvalds unsigned long end; 2787146425a3SHugh Dickins struct vm_area_struct *vma, *prev, *last; 27881da177e4SLinus Torvalds 2789de1741a1SAlexander Kuleshov if ((offset_in_page(start)) || start > TASK_SIZE || len > TASK_SIZE-start) 27901da177e4SLinus Torvalds return -EINVAL; 27911da177e4SLinus Torvalds 2792cc71aba3Svishnu.ps len = PAGE_ALIGN(len); 27935a28fc94SDave Hansen end = start + len; 2794cc71aba3Svishnu.ps if (len == 0) 27951da177e4SLinus Torvalds return -EINVAL; 27961da177e4SLinus Torvalds 27975a28fc94SDave Hansen /* 27985a28fc94SDave Hansen * arch_unmap() might do unmaps itself. It must be called 27995a28fc94SDave Hansen * and finish any rbtree manipulation before this code 28005a28fc94SDave Hansen * runs and also starts to manipulate the rbtree. 28015a28fc94SDave Hansen */ 28025a28fc94SDave Hansen arch_unmap(mm, start, end); 28035a28fc94SDave Hansen 280478d9cf60SGonzalo Matias Juarez Tello /* Find the first overlapping VMA where start < vma->vm_end */ 280578d9cf60SGonzalo Matias Juarez Tello vma = find_vma_intersection(mm, start, end); 2806146425a3SHugh Dickins if (!vma) 28071da177e4SLinus Torvalds return 0; 28089be34c9dSLinus Torvalds prev = vma->vm_prev; 28091da177e4SLinus Torvalds 28101da177e4SLinus Torvalds /* 28111da177e4SLinus Torvalds * If we need to split any vma, do it now to save pain later. 28121da177e4SLinus Torvalds * 28131da177e4SLinus Torvalds * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially 28141da177e4SLinus Torvalds * unmapped vm_area_struct will remain in use: so lower split_vma 28151da177e4SLinus Torvalds * places tmp vma above, and higher split_vma places tmp vma below. 28161da177e4SLinus Torvalds */ 2817146425a3SHugh Dickins if (start > vma->vm_start) { 2818659ace58SKOSAKI Motohiro int error; 2819659ace58SKOSAKI Motohiro 2820659ace58SKOSAKI Motohiro /* 2821659ace58SKOSAKI Motohiro * Make sure that map_count on return from munmap() will 2822659ace58SKOSAKI Motohiro * not exceed its limit; but let map_count go just above 2823659ace58SKOSAKI Motohiro * its limit temporarily, to help free resources as expected. 2824659ace58SKOSAKI Motohiro */ 2825659ace58SKOSAKI Motohiro if (end < vma->vm_end && mm->map_count >= sysctl_max_map_count) 2826659ace58SKOSAKI Motohiro return -ENOMEM; 2827659ace58SKOSAKI Motohiro 2828659ace58SKOSAKI Motohiro error = __split_vma(mm, vma, start, 0); 28291da177e4SLinus Torvalds if (error) 28301da177e4SLinus Torvalds return error; 2831146425a3SHugh Dickins prev = vma; 28321da177e4SLinus Torvalds } 28331da177e4SLinus Torvalds 28341da177e4SLinus Torvalds /* Does it split the last one? */ 28351da177e4SLinus Torvalds last = find_vma(mm, end); 28361da177e4SLinus Torvalds if (last && end > last->vm_start) { 2837659ace58SKOSAKI Motohiro int error = __split_vma(mm, last, end, 1); 28381da177e4SLinus Torvalds if (error) 28391da177e4SLinus Torvalds return error; 28401da177e4SLinus Torvalds } 28413903b55aSLiam R. Howlett vma = vma_next(mm, prev); 28421da177e4SLinus Torvalds 28432376dd7cSAndrea Arcangeli if (unlikely(uf)) { 28442376dd7cSAndrea Arcangeli /* 28452376dd7cSAndrea Arcangeli * If userfaultfd_unmap_prep returns an error the vmas 2846f0953a1bSIngo Molnar * will remain split, but userland will get a 28472376dd7cSAndrea Arcangeli * highly unexpected error anyway. This is no 28482376dd7cSAndrea Arcangeli * different than the case where the first of the two 28492376dd7cSAndrea Arcangeli * __split_vma fails, but we don't undo the first 28502376dd7cSAndrea Arcangeli * split, despite we could. This is unlikely enough 28512376dd7cSAndrea Arcangeli * failure that it's not worth optimizing it for. 28522376dd7cSAndrea Arcangeli */ 28532376dd7cSAndrea Arcangeli int error = userfaultfd_unmap_prep(vma, start, end, uf); 28542376dd7cSAndrea Arcangeli if (error) 28552376dd7cSAndrea Arcangeli return error; 28562376dd7cSAndrea Arcangeli } 28572376dd7cSAndrea Arcangeli 2858dd2283f2SYang Shi /* Detach vmas from rbtree */ 2859246c320aSKirill A. Shutemov if (!detach_vmas_to_be_unmapped(mm, vma, prev, end)) 2860246c320aSKirill A. Shutemov downgrade = false; 28611da177e4SLinus Torvalds 2862dd2283f2SYang Shi if (downgrade) 2863d8ed45c5SMichel Lespinasse mmap_write_downgrade(mm); 2864dd2283f2SYang Shi 2865dd2283f2SYang Shi unmap_region(mm, vma, prev, start, end); 2866dd2283f2SYang Shi 28671da177e4SLinus Torvalds /* Fix up all other VM information */ 28682c0b3814SHugh Dickins remove_vma_list(mm, vma); 28691da177e4SLinus Torvalds 2870dd2283f2SYang Shi return downgrade ? 1 : 0; 28711da177e4SLinus Torvalds } 28721da177e4SLinus Torvalds 2873dd2283f2SYang Shi int do_munmap(struct mm_struct *mm, unsigned long start, size_t len, 2874dd2283f2SYang Shi struct list_head *uf) 2875dd2283f2SYang Shi { 2876dd2283f2SYang Shi return __do_munmap(mm, start, len, uf, false); 2877dd2283f2SYang Shi } 2878dd2283f2SYang Shi 2879dd2283f2SYang Shi static int __vm_munmap(unsigned long start, size_t len, bool downgrade) 2880a46ef99dSLinus Torvalds { 2881a46ef99dSLinus Torvalds int ret; 2882bfce281cSAl Viro struct mm_struct *mm = current->mm; 2883897ab3e0SMike Rapoport LIST_HEAD(uf); 2884a46ef99dSLinus Torvalds 2885d8ed45c5SMichel Lespinasse if (mmap_write_lock_killable(mm)) 2886ae798783SMichal Hocko return -EINTR; 2887ae798783SMichal Hocko 2888dd2283f2SYang Shi ret = __do_munmap(mm, start, len, &uf, downgrade); 2889dd2283f2SYang Shi /* 2890c1e8d7c6SMichel Lespinasse * Returning 1 indicates mmap_lock is downgraded. 2891dd2283f2SYang Shi * But 1 is not legal return value of vm_munmap() and munmap(), reset 2892dd2283f2SYang Shi * it to 0 before return. 2893dd2283f2SYang Shi */ 2894dd2283f2SYang Shi if (ret == 1) { 2895d8ed45c5SMichel Lespinasse mmap_read_unlock(mm); 2896dd2283f2SYang Shi ret = 0; 2897dd2283f2SYang Shi } else 2898d8ed45c5SMichel Lespinasse mmap_write_unlock(mm); 2899dd2283f2SYang Shi 2900897ab3e0SMike Rapoport userfaultfd_unmap_complete(mm, &uf); 2901a46ef99dSLinus Torvalds return ret; 2902a46ef99dSLinus Torvalds } 2903dd2283f2SYang Shi 2904dd2283f2SYang Shi int vm_munmap(unsigned long start, size_t len) 2905dd2283f2SYang Shi { 2906dd2283f2SYang Shi return __vm_munmap(start, len, false); 2907dd2283f2SYang Shi } 2908a46ef99dSLinus Torvalds EXPORT_SYMBOL(vm_munmap); 2909a46ef99dSLinus Torvalds 29106a6160a7SHeiko Carstens SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len) 29111da177e4SLinus Torvalds { 2912ce18d171SCatalin Marinas addr = untagged_addr(addr); 2913dd2283f2SYang Shi return __vm_munmap(addr, len, true); 29141da177e4SLinus Torvalds } 29151da177e4SLinus Torvalds 2916c8d78c18SKirill A. Shutemov 2917c8d78c18SKirill A. Shutemov /* 2918c8d78c18SKirill A. Shutemov * Emulation of deprecated remap_file_pages() syscall. 2919c8d78c18SKirill A. Shutemov */ 2920c8d78c18SKirill A. Shutemov SYSCALL_DEFINE5(remap_file_pages, unsigned long, start, unsigned long, size, 2921c8d78c18SKirill A. Shutemov unsigned long, prot, unsigned long, pgoff, unsigned long, flags) 2922c8d78c18SKirill A. Shutemov { 2923c8d78c18SKirill A. Shutemov 2924c8d78c18SKirill A. Shutemov struct mm_struct *mm = current->mm; 2925c8d78c18SKirill A. Shutemov struct vm_area_struct *vma; 2926c8d78c18SKirill A. Shutemov unsigned long populate = 0; 2927c8d78c18SKirill A. Shutemov unsigned long ret = -EINVAL; 2928c8d78c18SKirill A. Shutemov struct file *file; 2929c8d78c18SKirill A. Shutemov 2930ad56b738SMike Rapoport pr_warn_once("%s (%d) uses deprecated remap_file_pages() syscall. See Documentation/vm/remap_file_pages.rst.\n", 2931c8d78c18SKirill A. Shutemov current->comm, current->pid); 2932c8d78c18SKirill A. Shutemov 2933c8d78c18SKirill A. Shutemov if (prot) 2934c8d78c18SKirill A. Shutemov return ret; 2935c8d78c18SKirill A. Shutemov start = start & PAGE_MASK; 2936c8d78c18SKirill A. Shutemov size = size & PAGE_MASK; 2937c8d78c18SKirill A. Shutemov 2938c8d78c18SKirill A. Shutemov if (start + size <= start) 2939c8d78c18SKirill A. Shutemov return ret; 2940c8d78c18SKirill A. Shutemov 2941c8d78c18SKirill A. Shutemov /* Does pgoff wrap? */ 2942c8d78c18SKirill A. Shutemov if (pgoff + (size >> PAGE_SHIFT) < pgoff) 2943c8d78c18SKirill A. Shutemov return ret; 2944c8d78c18SKirill A. Shutemov 2945d8ed45c5SMichel Lespinasse if (mmap_write_lock_killable(mm)) 2946dc0ef0dfSMichal Hocko return -EINTR; 2947dc0ef0dfSMichal Hocko 29489b593cb2SLiam R. Howlett vma = vma_lookup(mm, start); 2949c8d78c18SKirill A. Shutemov 2950c8d78c18SKirill A. Shutemov if (!vma || !(vma->vm_flags & VM_SHARED)) 2951c8d78c18SKirill A. Shutemov goto out; 2952c8d78c18SKirill A. Shutemov 295348f7df32SKirill A. Shutemov if (start + size > vma->vm_end) { 295448f7df32SKirill A. Shutemov struct vm_area_struct *next; 295548f7df32SKirill A. Shutemov 295648f7df32SKirill A. Shutemov for (next = vma->vm_next; next; next = next->vm_next) { 295748f7df32SKirill A. Shutemov /* hole between vmas ? */ 295848f7df32SKirill A. Shutemov if (next->vm_start != next->vm_prev->vm_end) 295948f7df32SKirill A. Shutemov goto out; 296048f7df32SKirill A. Shutemov 296148f7df32SKirill A. Shutemov if (next->vm_file != vma->vm_file) 296248f7df32SKirill A. Shutemov goto out; 296348f7df32SKirill A. Shutemov 296448f7df32SKirill A. Shutemov if (next->vm_flags != vma->vm_flags) 296548f7df32SKirill A. Shutemov goto out; 296648f7df32SKirill A. Shutemov 296748f7df32SKirill A. Shutemov if (start + size <= next->vm_end) 296848f7df32SKirill A. Shutemov break; 296948f7df32SKirill A. Shutemov } 297048f7df32SKirill A. Shutemov 297148f7df32SKirill A. Shutemov if (!next) 2972c8d78c18SKirill A. Shutemov goto out; 2973c8d78c18SKirill A. Shutemov } 2974c8d78c18SKirill A. Shutemov 2975c8d78c18SKirill A. Shutemov prot |= vma->vm_flags & VM_READ ? PROT_READ : 0; 2976c8d78c18SKirill A. Shutemov prot |= vma->vm_flags & VM_WRITE ? PROT_WRITE : 0; 2977c8d78c18SKirill A. Shutemov prot |= vma->vm_flags & VM_EXEC ? PROT_EXEC : 0; 2978c8d78c18SKirill A. Shutemov 2979c8d78c18SKirill A. Shutemov flags &= MAP_NONBLOCK; 2980c8d78c18SKirill A. Shutemov flags |= MAP_SHARED | MAP_FIXED | MAP_POPULATE; 2981fce000b1SLiam Howlett if (vma->vm_flags & VM_LOCKED) 2982c8d78c18SKirill A. Shutemov flags |= MAP_LOCKED; 298348f7df32SKirill A. Shutemov 2984c8d78c18SKirill A. Shutemov file = get_file(vma->vm_file); 298545e55300SPeter Collingbourne ret = do_mmap(vma->vm_file, start, size, 2986897ab3e0SMike Rapoport prot, flags, pgoff, &populate, NULL); 2987c8d78c18SKirill A. Shutemov fput(file); 2988c8d78c18SKirill A. Shutemov out: 2989d8ed45c5SMichel Lespinasse mmap_write_unlock(mm); 2990c8d78c18SKirill A. Shutemov if (populate) 2991c8d78c18SKirill A. Shutemov mm_populate(ret, populate); 2992c8d78c18SKirill A. Shutemov if (!IS_ERR_VALUE(ret)) 2993c8d78c18SKirill A. Shutemov ret = 0; 2994c8d78c18SKirill A. Shutemov return ret; 2995c8d78c18SKirill A. Shutemov } 2996c8d78c18SKirill A. Shutemov 29971da177e4SLinus Torvalds /* 29981da177e4SLinus Torvalds * this is really a simplified "do_mmap". it only handles 29991da177e4SLinus Torvalds * anonymous maps. eventually we may be able to do some 30001da177e4SLinus Torvalds * brk-specific accounting here. 30011da177e4SLinus Torvalds */ 3002bb177a73SMichal Hocko static int do_brk_flags(unsigned long addr, unsigned long len, unsigned long flags, struct list_head *uf) 30031da177e4SLinus Torvalds { 30041da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 30051da177e4SLinus Torvalds struct vm_area_struct *vma, *prev; 30061da177e4SLinus Torvalds struct rb_node **rb_link, *rb_parent; 30071da177e4SLinus Torvalds pgoff_t pgoff = addr >> PAGE_SHIFT; 30083a459756SKirill Korotaev int error; 3009ff68dac6SGaowei Pu unsigned long mapped_addr; 30101da177e4SLinus Torvalds 301116e72e9bSDenys Vlasenko /* Until we need other flags, refuse anything except VM_EXEC. */ 301216e72e9bSDenys Vlasenko if ((flags & (~VM_EXEC)) != 0) 301316e72e9bSDenys Vlasenko return -EINVAL; 301416e72e9bSDenys Vlasenko flags |= VM_DATA_DEFAULT_FLAGS | VM_ACCOUNT | mm->def_flags; 30153a459756SKirill Korotaev 3016ff68dac6SGaowei Pu mapped_addr = get_unmapped_area(NULL, addr, len, 0, MAP_FIXED); 3017ff68dac6SGaowei Pu if (IS_ERR_VALUE(mapped_addr)) 3018ff68dac6SGaowei Pu return mapped_addr; 30193a459756SKirill Korotaev 3020363ee17fSDavidlohr Bueso error = mlock_future_check(mm, mm->def_flags, len); 3021363ee17fSDavidlohr Bueso if (error) 3022363ee17fSDavidlohr Bueso return error; 30231da177e4SLinus Torvalds 3024fb8090b6SLiam R. Howlett /* Clear old maps, set up prev, rb_link, rb_parent, and uf */ 3025fb8090b6SLiam R. Howlett if (munmap_vma_range(mm, addr, len, &prev, &rb_link, &rb_parent, uf)) 30261da177e4SLinus Torvalds return -ENOMEM; 30271da177e4SLinus Torvalds 30281da177e4SLinus Torvalds /* Check against address space limits *after* clearing old maps... */ 302984638335SKonstantin Khlebnikov if (!may_expand_vm(mm, flags, len >> PAGE_SHIFT)) 30301da177e4SLinus Torvalds return -ENOMEM; 30311da177e4SLinus Torvalds 30321da177e4SLinus Torvalds if (mm->map_count > sysctl_max_map_count) 30331da177e4SLinus Torvalds return -ENOMEM; 30341da177e4SLinus Torvalds 3035191c5424SAl Viro if (security_vm_enough_memory_mm(mm, len >> PAGE_SHIFT)) 30361da177e4SLinus Torvalds return -ENOMEM; 30371da177e4SLinus Torvalds 30381da177e4SLinus Torvalds /* Can we just expand an old private anonymous mapping? */ 3039ba470de4SRik van Riel vma = vma_merge(mm, prev, addr, addr + len, flags, 30409a10064fSColin Cross NULL, NULL, pgoff, NULL, NULL_VM_UFFD_CTX, NULL); 3041ba470de4SRik van Riel if (vma) 30421da177e4SLinus Torvalds goto out; 30431da177e4SLinus Torvalds 30441da177e4SLinus Torvalds /* 30451da177e4SLinus Torvalds * create a vma struct for an anonymous mapping 30461da177e4SLinus Torvalds */ 3047490fc053SLinus Torvalds vma = vm_area_alloc(mm); 30481da177e4SLinus Torvalds if (!vma) { 30491da177e4SLinus Torvalds vm_unacct_memory(len >> PAGE_SHIFT); 30501da177e4SLinus Torvalds return -ENOMEM; 30511da177e4SLinus Torvalds } 30521da177e4SLinus Torvalds 3053bfd40eafSKirill A. Shutemov vma_set_anonymous(vma); 30541da177e4SLinus Torvalds vma->vm_start = addr; 30551da177e4SLinus Torvalds vma->vm_end = addr + len; 30561da177e4SLinus Torvalds vma->vm_pgoff = pgoff; 30571da177e4SLinus Torvalds vma->vm_flags = flags; 30583ed75eb8SColy Li vma->vm_page_prot = vm_get_page_prot(flags); 30591da177e4SLinus Torvalds vma_link(mm, vma, prev, rb_link, rb_parent); 30601da177e4SLinus Torvalds out: 30613af9e859SEric B Munson perf_event_mmap(vma); 30621da177e4SLinus Torvalds mm->total_vm += len >> PAGE_SHIFT; 306384638335SKonstantin Khlebnikov mm->data_vm += len >> PAGE_SHIFT; 3064128557ffSMichel Lespinasse if (flags & VM_LOCKED) 3065ba470de4SRik van Riel mm->locked_vm += (len >> PAGE_SHIFT); 3066d9104d1cSCyrill Gorcunov vma->vm_flags |= VM_SOFTDIRTY; 30675d22fc25SLinus Torvalds return 0; 30681da177e4SLinus Torvalds } 30691da177e4SLinus Torvalds 3070bb177a73SMichal Hocko int vm_brk_flags(unsigned long addr, unsigned long request, unsigned long flags) 3071e4eb1ff6SLinus Torvalds { 3072e4eb1ff6SLinus Torvalds struct mm_struct *mm = current->mm; 3073bb177a73SMichal Hocko unsigned long len; 30745d22fc25SLinus Torvalds int ret; 3075128557ffSMichel Lespinasse bool populate; 3076897ab3e0SMike Rapoport LIST_HEAD(uf); 3077e4eb1ff6SLinus Torvalds 3078bb177a73SMichal Hocko len = PAGE_ALIGN(request); 3079bb177a73SMichal Hocko if (len < request) 3080bb177a73SMichal Hocko return -ENOMEM; 3081bb177a73SMichal Hocko if (!len) 3082bb177a73SMichal Hocko return 0; 3083bb177a73SMichal Hocko 3084d8ed45c5SMichel Lespinasse if (mmap_write_lock_killable(mm)) 30852d6c9282SMichal Hocko return -EINTR; 30862d6c9282SMichal Hocko 3087897ab3e0SMike Rapoport ret = do_brk_flags(addr, len, flags, &uf); 3088128557ffSMichel Lespinasse populate = ((mm->def_flags & VM_LOCKED) != 0); 3089d8ed45c5SMichel Lespinasse mmap_write_unlock(mm); 3090897ab3e0SMike Rapoport userfaultfd_unmap_complete(mm, &uf); 30915d22fc25SLinus Torvalds if (populate && !ret) 3092128557ffSMichel Lespinasse mm_populate(addr, len); 3093e4eb1ff6SLinus Torvalds return ret; 3094e4eb1ff6SLinus Torvalds } 309516e72e9bSDenys Vlasenko EXPORT_SYMBOL(vm_brk_flags); 309616e72e9bSDenys Vlasenko 309716e72e9bSDenys Vlasenko int vm_brk(unsigned long addr, unsigned long len) 309816e72e9bSDenys Vlasenko { 309916e72e9bSDenys Vlasenko return vm_brk_flags(addr, len, 0); 310016e72e9bSDenys Vlasenko } 3101e4eb1ff6SLinus Torvalds EXPORT_SYMBOL(vm_brk); 31021da177e4SLinus Torvalds 31031da177e4SLinus Torvalds /* Release all mmaps. */ 31041da177e4SLinus Torvalds void exit_mmap(struct mm_struct *mm) 31051da177e4SLinus Torvalds { 3106d16dfc55SPeter Zijlstra struct mmu_gather tlb; 3107ba470de4SRik van Riel struct vm_area_struct *vma; 31081da177e4SLinus Torvalds unsigned long nr_accounted = 0; 31091da177e4SLinus Torvalds 3110d6dd61c8SJeremy Fitzhardinge /* mm's last user has gone, and its about to be pulled down */ 3111cddb8a5cSAndrea Arcangeli mmu_notifier_release(mm); 3112d6dd61c8SJeremy Fitzhardinge 311327ae357fSDavid Rientjes if (unlikely(mm_is_oom_victim(mm))) { 311427ae357fSDavid Rientjes /* 311527ae357fSDavid Rientjes * Manually reap the mm to free as much memory as possible. 311627ae357fSDavid Rientjes * Then, as the oom reaper does, set MMF_OOM_SKIP to disregard 3117c1e8d7c6SMichel Lespinasse * this mm from further consideration. Taking mm->mmap_lock for 311827ae357fSDavid Rientjes * write after setting MMF_OOM_SKIP will guarantee that the oom 3119c1e8d7c6SMichel Lespinasse * reaper will not run on this mm again after mmap_lock is 312027ae357fSDavid Rientjes * dropped. 312127ae357fSDavid Rientjes * 3122c1e8d7c6SMichel Lespinasse * Nothing can be holding mm->mmap_lock here and the above call 312327ae357fSDavid Rientjes * to mmu_notifier_release(mm) ensures mmu notifier callbacks in 312427ae357fSDavid Rientjes * __oom_reap_task_mm() will not block. 312527ae357fSDavid Rientjes */ 312693065ac7SMichal Hocko (void)__oom_reap_task_mm(mm); 312727ae357fSDavid Rientjes set_bit(MMF_OOM_SKIP, &mm->flags); 312827ae357fSDavid Rientjes } 312927ae357fSDavid Rientjes 313064591e86SSuren Baghdasaryan mmap_write_lock(mm); 31319480c53eSJeremy Fitzhardinge arch_exit_mmap(mm); 31329480c53eSJeremy Fitzhardinge 3133ba470de4SRik van Riel vma = mm->mmap; 313464591e86SSuren Baghdasaryan if (!vma) { 313564591e86SSuren Baghdasaryan /* Can happen if dup_mmap() received an OOM */ 313664591e86SSuren Baghdasaryan mmap_write_unlock(mm); 31379480c53eSJeremy Fitzhardinge return; 313864591e86SSuren Baghdasaryan } 31399480c53eSJeremy Fitzhardinge 31401da177e4SLinus Torvalds lru_add_drain(); 31411da177e4SLinus Torvalds flush_cache_mm(mm); 3142d8b45053SWill Deacon tlb_gather_mmu_fullmm(&tlb, mm); 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); 3147ae8eba8bSWill Deacon tlb_finish_mmu(&tlb); 31481da177e4SLinus Torvalds 314964591e86SSuren Baghdasaryan /* Walk the list again, actually closing and freeing it. */ 31504f74d2c8SLinus Torvalds while (vma) { 31514f74d2c8SLinus Torvalds if (vma->vm_flags & VM_ACCOUNT) 31524f74d2c8SLinus Torvalds nr_accounted += vma_pages(vma); 3153a8fb5618SHugh Dickins vma = remove_vma(vma); 31540a3b3c25SPaul E. McKenney cond_resched(); 31554f74d2c8SLinus Torvalds } 3156f798a1d4SSuren Baghdasaryan mm->mmap = NULL; 315764591e86SSuren Baghdasaryan mmap_write_unlock(mm); 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. 31878332326eSLiao Pingfang * Similarly in do_mmap and in do_brk_flags. 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), 32265c26f6acSSuren Baghdasaryan vma->vm_userfaultfd_ctx, anon_vma_name(vma)); 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 { 33087866076bSPeng Liu WRITE_ONCE(mm->total_vm, READ_ONCE(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 333214d07113SBrian Geffon 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 3345871402e0SDmitry Safonov static int special_mapping_split(struct vm_area_struct *vma, unsigned long addr) 3346871402e0SDmitry Safonov { 3347871402e0SDmitry Safonov /* 3348871402e0SDmitry Safonov * Forbid splitting special mappings - kernel has expectations over 3349871402e0SDmitry Safonov * the number of pages in mapping. Together with VM_DONTEXPAND 3350871402e0SDmitry Safonov * the size of vma should stay the same over the special mapping's 3351871402e0SDmitry Safonov * lifetime. 3352871402e0SDmitry Safonov */ 3353871402e0SDmitry Safonov return -EINVAL; 3354871402e0SDmitry Safonov } 3355871402e0SDmitry Safonov 3356a62c34bdSAndy Lutomirski static const struct vm_operations_struct special_mapping_vmops = { 3357a62c34bdSAndy Lutomirski .close = special_mapping_close, 3358a62c34bdSAndy Lutomirski .fault = special_mapping_fault, 3359b059a453SDmitry Safonov .mremap = special_mapping_mremap, 3360a62c34bdSAndy Lutomirski .name = special_mapping_name, 3361af34ebebSDmitry Safonov /* vDSO code relies that VVAR can't be accessed remotely */ 3362af34ebebSDmitry Safonov .access = NULL, 3363871402e0SDmitry Safonov .may_split = special_mapping_split, 3364a62c34bdSAndy Lutomirski }; 3365a62c34bdSAndy Lutomirski 3366a62c34bdSAndy Lutomirski static const struct vm_operations_struct legacy_special_mapping_vmops = { 3367a62c34bdSAndy Lutomirski .close = special_mapping_close, 3368a62c34bdSAndy Lutomirski .fault = special_mapping_fault, 3369a62c34bdSAndy Lutomirski }; 3370fa5dc22fSRoland McGrath 3371b3ec9f33SSouptick Joarder static vm_fault_t special_mapping_fault(struct vm_fault *vmf) 3372fa5dc22fSRoland McGrath { 337311bac800SDave Jiang struct vm_area_struct *vma = vmf->vma; 3374b1d0e4f5SNick Piggin pgoff_t pgoff; 3375fa5dc22fSRoland McGrath struct page **pages; 3376fa5dc22fSRoland McGrath 3377f872f540SAndy Lutomirski if (vma->vm_ops == &legacy_special_mapping_vmops) { 3378a62c34bdSAndy Lutomirski pages = vma->vm_private_data; 3379f872f540SAndy Lutomirski } else { 3380f872f540SAndy Lutomirski struct vm_special_mapping *sm = vma->vm_private_data; 3381f872f540SAndy Lutomirski 3382f872f540SAndy Lutomirski if (sm->fault) 338311bac800SDave Jiang return sm->fault(sm, vmf->vma, vmf); 3384f872f540SAndy Lutomirski 3385f872f540SAndy Lutomirski pages = sm->pages; 3386f872f540SAndy Lutomirski } 3387a62c34bdSAndy Lutomirski 33888a9cc3b5SOleg Nesterov for (pgoff = vmf->pgoff; pgoff && *pages; ++pages) 3389b1d0e4f5SNick Piggin pgoff--; 3390fa5dc22fSRoland McGrath 3391fa5dc22fSRoland McGrath if (*pages) { 3392fa5dc22fSRoland McGrath struct page *page = *pages; 3393fa5dc22fSRoland McGrath get_page(page); 3394b1d0e4f5SNick Piggin vmf->page = page; 3395b1d0e4f5SNick Piggin return 0; 3396fa5dc22fSRoland McGrath } 3397fa5dc22fSRoland McGrath 3398b1d0e4f5SNick Piggin return VM_FAULT_SIGBUS; 3399fa5dc22fSRoland McGrath } 3400fa5dc22fSRoland McGrath 3401a62c34bdSAndy Lutomirski static struct vm_area_struct *__install_special_mapping( 3402a62c34bdSAndy Lutomirski struct mm_struct *mm, 3403fa5dc22fSRoland McGrath unsigned long addr, unsigned long len, 340427f28b97SChen Gang unsigned long vm_flags, void *priv, 340527f28b97SChen Gang const struct vm_operations_struct *ops) 3406fa5dc22fSRoland McGrath { 3407462e635eSTavis Ormandy int ret; 3408fa5dc22fSRoland McGrath struct vm_area_struct *vma; 3409fa5dc22fSRoland McGrath 3410490fc053SLinus Torvalds vma = vm_area_alloc(mm); 3411fa5dc22fSRoland McGrath if (unlikely(vma == NULL)) 34123935ed6aSStefani Seibold return ERR_PTR(-ENOMEM); 3413fa5dc22fSRoland McGrath 3414fa5dc22fSRoland McGrath vma->vm_start = addr; 3415fa5dc22fSRoland McGrath vma->vm_end = addr + len; 3416fa5dc22fSRoland McGrath 3417d9104d1cSCyrill Gorcunov vma->vm_flags = vm_flags | mm->def_flags | VM_DONTEXPAND | VM_SOFTDIRTY; 34181fc09228SHugh Dickins vma->vm_flags &= VM_LOCKED_CLEAR_MASK; 34193ed75eb8SColy Li vma->vm_page_prot = vm_get_page_prot(vma->vm_flags); 3420fa5dc22fSRoland McGrath 3421a62c34bdSAndy Lutomirski vma->vm_ops = ops; 3422a62c34bdSAndy Lutomirski vma->vm_private_data = priv; 3423fa5dc22fSRoland McGrath 3424462e635eSTavis Ormandy ret = insert_vm_struct(mm, vma); 3425462e635eSTavis Ormandy if (ret) 3426462e635eSTavis Ormandy goto out; 3427fa5dc22fSRoland McGrath 342884638335SKonstantin Khlebnikov vm_stat_account(mm, vma->vm_flags, len >> PAGE_SHIFT); 3429fa5dc22fSRoland McGrath 3430cdd6c482SIngo Molnar perf_event_mmap(vma); 3431089dd79dSPeter Zijlstra 34323935ed6aSStefani Seibold return vma; 3433462e635eSTavis Ormandy 3434462e635eSTavis Ormandy out: 34353928d4f5SLinus Torvalds vm_area_free(vma); 34363935ed6aSStefani Seibold return ERR_PTR(ret); 34373935ed6aSStefani Seibold } 34383935ed6aSStefani Seibold 34392eefd878SDmitry Safonov bool vma_is_special_mapping(const struct vm_area_struct *vma, 34402eefd878SDmitry Safonov const struct vm_special_mapping *sm) 34412eefd878SDmitry Safonov { 34422eefd878SDmitry Safonov return vma->vm_private_data == sm && 34432eefd878SDmitry Safonov (vma->vm_ops == &special_mapping_vmops || 34442eefd878SDmitry Safonov vma->vm_ops == &legacy_special_mapping_vmops); 34452eefd878SDmitry Safonov } 34462eefd878SDmitry Safonov 3447a62c34bdSAndy Lutomirski /* 3448c1e8d7c6SMichel Lespinasse * Called with mm->mmap_lock held for writing. 3449a62c34bdSAndy Lutomirski * Insert a new vma covering the given region, with the given flags. 3450a62c34bdSAndy Lutomirski * Its pages are supplied by the given array of struct page *. 3451a62c34bdSAndy Lutomirski * The array can be shorter than len >> PAGE_SHIFT if it's null-terminated. 3452a62c34bdSAndy Lutomirski * The region past the last page supplied will always produce SIGBUS. 3453a62c34bdSAndy Lutomirski * The array pointer and the pages it points to are assumed to stay alive 3454a62c34bdSAndy Lutomirski * for as long as this mapping might exist. 3455a62c34bdSAndy Lutomirski */ 3456a62c34bdSAndy Lutomirski struct vm_area_struct *_install_special_mapping( 3457a62c34bdSAndy Lutomirski struct mm_struct *mm, 3458a62c34bdSAndy Lutomirski unsigned long addr, unsigned long len, 3459a62c34bdSAndy Lutomirski unsigned long vm_flags, const struct vm_special_mapping *spec) 3460a62c34bdSAndy Lutomirski { 346127f28b97SChen Gang return __install_special_mapping(mm, addr, len, vm_flags, (void *)spec, 346227f28b97SChen Gang &special_mapping_vmops); 3463a62c34bdSAndy Lutomirski } 3464a62c34bdSAndy Lutomirski 34653935ed6aSStefani Seibold int install_special_mapping(struct mm_struct *mm, 34663935ed6aSStefani Seibold unsigned long addr, unsigned long len, 34673935ed6aSStefani Seibold unsigned long vm_flags, struct page **pages) 34683935ed6aSStefani Seibold { 3469a62c34bdSAndy Lutomirski struct vm_area_struct *vma = __install_special_mapping( 347027f28b97SChen Gang mm, addr, len, vm_flags, (void *)pages, 347127f28b97SChen Gang &legacy_special_mapping_vmops); 34723935ed6aSStefani Seibold 347314bd5b45SDuan Jiong return PTR_ERR_OR_ZERO(vma); 3474fa5dc22fSRoland McGrath } 34757906d00cSAndrea Arcangeli 34767906d00cSAndrea Arcangeli static DEFINE_MUTEX(mm_all_locks_mutex); 34777906d00cSAndrea Arcangeli 3478454ed842SPeter Zijlstra static void vm_lock_anon_vma(struct mm_struct *mm, struct anon_vma *anon_vma) 34797906d00cSAndrea Arcangeli { 3480f808c13fSDavidlohr Bueso if (!test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_root.rb_node)) { 34817906d00cSAndrea Arcangeli /* 34827906d00cSAndrea Arcangeli * The LSB of head.next can't change from under us 34837906d00cSAndrea Arcangeli * because we hold the mm_all_locks_mutex. 34847906d00cSAndrea Arcangeli */ 3485da1c55f1SMichel Lespinasse down_write_nest_lock(&anon_vma->root->rwsem, &mm->mmap_lock); 34867906d00cSAndrea Arcangeli /* 34877906d00cSAndrea Arcangeli * We can safely modify head.next after taking the 34885a505085SIngo Molnar * anon_vma->root->rwsem. If some other vma in this mm shares 34897906d00cSAndrea Arcangeli * the same anon_vma we won't take it again. 34907906d00cSAndrea Arcangeli * 34917906d00cSAndrea Arcangeli * No need of atomic instructions here, head.next 34927906d00cSAndrea Arcangeli * can't change from under us thanks to the 34935a505085SIngo Molnar * anon_vma->root->rwsem. 34947906d00cSAndrea Arcangeli */ 34957906d00cSAndrea Arcangeli if (__test_and_set_bit(0, (unsigned long *) 3496f808c13fSDavidlohr Bueso &anon_vma->root->rb_root.rb_root.rb_node)) 34977906d00cSAndrea Arcangeli BUG(); 34987906d00cSAndrea Arcangeli } 34997906d00cSAndrea Arcangeli } 35007906d00cSAndrea Arcangeli 3501454ed842SPeter Zijlstra static void vm_lock_mapping(struct mm_struct *mm, struct address_space *mapping) 35027906d00cSAndrea Arcangeli { 35037906d00cSAndrea Arcangeli if (!test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) { 35047906d00cSAndrea Arcangeli /* 35057906d00cSAndrea Arcangeli * AS_MM_ALL_LOCKS can't change from under us because 35067906d00cSAndrea Arcangeli * we hold the mm_all_locks_mutex. 35077906d00cSAndrea Arcangeli * 35087906d00cSAndrea Arcangeli * Operations on ->flags have to be atomic because 35097906d00cSAndrea Arcangeli * even if AS_MM_ALL_LOCKS is stable thanks to the 35107906d00cSAndrea Arcangeli * mm_all_locks_mutex, there may be other cpus 35117906d00cSAndrea Arcangeli * changing other bitflags in parallel to us. 35127906d00cSAndrea Arcangeli */ 35137906d00cSAndrea Arcangeli if (test_and_set_bit(AS_MM_ALL_LOCKS, &mapping->flags)) 35147906d00cSAndrea Arcangeli BUG(); 3515da1c55f1SMichel Lespinasse down_write_nest_lock(&mapping->i_mmap_rwsem, &mm->mmap_lock); 35167906d00cSAndrea Arcangeli } 35177906d00cSAndrea Arcangeli } 35187906d00cSAndrea Arcangeli 35197906d00cSAndrea Arcangeli /* 35207906d00cSAndrea Arcangeli * This operation locks against the VM for all pte/vma/mm related 35217906d00cSAndrea Arcangeli * operations that could ever happen on a certain mm. This includes 35227906d00cSAndrea Arcangeli * vmtruncate, try_to_unmap, and all page faults. 35237906d00cSAndrea Arcangeli * 3524c1e8d7c6SMichel Lespinasse * The caller must take the mmap_lock in write mode before calling 35257906d00cSAndrea Arcangeli * mm_take_all_locks(). The caller isn't allowed to release the 3526c1e8d7c6SMichel Lespinasse * mmap_lock until mm_drop_all_locks() returns. 35277906d00cSAndrea Arcangeli * 3528c1e8d7c6SMichel Lespinasse * mmap_lock in write mode is required in order to block all operations 35297906d00cSAndrea Arcangeli * that could modify pagetables and free pages without need of 353027ba0644SKirill A. Shutemov * altering the vma layout. It's also needed in write mode to avoid new 35317906d00cSAndrea Arcangeli * anon_vmas to be associated with existing vmas. 35327906d00cSAndrea Arcangeli * 35337906d00cSAndrea Arcangeli * A single task can't take more than one mm_take_all_locks() in a row 35347906d00cSAndrea Arcangeli * or it would deadlock. 35357906d00cSAndrea Arcangeli * 3536bf181b9fSMichel Lespinasse * The LSB in anon_vma->rb_root.rb_node and the AS_MM_ALL_LOCKS bitflag in 35377906d00cSAndrea Arcangeli * mapping->flags avoid to take the same lock twice, if more than one 35387906d00cSAndrea Arcangeli * vma in this mm is backed by the same anon_vma or address_space. 35397906d00cSAndrea Arcangeli * 354088f306b6SKirill A. Shutemov * We take locks in following order, accordingly to comment at beginning 354188f306b6SKirill A. Shutemov * of mm/rmap.c: 354288f306b6SKirill A. Shutemov * - all hugetlbfs_i_mmap_rwsem_key locks (aka mapping->i_mmap_rwsem for 354388f306b6SKirill A. Shutemov * hugetlb mapping); 354488f306b6SKirill A. Shutemov * - all i_mmap_rwsem locks; 354588f306b6SKirill A. Shutemov * - all anon_vma->rwseml 354688f306b6SKirill A. Shutemov * 354788f306b6SKirill A. Shutemov * We can take all locks within these types randomly because the VM code 354888f306b6SKirill A. Shutemov * doesn't nest them and we protected from parallel mm_take_all_locks() by 354988f306b6SKirill A. Shutemov * mm_all_locks_mutex. 35507906d00cSAndrea Arcangeli * 35517906d00cSAndrea Arcangeli * mm_take_all_locks() and mm_drop_all_locks are expensive operations 35527906d00cSAndrea Arcangeli * that may have to take thousand of locks. 35537906d00cSAndrea Arcangeli * 35547906d00cSAndrea Arcangeli * mm_take_all_locks() can fail if it's interrupted by signals. 35557906d00cSAndrea Arcangeli */ 35567906d00cSAndrea Arcangeli int mm_take_all_locks(struct mm_struct *mm) 35577906d00cSAndrea Arcangeli { 35587906d00cSAndrea Arcangeli struct vm_area_struct *vma; 35595beb4930SRik van Riel struct anon_vma_chain *avc; 35607906d00cSAndrea Arcangeli 3561325bca1fSRolf Eike Beer mmap_assert_write_locked(mm); 35627906d00cSAndrea Arcangeli 35637906d00cSAndrea Arcangeli mutex_lock(&mm_all_locks_mutex); 35647906d00cSAndrea Arcangeli 35657906d00cSAndrea Arcangeli for (vma = mm->mmap; vma; vma = vma->vm_next) { 35667906d00cSAndrea Arcangeli if (signal_pending(current)) 35677906d00cSAndrea Arcangeli goto out_unlock; 356888f306b6SKirill A. Shutemov if (vma->vm_file && vma->vm_file->f_mapping && 356988f306b6SKirill A. Shutemov is_vm_hugetlb_page(vma)) 357088f306b6SKirill A. Shutemov vm_lock_mapping(mm, vma->vm_file->f_mapping); 357188f306b6SKirill A. Shutemov } 357288f306b6SKirill A. Shutemov 357388f306b6SKirill A. Shutemov for (vma = mm->mmap; vma; vma = vma->vm_next) { 357488f306b6SKirill A. Shutemov if (signal_pending(current)) 357588f306b6SKirill A. Shutemov goto out_unlock; 357688f306b6SKirill A. Shutemov if (vma->vm_file && vma->vm_file->f_mapping && 357788f306b6SKirill A. Shutemov !is_vm_hugetlb_page(vma)) 3578454ed842SPeter Zijlstra vm_lock_mapping(mm, vma->vm_file->f_mapping); 35797906d00cSAndrea Arcangeli } 35807cd5a02fSPeter Zijlstra 35817cd5a02fSPeter Zijlstra for (vma = mm->mmap; vma; vma = vma->vm_next) { 35827cd5a02fSPeter Zijlstra if (signal_pending(current)) 35837cd5a02fSPeter Zijlstra goto out_unlock; 35847cd5a02fSPeter Zijlstra if (vma->anon_vma) 35855beb4930SRik van Riel list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) 35865beb4930SRik van Riel vm_lock_anon_vma(mm, avc->anon_vma); 35877cd5a02fSPeter Zijlstra } 35887cd5a02fSPeter Zijlstra 3589584cff54SKautuk Consul return 0; 35907906d00cSAndrea Arcangeli 35917906d00cSAndrea Arcangeli out_unlock: 35927906d00cSAndrea Arcangeli mm_drop_all_locks(mm); 3593584cff54SKautuk Consul return -EINTR; 35947906d00cSAndrea Arcangeli } 35957906d00cSAndrea Arcangeli 35967906d00cSAndrea Arcangeli static void vm_unlock_anon_vma(struct anon_vma *anon_vma) 35977906d00cSAndrea Arcangeli { 3598f808c13fSDavidlohr Bueso if (test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_root.rb_node)) { 35997906d00cSAndrea Arcangeli /* 36007906d00cSAndrea Arcangeli * The LSB of head.next can't change to 0 from under 36017906d00cSAndrea Arcangeli * us because we hold the mm_all_locks_mutex. 36027906d00cSAndrea Arcangeli * 36037906d00cSAndrea Arcangeli * We must however clear the bitflag before unlocking 3604bf181b9fSMichel Lespinasse * the vma so the users using the anon_vma->rb_root will 36057906d00cSAndrea Arcangeli * never see our bitflag. 36067906d00cSAndrea Arcangeli * 36077906d00cSAndrea Arcangeli * No need of atomic instructions here, head.next 36087906d00cSAndrea Arcangeli * can't change from under us until we release the 36095a505085SIngo Molnar * anon_vma->root->rwsem. 36107906d00cSAndrea Arcangeli */ 36117906d00cSAndrea Arcangeli if (!__test_and_clear_bit(0, (unsigned long *) 3612f808c13fSDavidlohr Bueso &anon_vma->root->rb_root.rb_root.rb_node)) 36137906d00cSAndrea Arcangeli BUG(); 361408b52706SKonstantin Khlebnikov anon_vma_unlock_write(anon_vma); 36157906d00cSAndrea Arcangeli } 36167906d00cSAndrea Arcangeli } 36177906d00cSAndrea Arcangeli 36187906d00cSAndrea Arcangeli static void vm_unlock_mapping(struct address_space *mapping) 36197906d00cSAndrea Arcangeli { 36207906d00cSAndrea Arcangeli if (test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) { 36217906d00cSAndrea Arcangeli /* 36227906d00cSAndrea Arcangeli * AS_MM_ALL_LOCKS can't change to 0 from under us 36237906d00cSAndrea Arcangeli * because we hold the mm_all_locks_mutex. 36247906d00cSAndrea Arcangeli */ 362583cde9e8SDavidlohr Bueso i_mmap_unlock_write(mapping); 36267906d00cSAndrea Arcangeli if (!test_and_clear_bit(AS_MM_ALL_LOCKS, 36277906d00cSAndrea Arcangeli &mapping->flags)) 36287906d00cSAndrea Arcangeli BUG(); 36297906d00cSAndrea Arcangeli } 36307906d00cSAndrea Arcangeli } 36317906d00cSAndrea Arcangeli 36327906d00cSAndrea Arcangeli /* 3633c1e8d7c6SMichel Lespinasse * The mmap_lock cannot be released by the caller until 36347906d00cSAndrea Arcangeli * mm_drop_all_locks() returns. 36357906d00cSAndrea Arcangeli */ 36367906d00cSAndrea Arcangeli void mm_drop_all_locks(struct mm_struct *mm) 36377906d00cSAndrea Arcangeli { 36387906d00cSAndrea Arcangeli struct vm_area_struct *vma; 36395beb4930SRik van Riel struct anon_vma_chain *avc; 36407906d00cSAndrea Arcangeli 3641325bca1fSRolf Eike Beer mmap_assert_write_locked(mm); 36427906d00cSAndrea Arcangeli BUG_ON(!mutex_is_locked(&mm_all_locks_mutex)); 36437906d00cSAndrea Arcangeli 36447906d00cSAndrea Arcangeli for (vma = mm->mmap; vma; vma = vma->vm_next) { 36457906d00cSAndrea Arcangeli if (vma->anon_vma) 36465beb4930SRik van Riel list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) 36475beb4930SRik van Riel vm_unlock_anon_vma(avc->anon_vma); 36487906d00cSAndrea Arcangeli if (vma->vm_file && vma->vm_file->f_mapping) 36497906d00cSAndrea Arcangeli vm_unlock_mapping(vma->vm_file->f_mapping); 36507906d00cSAndrea Arcangeli } 36517906d00cSAndrea Arcangeli 36527906d00cSAndrea Arcangeli mutex_unlock(&mm_all_locks_mutex); 36537906d00cSAndrea Arcangeli } 36548feae131SDavid Howells 36558feae131SDavid Howells /* 36563edf41d8Sseokhoon.yoon * initialise the percpu counter for VM 36578feae131SDavid Howells */ 36588feae131SDavid Howells void __init mmap_init(void) 36598feae131SDavid Howells { 366000a62ce9SKOSAKI Motohiro int ret; 366100a62ce9SKOSAKI Motohiro 3662908c7f19STejun Heo ret = percpu_counter_init(&vm_committed_as, 0, GFP_KERNEL); 366300a62ce9SKOSAKI Motohiro VM_BUG_ON(ret); 36648feae131SDavid Howells } 3665c9b1d098SAndrew Shewmaker 3666c9b1d098SAndrew Shewmaker /* 3667c9b1d098SAndrew Shewmaker * Initialise sysctl_user_reserve_kbytes. 3668c9b1d098SAndrew Shewmaker * 3669c9b1d098SAndrew Shewmaker * This is intended to prevent a user from starting a single memory hogging 3670c9b1d098SAndrew Shewmaker * process, such that they cannot recover (kill the hog) in OVERCOMMIT_NEVER 3671c9b1d098SAndrew Shewmaker * mode. 3672c9b1d098SAndrew Shewmaker * 3673c9b1d098SAndrew Shewmaker * The default value is min(3% of free memory, 128MB) 3674c9b1d098SAndrew Shewmaker * 128MB is enough to recover with sshd/login, bash, and top/kill. 3675c9b1d098SAndrew Shewmaker */ 36761640879aSAndrew Shewmaker static int init_user_reserve(void) 3677c9b1d098SAndrew Shewmaker { 3678c9b1d098SAndrew Shewmaker unsigned long free_kbytes; 3679c9b1d098SAndrew Shewmaker 3680c41f012aSMichal Hocko free_kbytes = global_zone_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10); 3681c9b1d098SAndrew Shewmaker 3682c9b1d098SAndrew Shewmaker sysctl_user_reserve_kbytes = min(free_kbytes / 32, 1UL << 17); 3683c9b1d098SAndrew Shewmaker return 0; 3684c9b1d098SAndrew Shewmaker } 3685a64fb3cdSPaul Gortmaker subsys_initcall(init_user_reserve); 36864eeab4f5SAndrew Shewmaker 36874eeab4f5SAndrew Shewmaker /* 36884eeab4f5SAndrew Shewmaker * Initialise sysctl_admin_reserve_kbytes. 36894eeab4f5SAndrew Shewmaker * 36904eeab4f5SAndrew Shewmaker * The purpose of sysctl_admin_reserve_kbytes is to allow the sys admin 36914eeab4f5SAndrew Shewmaker * to log in and kill a memory hogging process. 36924eeab4f5SAndrew Shewmaker * 36934eeab4f5SAndrew Shewmaker * Systems with more than 256MB will reserve 8MB, enough to recover 36944eeab4f5SAndrew Shewmaker * with sshd, bash, and top in OVERCOMMIT_GUESS. Smaller systems will 36954eeab4f5SAndrew Shewmaker * only reserve 3% of free pages by default. 36964eeab4f5SAndrew Shewmaker */ 36971640879aSAndrew Shewmaker static int init_admin_reserve(void) 36984eeab4f5SAndrew Shewmaker { 36994eeab4f5SAndrew Shewmaker unsigned long free_kbytes; 37004eeab4f5SAndrew Shewmaker 3701c41f012aSMichal Hocko free_kbytes = global_zone_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10); 37024eeab4f5SAndrew Shewmaker 37034eeab4f5SAndrew Shewmaker sysctl_admin_reserve_kbytes = min(free_kbytes / 32, 1UL << 13); 37044eeab4f5SAndrew Shewmaker return 0; 37054eeab4f5SAndrew Shewmaker } 3706a64fb3cdSPaul Gortmaker subsys_initcall(init_admin_reserve); 37071640879aSAndrew Shewmaker 37081640879aSAndrew Shewmaker /* 37091640879aSAndrew Shewmaker * Reinititalise user and admin reserves if memory is added or removed. 37101640879aSAndrew Shewmaker * 37111640879aSAndrew Shewmaker * The default user reserve max is 128MB, and the default max for the 37121640879aSAndrew Shewmaker * admin reserve is 8MB. These are usually, but not always, enough to 37131640879aSAndrew Shewmaker * enable recovery from a memory hogging process using login/sshd, a shell, 37141640879aSAndrew Shewmaker * and tools like top. It may make sense to increase or even disable the 37151640879aSAndrew Shewmaker * reserve depending on the existence of swap or variations in the recovery 37161640879aSAndrew Shewmaker * tools. So, the admin may have changed them. 37171640879aSAndrew Shewmaker * 37181640879aSAndrew Shewmaker * If memory is added and the reserves have been eliminated or increased above 37191640879aSAndrew Shewmaker * the default max, then we'll trust the admin. 37201640879aSAndrew Shewmaker * 37211640879aSAndrew Shewmaker * If memory is removed and there isn't enough free memory, then we 37221640879aSAndrew Shewmaker * need to reset the reserves. 37231640879aSAndrew Shewmaker * 37241640879aSAndrew Shewmaker * Otherwise keep the reserve set by the admin. 37251640879aSAndrew Shewmaker */ 37261640879aSAndrew Shewmaker static int reserve_mem_notifier(struct notifier_block *nb, 37271640879aSAndrew Shewmaker unsigned long action, void *data) 37281640879aSAndrew Shewmaker { 37291640879aSAndrew Shewmaker unsigned long tmp, free_kbytes; 37301640879aSAndrew Shewmaker 37311640879aSAndrew Shewmaker switch (action) { 37321640879aSAndrew Shewmaker case MEM_ONLINE: 37331640879aSAndrew Shewmaker /* Default max is 128MB. Leave alone if modified by operator. */ 37341640879aSAndrew Shewmaker tmp = sysctl_user_reserve_kbytes; 37351640879aSAndrew Shewmaker if (0 < tmp && tmp < (1UL << 17)) 37361640879aSAndrew Shewmaker init_user_reserve(); 37371640879aSAndrew Shewmaker 37381640879aSAndrew Shewmaker /* Default max is 8MB. Leave alone if modified by operator. */ 37391640879aSAndrew Shewmaker tmp = sysctl_admin_reserve_kbytes; 37401640879aSAndrew Shewmaker if (0 < tmp && tmp < (1UL << 13)) 37411640879aSAndrew Shewmaker init_admin_reserve(); 37421640879aSAndrew Shewmaker 37431640879aSAndrew Shewmaker break; 37441640879aSAndrew Shewmaker case MEM_OFFLINE: 3745c41f012aSMichal Hocko free_kbytes = global_zone_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10); 37461640879aSAndrew Shewmaker 37471640879aSAndrew Shewmaker if (sysctl_user_reserve_kbytes > free_kbytes) { 37481640879aSAndrew Shewmaker init_user_reserve(); 37491640879aSAndrew Shewmaker pr_info("vm.user_reserve_kbytes reset to %lu\n", 37501640879aSAndrew Shewmaker sysctl_user_reserve_kbytes); 37511640879aSAndrew Shewmaker } 37521640879aSAndrew Shewmaker 37531640879aSAndrew Shewmaker if (sysctl_admin_reserve_kbytes > free_kbytes) { 37541640879aSAndrew Shewmaker init_admin_reserve(); 37551640879aSAndrew Shewmaker pr_info("vm.admin_reserve_kbytes reset to %lu\n", 37561640879aSAndrew Shewmaker sysctl_admin_reserve_kbytes); 37571640879aSAndrew Shewmaker } 37581640879aSAndrew Shewmaker break; 37591640879aSAndrew Shewmaker default: 37601640879aSAndrew Shewmaker break; 37611640879aSAndrew Shewmaker } 37621640879aSAndrew Shewmaker return NOTIFY_OK; 37631640879aSAndrew Shewmaker } 37641640879aSAndrew Shewmaker 37651640879aSAndrew Shewmaker static struct notifier_block reserve_mem_nb = { 37661640879aSAndrew Shewmaker .notifier_call = reserve_mem_notifier, 37671640879aSAndrew Shewmaker }; 37681640879aSAndrew Shewmaker 37691640879aSAndrew Shewmaker static int __meminit init_reserve_notifier(void) 37701640879aSAndrew Shewmaker { 37711640879aSAndrew Shewmaker if (register_hotmemory_notifier(&reserve_mem_nb)) 3772b1de0d13SMitchel Humpherys pr_err("Failed registering memory add/remove notifier for admin reserve\n"); 37731640879aSAndrew Shewmaker 37741640879aSAndrew Shewmaker return 0; 37751640879aSAndrew Shewmaker } 3776a64fb3cdSPaul Gortmaker subsys_initcall(init_reserve_notifier); 3777