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> 16615d6e87SDavidlohr Bueso #include <linux/vmacache.h> 171da177e4SLinus Torvalds #include <linux/shm.h> 181da177e4SLinus Torvalds #include <linux/mman.h> 191da177e4SLinus Torvalds #include <linux/pagemap.h> 201da177e4SLinus Torvalds #include <linux/swap.h> 211da177e4SLinus Torvalds #include <linux/syscalls.h> 22c59ede7bSRandy.Dunlap #include <linux/capability.h> 231da177e4SLinus Torvalds #include <linux/init.h> 241da177e4SLinus Torvalds #include <linux/file.h> 251da177e4SLinus Torvalds #include <linux/fs.h> 261da177e4SLinus Torvalds #include <linux/personality.h> 271da177e4SLinus Torvalds #include <linux/security.h> 281da177e4SLinus Torvalds #include <linux/hugetlb.h> 29c01d5b30SHugh Dickins #include <linux/shmem_fs.h> 301da177e4SLinus Torvalds #include <linux/profile.h> 31b95f1b31SPaul Gortmaker #include <linux/export.h> 321da177e4SLinus Torvalds #include <linux/mount.h> 331da177e4SLinus Torvalds #include <linux/mempolicy.h> 341da177e4SLinus Torvalds #include <linux/rmap.h> 35cddb8a5cSAndrea Arcangeli #include <linux/mmu_notifier.h> 3682f71ae4SKonstantin Khlebnikov #include <linux/mmdebug.h> 37cdd6c482SIngo Molnar #include <linux/perf_event.h> 38120a795dSAl Viro #include <linux/audit.h> 39b15d00b6SAndrea Arcangeli #include <linux/khugepaged.h> 402b144498SSrikar Dronamraju #include <linux/uprobes.h> 41d3737187SMichel Lespinasse #include <linux/rbtree_augmented.h> 421640879aSAndrew Shewmaker #include <linux/notifier.h> 431640879aSAndrew Shewmaker #include <linux/memory.h> 44b1de0d13SMitchel Humpherys #include <linux/printk.h> 4519a809afSAndrea Arcangeli #include <linux/userfaultfd_k.h> 46d977d56cSKonstantin Khlebnikov #include <linux/moduleparam.h> 4762b5f7d0SDave Hansen #include <linux/pkeys.h> 4821292580SAndrea Arcangeli #include <linux/oom.h> 4904f5866eSAndrea Arcangeli #include <linux/sched/mm.h> 501da177e4SLinus Torvalds 517c0f6ba6SLinus Torvalds #include <linux/uaccess.h> 521da177e4SLinus Torvalds #include <asm/cacheflush.h> 531da177e4SLinus Torvalds #include <asm/tlb.h> 54d6dd61c8SJeremy Fitzhardinge #include <asm/mmu_context.h> 551da177e4SLinus Torvalds 56df529cabSJaewon Kim #define CREATE_TRACE_POINTS 57df529cabSJaewon Kim #include <trace/events/mmap.h> 58df529cabSJaewon Kim 5942b77728SJan Beulich #include "internal.h" 6042b77728SJan Beulich 613a459756SKirill Korotaev #ifndef arch_mmap_check 623a459756SKirill Korotaev #define arch_mmap_check(addr, len, flags) (0) 633a459756SKirill Korotaev #endif 643a459756SKirill Korotaev 65d07e2259SDaniel Cashman #ifdef CONFIG_HAVE_ARCH_MMAP_RND_BITS 66d07e2259SDaniel Cashman const int mmap_rnd_bits_min = CONFIG_ARCH_MMAP_RND_BITS_MIN; 67d07e2259SDaniel Cashman const int mmap_rnd_bits_max = CONFIG_ARCH_MMAP_RND_BITS_MAX; 68d07e2259SDaniel Cashman int mmap_rnd_bits __read_mostly = CONFIG_ARCH_MMAP_RND_BITS; 69d07e2259SDaniel Cashman #endif 70d07e2259SDaniel Cashman #ifdef CONFIG_HAVE_ARCH_MMAP_RND_COMPAT_BITS 71d07e2259SDaniel Cashman const int mmap_rnd_compat_bits_min = CONFIG_ARCH_MMAP_RND_COMPAT_BITS_MIN; 72d07e2259SDaniel Cashman const int mmap_rnd_compat_bits_max = CONFIG_ARCH_MMAP_RND_COMPAT_BITS_MAX; 73d07e2259SDaniel Cashman int mmap_rnd_compat_bits __read_mostly = CONFIG_ARCH_MMAP_RND_COMPAT_BITS; 74d07e2259SDaniel Cashman #endif 75d07e2259SDaniel Cashman 76f4fcd558SKonstantin Khlebnikov static bool ignore_rlimit_data; 77d977d56cSKonstantin Khlebnikov core_param(ignore_rlimit_data, ignore_rlimit_data, bool, 0644); 78d07e2259SDaniel Cashman 79e0da382cSHugh Dickins static void unmap_region(struct mm_struct *mm, 80e0da382cSHugh Dickins struct vm_area_struct *vma, struct vm_area_struct *prev, 81e0da382cSHugh Dickins unsigned long start, unsigned long end); 82e0da382cSHugh Dickins 831da177e4SLinus Torvalds /* description of effects of mapping type and prot in current implementation. 841da177e4SLinus Torvalds * this is due to the limited x86 page protection hardware. The expected 851da177e4SLinus Torvalds * behavior is in parens: 861da177e4SLinus Torvalds * 871da177e4SLinus Torvalds * map_type prot 881da177e4SLinus Torvalds * PROT_NONE PROT_READ PROT_WRITE PROT_EXEC 891da177e4SLinus Torvalds * MAP_SHARED r: (no) no r: (yes) yes r: (no) yes r: (no) yes 901da177e4SLinus Torvalds * w: (no) no w: (no) no w: (yes) yes w: (no) no 911da177e4SLinus Torvalds * x: (no) no x: (no) yes x: (no) yes x: (yes) yes 921da177e4SLinus Torvalds * 931da177e4SLinus Torvalds * MAP_PRIVATE r: (no) no r: (yes) yes r: (no) yes r: (no) yes 941da177e4SLinus Torvalds * w: (no) no w: (no) no w: (copy) copy w: (no) no 951da177e4SLinus Torvalds * x: (no) no x: (no) yes x: (no) yes x: (yes) yes 961da177e4SLinus Torvalds */ 97ac34ceafSDaniel Micay pgprot_t protection_map[16] __ro_after_init = { 981da177e4SLinus Torvalds __P000, __P001, __P010, __P011, __P100, __P101, __P110, __P111, 991da177e4SLinus Torvalds __S000, __S001, __S010, __S011, __S100, __S101, __S110, __S111 1001da177e4SLinus Torvalds }; 1011da177e4SLinus Torvalds 102316d097cSDave Hansen #ifndef CONFIG_ARCH_HAS_FILTER_PGPROT 103316d097cSDave Hansen static inline pgprot_t arch_filter_pgprot(pgprot_t prot) 104316d097cSDave Hansen { 105316d097cSDave Hansen return prot; 106316d097cSDave Hansen } 107316d097cSDave Hansen #endif 108316d097cSDave Hansen 109804af2cfSHugh Dickins pgprot_t vm_get_page_prot(unsigned long vm_flags) 110804af2cfSHugh Dickins { 111316d097cSDave Hansen pgprot_t ret = __pgprot(pgprot_val(protection_map[vm_flags & 112b845f313SDave Kleikamp (VM_READ|VM_WRITE|VM_EXEC|VM_SHARED)]) | 113b845f313SDave Kleikamp pgprot_val(arch_vm_get_page_prot(vm_flags))); 114316d097cSDave Hansen 115316d097cSDave Hansen return arch_filter_pgprot(ret); 116804af2cfSHugh Dickins } 117804af2cfSHugh Dickins EXPORT_SYMBOL(vm_get_page_prot); 118804af2cfSHugh Dickins 11964e45507SPeter Feiner static pgprot_t vm_pgprot_modify(pgprot_t oldprot, unsigned long vm_flags) 12064e45507SPeter Feiner { 12164e45507SPeter Feiner return pgprot_modify(oldprot, vm_get_page_prot(vm_flags)); 12264e45507SPeter Feiner } 12364e45507SPeter Feiner 12464e45507SPeter Feiner /* Update vma->vm_page_prot to reflect vma->vm_flags. */ 12564e45507SPeter Feiner void vma_set_page_prot(struct vm_area_struct *vma) 12664e45507SPeter Feiner { 12764e45507SPeter Feiner unsigned long vm_flags = vma->vm_flags; 1286d2329f8SAndrea Arcangeli pgprot_t vm_page_prot; 12964e45507SPeter Feiner 1306d2329f8SAndrea Arcangeli vm_page_prot = vm_pgprot_modify(vma->vm_page_prot, vm_flags); 1316d2329f8SAndrea Arcangeli if (vma_wants_writenotify(vma, vm_page_prot)) { 13264e45507SPeter Feiner vm_flags &= ~VM_SHARED; 1336d2329f8SAndrea Arcangeli vm_page_prot = vm_pgprot_modify(vm_page_prot, vm_flags); 13464e45507SPeter Feiner } 135c1e8d7c6SMichel Lespinasse /* remove_protection_ptes reads vma->vm_page_prot without mmap_lock */ 1366d2329f8SAndrea Arcangeli WRITE_ONCE(vma->vm_page_prot, vm_page_prot); 13764e45507SPeter Feiner } 13864e45507SPeter Feiner 1391da177e4SLinus Torvalds /* 140c8c06efaSDavidlohr Bueso * Requires inode->i_mapping->i_mmap_rwsem 1411da177e4SLinus Torvalds */ 1421da177e4SLinus Torvalds static void __remove_shared_vm_struct(struct vm_area_struct *vma, 1431da177e4SLinus Torvalds struct file *file, struct address_space *mapping) 1441da177e4SLinus Torvalds { 1451da177e4SLinus Torvalds if (vma->vm_flags & VM_DENYWRITE) 146cb48841fSMiaohe Lin allow_write_access(file); 1471da177e4SLinus Torvalds if (vma->vm_flags & VM_SHARED) 1484bb5f5d9SDavid Herrmann mapping_unmap_writable(mapping); 1491da177e4SLinus Torvalds 1501da177e4SLinus Torvalds flush_dcache_mmap_lock(mapping); 1516b2dbba8SMichel Lespinasse vma_interval_tree_remove(vma, &mapping->i_mmap); 1521da177e4SLinus Torvalds flush_dcache_mmap_unlock(mapping); 1531da177e4SLinus Torvalds } 1541da177e4SLinus Torvalds 1551da177e4SLinus Torvalds /* 1566b2dbba8SMichel Lespinasse * Unlink a file-based vm structure from its interval tree, to hide 157a8fb5618SHugh Dickins * vma from rmap and vmtruncate before freeing its page tables. 1581da177e4SLinus Torvalds */ 159a8fb5618SHugh Dickins void unlink_file_vma(struct vm_area_struct *vma) 1601da177e4SLinus Torvalds { 1611da177e4SLinus Torvalds struct file *file = vma->vm_file; 1621da177e4SLinus Torvalds 1631da177e4SLinus Torvalds if (file) { 1641da177e4SLinus Torvalds struct address_space *mapping = file->f_mapping; 16583cde9e8SDavidlohr Bueso i_mmap_lock_write(mapping); 1661da177e4SLinus Torvalds __remove_shared_vm_struct(vma, file, mapping); 16783cde9e8SDavidlohr Bueso i_mmap_unlock_write(mapping); 1681da177e4SLinus Torvalds } 169a8fb5618SHugh Dickins } 170a8fb5618SHugh Dickins 171a8fb5618SHugh Dickins /* 172a8fb5618SHugh Dickins * Close a vm structure and free it, returning the next. 173a8fb5618SHugh Dickins */ 174a8fb5618SHugh Dickins static struct vm_area_struct *remove_vma(struct vm_area_struct *vma) 175a8fb5618SHugh Dickins { 176a8fb5618SHugh Dickins struct vm_area_struct *next = vma->vm_next; 177a8fb5618SHugh Dickins 178a8fb5618SHugh Dickins might_sleep(); 1791da177e4SLinus Torvalds if (vma->vm_ops && vma->vm_ops->close) 1801da177e4SLinus Torvalds vma->vm_ops->close(vma); 181e9714acfSKonstantin Khlebnikov if (vma->vm_file) 182a8fb5618SHugh Dickins fput(vma->vm_file); 183f0be3d32SLee Schermerhorn mpol_put(vma_policy(vma)); 1843928d4f5SLinus Torvalds vm_area_free(vma); 185a8fb5618SHugh Dickins return next; 1861da177e4SLinus Torvalds } 1871da177e4SLinus Torvalds 188bb177a73SMichal Hocko static int do_brk_flags(unsigned long addr, unsigned long request, unsigned long flags, 189bb177a73SMichal Hocko struct list_head *uf); 1906a6160a7SHeiko Carstens SYSCALL_DEFINE1(brk, unsigned long, brk) 1911da177e4SLinus Torvalds { 1928764b338SCyrill Gorcunov unsigned long retval; 1939bc8039eSYang Shi unsigned long newbrk, oldbrk, origbrk; 1941da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 1951be7107fSHugh Dickins struct vm_area_struct *next; 196a5b4592cSJiri Kosina unsigned long min_brk; 197128557ffSMichel Lespinasse bool populate; 1989bc8039eSYang Shi bool downgraded = false; 199897ab3e0SMike Rapoport LIST_HEAD(uf); 2001da177e4SLinus Torvalds 201d8ed45c5SMichel Lespinasse if (mmap_write_lock_killable(mm)) 202dc0ef0dfSMichal Hocko return -EINTR; 2031da177e4SLinus Torvalds 2049bc8039eSYang Shi origbrk = mm->brk; 2059bc8039eSYang Shi 206a5b4592cSJiri Kosina #ifdef CONFIG_COMPAT_BRK 2075520e894SJiri Kosina /* 2085520e894SJiri Kosina * CONFIG_COMPAT_BRK can still be overridden by setting 2095520e894SJiri Kosina * randomize_va_space to 2, which will still cause mm->start_brk 2105520e894SJiri Kosina * to be arbitrarily shifted 2115520e894SJiri Kosina */ 2124471a675SJiri Kosina if (current->brk_randomized) 2135520e894SJiri Kosina min_brk = mm->start_brk; 2145520e894SJiri Kosina else 2155520e894SJiri Kosina min_brk = mm->end_data; 216a5b4592cSJiri Kosina #else 217a5b4592cSJiri Kosina min_brk = mm->start_brk; 218a5b4592cSJiri Kosina #endif 219a5b4592cSJiri Kosina if (brk < min_brk) 2201da177e4SLinus Torvalds goto out; 2211e624196SRam Gupta 2221e624196SRam Gupta /* 2231e624196SRam Gupta * Check against rlimit here. If this check is done later after the test 2241e624196SRam Gupta * of oldbrk with newbrk then it can escape the test and let the data 2251e624196SRam Gupta * segment grow beyond its set limit the in case where the limit is 2261e624196SRam Gupta * not page aligned -Ram Gupta 2271e624196SRam Gupta */ 2288764b338SCyrill Gorcunov if (check_data_rlimit(rlimit(RLIMIT_DATA), brk, mm->start_brk, 2298764b338SCyrill Gorcunov mm->end_data, mm->start_data)) 2301e624196SRam Gupta goto out; 2311e624196SRam Gupta 2321da177e4SLinus Torvalds newbrk = PAGE_ALIGN(brk); 2331da177e4SLinus Torvalds oldbrk = PAGE_ALIGN(mm->brk); 2349bc8039eSYang Shi if (oldbrk == newbrk) { 2359bc8039eSYang Shi mm->brk = brk; 2369bc8039eSYang Shi goto success; 2379bc8039eSYang Shi } 2381da177e4SLinus Torvalds 2399bc8039eSYang Shi /* 2409bc8039eSYang Shi * Always allow shrinking brk. 241c1e8d7c6SMichel Lespinasse * __do_munmap() may downgrade mmap_lock to read. 2429bc8039eSYang Shi */ 2431da177e4SLinus Torvalds if (brk <= mm->brk) { 2449bc8039eSYang Shi int ret; 2459bc8039eSYang Shi 2469bc8039eSYang Shi /* 247c1e8d7c6SMichel Lespinasse * mm->brk must to be protected by write mmap_lock so update it 248c1e8d7c6SMichel Lespinasse * before downgrading mmap_lock. When __do_munmap() fails, 2499bc8039eSYang Shi * mm->brk will be restored from origbrk. 2509bc8039eSYang Shi */ 2519bc8039eSYang Shi mm->brk = brk; 2529bc8039eSYang Shi ret = __do_munmap(mm, newbrk, oldbrk-newbrk, &uf, true); 2539bc8039eSYang Shi if (ret < 0) { 2549bc8039eSYang Shi mm->brk = origbrk; 2551da177e4SLinus Torvalds goto out; 2569bc8039eSYang Shi } else if (ret == 1) { 2579bc8039eSYang Shi downgraded = true; 2589bc8039eSYang Shi } 2599bc8039eSYang Shi goto success; 2601da177e4SLinus Torvalds } 2611da177e4SLinus Torvalds 2621da177e4SLinus Torvalds /* Check against existing mmap mappings. */ 2631be7107fSHugh Dickins next = find_vma(mm, oldbrk); 2641be7107fSHugh Dickins if (next && newbrk + PAGE_SIZE > vm_start_gap(next)) 2651da177e4SLinus Torvalds goto out; 2661da177e4SLinus Torvalds 2671da177e4SLinus Torvalds /* Ok, looks good - let it rip. */ 268bb177a73SMichal Hocko if (do_brk_flags(oldbrk, newbrk-oldbrk, 0, &uf) < 0) 2691da177e4SLinus Torvalds goto out; 2701da177e4SLinus Torvalds mm->brk = brk; 2719bc8039eSYang Shi 2729bc8039eSYang Shi success: 273128557ffSMichel Lespinasse populate = newbrk > oldbrk && (mm->def_flags & VM_LOCKED) != 0; 2749bc8039eSYang Shi if (downgraded) 275d8ed45c5SMichel Lespinasse mmap_read_unlock(mm); 2769bc8039eSYang Shi else 277d8ed45c5SMichel Lespinasse mmap_write_unlock(mm); 278897ab3e0SMike Rapoport userfaultfd_unmap_complete(mm, &uf); 279128557ffSMichel Lespinasse if (populate) 280128557ffSMichel Lespinasse mm_populate(oldbrk, newbrk - oldbrk); 281128557ffSMichel Lespinasse return brk; 282128557ffSMichel Lespinasse 2831da177e4SLinus Torvalds out: 2849bc8039eSYang Shi retval = origbrk; 285d8ed45c5SMichel Lespinasse mmap_write_unlock(mm); 2861da177e4SLinus Torvalds return retval; 2871da177e4SLinus Torvalds } 2881da177e4SLinus Torvalds 289315cc066SMichel Lespinasse static inline unsigned long vma_compute_gap(struct vm_area_struct *vma) 290d3737187SMichel Lespinasse { 291315cc066SMichel Lespinasse unsigned long gap, prev_end; 2921be7107fSHugh Dickins 2931be7107fSHugh Dickins /* 2941be7107fSHugh Dickins * Note: in the rare case of a VM_GROWSDOWN above a VM_GROWSUP, we 2951be7107fSHugh Dickins * allow two stack_guard_gaps between them here, and when choosing 2961be7107fSHugh Dickins * an unmapped area; whereas when expanding we only require one. 2971be7107fSHugh Dickins * That's a little inconsistent, but keeps the code here simpler. 2981be7107fSHugh Dickins */ 299315cc066SMichel Lespinasse gap = vm_start_gap(vma); 3001be7107fSHugh Dickins if (vma->vm_prev) { 3011be7107fSHugh Dickins prev_end = vm_end_gap(vma->vm_prev); 302315cc066SMichel Lespinasse if (gap > prev_end) 303315cc066SMichel Lespinasse gap -= prev_end; 3041be7107fSHugh Dickins else 305315cc066SMichel Lespinasse gap = 0; 3061be7107fSHugh Dickins } 307315cc066SMichel Lespinasse return gap; 308315cc066SMichel Lespinasse } 309315cc066SMichel Lespinasse 310315cc066SMichel Lespinasse #ifdef CONFIG_DEBUG_VM_RB 311315cc066SMichel Lespinasse static unsigned long vma_compute_subtree_gap(struct vm_area_struct *vma) 312315cc066SMichel Lespinasse { 313315cc066SMichel Lespinasse unsigned long max = vma_compute_gap(vma), subtree_gap; 314d3737187SMichel Lespinasse if (vma->vm_rb.rb_left) { 315d3737187SMichel Lespinasse subtree_gap = rb_entry(vma->vm_rb.rb_left, 316d3737187SMichel Lespinasse struct vm_area_struct, vm_rb)->rb_subtree_gap; 317d3737187SMichel Lespinasse if (subtree_gap > max) 318d3737187SMichel Lespinasse max = subtree_gap; 319d3737187SMichel Lespinasse } 320d3737187SMichel Lespinasse if (vma->vm_rb.rb_right) { 321d3737187SMichel Lespinasse subtree_gap = rb_entry(vma->vm_rb.rb_right, 322d3737187SMichel Lespinasse struct vm_area_struct, vm_rb)->rb_subtree_gap; 323d3737187SMichel Lespinasse if (subtree_gap > max) 324d3737187SMichel Lespinasse max = subtree_gap; 325d3737187SMichel Lespinasse } 326d3737187SMichel Lespinasse return max; 327d3737187SMichel Lespinasse } 328d3737187SMichel Lespinasse 329acf128d0SAndrea Arcangeli static int browse_rb(struct mm_struct *mm) 3301da177e4SLinus Torvalds { 331acf128d0SAndrea Arcangeli struct rb_root *root = &mm->mm_rb; 3325a0768f6SMichel Lespinasse int i = 0, j, bug = 0; 3331da177e4SLinus Torvalds struct rb_node *nd, *pn = NULL; 3341da177e4SLinus Torvalds unsigned long prev = 0, pend = 0; 3351da177e4SLinus Torvalds 3361da177e4SLinus Torvalds for (nd = rb_first(root); nd; nd = rb_next(nd)) { 3371da177e4SLinus Torvalds struct vm_area_struct *vma; 3381da177e4SLinus Torvalds vma = rb_entry(nd, struct vm_area_struct, vm_rb); 3395a0768f6SMichel Lespinasse if (vma->vm_start < prev) { 340ff26f70fSAndrew Morton pr_emerg("vm_start %lx < prev %lx\n", 341ff26f70fSAndrew Morton vma->vm_start, prev); 3425a0768f6SMichel Lespinasse bug = 1; 3435a0768f6SMichel Lespinasse } 3445a0768f6SMichel Lespinasse if (vma->vm_start < pend) { 345ff26f70fSAndrew Morton pr_emerg("vm_start %lx < pend %lx\n", 346ff26f70fSAndrew Morton vma->vm_start, pend); 3475a0768f6SMichel Lespinasse bug = 1; 3485a0768f6SMichel Lespinasse } 3495a0768f6SMichel Lespinasse if (vma->vm_start > vma->vm_end) { 350ff26f70fSAndrew Morton pr_emerg("vm_start %lx > vm_end %lx\n", 351ff26f70fSAndrew Morton vma->vm_start, vma->vm_end); 3525a0768f6SMichel Lespinasse bug = 1; 3535a0768f6SMichel Lespinasse } 354acf128d0SAndrea Arcangeli spin_lock(&mm->page_table_lock); 3555a0768f6SMichel Lespinasse if (vma->rb_subtree_gap != vma_compute_subtree_gap(vma)) { 3568542bdfcSSasha Levin pr_emerg("free gap %lx, correct %lx\n", 3575a0768f6SMichel Lespinasse vma->rb_subtree_gap, 3585a0768f6SMichel Lespinasse vma_compute_subtree_gap(vma)); 3595a0768f6SMichel Lespinasse bug = 1; 3605a0768f6SMichel Lespinasse } 361acf128d0SAndrea Arcangeli spin_unlock(&mm->page_table_lock); 3621da177e4SLinus Torvalds i++; 3631da177e4SLinus Torvalds pn = nd; 364d1af65d1SDavid Miller prev = vma->vm_start; 365d1af65d1SDavid Miller pend = vma->vm_end; 3661da177e4SLinus Torvalds } 3671da177e4SLinus Torvalds j = 0; 3685a0768f6SMichel Lespinasse for (nd = pn; nd; nd = rb_prev(nd)) 3691da177e4SLinus Torvalds j++; 3705a0768f6SMichel Lespinasse if (i != j) { 3718542bdfcSSasha Levin pr_emerg("backwards %d, forwards %d\n", j, i); 3725a0768f6SMichel Lespinasse bug = 1; 3731da177e4SLinus Torvalds } 3745a0768f6SMichel Lespinasse return bug ? -1 : i; 3751da177e4SLinus Torvalds } 3761da177e4SLinus Torvalds 377d3737187SMichel Lespinasse static void validate_mm_rb(struct rb_root *root, struct vm_area_struct *ignore) 378d3737187SMichel Lespinasse { 379d3737187SMichel Lespinasse struct rb_node *nd; 380d3737187SMichel Lespinasse 381d3737187SMichel Lespinasse for (nd = rb_first(root); nd; nd = rb_next(nd)) { 382d3737187SMichel Lespinasse struct vm_area_struct *vma; 383d3737187SMichel Lespinasse vma = rb_entry(nd, struct vm_area_struct, vm_rb); 38496dad67fSSasha Levin VM_BUG_ON_VMA(vma != ignore && 38596dad67fSSasha Levin vma->rb_subtree_gap != vma_compute_subtree_gap(vma), 38696dad67fSSasha Levin vma); 387d3737187SMichel Lespinasse } 3881da177e4SLinus Torvalds } 3891da177e4SLinus Torvalds 390eafd4dc4SRashika Kheria static void validate_mm(struct mm_struct *mm) 3911da177e4SLinus Torvalds { 3921da177e4SLinus Torvalds int bug = 0; 3931da177e4SLinus Torvalds int i = 0; 3945a0768f6SMichel Lespinasse unsigned long highest_address = 0; 395ed8ea815SMichel Lespinasse struct vm_area_struct *vma = mm->mmap; 396ff26f70fSAndrew Morton 397ed8ea815SMichel Lespinasse while (vma) { 39812352d3cSKonstantin Khlebnikov struct anon_vma *anon_vma = vma->anon_vma; 399ed8ea815SMichel Lespinasse struct anon_vma_chain *avc; 400ff26f70fSAndrew Morton 40112352d3cSKonstantin Khlebnikov if (anon_vma) { 40212352d3cSKonstantin Khlebnikov anon_vma_lock_read(anon_vma); 403ed8ea815SMichel Lespinasse list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) 404ed8ea815SMichel Lespinasse anon_vma_interval_tree_verify(avc); 40512352d3cSKonstantin Khlebnikov anon_vma_unlock_read(anon_vma); 40612352d3cSKonstantin Khlebnikov } 40712352d3cSKonstantin Khlebnikov 4081be7107fSHugh Dickins highest_address = vm_end_gap(vma); 409ed8ea815SMichel Lespinasse vma = vma->vm_next; 4101da177e4SLinus Torvalds i++; 4111da177e4SLinus Torvalds } 4125a0768f6SMichel Lespinasse if (i != mm->map_count) { 4138542bdfcSSasha Levin pr_emerg("map_count %d vm_next %d\n", mm->map_count, i); 4145a0768f6SMichel Lespinasse bug = 1; 4155a0768f6SMichel Lespinasse } 4165a0768f6SMichel Lespinasse if (highest_address != mm->highest_vm_end) { 4178542bdfcSSasha Levin pr_emerg("mm->highest_vm_end %lx, found %lx\n", 4185a0768f6SMichel Lespinasse mm->highest_vm_end, highest_address); 4195a0768f6SMichel Lespinasse bug = 1; 4205a0768f6SMichel Lespinasse } 421acf128d0SAndrea Arcangeli i = browse_rb(mm); 4225a0768f6SMichel Lespinasse if (i != mm->map_count) { 423ff26f70fSAndrew Morton if (i != -1) 4248542bdfcSSasha Levin pr_emerg("map_count %d rb %d\n", mm->map_count, i); 4255a0768f6SMichel Lespinasse bug = 1; 4265a0768f6SMichel Lespinasse } 42796dad67fSSasha Levin VM_BUG_ON_MM(bug, mm); 4281da177e4SLinus Torvalds } 4291da177e4SLinus Torvalds #else 430d3737187SMichel Lespinasse #define validate_mm_rb(root, ignore) do { } while (0) 4311da177e4SLinus Torvalds #define validate_mm(mm) do { } while (0) 4321da177e4SLinus Torvalds #endif 4331da177e4SLinus Torvalds 434315cc066SMichel Lespinasse RB_DECLARE_CALLBACKS_MAX(static, vma_gap_callbacks, 435315cc066SMichel Lespinasse struct vm_area_struct, vm_rb, 436315cc066SMichel Lespinasse unsigned long, rb_subtree_gap, vma_compute_gap) 437d3737187SMichel Lespinasse 438d3737187SMichel Lespinasse /* 439d3737187SMichel Lespinasse * Update augmented rbtree rb_subtree_gap values after vma->vm_start or 440d3737187SMichel Lespinasse * vma->vm_prev->vm_end values changed, without modifying the vma's position 441d3737187SMichel Lespinasse * in the rbtree. 442d3737187SMichel Lespinasse */ 443d3737187SMichel Lespinasse static void vma_gap_update(struct vm_area_struct *vma) 444d3737187SMichel Lespinasse { 445d3737187SMichel Lespinasse /* 446315cc066SMichel Lespinasse * As it turns out, RB_DECLARE_CALLBACKS_MAX() already created 447315cc066SMichel Lespinasse * a callback function that does exactly what we want. 448d3737187SMichel Lespinasse */ 449d3737187SMichel Lespinasse vma_gap_callbacks_propagate(&vma->vm_rb, NULL); 450d3737187SMichel Lespinasse } 451d3737187SMichel Lespinasse 452d3737187SMichel Lespinasse static inline void vma_rb_insert(struct vm_area_struct *vma, 453d3737187SMichel Lespinasse struct rb_root *root) 454d3737187SMichel Lespinasse { 455d3737187SMichel Lespinasse /* All rb_subtree_gap values must be consistent prior to insertion */ 456d3737187SMichel Lespinasse validate_mm_rb(root, NULL); 457d3737187SMichel Lespinasse 458d3737187SMichel Lespinasse rb_insert_augmented(&vma->vm_rb, root, &vma_gap_callbacks); 459d3737187SMichel Lespinasse } 460d3737187SMichel Lespinasse 4618f26e0b1SAndrea Arcangeli static void __vma_rb_erase(struct vm_area_struct *vma, struct rb_root *root) 4628f26e0b1SAndrea Arcangeli { 4638f26e0b1SAndrea Arcangeli /* 4648f26e0b1SAndrea Arcangeli * Note rb_erase_augmented is a fairly large inline function, 4658f26e0b1SAndrea Arcangeli * so make sure we instantiate it only once with our desired 4668f26e0b1SAndrea Arcangeli * augmented rbtree callbacks. 4678f26e0b1SAndrea Arcangeli */ 4688f26e0b1SAndrea Arcangeli rb_erase_augmented(&vma->vm_rb, root, &vma_gap_callbacks); 4698f26e0b1SAndrea Arcangeli } 4708f26e0b1SAndrea Arcangeli 4718f26e0b1SAndrea Arcangeli static __always_inline void vma_rb_erase_ignore(struct vm_area_struct *vma, 4728f26e0b1SAndrea Arcangeli struct rb_root *root, 4738f26e0b1SAndrea Arcangeli struct vm_area_struct *ignore) 4748f26e0b1SAndrea Arcangeli { 4758f26e0b1SAndrea Arcangeli /* 4768f26e0b1SAndrea Arcangeli * All rb_subtree_gap values must be consistent prior to erase, 4774d1e7243SWei Yang * with the possible exception of 4784d1e7243SWei Yang * 4794d1e7243SWei Yang * a. the "next" vma being erased if next->vm_start was reduced in 4804d1e7243SWei Yang * __vma_adjust() -> __vma_unlink() 4814d1e7243SWei Yang * b. the vma being erased in detach_vmas_to_be_unmapped() -> 4824d1e7243SWei Yang * vma_rb_erase() 4838f26e0b1SAndrea Arcangeli */ 4848f26e0b1SAndrea Arcangeli validate_mm_rb(root, ignore); 4858f26e0b1SAndrea Arcangeli 4868f26e0b1SAndrea Arcangeli __vma_rb_erase(vma, root); 4878f26e0b1SAndrea Arcangeli } 4888f26e0b1SAndrea Arcangeli 4898f26e0b1SAndrea Arcangeli static __always_inline void vma_rb_erase(struct vm_area_struct *vma, 4908f26e0b1SAndrea Arcangeli struct rb_root *root) 491d3737187SMichel Lespinasse { 4924d1e7243SWei Yang vma_rb_erase_ignore(vma, root, vma); 493d3737187SMichel Lespinasse } 494d3737187SMichel Lespinasse 495bf181b9fSMichel Lespinasse /* 496bf181b9fSMichel Lespinasse * vma has some anon_vma assigned, and is already inserted on that 497bf181b9fSMichel Lespinasse * anon_vma's interval trees. 498bf181b9fSMichel Lespinasse * 499bf181b9fSMichel Lespinasse * Before updating the vma's vm_start / vm_end / vm_pgoff fields, the 500bf181b9fSMichel Lespinasse * vma must be removed from the anon_vma's interval trees using 501bf181b9fSMichel Lespinasse * anon_vma_interval_tree_pre_update_vma(). 502bf181b9fSMichel Lespinasse * 503bf181b9fSMichel Lespinasse * After the update, the vma will be reinserted using 504bf181b9fSMichel Lespinasse * anon_vma_interval_tree_post_update_vma(). 505bf181b9fSMichel Lespinasse * 506c1e8d7c6SMichel Lespinasse * The entire update must be protected by exclusive mmap_lock and by 507bf181b9fSMichel Lespinasse * the root anon_vma's mutex. 508bf181b9fSMichel Lespinasse */ 509bf181b9fSMichel Lespinasse static inline void 510bf181b9fSMichel Lespinasse anon_vma_interval_tree_pre_update_vma(struct vm_area_struct *vma) 511bf181b9fSMichel Lespinasse { 512bf181b9fSMichel Lespinasse struct anon_vma_chain *avc; 513bf181b9fSMichel Lespinasse 514bf181b9fSMichel Lespinasse list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) 515bf181b9fSMichel Lespinasse anon_vma_interval_tree_remove(avc, &avc->anon_vma->rb_root); 516bf181b9fSMichel Lespinasse } 517bf181b9fSMichel Lespinasse 518bf181b9fSMichel Lespinasse static inline void 519bf181b9fSMichel Lespinasse anon_vma_interval_tree_post_update_vma(struct vm_area_struct *vma) 520bf181b9fSMichel Lespinasse { 521bf181b9fSMichel Lespinasse struct anon_vma_chain *avc; 522bf181b9fSMichel Lespinasse 523bf181b9fSMichel Lespinasse list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) 524bf181b9fSMichel Lespinasse anon_vma_interval_tree_insert(avc, &avc->anon_vma->rb_root); 525bf181b9fSMichel Lespinasse } 526bf181b9fSMichel Lespinasse 5276597d783SHugh Dickins static int find_vma_links(struct mm_struct *mm, unsigned long addr, 5286597d783SHugh Dickins unsigned long end, struct vm_area_struct **pprev, 5296597d783SHugh Dickins struct rb_node ***rb_link, struct rb_node **rb_parent) 5301da177e4SLinus Torvalds { 5311da177e4SLinus Torvalds struct rb_node **__rb_link, *__rb_parent, *rb_prev; 5321da177e4SLinus Torvalds 5331da177e4SLinus Torvalds __rb_link = &mm->mm_rb.rb_node; 5341da177e4SLinus Torvalds rb_prev = __rb_parent = NULL; 5351da177e4SLinus Torvalds 5361da177e4SLinus Torvalds while (*__rb_link) { 5371da177e4SLinus Torvalds struct vm_area_struct *vma_tmp; 5381da177e4SLinus Torvalds 5391da177e4SLinus Torvalds __rb_parent = *__rb_link; 5401da177e4SLinus Torvalds vma_tmp = rb_entry(__rb_parent, struct vm_area_struct, vm_rb); 5411da177e4SLinus Torvalds 5421da177e4SLinus Torvalds if (vma_tmp->vm_end > addr) { 5436597d783SHugh Dickins /* Fail if an existing vma overlaps the area */ 5446597d783SHugh Dickins if (vma_tmp->vm_start < end) 5456597d783SHugh Dickins return -ENOMEM; 5461da177e4SLinus Torvalds __rb_link = &__rb_parent->rb_left; 5471da177e4SLinus Torvalds } else { 5481da177e4SLinus Torvalds rb_prev = __rb_parent; 5491da177e4SLinus Torvalds __rb_link = &__rb_parent->rb_right; 5501da177e4SLinus Torvalds } 5511da177e4SLinus Torvalds } 5521da177e4SLinus Torvalds 5531da177e4SLinus Torvalds *pprev = NULL; 5541da177e4SLinus Torvalds if (rb_prev) 5551da177e4SLinus Torvalds *pprev = rb_entry(rb_prev, struct vm_area_struct, vm_rb); 5561da177e4SLinus Torvalds *rb_link = __rb_link; 5571da177e4SLinus Torvalds *rb_parent = __rb_parent; 5586597d783SHugh Dickins return 0; 5591da177e4SLinus Torvalds } 5601da177e4SLinus Torvalds 5613903b55aSLiam R. Howlett /* 5623903b55aSLiam R. Howlett * vma_next() - Get the next VMA. 5633903b55aSLiam R. Howlett * @mm: The mm_struct. 5643903b55aSLiam R. Howlett * @vma: The current vma. 5653903b55aSLiam R. Howlett * 5663903b55aSLiam R. Howlett * If @vma is NULL, return the first vma in the mm. 5673903b55aSLiam R. Howlett * 5683903b55aSLiam R. Howlett * Returns: The next VMA after @vma. 5693903b55aSLiam R. Howlett */ 5703903b55aSLiam R. Howlett static inline struct vm_area_struct *vma_next(struct mm_struct *mm, 5713903b55aSLiam R. Howlett struct vm_area_struct *vma) 5723903b55aSLiam R. Howlett { 5733903b55aSLiam R. Howlett if (!vma) 5743903b55aSLiam R. Howlett return mm->mmap; 5753903b55aSLiam R. Howlett 5763903b55aSLiam R. Howlett return vma->vm_next; 5773903b55aSLiam R. Howlett } 578fb8090b6SLiam R. Howlett 579fb8090b6SLiam R. Howlett /* 580fb8090b6SLiam R. Howlett * munmap_vma_range() - munmap VMAs that overlap a range. 581fb8090b6SLiam R. Howlett * @mm: The mm struct 582fb8090b6SLiam R. Howlett * @start: The start of the range. 583fb8090b6SLiam R. Howlett * @len: The length of the range. 584fb8090b6SLiam R. Howlett * @pprev: pointer to the pointer that will be set to previous vm_area_struct 585fb8090b6SLiam R. Howlett * @rb_link: the rb_node 586fb8090b6SLiam R. Howlett * @rb_parent: the parent rb_node 587fb8090b6SLiam R. Howlett * 588fb8090b6SLiam R. Howlett * Find all the vm_area_struct that overlap from @start to 589fb8090b6SLiam R. Howlett * @end and munmap them. Set @pprev to the previous vm_area_struct. 590fb8090b6SLiam R. Howlett * 591fb8090b6SLiam R. Howlett * Returns: -ENOMEM on munmap failure or 0 on success. 592fb8090b6SLiam R. Howlett */ 593fb8090b6SLiam R. Howlett static inline int 594fb8090b6SLiam R. Howlett munmap_vma_range(struct mm_struct *mm, unsigned long start, unsigned long len, 595fb8090b6SLiam R. Howlett struct vm_area_struct **pprev, struct rb_node ***link, 596fb8090b6SLiam R. Howlett struct rb_node **parent, struct list_head *uf) 597fb8090b6SLiam R. Howlett { 598fb8090b6SLiam R. Howlett 599fb8090b6SLiam R. Howlett while (find_vma_links(mm, start, start + len, pprev, link, parent)) 600fb8090b6SLiam R. Howlett if (do_munmap(mm, start, len, uf)) 601fb8090b6SLiam R. Howlett return -ENOMEM; 602fb8090b6SLiam R. Howlett 603fb8090b6SLiam R. Howlett return 0; 604fb8090b6SLiam R. Howlett } 605e8420a8eSCyril Hrubis static unsigned long count_vma_pages_range(struct mm_struct *mm, 606e8420a8eSCyril Hrubis unsigned long addr, unsigned long end) 607e8420a8eSCyril Hrubis { 608e8420a8eSCyril Hrubis unsigned long nr_pages = 0; 609e8420a8eSCyril Hrubis struct vm_area_struct *vma; 610e8420a8eSCyril Hrubis 611e8420a8eSCyril Hrubis /* Find first overlaping mapping */ 612e8420a8eSCyril Hrubis vma = find_vma_intersection(mm, addr, end); 613e8420a8eSCyril Hrubis if (!vma) 614e8420a8eSCyril Hrubis return 0; 615e8420a8eSCyril Hrubis 616e8420a8eSCyril Hrubis nr_pages = (min(end, vma->vm_end) - 617e8420a8eSCyril Hrubis max(addr, vma->vm_start)) >> PAGE_SHIFT; 618e8420a8eSCyril Hrubis 619e8420a8eSCyril Hrubis /* Iterate over the rest of the overlaps */ 620e8420a8eSCyril Hrubis for (vma = vma->vm_next; vma; vma = vma->vm_next) { 621e8420a8eSCyril Hrubis unsigned long overlap_len; 622e8420a8eSCyril Hrubis 623e8420a8eSCyril Hrubis if (vma->vm_start > end) 624e8420a8eSCyril Hrubis break; 625e8420a8eSCyril Hrubis 626e8420a8eSCyril Hrubis overlap_len = min(end, vma->vm_end) - vma->vm_start; 627e8420a8eSCyril Hrubis nr_pages += overlap_len >> PAGE_SHIFT; 628e8420a8eSCyril Hrubis } 629e8420a8eSCyril Hrubis 630e8420a8eSCyril Hrubis return nr_pages; 631e8420a8eSCyril Hrubis } 632e8420a8eSCyril Hrubis 6331da177e4SLinus Torvalds void __vma_link_rb(struct mm_struct *mm, struct vm_area_struct *vma, 6341da177e4SLinus Torvalds struct rb_node **rb_link, struct rb_node *rb_parent) 6351da177e4SLinus Torvalds { 636d3737187SMichel Lespinasse /* Update tracking information for the gap following the new vma. */ 637d3737187SMichel Lespinasse if (vma->vm_next) 638d3737187SMichel Lespinasse vma_gap_update(vma->vm_next); 639d3737187SMichel Lespinasse else 6401be7107fSHugh Dickins mm->highest_vm_end = vm_end_gap(vma); 641d3737187SMichel Lespinasse 642d3737187SMichel Lespinasse /* 643d3737187SMichel Lespinasse * vma->vm_prev wasn't known when we followed the rbtree to find the 644d3737187SMichel Lespinasse * correct insertion point for that vma. As a result, we could not 645d3737187SMichel Lespinasse * update the vma vm_rb parents rb_subtree_gap values on the way down. 646d3737187SMichel Lespinasse * So, we first insert the vma with a zero rb_subtree_gap value 647d3737187SMichel Lespinasse * (to be consistent with what we did on the way down), and then 648d3737187SMichel Lespinasse * immediately update the gap to the correct value. Finally we 649d3737187SMichel Lespinasse * rebalance the rbtree after all augmented values have been set. 650d3737187SMichel Lespinasse */ 6511da177e4SLinus Torvalds rb_link_node(&vma->vm_rb, rb_parent, rb_link); 652d3737187SMichel Lespinasse vma->rb_subtree_gap = 0; 653d3737187SMichel Lespinasse vma_gap_update(vma); 654d3737187SMichel Lespinasse vma_rb_insert(vma, &mm->mm_rb); 6551da177e4SLinus Torvalds } 6561da177e4SLinus Torvalds 657cb8f488cSDenys Vlasenko static void __vma_link_file(struct vm_area_struct *vma) 6581da177e4SLinus Torvalds { 6591da177e4SLinus Torvalds struct file *file; 6601da177e4SLinus Torvalds 6611da177e4SLinus Torvalds file = vma->vm_file; 6621da177e4SLinus Torvalds if (file) { 6631da177e4SLinus Torvalds struct address_space *mapping = file->f_mapping; 6641da177e4SLinus Torvalds 6651da177e4SLinus Torvalds if (vma->vm_flags & VM_DENYWRITE) 66673eb7f9aSMiaohe Lin put_write_access(file_inode(file)); 6671da177e4SLinus Torvalds if (vma->vm_flags & VM_SHARED) 668cf508b58SMiaohe Lin mapping_allow_writable(mapping); 6691da177e4SLinus Torvalds 6701da177e4SLinus Torvalds flush_dcache_mmap_lock(mapping); 6716b2dbba8SMichel Lespinasse vma_interval_tree_insert(vma, &mapping->i_mmap); 6721da177e4SLinus Torvalds flush_dcache_mmap_unlock(mapping); 6731da177e4SLinus Torvalds } 6741da177e4SLinus Torvalds } 6751da177e4SLinus Torvalds 6761da177e4SLinus Torvalds static void 6771da177e4SLinus Torvalds __vma_link(struct mm_struct *mm, struct vm_area_struct *vma, 6781da177e4SLinus Torvalds struct vm_area_struct *prev, struct rb_node **rb_link, 6791da177e4SLinus Torvalds struct rb_node *rb_parent) 6801da177e4SLinus Torvalds { 681aba6dfb7SWei Yang __vma_link_list(mm, vma, prev); 6821da177e4SLinus Torvalds __vma_link_rb(mm, vma, rb_link, rb_parent); 6831da177e4SLinus Torvalds } 6841da177e4SLinus Torvalds 6851da177e4SLinus Torvalds static void vma_link(struct mm_struct *mm, struct vm_area_struct *vma, 6861da177e4SLinus Torvalds struct vm_area_struct *prev, struct rb_node **rb_link, 6871da177e4SLinus Torvalds struct rb_node *rb_parent) 6881da177e4SLinus Torvalds { 6891da177e4SLinus Torvalds struct address_space *mapping = NULL; 6901da177e4SLinus Torvalds 69164ac4940SHuang Shijie if (vma->vm_file) { 6921da177e4SLinus Torvalds mapping = vma->vm_file->f_mapping; 69383cde9e8SDavidlohr Bueso i_mmap_lock_write(mapping); 69464ac4940SHuang Shijie } 6951da177e4SLinus Torvalds 6961da177e4SLinus Torvalds __vma_link(mm, vma, prev, rb_link, rb_parent); 6971da177e4SLinus Torvalds __vma_link_file(vma); 6981da177e4SLinus Torvalds 6991da177e4SLinus Torvalds if (mapping) 70083cde9e8SDavidlohr Bueso i_mmap_unlock_write(mapping); 7011da177e4SLinus Torvalds 7021da177e4SLinus Torvalds mm->map_count++; 7031da177e4SLinus Torvalds validate_mm(mm); 7041da177e4SLinus Torvalds } 7051da177e4SLinus Torvalds 7061da177e4SLinus Torvalds /* 70788f6b4c3SKautuk Consul * Helper for vma_adjust() in the split_vma insert case: insert a vma into the 7086b2dbba8SMichel Lespinasse * mm's list and rbtree. It has already been inserted into the interval tree. 7091da177e4SLinus Torvalds */ 71048aae425SZhenwenXu static void __insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma) 7111da177e4SLinus Torvalds { 7126597d783SHugh Dickins struct vm_area_struct *prev; 7131da177e4SLinus Torvalds struct rb_node **rb_link, *rb_parent; 7141da177e4SLinus Torvalds 7156597d783SHugh Dickins if (find_vma_links(mm, vma->vm_start, vma->vm_end, 7166597d783SHugh Dickins &prev, &rb_link, &rb_parent)) 7176597d783SHugh Dickins BUG(); 7181da177e4SLinus Torvalds __vma_link(mm, vma, prev, rb_link, rb_parent); 7191da177e4SLinus Torvalds mm->map_count++; 7201da177e4SLinus Torvalds } 7211da177e4SLinus Torvalds 7227c61f917SWei Yang static __always_inline void __vma_unlink(struct mm_struct *mm, 723e86f15eeSAndrea Arcangeli struct vm_area_struct *vma, 7248f26e0b1SAndrea Arcangeli struct vm_area_struct *ignore) 7251da177e4SLinus Torvalds { 7268f26e0b1SAndrea Arcangeli vma_rb_erase_ignore(vma, &mm->mm_rb, ignore); 7271b9fc5b2SWei Yang __vma_unlink_list(mm, vma); 728615d6e87SDavidlohr Bueso /* Kill the cache */ 729615d6e87SDavidlohr Bueso vmacache_invalidate(mm); 7301da177e4SLinus Torvalds } 7311da177e4SLinus Torvalds 7321da177e4SLinus Torvalds /* 7331da177e4SLinus Torvalds * We cannot adjust vm_start, vm_end, vm_pgoff fields of a vma that 7341da177e4SLinus Torvalds * is already present in an i_mmap tree without adjusting the tree. 7351da177e4SLinus Torvalds * The following helper function should be used when such adjustments 7361da177e4SLinus Torvalds * are necessary. The "insert" vma (if any) is to be inserted 7371da177e4SLinus Torvalds * before we drop the necessary locks. 7381da177e4SLinus Torvalds */ 739e86f15eeSAndrea Arcangeli int __vma_adjust(struct vm_area_struct *vma, unsigned long start, 740e86f15eeSAndrea Arcangeli unsigned long end, pgoff_t pgoff, struct vm_area_struct *insert, 741e86f15eeSAndrea Arcangeli struct vm_area_struct *expand) 7421da177e4SLinus Torvalds { 7431da177e4SLinus Torvalds struct mm_struct *mm = vma->vm_mm; 744e86f15eeSAndrea Arcangeli struct vm_area_struct *next = vma->vm_next, *orig_vma = vma; 7451da177e4SLinus Torvalds struct address_space *mapping = NULL; 746f808c13fSDavidlohr Bueso struct rb_root_cached *root = NULL; 747012f1800SRik van Riel struct anon_vma *anon_vma = NULL; 7481da177e4SLinus Torvalds struct file *file = vma->vm_file; 749d3737187SMichel Lespinasse bool start_changed = false, end_changed = false; 7501da177e4SLinus Torvalds long adjust_next = 0; 7511da177e4SLinus Torvalds int remove_next = 0; 7521da177e4SLinus Torvalds 7531da177e4SLinus Torvalds if (next && !insert) { 754734537c9SKirill A. Shutemov struct vm_area_struct *exporter = NULL, *importer = NULL; 755287d97acSLinus Torvalds 7561da177e4SLinus Torvalds if (end >= next->vm_end) { 7571da177e4SLinus Torvalds /* 7581da177e4SLinus Torvalds * vma expands, overlapping all the next, and 7591da177e4SLinus Torvalds * perhaps the one after too (mprotect case 6). 76086d12e47SAndrea Arcangeli * The only other cases that gets here are 761e86f15eeSAndrea Arcangeli * case 1, case 7 and case 8. 762e86f15eeSAndrea Arcangeli */ 763e86f15eeSAndrea Arcangeli if (next == expand) { 764e86f15eeSAndrea Arcangeli /* 765e86f15eeSAndrea Arcangeli * The only case where we don't expand "vma" 766e86f15eeSAndrea Arcangeli * and we expand "next" instead is case 8. 767e86f15eeSAndrea Arcangeli */ 768e86f15eeSAndrea Arcangeli VM_WARN_ON(end != next->vm_end); 769e86f15eeSAndrea Arcangeli /* 770e86f15eeSAndrea Arcangeli * remove_next == 3 means we're 771e86f15eeSAndrea Arcangeli * removing "vma" and that to do so we 772e86f15eeSAndrea Arcangeli * swapped "vma" and "next". 773e86f15eeSAndrea Arcangeli */ 774e86f15eeSAndrea Arcangeli remove_next = 3; 775e86f15eeSAndrea Arcangeli VM_WARN_ON(file != next->vm_file); 776e86f15eeSAndrea Arcangeli swap(vma, next); 777e86f15eeSAndrea Arcangeli } else { 778e86f15eeSAndrea Arcangeli VM_WARN_ON(expand != vma); 779e86f15eeSAndrea Arcangeli /* 780e86f15eeSAndrea Arcangeli * case 1, 6, 7, remove_next == 2 is case 6, 781e86f15eeSAndrea Arcangeli * remove_next == 1 is case 1 or 7. 7821da177e4SLinus Torvalds */ 783734537c9SKirill A. Shutemov remove_next = 1 + (end > next->vm_end); 784e86f15eeSAndrea Arcangeli VM_WARN_ON(remove_next == 2 && 785e86f15eeSAndrea Arcangeli end != next->vm_next->vm_end); 786e86f15eeSAndrea Arcangeli /* trim end to next, for case 6 first pass */ 7871da177e4SLinus Torvalds end = next->vm_end; 788e86f15eeSAndrea Arcangeli } 789e86f15eeSAndrea Arcangeli 790287d97acSLinus Torvalds exporter = next; 7911da177e4SLinus Torvalds importer = vma; 792734537c9SKirill A. Shutemov 793734537c9SKirill A. Shutemov /* 794734537c9SKirill A. Shutemov * If next doesn't have anon_vma, import from vma after 795734537c9SKirill A. Shutemov * next, if the vma overlaps with it. 796734537c9SKirill A. Shutemov */ 79797a42cd4SAndrea Arcangeli if (remove_next == 2 && !next->anon_vma) 798734537c9SKirill A. Shutemov exporter = next->vm_next; 799734537c9SKirill A. Shutemov 8001da177e4SLinus Torvalds } else if (end > next->vm_start) { 8011da177e4SLinus Torvalds /* 8021da177e4SLinus Torvalds * vma expands, overlapping part of the next: 8031da177e4SLinus Torvalds * mprotect case 5 shifting the boundary up. 8041da177e4SLinus Torvalds */ 805f9d86a60SWei Yang adjust_next = (end - next->vm_start); 806287d97acSLinus Torvalds exporter = next; 8071da177e4SLinus Torvalds importer = vma; 808e86f15eeSAndrea Arcangeli VM_WARN_ON(expand != importer); 8091da177e4SLinus Torvalds } else if (end < vma->vm_end) { 8101da177e4SLinus Torvalds /* 8111da177e4SLinus Torvalds * vma shrinks, and !insert tells it's not 8121da177e4SLinus Torvalds * split_vma inserting another: so it must be 8131da177e4SLinus Torvalds * mprotect case 4 shifting the boundary down. 8141da177e4SLinus Torvalds */ 815f9d86a60SWei Yang adjust_next = -(vma->vm_end - end); 816287d97acSLinus Torvalds exporter = vma; 8171da177e4SLinus Torvalds importer = next; 818e86f15eeSAndrea Arcangeli VM_WARN_ON(expand != importer); 8191da177e4SLinus Torvalds } 8201da177e4SLinus Torvalds 8215beb4930SRik van Riel /* 8225beb4930SRik van Riel * Easily overlooked: when mprotect shifts the boundary, 8235beb4930SRik van Riel * make sure the expanding vma has anon_vma set if the 8245beb4930SRik van Riel * shrinking vma had, to cover any anon pages imported. 8255beb4930SRik van Riel */ 826287d97acSLinus Torvalds if (exporter && exporter->anon_vma && !importer->anon_vma) { 827c4ea95d7SDaniel Forrest int error; 828c4ea95d7SDaniel Forrest 829287d97acSLinus Torvalds importer->anon_vma = exporter->anon_vma; 830b800c91aSKonstantin Khlebnikov error = anon_vma_clone(importer, exporter); 8313fe89b3eSLeon Yu if (error) 832b800c91aSKonstantin Khlebnikov return error; 833b800c91aSKonstantin Khlebnikov } 8345beb4930SRik van Riel } 835734537c9SKirill A. Shutemov again: 836e86f15eeSAndrea Arcangeli vma_adjust_trans_huge(orig_vma, start, end, adjust_next); 83737f9f559SKirill A. Shutemov 8381da177e4SLinus Torvalds if (file) { 8391da177e4SLinus Torvalds mapping = file->f_mapping; 8401da177e4SLinus Torvalds root = &mapping->i_mmap; 841cbc91f71SSrikar Dronamraju uprobe_munmap(vma, vma->vm_start, vma->vm_end); 842682968e0SSrikar Dronamraju 843682968e0SSrikar Dronamraju if (adjust_next) 84427ba0644SKirill A. Shutemov uprobe_munmap(next, next->vm_start, next->vm_end); 845682968e0SSrikar Dronamraju 84683cde9e8SDavidlohr Bueso i_mmap_lock_write(mapping); 8471da177e4SLinus Torvalds if (insert) { 8481da177e4SLinus Torvalds /* 8496b2dbba8SMichel Lespinasse * Put into interval tree now, so instantiated pages 8501da177e4SLinus Torvalds * are visible to arm/parisc __flush_dcache_page 8511da177e4SLinus Torvalds * throughout; but we cannot insert into address 8521da177e4SLinus Torvalds * space until vma start or end is updated. 8531da177e4SLinus Torvalds */ 8541da177e4SLinus Torvalds __vma_link_file(insert); 8551da177e4SLinus Torvalds } 8561da177e4SLinus Torvalds } 8571da177e4SLinus Torvalds 858012f1800SRik van Riel anon_vma = vma->anon_vma; 859bf181b9fSMichel Lespinasse if (!anon_vma && adjust_next) 860bf181b9fSMichel Lespinasse anon_vma = next->anon_vma; 861bf181b9fSMichel Lespinasse if (anon_vma) { 862e86f15eeSAndrea Arcangeli VM_WARN_ON(adjust_next && next->anon_vma && 863e86f15eeSAndrea Arcangeli anon_vma != next->anon_vma); 8644fc3f1d6SIngo Molnar anon_vma_lock_write(anon_vma); 865bf181b9fSMichel Lespinasse anon_vma_interval_tree_pre_update_vma(vma); 866bf181b9fSMichel Lespinasse if (adjust_next) 867bf181b9fSMichel Lespinasse anon_vma_interval_tree_pre_update_vma(next); 868bf181b9fSMichel Lespinasse } 869012f1800SRik van Riel 8700fc48a6eSWei Yang if (file) { 8711da177e4SLinus Torvalds flush_dcache_mmap_lock(mapping); 8726b2dbba8SMichel Lespinasse vma_interval_tree_remove(vma, root); 8731da177e4SLinus Torvalds if (adjust_next) 8746b2dbba8SMichel Lespinasse vma_interval_tree_remove(next, root); 8751da177e4SLinus Torvalds } 8761da177e4SLinus Torvalds 877d3737187SMichel Lespinasse if (start != vma->vm_start) { 8781da177e4SLinus Torvalds vma->vm_start = start; 879d3737187SMichel Lespinasse start_changed = true; 880d3737187SMichel Lespinasse } 881d3737187SMichel Lespinasse if (end != vma->vm_end) { 8821da177e4SLinus Torvalds vma->vm_end = end; 883d3737187SMichel Lespinasse end_changed = true; 884d3737187SMichel Lespinasse } 8851da177e4SLinus Torvalds vma->vm_pgoff = pgoff; 8861da177e4SLinus Torvalds if (adjust_next) { 887f9d86a60SWei Yang next->vm_start += adjust_next; 888f9d86a60SWei Yang next->vm_pgoff += adjust_next >> PAGE_SHIFT; 8891da177e4SLinus Torvalds } 8901da177e4SLinus Torvalds 8910fc48a6eSWei Yang if (file) { 8921da177e4SLinus Torvalds if (adjust_next) 8936b2dbba8SMichel Lespinasse vma_interval_tree_insert(next, root); 8946b2dbba8SMichel Lespinasse vma_interval_tree_insert(vma, root); 8951da177e4SLinus Torvalds flush_dcache_mmap_unlock(mapping); 8961da177e4SLinus Torvalds } 8971da177e4SLinus Torvalds 8981da177e4SLinus Torvalds if (remove_next) { 8991da177e4SLinus Torvalds /* 9001da177e4SLinus Torvalds * vma_merge has merged next into vma, and needs 9011da177e4SLinus Torvalds * us to remove next before dropping the locks. 9021da177e4SLinus Torvalds */ 903e86f15eeSAndrea Arcangeli if (remove_next != 3) 9047c61f917SWei Yang __vma_unlink(mm, next, next); 905e86f15eeSAndrea Arcangeli else 9068f26e0b1SAndrea Arcangeli /* 9078f26e0b1SAndrea Arcangeli * vma is not before next if they've been 9088f26e0b1SAndrea Arcangeli * swapped. 9098f26e0b1SAndrea Arcangeli * 9108f26e0b1SAndrea Arcangeli * pre-swap() next->vm_start was reduced so 9118f26e0b1SAndrea Arcangeli * tell validate_mm_rb to ignore pre-swap() 9128f26e0b1SAndrea Arcangeli * "next" (which is stored in post-swap() 9138f26e0b1SAndrea Arcangeli * "vma"). 9148f26e0b1SAndrea Arcangeli */ 9157c61f917SWei Yang __vma_unlink(mm, next, vma); 9161da177e4SLinus Torvalds if (file) 9171da177e4SLinus Torvalds __remove_shared_vm_struct(next, file, mapping); 9181da177e4SLinus Torvalds } else if (insert) { 9191da177e4SLinus Torvalds /* 9201da177e4SLinus Torvalds * split_vma has split insert from vma, and needs 9211da177e4SLinus Torvalds * us to insert it before dropping the locks 9221da177e4SLinus Torvalds * (it may either follow vma or precede it). 9231da177e4SLinus Torvalds */ 9241da177e4SLinus Torvalds __insert_vm_struct(mm, insert); 925d3737187SMichel Lespinasse } else { 926d3737187SMichel Lespinasse if (start_changed) 927d3737187SMichel Lespinasse vma_gap_update(vma); 928d3737187SMichel Lespinasse if (end_changed) { 929d3737187SMichel Lespinasse if (!next) 9301be7107fSHugh Dickins mm->highest_vm_end = vm_end_gap(vma); 931d3737187SMichel Lespinasse else if (!adjust_next) 932d3737187SMichel Lespinasse vma_gap_update(next); 933d3737187SMichel Lespinasse } 9341da177e4SLinus Torvalds } 9351da177e4SLinus Torvalds 936bf181b9fSMichel Lespinasse if (anon_vma) { 937bf181b9fSMichel Lespinasse anon_vma_interval_tree_post_update_vma(vma); 938bf181b9fSMichel Lespinasse if (adjust_next) 939bf181b9fSMichel Lespinasse anon_vma_interval_tree_post_update_vma(next); 94008b52706SKonstantin Khlebnikov anon_vma_unlock_write(anon_vma); 941bf181b9fSMichel Lespinasse } 9421da177e4SLinus Torvalds 9430fc48a6eSWei Yang if (file) { 944808fbdbeSWei Yang i_mmap_unlock_write(mapping); 9457b2d81d4SIngo Molnar uprobe_mmap(vma); 9462b144498SSrikar Dronamraju 9472b144498SSrikar Dronamraju if (adjust_next) 9487b2d81d4SIngo Molnar uprobe_mmap(next); 9492b144498SSrikar Dronamraju } 9502b144498SSrikar Dronamraju 9511da177e4SLinus Torvalds if (remove_next) { 952925d1c40SMatt Helsley if (file) { 953cbc91f71SSrikar Dronamraju uprobe_munmap(next, next->vm_start, next->vm_end); 9541da177e4SLinus Torvalds fput(file); 955925d1c40SMatt Helsley } 9565beb4930SRik van Riel if (next->anon_vma) 9575beb4930SRik van Riel anon_vma_merge(vma, next); 9581da177e4SLinus Torvalds mm->map_count--; 9593964acd0SOleg Nesterov mpol_put(vma_policy(next)); 9603928d4f5SLinus Torvalds vm_area_free(next); 9611da177e4SLinus Torvalds /* 9621da177e4SLinus Torvalds * In mprotect's case 6 (see comments on vma_merge), 9631da177e4SLinus Torvalds * we must remove another next too. It would clutter 9641da177e4SLinus Torvalds * up the code too much to do both in one go. 9651da177e4SLinus Torvalds */ 966e86f15eeSAndrea Arcangeli if (remove_next != 3) { 967e86f15eeSAndrea Arcangeli /* 968e86f15eeSAndrea Arcangeli * If "next" was removed and vma->vm_end was 969e86f15eeSAndrea Arcangeli * expanded (up) over it, in turn 970e86f15eeSAndrea Arcangeli * "next->vm_prev->vm_end" changed and the 971e86f15eeSAndrea Arcangeli * "vma->vm_next" gap must be updated. 972e86f15eeSAndrea Arcangeli */ 9731da177e4SLinus Torvalds next = vma->vm_next; 974e86f15eeSAndrea Arcangeli } else { 975e86f15eeSAndrea Arcangeli /* 976e86f15eeSAndrea Arcangeli * For the scope of the comment "next" and 977e86f15eeSAndrea Arcangeli * "vma" considered pre-swap(): if "vma" was 978e86f15eeSAndrea Arcangeli * removed, next->vm_start was expanded (down) 979e86f15eeSAndrea Arcangeli * over it and the "next" gap must be updated. 980e86f15eeSAndrea Arcangeli * Because of the swap() the post-swap() "vma" 981e86f15eeSAndrea Arcangeli * actually points to pre-swap() "next" 982e86f15eeSAndrea Arcangeli * (post-swap() "next" as opposed is now a 983e86f15eeSAndrea Arcangeli * dangling pointer). 984e86f15eeSAndrea Arcangeli */ 985e86f15eeSAndrea Arcangeli next = vma; 986e86f15eeSAndrea Arcangeli } 987734537c9SKirill A. Shutemov if (remove_next == 2) { 988734537c9SKirill A. Shutemov remove_next = 1; 989734537c9SKirill A. Shutemov end = next->vm_end; 9901da177e4SLinus Torvalds goto again; 991734537c9SKirill A. Shutemov } 992d3737187SMichel Lespinasse else if (next) 993d3737187SMichel Lespinasse vma_gap_update(next); 994fb8c41e9SAndrea Arcangeli else { 995fb8c41e9SAndrea Arcangeli /* 996fb8c41e9SAndrea Arcangeli * If remove_next == 2 we obviously can't 997fb8c41e9SAndrea Arcangeli * reach this path. 998fb8c41e9SAndrea Arcangeli * 999fb8c41e9SAndrea Arcangeli * If remove_next == 3 we can't reach this 1000fb8c41e9SAndrea Arcangeli * path because pre-swap() next is always not 1001fb8c41e9SAndrea Arcangeli * NULL. pre-swap() "next" is not being 1002fb8c41e9SAndrea Arcangeli * removed and its next->vm_end is not altered 1003fb8c41e9SAndrea Arcangeli * (and furthermore "end" already matches 1004fb8c41e9SAndrea Arcangeli * next->vm_end in remove_next == 3). 1005fb8c41e9SAndrea Arcangeli * 1006fb8c41e9SAndrea Arcangeli * We reach this only in the remove_next == 1 1007fb8c41e9SAndrea Arcangeli * case if the "next" vma that was removed was 1008fb8c41e9SAndrea Arcangeli * the highest vma of the mm. However in such 1009fb8c41e9SAndrea Arcangeli * case next->vm_end == "end" and the extended 1010fb8c41e9SAndrea Arcangeli * "vma" has vma->vm_end == next->vm_end so 1011fb8c41e9SAndrea Arcangeli * mm->highest_vm_end doesn't need any update 1012fb8c41e9SAndrea Arcangeli * in remove_next == 1 case. 1013fb8c41e9SAndrea Arcangeli */ 10141be7107fSHugh Dickins VM_WARN_ON(mm->highest_vm_end != vm_end_gap(vma)); 1015fb8c41e9SAndrea Arcangeli } 10161da177e4SLinus Torvalds } 10172b144498SSrikar Dronamraju if (insert && file) 10187b2d81d4SIngo Molnar uprobe_mmap(insert); 10191da177e4SLinus Torvalds 10201da177e4SLinus Torvalds validate_mm(mm); 10215beb4930SRik van Riel 10225beb4930SRik van Riel return 0; 10231da177e4SLinus Torvalds } 10241da177e4SLinus Torvalds 10251da177e4SLinus Torvalds /* 10261da177e4SLinus Torvalds * If the vma has a ->close operation then the driver probably needs to release 10271da177e4SLinus Torvalds * per-vma resources, so we don't attempt to merge those. 10281da177e4SLinus Torvalds */ 10291da177e4SLinus Torvalds static inline int is_mergeable_vma(struct vm_area_struct *vma, 103019a809afSAndrea Arcangeli struct file *file, unsigned long vm_flags, 103119a809afSAndrea Arcangeli struct vm_userfaultfd_ctx vm_userfaultfd_ctx) 10321da177e4SLinus Torvalds { 103334228d47SCyrill Gorcunov /* 103434228d47SCyrill Gorcunov * VM_SOFTDIRTY should not prevent from VMA merging, if we 103534228d47SCyrill Gorcunov * match the flags but dirty bit -- the caller should mark 103634228d47SCyrill Gorcunov * merged VMA as dirty. If dirty bit won't be excluded from 10378bb4e7a2SWei Yang * comparison, we increase pressure on the memory system forcing 103834228d47SCyrill Gorcunov * the kernel to generate new VMAs when old one could be 103934228d47SCyrill Gorcunov * extended instead. 104034228d47SCyrill Gorcunov */ 104134228d47SCyrill Gorcunov if ((vma->vm_flags ^ vm_flags) & ~VM_SOFTDIRTY) 10421da177e4SLinus Torvalds return 0; 10431da177e4SLinus Torvalds if (vma->vm_file != file) 10441da177e4SLinus Torvalds return 0; 10451da177e4SLinus Torvalds if (vma->vm_ops && vma->vm_ops->close) 10461da177e4SLinus Torvalds return 0; 104719a809afSAndrea Arcangeli if (!is_mergeable_vm_userfaultfd_ctx(vma, vm_userfaultfd_ctx)) 104819a809afSAndrea Arcangeli return 0; 10491da177e4SLinus Torvalds return 1; 10501da177e4SLinus Torvalds } 10511da177e4SLinus Torvalds 10521da177e4SLinus Torvalds static inline int is_mergeable_anon_vma(struct anon_vma *anon_vma1, 1053965f55deSShaohua Li struct anon_vma *anon_vma2, 1054965f55deSShaohua Li struct vm_area_struct *vma) 10551da177e4SLinus Torvalds { 1056965f55deSShaohua Li /* 1057965f55deSShaohua Li * The list_is_singular() test is to avoid merging VMA cloned from 1058965f55deSShaohua Li * parents. This can improve scalability caused by anon_vma lock. 1059965f55deSShaohua Li */ 1060965f55deSShaohua Li if ((!anon_vma1 || !anon_vma2) && (!vma || 1061965f55deSShaohua Li list_is_singular(&vma->anon_vma_chain))) 1062965f55deSShaohua Li return 1; 1063965f55deSShaohua Li return anon_vma1 == anon_vma2; 10641da177e4SLinus Torvalds } 10651da177e4SLinus Torvalds 10661da177e4SLinus Torvalds /* 10671da177e4SLinus Torvalds * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff) 10681da177e4SLinus Torvalds * in front of (at a lower virtual address and file offset than) the vma. 10691da177e4SLinus Torvalds * 10701da177e4SLinus Torvalds * We cannot merge two vmas if they have differently assigned (non-NULL) 10711da177e4SLinus Torvalds * anon_vmas, nor if same anon_vma is assigned but offsets incompatible. 10721da177e4SLinus Torvalds * 10731da177e4SLinus Torvalds * We don't check here for the merged mmap wrapping around the end of pagecache 107445e55300SPeter Collingbourne * indices (16TB on ia32) because do_mmap() does not permit mmap's which 10751da177e4SLinus Torvalds * wrap, nor mmaps which cover the final page at index -1UL. 10761da177e4SLinus Torvalds */ 10771da177e4SLinus Torvalds static int 10781da177e4SLinus Torvalds can_vma_merge_before(struct vm_area_struct *vma, unsigned long vm_flags, 107919a809afSAndrea Arcangeli struct anon_vma *anon_vma, struct file *file, 108019a809afSAndrea Arcangeli pgoff_t vm_pgoff, 108119a809afSAndrea Arcangeli struct vm_userfaultfd_ctx vm_userfaultfd_ctx) 10821da177e4SLinus Torvalds { 108319a809afSAndrea Arcangeli if (is_mergeable_vma(vma, file, vm_flags, vm_userfaultfd_ctx) && 1084965f55deSShaohua Li is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) { 10851da177e4SLinus Torvalds if (vma->vm_pgoff == vm_pgoff) 10861da177e4SLinus Torvalds return 1; 10871da177e4SLinus Torvalds } 10881da177e4SLinus Torvalds return 0; 10891da177e4SLinus Torvalds } 10901da177e4SLinus Torvalds 10911da177e4SLinus Torvalds /* 10921da177e4SLinus Torvalds * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff) 10931da177e4SLinus Torvalds * beyond (at a higher virtual address and file offset than) the vma. 10941da177e4SLinus Torvalds * 10951da177e4SLinus Torvalds * We cannot merge two vmas if they have differently assigned (non-NULL) 10961da177e4SLinus Torvalds * anon_vmas, nor if same anon_vma is assigned but offsets incompatible. 10971da177e4SLinus Torvalds */ 10981da177e4SLinus Torvalds static int 10991da177e4SLinus Torvalds can_vma_merge_after(struct vm_area_struct *vma, unsigned long vm_flags, 110019a809afSAndrea Arcangeli struct anon_vma *anon_vma, struct file *file, 110119a809afSAndrea Arcangeli pgoff_t vm_pgoff, 110219a809afSAndrea Arcangeli struct vm_userfaultfd_ctx vm_userfaultfd_ctx) 11031da177e4SLinus Torvalds { 110419a809afSAndrea Arcangeli if (is_mergeable_vma(vma, file, vm_flags, vm_userfaultfd_ctx) && 1105965f55deSShaohua Li is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) { 11061da177e4SLinus Torvalds pgoff_t vm_pglen; 1107d6e93217SLibin vm_pglen = vma_pages(vma); 11081da177e4SLinus Torvalds if (vma->vm_pgoff + vm_pglen == vm_pgoff) 11091da177e4SLinus Torvalds return 1; 11101da177e4SLinus Torvalds } 11111da177e4SLinus Torvalds return 0; 11121da177e4SLinus Torvalds } 11131da177e4SLinus Torvalds 11141da177e4SLinus Torvalds /* 11151da177e4SLinus Torvalds * Given a mapping request (addr,end,vm_flags,file,pgoff), figure out 11161da177e4SLinus Torvalds * whether that can be merged with its predecessor or its successor. 11171da177e4SLinus Torvalds * Or both (it neatly fills a hole). 11181da177e4SLinus Torvalds * 11191da177e4SLinus Torvalds * In most cases - when called for mmap, brk or mremap - [addr,end) is 11201da177e4SLinus Torvalds * certain not to be mapped by the time vma_merge is called; but when 11211da177e4SLinus Torvalds * called for mprotect, it is certain to be already mapped (either at 11221da177e4SLinus Torvalds * an offset within prev, or at the start of next), and the flags of 11231da177e4SLinus Torvalds * this area are about to be changed to vm_flags - and the no-change 11241da177e4SLinus Torvalds * case has already been eliminated. 11251da177e4SLinus Torvalds * 11261da177e4SLinus Torvalds * The following mprotect cases have to be considered, where AAAA is 11271da177e4SLinus Torvalds * the area passed down from mprotect_fixup, never extending beyond one 11281da177e4SLinus Torvalds * vma, PPPPPP is the prev vma specified, and NNNNNN the next vma after: 11291da177e4SLinus Torvalds * 11305d42ab29SWei Yang * AAAA AAAA AAAA 11315d42ab29SWei Yang * PPPPPPNNNNNN PPPPPPNNNNNN PPPPPPNNNNNN 11325d42ab29SWei Yang * cannot merge might become might become 11335d42ab29SWei Yang * PPNNNNNNNNNN PPPPPPPPPPNN 11345d42ab29SWei Yang * mmap, brk or case 4 below case 5 below 11355d42ab29SWei Yang * mremap move: 11365d42ab29SWei Yang * AAAA AAAA 11375d42ab29SWei Yang * PPPP NNNN PPPPNNNNXXXX 11385d42ab29SWei Yang * might become might become 11395d42ab29SWei Yang * PPPPPPPPPPPP 1 or PPPPPPPPPPPP 6 or 11405d42ab29SWei Yang * PPPPPPPPNNNN 2 or PPPPPPPPXXXX 7 or 11415d42ab29SWei Yang * PPPPNNNNNNNN 3 PPPPXXXXXXXX 8 11421da177e4SLinus Torvalds * 11438bb4e7a2SWei Yang * It is important for case 8 that the vma NNNN overlapping the 1144e86f15eeSAndrea Arcangeli * region AAAA is never going to extended over XXXX. Instead XXXX must 1145e86f15eeSAndrea Arcangeli * be extended in region AAAA and NNNN must be removed. This way in 1146e86f15eeSAndrea Arcangeli * all cases where vma_merge succeeds, the moment vma_adjust drops the 1147e86f15eeSAndrea Arcangeli * rmap_locks, the properties of the merged vma will be already 1148e86f15eeSAndrea Arcangeli * correct for the whole merged range. Some of those properties like 1149e86f15eeSAndrea Arcangeli * vm_page_prot/vm_flags may be accessed by rmap_walks and they must 1150e86f15eeSAndrea Arcangeli * be correct for the whole merged range immediately after the 1151e86f15eeSAndrea Arcangeli * rmap_locks are released. Otherwise if XXXX would be removed and 1152e86f15eeSAndrea Arcangeli * NNNN would be extended over the XXXX range, remove_migration_ptes 1153e86f15eeSAndrea Arcangeli * or other rmap walkers (if working on addresses beyond the "end" 1154e86f15eeSAndrea Arcangeli * parameter) may establish ptes with the wrong permissions of NNNN 1155e86f15eeSAndrea Arcangeli * instead of the right permissions of XXXX. 11561da177e4SLinus Torvalds */ 11571da177e4SLinus Torvalds struct vm_area_struct *vma_merge(struct mm_struct *mm, 11581da177e4SLinus Torvalds struct vm_area_struct *prev, unsigned long addr, 11591da177e4SLinus Torvalds unsigned long end, unsigned long vm_flags, 11601da177e4SLinus Torvalds struct anon_vma *anon_vma, struct file *file, 116119a809afSAndrea Arcangeli pgoff_t pgoff, struct mempolicy *policy, 116219a809afSAndrea Arcangeli struct vm_userfaultfd_ctx vm_userfaultfd_ctx) 11631da177e4SLinus Torvalds { 11641da177e4SLinus Torvalds pgoff_t pglen = (end - addr) >> PAGE_SHIFT; 11651da177e4SLinus Torvalds struct vm_area_struct *area, *next; 11665beb4930SRik van Riel int err; 11671da177e4SLinus Torvalds 11681da177e4SLinus Torvalds /* 11691da177e4SLinus Torvalds * We later require that vma->vm_flags == vm_flags, 11701da177e4SLinus Torvalds * so this tests vma->vm_flags & VM_SPECIAL, too. 11711da177e4SLinus Torvalds */ 11721da177e4SLinus Torvalds if (vm_flags & VM_SPECIAL) 11731da177e4SLinus Torvalds return NULL; 11741da177e4SLinus Torvalds 11753903b55aSLiam R. Howlett next = vma_next(mm, prev); 11761da177e4SLinus Torvalds area = next; 1177e86f15eeSAndrea Arcangeli if (area && area->vm_end == end) /* cases 6, 7, 8 */ 11781da177e4SLinus Torvalds next = next->vm_next; 11791da177e4SLinus Torvalds 1180e86f15eeSAndrea Arcangeli /* verify some invariant that must be enforced by the caller */ 1181e86f15eeSAndrea Arcangeli VM_WARN_ON(prev && addr <= prev->vm_start); 1182e86f15eeSAndrea Arcangeli VM_WARN_ON(area && end > area->vm_end); 1183e86f15eeSAndrea Arcangeli VM_WARN_ON(addr >= end); 1184e86f15eeSAndrea Arcangeli 11851da177e4SLinus Torvalds /* 11861da177e4SLinus Torvalds * Can it merge with the predecessor? 11871da177e4SLinus Torvalds */ 11881da177e4SLinus Torvalds if (prev && prev->vm_end == addr && 11891da177e4SLinus Torvalds mpol_equal(vma_policy(prev), policy) && 11901da177e4SLinus Torvalds can_vma_merge_after(prev, vm_flags, 119119a809afSAndrea Arcangeli anon_vma, file, pgoff, 119219a809afSAndrea Arcangeli vm_userfaultfd_ctx)) { 11931da177e4SLinus Torvalds /* 11941da177e4SLinus Torvalds * OK, it can. Can we now merge in the successor as well? 11951da177e4SLinus Torvalds */ 11961da177e4SLinus Torvalds if (next && end == next->vm_start && 11971da177e4SLinus Torvalds mpol_equal(policy, vma_policy(next)) && 11981da177e4SLinus Torvalds can_vma_merge_before(next, vm_flags, 119919a809afSAndrea Arcangeli anon_vma, file, 120019a809afSAndrea Arcangeli pgoff+pglen, 120119a809afSAndrea Arcangeli vm_userfaultfd_ctx) && 12021da177e4SLinus Torvalds is_mergeable_anon_vma(prev->anon_vma, 1203965f55deSShaohua Li next->anon_vma, NULL)) { 12041da177e4SLinus Torvalds /* cases 1, 6 */ 1205e86f15eeSAndrea Arcangeli err = __vma_adjust(prev, prev->vm_start, 1206e86f15eeSAndrea Arcangeli next->vm_end, prev->vm_pgoff, NULL, 1207e86f15eeSAndrea Arcangeli prev); 12081da177e4SLinus Torvalds } else /* cases 2, 5, 7 */ 1209e86f15eeSAndrea Arcangeli err = __vma_adjust(prev, prev->vm_start, 1210e86f15eeSAndrea Arcangeli end, prev->vm_pgoff, NULL, prev); 12115beb4930SRik van Riel if (err) 12125beb4930SRik van Riel return NULL; 12136d50e60cSDavid Rientjes khugepaged_enter_vma_merge(prev, vm_flags); 12141da177e4SLinus Torvalds return prev; 12151da177e4SLinus Torvalds } 12161da177e4SLinus Torvalds 12171da177e4SLinus Torvalds /* 12181da177e4SLinus Torvalds * Can this new request be merged in front of next? 12191da177e4SLinus Torvalds */ 12201da177e4SLinus Torvalds if (next && end == next->vm_start && 12211da177e4SLinus Torvalds mpol_equal(policy, vma_policy(next)) && 12221da177e4SLinus Torvalds can_vma_merge_before(next, vm_flags, 122319a809afSAndrea Arcangeli anon_vma, file, pgoff+pglen, 122419a809afSAndrea Arcangeli vm_userfaultfd_ctx)) { 12251da177e4SLinus Torvalds if (prev && addr < prev->vm_end) /* case 4 */ 1226e86f15eeSAndrea Arcangeli err = __vma_adjust(prev, prev->vm_start, 1227e86f15eeSAndrea Arcangeli addr, prev->vm_pgoff, NULL, next); 1228e86f15eeSAndrea Arcangeli else { /* cases 3, 8 */ 1229e86f15eeSAndrea Arcangeli err = __vma_adjust(area, addr, next->vm_end, 1230e86f15eeSAndrea Arcangeli next->vm_pgoff - pglen, NULL, next); 1231e86f15eeSAndrea Arcangeli /* 1232e86f15eeSAndrea Arcangeli * In case 3 area is already equal to next and 1233e86f15eeSAndrea Arcangeli * this is a noop, but in case 8 "area" has 1234e86f15eeSAndrea Arcangeli * been removed and next was expanded over it. 1235e86f15eeSAndrea Arcangeli */ 1236e86f15eeSAndrea Arcangeli area = next; 1237e86f15eeSAndrea Arcangeli } 12385beb4930SRik van Riel if (err) 12395beb4930SRik van Riel return NULL; 12406d50e60cSDavid Rientjes khugepaged_enter_vma_merge(area, vm_flags); 12411da177e4SLinus Torvalds return area; 12421da177e4SLinus Torvalds } 12431da177e4SLinus Torvalds 12441da177e4SLinus Torvalds return NULL; 12451da177e4SLinus Torvalds } 12461da177e4SLinus Torvalds 12471da177e4SLinus Torvalds /* 1248b4f315b4SEthon Paul * Rough compatibility check to quickly see if it's even worth looking 1249d0e9fe17SLinus Torvalds * at sharing an anon_vma. 1250d0e9fe17SLinus Torvalds * 1251d0e9fe17SLinus Torvalds * They need to have the same vm_file, and the flags can only differ 1252d0e9fe17SLinus Torvalds * in things that mprotect may change. 1253d0e9fe17SLinus Torvalds * 1254d0e9fe17SLinus Torvalds * NOTE! The fact that we share an anon_vma doesn't _have_ to mean that 1255d0e9fe17SLinus Torvalds * we can merge the two vma's. For example, we refuse to merge a vma if 1256d0e9fe17SLinus Torvalds * there is a vm_ops->close() function, because that indicates that the 1257d0e9fe17SLinus Torvalds * driver is doing some kind of reference counting. But that doesn't 1258d0e9fe17SLinus Torvalds * really matter for the anon_vma sharing case. 1259d0e9fe17SLinus Torvalds */ 1260d0e9fe17SLinus Torvalds static int anon_vma_compatible(struct vm_area_struct *a, struct vm_area_struct *b) 1261d0e9fe17SLinus Torvalds { 1262d0e9fe17SLinus Torvalds return a->vm_end == b->vm_start && 1263d0e9fe17SLinus Torvalds mpol_equal(vma_policy(a), vma_policy(b)) && 1264d0e9fe17SLinus Torvalds a->vm_file == b->vm_file && 12656cb4d9a2SAnshuman Khandual !((a->vm_flags ^ b->vm_flags) & ~(VM_ACCESS_FLAGS | VM_SOFTDIRTY)) && 1266d0e9fe17SLinus Torvalds b->vm_pgoff == a->vm_pgoff + ((b->vm_start - a->vm_start) >> PAGE_SHIFT); 1267d0e9fe17SLinus Torvalds } 1268d0e9fe17SLinus Torvalds 1269d0e9fe17SLinus Torvalds /* 1270d0e9fe17SLinus Torvalds * Do some basic sanity checking to see if we can re-use the anon_vma 1271d0e9fe17SLinus Torvalds * from 'old'. The 'a'/'b' vma's are in VM order - one of them will be 1272d0e9fe17SLinus Torvalds * the same as 'old', the other will be the new one that is trying 1273d0e9fe17SLinus Torvalds * to share the anon_vma. 1274d0e9fe17SLinus Torvalds * 1275d0e9fe17SLinus Torvalds * NOTE! This runs with mm_sem held for reading, so it is possible that 1276d0e9fe17SLinus Torvalds * the anon_vma of 'old' is concurrently in the process of being set up 1277d0e9fe17SLinus Torvalds * by another page fault trying to merge _that_. But that's ok: if it 1278d0e9fe17SLinus Torvalds * is being set up, that automatically means that it will be a singleton 1279d0e9fe17SLinus Torvalds * acceptable for merging, so we can do all of this optimistically. But 12804db0c3c2SJason Low * we do that READ_ONCE() to make sure that we never re-load the pointer. 1281d0e9fe17SLinus Torvalds * 1282d0e9fe17SLinus Torvalds * IOW: that the "list_is_singular()" test on the anon_vma_chain only 1283d0e9fe17SLinus Torvalds * matters for the 'stable anon_vma' case (ie the thing we want to avoid 1284d0e9fe17SLinus Torvalds * is to return an anon_vma that is "complex" due to having gone through 1285d0e9fe17SLinus Torvalds * a fork). 1286d0e9fe17SLinus Torvalds * 1287d0e9fe17SLinus Torvalds * We also make sure that the two vma's are compatible (adjacent, 1288d0e9fe17SLinus Torvalds * and with the same memory policies). That's all stable, even with just 1289d0e9fe17SLinus Torvalds * a read lock on the mm_sem. 1290d0e9fe17SLinus Torvalds */ 1291d0e9fe17SLinus Torvalds static struct anon_vma *reusable_anon_vma(struct vm_area_struct *old, struct vm_area_struct *a, struct vm_area_struct *b) 1292d0e9fe17SLinus Torvalds { 1293d0e9fe17SLinus Torvalds if (anon_vma_compatible(a, b)) { 12944db0c3c2SJason Low struct anon_vma *anon_vma = READ_ONCE(old->anon_vma); 1295d0e9fe17SLinus Torvalds 1296d0e9fe17SLinus Torvalds if (anon_vma && list_is_singular(&old->anon_vma_chain)) 1297d0e9fe17SLinus Torvalds return anon_vma; 1298d0e9fe17SLinus Torvalds } 1299d0e9fe17SLinus Torvalds return NULL; 1300d0e9fe17SLinus Torvalds } 1301d0e9fe17SLinus Torvalds 1302d0e9fe17SLinus Torvalds /* 13031da177e4SLinus Torvalds * find_mergeable_anon_vma is used by anon_vma_prepare, to check 13041da177e4SLinus Torvalds * neighbouring vmas for a suitable anon_vma, before it goes off 13051da177e4SLinus Torvalds * to allocate a new anon_vma. It checks because a repetitive 13061da177e4SLinus Torvalds * sequence of mprotects and faults may otherwise lead to distinct 13071da177e4SLinus Torvalds * anon_vmas being allocated, preventing vma merge in subsequent 13081da177e4SLinus Torvalds * mprotect. 13091da177e4SLinus Torvalds */ 13101da177e4SLinus Torvalds struct anon_vma *find_mergeable_anon_vma(struct vm_area_struct *vma) 13111da177e4SLinus Torvalds { 1312a67c8caaSMiaohe Lin struct anon_vma *anon_vma = NULL; 13131da177e4SLinus Torvalds 1314a67c8caaSMiaohe Lin /* Try next first. */ 1315a67c8caaSMiaohe Lin if (vma->vm_next) { 1316a67c8caaSMiaohe Lin anon_vma = reusable_anon_vma(vma->vm_next, vma, vma->vm_next); 1317d0e9fe17SLinus Torvalds if (anon_vma) 1318d0e9fe17SLinus Torvalds return anon_vma; 1319a67c8caaSMiaohe Lin } 13201da177e4SLinus Torvalds 1321a67c8caaSMiaohe Lin /* Try prev next. */ 1322a67c8caaSMiaohe Lin if (vma->vm_prev) 1323a67c8caaSMiaohe Lin anon_vma = reusable_anon_vma(vma->vm_prev, vma->vm_prev, vma); 1324a67c8caaSMiaohe Lin 13251da177e4SLinus Torvalds /* 1326a67c8caaSMiaohe Lin * We might reach here with anon_vma == NULL if we can't find 1327a67c8caaSMiaohe Lin * any reusable anon_vma. 13281da177e4SLinus Torvalds * There's no absolute need to look only at touching neighbours: 13291da177e4SLinus Torvalds * we could search further afield for "compatible" anon_vmas. 13301da177e4SLinus Torvalds * But it would probably just be a waste of time searching, 13311da177e4SLinus Torvalds * or lead to too many vmas hanging off the same anon_vma. 13321da177e4SLinus Torvalds * We're trying to allow mprotect remerging later on, 13331da177e4SLinus Torvalds * not trying to minimize memory used for anon_vmas. 13341da177e4SLinus Torvalds */ 1335a67c8caaSMiaohe Lin return anon_vma; 13361da177e4SLinus Torvalds } 13371da177e4SLinus Torvalds 13381da177e4SLinus Torvalds /* 133940401530SAl Viro * If a hint addr is less than mmap_min_addr change hint to be as 134040401530SAl Viro * low as possible but still greater than mmap_min_addr 134140401530SAl Viro */ 134240401530SAl Viro static inline unsigned long round_hint_to_min(unsigned long hint) 134340401530SAl Viro { 134440401530SAl Viro hint &= PAGE_MASK; 134540401530SAl Viro if (((void *)hint != NULL) && 134640401530SAl Viro (hint < mmap_min_addr)) 134740401530SAl Viro return PAGE_ALIGN(mmap_min_addr); 134840401530SAl Viro return hint; 134940401530SAl Viro } 135040401530SAl Viro 1351363ee17fSDavidlohr Bueso static inline int mlock_future_check(struct mm_struct *mm, 1352363ee17fSDavidlohr Bueso unsigned long flags, 1353363ee17fSDavidlohr Bueso unsigned long len) 1354363ee17fSDavidlohr Bueso { 1355363ee17fSDavidlohr Bueso unsigned long locked, lock_limit; 1356363ee17fSDavidlohr Bueso 1357363ee17fSDavidlohr Bueso /* mlock MCL_FUTURE? */ 1358363ee17fSDavidlohr Bueso if (flags & VM_LOCKED) { 1359363ee17fSDavidlohr Bueso locked = len >> PAGE_SHIFT; 1360363ee17fSDavidlohr Bueso locked += mm->locked_vm; 1361363ee17fSDavidlohr Bueso lock_limit = rlimit(RLIMIT_MEMLOCK); 1362363ee17fSDavidlohr Bueso lock_limit >>= PAGE_SHIFT; 1363363ee17fSDavidlohr Bueso if (locked > lock_limit && !capable(CAP_IPC_LOCK)) 1364363ee17fSDavidlohr Bueso return -EAGAIN; 1365363ee17fSDavidlohr Bueso } 1366363ee17fSDavidlohr Bueso return 0; 1367363ee17fSDavidlohr Bueso } 1368363ee17fSDavidlohr Bueso 1369be83bbf8SLinus Torvalds static inline u64 file_mmap_size_max(struct file *file, struct inode *inode) 1370be83bbf8SLinus Torvalds { 1371be83bbf8SLinus Torvalds if (S_ISREG(inode->i_mode)) 1372423913adSLinus Torvalds return MAX_LFS_FILESIZE; 1373be83bbf8SLinus Torvalds 1374be83bbf8SLinus Torvalds if (S_ISBLK(inode->i_mode)) 1375be83bbf8SLinus Torvalds return MAX_LFS_FILESIZE; 1376be83bbf8SLinus Torvalds 137776f34950SIvan Khoronzhuk if (S_ISSOCK(inode->i_mode)) 137876f34950SIvan Khoronzhuk return MAX_LFS_FILESIZE; 137976f34950SIvan Khoronzhuk 1380be83bbf8SLinus Torvalds /* Special "we do even unsigned file positions" case */ 1381be83bbf8SLinus Torvalds if (file->f_mode & FMODE_UNSIGNED_OFFSET) 1382be83bbf8SLinus Torvalds return 0; 1383be83bbf8SLinus Torvalds 1384be83bbf8SLinus Torvalds /* Yes, random drivers might want more. But I'm tired of buggy drivers */ 1385be83bbf8SLinus Torvalds return ULONG_MAX; 1386be83bbf8SLinus Torvalds } 1387be83bbf8SLinus Torvalds 1388be83bbf8SLinus Torvalds static inline bool file_mmap_ok(struct file *file, struct inode *inode, 1389be83bbf8SLinus Torvalds unsigned long pgoff, unsigned long len) 1390be83bbf8SLinus Torvalds { 1391be83bbf8SLinus Torvalds u64 maxsize = file_mmap_size_max(file, inode); 1392be83bbf8SLinus Torvalds 1393be83bbf8SLinus Torvalds if (maxsize && len > maxsize) 1394be83bbf8SLinus Torvalds return false; 1395be83bbf8SLinus Torvalds maxsize -= len; 1396be83bbf8SLinus Torvalds if (pgoff > maxsize >> PAGE_SHIFT) 1397be83bbf8SLinus Torvalds return false; 1398be83bbf8SLinus Torvalds return true; 1399be83bbf8SLinus Torvalds } 1400be83bbf8SLinus Torvalds 140140401530SAl Viro /* 14023e4e28c5SMichel Lespinasse * The caller must write-lock current->mm->mmap_lock. 14031da177e4SLinus Torvalds */ 14041fcfd8dbSOleg Nesterov unsigned long do_mmap(struct file *file, unsigned long addr, 14051da177e4SLinus Torvalds unsigned long len, unsigned long prot, 140645e55300SPeter Collingbourne unsigned long flags, unsigned long pgoff, 140745e55300SPeter Collingbourne unsigned long *populate, struct list_head *uf) 14081da177e4SLinus Torvalds { 14091da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 141045e55300SPeter Collingbourne vm_flags_t vm_flags; 141162b5f7d0SDave Hansen int pkey = 0; 14121da177e4SLinus Torvalds 141341badc15SMichel Lespinasse *populate = 0; 1414bebeb3d6SMichel Lespinasse 1415e37609bbSPiotr Kwapulinski if (!len) 1416e37609bbSPiotr Kwapulinski return -EINVAL; 1417e37609bbSPiotr Kwapulinski 14181da177e4SLinus Torvalds /* 14191da177e4SLinus Torvalds * Does the application expect PROT_READ to imply PROT_EXEC? 14201da177e4SLinus Torvalds * 14211da177e4SLinus Torvalds * (the exception is when the underlying filesystem is noexec 14221da177e4SLinus Torvalds * mounted, in which case we dont add PROT_EXEC.) 14231da177e4SLinus Torvalds */ 14241da177e4SLinus Torvalds if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC)) 142590f8572bSEric W. Biederman if (!(file && path_noexec(&file->f_path))) 14261da177e4SLinus Torvalds prot |= PROT_EXEC; 14271da177e4SLinus Torvalds 1428a4ff8e86SMichal Hocko /* force arch specific MAP_FIXED handling in get_unmapped_area */ 1429a4ff8e86SMichal Hocko if (flags & MAP_FIXED_NOREPLACE) 1430a4ff8e86SMichal Hocko flags |= MAP_FIXED; 1431a4ff8e86SMichal Hocko 14327cd94146SEric Paris if (!(flags & MAP_FIXED)) 14337cd94146SEric Paris addr = round_hint_to_min(addr); 14347cd94146SEric Paris 14351da177e4SLinus Torvalds /* Careful about overflows.. */ 14361da177e4SLinus Torvalds len = PAGE_ALIGN(len); 14379206de95SAl Viro if (!len) 14381da177e4SLinus Torvalds return -ENOMEM; 14391da177e4SLinus Torvalds 14401da177e4SLinus Torvalds /* offset overflow? */ 14411da177e4SLinus Torvalds if ((pgoff + (len >> PAGE_SHIFT)) < pgoff) 14421da177e4SLinus Torvalds return -EOVERFLOW; 14431da177e4SLinus Torvalds 14441da177e4SLinus Torvalds /* Too many mappings? */ 14451da177e4SLinus Torvalds if (mm->map_count > sysctl_max_map_count) 14461da177e4SLinus Torvalds return -ENOMEM; 14471da177e4SLinus Torvalds 14481da177e4SLinus Torvalds /* Obtain the address to map to. we verify (or select) it and ensure 14491da177e4SLinus Torvalds * that it represents a valid section of the address space. 14501da177e4SLinus Torvalds */ 14511da177e4SLinus Torvalds addr = get_unmapped_area(file, addr, len, pgoff, flags); 1452ff68dac6SGaowei Pu if (IS_ERR_VALUE(addr)) 14531da177e4SLinus Torvalds return addr; 14541da177e4SLinus Torvalds 1455a4ff8e86SMichal Hocko if (flags & MAP_FIXED_NOREPLACE) { 1456a4ff8e86SMichal Hocko struct vm_area_struct *vma = find_vma(mm, addr); 1457a4ff8e86SMichal Hocko 14587aa867ddSJann Horn if (vma && vma->vm_start < addr + len) 1459a4ff8e86SMichal Hocko return -EEXIST; 1460a4ff8e86SMichal Hocko } 1461a4ff8e86SMichal Hocko 146262b5f7d0SDave Hansen if (prot == PROT_EXEC) { 146362b5f7d0SDave Hansen pkey = execute_only_pkey(mm); 146462b5f7d0SDave Hansen if (pkey < 0) 146562b5f7d0SDave Hansen pkey = 0; 146662b5f7d0SDave Hansen } 146762b5f7d0SDave Hansen 14681da177e4SLinus Torvalds /* Do simple checking here so the lower-level routines won't have 14691da177e4SLinus Torvalds * to. we assume access permissions have been handled by the open 14701da177e4SLinus Torvalds * of the memory object, so we don't do any here. 14711da177e4SLinus Torvalds */ 147245e55300SPeter Collingbourne vm_flags = calc_vm_prot_bits(prot, pkey) | calc_vm_flag_bits(flags) | 14731da177e4SLinus Torvalds mm->def_flags | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC; 14741da177e4SLinus Torvalds 1475cdf7b341SHuang Shijie if (flags & MAP_LOCKED) 14761da177e4SLinus Torvalds if (!can_do_mlock()) 14771da177e4SLinus Torvalds return -EPERM; 1478ba470de4SRik van Riel 1479363ee17fSDavidlohr Bueso if (mlock_future_check(mm, vm_flags, len)) 14801da177e4SLinus Torvalds return -EAGAIN; 14811da177e4SLinus Torvalds 14821da177e4SLinus Torvalds if (file) { 1483077bf22bSOleg Nesterov struct inode *inode = file_inode(file); 14841c972597SDan Williams unsigned long flags_mask; 14851c972597SDan Williams 1486be83bbf8SLinus Torvalds if (!file_mmap_ok(file, inode, pgoff, len)) 1487be83bbf8SLinus Torvalds return -EOVERFLOW; 1488be83bbf8SLinus Torvalds 14891c972597SDan Williams flags_mask = LEGACY_MAP_MASK | file->f_op->mmap_supported_flags; 1490077bf22bSOleg Nesterov 14911da177e4SLinus Torvalds switch (flags & MAP_TYPE) { 14921da177e4SLinus Torvalds case MAP_SHARED: 14931c972597SDan Williams /* 14941c972597SDan Williams * Force use of MAP_SHARED_VALIDATE with non-legacy 14951c972597SDan Williams * flags. E.g. MAP_SYNC is dangerous to use with 14961c972597SDan Williams * MAP_SHARED as you don't know which consistency model 14971c972597SDan Williams * you will get. We silently ignore unsupported flags 14981c972597SDan Williams * with MAP_SHARED to preserve backward compatibility. 14991c972597SDan Williams */ 15001c972597SDan Williams flags &= LEGACY_MAP_MASK; 1501e4a9bc58SJoe Perches fallthrough; 15021c972597SDan Williams case MAP_SHARED_VALIDATE: 15031c972597SDan Williams if (flags & ~flags_mask) 15041c972597SDan Williams return -EOPNOTSUPP; 1505dc617f29SDarrick J. Wong if (prot & PROT_WRITE) { 1506dc617f29SDarrick J. Wong if (!(file->f_mode & FMODE_WRITE)) 15071da177e4SLinus Torvalds return -EACCES; 1508dc617f29SDarrick J. Wong if (IS_SWAPFILE(file->f_mapping->host)) 1509dc617f29SDarrick J. Wong return -ETXTBSY; 1510dc617f29SDarrick J. Wong } 15111da177e4SLinus Torvalds 15121da177e4SLinus Torvalds /* 15131da177e4SLinus Torvalds * Make sure we don't allow writing to an append-only 15141da177e4SLinus Torvalds * file.. 15151da177e4SLinus Torvalds */ 15161da177e4SLinus Torvalds if (IS_APPEND(inode) && (file->f_mode & FMODE_WRITE)) 15171da177e4SLinus Torvalds return -EACCES; 15181da177e4SLinus Torvalds 15191da177e4SLinus Torvalds /* 15201da177e4SLinus Torvalds * Make sure there are no mandatory locks on the file. 15211da177e4SLinus Torvalds */ 1522d7a06983SJeff Layton if (locks_verify_locked(file)) 15231da177e4SLinus Torvalds return -EAGAIN; 15241da177e4SLinus Torvalds 15251da177e4SLinus Torvalds vm_flags |= VM_SHARED | VM_MAYSHARE; 15261da177e4SLinus Torvalds if (!(file->f_mode & FMODE_WRITE)) 15271da177e4SLinus Torvalds vm_flags &= ~(VM_MAYWRITE | VM_SHARED); 1528e4a9bc58SJoe Perches fallthrough; 15291da177e4SLinus Torvalds case MAP_PRIVATE: 15301da177e4SLinus Torvalds if (!(file->f_mode & FMODE_READ)) 15311da177e4SLinus Torvalds return -EACCES; 153290f8572bSEric W. Biederman if (path_noexec(&file->f_path)) { 153380c5606cSLinus Torvalds if (vm_flags & VM_EXEC) 153480c5606cSLinus Torvalds return -EPERM; 153580c5606cSLinus Torvalds vm_flags &= ~VM_MAYEXEC; 153680c5606cSLinus Torvalds } 153780c5606cSLinus Torvalds 153872c2d531SAl Viro if (!file->f_op->mmap) 153980c5606cSLinus Torvalds return -ENODEV; 1540b2c56e4fSOleg Nesterov if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP)) 1541b2c56e4fSOleg Nesterov return -EINVAL; 15421da177e4SLinus Torvalds break; 15431da177e4SLinus Torvalds 15441da177e4SLinus Torvalds default: 15451da177e4SLinus Torvalds return -EINVAL; 15461da177e4SLinus Torvalds } 15471da177e4SLinus Torvalds } else { 15481da177e4SLinus Torvalds switch (flags & MAP_TYPE) { 15491da177e4SLinus Torvalds case MAP_SHARED: 1550b2c56e4fSOleg Nesterov if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP)) 1551b2c56e4fSOleg Nesterov return -EINVAL; 1552ce363942STejun Heo /* 1553ce363942STejun Heo * Ignore pgoff. 1554ce363942STejun Heo */ 1555ce363942STejun Heo pgoff = 0; 15561da177e4SLinus Torvalds vm_flags |= VM_SHARED | VM_MAYSHARE; 15571da177e4SLinus Torvalds break; 15581da177e4SLinus Torvalds case MAP_PRIVATE: 15591da177e4SLinus Torvalds /* 15601da177e4SLinus Torvalds * Set pgoff according to addr for anon_vma. 15611da177e4SLinus Torvalds */ 15621da177e4SLinus Torvalds pgoff = addr >> PAGE_SHIFT; 15631da177e4SLinus Torvalds break; 15641da177e4SLinus Torvalds default: 15651da177e4SLinus Torvalds return -EINVAL; 15661da177e4SLinus Torvalds } 15671da177e4SLinus Torvalds } 15681da177e4SLinus Torvalds 1569c22c0d63SMichel Lespinasse /* 1570c22c0d63SMichel Lespinasse * Set 'VM_NORESERVE' if we should not account for the 1571c22c0d63SMichel Lespinasse * memory use of this mapping. 1572c22c0d63SMichel Lespinasse */ 1573c22c0d63SMichel Lespinasse if (flags & MAP_NORESERVE) { 1574c22c0d63SMichel Lespinasse /* We honor MAP_NORESERVE if allowed to overcommit */ 1575c22c0d63SMichel Lespinasse if (sysctl_overcommit_memory != OVERCOMMIT_NEVER) 1576c22c0d63SMichel Lespinasse vm_flags |= VM_NORESERVE; 1577c22c0d63SMichel Lespinasse 1578c22c0d63SMichel Lespinasse /* hugetlb applies strict overcommit unless MAP_NORESERVE */ 1579c22c0d63SMichel Lespinasse if (file && is_file_hugepages(file)) 1580c22c0d63SMichel Lespinasse vm_flags |= VM_NORESERVE; 1581c22c0d63SMichel Lespinasse } 1582c22c0d63SMichel Lespinasse 1583897ab3e0SMike Rapoport addr = mmap_region(file, addr, len, vm_flags, pgoff, uf); 158409a9f1d2SMichel Lespinasse if (!IS_ERR_VALUE(addr) && 158509a9f1d2SMichel Lespinasse ((vm_flags & VM_LOCKED) || 158609a9f1d2SMichel Lespinasse (flags & (MAP_POPULATE | MAP_NONBLOCK)) == MAP_POPULATE)) 158741badc15SMichel Lespinasse *populate = len; 1588bebeb3d6SMichel Lespinasse return addr; 15890165ab44SMiklos Szeredi } 15906be5ceb0SLinus Torvalds 1591a90f590aSDominik Brodowski unsigned long ksys_mmap_pgoff(unsigned long addr, unsigned long len, 1592a90f590aSDominik Brodowski unsigned long prot, unsigned long flags, 1593a90f590aSDominik Brodowski unsigned long fd, unsigned long pgoff) 159466f0dc48SHugh Dickins { 159566f0dc48SHugh Dickins struct file *file = NULL; 15961e3ee14bSChen Gang unsigned long retval; 159766f0dc48SHugh Dickins 159866f0dc48SHugh Dickins if (!(flags & MAP_ANONYMOUS)) { 1599120a795dSAl Viro audit_mmap_fd(fd, flags); 160066f0dc48SHugh Dickins file = fget(fd); 160166f0dc48SHugh Dickins if (!file) 16021e3ee14bSChen Gang return -EBADF; 16037bba8f0eSZhen Lei if (is_file_hugepages(file)) { 1604af73e4d9SNaoya Horiguchi len = ALIGN(len, huge_page_size(hstate_file(file))); 16057bba8f0eSZhen Lei } else if (unlikely(flags & MAP_HUGETLB)) { 1606493af578SJörn Engel retval = -EINVAL; 1607493af578SJörn Engel goto out_fput; 16087bba8f0eSZhen Lei } 160966f0dc48SHugh Dickins } else if (flags & MAP_HUGETLB) { 161066f0dc48SHugh Dickins struct user_struct *user = NULL; 1611c103a4dcSAndrew Morton struct hstate *hs; 1612af73e4d9SNaoya Horiguchi 161320ac2893SAnshuman Khandual hs = hstate_sizelog((flags >> MAP_HUGE_SHIFT) & MAP_HUGE_MASK); 1614091d0d55SLi Zefan if (!hs) 1615091d0d55SLi Zefan return -EINVAL; 1616091d0d55SLi Zefan 1617091d0d55SLi Zefan len = ALIGN(len, huge_page_size(hs)); 161866f0dc48SHugh Dickins /* 161966f0dc48SHugh Dickins * VM_NORESERVE is used because the reservations will be 162066f0dc48SHugh Dickins * taken when vm_ops->mmap() is called 162166f0dc48SHugh Dickins * A dummy user value is used because we are not locking 162266f0dc48SHugh Dickins * memory so no accounting is necessary 162366f0dc48SHugh Dickins */ 1624af73e4d9SNaoya Horiguchi file = hugetlb_file_setup(HUGETLB_ANON_FILE, len, 162542d7395fSAndi Kleen VM_NORESERVE, 162642d7395fSAndi Kleen &user, HUGETLB_ANONHUGE_INODE, 162742d7395fSAndi Kleen (flags >> MAP_HUGE_SHIFT) & MAP_HUGE_MASK); 162866f0dc48SHugh Dickins if (IS_ERR(file)) 162966f0dc48SHugh Dickins return PTR_ERR(file); 163066f0dc48SHugh Dickins } 163166f0dc48SHugh Dickins 163266f0dc48SHugh Dickins flags &= ~(MAP_EXECUTABLE | MAP_DENYWRITE); 163366f0dc48SHugh Dickins 16349fbeb5abSMichal Hocko retval = vm_mmap_pgoff(file, addr, len, prot, flags, pgoff); 1635493af578SJörn Engel out_fput: 163666f0dc48SHugh Dickins if (file) 163766f0dc48SHugh Dickins fput(file); 163866f0dc48SHugh Dickins return retval; 163966f0dc48SHugh Dickins } 164066f0dc48SHugh Dickins 1641a90f590aSDominik Brodowski SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len, 1642a90f590aSDominik Brodowski unsigned long, prot, unsigned long, flags, 1643a90f590aSDominik Brodowski unsigned long, fd, unsigned long, pgoff) 1644a90f590aSDominik Brodowski { 1645a90f590aSDominik Brodowski return ksys_mmap_pgoff(addr, len, prot, flags, fd, pgoff); 1646a90f590aSDominik Brodowski } 1647a90f590aSDominik Brodowski 1648a4679373SChristoph Hellwig #ifdef __ARCH_WANT_SYS_OLD_MMAP 1649a4679373SChristoph Hellwig struct mmap_arg_struct { 1650a4679373SChristoph Hellwig unsigned long addr; 1651a4679373SChristoph Hellwig unsigned long len; 1652a4679373SChristoph Hellwig unsigned long prot; 1653a4679373SChristoph Hellwig unsigned long flags; 1654a4679373SChristoph Hellwig unsigned long fd; 1655a4679373SChristoph Hellwig unsigned long offset; 1656a4679373SChristoph Hellwig }; 1657a4679373SChristoph Hellwig 1658a4679373SChristoph Hellwig SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg) 1659a4679373SChristoph Hellwig { 1660a4679373SChristoph Hellwig struct mmap_arg_struct a; 1661a4679373SChristoph Hellwig 1662a4679373SChristoph Hellwig if (copy_from_user(&a, arg, sizeof(a))) 1663a4679373SChristoph Hellwig return -EFAULT; 1664de1741a1SAlexander Kuleshov if (offset_in_page(a.offset)) 1665a4679373SChristoph Hellwig return -EINVAL; 1666a4679373SChristoph Hellwig 1667a90f590aSDominik Brodowski return ksys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd, 1668a4679373SChristoph Hellwig a.offset >> PAGE_SHIFT); 1669a4679373SChristoph Hellwig } 1670a4679373SChristoph Hellwig #endif /* __ARCH_WANT_SYS_OLD_MMAP */ 1671a4679373SChristoph Hellwig 16724e950f6fSAlexey Dobriyan /* 16738bb4e7a2SWei Yang * Some shared mappings will want the pages marked read-only 16744e950f6fSAlexey Dobriyan * to track write events. If so, we'll downgrade vm_page_prot 16754e950f6fSAlexey Dobriyan * to the private version (using protection_map[] without the 16764e950f6fSAlexey Dobriyan * VM_SHARED bit). 16774e950f6fSAlexey Dobriyan */ 16786d2329f8SAndrea Arcangeli int vma_wants_writenotify(struct vm_area_struct *vma, pgprot_t vm_page_prot) 16794e950f6fSAlexey Dobriyan { 1680ca16d140SKOSAKI Motohiro vm_flags_t vm_flags = vma->vm_flags; 16818a04446aSKirill A. Shutemov const struct vm_operations_struct *vm_ops = vma->vm_ops; 16824e950f6fSAlexey Dobriyan 16834e950f6fSAlexey Dobriyan /* If it was private or non-writable, the write bit is already clear */ 16844e950f6fSAlexey Dobriyan if ((vm_flags & (VM_WRITE|VM_SHARED)) != ((VM_WRITE|VM_SHARED))) 16854e950f6fSAlexey Dobriyan return 0; 16864e950f6fSAlexey Dobriyan 16874e950f6fSAlexey Dobriyan /* The backer wishes to know when pages are first written to? */ 16888a04446aSKirill A. Shutemov if (vm_ops && (vm_ops->page_mkwrite || vm_ops->pfn_mkwrite)) 16894e950f6fSAlexey Dobriyan return 1; 16904e950f6fSAlexey Dobriyan 169164e45507SPeter Feiner /* The open routine did something to the protections that pgprot_modify 169264e45507SPeter Feiner * won't preserve? */ 16936d2329f8SAndrea Arcangeli if (pgprot_val(vm_page_prot) != 16946d2329f8SAndrea Arcangeli pgprot_val(vm_pgprot_modify(vm_page_prot, vm_flags))) 16954e950f6fSAlexey Dobriyan return 0; 16964e950f6fSAlexey Dobriyan 169764e45507SPeter Feiner /* Do we need to track softdirty? */ 169864e45507SPeter Feiner if (IS_ENABLED(CONFIG_MEM_SOFT_DIRTY) && !(vm_flags & VM_SOFTDIRTY)) 169964e45507SPeter Feiner return 1; 170064e45507SPeter Feiner 17014e950f6fSAlexey Dobriyan /* Specialty mapping? */ 17024b6e1e37SKonstantin Khlebnikov if (vm_flags & VM_PFNMAP) 17034e950f6fSAlexey Dobriyan return 0; 17044e950f6fSAlexey Dobriyan 17054e950f6fSAlexey Dobriyan /* Can the mapping track the dirty pages? */ 17064e950f6fSAlexey Dobriyan return vma->vm_file && vma->vm_file->f_mapping && 1707f56753acSChristoph Hellwig mapping_can_writeback(vma->vm_file->f_mapping); 17084e950f6fSAlexey Dobriyan } 17094e950f6fSAlexey Dobriyan 1710fc8744adSLinus Torvalds /* 1711fc8744adSLinus Torvalds * We account for memory if it's a private writeable mapping, 17125a6fe125SMel Gorman * not hugepages and VM_NORESERVE wasn't set. 1713fc8744adSLinus Torvalds */ 1714ca16d140SKOSAKI Motohiro static inline int accountable_mapping(struct file *file, vm_flags_t vm_flags) 1715fc8744adSLinus Torvalds { 17165a6fe125SMel Gorman /* 17175a6fe125SMel Gorman * hugetlb has its own accounting separate from the core VM 17185a6fe125SMel Gorman * VM_HUGETLB may not be set yet so we cannot check for that flag. 17195a6fe125SMel Gorman */ 17205a6fe125SMel Gorman if (file && is_file_hugepages(file)) 17215a6fe125SMel Gorman return 0; 17225a6fe125SMel Gorman 1723fc8744adSLinus Torvalds return (vm_flags & (VM_NORESERVE | VM_SHARED | VM_WRITE)) == VM_WRITE; 1724fc8744adSLinus Torvalds } 1725fc8744adSLinus Torvalds 17260165ab44SMiklos Szeredi unsigned long mmap_region(struct file *file, unsigned long addr, 1727897ab3e0SMike Rapoport unsigned long len, vm_flags_t vm_flags, unsigned long pgoff, 1728897ab3e0SMike Rapoport struct list_head *uf) 17290165ab44SMiklos Szeredi { 17300165ab44SMiklos Szeredi struct mm_struct *mm = current->mm; 1731d70cec89SMiaohe Lin struct vm_area_struct *vma, *prev, *merge; 17320165ab44SMiklos Szeredi int error; 17330165ab44SMiklos Szeredi struct rb_node **rb_link, *rb_parent; 17340165ab44SMiklos Szeredi unsigned long charged = 0; 17350165ab44SMiklos Szeredi 1736e8420a8eSCyril Hrubis /* Check against address space limit. */ 173784638335SKonstantin Khlebnikov if (!may_expand_vm(mm, vm_flags, len >> PAGE_SHIFT)) { 1738e8420a8eSCyril Hrubis unsigned long nr_pages; 1739e8420a8eSCyril Hrubis 1740e8420a8eSCyril Hrubis /* 1741e8420a8eSCyril Hrubis * MAP_FIXED may remove pages of mappings that intersects with 1742e8420a8eSCyril Hrubis * requested mapping. Account for the pages it would unmap. 1743e8420a8eSCyril Hrubis */ 1744e8420a8eSCyril Hrubis nr_pages = count_vma_pages_range(mm, addr, addr + len); 1745e8420a8eSCyril Hrubis 174684638335SKonstantin Khlebnikov if (!may_expand_vm(mm, vm_flags, 174784638335SKonstantin Khlebnikov (len >> PAGE_SHIFT) - nr_pages)) 1748e8420a8eSCyril Hrubis return -ENOMEM; 1749e8420a8eSCyril Hrubis } 1750e8420a8eSCyril Hrubis 1751fb8090b6SLiam R. Howlett /* Clear old maps, set up prev, rb_link, rb_parent, and uf */ 1752fb8090b6SLiam R. Howlett if (munmap_vma_range(mm, addr, len, &prev, &rb_link, &rb_parent, uf)) 17531da177e4SLinus Torvalds return -ENOMEM; 1754fc8744adSLinus Torvalds /* 17551da177e4SLinus Torvalds * Private writable mapping: check memory availability 17561da177e4SLinus Torvalds */ 17575a6fe125SMel Gorman if (accountable_mapping(file, vm_flags)) { 17581da177e4SLinus Torvalds charged = len >> PAGE_SHIFT; 1759191c5424SAl Viro if (security_vm_enough_memory_mm(mm, charged)) 17601da177e4SLinus Torvalds return -ENOMEM; 17611da177e4SLinus Torvalds vm_flags |= VM_ACCOUNT; 17621da177e4SLinus Torvalds } 17631da177e4SLinus Torvalds 17641da177e4SLinus Torvalds /* 1765de33c8dbSLinus Torvalds * Can we just expand an old mapping? 17661da177e4SLinus Torvalds */ 176719a809afSAndrea Arcangeli vma = vma_merge(mm, prev, addr, addr + len, vm_flags, 176819a809afSAndrea Arcangeli NULL, file, pgoff, NULL, NULL_VM_UFFD_CTX); 1769ba470de4SRik van Riel if (vma) 17701da177e4SLinus Torvalds goto out; 17711da177e4SLinus Torvalds 17721da177e4SLinus Torvalds /* 17731da177e4SLinus Torvalds * Determine the object being mapped and call the appropriate 17741da177e4SLinus Torvalds * specific mapper. the address has already been validated, but 17751da177e4SLinus Torvalds * not unmapped, but the maps are removed from the list. 17761da177e4SLinus Torvalds */ 1777490fc053SLinus Torvalds vma = vm_area_alloc(mm); 17781da177e4SLinus Torvalds if (!vma) { 17791da177e4SLinus Torvalds error = -ENOMEM; 17801da177e4SLinus Torvalds goto unacct_error; 17811da177e4SLinus Torvalds } 17821da177e4SLinus Torvalds 17831da177e4SLinus Torvalds vma->vm_start = addr; 17841da177e4SLinus Torvalds vma->vm_end = addr + len; 17851da177e4SLinus Torvalds vma->vm_flags = vm_flags; 17863ed75eb8SColy Li vma->vm_page_prot = vm_get_page_prot(vm_flags); 17871da177e4SLinus Torvalds vma->vm_pgoff = pgoff; 17881da177e4SLinus Torvalds 17891da177e4SLinus Torvalds if (file) { 17901da177e4SLinus Torvalds if (vm_flags & VM_DENYWRITE) { 17911da177e4SLinus Torvalds error = deny_write_access(file); 17921da177e4SLinus Torvalds if (error) 17931da177e4SLinus Torvalds goto free_vma; 17941da177e4SLinus Torvalds } 17954bb5f5d9SDavid Herrmann if (vm_flags & VM_SHARED) { 17964bb5f5d9SDavid Herrmann error = mapping_map_writable(file->f_mapping); 17974bb5f5d9SDavid Herrmann if (error) 17984bb5f5d9SDavid Herrmann goto allow_write_and_free_vma; 17994bb5f5d9SDavid Herrmann } 18004bb5f5d9SDavid Herrmann 18014bb5f5d9SDavid Herrmann /* ->mmap() can change vma->vm_file, but must guarantee that 18024bb5f5d9SDavid Herrmann * vma_link() below can deny write-access if VM_DENYWRITE is set 18034bb5f5d9SDavid Herrmann * and map writably if VM_SHARED is set. This usually means the 18044bb5f5d9SDavid Herrmann * new file must not have been exposed to user-space, yet. 18054bb5f5d9SDavid Herrmann */ 1806cb0942b8SAl Viro vma->vm_file = get_file(file); 1807f74ac015SMiklos Szeredi error = call_mmap(file, vma); 18081da177e4SLinus Torvalds if (error) 18091da177e4SLinus Torvalds goto unmap_and_free_vma; 18101da177e4SLinus Torvalds 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, 1816d70cec89SMiaohe Lin NULL, vma->vm_file, vma->vm_pgoff, NULL, NULL_VM_UFFD_CTX); 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; 1825d70cec89SMiaohe Lin /* Update vm_flags and possible addr to pick up the change. We don't 1826d70cec89SMiaohe Lin * warn here if addr changed as the vma is not linked by vma_link(). 1827d70cec89SMiaohe Lin */ 1828d70cec89SMiaohe Lin addr = vma->vm_start; 1829d70cec89SMiaohe Lin vm_flags = vma->vm_flags; 1830d70cec89SMiaohe Lin goto unmap_writable; 1831d70cec89SMiaohe Lin } 1832d70cec89SMiaohe Lin } 1833d70cec89SMiaohe Lin 18341da177e4SLinus Torvalds /* Can addr have changed?? 18351da177e4SLinus Torvalds * 18361da177e4SLinus Torvalds * Answer: Yes, several device drivers can do it in their 18371da177e4SLinus Torvalds * f_op->mmap method. -DaveM 18382897b4d2SJoonsoo Kim * Bug: If addr is changed, prev, rb_link, rb_parent should 18392897b4d2SJoonsoo Kim * be updated for vma_link() 18401da177e4SLinus Torvalds */ 18412897b4d2SJoonsoo Kim WARN_ON_ONCE(addr != vma->vm_start); 18422897b4d2SJoonsoo Kim 18431da177e4SLinus Torvalds addr = vma->vm_start; 18441da177e4SLinus Torvalds vm_flags = vma->vm_flags; 1845f8dbf0a7SHuang Shijie } else if (vm_flags & VM_SHARED) { 1846f8dbf0a7SHuang Shijie error = shmem_zero_setup(vma); 1847f8dbf0a7SHuang Shijie if (error) 1848f8dbf0a7SHuang Shijie goto free_vma; 1849bfd40eafSKirill A. Shutemov } else { 1850bfd40eafSKirill A. Shutemov vma_set_anonymous(vma); 1851f8dbf0a7SHuang Shijie } 18521da177e4SLinus Torvalds 1853c462ac28SCatalin Marinas /* Allow architectures to sanity-check the vm_flags */ 1854c462ac28SCatalin Marinas if (!arch_validate_flags(vma->vm_flags)) { 1855c462ac28SCatalin Marinas error = -EINVAL; 1856c462ac28SCatalin Marinas if (file) 1857c462ac28SCatalin Marinas goto unmap_and_free_vma; 1858c462ac28SCatalin Marinas else 1859c462ac28SCatalin Marinas goto free_vma; 1860c462ac28SCatalin Marinas } 1861c462ac28SCatalin Marinas 18624d3d5b41SOleg Nesterov vma_link(mm, vma, prev, rb_link, rb_parent); 18634d3d5b41SOleg Nesterov /* Once vma denies write, undo our temporary denial count */ 18644bb5f5d9SDavid Herrmann if (file) { 1865d70cec89SMiaohe Lin unmap_writable: 18664bb5f5d9SDavid Herrmann if (vm_flags & VM_SHARED) 18674bb5f5d9SDavid Herrmann mapping_unmap_writable(file->f_mapping); 1868e8686772SOleg Nesterov if (vm_flags & VM_DENYWRITE) 1869e8686772SOleg Nesterov allow_write_access(file); 18704bb5f5d9SDavid Herrmann } 1871e8686772SOleg Nesterov file = vma->vm_file; 18721da177e4SLinus Torvalds out: 1873cdd6c482SIngo Molnar perf_event_mmap(vma); 18740a4a9391SPeter Zijlstra 187584638335SKonstantin Khlebnikov vm_stat_account(mm, vm_flags, len >> PAGE_SHIFT); 18761da177e4SLinus Torvalds if (vm_flags & VM_LOCKED) { 1877e1fb4a08SDave Jiang if ((vm_flags & VM_SPECIAL) || vma_is_dax(vma) || 1878e1fb4a08SDave Jiang is_vm_hugetlb_page(vma) || 1879e1fb4a08SDave Jiang vma == get_gate_vma(current->mm)) 1880de60f5f1SEric B Munson vma->vm_flags &= VM_LOCKED_CLEAR_MASK; 1881e1fb4a08SDave Jiang else 1882e1fb4a08SDave Jiang mm->locked_vm += (len >> PAGE_SHIFT); 1883bebeb3d6SMichel Lespinasse } 18842b144498SSrikar Dronamraju 1885c7a3a88cSOleg Nesterov if (file) 1886c7a3a88cSOleg Nesterov uprobe_mmap(vma); 18872b144498SSrikar Dronamraju 1888d9104d1cSCyrill Gorcunov /* 1889d9104d1cSCyrill Gorcunov * New (or expanded) vma always get soft dirty status. 1890d9104d1cSCyrill Gorcunov * Otherwise user-space soft-dirty page tracker won't 1891d9104d1cSCyrill Gorcunov * be able to distinguish situation when vma area unmapped, 1892d9104d1cSCyrill Gorcunov * then new mapped in-place (which must be aimed as 1893d9104d1cSCyrill Gorcunov * a completely new data area). 1894d9104d1cSCyrill Gorcunov */ 1895d9104d1cSCyrill Gorcunov vma->vm_flags |= VM_SOFTDIRTY; 1896d9104d1cSCyrill Gorcunov 189764e45507SPeter Feiner vma_set_page_prot(vma); 189864e45507SPeter Feiner 18991da177e4SLinus Torvalds return addr; 19001da177e4SLinus Torvalds 19011da177e4SLinus Torvalds unmap_and_free_vma: 19021da177e4SLinus Torvalds vma->vm_file = NULL; 19031da177e4SLinus Torvalds fput(file); 19041da177e4SLinus Torvalds 19051da177e4SLinus Torvalds /* Undo any partial mapping done by a device driver. */ 1906e0da382cSHugh Dickins unmap_region(mm, vma, prev, vma->vm_start, vma->vm_end); 1907e0da382cSHugh Dickins charged = 0; 19084bb5f5d9SDavid Herrmann if (vm_flags & VM_SHARED) 19094bb5f5d9SDavid Herrmann mapping_unmap_writable(file->f_mapping); 19104bb5f5d9SDavid Herrmann allow_write_and_free_vma: 19114bb5f5d9SDavid Herrmann if (vm_flags & VM_DENYWRITE) 19124bb5f5d9SDavid Herrmann allow_write_access(file); 19131da177e4SLinus Torvalds free_vma: 19143928d4f5SLinus Torvalds vm_area_free(vma); 19151da177e4SLinus Torvalds unacct_error: 19161da177e4SLinus Torvalds if (charged) 19171da177e4SLinus Torvalds vm_unacct_memory(charged); 19181da177e4SLinus Torvalds return error; 19191da177e4SLinus Torvalds } 19201da177e4SLinus Torvalds 1921baceaf1cSJaewon Kim static unsigned long unmapped_area(struct vm_unmapped_area_info *info) 1922db4fbfb9SMichel Lespinasse { 1923db4fbfb9SMichel Lespinasse /* 1924db4fbfb9SMichel Lespinasse * We implement the search by looking for an rbtree node that 1925db4fbfb9SMichel Lespinasse * immediately follows a suitable gap. That is, 1926db4fbfb9SMichel Lespinasse * - gap_start = vma->vm_prev->vm_end <= info->high_limit - length; 1927db4fbfb9SMichel Lespinasse * - gap_end = vma->vm_start >= info->low_limit + length; 1928db4fbfb9SMichel Lespinasse * - gap_end - gap_start >= length 1929db4fbfb9SMichel Lespinasse */ 1930db4fbfb9SMichel Lespinasse 1931db4fbfb9SMichel Lespinasse struct mm_struct *mm = current->mm; 1932db4fbfb9SMichel Lespinasse struct vm_area_struct *vma; 1933db4fbfb9SMichel Lespinasse unsigned long length, low_limit, high_limit, gap_start, gap_end; 1934db4fbfb9SMichel Lespinasse 1935db4fbfb9SMichel Lespinasse /* Adjust search length to account for worst case alignment overhead */ 1936db4fbfb9SMichel Lespinasse length = info->length + info->align_mask; 1937db4fbfb9SMichel Lespinasse if (length < info->length) 1938db4fbfb9SMichel Lespinasse return -ENOMEM; 1939db4fbfb9SMichel Lespinasse 1940db4fbfb9SMichel Lespinasse /* Adjust search limits by the desired length */ 1941db4fbfb9SMichel Lespinasse if (info->high_limit < length) 1942db4fbfb9SMichel Lespinasse return -ENOMEM; 1943db4fbfb9SMichel Lespinasse high_limit = info->high_limit - length; 1944db4fbfb9SMichel Lespinasse 1945db4fbfb9SMichel Lespinasse if (info->low_limit > high_limit) 1946db4fbfb9SMichel Lespinasse return -ENOMEM; 1947db4fbfb9SMichel Lespinasse low_limit = info->low_limit + length; 1948db4fbfb9SMichel Lespinasse 1949db4fbfb9SMichel Lespinasse /* Check if rbtree root looks promising */ 1950db4fbfb9SMichel Lespinasse if (RB_EMPTY_ROOT(&mm->mm_rb)) 1951db4fbfb9SMichel Lespinasse goto check_highest; 1952db4fbfb9SMichel Lespinasse vma = rb_entry(mm->mm_rb.rb_node, struct vm_area_struct, vm_rb); 1953db4fbfb9SMichel Lespinasse if (vma->rb_subtree_gap < length) 1954db4fbfb9SMichel Lespinasse goto check_highest; 1955db4fbfb9SMichel Lespinasse 1956db4fbfb9SMichel Lespinasse while (true) { 1957db4fbfb9SMichel Lespinasse /* Visit left subtree if it looks promising */ 19581be7107fSHugh Dickins gap_end = vm_start_gap(vma); 1959db4fbfb9SMichel Lespinasse if (gap_end >= low_limit && vma->vm_rb.rb_left) { 1960db4fbfb9SMichel Lespinasse struct vm_area_struct *left = 1961db4fbfb9SMichel Lespinasse rb_entry(vma->vm_rb.rb_left, 1962db4fbfb9SMichel Lespinasse struct vm_area_struct, vm_rb); 1963db4fbfb9SMichel Lespinasse if (left->rb_subtree_gap >= length) { 1964db4fbfb9SMichel Lespinasse vma = left; 1965db4fbfb9SMichel Lespinasse continue; 1966db4fbfb9SMichel Lespinasse } 1967db4fbfb9SMichel Lespinasse } 1968db4fbfb9SMichel Lespinasse 19691be7107fSHugh Dickins gap_start = vma->vm_prev ? vm_end_gap(vma->vm_prev) : 0; 1970db4fbfb9SMichel Lespinasse check_current: 1971db4fbfb9SMichel Lespinasse /* Check if current node has a suitable gap */ 1972db4fbfb9SMichel Lespinasse if (gap_start > high_limit) 1973db4fbfb9SMichel Lespinasse return -ENOMEM; 1974f4cb767dSHugh Dickins if (gap_end >= low_limit && 1975f4cb767dSHugh Dickins gap_end > gap_start && gap_end - gap_start >= length) 1976db4fbfb9SMichel Lespinasse goto found; 1977db4fbfb9SMichel Lespinasse 1978db4fbfb9SMichel Lespinasse /* Visit right subtree if it looks promising */ 1979db4fbfb9SMichel Lespinasse if (vma->vm_rb.rb_right) { 1980db4fbfb9SMichel Lespinasse struct vm_area_struct *right = 1981db4fbfb9SMichel Lespinasse rb_entry(vma->vm_rb.rb_right, 1982db4fbfb9SMichel Lespinasse struct vm_area_struct, vm_rb); 1983db4fbfb9SMichel Lespinasse if (right->rb_subtree_gap >= length) { 1984db4fbfb9SMichel Lespinasse vma = right; 1985db4fbfb9SMichel Lespinasse continue; 1986db4fbfb9SMichel Lespinasse } 1987db4fbfb9SMichel Lespinasse } 1988db4fbfb9SMichel Lespinasse 1989db4fbfb9SMichel Lespinasse /* Go back up the rbtree to find next candidate node */ 1990db4fbfb9SMichel Lespinasse while (true) { 1991db4fbfb9SMichel Lespinasse struct rb_node *prev = &vma->vm_rb; 1992db4fbfb9SMichel Lespinasse if (!rb_parent(prev)) 1993db4fbfb9SMichel Lespinasse goto check_highest; 1994db4fbfb9SMichel Lespinasse vma = rb_entry(rb_parent(prev), 1995db4fbfb9SMichel Lespinasse struct vm_area_struct, vm_rb); 1996db4fbfb9SMichel Lespinasse if (prev == vma->vm_rb.rb_left) { 19971be7107fSHugh Dickins gap_start = vm_end_gap(vma->vm_prev); 19981be7107fSHugh Dickins gap_end = vm_start_gap(vma); 1999db4fbfb9SMichel Lespinasse goto check_current; 2000db4fbfb9SMichel Lespinasse } 2001db4fbfb9SMichel Lespinasse } 2002db4fbfb9SMichel Lespinasse } 2003db4fbfb9SMichel Lespinasse 2004db4fbfb9SMichel Lespinasse check_highest: 2005db4fbfb9SMichel Lespinasse /* Check highest gap, which does not precede any rbtree node */ 2006db4fbfb9SMichel Lespinasse gap_start = mm->highest_vm_end; 2007db4fbfb9SMichel Lespinasse gap_end = ULONG_MAX; /* Only for VM_BUG_ON below */ 2008db4fbfb9SMichel Lespinasse if (gap_start > high_limit) 2009db4fbfb9SMichel Lespinasse return -ENOMEM; 2010db4fbfb9SMichel Lespinasse 2011db4fbfb9SMichel Lespinasse found: 2012db4fbfb9SMichel Lespinasse /* We found a suitable gap. Clip it with the original low_limit. */ 2013db4fbfb9SMichel Lespinasse if (gap_start < info->low_limit) 2014db4fbfb9SMichel Lespinasse gap_start = info->low_limit; 2015db4fbfb9SMichel Lespinasse 2016db4fbfb9SMichel Lespinasse /* Adjust gap address to the desired alignment */ 2017db4fbfb9SMichel Lespinasse gap_start += (info->align_offset - gap_start) & info->align_mask; 2018db4fbfb9SMichel Lespinasse 2019db4fbfb9SMichel Lespinasse VM_BUG_ON(gap_start + info->length > info->high_limit); 2020db4fbfb9SMichel Lespinasse VM_BUG_ON(gap_start + info->length > gap_end); 2021db4fbfb9SMichel Lespinasse return gap_start; 2022db4fbfb9SMichel Lespinasse } 2023db4fbfb9SMichel Lespinasse 2024baceaf1cSJaewon Kim static unsigned long unmapped_area_topdown(struct vm_unmapped_area_info *info) 2025db4fbfb9SMichel Lespinasse { 2026db4fbfb9SMichel Lespinasse struct mm_struct *mm = current->mm; 2027db4fbfb9SMichel Lespinasse struct vm_area_struct *vma; 2028db4fbfb9SMichel Lespinasse unsigned long length, low_limit, high_limit, gap_start, gap_end; 2029db4fbfb9SMichel Lespinasse 2030db4fbfb9SMichel Lespinasse /* Adjust search length to account for worst case alignment overhead */ 2031db4fbfb9SMichel Lespinasse length = info->length + info->align_mask; 2032db4fbfb9SMichel Lespinasse if (length < info->length) 2033db4fbfb9SMichel Lespinasse return -ENOMEM; 2034db4fbfb9SMichel Lespinasse 2035db4fbfb9SMichel Lespinasse /* 2036db4fbfb9SMichel Lespinasse * Adjust search limits by the desired length. 2037db4fbfb9SMichel Lespinasse * See implementation comment at top of unmapped_area(). 2038db4fbfb9SMichel Lespinasse */ 2039db4fbfb9SMichel Lespinasse gap_end = info->high_limit; 2040db4fbfb9SMichel Lespinasse if (gap_end < length) 2041db4fbfb9SMichel Lespinasse return -ENOMEM; 2042db4fbfb9SMichel Lespinasse high_limit = gap_end - length; 2043db4fbfb9SMichel Lespinasse 2044db4fbfb9SMichel Lespinasse if (info->low_limit > high_limit) 2045db4fbfb9SMichel Lespinasse return -ENOMEM; 2046db4fbfb9SMichel Lespinasse low_limit = info->low_limit + length; 2047db4fbfb9SMichel Lespinasse 2048db4fbfb9SMichel Lespinasse /* Check highest gap, which does not precede any rbtree node */ 2049db4fbfb9SMichel Lespinasse gap_start = mm->highest_vm_end; 2050db4fbfb9SMichel Lespinasse if (gap_start <= high_limit) 2051db4fbfb9SMichel Lespinasse goto found_highest; 2052db4fbfb9SMichel Lespinasse 2053db4fbfb9SMichel Lespinasse /* Check if rbtree root looks promising */ 2054db4fbfb9SMichel Lespinasse if (RB_EMPTY_ROOT(&mm->mm_rb)) 2055db4fbfb9SMichel Lespinasse return -ENOMEM; 2056db4fbfb9SMichel Lespinasse vma = rb_entry(mm->mm_rb.rb_node, struct vm_area_struct, vm_rb); 2057db4fbfb9SMichel Lespinasse if (vma->rb_subtree_gap < length) 2058db4fbfb9SMichel Lespinasse return -ENOMEM; 2059db4fbfb9SMichel Lespinasse 2060db4fbfb9SMichel Lespinasse while (true) { 2061db4fbfb9SMichel Lespinasse /* Visit right subtree if it looks promising */ 20621be7107fSHugh Dickins gap_start = vma->vm_prev ? vm_end_gap(vma->vm_prev) : 0; 2063db4fbfb9SMichel Lespinasse if (gap_start <= high_limit && vma->vm_rb.rb_right) { 2064db4fbfb9SMichel Lespinasse struct vm_area_struct *right = 2065db4fbfb9SMichel Lespinasse rb_entry(vma->vm_rb.rb_right, 2066db4fbfb9SMichel Lespinasse struct vm_area_struct, vm_rb); 2067db4fbfb9SMichel Lespinasse if (right->rb_subtree_gap >= length) { 2068db4fbfb9SMichel Lespinasse vma = right; 2069db4fbfb9SMichel Lespinasse continue; 2070db4fbfb9SMichel Lespinasse } 2071db4fbfb9SMichel Lespinasse } 2072db4fbfb9SMichel Lespinasse 2073db4fbfb9SMichel Lespinasse check_current: 2074db4fbfb9SMichel Lespinasse /* Check if current node has a suitable gap */ 20751be7107fSHugh Dickins gap_end = vm_start_gap(vma); 2076db4fbfb9SMichel Lespinasse if (gap_end < low_limit) 2077db4fbfb9SMichel Lespinasse return -ENOMEM; 2078f4cb767dSHugh Dickins if (gap_start <= high_limit && 2079f4cb767dSHugh Dickins gap_end > gap_start && gap_end - gap_start >= length) 2080db4fbfb9SMichel Lespinasse goto found; 2081db4fbfb9SMichel Lespinasse 2082db4fbfb9SMichel Lespinasse /* Visit left subtree if it looks promising */ 2083db4fbfb9SMichel Lespinasse if (vma->vm_rb.rb_left) { 2084db4fbfb9SMichel Lespinasse struct vm_area_struct *left = 2085db4fbfb9SMichel Lespinasse rb_entry(vma->vm_rb.rb_left, 2086db4fbfb9SMichel Lespinasse struct vm_area_struct, vm_rb); 2087db4fbfb9SMichel Lespinasse if (left->rb_subtree_gap >= length) { 2088db4fbfb9SMichel Lespinasse vma = left; 2089db4fbfb9SMichel Lespinasse continue; 2090db4fbfb9SMichel Lespinasse } 2091db4fbfb9SMichel Lespinasse } 2092db4fbfb9SMichel Lespinasse 2093db4fbfb9SMichel Lespinasse /* Go back up the rbtree to find next candidate node */ 2094db4fbfb9SMichel Lespinasse while (true) { 2095db4fbfb9SMichel Lespinasse struct rb_node *prev = &vma->vm_rb; 2096db4fbfb9SMichel Lespinasse if (!rb_parent(prev)) 2097db4fbfb9SMichel Lespinasse return -ENOMEM; 2098db4fbfb9SMichel Lespinasse vma = rb_entry(rb_parent(prev), 2099db4fbfb9SMichel Lespinasse struct vm_area_struct, vm_rb); 2100db4fbfb9SMichel Lespinasse if (prev == vma->vm_rb.rb_right) { 2101db4fbfb9SMichel Lespinasse gap_start = vma->vm_prev ? 21021be7107fSHugh Dickins vm_end_gap(vma->vm_prev) : 0; 2103db4fbfb9SMichel Lespinasse goto check_current; 2104db4fbfb9SMichel Lespinasse } 2105db4fbfb9SMichel Lespinasse } 2106db4fbfb9SMichel Lespinasse } 2107db4fbfb9SMichel Lespinasse 2108db4fbfb9SMichel Lespinasse found: 2109db4fbfb9SMichel Lespinasse /* We found a suitable gap. Clip it with the original high_limit. */ 2110db4fbfb9SMichel Lespinasse if (gap_end > info->high_limit) 2111db4fbfb9SMichel Lespinasse gap_end = info->high_limit; 2112db4fbfb9SMichel Lespinasse 2113db4fbfb9SMichel Lespinasse found_highest: 2114db4fbfb9SMichel Lespinasse /* Compute highest gap address at the desired alignment */ 2115db4fbfb9SMichel Lespinasse gap_end -= info->length; 2116db4fbfb9SMichel Lespinasse gap_end -= (gap_end - info->align_offset) & info->align_mask; 2117db4fbfb9SMichel Lespinasse 2118db4fbfb9SMichel Lespinasse VM_BUG_ON(gap_end < info->low_limit); 2119db4fbfb9SMichel Lespinasse VM_BUG_ON(gap_end < gap_start); 2120db4fbfb9SMichel Lespinasse return gap_end; 2121db4fbfb9SMichel Lespinasse } 2122db4fbfb9SMichel Lespinasse 2123baceaf1cSJaewon Kim /* 2124baceaf1cSJaewon Kim * Search for an unmapped address range. 2125baceaf1cSJaewon Kim * 2126baceaf1cSJaewon Kim * We are looking for a range that: 2127baceaf1cSJaewon Kim * - does not intersect with any VMA; 2128baceaf1cSJaewon Kim * - is contained within the [low_limit, high_limit) interval; 2129baceaf1cSJaewon Kim * - is at least the desired size. 2130baceaf1cSJaewon Kim * - satisfies (begin_addr & align_mask) == (align_offset & align_mask) 2131baceaf1cSJaewon Kim */ 2132baceaf1cSJaewon Kim unsigned long vm_unmapped_area(struct vm_unmapped_area_info *info) 2133baceaf1cSJaewon Kim { 2134df529cabSJaewon Kim unsigned long addr; 2135df529cabSJaewon Kim 2136baceaf1cSJaewon Kim if (info->flags & VM_UNMAPPED_AREA_TOPDOWN) 2137df529cabSJaewon Kim addr = unmapped_area_topdown(info); 2138baceaf1cSJaewon Kim else 2139df529cabSJaewon Kim addr = unmapped_area(info); 2140df529cabSJaewon Kim 2141df529cabSJaewon Kim trace_vm_unmapped_area(addr, info); 2142df529cabSJaewon Kim return addr; 2143baceaf1cSJaewon Kim } 2144f6795053SSteve Capper 2145f6795053SSteve Capper #ifndef arch_get_mmap_end 2146f6795053SSteve Capper #define arch_get_mmap_end(addr) (TASK_SIZE) 2147f6795053SSteve Capper #endif 2148f6795053SSteve Capper 2149f6795053SSteve Capper #ifndef arch_get_mmap_base 2150f6795053SSteve Capper #define arch_get_mmap_base(addr, base) (base) 2151f6795053SSteve Capper #endif 2152f6795053SSteve Capper 21531da177e4SLinus Torvalds /* Get an address range which is currently unmapped. 21541da177e4SLinus Torvalds * For shmat() with addr=0. 21551da177e4SLinus Torvalds * 21561da177e4SLinus Torvalds * Ugly calling convention alert: 21571da177e4SLinus Torvalds * Return value with the low bits set means error value, 21581da177e4SLinus Torvalds * ie 21591da177e4SLinus Torvalds * if (ret & ~PAGE_MASK) 21601da177e4SLinus Torvalds * error = ret; 21611da177e4SLinus Torvalds * 21621da177e4SLinus Torvalds * This function "knows" that -ENOMEM has the bits set. 21631da177e4SLinus Torvalds */ 21641da177e4SLinus Torvalds #ifndef HAVE_ARCH_UNMAPPED_AREA 21651da177e4SLinus Torvalds unsigned long 21661da177e4SLinus Torvalds arch_get_unmapped_area(struct file *filp, unsigned long addr, 21671da177e4SLinus Torvalds unsigned long len, unsigned long pgoff, unsigned long flags) 21681da177e4SLinus Torvalds { 21691da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 21701be7107fSHugh Dickins struct vm_area_struct *vma, *prev; 2171db4fbfb9SMichel Lespinasse struct vm_unmapped_area_info info; 2172f6795053SSteve Capper const unsigned long mmap_end = arch_get_mmap_end(addr); 21731da177e4SLinus Torvalds 2174f6795053SSteve Capper if (len > mmap_end - mmap_min_addr) 21751da177e4SLinus Torvalds return -ENOMEM; 21761da177e4SLinus Torvalds 217706abdfb4SBenjamin Herrenschmidt if (flags & MAP_FIXED) 217806abdfb4SBenjamin Herrenschmidt return addr; 217906abdfb4SBenjamin Herrenschmidt 21801da177e4SLinus Torvalds if (addr) { 21811da177e4SLinus Torvalds addr = PAGE_ALIGN(addr); 21821be7107fSHugh Dickins vma = find_vma_prev(mm, addr, &prev); 2183f6795053SSteve Capper if (mmap_end - len >= addr && addr >= mmap_min_addr && 21841be7107fSHugh Dickins (!vma || addr + len <= vm_start_gap(vma)) && 21851be7107fSHugh Dickins (!prev || addr >= vm_end_gap(prev))) 21861da177e4SLinus Torvalds return addr; 21871da177e4SLinus Torvalds } 21881da177e4SLinus Torvalds 2189db4fbfb9SMichel Lespinasse info.flags = 0; 2190db4fbfb9SMichel Lespinasse info.length = len; 21914e99b021SHeiko Carstens info.low_limit = mm->mmap_base; 2192f6795053SSteve Capper info.high_limit = mmap_end; 2193db4fbfb9SMichel Lespinasse info.align_mask = 0; 219409ef5283SJaewon Kim info.align_offset = 0; 2195db4fbfb9SMichel Lespinasse return vm_unmapped_area(&info); 21961da177e4SLinus Torvalds } 21971da177e4SLinus Torvalds #endif 21981da177e4SLinus Torvalds 21991da177e4SLinus Torvalds /* 22001da177e4SLinus Torvalds * This mmap-allocator allocates new areas top-down from below the 22011da177e4SLinus Torvalds * stack's low limit (the base): 22021da177e4SLinus Torvalds */ 22031da177e4SLinus Torvalds #ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN 22041da177e4SLinus Torvalds unsigned long 220543cca0b1SYang Fan arch_get_unmapped_area_topdown(struct file *filp, unsigned long addr, 220643cca0b1SYang Fan unsigned long len, unsigned long pgoff, 220743cca0b1SYang Fan unsigned long flags) 22081da177e4SLinus Torvalds { 22091be7107fSHugh Dickins struct vm_area_struct *vma, *prev; 22101da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 2211db4fbfb9SMichel Lespinasse struct vm_unmapped_area_info info; 2212f6795053SSteve Capper const unsigned long mmap_end = arch_get_mmap_end(addr); 22131da177e4SLinus Torvalds 22141da177e4SLinus Torvalds /* requested length too big for entire address space */ 2215f6795053SSteve Capper if (len > mmap_end - mmap_min_addr) 22161da177e4SLinus Torvalds return -ENOMEM; 22171da177e4SLinus Torvalds 221806abdfb4SBenjamin Herrenschmidt if (flags & MAP_FIXED) 221906abdfb4SBenjamin Herrenschmidt return addr; 222006abdfb4SBenjamin Herrenschmidt 22211da177e4SLinus Torvalds /* requesting a specific address */ 22221da177e4SLinus Torvalds if (addr) { 22231da177e4SLinus Torvalds addr = PAGE_ALIGN(addr); 22241be7107fSHugh Dickins vma = find_vma_prev(mm, addr, &prev); 2225f6795053SSteve Capper if (mmap_end - len >= addr && addr >= mmap_min_addr && 22261be7107fSHugh Dickins (!vma || addr + len <= vm_start_gap(vma)) && 22271be7107fSHugh Dickins (!prev || addr >= vm_end_gap(prev))) 22281da177e4SLinus Torvalds return addr; 22291da177e4SLinus Torvalds } 22301da177e4SLinus Torvalds 2231db4fbfb9SMichel Lespinasse info.flags = VM_UNMAPPED_AREA_TOPDOWN; 2232db4fbfb9SMichel Lespinasse info.length = len; 22332afc745fSAkira Takeuchi info.low_limit = max(PAGE_SIZE, mmap_min_addr); 2234f6795053SSteve Capper info.high_limit = arch_get_mmap_base(addr, mm->mmap_base); 2235db4fbfb9SMichel Lespinasse info.align_mask = 0; 223609ef5283SJaewon Kim info.align_offset = 0; 2237db4fbfb9SMichel Lespinasse addr = vm_unmapped_area(&info); 2238b716ad95SXiao Guangrong 22391da177e4SLinus Torvalds /* 22401da177e4SLinus Torvalds * A failed mmap() very likely causes application failure, 22411da177e4SLinus Torvalds * so fall back to the bottom-up function here. This scenario 22421da177e4SLinus Torvalds * can happen with large stack limits and large mmap() 22431da177e4SLinus Torvalds * allocations. 22441da177e4SLinus Torvalds */ 2245de1741a1SAlexander Kuleshov if (offset_in_page(addr)) { 2246db4fbfb9SMichel Lespinasse VM_BUG_ON(addr != -ENOMEM); 2247db4fbfb9SMichel Lespinasse info.flags = 0; 2248db4fbfb9SMichel Lespinasse info.low_limit = TASK_UNMAPPED_BASE; 2249f6795053SSteve Capper info.high_limit = mmap_end; 2250db4fbfb9SMichel Lespinasse addr = vm_unmapped_area(&info); 2251db4fbfb9SMichel Lespinasse } 22521da177e4SLinus Torvalds 22531da177e4SLinus Torvalds return addr; 22541da177e4SLinus Torvalds } 22551da177e4SLinus Torvalds #endif 22561da177e4SLinus Torvalds 22571da177e4SLinus Torvalds unsigned long 22581da177e4SLinus Torvalds get_unmapped_area(struct file *file, unsigned long addr, unsigned long len, 22591da177e4SLinus Torvalds unsigned long pgoff, unsigned long flags) 22601da177e4SLinus Torvalds { 226106abdfb4SBenjamin Herrenschmidt unsigned long (*get_area)(struct file *, unsigned long, 226206abdfb4SBenjamin Herrenschmidt unsigned long, unsigned long, unsigned long); 226307ab67c8SLinus Torvalds 22649206de95SAl Viro unsigned long error = arch_mmap_check(addr, len, flags); 22659206de95SAl Viro if (error) 22669206de95SAl Viro return error; 22679206de95SAl Viro 22689206de95SAl Viro /* Careful about overflows.. */ 22699206de95SAl Viro if (len > TASK_SIZE) 22709206de95SAl Viro return -ENOMEM; 22719206de95SAl Viro 227207ab67c8SLinus Torvalds get_area = current->mm->get_unmapped_area; 2273c01d5b30SHugh Dickins if (file) { 2274c01d5b30SHugh Dickins if (file->f_op->get_unmapped_area) 227507ab67c8SLinus Torvalds get_area = file->f_op->get_unmapped_area; 2276c01d5b30SHugh Dickins } else if (flags & MAP_SHARED) { 2277c01d5b30SHugh Dickins /* 2278c01d5b30SHugh Dickins * mmap_region() will call shmem_zero_setup() to create a file, 2279c01d5b30SHugh Dickins * so use shmem's get_unmapped_area in case it can be huge. 228045e55300SPeter Collingbourne * do_mmap() will clear pgoff, so match alignment. 2281c01d5b30SHugh Dickins */ 2282c01d5b30SHugh Dickins pgoff = 0; 2283c01d5b30SHugh Dickins get_area = shmem_get_unmapped_area; 2284c01d5b30SHugh Dickins } 2285c01d5b30SHugh Dickins 228607ab67c8SLinus Torvalds addr = get_area(file, addr, len, pgoff, flags); 228707ab67c8SLinus Torvalds if (IS_ERR_VALUE(addr)) 228807ab67c8SLinus Torvalds return addr; 228907ab67c8SLinus Torvalds 22901da177e4SLinus Torvalds if (addr > TASK_SIZE - len) 22911da177e4SLinus Torvalds return -ENOMEM; 2292de1741a1SAlexander Kuleshov if (offset_in_page(addr)) 22931da177e4SLinus Torvalds return -EINVAL; 229406abdfb4SBenjamin Herrenschmidt 22959ac4ed4bSAl Viro error = security_mmap_addr(addr); 22969ac4ed4bSAl Viro return error ? error : addr; 22971da177e4SLinus Torvalds } 22981da177e4SLinus Torvalds 22991da177e4SLinus Torvalds EXPORT_SYMBOL(get_unmapped_area); 23001da177e4SLinus Torvalds 23011da177e4SLinus Torvalds /* Look up the first VMA which satisfies addr < vm_end, NULL if none. */ 23021da177e4SLinus Torvalds struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr) 23031da177e4SLinus Torvalds { 2304615d6e87SDavidlohr Bueso struct rb_node *rb_node; 2305615d6e87SDavidlohr Bueso struct vm_area_struct *vma; 23061da177e4SLinus Torvalds 23071da177e4SLinus Torvalds /* Check the cache first. */ 2308615d6e87SDavidlohr Bueso vma = vmacache_find(mm, addr); 2309615d6e87SDavidlohr Bueso if (likely(vma)) 2310615d6e87SDavidlohr Bueso return vma; 23111da177e4SLinus Torvalds 23121da177e4SLinus Torvalds rb_node = mm->mm_rb.rb_node; 23131da177e4SLinus Torvalds 23141da177e4SLinus Torvalds while (rb_node) { 2315615d6e87SDavidlohr Bueso struct vm_area_struct *tmp; 23161da177e4SLinus Torvalds 2317615d6e87SDavidlohr Bueso tmp = rb_entry(rb_node, struct vm_area_struct, vm_rb); 23181da177e4SLinus Torvalds 2319615d6e87SDavidlohr Bueso if (tmp->vm_end > addr) { 2320615d6e87SDavidlohr Bueso vma = tmp; 2321615d6e87SDavidlohr Bueso if (tmp->vm_start <= addr) 23221da177e4SLinus Torvalds break; 23231da177e4SLinus Torvalds rb_node = rb_node->rb_left; 23241da177e4SLinus Torvalds } else 23251da177e4SLinus Torvalds rb_node = rb_node->rb_right; 23261da177e4SLinus Torvalds } 2327615d6e87SDavidlohr Bueso 23281da177e4SLinus Torvalds if (vma) 2329615d6e87SDavidlohr Bueso vmacache_update(addr, vma); 23301da177e4SLinus Torvalds return vma; 23311da177e4SLinus Torvalds } 23321da177e4SLinus Torvalds 23331da177e4SLinus Torvalds EXPORT_SYMBOL(find_vma); 23341da177e4SLinus Torvalds 23356bd4837dSKOSAKI Motohiro /* 23366bd4837dSKOSAKI Motohiro * Same as find_vma, but also return a pointer to the previous VMA in *pprev. 23376bd4837dSKOSAKI Motohiro */ 23381da177e4SLinus Torvalds struct vm_area_struct * 23391da177e4SLinus Torvalds find_vma_prev(struct mm_struct *mm, unsigned long addr, 23401da177e4SLinus Torvalds struct vm_area_struct **pprev) 23411da177e4SLinus Torvalds { 23426bd4837dSKOSAKI Motohiro struct vm_area_struct *vma; 23431da177e4SLinus Torvalds 23446bd4837dSKOSAKI Motohiro vma = find_vma(mm, addr); 234583cd904dSMikulas Patocka if (vma) { 234683cd904dSMikulas Patocka *pprev = vma->vm_prev; 234783cd904dSMikulas Patocka } else { 234873848a97SWei Yang struct rb_node *rb_node = rb_last(&mm->mm_rb); 234973848a97SWei Yang 235073848a97SWei Yang *pprev = rb_node ? rb_entry(rb_node, struct vm_area_struct, vm_rb) : NULL; 235183cd904dSMikulas Patocka } 23526bd4837dSKOSAKI Motohiro return vma; 23531da177e4SLinus Torvalds } 23541da177e4SLinus Torvalds 23551da177e4SLinus Torvalds /* 23561da177e4SLinus Torvalds * Verify that the stack growth is acceptable and 23571da177e4SLinus Torvalds * update accounting. This is shared with both the 23581da177e4SLinus Torvalds * grow-up and grow-down cases. 23591da177e4SLinus Torvalds */ 23601be7107fSHugh Dickins static int acct_stack_growth(struct vm_area_struct *vma, 23611be7107fSHugh Dickins unsigned long size, unsigned long grow) 23621da177e4SLinus Torvalds { 23631da177e4SLinus Torvalds struct mm_struct *mm = vma->vm_mm; 23641be7107fSHugh Dickins unsigned long new_start; 23651da177e4SLinus Torvalds 23661da177e4SLinus Torvalds /* address space limit tests */ 236784638335SKonstantin Khlebnikov if (!may_expand_vm(mm, vma->vm_flags, grow)) 23681da177e4SLinus Torvalds return -ENOMEM; 23691da177e4SLinus Torvalds 23701da177e4SLinus Torvalds /* Stack limit test */ 237124c79d8eSKrzysztof Opasiak if (size > rlimit(RLIMIT_STACK)) 23721da177e4SLinus Torvalds return -ENOMEM; 23731da177e4SLinus Torvalds 23741da177e4SLinus Torvalds /* mlock limit tests */ 23751da177e4SLinus Torvalds if (vma->vm_flags & VM_LOCKED) { 23761da177e4SLinus Torvalds unsigned long locked; 23771da177e4SLinus Torvalds unsigned long limit; 23781da177e4SLinus Torvalds locked = mm->locked_vm + grow; 237924c79d8eSKrzysztof Opasiak limit = rlimit(RLIMIT_MEMLOCK); 238059e99e5bSJiri Slaby limit >>= PAGE_SHIFT; 23811da177e4SLinus Torvalds if (locked > limit && !capable(CAP_IPC_LOCK)) 23821da177e4SLinus Torvalds return -ENOMEM; 23831da177e4SLinus Torvalds } 23841da177e4SLinus Torvalds 23850d59a01bSAdam Litke /* Check to ensure the stack will not grow into a hugetlb-only region */ 23860d59a01bSAdam Litke new_start = (vma->vm_flags & VM_GROWSUP) ? vma->vm_start : 23870d59a01bSAdam Litke vma->vm_end - size; 23880d59a01bSAdam Litke if (is_hugepage_only_range(vma->vm_mm, new_start, size)) 23890d59a01bSAdam Litke return -EFAULT; 23900d59a01bSAdam Litke 23911da177e4SLinus Torvalds /* 23921da177e4SLinus Torvalds * Overcommit.. This must be the final test, as it will 23931da177e4SLinus Torvalds * update security statistics. 23941da177e4SLinus Torvalds */ 239505fa199dSHugh Dickins if (security_vm_enough_memory_mm(mm, grow)) 23961da177e4SLinus Torvalds return -ENOMEM; 23971da177e4SLinus Torvalds 23981da177e4SLinus Torvalds return 0; 23991da177e4SLinus Torvalds } 24001da177e4SLinus Torvalds 240146dea3d0SHugh Dickins #if defined(CONFIG_STACK_GROWSUP) || defined(CONFIG_IA64) 24021da177e4SLinus Torvalds /* 240346dea3d0SHugh Dickins * PA-RISC uses this for its stack; IA64 for its Register Backing Store. 240446dea3d0SHugh Dickins * vma is the last one with address > vma->vm_end. Have to extend vma. 24051da177e4SLinus Torvalds */ 240646dea3d0SHugh Dickins int expand_upwards(struct vm_area_struct *vma, unsigned long address) 24071da177e4SLinus Torvalds { 240809357814SOleg Nesterov struct mm_struct *mm = vma->vm_mm; 24091be7107fSHugh Dickins struct vm_area_struct *next; 24101be7107fSHugh Dickins unsigned long gap_addr; 241112352d3cSKonstantin Khlebnikov int error = 0; 24121da177e4SLinus Torvalds 24131da177e4SLinus Torvalds if (!(vma->vm_flags & VM_GROWSUP)) 24141da177e4SLinus Torvalds return -EFAULT; 24151da177e4SLinus Torvalds 2416bd726c90SHelge Deller /* Guard against exceeding limits of the address space. */ 24171be7107fSHugh Dickins address &= PAGE_MASK; 241837511fb5SHelge Deller if (address >= (TASK_SIZE & PAGE_MASK)) 241912352d3cSKonstantin Khlebnikov return -ENOMEM; 2420bd726c90SHelge Deller address += PAGE_SIZE; 242112352d3cSKonstantin Khlebnikov 24221be7107fSHugh Dickins /* Enforce stack_guard_gap */ 24231be7107fSHugh Dickins gap_addr = address + stack_guard_gap; 2424bd726c90SHelge Deller 2425bd726c90SHelge Deller /* Guard against overflow */ 2426bd726c90SHelge Deller if (gap_addr < address || gap_addr > TASK_SIZE) 2427bd726c90SHelge Deller gap_addr = TASK_SIZE; 2428bd726c90SHelge Deller 24291be7107fSHugh Dickins next = vma->vm_next; 24303122e80eSAnshuman Khandual if (next && next->vm_start < gap_addr && vma_is_accessible(next)) { 24311be7107fSHugh Dickins if (!(next->vm_flags & VM_GROWSUP)) 24321be7107fSHugh Dickins return -ENOMEM; 24331be7107fSHugh Dickins /* Check that both stack segments have the same anon_vma? */ 24341be7107fSHugh Dickins } 24351be7107fSHugh Dickins 243612352d3cSKonstantin Khlebnikov /* We must make sure the anon_vma is allocated. */ 24371da177e4SLinus Torvalds if (unlikely(anon_vma_prepare(vma))) 24381da177e4SLinus Torvalds return -ENOMEM; 24391da177e4SLinus Torvalds 24401da177e4SLinus Torvalds /* 24411da177e4SLinus Torvalds * vma->vm_start/vm_end cannot change under us because the caller 2442c1e8d7c6SMichel Lespinasse * is required to hold the mmap_lock in read mode. We need the 24431da177e4SLinus Torvalds * anon_vma lock to serialize against concurrent expand_stacks. 24441da177e4SLinus Torvalds */ 244512352d3cSKonstantin Khlebnikov anon_vma_lock_write(vma->anon_vma); 24461da177e4SLinus Torvalds 24471da177e4SLinus Torvalds /* Somebody else might have raced and expanded it already */ 24481da177e4SLinus Torvalds if (address > vma->vm_end) { 24491da177e4SLinus Torvalds unsigned long size, grow; 24501da177e4SLinus Torvalds 24511da177e4SLinus Torvalds size = address - vma->vm_start; 24521da177e4SLinus Torvalds grow = (address - vma->vm_end) >> PAGE_SHIFT; 24531da177e4SLinus Torvalds 245442c36f63SHugh Dickins error = -ENOMEM; 245542c36f63SHugh Dickins if (vma->vm_pgoff + (size >> PAGE_SHIFT) >= vma->vm_pgoff) { 24561da177e4SLinus Torvalds error = acct_stack_growth(vma, size, grow); 24573af9e859SEric B Munson if (!error) { 24584128997bSMichel Lespinasse /* 24594128997bSMichel Lespinasse * vma_gap_update() doesn't support concurrent 2460c1e8d7c6SMichel Lespinasse * updates, but we only hold a shared mmap_lock 24614128997bSMichel Lespinasse * lock here, so we need to protect against 24624128997bSMichel Lespinasse * concurrent vma expansions. 246312352d3cSKonstantin Khlebnikov * anon_vma_lock_write() doesn't help here, as 24644128997bSMichel Lespinasse * we don't guarantee that all growable vmas 24654128997bSMichel Lespinasse * in a mm share the same root anon vma. 24664128997bSMichel Lespinasse * So, we reuse mm->page_table_lock to guard 24674128997bSMichel Lespinasse * against concurrent vma expansions. 24684128997bSMichel Lespinasse */ 246909357814SOleg Nesterov spin_lock(&mm->page_table_lock); 247087e8827bSOleg Nesterov if (vma->vm_flags & VM_LOCKED) 247109357814SOleg Nesterov mm->locked_vm += grow; 247284638335SKonstantin Khlebnikov vm_stat_account(mm, vma->vm_flags, grow); 2473bf181b9fSMichel Lespinasse anon_vma_interval_tree_pre_update_vma(vma); 24741da177e4SLinus Torvalds vma->vm_end = address; 2475bf181b9fSMichel Lespinasse anon_vma_interval_tree_post_update_vma(vma); 2476d3737187SMichel Lespinasse if (vma->vm_next) 2477d3737187SMichel Lespinasse vma_gap_update(vma->vm_next); 2478d3737187SMichel Lespinasse else 24791be7107fSHugh Dickins mm->highest_vm_end = vm_end_gap(vma); 248009357814SOleg Nesterov spin_unlock(&mm->page_table_lock); 24814128997bSMichel Lespinasse 24823af9e859SEric B Munson perf_event_mmap(vma); 24833af9e859SEric B Munson } 24841da177e4SLinus Torvalds } 248542c36f63SHugh Dickins } 248612352d3cSKonstantin Khlebnikov anon_vma_unlock_write(vma->anon_vma); 24876d50e60cSDavid Rientjes khugepaged_enter_vma_merge(vma, vma->vm_flags); 248809357814SOleg Nesterov validate_mm(mm); 24891da177e4SLinus Torvalds return error; 24901da177e4SLinus Torvalds } 249146dea3d0SHugh Dickins #endif /* CONFIG_STACK_GROWSUP || CONFIG_IA64 */ 249246dea3d0SHugh Dickins 24931da177e4SLinus Torvalds /* 24941da177e4SLinus Torvalds * vma is the first one with address < vma->vm_start. Have to extend vma. 24951da177e4SLinus Torvalds */ 2496d05f3169SMichal Hocko int expand_downwards(struct vm_area_struct *vma, 2497b6a2fea3SOllie Wild unsigned long address) 24981da177e4SLinus Torvalds { 249909357814SOleg Nesterov struct mm_struct *mm = vma->vm_mm; 25001be7107fSHugh Dickins struct vm_area_struct *prev; 25010a1d5299SJann Horn int error = 0; 25021da177e4SLinus Torvalds 25038869477aSEric Paris address &= PAGE_MASK; 25040a1d5299SJann Horn if (address < mmap_min_addr) 25050a1d5299SJann Horn return -EPERM; 25068869477aSEric Paris 25071be7107fSHugh Dickins /* Enforce stack_guard_gap */ 25081be7107fSHugh Dickins prev = vma->vm_prev; 25091be7107fSHugh Dickins /* Check that both stack segments have the same anon_vma? */ 251032e4e6d5SOleg Nesterov if (prev && !(prev->vm_flags & VM_GROWSDOWN) && 25113122e80eSAnshuman Khandual vma_is_accessible(prev)) { 251232e4e6d5SOleg Nesterov if (address - prev->vm_end < stack_guard_gap) 251332e4e6d5SOleg Nesterov return -ENOMEM; 25141be7107fSHugh Dickins } 25151be7107fSHugh Dickins 251612352d3cSKonstantin Khlebnikov /* We must make sure the anon_vma is allocated. */ 251712352d3cSKonstantin Khlebnikov if (unlikely(anon_vma_prepare(vma))) 251812352d3cSKonstantin Khlebnikov return -ENOMEM; 25191da177e4SLinus Torvalds 25201da177e4SLinus Torvalds /* 25211da177e4SLinus Torvalds * vma->vm_start/vm_end cannot change under us because the caller 2522c1e8d7c6SMichel Lespinasse * is required to hold the mmap_lock in read mode. We need the 25231da177e4SLinus Torvalds * anon_vma lock to serialize against concurrent expand_stacks. 25241da177e4SLinus Torvalds */ 252512352d3cSKonstantin Khlebnikov anon_vma_lock_write(vma->anon_vma); 25261da177e4SLinus Torvalds 25271da177e4SLinus Torvalds /* Somebody else might have raced and expanded it already */ 25281da177e4SLinus Torvalds if (address < vma->vm_start) { 25291da177e4SLinus Torvalds unsigned long size, grow; 25301da177e4SLinus Torvalds 25311da177e4SLinus Torvalds size = vma->vm_end - address; 25321da177e4SLinus Torvalds grow = (vma->vm_start - address) >> PAGE_SHIFT; 25331da177e4SLinus Torvalds 2534a626ca6aSLinus Torvalds error = -ENOMEM; 2535a626ca6aSLinus Torvalds if (grow <= vma->vm_pgoff) { 25361da177e4SLinus Torvalds error = acct_stack_growth(vma, size, grow); 25371da177e4SLinus Torvalds if (!error) { 25384128997bSMichel Lespinasse /* 25394128997bSMichel Lespinasse * vma_gap_update() doesn't support concurrent 2540c1e8d7c6SMichel Lespinasse * updates, but we only hold a shared mmap_lock 25414128997bSMichel Lespinasse * lock here, so we need to protect against 25424128997bSMichel Lespinasse * concurrent vma expansions. 254312352d3cSKonstantin Khlebnikov * anon_vma_lock_write() doesn't help here, as 25444128997bSMichel Lespinasse * we don't guarantee that all growable vmas 25454128997bSMichel Lespinasse * in a mm share the same root anon vma. 25464128997bSMichel Lespinasse * So, we reuse mm->page_table_lock to guard 25474128997bSMichel Lespinasse * against concurrent vma expansions. 25484128997bSMichel Lespinasse */ 254909357814SOleg Nesterov spin_lock(&mm->page_table_lock); 255087e8827bSOleg Nesterov if (vma->vm_flags & VM_LOCKED) 255109357814SOleg Nesterov mm->locked_vm += grow; 255284638335SKonstantin Khlebnikov vm_stat_account(mm, vma->vm_flags, grow); 2553bf181b9fSMichel Lespinasse anon_vma_interval_tree_pre_update_vma(vma); 25541da177e4SLinus Torvalds vma->vm_start = address; 25551da177e4SLinus Torvalds vma->vm_pgoff -= grow; 2556bf181b9fSMichel Lespinasse anon_vma_interval_tree_post_update_vma(vma); 2557d3737187SMichel Lespinasse vma_gap_update(vma); 255809357814SOleg Nesterov spin_unlock(&mm->page_table_lock); 25594128997bSMichel Lespinasse 25603af9e859SEric B Munson perf_event_mmap(vma); 25611da177e4SLinus Torvalds } 25621da177e4SLinus Torvalds } 2563a626ca6aSLinus Torvalds } 256412352d3cSKonstantin Khlebnikov anon_vma_unlock_write(vma->anon_vma); 25656d50e60cSDavid Rientjes khugepaged_enter_vma_merge(vma, vma->vm_flags); 256609357814SOleg Nesterov validate_mm(mm); 25671da177e4SLinus Torvalds return error; 25681da177e4SLinus Torvalds } 25691da177e4SLinus Torvalds 25701be7107fSHugh Dickins /* enforced gap between the expanding stack and other mappings. */ 25711be7107fSHugh Dickins unsigned long stack_guard_gap = 256UL<<PAGE_SHIFT; 25721be7107fSHugh Dickins 25731be7107fSHugh Dickins static int __init cmdline_parse_stack_guard_gap(char *p) 25741be7107fSHugh Dickins { 25751be7107fSHugh Dickins unsigned long val; 25761be7107fSHugh Dickins char *endptr; 25771be7107fSHugh Dickins 25781be7107fSHugh Dickins val = simple_strtoul(p, &endptr, 10); 25791be7107fSHugh Dickins if (!*endptr) 25801be7107fSHugh Dickins stack_guard_gap = val << PAGE_SHIFT; 25811be7107fSHugh Dickins 25821be7107fSHugh Dickins return 0; 25831be7107fSHugh Dickins } 25841be7107fSHugh Dickins __setup("stack_guard_gap=", cmdline_parse_stack_guard_gap); 25851be7107fSHugh Dickins 2586b6a2fea3SOllie Wild #ifdef CONFIG_STACK_GROWSUP 2587b6a2fea3SOllie Wild int expand_stack(struct vm_area_struct *vma, unsigned long address) 2588b6a2fea3SOllie Wild { 2589b6a2fea3SOllie Wild return expand_upwards(vma, address); 2590b6a2fea3SOllie Wild } 2591b6a2fea3SOllie Wild 2592b6a2fea3SOllie Wild struct vm_area_struct * 2593b6a2fea3SOllie Wild find_extend_vma(struct mm_struct *mm, unsigned long addr) 2594b6a2fea3SOllie Wild { 2595b6a2fea3SOllie Wild struct vm_area_struct *vma, *prev; 2596b6a2fea3SOllie Wild 2597b6a2fea3SOllie Wild addr &= PAGE_MASK; 2598b6a2fea3SOllie Wild vma = find_vma_prev(mm, addr, &prev); 2599b6a2fea3SOllie Wild if (vma && (vma->vm_start <= addr)) 2600b6a2fea3SOllie Wild return vma; 260104f5866eSAndrea Arcangeli /* don't alter vm_end if the coredump is running */ 26024d45e75aSJann Horn if (!prev || expand_stack(prev, addr)) 2603b6a2fea3SOllie Wild return NULL; 2604cea10a19SMichel Lespinasse if (prev->vm_flags & VM_LOCKED) 2605fc05f566SKirill A. Shutemov populate_vma_page_range(prev, addr, prev->vm_end, NULL); 2606b6a2fea3SOllie Wild return prev; 2607b6a2fea3SOllie Wild } 2608b6a2fea3SOllie Wild #else 2609b6a2fea3SOllie Wild int expand_stack(struct vm_area_struct *vma, unsigned long address) 2610b6a2fea3SOllie Wild { 2611b6a2fea3SOllie Wild return expand_downwards(vma, address); 2612b6a2fea3SOllie Wild } 2613b6a2fea3SOllie Wild 26141da177e4SLinus Torvalds struct vm_area_struct * 26151da177e4SLinus Torvalds find_extend_vma(struct mm_struct *mm, unsigned long addr) 26161da177e4SLinus Torvalds { 26171da177e4SLinus Torvalds struct vm_area_struct *vma; 26181da177e4SLinus Torvalds unsigned long start; 26191da177e4SLinus Torvalds 26201da177e4SLinus Torvalds addr &= PAGE_MASK; 26211da177e4SLinus Torvalds vma = find_vma(mm, addr); 26221da177e4SLinus Torvalds if (!vma) 26231da177e4SLinus Torvalds return NULL; 26241da177e4SLinus Torvalds if (vma->vm_start <= addr) 26251da177e4SLinus Torvalds return vma; 26261da177e4SLinus Torvalds if (!(vma->vm_flags & VM_GROWSDOWN)) 26271da177e4SLinus Torvalds return NULL; 26281da177e4SLinus Torvalds start = vma->vm_start; 26291da177e4SLinus Torvalds if (expand_stack(vma, addr)) 26301da177e4SLinus Torvalds return NULL; 2631cea10a19SMichel Lespinasse if (vma->vm_flags & VM_LOCKED) 2632fc05f566SKirill A. Shutemov populate_vma_page_range(vma, addr, start, NULL); 26331da177e4SLinus Torvalds return vma; 26341da177e4SLinus Torvalds } 26351da177e4SLinus Torvalds #endif 26361da177e4SLinus Torvalds 2637e1d6d01aSJesse Barnes EXPORT_SYMBOL_GPL(find_extend_vma); 2638e1d6d01aSJesse Barnes 26392c0b3814SHugh Dickins /* 26402c0b3814SHugh Dickins * Ok - we have the memory areas we should free on the vma list, 26412c0b3814SHugh Dickins * so release them, and do the vma updates. 26421da177e4SLinus Torvalds * 26432c0b3814SHugh Dickins * Called with the mm semaphore held. 26441da177e4SLinus Torvalds */ 26452c0b3814SHugh Dickins static void remove_vma_list(struct mm_struct *mm, struct vm_area_struct *vma) 26461da177e4SLinus Torvalds { 26474f74d2c8SLinus Torvalds unsigned long nr_accounted = 0; 26484f74d2c8SLinus Torvalds 2649365e9c87SHugh Dickins /* Update high watermark before we lower total_vm */ 2650365e9c87SHugh Dickins update_hiwater_vm(mm); 26512c0b3814SHugh Dickins do { 2652ab50b8edSHugh Dickins long nrpages = vma_pages(vma); 26531da177e4SLinus Torvalds 26544f74d2c8SLinus Torvalds if (vma->vm_flags & VM_ACCOUNT) 26554f74d2c8SLinus Torvalds nr_accounted += nrpages; 265684638335SKonstantin Khlebnikov vm_stat_account(mm, vma->vm_flags, -nrpages); 2657a8fb5618SHugh Dickins vma = remove_vma(vma); 2658146425a3SHugh Dickins } while (vma); 26594f74d2c8SLinus Torvalds vm_unacct_memory(nr_accounted); 26601da177e4SLinus Torvalds validate_mm(mm); 26611da177e4SLinus Torvalds } 26621da177e4SLinus Torvalds 26631da177e4SLinus Torvalds /* 26641da177e4SLinus Torvalds * Get rid of page table information in the indicated region. 26651da177e4SLinus Torvalds * 2666f10df686SPaolo 'Blaisorblade' Giarrusso * Called with the mm semaphore held. 26671da177e4SLinus Torvalds */ 26681da177e4SLinus Torvalds static void unmap_region(struct mm_struct *mm, 2669e0da382cSHugh Dickins struct vm_area_struct *vma, struct vm_area_struct *prev, 2670e0da382cSHugh Dickins unsigned long start, unsigned long end) 26711da177e4SLinus Torvalds { 26723903b55aSLiam R. Howlett struct vm_area_struct *next = vma_next(mm, prev); 2673d16dfc55SPeter Zijlstra struct mmu_gather tlb; 26741da177e4SLinus Torvalds 26751da177e4SLinus Torvalds lru_add_drain(); 26762b047252SLinus Torvalds tlb_gather_mmu(&tlb, mm, start, end); 2677365e9c87SHugh Dickins update_hiwater_rss(mm); 26784f74d2c8SLinus Torvalds unmap_vmas(&tlb, vma, start, end); 2679d16dfc55SPeter Zijlstra free_pgtables(&tlb, vma, prev ? prev->vm_end : FIRST_USER_ADDRESS, 26806ee8630eSHugh Dickins next ? next->vm_start : USER_PGTABLES_CEILING); 2681d16dfc55SPeter Zijlstra tlb_finish_mmu(&tlb, start, end); 26821da177e4SLinus Torvalds } 26831da177e4SLinus Torvalds 26841da177e4SLinus Torvalds /* 26851da177e4SLinus Torvalds * Create a list of vma's touched by the unmap, removing them from the mm's 26861da177e4SLinus Torvalds * vma list as we go.. 26871da177e4SLinus Torvalds */ 2688246c320aSKirill A. Shutemov static bool 26891da177e4SLinus Torvalds detach_vmas_to_be_unmapped(struct mm_struct *mm, struct vm_area_struct *vma, 26901da177e4SLinus Torvalds struct vm_area_struct *prev, unsigned long end) 26911da177e4SLinus Torvalds { 26921da177e4SLinus Torvalds struct vm_area_struct **insertion_point; 26931da177e4SLinus Torvalds struct vm_area_struct *tail_vma = NULL; 26941da177e4SLinus Torvalds 26951da177e4SLinus Torvalds insertion_point = (prev ? &prev->vm_next : &mm->mmap); 2696297c5eeeSLinus Torvalds vma->vm_prev = NULL; 26971da177e4SLinus Torvalds do { 2698d3737187SMichel Lespinasse vma_rb_erase(vma, &mm->mm_rb); 26991da177e4SLinus Torvalds mm->map_count--; 27001da177e4SLinus Torvalds tail_vma = vma; 27011da177e4SLinus Torvalds vma = vma->vm_next; 27021da177e4SLinus Torvalds } while (vma && vma->vm_start < end); 27031da177e4SLinus Torvalds *insertion_point = vma; 2704d3737187SMichel Lespinasse if (vma) { 2705297c5eeeSLinus Torvalds vma->vm_prev = prev; 2706d3737187SMichel Lespinasse vma_gap_update(vma); 2707d3737187SMichel Lespinasse } else 27081be7107fSHugh Dickins mm->highest_vm_end = prev ? vm_end_gap(prev) : 0; 27091da177e4SLinus Torvalds tail_vma->vm_next = NULL; 2710615d6e87SDavidlohr Bueso 2711615d6e87SDavidlohr Bueso /* Kill the cache */ 2712615d6e87SDavidlohr Bueso vmacache_invalidate(mm); 2713246c320aSKirill A. Shutemov 2714246c320aSKirill A. Shutemov /* 2715246c320aSKirill A. Shutemov * Do not downgrade mmap_lock if we are next to VM_GROWSDOWN or 2716246c320aSKirill A. Shutemov * VM_GROWSUP VMA. Such VMAs can change their size under 2717246c320aSKirill A. Shutemov * down_read(mmap_lock) and collide with the VMA we are about to unmap. 2718246c320aSKirill A. Shutemov */ 2719246c320aSKirill A. Shutemov if (vma && (vma->vm_flags & VM_GROWSDOWN)) 2720246c320aSKirill A. Shutemov return false; 2721246c320aSKirill A. Shutemov if (prev && (prev->vm_flags & VM_GROWSUP)) 2722246c320aSKirill A. Shutemov return false; 2723246c320aSKirill A. Shutemov return true; 27241da177e4SLinus Torvalds } 27251da177e4SLinus Torvalds 27261da177e4SLinus Torvalds /* 2727def5efe0SDavid Rientjes * __split_vma() bypasses sysctl_max_map_count checking. We use this where it 2728def5efe0SDavid Rientjes * has already been checked or doesn't make sense to fail. 27291da177e4SLinus Torvalds */ 2730def5efe0SDavid Rientjes int __split_vma(struct mm_struct *mm, struct vm_area_struct *vma, 27311da177e4SLinus Torvalds unsigned long addr, int new_below) 27321da177e4SLinus Torvalds { 27331da177e4SLinus Torvalds struct vm_area_struct *new; 2734e3975891SChen Gang int err; 27351da177e4SLinus Torvalds 273631383c68SDan Williams if (vma->vm_ops && vma->vm_ops->split) { 273731383c68SDan Williams err = vma->vm_ops->split(vma, addr); 273831383c68SDan Williams if (err) 273931383c68SDan Williams return err; 274031383c68SDan Williams } 27411da177e4SLinus Torvalds 27423928d4f5SLinus Torvalds new = vm_area_dup(vma); 27431da177e4SLinus Torvalds if (!new) 2744e3975891SChen Gang return -ENOMEM; 27451da177e4SLinus Torvalds 27461da177e4SLinus Torvalds if (new_below) 27471da177e4SLinus Torvalds new->vm_end = addr; 27481da177e4SLinus Torvalds else { 27491da177e4SLinus Torvalds new->vm_start = addr; 27501da177e4SLinus Torvalds new->vm_pgoff += ((addr - vma->vm_start) >> PAGE_SHIFT); 27511da177e4SLinus Torvalds } 27521da177e4SLinus Torvalds 2753ef0855d3SOleg Nesterov err = vma_dup_policy(vma, new); 2754ef0855d3SOleg Nesterov if (err) 27555beb4930SRik van Riel goto out_free_vma; 27561da177e4SLinus Torvalds 2757c4ea95d7SDaniel Forrest err = anon_vma_clone(new, vma); 2758c4ea95d7SDaniel Forrest if (err) 27595beb4930SRik van Riel goto out_free_mpol; 27605beb4930SRik van Riel 2761e9714acfSKonstantin Khlebnikov if (new->vm_file) 27621da177e4SLinus Torvalds get_file(new->vm_file); 27631da177e4SLinus Torvalds 27641da177e4SLinus Torvalds if (new->vm_ops && new->vm_ops->open) 27651da177e4SLinus Torvalds new->vm_ops->open(new); 27661da177e4SLinus Torvalds 27671da177e4SLinus Torvalds if (new_below) 27685beb4930SRik van Riel err = vma_adjust(vma, addr, vma->vm_end, vma->vm_pgoff + 27691da177e4SLinus Torvalds ((addr - new->vm_start) >> PAGE_SHIFT), new); 27701da177e4SLinus Torvalds else 27715beb4930SRik van Riel err = vma_adjust(vma, vma->vm_start, addr, vma->vm_pgoff, new); 27721da177e4SLinus Torvalds 27735beb4930SRik van Riel /* Success. */ 27745beb4930SRik van Riel if (!err) 27751da177e4SLinus Torvalds return 0; 27765beb4930SRik van Riel 27775beb4930SRik van Riel /* Clean everything up if vma_adjust failed. */ 277858927533SRik van Riel if (new->vm_ops && new->vm_ops->close) 27795beb4930SRik van Riel new->vm_ops->close(new); 2780e9714acfSKonstantin Khlebnikov if (new->vm_file) 27815beb4930SRik van Riel fput(new->vm_file); 27822aeadc30SAndrea Arcangeli unlink_anon_vmas(new); 27835beb4930SRik van Riel out_free_mpol: 2784ef0855d3SOleg Nesterov mpol_put(vma_policy(new)); 27855beb4930SRik van Riel out_free_vma: 27863928d4f5SLinus Torvalds vm_area_free(new); 27875beb4930SRik van Riel return err; 27881da177e4SLinus Torvalds } 27891da177e4SLinus Torvalds 2790659ace58SKOSAKI Motohiro /* 2791659ace58SKOSAKI Motohiro * Split a vma into two pieces at address 'addr', a new vma is allocated 2792659ace58SKOSAKI Motohiro * either for the first part or the tail. 2793659ace58SKOSAKI Motohiro */ 2794659ace58SKOSAKI Motohiro int split_vma(struct mm_struct *mm, struct vm_area_struct *vma, 2795659ace58SKOSAKI Motohiro unsigned long addr, int new_below) 2796659ace58SKOSAKI Motohiro { 2797659ace58SKOSAKI Motohiro if (mm->map_count >= sysctl_max_map_count) 2798659ace58SKOSAKI Motohiro return -ENOMEM; 2799659ace58SKOSAKI Motohiro 2800659ace58SKOSAKI Motohiro return __split_vma(mm, vma, addr, new_below); 2801659ace58SKOSAKI Motohiro } 2802659ace58SKOSAKI Motohiro 28031da177e4SLinus Torvalds /* Munmap is split into 2 main parts -- this part which finds 28041da177e4SLinus Torvalds * what needs doing, and the areas themselves, which do the 28051da177e4SLinus Torvalds * work. This now handles partial unmappings. 28061da177e4SLinus Torvalds * Jeremy Fitzhardinge <jeremy@goop.org> 28071da177e4SLinus Torvalds */ 280885a06835SYang Shi int __do_munmap(struct mm_struct *mm, unsigned long start, size_t len, 2809dd2283f2SYang Shi struct list_head *uf, bool downgrade) 28101da177e4SLinus Torvalds { 28111da177e4SLinus Torvalds unsigned long end; 2812146425a3SHugh Dickins struct vm_area_struct *vma, *prev, *last; 28131da177e4SLinus Torvalds 2814de1741a1SAlexander Kuleshov if ((offset_in_page(start)) || start > TASK_SIZE || len > TASK_SIZE-start) 28151da177e4SLinus Torvalds return -EINVAL; 28161da177e4SLinus Torvalds 2817cc71aba3Svishnu.ps len = PAGE_ALIGN(len); 28185a28fc94SDave Hansen end = start + len; 2819cc71aba3Svishnu.ps if (len == 0) 28201da177e4SLinus Torvalds return -EINVAL; 28211da177e4SLinus Torvalds 28225a28fc94SDave Hansen /* 28235a28fc94SDave Hansen * arch_unmap() might do unmaps itself. It must be called 28245a28fc94SDave Hansen * and finish any rbtree manipulation before this code 28255a28fc94SDave Hansen * runs and also starts to manipulate the rbtree. 28265a28fc94SDave Hansen */ 28275a28fc94SDave Hansen arch_unmap(mm, start, end); 28285a28fc94SDave Hansen 28291da177e4SLinus Torvalds /* Find the first overlapping VMA */ 28309be34c9dSLinus Torvalds vma = find_vma(mm, start); 2831146425a3SHugh Dickins if (!vma) 28321da177e4SLinus Torvalds return 0; 28339be34c9dSLinus Torvalds prev = vma->vm_prev; 2834146425a3SHugh Dickins /* we have start < vma->vm_end */ 28351da177e4SLinus Torvalds 28361da177e4SLinus Torvalds /* if it doesn't overlap, we have nothing.. */ 2837146425a3SHugh Dickins if (vma->vm_start >= end) 28381da177e4SLinus Torvalds return 0; 28391da177e4SLinus Torvalds 28401da177e4SLinus Torvalds /* 28411da177e4SLinus Torvalds * If we need to split any vma, do it now to save pain later. 28421da177e4SLinus Torvalds * 28431da177e4SLinus Torvalds * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially 28441da177e4SLinus Torvalds * unmapped vm_area_struct will remain in use: so lower split_vma 28451da177e4SLinus Torvalds * places tmp vma above, and higher split_vma places tmp vma below. 28461da177e4SLinus Torvalds */ 2847146425a3SHugh Dickins if (start > vma->vm_start) { 2848659ace58SKOSAKI Motohiro int error; 2849659ace58SKOSAKI Motohiro 2850659ace58SKOSAKI Motohiro /* 2851659ace58SKOSAKI Motohiro * Make sure that map_count on return from munmap() will 2852659ace58SKOSAKI Motohiro * not exceed its limit; but let map_count go just above 2853659ace58SKOSAKI Motohiro * its limit temporarily, to help free resources as expected. 2854659ace58SKOSAKI Motohiro */ 2855659ace58SKOSAKI Motohiro if (end < vma->vm_end && mm->map_count >= sysctl_max_map_count) 2856659ace58SKOSAKI Motohiro return -ENOMEM; 2857659ace58SKOSAKI Motohiro 2858659ace58SKOSAKI Motohiro error = __split_vma(mm, vma, start, 0); 28591da177e4SLinus Torvalds if (error) 28601da177e4SLinus Torvalds return error; 2861146425a3SHugh Dickins prev = vma; 28621da177e4SLinus Torvalds } 28631da177e4SLinus Torvalds 28641da177e4SLinus Torvalds /* Does it split the last one? */ 28651da177e4SLinus Torvalds last = find_vma(mm, end); 28661da177e4SLinus Torvalds if (last && end > last->vm_start) { 2867659ace58SKOSAKI Motohiro int error = __split_vma(mm, last, end, 1); 28681da177e4SLinus Torvalds if (error) 28691da177e4SLinus Torvalds return error; 28701da177e4SLinus Torvalds } 28713903b55aSLiam R. Howlett vma = vma_next(mm, prev); 28721da177e4SLinus Torvalds 28732376dd7cSAndrea Arcangeli if (unlikely(uf)) { 28742376dd7cSAndrea Arcangeli /* 28752376dd7cSAndrea Arcangeli * If userfaultfd_unmap_prep returns an error the vmas 28762376dd7cSAndrea Arcangeli * will remain splitted, but userland will get a 28772376dd7cSAndrea Arcangeli * highly unexpected error anyway. This is no 28782376dd7cSAndrea Arcangeli * different than the case where the first of the two 28792376dd7cSAndrea Arcangeli * __split_vma fails, but we don't undo the first 28802376dd7cSAndrea Arcangeli * split, despite we could. This is unlikely enough 28812376dd7cSAndrea Arcangeli * failure that it's not worth optimizing it for. 28822376dd7cSAndrea Arcangeli */ 28832376dd7cSAndrea Arcangeli int error = userfaultfd_unmap_prep(vma, start, end, uf); 28842376dd7cSAndrea Arcangeli if (error) 28852376dd7cSAndrea Arcangeli return error; 28862376dd7cSAndrea Arcangeli } 28872376dd7cSAndrea Arcangeli 28881da177e4SLinus Torvalds /* 2889ba470de4SRik van Riel * unlock any mlock()ed ranges before detaching vmas 2890ba470de4SRik van Riel */ 2891ba470de4SRik van Riel if (mm->locked_vm) { 2892ba470de4SRik van Riel struct vm_area_struct *tmp = vma; 2893ba470de4SRik van Riel while (tmp && tmp->vm_start < end) { 2894ba470de4SRik van Riel if (tmp->vm_flags & VM_LOCKED) { 2895ba470de4SRik van Riel mm->locked_vm -= vma_pages(tmp); 2896ba470de4SRik van Riel munlock_vma_pages_all(tmp); 2897ba470de4SRik van Riel } 2898dd2283f2SYang Shi 2899ba470de4SRik van Riel tmp = tmp->vm_next; 2900ba470de4SRik van Riel } 2901ba470de4SRik van Riel } 2902ba470de4SRik van Riel 2903dd2283f2SYang Shi /* Detach vmas from rbtree */ 2904246c320aSKirill A. Shutemov if (!detach_vmas_to_be_unmapped(mm, vma, prev, end)) 2905246c320aSKirill A. Shutemov downgrade = false; 29061da177e4SLinus Torvalds 2907dd2283f2SYang Shi if (downgrade) 2908d8ed45c5SMichel Lespinasse mmap_write_downgrade(mm); 2909dd2283f2SYang Shi 2910dd2283f2SYang Shi unmap_region(mm, vma, prev, start, end); 2911dd2283f2SYang Shi 29121da177e4SLinus Torvalds /* Fix up all other VM information */ 29132c0b3814SHugh Dickins remove_vma_list(mm, vma); 29141da177e4SLinus Torvalds 2915dd2283f2SYang Shi return downgrade ? 1 : 0; 29161da177e4SLinus Torvalds } 29171da177e4SLinus Torvalds 2918dd2283f2SYang Shi int do_munmap(struct mm_struct *mm, unsigned long start, size_t len, 2919dd2283f2SYang Shi struct list_head *uf) 2920dd2283f2SYang Shi { 2921dd2283f2SYang Shi return __do_munmap(mm, start, len, uf, false); 2922dd2283f2SYang Shi } 2923dd2283f2SYang Shi 2924dd2283f2SYang Shi static int __vm_munmap(unsigned long start, size_t len, bool downgrade) 2925a46ef99dSLinus Torvalds { 2926a46ef99dSLinus Torvalds int ret; 2927bfce281cSAl Viro struct mm_struct *mm = current->mm; 2928897ab3e0SMike Rapoport LIST_HEAD(uf); 2929a46ef99dSLinus Torvalds 2930d8ed45c5SMichel Lespinasse if (mmap_write_lock_killable(mm)) 2931ae798783SMichal Hocko return -EINTR; 2932ae798783SMichal Hocko 2933dd2283f2SYang Shi ret = __do_munmap(mm, start, len, &uf, downgrade); 2934dd2283f2SYang Shi /* 2935c1e8d7c6SMichel Lespinasse * Returning 1 indicates mmap_lock is downgraded. 2936dd2283f2SYang Shi * But 1 is not legal return value of vm_munmap() and munmap(), reset 2937dd2283f2SYang Shi * it to 0 before return. 2938dd2283f2SYang Shi */ 2939dd2283f2SYang Shi if (ret == 1) { 2940d8ed45c5SMichel Lespinasse mmap_read_unlock(mm); 2941dd2283f2SYang Shi ret = 0; 2942dd2283f2SYang Shi } else 2943d8ed45c5SMichel Lespinasse mmap_write_unlock(mm); 2944dd2283f2SYang Shi 2945897ab3e0SMike Rapoport userfaultfd_unmap_complete(mm, &uf); 2946a46ef99dSLinus Torvalds return ret; 2947a46ef99dSLinus Torvalds } 2948dd2283f2SYang Shi 2949dd2283f2SYang Shi int vm_munmap(unsigned long start, size_t len) 2950dd2283f2SYang Shi { 2951dd2283f2SYang Shi return __vm_munmap(start, len, false); 2952dd2283f2SYang Shi } 2953a46ef99dSLinus Torvalds EXPORT_SYMBOL(vm_munmap); 2954a46ef99dSLinus Torvalds 29556a6160a7SHeiko Carstens SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len) 29561da177e4SLinus Torvalds { 2957ce18d171SCatalin Marinas addr = untagged_addr(addr); 29581da177e4SLinus Torvalds profile_munmap(addr); 2959dd2283f2SYang Shi return __vm_munmap(addr, len, true); 29601da177e4SLinus Torvalds } 29611da177e4SLinus Torvalds 2962c8d78c18SKirill A. Shutemov 2963c8d78c18SKirill A. Shutemov /* 2964c8d78c18SKirill A. Shutemov * Emulation of deprecated remap_file_pages() syscall. 2965c8d78c18SKirill A. Shutemov */ 2966c8d78c18SKirill A. Shutemov SYSCALL_DEFINE5(remap_file_pages, unsigned long, start, unsigned long, size, 2967c8d78c18SKirill A. Shutemov unsigned long, prot, unsigned long, pgoff, unsigned long, flags) 2968c8d78c18SKirill A. Shutemov { 2969c8d78c18SKirill A. Shutemov 2970c8d78c18SKirill A. Shutemov struct mm_struct *mm = current->mm; 2971c8d78c18SKirill A. Shutemov struct vm_area_struct *vma; 2972c8d78c18SKirill A. Shutemov unsigned long populate = 0; 2973c8d78c18SKirill A. Shutemov unsigned long ret = -EINVAL; 2974c8d78c18SKirill A. Shutemov struct file *file; 2975c8d78c18SKirill A. Shutemov 2976ad56b738SMike Rapoport pr_warn_once("%s (%d) uses deprecated remap_file_pages() syscall. See Documentation/vm/remap_file_pages.rst.\n", 2977c8d78c18SKirill A. Shutemov current->comm, current->pid); 2978c8d78c18SKirill A. Shutemov 2979c8d78c18SKirill A. Shutemov if (prot) 2980c8d78c18SKirill A. Shutemov return ret; 2981c8d78c18SKirill A. Shutemov start = start & PAGE_MASK; 2982c8d78c18SKirill A. Shutemov size = size & PAGE_MASK; 2983c8d78c18SKirill A. Shutemov 2984c8d78c18SKirill A. Shutemov if (start + size <= start) 2985c8d78c18SKirill A. Shutemov return ret; 2986c8d78c18SKirill A. Shutemov 2987c8d78c18SKirill A. Shutemov /* Does pgoff wrap? */ 2988c8d78c18SKirill A. Shutemov if (pgoff + (size >> PAGE_SHIFT) < pgoff) 2989c8d78c18SKirill A. Shutemov return ret; 2990c8d78c18SKirill A. Shutemov 2991d8ed45c5SMichel Lespinasse if (mmap_write_lock_killable(mm)) 2992dc0ef0dfSMichal Hocko return -EINTR; 2993dc0ef0dfSMichal Hocko 2994c8d78c18SKirill A. Shutemov vma = find_vma(mm, start); 2995c8d78c18SKirill A. Shutemov 2996c8d78c18SKirill A. Shutemov if (!vma || !(vma->vm_flags & VM_SHARED)) 2997c8d78c18SKirill A. Shutemov goto out; 2998c8d78c18SKirill A. Shutemov 299948f7df32SKirill A. Shutemov if (start < vma->vm_start) 3000c8d78c18SKirill A. Shutemov goto out; 3001c8d78c18SKirill A. Shutemov 300248f7df32SKirill A. Shutemov if (start + size > vma->vm_end) { 300348f7df32SKirill A. Shutemov struct vm_area_struct *next; 300448f7df32SKirill A. Shutemov 300548f7df32SKirill A. Shutemov for (next = vma->vm_next; next; next = next->vm_next) { 300648f7df32SKirill A. Shutemov /* hole between vmas ? */ 300748f7df32SKirill A. Shutemov if (next->vm_start != next->vm_prev->vm_end) 300848f7df32SKirill A. Shutemov goto out; 300948f7df32SKirill A. Shutemov 301048f7df32SKirill A. Shutemov if (next->vm_file != vma->vm_file) 301148f7df32SKirill A. Shutemov goto out; 301248f7df32SKirill A. Shutemov 301348f7df32SKirill A. Shutemov if (next->vm_flags != vma->vm_flags) 301448f7df32SKirill A. Shutemov goto out; 301548f7df32SKirill A. Shutemov 301648f7df32SKirill A. Shutemov if (start + size <= next->vm_end) 301748f7df32SKirill A. Shutemov break; 301848f7df32SKirill A. Shutemov } 301948f7df32SKirill A. Shutemov 302048f7df32SKirill A. Shutemov if (!next) 3021c8d78c18SKirill A. Shutemov goto out; 3022c8d78c18SKirill A. Shutemov } 3023c8d78c18SKirill A. Shutemov 3024c8d78c18SKirill A. Shutemov prot |= vma->vm_flags & VM_READ ? PROT_READ : 0; 3025c8d78c18SKirill A. Shutemov prot |= vma->vm_flags & VM_WRITE ? PROT_WRITE : 0; 3026c8d78c18SKirill A. Shutemov prot |= vma->vm_flags & VM_EXEC ? PROT_EXEC : 0; 3027c8d78c18SKirill A. Shutemov 3028c8d78c18SKirill A. Shutemov flags &= MAP_NONBLOCK; 3029c8d78c18SKirill A. Shutemov flags |= MAP_SHARED | MAP_FIXED | MAP_POPULATE; 3030c8d78c18SKirill A. Shutemov if (vma->vm_flags & VM_LOCKED) { 303148f7df32SKirill A. Shutemov struct vm_area_struct *tmp; 3032c8d78c18SKirill A. Shutemov flags |= MAP_LOCKED; 303348f7df32SKirill A. Shutemov 3034c8d78c18SKirill A. Shutemov /* drop PG_Mlocked flag for over-mapped range */ 303548f7df32SKirill A. Shutemov for (tmp = vma; tmp->vm_start >= start + size; 303648f7df32SKirill A. Shutemov tmp = tmp->vm_next) { 30379a73f61bSKirill A. Shutemov /* 30389a73f61bSKirill A. Shutemov * Split pmd and munlock page on the border 30399a73f61bSKirill A. Shutemov * of the range. 30409a73f61bSKirill A. Shutemov */ 30419a73f61bSKirill A. Shutemov vma_adjust_trans_huge(tmp, start, start + size, 0); 30429a73f61bSKirill A. Shutemov 304348f7df32SKirill A. Shutemov munlock_vma_pages_range(tmp, 304448f7df32SKirill A. Shutemov max(tmp->vm_start, start), 304548f7df32SKirill A. Shutemov min(tmp->vm_end, start + size)); 304648f7df32SKirill A. Shutemov } 3047c8d78c18SKirill A. Shutemov } 3048c8d78c18SKirill A. Shutemov 3049c8d78c18SKirill A. Shutemov file = get_file(vma->vm_file); 305045e55300SPeter Collingbourne ret = do_mmap(vma->vm_file, start, size, 3051897ab3e0SMike Rapoport prot, flags, pgoff, &populate, NULL); 3052c8d78c18SKirill A. Shutemov fput(file); 3053c8d78c18SKirill A. Shutemov out: 3054d8ed45c5SMichel Lespinasse mmap_write_unlock(mm); 3055c8d78c18SKirill A. Shutemov if (populate) 3056c8d78c18SKirill A. Shutemov mm_populate(ret, populate); 3057c8d78c18SKirill A. Shutemov if (!IS_ERR_VALUE(ret)) 3058c8d78c18SKirill A. Shutemov ret = 0; 3059c8d78c18SKirill A. Shutemov return ret; 3060c8d78c18SKirill A. Shutemov } 3061c8d78c18SKirill A. Shutemov 30621da177e4SLinus Torvalds /* 30631da177e4SLinus Torvalds * this is really a simplified "do_mmap". it only handles 30641da177e4SLinus Torvalds * anonymous maps. eventually we may be able to do some 30651da177e4SLinus Torvalds * brk-specific accounting here. 30661da177e4SLinus Torvalds */ 3067bb177a73SMichal Hocko static int do_brk_flags(unsigned long addr, unsigned long len, unsigned long flags, struct list_head *uf) 30681da177e4SLinus Torvalds { 30691da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 30701da177e4SLinus Torvalds struct vm_area_struct *vma, *prev; 30711da177e4SLinus Torvalds struct rb_node **rb_link, *rb_parent; 30721da177e4SLinus Torvalds pgoff_t pgoff = addr >> PAGE_SHIFT; 30733a459756SKirill Korotaev int error; 3074ff68dac6SGaowei Pu unsigned long mapped_addr; 30751da177e4SLinus Torvalds 307616e72e9bSDenys Vlasenko /* Until we need other flags, refuse anything except VM_EXEC. */ 307716e72e9bSDenys Vlasenko if ((flags & (~VM_EXEC)) != 0) 307816e72e9bSDenys Vlasenko return -EINVAL; 307916e72e9bSDenys Vlasenko flags |= VM_DATA_DEFAULT_FLAGS | VM_ACCOUNT | mm->def_flags; 30803a459756SKirill Korotaev 3081ff68dac6SGaowei Pu mapped_addr = get_unmapped_area(NULL, addr, len, 0, MAP_FIXED); 3082ff68dac6SGaowei Pu if (IS_ERR_VALUE(mapped_addr)) 3083ff68dac6SGaowei Pu return mapped_addr; 30843a459756SKirill Korotaev 3085363ee17fSDavidlohr Bueso error = mlock_future_check(mm, mm->def_flags, len); 3086363ee17fSDavidlohr Bueso if (error) 3087363ee17fSDavidlohr Bueso return error; 30881da177e4SLinus Torvalds 3089fb8090b6SLiam R. Howlett /* Clear old maps, set up prev, rb_link, rb_parent, and uf */ 3090fb8090b6SLiam R. Howlett if (munmap_vma_range(mm, addr, len, &prev, &rb_link, &rb_parent, uf)) 30911da177e4SLinus Torvalds return -ENOMEM; 30921da177e4SLinus Torvalds 30931da177e4SLinus Torvalds /* Check against address space limits *after* clearing old maps... */ 309484638335SKonstantin Khlebnikov if (!may_expand_vm(mm, flags, len >> PAGE_SHIFT)) 30951da177e4SLinus Torvalds return -ENOMEM; 30961da177e4SLinus Torvalds 30971da177e4SLinus Torvalds if (mm->map_count > sysctl_max_map_count) 30981da177e4SLinus Torvalds return -ENOMEM; 30991da177e4SLinus Torvalds 3100191c5424SAl Viro if (security_vm_enough_memory_mm(mm, len >> PAGE_SHIFT)) 31011da177e4SLinus Torvalds return -ENOMEM; 31021da177e4SLinus Torvalds 31031da177e4SLinus Torvalds /* Can we just expand an old private anonymous mapping? */ 3104ba470de4SRik van Riel vma = vma_merge(mm, prev, addr, addr + len, flags, 310519a809afSAndrea Arcangeli NULL, NULL, pgoff, NULL, NULL_VM_UFFD_CTX); 3106ba470de4SRik van Riel if (vma) 31071da177e4SLinus Torvalds goto out; 31081da177e4SLinus Torvalds 31091da177e4SLinus Torvalds /* 31101da177e4SLinus Torvalds * create a vma struct for an anonymous mapping 31111da177e4SLinus Torvalds */ 3112490fc053SLinus Torvalds vma = vm_area_alloc(mm); 31131da177e4SLinus Torvalds if (!vma) { 31141da177e4SLinus Torvalds vm_unacct_memory(len >> PAGE_SHIFT); 31151da177e4SLinus Torvalds return -ENOMEM; 31161da177e4SLinus Torvalds } 31171da177e4SLinus Torvalds 3118bfd40eafSKirill A. Shutemov vma_set_anonymous(vma); 31191da177e4SLinus Torvalds vma->vm_start = addr; 31201da177e4SLinus Torvalds vma->vm_end = addr + len; 31211da177e4SLinus Torvalds vma->vm_pgoff = pgoff; 31221da177e4SLinus Torvalds vma->vm_flags = flags; 31233ed75eb8SColy Li vma->vm_page_prot = vm_get_page_prot(flags); 31241da177e4SLinus Torvalds vma_link(mm, vma, prev, rb_link, rb_parent); 31251da177e4SLinus Torvalds out: 31263af9e859SEric B Munson perf_event_mmap(vma); 31271da177e4SLinus Torvalds mm->total_vm += len >> PAGE_SHIFT; 312884638335SKonstantin Khlebnikov mm->data_vm += len >> PAGE_SHIFT; 3129128557ffSMichel Lespinasse if (flags & VM_LOCKED) 3130ba470de4SRik van Riel mm->locked_vm += (len >> PAGE_SHIFT); 3131d9104d1cSCyrill Gorcunov vma->vm_flags |= VM_SOFTDIRTY; 31325d22fc25SLinus Torvalds return 0; 31331da177e4SLinus Torvalds } 31341da177e4SLinus Torvalds 3135bb177a73SMichal Hocko int vm_brk_flags(unsigned long addr, unsigned long request, unsigned long flags) 3136e4eb1ff6SLinus Torvalds { 3137e4eb1ff6SLinus Torvalds struct mm_struct *mm = current->mm; 3138bb177a73SMichal Hocko unsigned long len; 31395d22fc25SLinus Torvalds int ret; 3140128557ffSMichel Lespinasse bool populate; 3141897ab3e0SMike Rapoport LIST_HEAD(uf); 3142e4eb1ff6SLinus Torvalds 3143bb177a73SMichal Hocko len = PAGE_ALIGN(request); 3144bb177a73SMichal Hocko if (len < request) 3145bb177a73SMichal Hocko return -ENOMEM; 3146bb177a73SMichal Hocko if (!len) 3147bb177a73SMichal Hocko return 0; 3148bb177a73SMichal Hocko 3149d8ed45c5SMichel Lespinasse if (mmap_write_lock_killable(mm)) 31502d6c9282SMichal Hocko return -EINTR; 31512d6c9282SMichal Hocko 3152897ab3e0SMike Rapoport ret = do_brk_flags(addr, len, flags, &uf); 3153128557ffSMichel Lespinasse populate = ((mm->def_flags & VM_LOCKED) != 0); 3154d8ed45c5SMichel Lespinasse mmap_write_unlock(mm); 3155897ab3e0SMike Rapoport userfaultfd_unmap_complete(mm, &uf); 31565d22fc25SLinus Torvalds if (populate && !ret) 3157128557ffSMichel Lespinasse mm_populate(addr, len); 3158e4eb1ff6SLinus Torvalds return ret; 3159e4eb1ff6SLinus Torvalds } 316016e72e9bSDenys Vlasenko EXPORT_SYMBOL(vm_brk_flags); 316116e72e9bSDenys Vlasenko 316216e72e9bSDenys Vlasenko int vm_brk(unsigned long addr, unsigned long len) 316316e72e9bSDenys Vlasenko { 316416e72e9bSDenys Vlasenko return vm_brk_flags(addr, len, 0); 316516e72e9bSDenys Vlasenko } 3166e4eb1ff6SLinus Torvalds EXPORT_SYMBOL(vm_brk); 31671da177e4SLinus Torvalds 31681da177e4SLinus Torvalds /* Release all mmaps. */ 31691da177e4SLinus Torvalds void exit_mmap(struct mm_struct *mm) 31701da177e4SLinus Torvalds { 3171d16dfc55SPeter Zijlstra struct mmu_gather tlb; 3172ba470de4SRik van Riel struct vm_area_struct *vma; 31731da177e4SLinus Torvalds unsigned long nr_accounted = 0; 31741da177e4SLinus Torvalds 3175d6dd61c8SJeremy Fitzhardinge /* mm's last user has gone, and its about to be pulled down */ 3176cddb8a5cSAndrea Arcangeli mmu_notifier_release(mm); 3177d6dd61c8SJeremy Fitzhardinge 317827ae357fSDavid Rientjes if (unlikely(mm_is_oom_victim(mm))) { 317927ae357fSDavid Rientjes /* 318027ae357fSDavid Rientjes * Manually reap the mm to free as much memory as possible. 318127ae357fSDavid Rientjes * Then, as the oom reaper does, set MMF_OOM_SKIP to disregard 3182c1e8d7c6SMichel Lespinasse * this mm from further consideration. Taking mm->mmap_lock for 318327ae357fSDavid Rientjes * write after setting MMF_OOM_SKIP will guarantee that the oom 3184c1e8d7c6SMichel Lespinasse * reaper will not run on this mm again after mmap_lock is 318527ae357fSDavid Rientjes * dropped. 318627ae357fSDavid Rientjes * 3187c1e8d7c6SMichel Lespinasse * Nothing can be holding mm->mmap_lock here and the above call 318827ae357fSDavid Rientjes * to mmu_notifier_release(mm) ensures mmu notifier callbacks in 318927ae357fSDavid Rientjes * __oom_reap_task_mm() will not block. 319027ae357fSDavid Rientjes * 319127ae357fSDavid Rientjes * This needs to be done before calling munlock_vma_pages_all(), 319227ae357fSDavid Rientjes * which clears VM_LOCKED, otherwise the oom reaper cannot 319327ae357fSDavid Rientjes * reliably test it. 319427ae357fSDavid Rientjes */ 319593065ac7SMichal Hocko (void)__oom_reap_task_mm(mm); 319627ae357fSDavid Rientjes 319727ae357fSDavid Rientjes set_bit(MMF_OOM_SKIP, &mm->flags); 3198d8ed45c5SMichel Lespinasse mmap_write_lock(mm); 3199d8ed45c5SMichel Lespinasse mmap_write_unlock(mm); 320027ae357fSDavid Rientjes } 320127ae357fSDavid Rientjes 3202ba470de4SRik van Riel if (mm->locked_vm) { 3203ba470de4SRik van Riel vma = mm->mmap; 3204ba470de4SRik van Riel while (vma) { 3205ba470de4SRik van Riel if (vma->vm_flags & VM_LOCKED) 3206ba470de4SRik van Riel munlock_vma_pages_all(vma); 3207ba470de4SRik van Riel vma = vma->vm_next; 3208ba470de4SRik van Riel } 3209ba470de4SRik van Riel } 32109480c53eSJeremy Fitzhardinge 32119480c53eSJeremy Fitzhardinge arch_exit_mmap(mm); 32129480c53eSJeremy Fitzhardinge 3213ba470de4SRik van Riel vma = mm->mmap; 32149480c53eSJeremy Fitzhardinge if (!vma) /* Can happen if dup_mmap() received an OOM */ 32159480c53eSJeremy Fitzhardinge return; 32169480c53eSJeremy Fitzhardinge 32171da177e4SLinus Torvalds lru_add_drain(); 32181da177e4SLinus Torvalds flush_cache_mm(mm); 32192b047252SLinus Torvalds tlb_gather_mmu(&tlb, mm, 0, -1); 3220901608d9SOleg Nesterov /* update_hiwater_rss(mm) here? but nobody should be looking */ 3221e0da382cSHugh Dickins /* Use -1 here to ensure all VMAs in the mm are unmapped */ 32224f74d2c8SLinus Torvalds unmap_vmas(&tlb, vma, 0, -1); 32236ee8630eSHugh Dickins free_pgtables(&tlb, vma, FIRST_USER_ADDRESS, USER_PGTABLES_CEILING); 3224853f5e26SAl Viro tlb_finish_mmu(&tlb, 0, -1); 32251da177e4SLinus Torvalds 32261da177e4SLinus Torvalds /* 32278f4f8c16SHugh Dickins * Walk the list again, actually closing and freeing it, 32288f4f8c16SHugh Dickins * with preemption enabled, without holding any MM locks. 32291da177e4SLinus Torvalds */ 32304f74d2c8SLinus Torvalds while (vma) { 32314f74d2c8SLinus Torvalds if (vma->vm_flags & VM_ACCOUNT) 32324f74d2c8SLinus Torvalds nr_accounted += vma_pages(vma); 3233a8fb5618SHugh Dickins vma = remove_vma(vma); 32340a3b3c25SPaul E. McKenney cond_resched(); 32354f74d2c8SLinus Torvalds } 32364f74d2c8SLinus Torvalds vm_unacct_memory(nr_accounted); 32371da177e4SLinus Torvalds } 32381da177e4SLinus Torvalds 32391da177e4SLinus Torvalds /* Insert vm structure into process list sorted by address 32401da177e4SLinus Torvalds * and into the inode's i_mmap tree. If vm_file is non-NULL 3241c8c06efaSDavidlohr Bueso * then i_mmap_rwsem is taken here. 32421da177e4SLinus Torvalds */ 32431da177e4SLinus Torvalds int insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma) 32441da177e4SLinus Torvalds { 32456597d783SHugh Dickins struct vm_area_struct *prev; 32461da177e4SLinus Torvalds struct rb_node **rb_link, *rb_parent; 32471da177e4SLinus Torvalds 3248c9d13f5fSChen Gang if (find_vma_links(mm, vma->vm_start, vma->vm_end, 3249c9d13f5fSChen Gang &prev, &rb_link, &rb_parent)) 3250c9d13f5fSChen Gang return -ENOMEM; 3251c9d13f5fSChen Gang if ((vma->vm_flags & VM_ACCOUNT) && 3252c9d13f5fSChen Gang security_vm_enough_memory_mm(mm, vma_pages(vma))) 3253c9d13f5fSChen Gang return -ENOMEM; 3254c9d13f5fSChen Gang 32551da177e4SLinus Torvalds /* 32561da177e4SLinus Torvalds * The vm_pgoff of a purely anonymous vma should be irrelevant 32571da177e4SLinus Torvalds * until its first write fault, when page's anon_vma and index 32581da177e4SLinus Torvalds * are set. But now set the vm_pgoff it will almost certainly 32591da177e4SLinus Torvalds * end up with (unless mremap moves it elsewhere before that 32601da177e4SLinus Torvalds * first wfault), so /proc/pid/maps tells a consistent story. 32611da177e4SLinus Torvalds * 32621da177e4SLinus Torvalds * By setting it to reflect the virtual start address of the 32631da177e4SLinus Torvalds * vma, merges and splits can happen in a seamless way, just 32641da177e4SLinus Torvalds * using the existing file pgoff checks and manipulations. 32658332326eSLiao Pingfang * Similarly in do_mmap and in do_brk_flags. 32661da177e4SLinus Torvalds */ 32678a9cc3b5SOleg Nesterov if (vma_is_anonymous(vma)) { 32681da177e4SLinus Torvalds BUG_ON(vma->anon_vma); 32691da177e4SLinus Torvalds vma->vm_pgoff = vma->vm_start >> PAGE_SHIFT; 32701da177e4SLinus Torvalds } 32712b144498SSrikar Dronamraju 32721da177e4SLinus Torvalds vma_link(mm, vma, prev, rb_link, rb_parent); 32731da177e4SLinus Torvalds return 0; 32741da177e4SLinus Torvalds } 32751da177e4SLinus Torvalds 32761da177e4SLinus Torvalds /* 32771da177e4SLinus Torvalds * Copy the vma structure to a new location in the same mm, 32781da177e4SLinus Torvalds * prior to moving page table entries, to effect an mremap move. 32791da177e4SLinus Torvalds */ 32801da177e4SLinus Torvalds struct vm_area_struct *copy_vma(struct vm_area_struct **vmap, 328138a76013SMichel Lespinasse unsigned long addr, unsigned long len, pgoff_t pgoff, 328238a76013SMichel Lespinasse bool *need_rmap_locks) 32831da177e4SLinus Torvalds { 32841da177e4SLinus Torvalds struct vm_area_struct *vma = *vmap; 32851da177e4SLinus Torvalds unsigned long vma_start = vma->vm_start; 32861da177e4SLinus Torvalds struct mm_struct *mm = vma->vm_mm; 32871da177e4SLinus Torvalds struct vm_area_struct *new_vma, *prev; 32881da177e4SLinus Torvalds struct rb_node **rb_link, *rb_parent; 3289948f017bSAndrea Arcangeli bool faulted_in_anon_vma = true; 32901da177e4SLinus Torvalds 32911da177e4SLinus Torvalds /* 32921da177e4SLinus Torvalds * If anonymous vma has not yet been faulted, update new pgoff 32931da177e4SLinus Torvalds * to match new location, to increase its chance of merging. 32941da177e4SLinus Torvalds */ 3295ce75799bSOleg Nesterov if (unlikely(vma_is_anonymous(vma) && !vma->anon_vma)) { 32961da177e4SLinus Torvalds pgoff = addr >> PAGE_SHIFT; 3297948f017bSAndrea Arcangeli faulted_in_anon_vma = false; 3298948f017bSAndrea Arcangeli } 32991da177e4SLinus Torvalds 33006597d783SHugh Dickins if (find_vma_links(mm, addr, addr + len, &prev, &rb_link, &rb_parent)) 33016597d783SHugh Dickins return NULL; /* should never get here */ 33021da177e4SLinus Torvalds new_vma = vma_merge(mm, prev, addr, addr + len, vma->vm_flags, 330319a809afSAndrea Arcangeli vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma), 330419a809afSAndrea Arcangeli vma->vm_userfaultfd_ctx); 33051da177e4SLinus Torvalds if (new_vma) { 33061da177e4SLinus Torvalds /* 33071da177e4SLinus Torvalds * Source vma may have been merged into new_vma 33081da177e4SLinus Torvalds */ 3309948f017bSAndrea Arcangeli if (unlikely(vma_start >= new_vma->vm_start && 3310948f017bSAndrea Arcangeli vma_start < new_vma->vm_end)) { 3311948f017bSAndrea Arcangeli /* 3312948f017bSAndrea Arcangeli * The only way we can get a vma_merge with 3313948f017bSAndrea Arcangeli * self during an mremap is if the vma hasn't 3314948f017bSAndrea Arcangeli * been faulted in yet and we were allowed to 3315948f017bSAndrea Arcangeli * reset the dst vma->vm_pgoff to the 3316948f017bSAndrea Arcangeli * destination address of the mremap to allow 3317948f017bSAndrea Arcangeli * the merge to happen. mremap must change the 3318948f017bSAndrea Arcangeli * vm_pgoff linearity between src and dst vmas 3319948f017bSAndrea Arcangeli * (in turn preventing a vma_merge) to be 3320948f017bSAndrea Arcangeli * safe. It is only safe to keep the vm_pgoff 3321948f017bSAndrea Arcangeli * linear if there are no pages mapped yet. 3322948f017bSAndrea Arcangeli */ 332381d1b09cSSasha Levin VM_BUG_ON_VMA(faulted_in_anon_vma, new_vma); 332438a76013SMichel Lespinasse *vmap = vma = new_vma; 3325108d6642SMichel Lespinasse } 332638a76013SMichel Lespinasse *need_rmap_locks = (new_vma->vm_pgoff <= vma->vm_pgoff); 33271da177e4SLinus Torvalds } else { 33283928d4f5SLinus Torvalds new_vma = vm_area_dup(vma); 3329e3975891SChen Gang if (!new_vma) 3330e3975891SChen Gang goto out; 33311da177e4SLinus Torvalds new_vma->vm_start = addr; 33321da177e4SLinus Torvalds new_vma->vm_end = addr + len; 33331da177e4SLinus Torvalds new_vma->vm_pgoff = pgoff; 3334ef0855d3SOleg Nesterov if (vma_dup_policy(vma, new_vma)) 3335523d4e20SMichel Lespinasse goto out_free_vma; 3336523d4e20SMichel Lespinasse if (anon_vma_clone(new_vma, vma)) 3337523d4e20SMichel Lespinasse goto out_free_mempol; 3338e9714acfSKonstantin Khlebnikov if (new_vma->vm_file) 33391da177e4SLinus Torvalds get_file(new_vma->vm_file); 33401da177e4SLinus Torvalds if (new_vma->vm_ops && new_vma->vm_ops->open) 33411da177e4SLinus Torvalds new_vma->vm_ops->open(new_vma); 33421da177e4SLinus Torvalds vma_link(mm, new_vma, prev, rb_link, rb_parent); 334338a76013SMichel Lespinasse *need_rmap_locks = false; 33441da177e4SLinus Torvalds } 33451da177e4SLinus Torvalds return new_vma; 33465beb4930SRik van Riel 33475beb4930SRik van Riel out_free_mempol: 3348ef0855d3SOleg Nesterov mpol_put(vma_policy(new_vma)); 33495beb4930SRik van Riel out_free_vma: 33503928d4f5SLinus Torvalds vm_area_free(new_vma); 3351e3975891SChen Gang out: 33525beb4930SRik van Riel return NULL; 33531da177e4SLinus Torvalds } 3354119f657cSakpm@osdl.org 3355119f657cSakpm@osdl.org /* 3356119f657cSakpm@osdl.org * Return true if the calling process may expand its vm space by the passed 3357119f657cSakpm@osdl.org * number of pages 3358119f657cSakpm@osdl.org */ 335984638335SKonstantin Khlebnikov bool may_expand_vm(struct mm_struct *mm, vm_flags_t flags, unsigned long npages) 3360119f657cSakpm@osdl.org { 336184638335SKonstantin Khlebnikov if (mm->total_vm + npages > rlimit(RLIMIT_AS) >> PAGE_SHIFT) 336284638335SKonstantin Khlebnikov return false; 3363119f657cSakpm@osdl.org 3364d977d56cSKonstantin Khlebnikov if (is_data_mapping(flags) && 3365d977d56cSKonstantin Khlebnikov mm->data_vm + npages > rlimit(RLIMIT_DATA) >> PAGE_SHIFT) { 3366f4fcd558SKonstantin Khlebnikov /* Workaround for Valgrind */ 3367f4fcd558SKonstantin Khlebnikov if (rlimit(RLIMIT_DATA) == 0 && 3368f4fcd558SKonstantin Khlebnikov mm->data_vm + npages <= rlimit_max(RLIMIT_DATA) >> PAGE_SHIFT) 3369f4fcd558SKonstantin Khlebnikov return true; 337057a7702bSDavid Woodhouse 337157a7702bSDavid Woodhouse pr_warn_once("%s (%d): VmData %lu exceed data ulimit %lu. Update limits%s.\n", 3372d977d56cSKonstantin Khlebnikov current->comm, current->pid, 3373d977d56cSKonstantin Khlebnikov (mm->data_vm + npages) << PAGE_SHIFT, 337457a7702bSDavid Woodhouse rlimit(RLIMIT_DATA), 337557a7702bSDavid Woodhouse ignore_rlimit_data ? "" : " or use boot option ignore_rlimit_data"); 337657a7702bSDavid Woodhouse 337757a7702bSDavid Woodhouse if (!ignore_rlimit_data) 3378d977d56cSKonstantin Khlebnikov return false; 3379d977d56cSKonstantin Khlebnikov } 3380119f657cSakpm@osdl.org 338184638335SKonstantin Khlebnikov return true; 338284638335SKonstantin Khlebnikov } 338384638335SKonstantin Khlebnikov 338484638335SKonstantin Khlebnikov void vm_stat_account(struct mm_struct *mm, vm_flags_t flags, long npages) 338584638335SKonstantin Khlebnikov { 338684638335SKonstantin Khlebnikov mm->total_vm += npages; 338784638335SKonstantin Khlebnikov 3388d977d56cSKonstantin Khlebnikov if (is_exec_mapping(flags)) 338984638335SKonstantin Khlebnikov mm->exec_vm += npages; 3390d977d56cSKonstantin Khlebnikov else if (is_stack_mapping(flags)) 339184638335SKonstantin Khlebnikov mm->stack_vm += npages; 3392d977d56cSKonstantin Khlebnikov else if (is_data_mapping(flags)) 339384638335SKonstantin Khlebnikov mm->data_vm += npages; 3394119f657cSakpm@osdl.org } 3395fa5dc22fSRoland McGrath 3396b3ec9f33SSouptick Joarder static vm_fault_t special_mapping_fault(struct vm_fault *vmf); 3397a62c34bdSAndy Lutomirski 3398a62c34bdSAndy Lutomirski /* 3399a62c34bdSAndy Lutomirski * Having a close hook prevents vma merging regardless of flags. 3400a62c34bdSAndy Lutomirski */ 3401a62c34bdSAndy Lutomirski static void special_mapping_close(struct vm_area_struct *vma) 3402a62c34bdSAndy Lutomirski { 3403a62c34bdSAndy Lutomirski } 3404a62c34bdSAndy Lutomirski 3405a62c34bdSAndy Lutomirski static const char *special_mapping_name(struct vm_area_struct *vma) 3406a62c34bdSAndy Lutomirski { 3407a62c34bdSAndy Lutomirski return ((struct vm_special_mapping *)vma->vm_private_data)->name; 3408a62c34bdSAndy Lutomirski } 3409a62c34bdSAndy Lutomirski 3410b059a453SDmitry Safonov static int special_mapping_mremap(struct vm_area_struct *new_vma) 3411b059a453SDmitry Safonov { 3412b059a453SDmitry Safonov struct vm_special_mapping *sm = new_vma->vm_private_data; 3413b059a453SDmitry Safonov 3414280e87e9SDmitry Safonov if (WARN_ON_ONCE(current->mm != new_vma->vm_mm)) 3415280e87e9SDmitry Safonov return -EFAULT; 3416280e87e9SDmitry Safonov 3417b059a453SDmitry Safonov if (sm->mremap) 3418b059a453SDmitry Safonov return sm->mremap(sm, new_vma); 3419280e87e9SDmitry Safonov 3420b059a453SDmitry Safonov return 0; 3421b059a453SDmitry Safonov } 3422b059a453SDmitry Safonov 3423a62c34bdSAndy Lutomirski static const struct vm_operations_struct special_mapping_vmops = { 3424a62c34bdSAndy Lutomirski .close = special_mapping_close, 3425a62c34bdSAndy Lutomirski .fault = special_mapping_fault, 3426b059a453SDmitry Safonov .mremap = special_mapping_mremap, 3427a62c34bdSAndy Lutomirski .name = special_mapping_name, 3428af34ebebSDmitry Safonov /* vDSO code relies that VVAR can't be accessed remotely */ 3429af34ebebSDmitry Safonov .access = NULL, 3430a62c34bdSAndy Lutomirski }; 3431a62c34bdSAndy Lutomirski 3432a62c34bdSAndy Lutomirski static const struct vm_operations_struct legacy_special_mapping_vmops = { 3433a62c34bdSAndy Lutomirski .close = special_mapping_close, 3434a62c34bdSAndy Lutomirski .fault = special_mapping_fault, 3435a62c34bdSAndy Lutomirski }; 3436fa5dc22fSRoland McGrath 3437b3ec9f33SSouptick Joarder static vm_fault_t special_mapping_fault(struct vm_fault *vmf) 3438fa5dc22fSRoland McGrath { 343911bac800SDave Jiang struct vm_area_struct *vma = vmf->vma; 3440b1d0e4f5SNick Piggin pgoff_t pgoff; 3441fa5dc22fSRoland McGrath struct page **pages; 3442fa5dc22fSRoland McGrath 3443f872f540SAndy Lutomirski if (vma->vm_ops == &legacy_special_mapping_vmops) { 3444a62c34bdSAndy Lutomirski pages = vma->vm_private_data; 3445f872f540SAndy Lutomirski } else { 3446f872f540SAndy Lutomirski struct vm_special_mapping *sm = vma->vm_private_data; 3447f872f540SAndy Lutomirski 3448f872f540SAndy Lutomirski if (sm->fault) 344911bac800SDave Jiang return sm->fault(sm, vmf->vma, vmf); 3450f872f540SAndy Lutomirski 3451f872f540SAndy Lutomirski pages = sm->pages; 3452f872f540SAndy Lutomirski } 3453a62c34bdSAndy Lutomirski 34548a9cc3b5SOleg Nesterov for (pgoff = vmf->pgoff; pgoff && *pages; ++pages) 3455b1d0e4f5SNick Piggin pgoff--; 3456fa5dc22fSRoland McGrath 3457fa5dc22fSRoland McGrath if (*pages) { 3458fa5dc22fSRoland McGrath struct page *page = *pages; 3459fa5dc22fSRoland McGrath get_page(page); 3460b1d0e4f5SNick Piggin vmf->page = page; 3461b1d0e4f5SNick Piggin return 0; 3462fa5dc22fSRoland McGrath } 3463fa5dc22fSRoland McGrath 3464b1d0e4f5SNick Piggin return VM_FAULT_SIGBUS; 3465fa5dc22fSRoland McGrath } 3466fa5dc22fSRoland McGrath 3467a62c34bdSAndy Lutomirski static struct vm_area_struct *__install_special_mapping( 3468a62c34bdSAndy Lutomirski struct mm_struct *mm, 3469fa5dc22fSRoland McGrath unsigned long addr, unsigned long len, 347027f28b97SChen Gang unsigned long vm_flags, void *priv, 347127f28b97SChen Gang const struct vm_operations_struct *ops) 3472fa5dc22fSRoland McGrath { 3473462e635eSTavis Ormandy int ret; 3474fa5dc22fSRoland McGrath struct vm_area_struct *vma; 3475fa5dc22fSRoland McGrath 3476490fc053SLinus Torvalds vma = vm_area_alloc(mm); 3477fa5dc22fSRoland McGrath if (unlikely(vma == NULL)) 34783935ed6aSStefani Seibold return ERR_PTR(-ENOMEM); 3479fa5dc22fSRoland McGrath 3480fa5dc22fSRoland McGrath vma->vm_start = addr; 3481fa5dc22fSRoland McGrath vma->vm_end = addr + len; 3482fa5dc22fSRoland McGrath 3483d9104d1cSCyrill Gorcunov vma->vm_flags = vm_flags | mm->def_flags | VM_DONTEXPAND | VM_SOFTDIRTY; 34843ed75eb8SColy Li vma->vm_page_prot = vm_get_page_prot(vma->vm_flags); 3485fa5dc22fSRoland McGrath 3486a62c34bdSAndy Lutomirski vma->vm_ops = ops; 3487a62c34bdSAndy Lutomirski vma->vm_private_data = priv; 3488fa5dc22fSRoland McGrath 3489462e635eSTavis Ormandy ret = insert_vm_struct(mm, vma); 3490462e635eSTavis Ormandy if (ret) 3491462e635eSTavis Ormandy goto out; 3492fa5dc22fSRoland McGrath 349384638335SKonstantin Khlebnikov vm_stat_account(mm, vma->vm_flags, len >> PAGE_SHIFT); 3494fa5dc22fSRoland McGrath 3495cdd6c482SIngo Molnar perf_event_mmap(vma); 3496089dd79dSPeter Zijlstra 34973935ed6aSStefani Seibold return vma; 3498462e635eSTavis Ormandy 3499462e635eSTavis Ormandy out: 35003928d4f5SLinus Torvalds vm_area_free(vma); 35013935ed6aSStefani Seibold return ERR_PTR(ret); 35023935ed6aSStefani Seibold } 35033935ed6aSStefani Seibold 35042eefd878SDmitry Safonov bool vma_is_special_mapping(const struct vm_area_struct *vma, 35052eefd878SDmitry Safonov const struct vm_special_mapping *sm) 35062eefd878SDmitry Safonov { 35072eefd878SDmitry Safonov return vma->vm_private_data == sm && 35082eefd878SDmitry Safonov (vma->vm_ops == &special_mapping_vmops || 35092eefd878SDmitry Safonov vma->vm_ops == &legacy_special_mapping_vmops); 35102eefd878SDmitry Safonov } 35112eefd878SDmitry Safonov 3512a62c34bdSAndy Lutomirski /* 3513c1e8d7c6SMichel Lespinasse * Called with mm->mmap_lock held for writing. 3514a62c34bdSAndy Lutomirski * Insert a new vma covering the given region, with the given flags. 3515a62c34bdSAndy Lutomirski * Its pages are supplied by the given array of struct page *. 3516a62c34bdSAndy Lutomirski * The array can be shorter than len >> PAGE_SHIFT if it's null-terminated. 3517a62c34bdSAndy Lutomirski * The region past the last page supplied will always produce SIGBUS. 3518a62c34bdSAndy Lutomirski * The array pointer and the pages it points to are assumed to stay alive 3519a62c34bdSAndy Lutomirski * for as long as this mapping might exist. 3520a62c34bdSAndy Lutomirski */ 3521a62c34bdSAndy Lutomirski struct vm_area_struct *_install_special_mapping( 3522a62c34bdSAndy Lutomirski struct mm_struct *mm, 3523a62c34bdSAndy Lutomirski unsigned long addr, unsigned long len, 3524a62c34bdSAndy Lutomirski unsigned long vm_flags, const struct vm_special_mapping *spec) 3525a62c34bdSAndy Lutomirski { 352627f28b97SChen Gang return __install_special_mapping(mm, addr, len, vm_flags, (void *)spec, 352727f28b97SChen Gang &special_mapping_vmops); 3528a62c34bdSAndy Lutomirski } 3529a62c34bdSAndy Lutomirski 35303935ed6aSStefani Seibold int install_special_mapping(struct mm_struct *mm, 35313935ed6aSStefani Seibold unsigned long addr, unsigned long len, 35323935ed6aSStefani Seibold unsigned long vm_flags, struct page **pages) 35333935ed6aSStefani Seibold { 3534a62c34bdSAndy Lutomirski struct vm_area_struct *vma = __install_special_mapping( 353527f28b97SChen Gang mm, addr, len, vm_flags, (void *)pages, 353627f28b97SChen Gang &legacy_special_mapping_vmops); 35373935ed6aSStefani Seibold 353814bd5b45SDuan Jiong return PTR_ERR_OR_ZERO(vma); 3539fa5dc22fSRoland McGrath } 35407906d00cSAndrea Arcangeli 35417906d00cSAndrea Arcangeli static DEFINE_MUTEX(mm_all_locks_mutex); 35427906d00cSAndrea Arcangeli 3543454ed842SPeter Zijlstra static void vm_lock_anon_vma(struct mm_struct *mm, struct anon_vma *anon_vma) 35447906d00cSAndrea Arcangeli { 3545f808c13fSDavidlohr Bueso if (!test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_root.rb_node)) { 35467906d00cSAndrea Arcangeli /* 35477906d00cSAndrea Arcangeli * The LSB of head.next can't change from under us 35487906d00cSAndrea Arcangeli * because we hold the mm_all_locks_mutex. 35497906d00cSAndrea Arcangeli */ 3550da1c55f1SMichel Lespinasse down_write_nest_lock(&anon_vma->root->rwsem, &mm->mmap_lock); 35517906d00cSAndrea Arcangeli /* 35527906d00cSAndrea Arcangeli * We can safely modify head.next after taking the 35535a505085SIngo Molnar * anon_vma->root->rwsem. If some other vma in this mm shares 35547906d00cSAndrea Arcangeli * the same anon_vma we won't take it again. 35557906d00cSAndrea Arcangeli * 35567906d00cSAndrea Arcangeli * No need of atomic instructions here, head.next 35577906d00cSAndrea Arcangeli * can't change from under us thanks to the 35585a505085SIngo Molnar * anon_vma->root->rwsem. 35597906d00cSAndrea Arcangeli */ 35607906d00cSAndrea Arcangeli if (__test_and_set_bit(0, (unsigned long *) 3561f808c13fSDavidlohr Bueso &anon_vma->root->rb_root.rb_root.rb_node)) 35627906d00cSAndrea Arcangeli BUG(); 35637906d00cSAndrea Arcangeli } 35647906d00cSAndrea Arcangeli } 35657906d00cSAndrea Arcangeli 3566454ed842SPeter Zijlstra static void vm_lock_mapping(struct mm_struct *mm, struct address_space *mapping) 35677906d00cSAndrea Arcangeli { 35687906d00cSAndrea Arcangeli if (!test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) { 35697906d00cSAndrea Arcangeli /* 35707906d00cSAndrea Arcangeli * AS_MM_ALL_LOCKS can't change from under us because 35717906d00cSAndrea Arcangeli * we hold the mm_all_locks_mutex. 35727906d00cSAndrea Arcangeli * 35737906d00cSAndrea Arcangeli * Operations on ->flags have to be atomic because 35747906d00cSAndrea Arcangeli * even if AS_MM_ALL_LOCKS is stable thanks to the 35757906d00cSAndrea Arcangeli * mm_all_locks_mutex, there may be other cpus 35767906d00cSAndrea Arcangeli * changing other bitflags in parallel to us. 35777906d00cSAndrea Arcangeli */ 35787906d00cSAndrea Arcangeli if (test_and_set_bit(AS_MM_ALL_LOCKS, &mapping->flags)) 35797906d00cSAndrea Arcangeli BUG(); 3580da1c55f1SMichel Lespinasse down_write_nest_lock(&mapping->i_mmap_rwsem, &mm->mmap_lock); 35817906d00cSAndrea Arcangeli } 35827906d00cSAndrea Arcangeli } 35837906d00cSAndrea Arcangeli 35847906d00cSAndrea Arcangeli /* 35857906d00cSAndrea Arcangeli * This operation locks against the VM for all pte/vma/mm related 35867906d00cSAndrea Arcangeli * operations that could ever happen on a certain mm. This includes 35877906d00cSAndrea Arcangeli * vmtruncate, try_to_unmap, and all page faults. 35887906d00cSAndrea Arcangeli * 3589c1e8d7c6SMichel Lespinasse * The caller must take the mmap_lock in write mode before calling 35907906d00cSAndrea Arcangeli * mm_take_all_locks(). The caller isn't allowed to release the 3591c1e8d7c6SMichel Lespinasse * mmap_lock until mm_drop_all_locks() returns. 35927906d00cSAndrea Arcangeli * 3593c1e8d7c6SMichel Lespinasse * mmap_lock in write mode is required in order to block all operations 35947906d00cSAndrea Arcangeli * that could modify pagetables and free pages without need of 359527ba0644SKirill A. Shutemov * altering the vma layout. It's also needed in write mode to avoid new 35967906d00cSAndrea Arcangeli * anon_vmas to be associated with existing vmas. 35977906d00cSAndrea Arcangeli * 35987906d00cSAndrea Arcangeli * A single task can't take more than one mm_take_all_locks() in a row 35997906d00cSAndrea Arcangeli * or it would deadlock. 36007906d00cSAndrea Arcangeli * 3601bf181b9fSMichel Lespinasse * The LSB in anon_vma->rb_root.rb_node and the AS_MM_ALL_LOCKS bitflag in 36027906d00cSAndrea Arcangeli * mapping->flags avoid to take the same lock twice, if more than one 36037906d00cSAndrea Arcangeli * vma in this mm is backed by the same anon_vma or address_space. 36047906d00cSAndrea Arcangeli * 360588f306b6SKirill A. Shutemov * We take locks in following order, accordingly to comment at beginning 360688f306b6SKirill A. Shutemov * of mm/rmap.c: 360788f306b6SKirill A. Shutemov * - all hugetlbfs_i_mmap_rwsem_key locks (aka mapping->i_mmap_rwsem for 360888f306b6SKirill A. Shutemov * hugetlb mapping); 360988f306b6SKirill A. Shutemov * - all i_mmap_rwsem locks; 361088f306b6SKirill A. Shutemov * - all anon_vma->rwseml 361188f306b6SKirill A. Shutemov * 361288f306b6SKirill A. Shutemov * We can take all locks within these types randomly because the VM code 361388f306b6SKirill A. Shutemov * doesn't nest them and we protected from parallel mm_take_all_locks() by 361488f306b6SKirill A. Shutemov * mm_all_locks_mutex. 36157906d00cSAndrea Arcangeli * 36167906d00cSAndrea Arcangeli * mm_take_all_locks() and mm_drop_all_locks are expensive operations 36177906d00cSAndrea Arcangeli * that may have to take thousand of locks. 36187906d00cSAndrea Arcangeli * 36197906d00cSAndrea Arcangeli * mm_take_all_locks() can fail if it's interrupted by signals. 36207906d00cSAndrea Arcangeli */ 36217906d00cSAndrea Arcangeli int mm_take_all_locks(struct mm_struct *mm) 36227906d00cSAndrea Arcangeli { 36237906d00cSAndrea Arcangeli struct vm_area_struct *vma; 36245beb4930SRik van Riel struct anon_vma_chain *avc; 36257906d00cSAndrea Arcangeli 3626d8ed45c5SMichel Lespinasse BUG_ON(mmap_read_trylock(mm)); 36277906d00cSAndrea Arcangeli 36287906d00cSAndrea Arcangeli mutex_lock(&mm_all_locks_mutex); 36297906d00cSAndrea Arcangeli 36307906d00cSAndrea Arcangeli for (vma = mm->mmap; vma; vma = vma->vm_next) { 36317906d00cSAndrea Arcangeli if (signal_pending(current)) 36327906d00cSAndrea Arcangeli goto out_unlock; 363388f306b6SKirill A. Shutemov if (vma->vm_file && vma->vm_file->f_mapping && 363488f306b6SKirill A. Shutemov is_vm_hugetlb_page(vma)) 363588f306b6SKirill A. Shutemov vm_lock_mapping(mm, vma->vm_file->f_mapping); 363688f306b6SKirill A. Shutemov } 363788f306b6SKirill A. Shutemov 363888f306b6SKirill A. Shutemov for (vma = mm->mmap; vma; vma = vma->vm_next) { 363988f306b6SKirill A. Shutemov if (signal_pending(current)) 364088f306b6SKirill A. Shutemov goto out_unlock; 364188f306b6SKirill A. Shutemov if (vma->vm_file && vma->vm_file->f_mapping && 364288f306b6SKirill A. Shutemov !is_vm_hugetlb_page(vma)) 3643454ed842SPeter Zijlstra vm_lock_mapping(mm, vma->vm_file->f_mapping); 36447906d00cSAndrea Arcangeli } 36457cd5a02fSPeter Zijlstra 36467cd5a02fSPeter Zijlstra for (vma = mm->mmap; vma; vma = vma->vm_next) { 36477cd5a02fSPeter Zijlstra if (signal_pending(current)) 36487cd5a02fSPeter Zijlstra goto out_unlock; 36497cd5a02fSPeter Zijlstra if (vma->anon_vma) 36505beb4930SRik van Riel list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) 36515beb4930SRik van Riel vm_lock_anon_vma(mm, avc->anon_vma); 36527cd5a02fSPeter Zijlstra } 36537cd5a02fSPeter Zijlstra 3654584cff54SKautuk Consul return 0; 36557906d00cSAndrea Arcangeli 36567906d00cSAndrea Arcangeli out_unlock: 36577906d00cSAndrea Arcangeli mm_drop_all_locks(mm); 3658584cff54SKautuk Consul return -EINTR; 36597906d00cSAndrea Arcangeli } 36607906d00cSAndrea Arcangeli 36617906d00cSAndrea Arcangeli static void vm_unlock_anon_vma(struct anon_vma *anon_vma) 36627906d00cSAndrea Arcangeli { 3663f808c13fSDavidlohr Bueso if (test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_root.rb_node)) { 36647906d00cSAndrea Arcangeli /* 36657906d00cSAndrea Arcangeli * The LSB of head.next can't change to 0 from under 36667906d00cSAndrea Arcangeli * us because we hold the mm_all_locks_mutex. 36677906d00cSAndrea Arcangeli * 36687906d00cSAndrea Arcangeli * We must however clear the bitflag before unlocking 3669bf181b9fSMichel Lespinasse * the vma so the users using the anon_vma->rb_root will 36707906d00cSAndrea Arcangeli * never see our bitflag. 36717906d00cSAndrea Arcangeli * 36727906d00cSAndrea Arcangeli * No need of atomic instructions here, head.next 36737906d00cSAndrea Arcangeli * can't change from under us until we release the 36745a505085SIngo Molnar * anon_vma->root->rwsem. 36757906d00cSAndrea Arcangeli */ 36767906d00cSAndrea Arcangeli if (!__test_and_clear_bit(0, (unsigned long *) 3677f808c13fSDavidlohr Bueso &anon_vma->root->rb_root.rb_root.rb_node)) 36787906d00cSAndrea Arcangeli BUG(); 367908b52706SKonstantin Khlebnikov anon_vma_unlock_write(anon_vma); 36807906d00cSAndrea Arcangeli } 36817906d00cSAndrea Arcangeli } 36827906d00cSAndrea Arcangeli 36837906d00cSAndrea Arcangeli static void vm_unlock_mapping(struct address_space *mapping) 36847906d00cSAndrea Arcangeli { 36857906d00cSAndrea Arcangeli if (test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) { 36867906d00cSAndrea Arcangeli /* 36877906d00cSAndrea Arcangeli * AS_MM_ALL_LOCKS can't change to 0 from under us 36887906d00cSAndrea Arcangeli * because we hold the mm_all_locks_mutex. 36897906d00cSAndrea Arcangeli */ 369083cde9e8SDavidlohr Bueso i_mmap_unlock_write(mapping); 36917906d00cSAndrea Arcangeli if (!test_and_clear_bit(AS_MM_ALL_LOCKS, 36927906d00cSAndrea Arcangeli &mapping->flags)) 36937906d00cSAndrea Arcangeli BUG(); 36947906d00cSAndrea Arcangeli } 36957906d00cSAndrea Arcangeli } 36967906d00cSAndrea Arcangeli 36977906d00cSAndrea Arcangeli /* 3698c1e8d7c6SMichel Lespinasse * The mmap_lock cannot be released by the caller until 36997906d00cSAndrea Arcangeli * mm_drop_all_locks() returns. 37007906d00cSAndrea Arcangeli */ 37017906d00cSAndrea Arcangeli void mm_drop_all_locks(struct mm_struct *mm) 37027906d00cSAndrea Arcangeli { 37037906d00cSAndrea Arcangeli struct vm_area_struct *vma; 37045beb4930SRik van Riel struct anon_vma_chain *avc; 37057906d00cSAndrea Arcangeli 3706d8ed45c5SMichel Lespinasse BUG_ON(mmap_read_trylock(mm)); 37077906d00cSAndrea Arcangeli BUG_ON(!mutex_is_locked(&mm_all_locks_mutex)); 37087906d00cSAndrea Arcangeli 37097906d00cSAndrea Arcangeli for (vma = mm->mmap; vma; vma = vma->vm_next) { 37107906d00cSAndrea Arcangeli if (vma->anon_vma) 37115beb4930SRik van Riel list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) 37125beb4930SRik van Riel vm_unlock_anon_vma(avc->anon_vma); 37137906d00cSAndrea Arcangeli if (vma->vm_file && vma->vm_file->f_mapping) 37147906d00cSAndrea Arcangeli vm_unlock_mapping(vma->vm_file->f_mapping); 37157906d00cSAndrea Arcangeli } 37167906d00cSAndrea Arcangeli 37177906d00cSAndrea Arcangeli mutex_unlock(&mm_all_locks_mutex); 37187906d00cSAndrea Arcangeli } 37198feae131SDavid Howells 37208feae131SDavid Howells /* 37213edf41d8Sseokhoon.yoon * initialise the percpu counter for VM 37228feae131SDavid Howells */ 37238feae131SDavid Howells void __init mmap_init(void) 37248feae131SDavid Howells { 372500a62ce9SKOSAKI Motohiro int ret; 372600a62ce9SKOSAKI Motohiro 3727908c7f19STejun Heo ret = percpu_counter_init(&vm_committed_as, 0, GFP_KERNEL); 372800a62ce9SKOSAKI Motohiro VM_BUG_ON(ret); 37298feae131SDavid Howells } 3730c9b1d098SAndrew Shewmaker 3731c9b1d098SAndrew Shewmaker /* 3732c9b1d098SAndrew Shewmaker * Initialise sysctl_user_reserve_kbytes. 3733c9b1d098SAndrew Shewmaker * 3734c9b1d098SAndrew Shewmaker * This is intended to prevent a user from starting a single memory hogging 3735c9b1d098SAndrew Shewmaker * process, such that they cannot recover (kill the hog) in OVERCOMMIT_NEVER 3736c9b1d098SAndrew Shewmaker * mode. 3737c9b1d098SAndrew Shewmaker * 3738c9b1d098SAndrew Shewmaker * The default value is min(3% of free memory, 128MB) 3739c9b1d098SAndrew Shewmaker * 128MB is enough to recover with sshd/login, bash, and top/kill. 3740c9b1d098SAndrew Shewmaker */ 37411640879aSAndrew Shewmaker static int init_user_reserve(void) 3742c9b1d098SAndrew Shewmaker { 3743c9b1d098SAndrew Shewmaker unsigned long free_kbytes; 3744c9b1d098SAndrew Shewmaker 3745c41f012aSMichal Hocko free_kbytes = global_zone_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10); 3746c9b1d098SAndrew Shewmaker 3747c9b1d098SAndrew Shewmaker sysctl_user_reserve_kbytes = min(free_kbytes / 32, 1UL << 17); 3748c9b1d098SAndrew Shewmaker return 0; 3749c9b1d098SAndrew Shewmaker } 3750a64fb3cdSPaul Gortmaker subsys_initcall(init_user_reserve); 37514eeab4f5SAndrew Shewmaker 37524eeab4f5SAndrew Shewmaker /* 37534eeab4f5SAndrew Shewmaker * Initialise sysctl_admin_reserve_kbytes. 37544eeab4f5SAndrew Shewmaker * 37554eeab4f5SAndrew Shewmaker * The purpose of sysctl_admin_reserve_kbytes is to allow the sys admin 37564eeab4f5SAndrew Shewmaker * to log in and kill a memory hogging process. 37574eeab4f5SAndrew Shewmaker * 37584eeab4f5SAndrew Shewmaker * Systems with more than 256MB will reserve 8MB, enough to recover 37594eeab4f5SAndrew Shewmaker * with sshd, bash, and top in OVERCOMMIT_GUESS. Smaller systems will 37604eeab4f5SAndrew Shewmaker * only reserve 3% of free pages by default. 37614eeab4f5SAndrew Shewmaker */ 37621640879aSAndrew Shewmaker static int init_admin_reserve(void) 37634eeab4f5SAndrew Shewmaker { 37644eeab4f5SAndrew Shewmaker unsigned long free_kbytes; 37654eeab4f5SAndrew Shewmaker 3766c41f012aSMichal Hocko free_kbytes = global_zone_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10); 37674eeab4f5SAndrew Shewmaker 37684eeab4f5SAndrew Shewmaker sysctl_admin_reserve_kbytes = min(free_kbytes / 32, 1UL << 13); 37694eeab4f5SAndrew Shewmaker return 0; 37704eeab4f5SAndrew Shewmaker } 3771a64fb3cdSPaul Gortmaker subsys_initcall(init_admin_reserve); 37721640879aSAndrew Shewmaker 37731640879aSAndrew Shewmaker /* 37741640879aSAndrew Shewmaker * Reinititalise user and admin reserves if memory is added or removed. 37751640879aSAndrew Shewmaker * 37761640879aSAndrew Shewmaker * The default user reserve max is 128MB, and the default max for the 37771640879aSAndrew Shewmaker * admin reserve is 8MB. These are usually, but not always, enough to 37781640879aSAndrew Shewmaker * enable recovery from a memory hogging process using login/sshd, a shell, 37791640879aSAndrew Shewmaker * and tools like top. It may make sense to increase or even disable the 37801640879aSAndrew Shewmaker * reserve depending on the existence of swap or variations in the recovery 37811640879aSAndrew Shewmaker * tools. So, the admin may have changed them. 37821640879aSAndrew Shewmaker * 37831640879aSAndrew Shewmaker * If memory is added and the reserves have been eliminated or increased above 37841640879aSAndrew Shewmaker * the default max, then we'll trust the admin. 37851640879aSAndrew Shewmaker * 37861640879aSAndrew Shewmaker * If memory is removed and there isn't enough free memory, then we 37871640879aSAndrew Shewmaker * need to reset the reserves. 37881640879aSAndrew Shewmaker * 37891640879aSAndrew Shewmaker * Otherwise keep the reserve set by the admin. 37901640879aSAndrew Shewmaker */ 37911640879aSAndrew Shewmaker static int reserve_mem_notifier(struct notifier_block *nb, 37921640879aSAndrew Shewmaker unsigned long action, void *data) 37931640879aSAndrew Shewmaker { 37941640879aSAndrew Shewmaker unsigned long tmp, free_kbytes; 37951640879aSAndrew Shewmaker 37961640879aSAndrew Shewmaker switch (action) { 37971640879aSAndrew Shewmaker case MEM_ONLINE: 37981640879aSAndrew Shewmaker /* Default max is 128MB. Leave alone if modified by operator. */ 37991640879aSAndrew Shewmaker tmp = sysctl_user_reserve_kbytes; 38001640879aSAndrew Shewmaker if (0 < tmp && tmp < (1UL << 17)) 38011640879aSAndrew Shewmaker init_user_reserve(); 38021640879aSAndrew Shewmaker 38031640879aSAndrew Shewmaker /* Default max is 8MB. Leave alone if modified by operator. */ 38041640879aSAndrew Shewmaker tmp = sysctl_admin_reserve_kbytes; 38051640879aSAndrew Shewmaker if (0 < tmp && tmp < (1UL << 13)) 38061640879aSAndrew Shewmaker init_admin_reserve(); 38071640879aSAndrew Shewmaker 38081640879aSAndrew Shewmaker break; 38091640879aSAndrew Shewmaker case MEM_OFFLINE: 3810c41f012aSMichal Hocko free_kbytes = global_zone_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10); 38111640879aSAndrew Shewmaker 38121640879aSAndrew Shewmaker if (sysctl_user_reserve_kbytes > free_kbytes) { 38131640879aSAndrew Shewmaker init_user_reserve(); 38141640879aSAndrew Shewmaker pr_info("vm.user_reserve_kbytes reset to %lu\n", 38151640879aSAndrew Shewmaker sysctl_user_reserve_kbytes); 38161640879aSAndrew Shewmaker } 38171640879aSAndrew Shewmaker 38181640879aSAndrew Shewmaker if (sysctl_admin_reserve_kbytes > free_kbytes) { 38191640879aSAndrew Shewmaker init_admin_reserve(); 38201640879aSAndrew Shewmaker pr_info("vm.admin_reserve_kbytes reset to %lu\n", 38211640879aSAndrew Shewmaker sysctl_admin_reserve_kbytes); 38221640879aSAndrew Shewmaker } 38231640879aSAndrew Shewmaker break; 38241640879aSAndrew Shewmaker default: 38251640879aSAndrew Shewmaker break; 38261640879aSAndrew Shewmaker } 38271640879aSAndrew Shewmaker return NOTIFY_OK; 38281640879aSAndrew Shewmaker } 38291640879aSAndrew Shewmaker 38301640879aSAndrew Shewmaker static struct notifier_block reserve_mem_nb = { 38311640879aSAndrew Shewmaker .notifier_call = reserve_mem_notifier, 38321640879aSAndrew Shewmaker }; 38331640879aSAndrew Shewmaker 38341640879aSAndrew Shewmaker static int __meminit init_reserve_notifier(void) 38351640879aSAndrew Shewmaker { 38361640879aSAndrew Shewmaker if (register_hotmemory_notifier(&reserve_mem_nb)) 3837b1de0d13SMitchel Humpherys pr_err("Failed registering memory add/remove notifier for admin reserve\n"); 38381640879aSAndrew Shewmaker 38391640879aSAndrew Shewmaker return 0; 38401640879aSAndrew Shewmaker } 3841a64fb3cdSPaul Gortmaker subsys_initcall(init_reserve_notifier); 3842