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) 146496ad9aaSAl Viro atomic_inc(&file_inode(file)->i_writecount); 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, 477*4d1e7243SWei Yang * with the possible exception of 478*4d1e7243SWei Yang * 479*4d1e7243SWei Yang * a. the "next" vma being erased if next->vm_start was reduced in 480*4d1e7243SWei Yang * __vma_adjust() -> __vma_unlink() 481*4d1e7243SWei Yang * b. the vma being erased in detach_vmas_to_be_unmapped() -> 482*4d1e7243SWei 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 { 492*4d1e7243SWei 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 561e8420a8eSCyril Hrubis static unsigned long count_vma_pages_range(struct mm_struct *mm, 562e8420a8eSCyril Hrubis unsigned long addr, unsigned long end) 563e8420a8eSCyril Hrubis { 564e8420a8eSCyril Hrubis unsigned long nr_pages = 0; 565e8420a8eSCyril Hrubis struct vm_area_struct *vma; 566e8420a8eSCyril Hrubis 567e8420a8eSCyril Hrubis /* Find first overlaping mapping */ 568e8420a8eSCyril Hrubis vma = find_vma_intersection(mm, addr, end); 569e8420a8eSCyril Hrubis if (!vma) 570e8420a8eSCyril Hrubis return 0; 571e8420a8eSCyril Hrubis 572e8420a8eSCyril Hrubis nr_pages = (min(end, vma->vm_end) - 573e8420a8eSCyril Hrubis max(addr, vma->vm_start)) >> PAGE_SHIFT; 574e8420a8eSCyril Hrubis 575e8420a8eSCyril Hrubis /* Iterate over the rest of the overlaps */ 576e8420a8eSCyril Hrubis for (vma = vma->vm_next; vma; vma = vma->vm_next) { 577e8420a8eSCyril Hrubis unsigned long overlap_len; 578e8420a8eSCyril Hrubis 579e8420a8eSCyril Hrubis if (vma->vm_start > end) 580e8420a8eSCyril Hrubis break; 581e8420a8eSCyril Hrubis 582e8420a8eSCyril Hrubis overlap_len = min(end, vma->vm_end) - vma->vm_start; 583e8420a8eSCyril Hrubis nr_pages += overlap_len >> PAGE_SHIFT; 584e8420a8eSCyril Hrubis } 585e8420a8eSCyril Hrubis 586e8420a8eSCyril Hrubis return nr_pages; 587e8420a8eSCyril Hrubis } 588e8420a8eSCyril Hrubis 5891da177e4SLinus Torvalds void __vma_link_rb(struct mm_struct *mm, struct vm_area_struct *vma, 5901da177e4SLinus Torvalds struct rb_node **rb_link, struct rb_node *rb_parent) 5911da177e4SLinus Torvalds { 592d3737187SMichel Lespinasse /* Update tracking information for the gap following the new vma. */ 593d3737187SMichel Lespinasse if (vma->vm_next) 594d3737187SMichel Lespinasse vma_gap_update(vma->vm_next); 595d3737187SMichel Lespinasse else 5961be7107fSHugh Dickins mm->highest_vm_end = vm_end_gap(vma); 597d3737187SMichel Lespinasse 598d3737187SMichel Lespinasse /* 599d3737187SMichel Lespinasse * vma->vm_prev wasn't known when we followed the rbtree to find the 600d3737187SMichel Lespinasse * correct insertion point for that vma. As a result, we could not 601d3737187SMichel Lespinasse * update the vma vm_rb parents rb_subtree_gap values on the way down. 602d3737187SMichel Lespinasse * So, we first insert the vma with a zero rb_subtree_gap value 603d3737187SMichel Lespinasse * (to be consistent with what we did on the way down), and then 604d3737187SMichel Lespinasse * immediately update the gap to the correct value. Finally we 605d3737187SMichel Lespinasse * rebalance the rbtree after all augmented values have been set. 606d3737187SMichel Lespinasse */ 6071da177e4SLinus Torvalds rb_link_node(&vma->vm_rb, rb_parent, rb_link); 608d3737187SMichel Lespinasse vma->rb_subtree_gap = 0; 609d3737187SMichel Lespinasse vma_gap_update(vma); 610d3737187SMichel Lespinasse vma_rb_insert(vma, &mm->mm_rb); 6111da177e4SLinus Torvalds } 6121da177e4SLinus Torvalds 613cb8f488cSDenys Vlasenko static void __vma_link_file(struct vm_area_struct *vma) 6141da177e4SLinus Torvalds { 6151da177e4SLinus Torvalds struct file *file; 6161da177e4SLinus Torvalds 6171da177e4SLinus Torvalds file = vma->vm_file; 6181da177e4SLinus Torvalds if (file) { 6191da177e4SLinus Torvalds struct address_space *mapping = file->f_mapping; 6201da177e4SLinus Torvalds 6211da177e4SLinus Torvalds if (vma->vm_flags & VM_DENYWRITE) 622496ad9aaSAl Viro atomic_dec(&file_inode(file)->i_writecount); 6231da177e4SLinus Torvalds if (vma->vm_flags & VM_SHARED) 6244bb5f5d9SDavid Herrmann atomic_inc(&mapping->i_mmap_writable); 6251da177e4SLinus Torvalds 6261da177e4SLinus Torvalds flush_dcache_mmap_lock(mapping); 6276b2dbba8SMichel Lespinasse vma_interval_tree_insert(vma, &mapping->i_mmap); 6281da177e4SLinus Torvalds flush_dcache_mmap_unlock(mapping); 6291da177e4SLinus Torvalds } 6301da177e4SLinus Torvalds } 6311da177e4SLinus Torvalds 6321da177e4SLinus Torvalds static void 6331da177e4SLinus Torvalds __vma_link(struct mm_struct *mm, struct vm_area_struct *vma, 6341da177e4SLinus Torvalds struct vm_area_struct *prev, struct rb_node **rb_link, 6351da177e4SLinus Torvalds struct rb_node *rb_parent) 6361da177e4SLinus Torvalds { 637aba6dfb7SWei Yang __vma_link_list(mm, vma, prev); 6381da177e4SLinus Torvalds __vma_link_rb(mm, vma, rb_link, rb_parent); 6391da177e4SLinus Torvalds } 6401da177e4SLinus Torvalds 6411da177e4SLinus Torvalds static void vma_link(struct mm_struct *mm, struct vm_area_struct *vma, 6421da177e4SLinus Torvalds struct vm_area_struct *prev, struct rb_node **rb_link, 6431da177e4SLinus Torvalds struct rb_node *rb_parent) 6441da177e4SLinus Torvalds { 6451da177e4SLinus Torvalds struct address_space *mapping = NULL; 6461da177e4SLinus Torvalds 64764ac4940SHuang Shijie if (vma->vm_file) { 6481da177e4SLinus Torvalds mapping = vma->vm_file->f_mapping; 64983cde9e8SDavidlohr Bueso i_mmap_lock_write(mapping); 65064ac4940SHuang Shijie } 6511da177e4SLinus Torvalds 6521da177e4SLinus Torvalds __vma_link(mm, vma, prev, rb_link, rb_parent); 6531da177e4SLinus Torvalds __vma_link_file(vma); 6541da177e4SLinus Torvalds 6551da177e4SLinus Torvalds if (mapping) 65683cde9e8SDavidlohr Bueso i_mmap_unlock_write(mapping); 6571da177e4SLinus Torvalds 6581da177e4SLinus Torvalds mm->map_count++; 6591da177e4SLinus Torvalds validate_mm(mm); 6601da177e4SLinus Torvalds } 6611da177e4SLinus Torvalds 6621da177e4SLinus Torvalds /* 66388f6b4c3SKautuk Consul * Helper for vma_adjust() in the split_vma insert case: insert a vma into the 6646b2dbba8SMichel Lespinasse * mm's list and rbtree. It has already been inserted into the interval tree. 6651da177e4SLinus Torvalds */ 66648aae425SZhenwenXu static void __insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma) 6671da177e4SLinus Torvalds { 6686597d783SHugh Dickins struct vm_area_struct *prev; 6691da177e4SLinus Torvalds struct rb_node **rb_link, *rb_parent; 6701da177e4SLinus Torvalds 6716597d783SHugh Dickins if (find_vma_links(mm, vma->vm_start, vma->vm_end, 6726597d783SHugh Dickins &prev, &rb_link, &rb_parent)) 6736597d783SHugh Dickins BUG(); 6741da177e4SLinus Torvalds __vma_link(mm, vma, prev, rb_link, rb_parent); 6751da177e4SLinus Torvalds mm->map_count++; 6761da177e4SLinus Torvalds } 6771da177e4SLinus Torvalds 6787c61f917SWei Yang static __always_inline void __vma_unlink(struct mm_struct *mm, 679e86f15eeSAndrea Arcangeli struct vm_area_struct *vma, 6808f26e0b1SAndrea Arcangeli struct vm_area_struct *ignore) 6811da177e4SLinus Torvalds { 6828f26e0b1SAndrea Arcangeli vma_rb_erase_ignore(vma, &mm->mm_rb, ignore); 6831b9fc5b2SWei Yang __vma_unlink_list(mm, vma); 684615d6e87SDavidlohr Bueso /* Kill the cache */ 685615d6e87SDavidlohr Bueso vmacache_invalidate(mm); 6861da177e4SLinus Torvalds } 6871da177e4SLinus Torvalds 6881da177e4SLinus Torvalds /* 6891da177e4SLinus Torvalds * We cannot adjust vm_start, vm_end, vm_pgoff fields of a vma that 6901da177e4SLinus Torvalds * is already present in an i_mmap tree without adjusting the tree. 6911da177e4SLinus Torvalds * The following helper function should be used when such adjustments 6921da177e4SLinus Torvalds * are necessary. The "insert" vma (if any) is to be inserted 6931da177e4SLinus Torvalds * before we drop the necessary locks. 6941da177e4SLinus Torvalds */ 695e86f15eeSAndrea Arcangeli int __vma_adjust(struct vm_area_struct *vma, unsigned long start, 696e86f15eeSAndrea Arcangeli unsigned long end, pgoff_t pgoff, struct vm_area_struct *insert, 697e86f15eeSAndrea Arcangeli struct vm_area_struct *expand) 6981da177e4SLinus Torvalds { 6991da177e4SLinus Torvalds struct mm_struct *mm = vma->vm_mm; 700e86f15eeSAndrea Arcangeli struct vm_area_struct *next = vma->vm_next, *orig_vma = vma; 7011da177e4SLinus Torvalds struct address_space *mapping = NULL; 702f808c13fSDavidlohr Bueso struct rb_root_cached *root = NULL; 703012f1800SRik van Riel struct anon_vma *anon_vma = NULL; 7041da177e4SLinus Torvalds struct file *file = vma->vm_file; 705d3737187SMichel Lespinasse bool start_changed = false, end_changed = false; 7061da177e4SLinus Torvalds long adjust_next = 0; 7071da177e4SLinus Torvalds int remove_next = 0; 7081da177e4SLinus Torvalds 7091da177e4SLinus Torvalds if (next && !insert) { 710734537c9SKirill A. Shutemov struct vm_area_struct *exporter = NULL, *importer = NULL; 711287d97acSLinus Torvalds 7121da177e4SLinus Torvalds if (end >= next->vm_end) { 7131da177e4SLinus Torvalds /* 7141da177e4SLinus Torvalds * vma expands, overlapping all the next, and 7151da177e4SLinus Torvalds * perhaps the one after too (mprotect case 6). 71686d12e47SAndrea Arcangeli * The only other cases that gets here are 717e86f15eeSAndrea Arcangeli * case 1, case 7 and case 8. 718e86f15eeSAndrea Arcangeli */ 719e86f15eeSAndrea Arcangeli if (next == expand) { 720e86f15eeSAndrea Arcangeli /* 721e86f15eeSAndrea Arcangeli * The only case where we don't expand "vma" 722e86f15eeSAndrea Arcangeli * and we expand "next" instead is case 8. 723e86f15eeSAndrea Arcangeli */ 724e86f15eeSAndrea Arcangeli VM_WARN_ON(end != next->vm_end); 725e86f15eeSAndrea Arcangeli /* 726e86f15eeSAndrea Arcangeli * remove_next == 3 means we're 727e86f15eeSAndrea Arcangeli * removing "vma" and that to do so we 728e86f15eeSAndrea Arcangeli * swapped "vma" and "next". 729e86f15eeSAndrea Arcangeli */ 730e86f15eeSAndrea Arcangeli remove_next = 3; 731e86f15eeSAndrea Arcangeli VM_WARN_ON(file != next->vm_file); 732e86f15eeSAndrea Arcangeli swap(vma, next); 733e86f15eeSAndrea Arcangeli } else { 734e86f15eeSAndrea Arcangeli VM_WARN_ON(expand != vma); 735e86f15eeSAndrea Arcangeli /* 736e86f15eeSAndrea Arcangeli * case 1, 6, 7, remove_next == 2 is case 6, 737e86f15eeSAndrea Arcangeli * remove_next == 1 is case 1 or 7. 7381da177e4SLinus Torvalds */ 739734537c9SKirill A. Shutemov remove_next = 1 + (end > next->vm_end); 740e86f15eeSAndrea Arcangeli VM_WARN_ON(remove_next == 2 && 741e86f15eeSAndrea Arcangeli end != next->vm_next->vm_end); 742e86f15eeSAndrea Arcangeli /* trim end to next, for case 6 first pass */ 7431da177e4SLinus Torvalds end = next->vm_end; 744e86f15eeSAndrea Arcangeli } 745e86f15eeSAndrea Arcangeli 746287d97acSLinus Torvalds exporter = next; 7471da177e4SLinus Torvalds importer = vma; 748734537c9SKirill A. Shutemov 749734537c9SKirill A. Shutemov /* 750734537c9SKirill A. Shutemov * If next doesn't have anon_vma, import from vma after 751734537c9SKirill A. Shutemov * next, if the vma overlaps with it. 752734537c9SKirill A. Shutemov */ 75397a42cd4SAndrea Arcangeli if (remove_next == 2 && !next->anon_vma) 754734537c9SKirill A. Shutemov exporter = next->vm_next; 755734537c9SKirill A. Shutemov 7561da177e4SLinus Torvalds } else if (end > next->vm_start) { 7571da177e4SLinus Torvalds /* 7581da177e4SLinus Torvalds * vma expands, overlapping part of the next: 7591da177e4SLinus Torvalds * mprotect case 5 shifting the boundary up. 7601da177e4SLinus Torvalds */ 7611da177e4SLinus Torvalds adjust_next = (end - next->vm_start) >> PAGE_SHIFT; 762287d97acSLinus Torvalds exporter = next; 7631da177e4SLinus Torvalds importer = vma; 764e86f15eeSAndrea Arcangeli VM_WARN_ON(expand != importer); 7651da177e4SLinus Torvalds } else if (end < vma->vm_end) { 7661da177e4SLinus Torvalds /* 7671da177e4SLinus Torvalds * vma shrinks, and !insert tells it's not 7681da177e4SLinus Torvalds * split_vma inserting another: so it must be 7691da177e4SLinus Torvalds * mprotect case 4 shifting the boundary down. 7701da177e4SLinus Torvalds */ 7711da177e4SLinus Torvalds adjust_next = -((vma->vm_end - end) >> PAGE_SHIFT); 772287d97acSLinus Torvalds exporter = vma; 7731da177e4SLinus Torvalds importer = next; 774e86f15eeSAndrea Arcangeli VM_WARN_ON(expand != importer); 7751da177e4SLinus Torvalds } 7761da177e4SLinus Torvalds 7775beb4930SRik van Riel /* 7785beb4930SRik van Riel * Easily overlooked: when mprotect shifts the boundary, 7795beb4930SRik van Riel * make sure the expanding vma has anon_vma set if the 7805beb4930SRik van Riel * shrinking vma had, to cover any anon pages imported. 7815beb4930SRik van Riel */ 782287d97acSLinus Torvalds if (exporter && exporter->anon_vma && !importer->anon_vma) { 783c4ea95d7SDaniel Forrest int error; 784c4ea95d7SDaniel Forrest 785287d97acSLinus Torvalds importer->anon_vma = exporter->anon_vma; 786b800c91aSKonstantin Khlebnikov error = anon_vma_clone(importer, exporter); 7873fe89b3eSLeon Yu if (error) 788b800c91aSKonstantin Khlebnikov return error; 789b800c91aSKonstantin Khlebnikov } 7905beb4930SRik van Riel } 791734537c9SKirill A. Shutemov again: 792e86f15eeSAndrea Arcangeli vma_adjust_trans_huge(orig_vma, start, end, adjust_next); 79337f9f559SKirill A. Shutemov 7941da177e4SLinus Torvalds if (file) { 7951da177e4SLinus Torvalds mapping = file->f_mapping; 7961da177e4SLinus Torvalds root = &mapping->i_mmap; 797cbc91f71SSrikar Dronamraju uprobe_munmap(vma, vma->vm_start, vma->vm_end); 798682968e0SSrikar Dronamraju 799682968e0SSrikar Dronamraju if (adjust_next) 80027ba0644SKirill A. Shutemov uprobe_munmap(next, next->vm_start, next->vm_end); 801682968e0SSrikar Dronamraju 80283cde9e8SDavidlohr Bueso i_mmap_lock_write(mapping); 8031da177e4SLinus Torvalds if (insert) { 8041da177e4SLinus Torvalds /* 8056b2dbba8SMichel Lespinasse * Put into interval tree now, so instantiated pages 8061da177e4SLinus Torvalds * are visible to arm/parisc __flush_dcache_page 8071da177e4SLinus Torvalds * throughout; but we cannot insert into address 8081da177e4SLinus Torvalds * space until vma start or end is updated. 8091da177e4SLinus Torvalds */ 8101da177e4SLinus Torvalds __vma_link_file(insert); 8111da177e4SLinus Torvalds } 8121da177e4SLinus Torvalds } 8131da177e4SLinus Torvalds 814012f1800SRik van Riel anon_vma = vma->anon_vma; 815bf181b9fSMichel Lespinasse if (!anon_vma && adjust_next) 816bf181b9fSMichel Lespinasse anon_vma = next->anon_vma; 817bf181b9fSMichel Lespinasse if (anon_vma) { 818e86f15eeSAndrea Arcangeli VM_WARN_ON(adjust_next && next->anon_vma && 819e86f15eeSAndrea Arcangeli anon_vma != next->anon_vma); 8204fc3f1d6SIngo Molnar anon_vma_lock_write(anon_vma); 821bf181b9fSMichel Lespinasse anon_vma_interval_tree_pre_update_vma(vma); 822bf181b9fSMichel Lespinasse if (adjust_next) 823bf181b9fSMichel Lespinasse anon_vma_interval_tree_pre_update_vma(next); 824bf181b9fSMichel Lespinasse } 825012f1800SRik van Riel 8261da177e4SLinus Torvalds if (root) { 8271da177e4SLinus Torvalds flush_dcache_mmap_lock(mapping); 8286b2dbba8SMichel Lespinasse vma_interval_tree_remove(vma, root); 8291da177e4SLinus Torvalds if (adjust_next) 8306b2dbba8SMichel Lespinasse vma_interval_tree_remove(next, root); 8311da177e4SLinus Torvalds } 8321da177e4SLinus Torvalds 833d3737187SMichel Lespinasse if (start != vma->vm_start) { 8341da177e4SLinus Torvalds vma->vm_start = start; 835d3737187SMichel Lespinasse start_changed = true; 836d3737187SMichel Lespinasse } 837d3737187SMichel Lespinasse if (end != vma->vm_end) { 8381da177e4SLinus Torvalds vma->vm_end = end; 839d3737187SMichel Lespinasse end_changed = true; 840d3737187SMichel Lespinasse } 8411da177e4SLinus Torvalds vma->vm_pgoff = pgoff; 8421da177e4SLinus Torvalds if (adjust_next) { 8431da177e4SLinus Torvalds next->vm_start += adjust_next << PAGE_SHIFT; 8441da177e4SLinus Torvalds next->vm_pgoff += adjust_next; 8451da177e4SLinus Torvalds } 8461da177e4SLinus Torvalds 8471da177e4SLinus Torvalds if (root) { 8481da177e4SLinus Torvalds if (adjust_next) 8496b2dbba8SMichel Lespinasse vma_interval_tree_insert(next, root); 8506b2dbba8SMichel Lespinasse vma_interval_tree_insert(vma, root); 8511da177e4SLinus Torvalds flush_dcache_mmap_unlock(mapping); 8521da177e4SLinus Torvalds } 8531da177e4SLinus Torvalds 8541da177e4SLinus Torvalds if (remove_next) { 8551da177e4SLinus Torvalds /* 8561da177e4SLinus Torvalds * vma_merge has merged next into vma, and needs 8571da177e4SLinus Torvalds * us to remove next before dropping the locks. 8581da177e4SLinus Torvalds */ 859e86f15eeSAndrea Arcangeli if (remove_next != 3) 8607c61f917SWei Yang __vma_unlink(mm, next, next); 861e86f15eeSAndrea Arcangeli else 8628f26e0b1SAndrea Arcangeli /* 8638f26e0b1SAndrea Arcangeli * vma is not before next if they've been 8648f26e0b1SAndrea Arcangeli * swapped. 8658f26e0b1SAndrea Arcangeli * 8668f26e0b1SAndrea Arcangeli * pre-swap() next->vm_start was reduced so 8678f26e0b1SAndrea Arcangeli * tell validate_mm_rb to ignore pre-swap() 8688f26e0b1SAndrea Arcangeli * "next" (which is stored in post-swap() 8698f26e0b1SAndrea Arcangeli * "vma"). 8708f26e0b1SAndrea Arcangeli */ 8717c61f917SWei Yang __vma_unlink(mm, next, vma); 8721da177e4SLinus Torvalds if (file) 8731da177e4SLinus Torvalds __remove_shared_vm_struct(next, file, mapping); 8741da177e4SLinus Torvalds } else if (insert) { 8751da177e4SLinus Torvalds /* 8761da177e4SLinus Torvalds * split_vma has split insert from vma, and needs 8771da177e4SLinus Torvalds * us to insert it before dropping the locks 8781da177e4SLinus Torvalds * (it may either follow vma or precede it). 8791da177e4SLinus Torvalds */ 8801da177e4SLinus Torvalds __insert_vm_struct(mm, insert); 881d3737187SMichel Lespinasse } else { 882d3737187SMichel Lespinasse if (start_changed) 883d3737187SMichel Lespinasse vma_gap_update(vma); 884d3737187SMichel Lespinasse if (end_changed) { 885d3737187SMichel Lespinasse if (!next) 8861be7107fSHugh Dickins mm->highest_vm_end = vm_end_gap(vma); 887d3737187SMichel Lespinasse else if (!adjust_next) 888d3737187SMichel Lespinasse vma_gap_update(next); 889d3737187SMichel Lespinasse } 8901da177e4SLinus Torvalds } 8911da177e4SLinus Torvalds 892bf181b9fSMichel Lespinasse if (anon_vma) { 893bf181b9fSMichel Lespinasse anon_vma_interval_tree_post_update_vma(vma); 894bf181b9fSMichel Lespinasse if (adjust_next) 895bf181b9fSMichel Lespinasse anon_vma_interval_tree_post_update_vma(next); 89608b52706SKonstantin Khlebnikov anon_vma_unlock_write(anon_vma); 897bf181b9fSMichel Lespinasse } 8981da177e4SLinus Torvalds if (mapping) 89983cde9e8SDavidlohr Bueso i_mmap_unlock_write(mapping); 9001da177e4SLinus Torvalds 9012b144498SSrikar Dronamraju if (root) { 9027b2d81d4SIngo Molnar uprobe_mmap(vma); 9032b144498SSrikar Dronamraju 9042b144498SSrikar Dronamraju if (adjust_next) 9057b2d81d4SIngo Molnar uprobe_mmap(next); 9062b144498SSrikar Dronamraju } 9072b144498SSrikar Dronamraju 9081da177e4SLinus Torvalds if (remove_next) { 909925d1c40SMatt Helsley if (file) { 910cbc91f71SSrikar Dronamraju uprobe_munmap(next, next->vm_start, next->vm_end); 9111da177e4SLinus Torvalds fput(file); 912925d1c40SMatt Helsley } 9135beb4930SRik van Riel if (next->anon_vma) 9145beb4930SRik van Riel anon_vma_merge(vma, next); 9151da177e4SLinus Torvalds mm->map_count--; 9163964acd0SOleg Nesterov mpol_put(vma_policy(next)); 9173928d4f5SLinus Torvalds vm_area_free(next); 9181da177e4SLinus Torvalds /* 9191da177e4SLinus Torvalds * In mprotect's case 6 (see comments on vma_merge), 9201da177e4SLinus Torvalds * we must remove another next too. It would clutter 9211da177e4SLinus Torvalds * up the code too much to do both in one go. 9221da177e4SLinus Torvalds */ 923e86f15eeSAndrea Arcangeli if (remove_next != 3) { 924e86f15eeSAndrea Arcangeli /* 925e86f15eeSAndrea Arcangeli * If "next" was removed and vma->vm_end was 926e86f15eeSAndrea Arcangeli * expanded (up) over it, in turn 927e86f15eeSAndrea Arcangeli * "next->vm_prev->vm_end" changed and the 928e86f15eeSAndrea Arcangeli * "vma->vm_next" gap must be updated. 929e86f15eeSAndrea Arcangeli */ 9301da177e4SLinus Torvalds next = vma->vm_next; 931e86f15eeSAndrea Arcangeli } else { 932e86f15eeSAndrea Arcangeli /* 933e86f15eeSAndrea Arcangeli * For the scope of the comment "next" and 934e86f15eeSAndrea Arcangeli * "vma" considered pre-swap(): if "vma" was 935e86f15eeSAndrea Arcangeli * removed, next->vm_start was expanded (down) 936e86f15eeSAndrea Arcangeli * over it and the "next" gap must be updated. 937e86f15eeSAndrea Arcangeli * Because of the swap() the post-swap() "vma" 938e86f15eeSAndrea Arcangeli * actually points to pre-swap() "next" 939e86f15eeSAndrea Arcangeli * (post-swap() "next" as opposed is now a 940e86f15eeSAndrea Arcangeli * dangling pointer). 941e86f15eeSAndrea Arcangeli */ 942e86f15eeSAndrea Arcangeli next = vma; 943e86f15eeSAndrea Arcangeli } 944734537c9SKirill A. Shutemov if (remove_next == 2) { 945734537c9SKirill A. Shutemov remove_next = 1; 946734537c9SKirill A. Shutemov end = next->vm_end; 9471da177e4SLinus Torvalds goto again; 948734537c9SKirill A. Shutemov } 949d3737187SMichel Lespinasse else if (next) 950d3737187SMichel Lespinasse vma_gap_update(next); 951fb8c41e9SAndrea Arcangeli else { 952fb8c41e9SAndrea Arcangeli /* 953fb8c41e9SAndrea Arcangeli * If remove_next == 2 we obviously can't 954fb8c41e9SAndrea Arcangeli * reach this path. 955fb8c41e9SAndrea Arcangeli * 956fb8c41e9SAndrea Arcangeli * If remove_next == 3 we can't reach this 957fb8c41e9SAndrea Arcangeli * path because pre-swap() next is always not 958fb8c41e9SAndrea Arcangeli * NULL. pre-swap() "next" is not being 959fb8c41e9SAndrea Arcangeli * removed and its next->vm_end is not altered 960fb8c41e9SAndrea Arcangeli * (and furthermore "end" already matches 961fb8c41e9SAndrea Arcangeli * next->vm_end in remove_next == 3). 962fb8c41e9SAndrea Arcangeli * 963fb8c41e9SAndrea Arcangeli * We reach this only in the remove_next == 1 964fb8c41e9SAndrea Arcangeli * case if the "next" vma that was removed was 965fb8c41e9SAndrea Arcangeli * the highest vma of the mm. However in such 966fb8c41e9SAndrea Arcangeli * case next->vm_end == "end" and the extended 967fb8c41e9SAndrea Arcangeli * "vma" has vma->vm_end == next->vm_end so 968fb8c41e9SAndrea Arcangeli * mm->highest_vm_end doesn't need any update 969fb8c41e9SAndrea Arcangeli * in remove_next == 1 case. 970fb8c41e9SAndrea Arcangeli */ 9711be7107fSHugh Dickins VM_WARN_ON(mm->highest_vm_end != vm_end_gap(vma)); 972fb8c41e9SAndrea Arcangeli } 9731da177e4SLinus Torvalds } 9742b144498SSrikar Dronamraju if (insert && file) 9757b2d81d4SIngo Molnar uprobe_mmap(insert); 9761da177e4SLinus Torvalds 9771da177e4SLinus Torvalds validate_mm(mm); 9785beb4930SRik van Riel 9795beb4930SRik van Riel return 0; 9801da177e4SLinus Torvalds } 9811da177e4SLinus Torvalds 9821da177e4SLinus Torvalds /* 9831da177e4SLinus Torvalds * If the vma has a ->close operation then the driver probably needs to release 9841da177e4SLinus Torvalds * per-vma resources, so we don't attempt to merge those. 9851da177e4SLinus Torvalds */ 9861da177e4SLinus Torvalds static inline int is_mergeable_vma(struct vm_area_struct *vma, 98719a809afSAndrea Arcangeli struct file *file, unsigned long vm_flags, 98819a809afSAndrea Arcangeli struct vm_userfaultfd_ctx vm_userfaultfd_ctx) 9891da177e4SLinus Torvalds { 99034228d47SCyrill Gorcunov /* 99134228d47SCyrill Gorcunov * VM_SOFTDIRTY should not prevent from VMA merging, if we 99234228d47SCyrill Gorcunov * match the flags but dirty bit -- the caller should mark 99334228d47SCyrill Gorcunov * merged VMA as dirty. If dirty bit won't be excluded from 9948bb4e7a2SWei Yang * comparison, we increase pressure on the memory system forcing 99534228d47SCyrill Gorcunov * the kernel to generate new VMAs when old one could be 99634228d47SCyrill Gorcunov * extended instead. 99734228d47SCyrill Gorcunov */ 99834228d47SCyrill Gorcunov if ((vma->vm_flags ^ vm_flags) & ~VM_SOFTDIRTY) 9991da177e4SLinus Torvalds return 0; 10001da177e4SLinus Torvalds if (vma->vm_file != file) 10011da177e4SLinus Torvalds return 0; 10021da177e4SLinus Torvalds if (vma->vm_ops && vma->vm_ops->close) 10031da177e4SLinus Torvalds return 0; 100419a809afSAndrea Arcangeli if (!is_mergeable_vm_userfaultfd_ctx(vma, vm_userfaultfd_ctx)) 100519a809afSAndrea Arcangeli return 0; 10061da177e4SLinus Torvalds return 1; 10071da177e4SLinus Torvalds } 10081da177e4SLinus Torvalds 10091da177e4SLinus Torvalds static inline int is_mergeable_anon_vma(struct anon_vma *anon_vma1, 1010965f55deSShaohua Li struct anon_vma *anon_vma2, 1011965f55deSShaohua Li struct vm_area_struct *vma) 10121da177e4SLinus Torvalds { 1013965f55deSShaohua Li /* 1014965f55deSShaohua Li * The list_is_singular() test is to avoid merging VMA cloned from 1015965f55deSShaohua Li * parents. This can improve scalability caused by anon_vma lock. 1016965f55deSShaohua Li */ 1017965f55deSShaohua Li if ((!anon_vma1 || !anon_vma2) && (!vma || 1018965f55deSShaohua Li list_is_singular(&vma->anon_vma_chain))) 1019965f55deSShaohua Li return 1; 1020965f55deSShaohua Li return anon_vma1 == anon_vma2; 10211da177e4SLinus Torvalds } 10221da177e4SLinus Torvalds 10231da177e4SLinus Torvalds /* 10241da177e4SLinus Torvalds * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff) 10251da177e4SLinus Torvalds * in front of (at a lower virtual address and file offset than) the vma. 10261da177e4SLinus Torvalds * 10271da177e4SLinus Torvalds * We cannot merge two vmas if they have differently assigned (non-NULL) 10281da177e4SLinus Torvalds * anon_vmas, nor if same anon_vma is assigned but offsets incompatible. 10291da177e4SLinus Torvalds * 10301da177e4SLinus Torvalds * We don't check here for the merged mmap wrapping around the end of pagecache 103145e55300SPeter Collingbourne * indices (16TB on ia32) because do_mmap() does not permit mmap's which 10321da177e4SLinus Torvalds * wrap, nor mmaps which cover the final page at index -1UL. 10331da177e4SLinus Torvalds */ 10341da177e4SLinus Torvalds static int 10351da177e4SLinus Torvalds can_vma_merge_before(struct vm_area_struct *vma, unsigned long vm_flags, 103619a809afSAndrea Arcangeli struct anon_vma *anon_vma, struct file *file, 103719a809afSAndrea Arcangeli pgoff_t vm_pgoff, 103819a809afSAndrea Arcangeli struct vm_userfaultfd_ctx vm_userfaultfd_ctx) 10391da177e4SLinus Torvalds { 104019a809afSAndrea Arcangeli if (is_mergeable_vma(vma, file, vm_flags, vm_userfaultfd_ctx) && 1041965f55deSShaohua Li is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) { 10421da177e4SLinus Torvalds if (vma->vm_pgoff == vm_pgoff) 10431da177e4SLinus Torvalds return 1; 10441da177e4SLinus Torvalds } 10451da177e4SLinus Torvalds return 0; 10461da177e4SLinus Torvalds } 10471da177e4SLinus Torvalds 10481da177e4SLinus Torvalds /* 10491da177e4SLinus Torvalds * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff) 10501da177e4SLinus Torvalds * beyond (at a higher virtual address and file offset than) the vma. 10511da177e4SLinus Torvalds * 10521da177e4SLinus Torvalds * We cannot merge two vmas if they have differently assigned (non-NULL) 10531da177e4SLinus Torvalds * anon_vmas, nor if same anon_vma is assigned but offsets incompatible. 10541da177e4SLinus Torvalds */ 10551da177e4SLinus Torvalds static int 10561da177e4SLinus Torvalds can_vma_merge_after(struct vm_area_struct *vma, unsigned long vm_flags, 105719a809afSAndrea Arcangeli struct anon_vma *anon_vma, struct file *file, 105819a809afSAndrea Arcangeli pgoff_t vm_pgoff, 105919a809afSAndrea Arcangeli struct vm_userfaultfd_ctx vm_userfaultfd_ctx) 10601da177e4SLinus Torvalds { 106119a809afSAndrea Arcangeli if (is_mergeable_vma(vma, file, vm_flags, vm_userfaultfd_ctx) && 1062965f55deSShaohua Li is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) { 10631da177e4SLinus Torvalds pgoff_t vm_pglen; 1064d6e93217SLibin vm_pglen = vma_pages(vma); 10651da177e4SLinus Torvalds if (vma->vm_pgoff + vm_pglen == vm_pgoff) 10661da177e4SLinus Torvalds return 1; 10671da177e4SLinus Torvalds } 10681da177e4SLinus Torvalds return 0; 10691da177e4SLinus Torvalds } 10701da177e4SLinus Torvalds 10711da177e4SLinus Torvalds /* 10721da177e4SLinus Torvalds * Given a mapping request (addr,end,vm_flags,file,pgoff), figure out 10731da177e4SLinus Torvalds * whether that can be merged with its predecessor or its successor. 10741da177e4SLinus Torvalds * Or both (it neatly fills a hole). 10751da177e4SLinus Torvalds * 10761da177e4SLinus Torvalds * In most cases - when called for mmap, brk or mremap - [addr,end) is 10771da177e4SLinus Torvalds * certain not to be mapped by the time vma_merge is called; but when 10781da177e4SLinus Torvalds * called for mprotect, it is certain to be already mapped (either at 10791da177e4SLinus Torvalds * an offset within prev, or at the start of next), and the flags of 10801da177e4SLinus Torvalds * this area are about to be changed to vm_flags - and the no-change 10811da177e4SLinus Torvalds * case has already been eliminated. 10821da177e4SLinus Torvalds * 10831da177e4SLinus Torvalds * The following mprotect cases have to be considered, where AAAA is 10841da177e4SLinus Torvalds * the area passed down from mprotect_fixup, never extending beyond one 10851da177e4SLinus Torvalds * vma, PPPPPP is the prev vma specified, and NNNNNN the next vma after: 10861da177e4SLinus Torvalds * 10875d42ab29SWei Yang * AAAA AAAA AAAA 10885d42ab29SWei Yang * PPPPPPNNNNNN PPPPPPNNNNNN PPPPPPNNNNNN 10895d42ab29SWei Yang * cannot merge might become might become 10905d42ab29SWei Yang * PPNNNNNNNNNN PPPPPPPPPPNN 10915d42ab29SWei Yang * mmap, brk or case 4 below case 5 below 10925d42ab29SWei Yang * mremap move: 10935d42ab29SWei Yang * AAAA AAAA 10945d42ab29SWei Yang * PPPP NNNN PPPPNNNNXXXX 10955d42ab29SWei Yang * might become might become 10965d42ab29SWei Yang * PPPPPPPPPPPP 1 or PPPPPPPPPPPP 6 or 10975d42ab29SWei Yang * PPPPPPPPNNNN 2 or PPPPPPPPXXXX 7 or 10985d42ab29SWei Yang * PPPPNNNNNNNN 3 PPPPXXXXXXXX 8 10991da177e4SLinus Torvalds * 11008bb4e7a2SWei Yang * It is important for case 8 that the vma NNNN overlapping the 1101e86f15eeSAndrea Arcangeli * region AAAA is never going to extended over XXXX. Instead XXXX must 1102e86f15eeSAndrea Arcangeli * be extended in region AAAA and NNNN must be removed. This way in 1103e86f15eeSAndrea Arcangeli * all cases where vma_merge succeeds, the moment vma_adjust drops the 1104e86f15eeSAndrea Arcangeli * rmap_locks, the properties of the merged vma will be already 1105e86f15eeSAndrea Arcangeli * correct for the whole merged range. Some of those properties like 1106e86f15eeSAndrea Arcangeli * vm_page_prot/vm_flags may be accessed by rmap_walks and they must 1107e86f15eeSAndrea Arcangeli * be correct for the whole merged range immediately after the 1108e86f15eeSAndrea Arcangeli * rmap_locks are released. Otherwise if XXXX would be removed and 1109e86f15eeSAndrea Arcangeli * NNNN would be extended over the XXXX range, remove_migration_ptes 1110e86f15eeSAndrea Arcangeli * or other rmap walkers (if working on addresses beyond the "end" 1111e86f15eeSAndrea Arcangeli * parameter) may establish ptes with the wrong permissions of NNNN 1112e86f15eeSAndrea Arcangeli * instead of the right permissions of XXXX. 11131da177e4SLinus Torvalds */ 11141da177e4SLinus Torvalds struct vm_area_struct *vma_merge(struct mm_struct *mm, 11151da177e4SLinus Torvalds struct vm_area_struct *prev, unsigned long addr, 11161da177e4SLinus Torvalds unsigned long end, unsigned long vm_flags, 11171da177e4SLinus Torvalds struct anon_vma *anon_vma, struct file *file, 111819a809afSAndrea Arcangeli pgoff_t pgoff, struct mempolicy *policy, 111919a809afSAndrea Arcangeli struct vm_userfaultfd_ctx vm_userfaultfd_ctx) 11201da177e4SLinus Torvalds { 11211da177e4SLinus Torvalds pgoff_t pglen = (end - addr) >> PAGE_SHIFT; 11221da177e4SLinus Torvalds struct vm_area_struct *area, *next; 11235beb4930SRik van Riel int err; 11241da177e4SLinus Torvalds 11251da177e4SLinus Torvalds /* 11261da177e4SLinus Torvalds * We later require that vma->vm_flags == vm_flags, 11271da177e4SLinus Torvalds * so this tests vma->vm_flags & VM_SPECIAL, too. 11281da177e4SLinus Torvalds */ 11291da177e4SLinus Torvalds if (vm_flags & VM_SPECIAL) 11301da177e4SLinus Torvalds return NULL; 11311da177e4SLinus Torvalds 11321da177e4SLinus Torvalds if (prev) 11331da177e4SLinus Torvalds next = prev->vm_next; 11341da177e4SLinus Torvalds else 11351da177e4SLinus Torvalds next = mm->mmap; 11361da177e4SLinus Torvalds area = next; 1137e86f15eeSAndrea Arcangeli if (area && area->vm_end == end) /* cases 6, 7, 8 */ 11381da177e4SLinus Torvalds next = next->vm_next; 11391da177e4SLinus Torvalds 1140e86f15eeSAndrea Arcangeli /* verify some invariant that must be enforced by the caller */ 1141e86f15eeSAndrea Arcangeli VM_WARN_ON(prev && addr <= prev->vm_start); 1142e86f15eeSAndrea Arcangeli VM_WARN_ON(area && end > area->vm_end); 1143e86f15eeSAndrea Arcangeli VM_WARN_ON(addr >= end); 1144e86f15eeSAndrea Arcangeli 11451da177e4SLinus Torvalds /* 11461da177e4SLinus Torvalds * Can it merge with the predecessor? 11471da177e4SLinus Torvalds */ 11481da177e4SLinus Torvalds if (prev && prev->vm_end == addr && 11491da177e4SLinus Torvalds mpol_equal(vma_policy(prev), policy) && 11501da177e4SLinus Torvalds can_vma_merge_after(prev, vm_flags, 115119a809afSAndrea Arcangeli anon_vma, file, pgoff, 115219a809afSAndrea Arcangeli vm_userfaultfd_ctx)) { 11531da177e4SLinus Torvalds /* 11541da177e4SLinus Torvalds * OK, it can. Can we now merge in the successor as well? 11551da177e4SLinus Torvalds */ 11561da177e4SLinus Torvalds if (next && end == next->vm_start && 11571da177e4SLinus Torvalds mpol_equal(policy, vma_policy(next)) && 11581da177e4SLinus Torvalds can_vma_merge_before(next, vm_flags, 115919a809afSAndrea Arcangeli anon_vma, file, 116019a809afSAndrea Arcangeli pgoff+pglen, 116119a809afSAndrea Arcangeli vm_userfaultfd_ctx) && 11621da177e4SLinus Torvalds is_mergeable_anon_vma(prev->anon_vma, 1163965f55deSShaohua Li next->anon_vma, NULL)) { 11641da177e4SLinus Torvalds /* cases 1, 6 */ 1165e86f15eeSAndrea Arcangeli err = __vma_adjust(prev, prev->vm_start, 1166e86f15eeSAndrea Arcangeli next->vm_end, prev->vm_pgoff, NULL, 1167e86f15eeSAndrea Arcangeli prev); 11681da177e4SLinus Torvalds } else /* cases 2, 5, 7 */ 1169e86f15eeSAndrea Arcangeli err = __vma_adjust(prev, prev->vm_start, 1170e86f15eeSAndrea Arcangeli end, prev->vm_pgoff, NULL, prev); 11715beb4930SRik van Riel if (err) 11725beb4930SRik van Riel return NULL; 11736d50e60cSDavid Rientjes khugepaged_enter_vma_merge(prev, vm_flags); 11741da177e4SLinus Torvalds return prev; 11751da177e4SLinus Torvalds } 11761da177e4SLinus Torvalds 11771da177e4SLinus Torvalds /* 11781da177e4SLinus Torvalds * Can this new request be merged in front of next? 11791da177e4SLinus Torvalds */ 11801da177e4SLinus Torvalds if (next && end == next->vm_start && 11811da177e4SLinus Torvalds mpol_equal(policy, vma_policy(next)) && 11821da177e4SLinus Torvalds can_vma_merge_before(next, vm_flags, 118319a809afSAndrea Arcangeli anon_vma, file, pgoff+pglen, 118419a809afSAndrea Arcangeli vm_userfaultfd_ctx)) { 11851da177e4SLinus Torvalds if (prev && addr < prev->vm_end) /* case 4 */ 1186e86f15eeSAndrea Arcangeli err = __vma_adjust(prev, prev->vm_start, 1187e86f15eeSAndrea Arcangeli addr, prev->vm_pgoff, NULL, next); 1188e86f15eeSAndrea Arcangeli else { /* cases 3, 8 */ 1189e86f15eeSAndrea Arcangeli err = __vma_adjust(area, addr, next->vm_end, 1190e86f15eeSAndrea Arcangeli next->vm_pgoff - pglen, NULL, next); 1191e86f15eeSAndrea Arcangeli /* 1192e86f15eeSAndrea Arcangeli * In case 3 area is already equal to next and 1193e86f15eeSAndrea Arcangeli * this is a noop, but in case 8 "area" has 1194e86f15eeSAndrea Arcangeli * been removed and next was expanded over it. 1195e86f15eeSAndrea Arcangeli */ 1196e86f15eeSAndrea Arcangeli area = next; 1197e86f15eeSAndrea Arcangeli } 11985beb4930SRik van Riel if (err) 11995beb4930SRik van Riel return NULL; 12006d50e60cSDavid Rientjes khugepaged_enter_vma_merge(area, vm_flags); 12011da177e4SLinus Torvalds return area; 12021da177e4SLinus Torvalds } 12031da177e4SLinus Torvalds 12041da177e4SLinus Torvalds return NULL; 12051da177e4SLinus Torvalds } 12061da177e4SLinus Torvalds 12071da177e4SLinus Torvalds /* 1208b4f315b4SEthon Paul * Rough compatibility check to quickly see if it's even worth looking 1209d0e9fe17SLinus Torvalds * at sharing an anon_vma. 1210d0e9fe17SLinus Torvalds * 1211d0e9fe17SLinus Torvalds * They need to have the same vm_file, and the flags can only differ 1212d0e9fe17SLinus Torvalds * in things that mprotect may change. 1213d0e9fe17SLinus Torvalds * 1214d0e9fe17SLinus Torvalds * NOTE! The fact that we share an anon_vma doesn't _have_ to mean that 1215d0e9fe17SLinus Torvalds * we can merge the two vma's. For example, we refuse to merge a vma if 1216d0e9fe17SLinus Torvalds * there is a vm_ops->close() function, because that indicates that the 1217d0e9fe17SLinus Torvalds * driver is doing some kind of reference counting. But that doesn't 1218d0e9fe17SLinus Torvalds * really matter for the anon_vma sharing case. 1219d0e9fe17SLinus Torvalds */ 1220d0e9fe17SLinus Torvalds static int anon_vma_compatible(struct vm_area_struct *a, struct vm_area_struct *b) 1221d0e9fe17SLinus Torvalds { 1222d0e9fe17SLinus Torvalds return a->vm_end == b->vm_start && 1223d0e9fe17SLinus Torvalds mpol_equal(vma_policy(a), vma_policy(b)) && 1224d0e9fe17SLinus Torvalds a->vm_file == b->vm_file && 12256cb4d9a2SAnshuman Khandual !((a->vm_flags ^ b->vm_flags) & ~(VM_ACCESS_FLAGS | VM_SOFTDIRTY)) && 1226d0e9fe17SLinus Torvalds b->vm_pgoff == a->vm_pgoff + ((b->vm_start - a->vm_start) >> PAGE_SHIFT); 1227d0e9fe17SLinus Torvalds } 1228d0e9fe17SLinus Torvalds 1229d0e9fe17SLinus Torvalds /* 1230d0e9fe17SLinus Torvalds * Do some basic sanity checking to see if we can re-use the anon_vma 1231d0e9fe17SLinus Torvalds * from 'old'. The 'a'/'b' vma's are in VM order - one of them will be 1232d0e9fe17SLinus Torvalds * the same as 'old', the other will be the new one that is trying 1233d0e9fe17SLinus Torvalds * to share the anon_vma. 1234d0e9fe17SLinus Torvalds * 1235d0e9fe17SLinus Torvalds * NOTE! This runs with mm_sem held for reading, so it is possible that 1236d0e9fe17SLinus Torvalds * the anon_vma of 'old' is concurrently in the process of being set up 1237d0e9fe17SLinus Torvalds * by another page fault trying to merge _that_. But that's ok: if it 1238d0e9fe17SLinus Torvalds * is being set up, that automatically means that it will be a singleton 1239d0e9fe17SLinus Torvalds * acceptable for merging, so we can do all of this optimistically. But 12404db0c3c2SJason Low * we do that READ_ONCE() to make sure that we never re-load the pointer. 1241d0e9fe17SLinus Torvalds * 1242d0e9fe17SLinus Torvalds * IOW: that the "list_is_singular()" test on the anon_vma_chain only 1243d0e9fe17SLinus Torvalds * matters for the 'stable anon_vma' case (ie the thing we want to avoid 1244d0e9fe17SLinus Torvalds * is to return an anon_vma that is "complex" due to having gone through 1245d0e9fe17SLinus Torvalds * a fork). 1246d0e9fe17SLinus Torvalds * 1247d0e9fe17SLinus Torvalds * We also make sure that the two vma's are compatible (adjacent, 1248d0e9fe17SLinus Torvalds * and with the same memory policies). That's all stable, even with just 1249d0e9fe17SLinus Torvalds * a read lock on the mm_sem. 1250d0e9fe17SLinus Torvalds */ 1251d0e9fe17SLinus Torvalds static struct anon_vma *reusable_anon_vma(struct vm_area_struct *old, struct vm_area_struct *a, struct vm_area_struct *b) 1252d0e9fe17SLinus Torvalds { 1253d0e9fe17SLinus Torvalds if (anon_vma_compatible(a, b)) { 12544db0c3c2SJason Low struct anon_vma *anon_vma = READ_ONCE(old->anon_vma); 1255d0e9fe17SLinus Torvalds 1256d0e9fe17SLinus Torvalds if (anon_vma && list_is_singular(&old->anon_vma_chain)) 1257d0e9fe17SLinus Torvalds return anon_vma; 1258d0e9fe17SLinus Torvalds } 1259d0e9fe17SLinus Torvalds return NULL; 1260d0e9fe17SLinus Torvalds } 1261d0e9fe17SLinus Torvalds 1262d0e9fe17SLinus Torvalds /* 12631da177e4SLinus Torvalds * find_mergeable_anon_vma is used by anon_vma_prepare, to check 12641da177e4SLinus Torvalds * neighbouring vmas for a suitable anon_vma, before it goes off 12651da177e4SLinus Torvalds * to allocate a new anon_vma. It checks because a repetitive 12661da177e4SLinus Torvalds * sequence of mprotects and faults may otherwise lead to distinct 12671da177e4SLinus Torvalds * anon_vmas being allocated, preventing vma merge in subsequent 12681da177e4SLinus Torvalds * mprotect. 12691da177e4SLinus Torvalds */ 12701da177e4SLinus Torvalds struct anon_vma *find_mergeable_anon_vma(struct vm_area_struct *vma) 12711da177e4SLinus Torvalds { 1272a67c8caaSMiaohe Lin struct anon_vma *anon_vma = NULL; 12731da177e4SLinus Torvalds 1274a67c8caaSMiaohe Lin /* Try next first. */ 1275a67c8caaSMiaohe Lin if (vma->vm_next) { 1276a67c8caaSMiaohe Lin anon_vma = reusable_anon_vma(vma->vm_next, vma, vma->vm_next); 1277d0e9fe17SLinus Torvalds if (anon_vma) 1278d0e9fe17SLinus Torvalds return anon_vma; 1279a67c8caaSMiaohe Lin } 12801da177e4SLinus Torvalds 1281a67c8caaSMiaohe Lin /* Try prev next. */ 1282a67c8caaSMiaohe Lin if (vma->vm_prev) 1283a67c8caaSMiaohe Lin anon_vma = reusable_anon_vma(vma->vm_prev, vma->vm_prev, vma); 1284a67c8caaSMiaohe Lin 12851da177e4SLinus Torvalds /* 1286a67c8caaSMiaohe Lin * We might reach here with anon_vma == NULL if we can't find 1287a67c8caaSMiaohe Lin * any reusable anon_vma. 12881da177e4SLinus Torvalds * There's no absolute need to look only at touching neighbours: 12891da177e4SLinus Torvalds * we could search further afield for "compatible" anon_vmas. 12901da177e4SLinus Torvalds * But it would probably just be a waste of time searching, 12911da177e4SLinus Torvalds * or lead to too many vmas hanging off the same anon_vma. 12921da177e4SLinus Torvalds * We're trying to allow mprotect remerging later on, 12931da177e4SLinus Torvalds * not trying to minimize memory used for anon_vmas. 12941da177e4SLinus Torvalds */ 1295a67c8caaSMiaohe Lin return anon_vma; 12961da177e4SLinus Torvalds } 12971da177e4SLinus Torvalds 12981da177e4SLinus Torvalds /* 129940401530SAl Viro * If a hint addr is less than mmap_min_addr change hint to be as 130040401530SAl Viro * low as possible but still greater than mmap_min_addr 130140401530SAl Viro */ 130240401530SAl Viro static inline unsigned long round_hint_to_min(unsigned long hint) 130340401530SAl Viro { 130440401530SAl Viro hint &= PAGE_MASK; 130540401530SAl Viro if (((void *)hint != NULL) && 130640401530SAl Viro (hint < mmap_min_addr)) 130740401530SAl Viro return PAGE_ALIGN(mmap_min_addr); 130840401530SAl Viro return hint; 130940401530SAl Viro } 131040401530SAl Viro 1311363ee17fSDavidlohr Bueso static inline int mlock_future_check(struct mm_struct *mm, 1312363ee17fSDavidlohr Bueso unsigned long flags, 1313363ee17fSDavidlohr Bueso unsigned long len) 1314363ee17fSDavidlohr Bueso { 1315363ee17fSDavidlohr Bueso unsigned long locked, lock_limit; 1316363ee17fSDavidlohr Bueso 1317363ee17fSDavidlohr Bueso /* mlock MCL_FUTURE? */ 1318363ee17fSDavidlohr Bueso if (flags & VM_LOCKED) { 1319363ee17fSDavidlohr Bueso locked = len >> PAGE_SHIFT; 1320363ee17fSDavidlohr Bueso locked += mm->locked_vm; 1321363ee17fSDavidlohr Bueso lock_limit = rlimit(RLIMIT_MEMLOCK); 1322363ee17fSDavidlohr Bueso lock_limit >>= PAGE_SHIFT; 1323363ee17fSDavidlohr Bueso if (locked > lock_limit && !capable(CAP_IPC_LOCK)) 1324363ee17fSDavidlohr Bueso return -EAGAIN; 1325363ee17fSDavidlohr Bueso } 1326363ee17fSDavidlohr Bueso return 0; 1327363ee17fSDavidlohr Bueso } 1328363ee17fSDavidlohr Bueso 1329be83bbf8SLinus Torvalds static inline u64 file_mmap_size_max(struct file *file, struct inode *inode) 1330be83bbf8SLinus Torvalds { 1331be83bbf8SLinus Torvalds if (S_ISREG(inode->i_mode)) 1332423913adSLinus Torvalds return MAX_LFS_FILESIZE; 1333be83bbf8SLinus Torvalds 1334be83bbf8SLinus Torvalds if (S_ISBLK(inode->i_mode)) 1335be83bbf8SLinus Torvalds return MAX_LFS_FILESIZE; 1336be83bbf8SLinus Torvalds 133776f34950SIvan Khoronzhuk if (S_ISSOCK(inode->i_mode)) 133876f34950SIvan Khoronzhuk return MAX_LFS_FILESIZE; 133976f34950SIvan Khoronzhuk 1340be83bbf8SLinus Torvalds /* Special "we do even unsigned file positions" case */ 1341be83bbf8SLinus Torvalds if (file->f_mode & FMODE_UNSIGNED_OFFSET) 1342be83bbf8SLinus Torvalds return 0; 1343be83bbf8SLinus Torvalds 1344be83bbf8SLinus Torvalds /* Yes, random drivers might want more. But I'm tired of buggy drivers */ 1345be83bbf8SLinus Torvalds return ULONG_MAX; 1346be83bbf8SLinus Torvalds } 1347be83bbf8SLinus Torvalds 1348be83bbf8SLinus Torvalds static inline bool file_mmap_ok(struct file *file, struct inode *inode, 1349be83bbf8SLinus Torvalds unsigned long pgoff, unsigned long len) 1350be83bbf8SLinus Torvalds { 1351be83bbf8SLinus Torvalds u64 maxsize = file_mmap_size_max(file, inode); 1352be83bbf8SLinus Torvalds 1353be83bbf8SLinus Torvalds if (maxsize && len > maxsize) 1354be83bbf8SLinus Torvalds return false; 1355be83bbf8SLinus Torvalds maxsize -= len; 1356be83bbf8SLinus Torvalds if (pgoff > maxsize >> PAGE_SHIFT) 1357be83bbf8SLinus Torvalds return false; 1358be83bbf8SLinus Torvalds return true; 1359be83bbf8SLinus Torvalds } 1360be83bbf8SLinus Torvalds 136140401530SAl Viro /* 13623e4e28c5SMichel Lespinasse * The caller must write-lock current->mm->mmap_lock. 13631da177e4SLinus Torvalds */ 13641fcfd8dbSOleg Nesterov unsigned long do_mmap(struct file *file, unsigned long addr, 13651da177e4SLinus Torvalds unsigned long len, unsigned long prot, 136645e55300SPeter Collingbourne unsigned long flags, unsigned long pgoff, 136745e55300SPeter Collingbourne unsigned long *populate, struct list_head *uf) 13681da177e4SLinus Torvalds { 13691da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 137045e55300SPeter Collingbourne vm_flags_t vm_flags; 137162b5f7d0SDave Hansen int pkey = 0; 13721da177e4SLinus Torvalds 137341badc15SMichel Lespinasse *populate = 0; 1374bebeb3d6SMichel Lespinasse 1375e37609bbSPiotr Kwapulinski if (!len) 1376e37609bbSPiotr Kwapulinski return -EINVAL; 1377e37609bbSPiotr Kwapulinski 13781da177e4SLinus Torvalds /* 13791da177e4SLinus Torvalds * Does the application expect PROT_READ to imply PROT_EXEC? 13801da177e4SLinus Torvalds * 13811da177e4SLinus Torvalds * (the exception is when the underlying filesystem is noexec 13821da177e4SLinus Torvalds * mounted, in which case we dont add PROT_EXEC.) 13831da177e4SLinus Torvalds */ 13841da177e4SLinus Torvalds if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC)) 138590f8572bSEric W. Biederman if (!(file && path_noexec(&file->f_path))) 13861da177e4SLinus Torvalds prot |= PROT_EXEC; 13871da177e4SLinus Torvalds 1388a4ff8e86SMichal Hocko /* force arch specific MAP_FIXED handling in get_unmapped_area */ 1389a4ff8e86SMichal Hocko if (flags & MAP_FIXED_NOREPLACE) 1390a4ff8e86SMichal Hocko flags |= MAP_FIXED; 1391a4ff8e86SMichal Hocko 13927cd94146SEric Paris if (!(flags & MAP_FIXED)) 13937cd94146SEric Paris addr = round_hint_to_min(addr); 13947cd94146SEric Paris 13951da177e4SLinus Torvalds /* Careful about overflows.. */ 13961da177e4SLinus Torvalds len = PAGE_ALIGN(len); 13979206de95SAl Viro if (!len) 13981da177e4SLinus Torvalds return -ENOMEM; 13991da177e4SLinus Torvalds 14001da177e4SLinus Torvalds /* offset overflow? */ 14011da177e4SLinus Torvalds if ((pgoff + (len >> PAGE_SHIFT)) < pgoff) 14021da177e4SLinus Torvalds return -EOVERFLOW; 14031da177e4SLinus Torvalds 14041da177e4SLinus Torvalds /* Too many mappings? */ 14051da177e4SLinus Torvalds if (mm->map_count > sysctl_max_map_count) 14061da177e4SLinus Torvalds return -ENOMEM; 14071da177e4SLinus Torvalds 14081da177e4SLinus Torvalds /* Obtain the address to map to. we verify (or select) it and ensure 14091da177e4SLinus Torvalds * that it represents a valid section of the address space. 14101da177e4SLinus Torvalds */ 14111da177e4SLinus Torvalds addr = get_unmapped_area(file, addr, len, pgoff, flags); 1412ff68dac6SGaowei Pu if (IS_ERR_VALUE(addr)) 14131da177e4SLinus Torvalds return addr; 14141da177e4SLinus Torvalds 1415a4ff8e86SMichal Hocko if (flags & MAP_FIXED_NOREPLACE) { 1416a4ff8e86SMichal Hocko struct vm_area_struct *vma = find_vma(mm, addr); 1417a4ff8e86SMichal Hocko 14187aa867ddSJann Horn if (vma && vma->vm_start < addr + len) 1419a4ff8e86SMichal Hocko return -EEXIST; 1420a4ff8e86SMichal Hocko } 1421a4ff8e86SMichal Hocko 142262b5f7d0SDave Hansen if (prot == PROT_EXEC) { 142362b5f7d0SDave Hansen pkey = execute_only_pkey(mm); 142462b5f7d0SDave Hansen if (pkey < 0) 142562b5f7d0SDave Hansen pkey = 0; 142662b5f7d0SDave Hansen } 142762b5f7d0SDave Hansen 14281da177e4SLinus Torvalds /* Do simple checking here so the lower-level routines won't have 14291da177e4SLinus Torvalds * to. we assume access permissions have been handled by the open 14301da177e4SLinus Torvalds * of the memory object, so we don't do any here. 14311da177e4SLinus Torvalds */ 143245e55300SPeter Collingbourne vm_flags = calc_vm_prot_bits(prot, pkey) | calc_vm_flag_bits(flags) | 14331da177e4SLinus Torvalds mm->def_flags | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC; 14341da177e4SLinus Torvalds 1435cdf7b341SHuang Shijie if (flags & MAP_LOCKED) 14361da177e4SLinus Torvalds if (!can_do_mlock()) 14371da177e4SLinus Torvalds return -EPERM; 1438ba470de4SRik van Riel 1439363ee17fSDavidlohr Bueso if (mlock_future_check(mm, vm_flags, len)) 14401da177e4SLinus Torvalds return -EAGAIN; 14411da177e4SLinus Torvalds 14421da177e4SLinus Torvalds if (file) { 1443077bf22bSOleg Nesterov struct inode *inode = file_inode(file); 14441c972597SDan Williams unsigned long flags_mask; 14451c972597SDan Williams 1446be83bbf8SLinus Torvalds if (!file_mmap_ok(file, inode, pgoff, len)) 1447be83bbf8SLinus Torvalds return -EOVERFLOW; 1448be83bbf8SLinus Torvalds 14491c972597SDan Williams flags_mask = LEGACY_MAP_MASK | file->f_op->mmap_supported_flags; 1450077bf22bSOleg Nesterov 14511da177e4SLinus Torvalds switch (flags & MAP_TYPE) { 14521da177e4SLinus Torvalds case MAP_SHARED: 14531c972597SDan Williams /* 14541c972597SDan Williams * Force use of MAP_SHARED_VALIDATE with non-legacy 14551c972597SDan Williams * flags. E.g. MAP_SYNC is dangerous to use with 14561c972597SDan Williams * MAP_SHARED as you don't know which consistency model 14571c972597SDan Williams * you will get. We silently ignore unsupported flags 14581c972597SDan Williams * with MAP_SHARED to preserve backward compatibility. 14591c972597SDan Williams */ 14601c972597SDan Williams flags &= LEGACY_MAP_MASK; 1461e4a9bc58SJoe Perches fallthrough; 14621c972597SDan Williams case MAP_SHARED_VALIDATE: 14631c972597SDan Williams if (flags & ~flags_mask) 14641c972597SDan Williams return -EOPNOTSUPP; 1465dc617f29SDarrick J. Wong if (prot & PROT_WRITE) { 1466dc617f29SDarrick J. Wong if (!(file->f_mode & FMODE_WRITE)) 14671da177e4SLinus Torvalds return -EACCES; 1468dc617f29SDarrick J. Wong if (IS_SWAPFILE(file->f_mapping->host)) 1469dc617f29SDarrick J. Wong return -ETXTBSY; 1470dc617f29SDarrick J. Wong } 14711da177e4SLinus Torvalds 14721da177e4SLinus Torvalds /* 14731da177e4SLinus Torvalds * Make sure we don't allow writing to an append-only 14741da177e4SLinus Torvalds * file.. 14751da177e4SLinus Torvalds */ 14761da177e4SLinus Torvalds if (IS_APPEND(inode) && (file->f_mode & FMODE_WRITE)) 14771da177e4SLinus Torvalds return -EACCES; 14781da177e4SLinus Torvalds 14791da177e4SLinus Torvalds /* 14801da177e4SLinus Torvalds * Make sure there are no mandatory locks on the file. 14811da177e4SLinus Torvalds */ 1482d7a06983SJeff Layton if (locks_verify_locked(file)) 14831da177e4SLinus Torvalds return -EAGAIN; 14841da177e4SLinus Torvalds 14851da177e4SLinus Torvalds vm_flags |= VM_SHARED | VM_MAYSHARE; 14861da177e4SLinus Torvalds if (!(file->f_mode & FMODE_WRITE)) 14871da177e4SLinus Torvalds vm_flags &= ~(VM_MAYWRITE | VM_SHARED); 1488e4a9bc58SJoe Perches fallthrough; 14891da177e4SLinus Torvalds case MAP_PRIVATE: 14901da177e4SLinus Torvalds if (!(file->f_mode & FMODE_READ)) 14911da177e4SLinus Torvalds return -EACCES; 149290f8572bSEric W. Biederman if (path_noexec(&file->f_path)) { 149380c5606cSLinus Torvalds if (vm_flags & VM_EXEC) 149480c5606cSLinus Torvalds return -EPERM; 149580c5606cSLinus Torvalds vm_flags &= ~VM_MAYEXEC; 149680c5606cSLinus Torvalds } 149780c5606cSLinus Torvalds 149872c2d531SAl Viro if (!file->f_op->mmap) 149980c5606cSLinus Torvalds return -ENODEV; 1500b2c56e4fSOleg Nesterov if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP)) 1501b2c56e4fSOleg Nesterov return -EINVAL; 15021da177e4SLinus Torvalds break; 15031da177e4SLinus Torvalds 15041da177e4SLinus Torvalds default: 15051da177e4SLinus Torvalds return -EINVAL; 15061da177e4SLinus Torvalds } 15071da177e4SLinus Torvalds } else { 15081da177e4SLinus Torvalds switch (flags & MAP_TYPE) { 15091da177e4SLinus Torvalds case MAP_SHARED: 1510b2c56e4fSOleg Nesterov if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP)) 1511b2c56e4fSOleg Nesterov return -EINVAL; 1512ce363942STejun Heo /* 1513ce363942STejun Heo * Ignore pgoff. 1514ce363942STejun Heo */ 1515ce363942STejun Heo pgoff = 0; 15161da177e4SLinus Torvalds vm_flags |= VM_SHARED | VM_MAYSHARE; 15171da177e4SLinus Torvalds break; 15181da177e4SLinus Torvalds case MAP_PRIVATE: 15191da177e4SLinus Torvalds /* 15201da177e4SLinus Torvalds * Set pgoff according to addr for anon_vma. 15211da177e4SLinus Torvalds */ 15221da177e4SLinus Torvalds pgoff = addr >> PAGE_SHIFT; 15231da177e4SLinus Torvalds break; 15241da177e4SLinus Torvalds default: 15251da177e4SLinus Torvalds return -EINVAL; 15261da177e4SLinus Torvalds } 15271da177e4SLinus Torvalds } 15281da177e4SLinus Torvalds 1529c22c0d63SMichel Lespinasse /* 1530c22c0d63SMichel Lespinasse * Set 'VM_NORESERVE' if we should not account for the 1531c22c0d63SMichel Lespinasse * memory use of this mapping. 1532c22c0d63SMichel Lespinasse */ 1533c22c0d63SMichel Lespinasse if (flags & MAP_NORESERVE) { 1534c22c0d63SMichel Lespinasse /* We honor MAP_NORESERVE if allowed to overcommit */ 1535c22c0d63SMichel Lespinasse if (sysctl_overcommit_memory != OVERCOMMIT_NEVER) 1536c22c0d63SMichel Lespinasse vm_flags |= VM_NORESERVE; 1537c22c0d63SMichel Lespinasse 1538c22c0d63SMichel Lespinasse /* hugetlb applies strict overcommit unless MAP_NORESERVE */ 1539c22c0d63SMichel Lespinasse if (file && is_file_hugepages(file)) 1540c22c0d63SMichel Lespinasse vm_flags |= VM_NORESERVE; 1541c22c0d63SMichel Lespinasse } 1542c22c0d63SMichel Lespinasse 1543897ab3e0SMike Rapoport addr = mmap_region(file, addr, len, vm_flags, pgoff, uf); 154409a9f1d2SMichel Lespinasse if (!IS_ERR_VALUE(addr) && 154509a9f1d2SMichel Lespinasse ((vm_flags & VM_LOCKED) || 154609a9f1d2SMichel Lespinasse (flags & (MAP_POPULATE | MAP_NONBLOCK)) == MAP_POPULATE)) 154741badc15SMichel Lespinasse *populate = len; 1548bebeb3d6SMichel Lespinasse return addr; 15490165ab44SMiklos Szeredi } 15506be5ceb0SLinus Torvalds 1551a90f590aSDominik Brodowski unsigned long ksys_mmap_pgoff(unsigned long addr, unsigned long len, 1552a90f590aSDominik Brodowski unsigned long prot, unsigned long flags, 1553a90f590aSDominik Brodowski unsigned long fd, unsigned long pgoff) 155466f0dc48SHugh Dickins { 155566f0dc48SHugh Dickins struct file *file = NULL; 15561e3ee14bSChen Gang unsigned long retval; 155766f0dc48SHugh Dickins 155866f0dc48SHugh Dickins if (!(flags & MAP_ANONYMOUS)) { 1559120a795dSAl Viro audit_mmap_fd(fd, flags); 156066f0dc48SHugh Dickins file = fget(fd); 156166f0dc48SHugh Dickins if (!file) 15621e3ee14bSChen Gang return -EBADF; 15637bba8f0eSZhen Lei if (is_file_hugepages(file)) { 1564af73e4d9SNaoya Horiguchi len = ALIGN(len, huge_page_size(hstate_file(file))); 15657bba8f0eSZhen Lei } else if (unlikely(flags & MAP_HUGETLB)) { 1566493af578SJörn Engel retval = -EINVAL; 1567493af578SJörn Engel goto out_fput; 15687bba8f0eSZhen Lei } 156966f0dc48SHugh Dickins } else if (flags & MAP_HUGETLB) { 157066f0dc48SHugh Dickins struct user_struct *user = NULL; 1571c103a4dcSAndrew Morton struct hstate *hs; 1572af73e4d9SNaoya Horiguchi 157320ac2893SAnshuman Khandual hs = hstate_sizelog((flags >> MAP_HUGE_SHIFT) & MAP_HUGE_MASK); 1574091d0d55SLi Zefan if (!hs) 1575091d0d55SLi Zefan return -EINVAL; 1576091d0d55SLi Zefan 1577091d0d55SLi Zefan len = ALIGN(len, huge_page_size(hs)); 157866f0dc48SHugh Dickins /* 157966f0dc48SHugh Dickins * VM_NORESERVE is used because the reservations will be 158066f0dc48SHugh Dickins * taken when vm_ops->mmap() is called 158166f0dc48SHugh Dickins * A dummy user value is used because we are not locking 158266f0dc48SHugh Dickins * memory so no accounting is necessary 158366f0dc48SHugh Dickins */ 1584af73e4d9SNaoya Horiguchi file = hugetlb_file_setup(HUGETLB_ANON_FILE, len, 158542d7395fSAndi Kleen VM_NORESERVE, 158642d7395fSAndi Kleen &user, HUGETLB_ANONHUGE_INODE, 158742d7395fSAndi Kleen (flags >> MAP_HUGE_SHIFT) & MAP_HUGE_MASK); 158866f0dc48SHugh Dickins if (IS_ERR(file)) 158966f0dc48SHugh Dickins return PTR_ERR(file); 159066f0dc48SHugh Dickins } 159166f0dc48SHugh Dickins 159266f0dc48SHugh Dickins flags &= ~(MAP_EXECUTABLE | MAP_DENYWRITE); 159366f0dc48SHugh Dickins 15949fbeb5abSMichal Hocko retval = vm_mmap_pgoff(file, addr, len, prot, flags, pgoff); 1595493af578SJörn Engel out_fput: 159666f0dc48SHugh Dickins if (file) 159766f0dc48SHugh Dickins fput(file); 159866f0dc48SHugh Dickins return retval; 159966f0dc48SHugh Dickins } 160066f0dc48SHugh Dickins 1601a90f590aSDominik Brodowski SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len, 1602a90f590aSDominik Brodowski unsigned long, prot, unsigned long, flags, 1603a90f590aSDominik Brodowski unsigned long, fd, unsigned long, pgoff) 1604a90f590aSDominik Brodowski { 1605a90f590aSDominik Brodowski return ksys_mmap_pgoff(addr, len, prot, flags, fd, pgoff); 1606a90f590aSDominik Brodowski } 1607a90f590aSDominik Brodowski 1608a4679373SChristoph Hellwig #ifdef __ARCH_WANT_SYS_OLD_MMAP 1609a4679373SChristoph Hellwig struct mmap_arg_struct { 1610a4679373SChristoph Hellwig unsigned long addr; 1611a4679373SChristoph Hellwig unsigned long len; 1612a4679373SChristoph Hellwig unsigned long prot; 1613a4679373SChristoph Hellwig unsigned long flags; 1614a4679373SChristoph Hellwig unsigned long fd; 1615a4679373SChristoph Hellwig unsigned long offset; 1616a4679373SChristoph Hellwig }; 1617a4679373SChristoph Hellwig 1618a4679373SChristoph Hellwig SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg) 1619a4679373SChristoph Hellwig { 1620a4679373SChristoph Hellwig struct mmap_arg_struct a; 1621a4679373SChristoph Hellwig 1622a4679373SChristoph Hellwig if (copy_from_user(&a, arg, sizeof(a))) 1623a4679373SChristoph Hellwig return -EFAULT; 1624de1741a1SAlexander Kuleshov if (offset_in_page(a.offset)) 1625a4679373SChristoph Hellwig return -EINVAL; 1626a4679373SChristoph Hellwig 1627a90f590aSDominik Brodowski return ksys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd, 1628a4679373SChristoph Hellwig a.offset >> PAGE_SHIFT); 1629a4679373SChristoph Hellwig } 1630a4679373SChristoph Hellwig #endif /* __ARCH_WANT_SYS_OLD_MMAP */ 1631a4679373SChristoph Hellwig 16324e950f6fSAlexey Dobriyan /* 16338bb4e7a2SWei Yang * Some shared mappings will want the pages marked read-only 16344e950f6fSAlexey Dobriyan * to track write events. If so, we'll downgrade vm_page_prot 16354e950f6fSAlexey Dobriyan * to the private version (using protection_map[] without the 16364e950f6fSAlexey Dobriyan * VM_SHARED bit). 16374e950f6fSAlexey Dobriyan */ 16386d2329f8SAndrea Arcangeli int vma_wants_writenotify(struct vm_area_struct *vma, pgprot_t vm_page_prot) 16394e950f6fSAlexey Dobriyan { 1640ca16d140SKOSAKI Motohiro vm_flags_t vm_flags = vma->vm_flags; 16418a04446aSKirill A. Shutemov const struct vm_operations_struct *vm_ops = vma->vm_ops; 16424e950f6fSAlexey Dobriyan 16434e950f6fSAlexey Dobriyan /* If it was private or non-writable, the write bit is already clear */ 16444e950f6fSAlexey Dobriyan if ((vm_flags & (VM_WRITE|VM_SHARED)) != ((VM_WRITE|VM_SHARED))) 16454e950f6fSAlexey Dobriyan return 0; 16464e950f6fSAlexey Dobriyan 16474e950f6fSAlexey Dobriyan /* The backer wishes to know when pages are first written to? */ 16488a04446aSKirill A. Shutemov if (vm_ops && (vm_ops->page_mkwrite || vm_ops->pfn_mkwrite)) 16494e950f6fSAlexey Dobriyan return 1; 16504e950f6fSAlexey Dobriyan 165164e45507SPeter Feiner /* The open routine did something to the protections that pgprot_modify 165264e45507SPeter Feiner * won't preserve? */ 16536d2329f8SAndrea Arcangeli if (pgprot_val(vm_page_prot) != 16546d2329f8SAndrea Arcangeli pgprot_val(vm_pgprot_modify(vm_page_prot, vm_flags))) 16554e950f6fSAlexey Dobriyan return 0; 16564e950f6fSAlexey Dobriyan 165764e45507SPeter Feiner /* Do we need to track softdirty? */ 165864e45507SPeter Feiner if (IS_ENABLED(CONFIG_MEM_SOFT_DIRTY) && !(vm_flags & VM_SOFTDIRTY)) 165964e45507SPeter Feiner return 1; 166064e45507SPeter Feiner 16614e950f6fSAlexey Dobriyan /* Specialty mapping? */ 16624b6e1e37SKonstantin Khlebnikov if (vm_flags & VM_PFNMAP) 16634e950f6fSAlexey Dobriyan return 0; 16644e950f6fSAlexey Dobriyan 16654e950f6fSAlexey Dobriyan /* Can the mapping track the dirty pages? */ 16664e950f6fSAlexey Dobriyan return vma->vm_file && vma->vm_file->f_mapping && 1667f56753acSChristoph Hellwig mapping_can_writeback(vma->vm_file->f_mapping); 16684e950f6fSAlexey Dobriyan } 16694e950f6fSAlexey Dobriyan 1670fc8744adSLinus Torvalds /* 1671fc8744adSLinus Torvalds * We account for memory if it's a private writeable mapping, 16725a6fe125SMel Gorman * not hugepages and VM_NORESERVE wasn't set. 1673fc8744adSLinus Torvalds */ 1674ca16d140SKOSAKI Motohiro static inline int accountable_mapping(struct file *file, vm_flags_t vm_flags) 1675fc8744adSLinus Torvalds { 16765a6fe125SMel Gorman /* 16775a6fe125SMel Gorman * hugetlb has its own accounting separate from the core VM 16785a6fe125SMel Gorman * VM_HUGETLB may not be set yet so we cannot check for that flag. 16795a6fe125SMel Gorman */ 16805a6fe125SMel Gorman if (file && is_file_hugepages(file)) 16815a6fe125SMel Gorman return 0; 16825a6fe125SMel Gorman 1683fc8744adSLinus Torvalds return (vm_flags & (VM_NORESERVE | VM_SHARED | VM_WRITE)) == VM_WRITE; 1684fc8744adSLinus Torvalds } 1685fc8744adSLinus Torvalds 16860165ab44SMiklos Szeredi unsigned long mmap_region(struct file *file, unsigned long addr, 1687897ab3e0SMike Rapoport unsigned long len, vm_flags_t vm_flags, unsigned long pgoff, 1688897ab3e0SMike Rapoport struct list_head *uf) 16890165ab44SMiklos Szeredi { 16900165ab44SMiklos Szeredi struct mm_struct *mm = current->mm; 1691d70cec89SMiaohe Lin struct vm_area_struct *vma, *prev, *merge; 16920165ab44SMiklos Szeredi int error; 16930165ab44SMiklos Szeredi struct rb_node **rb_link, *rb_parent; 16940165ab44SMiklos Szeredi unsigned long charged = 0; 16950165ab44SMiklos Szeredi 1696e8420a8eSCyril Hrubis /* Check against address space limit. */ 169784638335SKonstantin Khlebnikov if (!may_expand_vm(mm, vm_flags, len >> PAGE_SHIFT)) { 1698e8420a8eSCyril Hrubis unsigned long nr_pages; 1699e8420a8eSCyril Hrubis 1700e8420a8eSCyril Hrubis /* 1701e8420a8eSCyril Hrubis * MAP_FIXED may remove pages of mappings that intersects with 1702e8420a8eSCyril Hrubis * requested mapping. Account for the pages it would unmap. 1703e8420a8eSCyril Hrubis */ 1704e8420a8eSCyril Hrubis nr_pages = count_vma_pages_range(mm, addr, addr + len); 1705e8420a8eSCyril Hrubis 170684638335SKonstantin Khlebnikov if (!may_expand_vm(mm, vm_flags, 170784638335SKonstantin Khlebnikov (len >> PAGE_SHIFT) - nr_pages)) 1708e8420a8eSCyril Hrubis return -ENOMEM; 1709e8420a8eSCyril Hrubis } 1710e8420a8eSCyril Hrubis 17111da177e4SLinus Torvalds /* Clear old maps */ 17129fcd1457SRasmus Villemoes while (find_vma_links(mm, addr, addr + len, &prev, &rb_link, 17139fcd1457SRasmus Villemoes &rb_parent)) { 1714897ab3e0SMike Rapoport if (do_munmap(mm, addr, len, uf)) 17151da177e4SLinus Torvalds return -ENOMEM; 17161da177e4SLinus Torvalds } 17171da177e4SLinus Torvalds 1718fc8744adSLinus Torvalds /* 17191da177e4SLinus Torvalds * Private writable mapping: check memory availability 17201da177e4SLinus Torvalds */ 17215a6fe125SMel Gorman if (accountable_mapping(file, vm_flags)) { 17221da177e4SLinus Torvalds charged = len >> PAGE_SHIFT; 1723191c5424SAl Viro if (security_vm_enough_memory_mm(mm, charged)) 17241da177e4SLinus Torvalds return -ENOMEM; 17251da177e4SLinus Torvalds vm_flags |= VM_ACCOUNT; 17261da177e4SLinus Torvalds } 17271da177e4SLinus Torvalds 17281da177e4SLinus Torvalds /* 1729de33c8dbSLinus Torvalds * Can we just expand an old mapping? 17301da177e4SLinus Torvalds */ 173119a809afSAndrea Arcangeli vma = vma_merge(mm, prev, addr, addr + len, vm_flags, 173219a809afSAndrea Arcangeli NULL, file, pgoff, NULL, NULL_VM_UFFD_CTX); 1733ba470de4SRik van Riel if (vma) 17341da177e4SLinus Torvalds goto out; 17351da177e4SLinus Torvalds 17361da177e4SLinus Torvalds /* 17371da177e4SLinus Torvalds * Determine the object being mapped and call the appropriate 17381da177e4SLinus Torvalds * specific mapper. the address has already been validated, but 17391da177e4SLinus Torvalds * not unmapped, but the maps are removed from the list. 17401da177e4SLinus Torvalds */ 1741490fc053SLinus Torvalds vma = vm_area_alloc(mm); 17421da177e4SLinus Torvalds if (!vma) { 17431da177e4SLinus Torvalds error = -ENOMEM; 17441da177e4SLinus Torvalds goto unacct_error; 17451da177e4SLinus Torvalds } 17461da177e4SLinus Torvalds 17471da177e4SLinus Torvalds vma->vm_start = addr; 17481da177e4SLinus Torvalds vma->vm_end = addr + len; 17491da177e4SLinus Torvalds vma->vm_flags = vm_flags; 17503ed75eb8SColy Li vma->vm_page_prot = vm_get_page_prot(vm_flags); 17511da177e4SLinus Torvalds vma->vm_pgoff = pgoff; 17521da177e4SLinus Torvalds 17531da177e4SLinus Torvalds if (file) { 17541da177e4SLinus Torvalds if (vm_flags & VM_DENYWRITE) { 17551da177e4SLinus Torvalds error = deny_write_access(file); 17561da177e4SLinus Torvalds if (error) 17571da177e4SLinus Torvalds goto free_vma; 17581da177e4SLinus Torvalds } 17594bb5f5d9SDavid Herrmann if (vm_flags & VM_SHARED) { 17604bb5f5d9SDavid Herrmann error = mapping_map_writable(file->f_mapping); 17614bb5f5d9SDavid Herrmann if (error) 17624bb5f5d9SDavid Herrmann goto allow_write_and_free_vma; 17634bb5f5d9SDavid Herrmann } 17644bb5f5d9SDavid Herrmann 17654bb5f5d9SDavid Herrmann /* ->mmap() can change vma->vm_file, but must guarantee that 17664bb5f5d9SDavid Herrmann * vma_link() below can deny write-access if VM_DENYWRITE is set 17674bb5f5d9SDavid Herrmann * and map writably if VM_SHARED is set. This usually means the 17684bb5f5d9SDavid Herrmann * new file must not have been exposed to user-space, yet. 17694bb5f5d9SDavid Herrmann */ 1770cb0942b8SAl Viro vma->vm_file = get_file(file); 1771f74ac015SMiklos Szeredi error = call_mmap(file, vma); 17721da177e4SLinus Torvalds if (error) 17731da177e4SLinus Torvalds goto unmap_and_free_vma; 17741da177e4SLinus Torvalds 1775d70cec89SMiaohe Lin /* If vm_flags changed after call_mmap(), we should try merge vma again 1776d70cec89SMiaohe Lin * as we may succeed this time. 1777d70cec89SMiaohe Lin */ 1778d70cec89SMiaohe Lin if (unlikely(vm_flags != vma->vm_flags && prev)) { 1779d70cec89SMiaohe Lin merge = vma_merge(mm, prev, vma->vm_start, vma->vm_end, vma->vm_flags, 1780d70cec89SMiaohe Lin NULL, vma->vm_file, vma->vm_pgoff, NULL, NULL_VM_UFFD_CTX); 1781d70cec89SMiaohe Lin if (merge) { 1782bc4fe4cdSMiaohe Lin /* ->mmap() can change vma->vm_file and fput the original file. So 1783bc4fe4cdSMiaohe Lin * fput the vma->vm_file here or we would add an extra fput for file 1784bc4fe4cdSMiaohe Lin * and cause general protection fault ultimately. 1785bc4fe4cdSMiaohe Lin */ 1786bc4fe4cdSMiaohe Lin fput(vma->vm_file); 1787d70cec89SMiaohe Lin vm_area_free(vma); 1788d70cec89SMiaohe Lin vma = merge; 1789d70cec89SMiaohe Lin /* Update vm_flags and possible addr to pick up the change. We don't 1790d70cec89SMiaohe Lin * warn here if addr changed as the vma is not linked by vma_link(). 1791d70cec89SMiaohe Lin */ 1792d70cec89SMiaohe Lin addr = vma->vm_start; 1793d70cec89SMiaohe Lin vm_flags = vma->vm_flags; 1794d70cec89SMiaohe Lin goto unmap_writable; 1795d70cec89SMiaohe Lin } 1796d70cec89SMiaohe Lin } 1797d70cec89SMiaohe Lin 17981da177e4SLinus Torvalds /* Can addr have changed?? 17991da177e4SLinus Torvalds * 18001da177e4SLinus Torvalds * Answer: Yes, several device drivers can do it in their 18011da177e4SLinus Torvalds * f_op->mmap method. -DaveM 18022897b4d2SJoonsoo Kim * Bug: If addr is changed, prev, rb_link, rb_parent should 18032897b4d2SJoonsoo Kim * be updated for vma_link() 18041da177e4SLinus Torvalds */ 18052897b4d2SJoonsoo Kim WARN_ON_ONCE(addr != vma->vm_start); 18062897b4d2SJoonsoo Kim 18071da177e4SLinus Torvalds addr = vma->vm_start; 18081da177e4SLinus Torvalds vm_flags = vma->vm_flags; 1809f8dbf0a7SHuang Shijie } else if (vm_flags & VM_SHARED) { 1810f8dbf0a7SHuang Shijie error = shmem_zero_setup(vma); 1811f8dbf0a7SHuang Shijie if (error) 1812f8dbf0a7SHuang Shijie goto free_vma; 1813bfd40eafSKirill A. Shutemov } else { 1814bfd40eafSKirill A. Shutemov vma_set_anonymous(vma); 1815f8dbf0a7SHuang Shijie } 18161da177e4SLinus Torvalds 1817c462ac28SCatalin Marinas /* Allow architectures to sanity-check the vm_flags */ 1818c462ac28SCatalin Marinas if (!arch_validate_flags(vma->vm_flags)) { 1819c462ac28SCatalin Marinas error = -EINVAL; 1820c462ac28SCatalin Marinas if (file) 1821c462ac28SCatalin Marinas goto unmap_and_free_vma; 1822c462ac28SCatalin Marinas else 1823c462ac28SCatalin Marinas goto free_vma; 1824c462ac28SCatalin Marinas } 1825c462ac28SCatalin Marinas 18264d3d5b41SOleg Nesterov vma_link(mm, vma, prev, rb_link, rb_parent); 18274d3d5b41SOleg Nesterov /* Once vma denies write, undo our temporary denial count */ 18284bb5f5d9SDavid Herrmann if (file) { 1829d70cec89SMiaohe Lin unmap_writable: 18304bb5f5d9SDavid Herrmann if (vm_flags & VM_SHARED) 18314bb5f5d9SDavid Herrmann mapping_unmap_writable(file->f_mapping); 1832e8686772SOleg Nesterov if (vm_flags & VM_DENYWRITE) 1833e8686772SOleg Nesterov allow_write_access(file); 18344bb5f5d9SDavid Herrmann } 1835e8686772SOleg Nesterov file = vma->vm_file; 18361da177e4SLinus Torvalds out: 1837cdd6c482SIngo Molnar perf_event_mmap(vma); 18380a4a9391SPeter Zijlstra 183984638335SKonstantin Khlebnikov vm_stat_account(mm, vm_flags, len >> PAGE_SHIFT); 18401da177e4SLinus Torvalds if (vm_flags & VM_LOCKED) { 1841e1fb4a08SDave Jiang if ((vm_flags & VM_SPECIAL) || vma_is_dax(vma) || 1842e1fb4a08SDave Jiang is_vm_hugetlb_page(vma) || 1843e1fb4a08SDave Jiang vma == get_gate_vma(current->mm)) 1844de60f5f1SEric B Munson vma->vm_flags &= VM_LOCKED_CLEAR_MASK; 1845e1fb4a08SDave Jiang else 1846e1fb4a08SDave Jiang mm->locked_vm += (len >> PAGE_SHIFT); 1847bebeb3d6SMichel Lespinasse } 18482b144498SSrikar Dronamraju 1849c7a3a88cSOleg Nesterov if (file) 1850c7a3a88cSOleg Nesterov uprobe_mmap(vma); 18512b144498SSrikar Dronamraju 1852d9104d1cSCyrill Gorcunov /* 1853d9104d1cSCyrill Gorcunov * New (or expanded) vma always get soft dirty status. 1854d9104d1cSCyrill Gorcunov * Otherwise user-space soft-dirty page tracker won't 1855d9104d1cSCyrill Gorcunov * be able to distinguish situation when vma area unmapped, 1856d9104d1cSCyrill Gorcunov * then new mapped in-place (which must be aimed as 1857d9104d1cSCyrill Gorcunov * a completely new data area). 1858d9104d1cSCyrill Gorcunov */ 1859d9104d1cSCyrill Gorcunov vma->vm_flags |= VM_SOFTDIRTY; 1860d9104d1cSCyrill Gorcunov 186164e45507SPeter Feiner vma_set_page_prot(vma); 186264e45507SPeter Feiner 18631da177e4SLinus Torvalds return addr; 18641da177e4SLinus Torvalds 18651da177e4SLinus Torvalds unmap_and_free_vma: 18661da177e4SLinus Torvalds vma->vm_file = NULL; 18671da177e4SLinus Torvalds fput(file); 18681da177e4SLinus Torvalds 18691da177e4SLinus Torvalds /* Undo any partial mapping done by a device driver. */ 1870e0da382cSHugh Dickins unmap_region(mm, vma, prev, vma->vm_start, vma->vm_end); 1871e0da382cSHugh Dickins charged = 0; 18724bb5f5d9SDavid Herrmann if (vm_flags & VM_SHARED) 18734bb5f5d9SDavid Herrmann mapping_unmap_writable(file->f_mapping); 18744bb5f5d9SDavid Herrmann allow_write_and_free_vma: 18754bb5f5d9SDavid Herrmann if (vm_flags & VM_DENYWRITE) 18764bb5f5d9SDavid Herrmann allow_write_access(file); 18771da177e4SLinus Torvalds free_vma: 18783928d4f5SLinus Torvalds vm_area_free(vma); 18791da177e4SLinus Torvalds unacct_error: 18801da177e4SLinus Torvalds if (charged) 18811da177e4SLinus Torvalds vm_unacct_memory(charged); 18821da177e4SLinus Torvalds return error; 18831da177e4SLinus Torvalds } 18841da177e4SLinus Torvalds 1885baceaf1cSJaewon Kim static unsigned long unmapped_area(struct vm_unmapped_area_info *info) 1886db4fbfb9SMichel Lespinasse { 1887db4fbfb9SMichel Lespinasse /* 1888db4fbfb9SMichel Lespinasse * We implement the search by looking for an rbtree node that 1889db4fbfb9SMichel Lespinasse * immediately follows a suitable gap. That is, 1890db4fbfb9SMichel Lespinasse * - gap_start = vma->vm_prev->vm_end <= info->high_limit - length; 1891db4fbfb9SMichel Lespinasse * - gap_end = vma->vm_start >= info->low_limit + length; 1892db4fbfb9SMichel Lespinasse * - gap_end - gap_start >= length 1893db4fbfb9SMichel Lespinasse */ 1894db4fbfb9SMichel Lespinasse 1895db4fbfb9SMichel Lespinasse struct mm_struct *mm = current->mm; 1896db4fbfb9SMichel Lespinasse struct vm_area_struct *vma; 1897db4fbfb9SMichel Lespinasse unsigned long length, low_limit, high_limit, gap_start, gap_end; 1898db4fbfb9SMichel Lespinasse 1899db4fbfb9SMichel Lespinasse /* Adjust search length to account for worst case alignment overhead */ 1900db4fbfb9SMichel Lespinasse length = info->length + info->align_mask; 1901db4fbfb9SMichel Lespinasse if (length < info->length) 1902db4fbfb9SMichel Lespinasse return -ENOMEM; 1903db4fbfb9SMichel Lespinasse 1904db4fbfb9SMichel Lespinasse /* Adjust search limits by the desired length */ 1905db4fbfb9SMichel Lespinasse if (info->high_limit < length) 1906db4fbfb9SMichel Lespinasse return -ENOMEM; 1907db4fbfb9SMichel Lespinasse high_limit = info->high_limit - length; 1908db4fbfb9SMichel Lespinasse 1909db4fbfb9SMichel Lespinasse if (info->low_limit > high_limit) 1910db4fbfb9SMichel Lespinasse return -ENOMEM; 1911db4fbfb9SMichel Lespinasse low_limit = info->low_limit + length; 1912db4fbfb9SMichel Lespinasse 1913db4fbfb9SMichel Lespinasse /* Check if rbtree root looks promising */ 1914db4fbfb9SMichel Lespinasse if (RB_EMPTY_ROOT(&mm->mm_rb)) 1915db4fbfb9SMichel Lespinasse goto check_highest; 1916db4fbfb9SMichel Lespinasse vma = rb_entry(mm->mm_rb.rb_node, struct vm_area_struct, vm_rb); 1917db4fbfb9SMichel Lespinasse if (vma->rb_subtree_gap < length) 1918db4fbfb9SMichel Lespinasse goto check_highest; 1919db4fbfb9SMichel Lespinasse 1920db4fbfb9SMichel Lespinasse while (true) { 1921db4fbfb9SMichel Lespinasse /* Visit left subtree if it looks promising */ 19221be7107fSHugh Dickins gap_end = vm_start_gap(vma); 1923db4fbfb9SMichel Lespinasse if (gap_end >= low_limit && vma->vm_rb.rb_left) { 1924db4fbfb9SMichel Lespinasse struct vm_area_struct *left = 1925db4fbfb9SMichel Lespinasse rb_entry(vma->vm_rb.rb_left, 1926db4fbfb9SMichel Lespinasse struct vm_area_struct, vm_rb); 1927db4fbfb9SMichel Lespinasse if (left->rb_subtree_gap >= length) { 1928db4fbfb9SMichel Lespinasse vma = left; 1929db4fbfb9SMichel Lespinasse continue; 1930db4fbfb9SMichel Lespinasse } 1931db4fbfb9SMichel Lespinasse } 1932db4fbfb9SMichel Lespinasse 19331be7107fSHugh Dickins gap_start = vma->vm_prev ? vm_end_gap(vma->vm_prev) : 0; 1934db4fbfb9SMichel Lespinasse check_current: 1935db4fbfb9SMichel Lespinasse /* Check if current node has a suitable gap */ 1936db4fbfb9SMichel Lespinasse if (gap_start > high_limit) 1937db4fbfb9SMichel Lespinasse return -ENOMEM; 1938f4cb767dSHugh Dickins if (gap_end >= low_limit && 1939f4cb767dSHugh Dickins gap_end > gap_start && gap_end - gap_start >= length) 1940db4fbfb9SMichel Lespinasse goto found; 1941db4fbfb9SMichel Lespinasse 1942db4fbfb9SMichel Lespinasse /* Visit right subtree if it looks promising */ 1943db4fbfb9SMichel Lespinasse if (vma->vm_rb.rb_right) { 1944db4fbfb9SMichel Lespinasse struct vm_area_struct *right = 1945db4fbfb9SMichel Lespinasse rb_entry(vma->vm_rb.rb_right, 1946db4fbfb9SMichel Lespinasse struct vm_area_struct, vm_rb); 1947db4fbfb9SMichel Lespinasse if (right->rb_subtree_gap >= length) { 1948db4fbfb9SMichel Lespinasse vma = right; 1949db4fbfb9SMichel Lespinasse continue; 1950db4fbfb9SMichel Lespinasse } 1951db4fbfb9SMichel Lespinasse } 1952db4fbfb9SMichel Lespinasse 1953db4fbfb9SMichel Lespinasse /* Go back up the rbtree to find next candidate node */ 1954db4fbfb9SMichel Lespinasse while (true) { 1955db4fbfb9SMichel Lespinasse struct rb_node *prev = &vma->vm_rb; 1956db4fbfb9SMichel Lespinasse if (!rb_parent(prev)) 1957db4fbfb9SMichel Lespinasse goto check_highest; 1958db4fbfb9SMichel Lespinasse vma = rb_entry(rb_parent(prev), 1959db4fbfb9SMichel Lespinasse struct vm_area_struct, vm_rb); 1960db4fbfb9SMichel Lespinasse if (prev == vma->vm_rb.rb_left) { 19611be7107fSHugh Dickins gap_start = vm_end_gap(vma->vm_prev); 19621be7107fSHugh Dickins gap_end = vm_start_gap(vma); 1963db4fbfb9SMichel Lespinasse goto check_current; 1964db4fbfb9SMichel Lespinasse } 1965db4fbfb9SMichel Lespinasse } 1966db4fbfb9SMichel Lespinasse } 1967db4fbfb9SMichel Lespinasse 1968db4fbfb9SMichel Lespinasse check_highest: 1969db4fbfb9SMichel Lespinasse /* Check highest gap, which does not precede any rbtree node */ 1970db4fbfb9SMichel Lespinasse gap_start = mm->highest_vm_end; 1971db4fbfb9SMichel Lespinasse gap_end = ULONG_MAX; /* Only for VM_BUG_ON below */ 1972db4fbfb9SMichel Lespinasse if (gap_start > high_limit) 1973db4fbfb9SMichel Lespinasse return -ENOMEM; 1974db4fbfb9SMichel Lespinasse 1975db4fbfb9SMichel Lespinasse found: 1976db4fbfb9SMichel Lespinasse /* We found a suitable gap. Clip it with the original low_limit. */ 1977db4fbfb9SMichel Lespinasse if (gap_start < info->low_limit) 1978db4fbfb9SMichel Lespinasse gap_start = info->low_limit; 1979db4fbfb9SMichel Lespinasse 1980db4fbfb9SMichel Lespinasse /* Adjust gap address to the desired alignment */ 1981db4fbfb9SMichel Lespinasse gap_start += (info->align_offset - gap_start) & info->align_mask; 1982db4fbfb9SMichel Lespinasse 1983db4fbfb9SMichel Lespinasse VM_BUG_ON(gap_start + info->length > info->high_limit); 1984db4fbfb9SMichel Lespinasse VM_BUG_ON(gap_start + info->length > gap_end); 1985db4fbfb9SMichel Lespinasse return gap_start; 1986db4fbfb9SMichel Lespinasse } 1987db4fbfb9SMichel Lespinasse 1988baceaf1cSJaewon Kim static unsigned long unmapped_area_topdown(struct vm_unmapped_area_info *info) 1989db4fbfb9SMichel Lespinasse { 1990db4fbfb9SMichel Lespinasse struct mm_struct *mm = current->mm; 1991db4fbfb9SMichel Lespinasse struct vm_area_struct *vma; 1992db4fbfb9SMichel Lespinasse unsigned long length, low_limit, high_limit, gap_start, gap_end; 1993db4fbfb9SMichel Lespinasse 1994db4fbfb9SMichel Lespinasse /* Adjust search length to account for worst case alignment overhead */ 1995db4fbfb9SMichel Lespinasse length = info->length + info->align_mask; 1996db4fbfb9SMichel Lespinasse if (length < info->length) 1997db4fbfb9SMichel Lespinasse return -ENOMEM; 1998db4fbfb9SMichel Lespinasse 1999db4fbfb9SMichel Lespinasse /* 2000db4fbfb9SMichel Lespinasse * Adjust search limits by the desired length. 2001db4fbfb9SMichel Lespinasse * See implementation comment at top of unmapped_area(). 2002db4fbfb9SMichel Lespinasse */ 2003db4fbfb9SMichel Lespinasse gap_end = info->high_limit; 2004db4fbfb9SMichel Lespinasse if (gap_end < length) 2005db4fbfb9SMichel Lespinasse return -ENOMEM; 2006db4fbfb9SMichel Lespinasse high_limit = gap_end - length; 2007db4fbfb9SMichel Lespinasse 2008db4fbfb9SMichel Lespinasse if (info->low_limit > high_limit) 2009db4fbfb9SMichel Lespinasse return -ENOMEM; 2010db4fbfb9SMichel Lespinasse low_limit = info->low_limit + length; 2011db4fbfb9SMichel Lespinasse 2012db4fbfb9SMichel Lespinasse /* Check highest gap, which does not precede any rbtree node */ 2013db4fbfb9SMichel Lespinasse gap_start = mm->highest_vm_end; 2014db4fbfb9SMichel Lespinasse if (gap_start <= high_limit) 2015db4fbfb9SMichel Lespinasse goto found_highest; 2016db4fbfb9SMichel Lespinasse 2017db4fbfb9SMichel Lespinasse /* Check if rbtree root looks promising */ 2018db4fbfb9SMichel Lespinasse if (RB_EMPTY_ROOT(&mm->mm_rb)) 2019db4fbfb9SMichel Lespinasse return -ENOMEM; 2020db4fbfb9SMichel Lespinasse vma = rb_entry(mm->mm_rb.rb_node, struct vm_area_struct, vm_rb); 2021db4fbfb9SMichel Lespinasse if (vma->rb_subtree_gap < length) 2022db4fbfb9SMichel Lespinasse return -ENOMEM; 2023db4fbfb9SMichel Lespinasse 2024db4fbfb9SMichel Lespinasse while (true) { 2025db4fbfb9SMichel Lespinasse /* Visit right subtree if it looks promising */ 20261be7107fSHugh Dickins gap_start = vma->vm_prev ? vm_end_gap(vma->vm_prev) : 0; 2027db4fbfb9SMichel Lespinasse if (gap_start <= high_limit && vma->vm_rb.rb_right) { 2028db4fbfb9SMichel Lespinasse struct vm_area_struct *right = 2029db4fbfb9SMichel Lespinasse rb_entry(vma->vm_rb.rb_right, 2030db4fbfb9SMichel Lespinasse struct vm_area_struct, vm_rb); 2031db4fbfb9SMichel Lespinasse if (right->rb_subtree_gap >= length) { 2032db4fbfb9SMichel Lespinasse vma = right; 2033db4fbfb9SMichel Lespinasse continue; 2034db4fbfb9SMichel Lespinasse } 2035db4fbfb9SMichel Lespinasse } 2036db4fbfb9SMichel Lespinasse 2037db4fbfb9SMichel Lespinasse check_current: 2038db4fbfb9SMichel Lespinasse /* Check if current node has a suitable gap */ 20391be7107fSHugh Dickins gap_end = vm_start_gap(vma); 2040db4fbfb9SMichel Lespinasse if (gap_end < low_limit) 2041db4fbfb9SMichel Lespinasse return -ENOMEM; 2042f4cb767dSHugh Dickins if (gap_start <= high_limit && 2043f4cb767dSHugh Dickins gap_end > gap_start && gap_end - gap_start >= length) 2044db4fbfb9SMichel Lespinasse goto found; 2045db4fbfb9SMichel Lespinasse 2046db4fbfb9SMichel Lespinasse /* Visit left subtree if it looks promising */ 2047db4fbfb9SMichel Lespinasse if (vma->vm_rb.rb_left) { 2048db4fbfb9SMichel Lespinasse struct vm_area_struct *left = 2049db4fbfb9SMichel Lespinasse rb_entry(vma->vm_rb.rb_left, 2050db4fbfb9SMichel Lespinasse struct vm_area_struct, vm_rb); 2051db4fbfb9SMichel Lespinasse if (left->rb_subtree_gap >= length) { 2052db4fbfb9SMichel Lespinasse vma = left; 2053db4fbfb9SMichel Lespinasse continue; 2054db4fbfb9SMichel Lespinasse } 2055db4fbfb9SMichel Lespinasse } 2056db4fbfb9SMichel Lespinasse 2057db4fbfb9SMichel Lespinasse /* Go back up the rbtree to find next candidate node */ 2058db4fbfb9SMichel Lespinasse while (true) { 2059db4fbfb9SMichel Lespinasse struct rb_node *prev = &vma->vm_rb; 2060db4fbfb9SMichel Lespinasse if (!rb_parent(prev)) 2061db4fbfb9SMichel Lespinasse return -ENOMEM; 2062db4fbfb9SMichel Lespinasse vma = rb_entry(rb_parent(prev), 2063db4fbfb9SMichel Lespinasse struct vm_area_struct, vm_rb); 2064db4fbfb9SMichel Lespinasse if (prev == vma->vm_rb.rb_right) { 2065db4fbfb9SMichel Lespinasse gap_start = vma->vm_prev ? 20661be7107fSHugh Dickins vm_end_gap(vma->vm_prev) : 0; 2067db4fbfb9SMichel Lespinasse goto check_current; 2068db4fbfb9SMichel Lespinasse } 2069db4fbfb9SMichel Lespinasse } 2070db4fbfb9SMichel Lespinasse } 2071db4fbfb9SMichel Lespinasse 2072db4fbfb9SMichel Lespinasse found: 2073db4fbfb9SMichel Lespinasse /* We found a suitable gap. Clip it with the original high_limit. */ 2074db4fbfb9SMichel Lespinasse if (gap_end > info->high_limit) 2075db4fbfb9SMichel Lespinasse gap_end = info->high_limit; 2076db4fbfb9SMichel Lespinasse 2077db4fbfb9SMichel Lespinasse found_highest: 2078db4fbfb9SMichel Lespinasse /* Compute highest gap address at the desired alignment */ 2079db4fbfb9SMichel Lespinasse gap_end -= info->length; 2080db4fbfb9SMichel Lespinasse gap_end -= (gap_end - info->align_offset) & info->align_mask; 2081db4fbfb9SMichel Lespinasse 2082db4fbfb9SMichel Lespinasse VM_BUG_ON(gap_end < info->low_limit); 2083db4fbfb9SMichel Lespinasse VM_BUG_ON(gap_end < gap_start); 2084db4fbfb9SMichel Lespinasse return gap_end; 2085db4fbfb9SMichel Lespinasse } 2086db4fbfb9SMichel Lespinasse 2087baceaf1cSJaewon Kim /* 2088baceaf1cSJaewon Kim * Search for an unmapped address range. 2089baceaf1cSJaewon Kim * 2090baceaf1cSJaewon Kim * We are looking for a range that: 2091baceaf1cSJaewon Kim * - does not intersect with any VMA; 2092baceaf1cSJaewon Kim * - is contained within the [low_limit, high_limit) interval; 2093baceaf1cSJaewon Kim * - is at least the desired size. 2094baceaf1cSJaewon Kim * - satisfies (begin_addr & align_mask) == (align_offset & align_mask) 2095baceaf1cSJaewon Kim */ 2096baceaf1cSJaewon Kim unsigned long vm_unmapped_area(struct vm_unmapped_area_info *info) 2097baceaf1cSJaewon Kim { 2098df529cabSJaewon Kim unsigned long addr; 2099df529cabSJaewon Kim 2100baceaf1cSJaewon Kim if (info->flags & VM_UNMAPPED_AREA_TOPDOWN) 2101df529cabSJaewon Kim addr = unmapped_area_topdown(info); 2102baceaf1cSJaewon Kim else 2103df529cabSJaewon Kim addr = unmapped_area(info); 2104df529cabSJaewon Kim 2105df529cabSJaewon Kim trace_vm_unmapped_area(addr, info); 2106df529cabSJaewon Kim return addr; 2107baceaf1cSJaewon Kim } 2108f6795053SSteve Capper 2109f6795053SSteve Capper #ifndef arch_get_mmap_end 2110f6795053SSteve Capper #define arch_get_mmap_end(addr) (TASK_SIZE) 2111f6795053SSteve Capper #endif 2112f6795053SSteve Capper 2113f6795053SSteve Capper #ifndef arch_get_mmap_base 2114f6795053SSteve Capper #define arch_get_mmap_base(addr, base) (base) 2115f6795053SSteve Capper #endif 2116f6795053SSteve Capper 21171da177e4SLinus Torvalds /* Get an address range which is currently unmapped. 21181da177e4SLinus Torvalds * For shmat() with addr=0. 21191da177e4SLinus Torvalds * 21201da177e4SLinus Torvalds * Ugly calling convention alert: 21211da177e4SLinus Torvalds * Return value with the low bits set means error value, 21221da177e4SLinus Torvalds * ie 21231da177e4SLinus Torvalds * if (ret & ~PAGE_MASK) 21241da177e4SLinus Torvalds * error = ret; 21251da177e4SLinus Torvalds * 21261da177e4SLinus Torvalds * This function "knows" that -ENOMEM has the bits set. 21271da177e4SLinus Torvalds */ 21281da177e4SLinus Torvalds #ifndef HAVE_ARCH_UNMAPPED_AREA 21291da177e4SLinus Torvalds unsigned long 21301da177e4SLinus Torvalds arch_get_unmapped_area(struct file *filp, unsigned long addr, 21311da177e4SLinus Torvalds unsigned long len, unsigned long pgoff, unsigned long flags) 21321da177e4SLinus Torvalds { 21331da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 21341be7107fSHugh Dickins struct vm_area_struct *vma, *prev; 2135db4fbfb9SMichel Lespinasse struct vm_unmapped_area_info info; 2136f6795053SSteve Capper const unsigned long mmap_end = arch_get_mmap_end(addr); 21371da177e4SLinus Torvalds 2138f6795053SSteve Capper if (len > mmap_end - mmap_min_addr) 21391da177e4SLinus Torvalds return -ENOMEM; 21401da177e4SLinus Torvalds 214106abdfb4SBenjamin Herrenschmidt if (flags & MAP_FIXED) 214206abdfb4SBenjamin Herrenschmidt return addr; 214306abdfb4SBenjamin Herrenschmidt 21441da177e4SLinus Torvalds if (addr) { 21451da177e4SLinus Torvalds addr = PAGE_ALIGN(addr); 21461be7107fSHugh Dickins vma = find_vma_prev(mm, addr, &prev); 2147f6795053SSteve Capper if (mmap_end - len >= addr && addr >= mmap_min_addr && 21481be7107fSHugh Dickins (!vma || addr + len <= vm_start_gap(vma)) && 21491be7107fSHugh Dickins (!prev || addr >= vm_end_gap(prev))) 21501da177e4SLinus Torvalds return addr; 21511da177e4SLinus Torvalds } 21521da177e4SLinus Torvalds 2153db4fbfb9SMichel Lespinasse info.flags = 0; 2154db4fbfb9SMichel Lespinasse info.length = len; 21554e99b021SHeiko Carstens info.low_limit = mm->mmap_base; 2156f6795053SSteve Capper info.high_limit = mmap_end; 2157db4fbfb9SMichel Lespinasse info.align_mask = 0; 215809ef5283SJaewon Kim info.align_offset = 0; 2159db4fbfb9SMichel Lespinasse return vm_unmapped_area(&info); 21601da177e4SLinus Torvalds } 21611da177e4SLinus Torvalds #endif 21621da177e4SLinus Torvalds 21631da177e4SLinus Torvalds /* 21641da177e4SLinus Torvalds * This mmap-allocator allocates new areas top-down from below the 21651da177e4SLinus Torvalds * stack's low limit (the base): 21661da177e4SLinus Torvalds */ 21671da177e4SLinus Torvalds #ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN 21681da177e4SLinus Torvalds unsigned long 216943cca0b1SYang Fan arch_get_unmapped_area_topdown(struct file *filp, unsigned long addr, 217043cca0b1SYang Fan unsigned long len, unsigned long pgoff, 217143cca0b1SYang Fan unsigned long flags) 21721da177e4SLinus Torvalds { 21731be7107fSHugh Dickins struct vm_area_struct *vma, *prev; 21741da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 2175db4fbfb9SMichel Lespinasse struct vm_unmapped_area_info info; 2176f6795053SSteve Capper const unsigned long mmap_end = arch_get_mmap_end(addr); 21771da177e4SLinus Torvalds 21781da177e4SLinus Torvalds /* requested length too big for entire address space */ 2179f6795053SSteve Capper if (len > mmap_end - mmap_min_addr) 21801da177e4SLinus Torvalds return -ENOMEM; 21811da177e4SLinus Torvalds 218206abdfb4SBenjamin Herrenschmidt if (flags & MAP_FIXED) 218306abdfb4SBenjamin Herrenschmidt return addr; 218406abdfb4SBenjamin Herrenschmidt 21851da177e4SLinus Torvalds /* requesting a specific address */ 21861da177e4SLinus Torvalds if (addr) { 21871da177e4SLinus Torvalds addr = PAGE_ALIGN(addr); 21881be7107fSHugh Dickins vma = find_vma_prev(mm, addr, &prev); 2189f6795053SSteve Capper if (mmap_end - len >= addr && addr >= mmap_min_addr && 21901be7107fSHugh Dickins (!vma || addr + len <= vm_start_gap(vma)) && 21911be7107fSHugh Dickins (!prev || addr >= vm_end_gap(prev))) 21921da177e4SLinus Torvalds return addr; 21931da177e4SLinus Torvalds } 21941da177e4SLinus Torvalds 2195db4fbfb9SMichel Lespinasse info.flags = VM_UNMAPPED_AREA_TOPDOWN; 2196db4fbfb9SMichel Lespinasse info.length = len; 21972afc745fSAkira Takeuchi info.low_limit = max(PAGE_SIZE, mmap_min_addr); 2198f6795053SSteve Capper info.high_limit = arch_get_mmap_base(addr, mm->mmap_base); 2199db4fbfb9SMichel Lespinasse info.align_mask = 0; 220009ef5283SJaewon Kim info.align_offset = 0; 2201db4fbfb9SMichel Lespinasse addr = vm_unmapped_area(&info); 2202b716ad95SXiao Guangrong 22031da177e4SLinus Torvalds /* 22041da177e4SLinus Torvalds * A failed mmap() very likely causes application failure, 22051da177e4SLinus Torvalds * so fall back to the bottom-up function here. This scenario 22061da177e4SLinus Torvalds * can happen with large stack limits and large mmap() 22071da177e4SLinus Torvalds * allocations. 22081da177e4SLinus Torvalds */ 2209de1741a1SAlexander Kuleshov if (offset_in_page(addr)) { 2210db4fbfb9SMichel Lespinasse VM_BUG_ON(addr != -ENOMEM); 2211db4fbfb9SMichel Lespinasse info.flags = 0; 2212db4fbfb9SMichel Lespinasse info.low_limit = TASK_UNMAPPED_BASE; 2213f6795053SSteve Capper info.high_limit = mmap_end; 2214db4fbfb9SMichel Lespinasse addr = vm_unmapped_area(&info); 2215db4fbfb9SMichel Lespinasse } 22161da177e4SLinus Torvalds 22171da177e4SLinus Torvalds return addr; 22181da177e4SLinus Torvalds } 22191da177e4SLinus Torvalds #endif 22201da177e4SLinus Torvalds 22211da177e4SLinus Torvalds unsigned long 22221da177e4SLinus Torvalds get_unmapped_area(struct file *file, unsigned long addr, unsigned long len, 22231da177e4SLinus Torvalds unsigned long pgoff, unsigned long flags) 22241da177e4SLinus Torvalds { 222506abdfb4SBenjamin Herrenschmidt unsigned long (*get_area)(struct file *, unsigned long, 222606abdfb4SBenjamin Herrenschmidt unsigned long, unsigned long, unsigned long); 222707ab67c8SLinus Torvalds 22289206de95SAl Viro unsigned long error = arch_mmap_check(addr, len, flags); 22299206de95SAl Viro if (error) 22309206de95SAl Viro return error; 22319206de95SAl Viro 22329206de95SAl Viro /* Careful about overflows.. */ 22339206de95SAl Viro if (len > TASK_SIZE) 22349206de95SAl Viro return -ENOMEM; 22359206de95SAl Viro 223607ab67c8SLinus Torvalds get_area = current->mm->get_unmapped_area; 2237c01d5b30SHugh Dickins if (file) { 2238c01d5b30SHugh Dickins if (file->f_op->get_unmapped_area) 223907ab67c8SLinus Torvalds get_area = file->f_op->get_unmapped_area; 2240c01d5b30SHugh Dickins } else if (flags & MAP_SHARED) { 2241c01d5b30SHugh Dickins /* 2242c01d5b30SHugh Dickins * mmap_region() will call shmem_zero_setup() to create a file, 2243c01d5b30SHugh Dickins * so use shmem's get_unmapped_area in case it can be huge. 224445e55300SPeter Collingbourne * do_mmap() will clear pgoff, so match alignment. 2245c01d5b30SHugh Dickins */ 2246c01d5b30SHugh Dickins pgoff = 0; 2247c01d5b30SHugh Dickins get_area = shmem_get_unmapped_area; 2248c01d5b30SHugh Dickins } 2249c01d5b30SHugh Dickins 225007ab67c8SLinus Torvalds addr = get_area(file, addr, len, pgoff, flags); 225107ab67c8SLinus Torvalds if (IS_ERR_VALUE(addr)) 225207ab67c8SLinus Torvalds return addr; 225307ab67c8SLinus Torvalds 22541da177e4SLinus Torvalds if (addr > TASK_SIZE - len) 22551da177e4SLinus Torvalds return -ENOMEM; 2256de1741a1SAlexander Kuleshov if (offset_in_page(addr)) 22571da177e4SLinus Torvalds return -EINVAL; 225806abdfb4SBenjamin Herrenschmidt 22599ac4ed4bSAl Viro error = security_mmap_addr(addr); 22609ac4ed4bSAl Viro return error ? error : addr; 22611da177e4SLinus Torvalds } 22621da177e4SLinus Torvalds 22631da177e4SLinus Torvalds EXPORT_SYMBOL(get_unmapped_area); 22641da177e4SLinus Torvalds 22651da177e4SLinus Torvalds /* Look up the first VMA which satisfies addr < vm_end, NULL if none. */ 22661da177e4SLinus Torvalds struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr) 22671da177e4SLinus Torvalds { 2268615d6e87SDavidlohr Bueso struct rb_node *rb_node; 2269615d6e87SDavidlohr Bueso struct vm_area_struct *vma; 22701da177e4SLinus Torvalds 22711da177e4SLinus Torvalds /* Check the cache first. */ 2272615d6e87SDavidlohr Bueso vma = vmacache_find(mm, addr); 2273615d6e87SDavidlohr Bueso if (likely(vma)) 2274615d6e87SDavidlohr Bueso return vma; 22751da177e4SLinus Torvalds 22761da177e4SLinus Torvalds rb_node = mm->mm_rb.rb_node; 22771da177e4SLinus Torvalds 22781da177e4SLinus Torvalds while (rb_node) { 2279615d6e87SDavidlohr Bueso struct vm_area_struct *tmp; 22801da177e4SLinus Torvalds 2281615d6e87SDavidlohr Bueso tmp = rb_entry(rb_node, struct vm_area_struct, vm_rb); 22821da177e4SLinus Torvalds 2283615d6e87SDavidlohr Bueso if (tmp->vm_end > addr) { 2284615d6e87SDavidlohr Bueso vma = tmp; 2285615d6e87SDavidlohr Bueso if (tmp->vm_start <= addr) 22861da177e4SLinus Torvalds break; 22871da177e4SLinus Torvalds rb_node = rb_node->rb_left; 22881da177e4SLinus Torvalds } else 22891da177e4SLinus Torvalds rb_node = rb_node->rb_right; 22901da177e4SLinus Torvalds } 2291615d6e87SDavidlohr Bueso 22921da177e4SLinus Torvalds if (vma) 2293615d6e87SDavidlohr Bueso vmacache_update(addr, vma); 22941da177e4SLinus Torvalds return vma; 22951da177e4SLinus Torvalds } 22961da177e4SLinus Torvalds 22971da177e4SLinus Torvalds EXPORT_SYMBOL(find_vma); 22981da177e4SLinus Torvalds 22996bd4837dSKOSAKI Motohiro /* 23006bd4837dSKOSAKI Motohiro * Same as find_vma, but also return a pointer to the previous VMA in *pprev. 23016bd4837dSKOSAKI Motohiro */ 23021da177e4SLinus Torvalds struct vm_area_struct * 23031da177e4SLinus Torvalds find_vma_prev(struct mm_struct *mm, unsigned long addr, 23041da177e4SLinus Torvalds struct vm_area_struct **pprev) 23051da177e4SLinus Torvalds { 23066bd4837dSKOSAKI Motohiro struct vm_area_struct *vma; 23071da177e4SLinus Torvalds 23086bd4837dSKOSAKI Motohiro vma = find_vma(mm, addr); 230983cd904dSMikulas Patocka if (vma) { 231083cd904dSMikulas Patocka *pprev = vma->vm_prev; 231183cd904dSMikulas Patocka } else { 231273848a97SWei Yang struct rb_node *rb_node = rb_last(&mm->mm_rb); 231373848a97SWei Yang 231473848a97SWei Yang *pprev = rb_node ? rb_entry(rb_node, struct vm_area_struct, vm_rb) : NULL; 231583cd904dSMikulas Patocka } 23166bd4837dSKOSAKI Motohiro return vma; 23171da177e4SLinus Torvalds } 23181da177e4SLinus Torvalds 23191da177e4SLinus Torvalds /* 23201da177e4SLinus Torvalds * Verify that the stack growth is acceptable and 23211da177e4SLinus Torvalds * update accounting. This is shared with both the 23221da177e4SLinus Torvalds * grow-up and grow-down cases. 23231da177e4SLinus Torvalds */ 23241be7107fSHugh Dickins static int acct_stack_growth(struct vm_area_struct *vma, 23251be7107fSHugh Dickins unsigned long size, unsigned long grow) 23261da177e4SLinus Torvalds { 23271da177e4SLinus Torvalds struct mm_struct *mm = vma->vm_mm; 23281be7107fSHugh Dickins unsigned long new_start; 23291da177e4SLinus Torvalds 23301da177e4SLinus Torvalds /* address space limit tests */ 233184638335SKonstantin Khlebnikov if (!may_expand_vm(mm, vma->vm_flags, grow)) 23321da177e4SLinus Torvalds return -ENOMEM; 23331da177e4SLinus Torvalds 23341da177e4SLinus Torvalds /* Stack limit test */ 233524c79d8eSKrzysztof Opasiak if (size > rlimit(RLIMIT_STACK)) 23361da177e4SLinus Torvalds return -ENOMEM; 23371da177e4SLinus Torvalds 23381da177e4SLinus Torvalds /* mlock limit tests */ 23391da177e4SLinus Torvalds if (vma->vm_flags & VM_LOCKED) { 23401da177e4SLinus Torvalds unsigned long locked; 23411da177e4SLinus Torvalds unsigned long limit; 23421da177e4SLinus Torvalds locked = mm->locked_vm + grow; 234324c79d8eSKrzysztof Opasiak limit = rlimit(RLIMIT_MEMLOCK); 234459e99e5bSJiri Slaby limit >>= PAGE_SHIFT; 23451da177e4SLinus Torvalds if (locked > limit && !capable(CAP_IPC_LOCK)) 23461da177e4SLinus Torvalds return -ENOMEM; 23471da177e4SLinus Torvalds } 23481da177e4SLinus Torvalds 23490d59a01bSAdam Litke /* Check to ensure the stack will not grow into a hugetlb-only region */ 23500d59a01bSAdam Litke new_start = (vma->vm_flags & VM_GROWSUP) ? vma->vm_start : 23510d59a01bSAdam Litke vma->vm_end - size; 23520d59a01bSAdam Litke if (is_hugepage_only_range(vma->vm_mm, new_start, size)) 23530d59a01bSAdam Litke return -EFAULT; 23540d59a01bSAdam Litke 23551da177e4SLinus Torvalds /* 23561da177e4SLinus Torvalds * Overcommit.. This must be the final test, as it will 23571da177e4SLinus Torvalds * update security statistics. 23581da177e4SLinus Torvalds */ 235905fa199dSHugh Dickins if (security_vm_enough_memory_mm(mm, grow)) 23601da177e4SLinus Torvalds return -ENOMEM; 23611da177e4SLinus Torvalds 23621da177e4SLinus Torvalds return 0; 23631da177e4SLinus Torvalds } 23641da177e4SLinus Torvalds 236546dea3d0SHugh Dickins #if defined(CONFIG_STACK_GROWSUP) || defined(CONFIG_IA64) 23661da177e4SLinus Torvalds /* 236746dea3d0SHugh Dickins * PA-RISC uses this for its stack; IA64 for its Register Backing Store. 236846dea3d0SHugh Dickins * vma is the last one with address > vma->vm_end. Have to extend vma. 23691da177e4SLinus Torvalds */ 237046dea3d0SHugh Dickins int expand_upwards(struct vm_area_struct *vma, unsigned long address) 23711da177e4SLinus Torvalds { 237209357814SOleg Nesterov struct mm_struct *mm = vma->vm_mm; 23731be7107fSHugh Dickins struct vm_area_struct *next; 23741be7107fSHugh Dickins unsigned long gap_addr; 237512352d3cSKonstantin Khlebnikov int error = 0; 23761da177e4SLinus Torvalds 23771da177e4SLinus Torvalds if (!(vma->vm_flags & VM_GROWSUP)) 23781da177e4SLinus Torvalds return -EFAULT; 23791da177e4SLinus Torvalds 2380bd726c90SHelge Deller /* Guard against exceeding limits of the address space. */ 23811be7107fSHugh Dickins address &= PAGE_MASK; 238237511fb5SHelge Deller if (address >= (TASK_SIZE & PAGE_MASK)) 238312352d3cSKonstantin Khlebnikov return -ENOMEM; 2384bd726c90SHelge Deller address += PAGE_SIZE; 238512352d3cSKonstantin Khlebnikov 23861be7107fSHugh Dickins /* Enforce stack_guard_gap */ 23871be7107fSHugh Dickins gap_addr = address + stack_guard_gap; 2388bd726c90SHelge Deller 2389bd726c90SHelge Deller /* Guard against overflow */ 2390bd726c90SHelge Deller if (gap_addr < address || gap_addr > TASK_SIZE) 2391bd726c90SHelge Deller gap_addr = TASK_SIZE; 2392bd726c90SHelge Deller 23931be7107fSHugh Dickins next = vma->vm_next; 23943122e80eSAnshuman Khandual if (next && next->vm_start < gap_addr && vma_is_accessible(next)) { 23951be7107fSHugh Dickins if (!(next->vm_flags & VM_GROWSUP)) 23961be7107fSHugh Dickins return -ENOMEM; 23971be7107fSHugh Dickins /* Check that both stack segments have the same anon_vma? */ 23981be7107fSHugh Dickins } 23991be7107fSHugh Dickins 240012352d3cSKonstantin Khlebnikov /* We must make sure the anon_vma is allocated. */ 24011da177e4SLinus Torvalds if (unlikely(anon_vma_prepare(vma))) 24021da177e4SLinus Torvalds return -ENOMEM; 24031da177e4SLinus Torvalds 24041da177e4SLinus Torvalds /* 24051da177e4SLinus Torvalds * vma->vm_start/vm_end cannot change under us because the caller 2406c1e8d7c6SMichel Lespinasse * is required to hold the mmap_lock in read mode. We need the 24071da177e4SLinus Torvalds * anon_vma lock to serialize against concurrent expand_stacks. 24081da177e4SLinus Torvalds */ 240912352d3cSKonstantin Khlebnikov anon_vma_lock_write(vma->anon_vma); 24101da177e4SLinus Torvalds 24111da177e4SLinus Torvalds /* Somebody else might have raced and expanded it already */ 24121da177e4SLinus Torvalds if (address > vma->vm_end) { 24131da177e4SLinus Torvalds unsigned long size, grow; 24141da177e4SLinus Torvalds 24151da177e4SLinus Torvalds size = address - vma->vm_start; 24161da177e4SLinus Torvalds grow = (address - vma->vm_end) >> PAGE_SHIFT; 24171da177e4SLinus Torvalds 241842c36f63SHugh Dickins error = -ENOMEM; 241942c36f63SHugh Dickins if (vma->vm_pgoff + (size >> PAGE_SHIFT) >= vma->vm_pgoff) { 24201da177e4SLinus Torvalds error = acct_stack_growth(vma, size, grow); 24213af9e859SEric B Munson if (!error) { 24224128997bSMichel Lespinasse /* 24234128997bSMichel Lespinasse * vma_gap_update() doesn't support concurrent 2424c1e8d7c6SMichel Lespinasse * updates, but we only hold a shared mmap_lock 24254128997bSMichel Lespinasse * lock here, so we need to protect against 24264128997bSMichel Lespinasse * concurrent vma expansions. 242712352d3cSKonstantin Khlebnikov * anon_vma_lock_write() doesn't help here, as 24284128997bSMichel Lespinasse * we don't guarantee that all growable vmas 24294128997bSMichel Lespinasse * in a mm share the same root anon vma. 24304128997bSMichel Lespinasse * So, we reuse mm->page_table_lock to guard 24314128997bSMichel Lespinasse * against concurrent vma expansions. 24324128997bSMichel Lespinasse */ 243309357814SOleg Nesterov spin_lock(&mm->page_table_lock); 243487e8827bSOleg Nesterov if (vma->vm_flags & VM_LOCKED) 243509357814SOleg Nesterov mm->locked_vm += grow; 243684638335SKonstantin Khlebnikov vm_stat_account(mm, vma->vm_flags, grow); 2437bf181b9fSMichel Lespinasse anon_vma_interval_tree_pre_update_vma(vma); 24381da177e4SLinus Torvalds vma->vm_end = address; 2439bf181b9fSMichel Lespinasse anon_vma_interval_tree_post_update_vma(vma); 2440d3737187SMichel Lespinasse if (vma->vm_next) 2441d3737187SMichel Lespinasse vma_gap_update(vma->vm_next); 2442d3737187SMichel Lespinasse else 24431be7107fSHugh Dickins mm->highest_vm_end = vm_end_gap(vma); 244409357814SOleg Nesterov spin_unlock(&mm->page_table_lock); 24454128997bSMichel Lespinasse 24463af9e859SEric B Munson perf_event_mmap(vma); 24473af9e859SEric B Munson } 24481da177e4SLinus Torvalds } 244942c36f63SHugh Dickins } 245012352d3cSKonstantin Khlebnikov anon_vma_unlock_write(vma->anon_vma); 24516d50e60cSDavid Rientjes khugepaged_enter_vma_merge(vma, vma->vm_flags); 245209357814SOleg Nesterov validate_mm(mm); 24531da177e4SLinus Torvalds return error; 24541da177e4SLinus Torvalds } 245546dea3d0SHugh Dickins #endif /* CONFIG_STACK_GROWSUP || CONFIG_IA64 */ 245646dea3d0SHugh Dickins 24571da177e4SLinus Torvalds /* 24581da177e4SLinus Torvalds * vma is the first one with address < vma->vm_start. Have to extend vma. 24591da177e4SLinus Torvalds */ 2460d05f3169SMichal Hocko int expand_downwards(struct vm_area_struct *vma, 2461b6a2fea3SOllie Wild unsigned long address) 24621da177e4SLinus Torvalds { 246309357814SOleg Nesterov struct mm_struct *mm = vma->vm_mm; 24641be7107fSHugh Dickins struct vm_area_struct *prev; 24650a1d5299SJann Horn int error = 0; 24661da177e4SLinus Torvalds 24678869477aSEric Paris address &= PAGE_MASK; 24680a1d5299SJann Horn if (address < mmap_min_addr) 24690a1d5299SJann Horn return -EPERM; 24708869477aSEric Paris 24711be7107fSHugh Dickins /* Enforce stack_guard_gap */ 24721be7107fSHugh Dickins prev = vma->vm_prev; 24731be7107fSHugh Dickins /* Check that both stack segments have the same anon_vma? */ 247432e4e6d5SOleg Nesterov if (prev && !(prev->vm_flags & VM_GROWSDOWN) && 24753122e80eSAnshuman Khandual vma_is_accessible(prev)) { 247632e4e6d5SOleg Nesterov if (address - prev->vm_end < stack_guard_gap) 247732e4e6d5SOleg Nesterov return -ENOMEM; 24781be7107fSHugh Dickins } 24791be7107fSHugh Dickins 248012352d3cSKonstantin Khlebnikov /* We must make sure the anon_vma is allocated. */ 248112352d3cSKonstantin Khlebnikov if (unlikely(anon_vma_prepare(vma))) 248212352d3cSKonstantin Khlebnikov return -ENOMEM; 24831da177e4SLinus Torvalds 24841da177e4SLinus Torvalds /* 24851da177e4SLinus Torvalds * vma->vm_start/vm_end cannot change under us because the caller 2486c1e8d7c6SMichel Lespinasse * is required to hold the mmap_lock in read mode. We need the 24871da177e4SLinus Torvalds * anon_vma lock to serialize against concurrent expand_stacks. 24881da177e4SLinus Torvalds */ 248912352d3cSKonstantin Khlebnikov anon_vma_lock_write(vma->anon_vma); 24901da177e4SLinus Torvalds 24911da177e4SLinus Torvalds /* Somebody else might have raced and expanded it already */ 24921da177e4SLinus Torvalds if (address < vma->vm_start) { 24931da177e4SLinus Torvalds unsigned long size, grow; 24941da177e4SLinus Torvalds 24951da177e4SLinus Torvalds size = vma->vm_end - address; 24961da177e4SLinus Torvalds grow = (vma->vm_start - address) >> PAGE_SHIFT; 24971da177e4SLinus Torvalds 2498a626ca6aSLinus Torvalds error = -ENOMEM; 2499a626ca6aSLinus Torvalds if (grow <= vma->vm_pgoff) { 25001da177e4SLinus Torvalds error = acct_stack_growth(vma, size, grow); 25011da177e4SLinus Torvalds if (!error) { 25024128997bSMichel Lespinasse /* 25034128997bSMichel Lespinasse * vma_gap_update() doesn't support concurrent 2504c1e8d7c6SMichel Lespinasse * updates, but we only hold a shared mmap_lock 25054128997bSMichel Lespinasse * lock here, so we need to protect against 25064128997bSMichel Lespinasse * concurrent vma expansions. 250712352d3cSKonstantin Khlebnikov * anon_vma_lock_write() doesn't help here, as 25084128997bSMichel Lespinasse * we don't guarantee that all growable vmas 25094128997bSMichel Lespinasse * in a mm share the same root anon vma. 25104128997bSMichel Lespinasse * So, we reuse mm->page_table_lock to guard 25114128997bSMichel Lespinasse * against concurrent vma expansions. 25124128997bSMichel Lespinasse */ 251309357814SOleg Nesterov spin_lock(&mm->page_table_lock); 251487e8827bSOleg Nesterov if (vma->vm_flags & VM_LOCKED) 251509357814SOleg Nesterov mm->locked_vm += grow; 251684638335SKonstantin Khlebnikov vm_stat_account(mm, vma->vm_flags, grow); 2517bf181b9fSMichel Lespinasse anon_vma_interval_tree_pre_update_vma(vma); 25181da177e4SLinus Torvalds vma->vm_start = address; 25191da177e4SLinus Torvalds vma->vm_pgoff -= grow; 2520bf181b9fSMichel Lespinasse anon_vma_interval_tree_post_update_vma(vma); 2521d3737187SMichel Lespinasse vma_gap_update(vma); 252209357814SOleg Nesterov spin_unlock(&mm->page_table_lock); 25234128997bSMichel Lespinasse 25243af9e859SEric B Munson perf_event_mmap(vma); 25251da177e4SLinus Torvalds } 25261da177e4SLinus Torvalds } 2527a626ca6aSLinus Torvalds } 252812352d3cSKonstantin Khlebnikov anon_vma_unlock_write(vma->anon_vma); 25296d50e60cSDavid Rientjes khugepaged_enter_vma_merge(vma, vma->vm_flags); 253009357814SOleg Nesterov validate_mm(mm); 25311da177e4SLinus Torvalds return error; 25321da177e4SLinus Torvalds } 25331da177e4SLinus Torvalds 25341be7107fSHugh Dickins /* enforced gap between the expanding stack and other mappings. */ 25351be7107fSHugh Dickins unsigned long stack_guard_gap = 256UL<<PAGE_SHIFT; 25361be7107fSHugh Dickins 25371be7107fSHugh Dickins static int __init cmdline_parse_stack_guard_gap(char *p) 25381be7107fSHugh Dickins { 25391be7107fSHugh Dickins unsigned long val; 25401be7107fSHugh Dickins char *endptr; 25411be7107fSHugh Dickins 25421be7107fSHugh Dickins val = simple_strtoul(p, &endptr, 10); 25431be7107fSHugh Dickins if (!*endptr) 25441be7107fSHugh Dickins stack_guard_gap = val << PAGE_SHIFT; 25451be7107fSHugh Dickins 25461be7107fSHugh Dickins return 0; 25471be7107fSHugh Dickins } 25481be7107fSHugh Dickins __setup("stack_guard_gap=", cmdline_parse_stack_guard_gap); 25491be7107fSHugh Dickins 2550b6a2fea3SOllie Wild #ifdef CONFIG_STACK_GROWSUP 2551b6a2fea3SOllie Wild int expand_stack(struct vm_area_struct *vma, unsigned long address) 2552b6a2fea3SOllie Wild { 2553b6a2fea3SOllie Wild return expand_upwards(vma, address); 2554b6a2fea3SOllie Wild } 2555b6a2fea3SOllie Wild 2556b6a2fea3SOllie Wild struct vm_area_struct * 2557b6a2fea3SOllie Wild find_extend_vma(struct mm_struct *mm, unsigned long addr) 2558b6a2fea3SOllie Wild { 2559b6a2fea3SOllie Wild struct vm_area_struct *vma, *prev; 2560b6a2fea3SOllie Wild 2561b6a2fea3SOllie Wild addr &= PAGE_MASK; 2562b6a2fea3SOllie Wild vma = find_vma_prev(mm, addr, &prev); 2563b6a2fea3SOllie Wild if (vma && (vma->vm_start <= addr)) 2564b6a2fea3SOllie Wild return vma; 256504f5866eSAndrea Arcangeli /* don't alter vm_end if the coredump is running */ 256604f5866eSAndrea Arcangeli if (!prev || !mmget_still_valid(mm) || expand_stack(prev, addr)) 2567b6a2fea3SOllie Wild return NULL; 2568cea10a19SMichel Lespinasse if (prev->vm_flags & VM_LOCKED) 2569fc05f566SKirill A. Shutemov populate_vma_page_range(prev, addr, prev->vm_end, NULL); 2570b6a2fea3SOllie Wild return prev; 2571b6a2fea3SOllie Wild } 2572b6a2fea3SOllie Wild #else 2573b6a2fea3SOllie Wild int expand_stack(struct vm_area_struct *vma, unsigned long address) 2574b6a2fea3SOllie Wild { 2575b6a2fea3SOllie Wild return expand_downwards(vma, address); 2576b6a2fea3SOllie Wild } 2577b6a2fea3SOllie Wild 25781da177e4SLinus Torvalds struct vm_area_struct * 25791da177e4SLinus Torvalds find_extend_vma(struct mm_struct *mm, unsigned long addr) 25801da177e4SLinus Torvalds { 25811da177e4SLinus Torvalds struct vm_area_struct *vma; 25821da177e4SLinus Torvalds unsigned long start; 25831da177e4SLinus Torvalds 25841da177e4SLinus Torvalds addr &= PAGE_MASK; 25851da177e4SLinus Torvalds vma = find_vma(mm, addr); 25861da177e4SLinus Torvalds if (!vma) 25871da177e4SLinus Torvalds return NULL; 25881da177e4SLinus Torvalds if (vma->vm_start <= addr) 25891da177e4SLinus Torvalds return vma; 25901da177e4SLinus Torvalds if (!(vma->vm_flags & VM_GROWSDOWN)) 25911da177e4SLinus Torvalds return NULL; 259204f5866eSAndrea Arcangeli /* don't alter vm_start if the coredump is running */ 259304f5866eSAndrea Arcangeli if (!mmget_still_valid(mm)) 259404f5866eSAndrea Arcangeli return NULL; 25951da177e4SLinus Torvalds start = vma->vm_start; 25961da177e4SLinus Torvalds if (expand_stack(vma, addr)) 25971da177e4SLinus Torvalds return NULL; 2598cea10a19SMichel Lespinasse if (vma->vm_flags & VM_LOCKED) 2599fc05f566SKirill A. Shutemov populate_vma_page_range(vma, addr, start, NULL); 26001da177e4SLinus Torvalds return vma; 26011da177e4SLinus Torvalds } 26021da177e4SLinus Torvalds #endif 26031da177e4SLinus Torvalds 2604e1d6d01aSJesse Barnes EXPORT_SYMBOL_GPL(find_extend_vma); 2605e1d6d01aSJesse Barnes 26062c0b3814SHugh Dickins /* 26072c0b3814SHugh Dickins * Ok - we have the memory areas we should free on the vma list, 26082c0b3814SHugh Dickins * so release them, and do the vma updates. 26091da177e4SLinus Torvalds * 26102c0b3814SHugh Dickins * Called with the mm semaphore held. 26111da177e4SLinus Torvalds */ 26122c0b3814SHugh Dickins static void remove_vma_list(struct mm_struct *mm, struct vm_area_struct *vma) 26131da177e4SLinus Torvalds { 26144f74d2c8SLinus Torvalds unsigned long nr_accounted = 0; 26154f74d2c8SLinus Torvalds 2616365e9c87SHugh Dickins /* Update high watermark before we lower total_vm */ 2617365e9c87SHugh Dickins update_hiwater_vm(mm); 26182c0b3814SHugh Dickins do { 2619ab50b8edSHugh Dickins long nrpages = vma_pages(vma); 26201da177e4SLinus Torvalds 26214f74d2c8SLinus Torvalds if (vma->vm_flags & VM_ACCOUNT) 26224f74d2c8SLinus Torvalds nr_accounted += nrpages; 262384638335SKonstantin Khlebnikov vm_stat_account(mm, vma->vm_flags, -nrpages); 2624a8fb5618SHugh Dickins vma = remove_vma(vma); 2625146425a3SHugh Dickins } while (vma); 26264f74d2c8SLinus Torvalds vm_unacct_memory(nr_accounted); 26271da177e4SLinus Torvalds validate_mm(mm); 26281da177e4SLinus Torvalds } 26291da177e4SLinus Torvalds 26301da177e4SLinus Torvalds /* 26311da177e4SLinus Torvalds * Get rid of page table information in the indicated region. 26321da177e4SLinus Torvalds * 2633f10df686SPaolo 'Blaisorblade' Giarrusso * Called with the mm semaphore held. 26341da177e4SLinus Torvalds */ 26351da177e4SLinus Torvalds static void unmap_region(struct mm_struct *mm, 2636e0da382cSHugh Dickins struct vm_area_struct *vma, struct vm_area_struct *prev, 2637e0da382cSHugh Dickins unsigned long start, unsigned long end) 26381da177e4SLinus Torvalds { 2639e0da382cSHugh Dickins struct vm_area_struct *next = prev ? prev->vm_next : mm->mmap; 2640d16dfc55SPeter Zijlstra struct mmu_gather tlb; 26411da177e4SLinus Torvalds 26421da177e4SLinus Torvalds lru_add_drain(); 26432b047252SLinus Torvalds tlb_gather_mmu(&tlb, mm, start, end); 2644365e9c87SHugh Dickins update_hiwater_rss(mm); 26454f74d2c8SLinus Torvalds unmap_vmas(&tlb, vma, start, end); 2646d16dfc55SPeter Zijlstra free_pgtables(&tlb, vma, prev ? prev->vm_end : FIRST_USER_ADDRESS, 26476ee8630eSHugh Dickins next ? next->vm_start : USER_PGTABLES_CEILING); 2648d16dfc55SPeter Zijlstra tlb_finish_mmu(&tlb, start, end); 26491da177e4SLinus Torvalds } 26501da177e4SLinus Torvalds 26511da177e4SLinus Torvalds /* 26521da177e4SLinus Torvalds * Create a list of vma's touched by the unmap, removing them from the mm's 26531da177e4SLinus Torvalds * vma list as we go.. 26541da177e4SLinus Torvalds */ 2655246c320aSKirill A. Shutemov static bool 26561da177e4SLinus Torvalds detach_vmas_to_be_unmapped(struct mm_struct *mm, struct vm_area_struct *vma, 26571da177e4SLinus Torvalds struct vm_area_struct *prev, unsigned long end) 26581da177e4SLinus Torvalds { 26591da177e4SLinus Torvalds struct vm_area_struct **insertion_point; 26601da177e4SLinus Torvalds struct vm_area_struct *tail_vma = NULL; 26611da177e4SLinus Torvalds 26621da177e4SLinus Torvalds insertion_point = (prev ? &prev->vm_next : &mm->mmap); 2663297c5eeeSLinus Torvalds vma->vm_prev = NULL; 26641da177e4SLinus Torvalds do { 2665d3737187SMichel Lespinasse vma_rb_erase(vma, &mm->mm_rb); 26661da177e4SLinus Torvalds mm->map_count--; 26671da177e4SLinus Torvalds tail_vma = vma; 26681da177e4SLinus Torvalds vma = vma->vm_next; 26691da177e4SLinus Torvalds } while (vma && vma->vm_start < end); 26701da177e4SLinus Torvalds *insertion_point = vma; 2671d3737187SMichel Lespinasse if (vma) { 2672297c5eeeSLinus Torvalds vma->vm_prev = prev; 2673d3737187SMichel Lespinasse vma_gap_update(vma); 2674d3737187SMichel Lespinasse } else 26751be7107fSHugh Dickins mm->highest_vm_end = prev ? vm_end_gap(prev) : 0; 26761da177e4SLinus Torvalds tail_vma->vm_next = NULL; 2677615d6e87SDavidlohr Bueso 2678615d6e87SDavidlohr Bueso /* Kill the cache */ 2679615d6e87SDavidlohr Bueso vmacache_invalidate(mm); 2680246c320aSKirill A. Shutemov 2681246c320aSKirill A. Shutemov /* 2682246c320aSKirill A. Shutemov * Do not downgrade mmap_lock if we are next to VM_GROWSDOWN or 2683246c320aSKirill A. Shutemov * VM_GROWSUP VMA. Such VMAs can change their size under 2684246c320aSKirill A. Shutemov * down_read(mmap_lock) and collide with the VMA we are about to unmap. 2685246c320aSKirill A. Shutemov */ 2686246c320aSKirill A. Shutemov if (vma && (vma->vm_flags & VM_GROWSDOWN)) 2687246c320aSKirill A. Shutemov return false; 2688246c320aSKirill A. Shutemov if (prev && (prev->vm_flags & VM_GROWSUP)) 2689246c320aSKirill A. Shutemov return false; 2690246c320aSKirill A. Shutemov return true; 26911da177e4SLinus Torvalds } 26921da177e4SLinus Torvalds 26931da177e4SLinus Torvalds /* 2694def5efe0SDavid Rientjes * __split_vma() bypasses sysctl_max_map_count checking. We use this where it 2695def5efe0SDavid Rientjes * has already been checked or doesn't make sense to fail. 26961da177e4SLinus Torvalds */ 2697def5efe0SDavid Rientjes int __split_vma(struct mm_struct *mm, struct vm_area_struct *vma, 26981da177e4SLinus Torvalds unsigned long addr, int new_below) 26991da177e4SLinus Torvalds { 27001da177e4SLinus Torvalds struct vm_area_struct *new; 2701e3975891SChen Gang int err; 27021da177e4SLinus Torvalds 270331383c68SDan Williams if (vma->vm_ops && vma->vm_ops->split) { 270431383c68SDan Williams err = vma->vm_ops->split(vma, addr); 270531383c68SDan Williams if (err) 270631383c68SDan Williams return err; 270731383c68SDan Williams } 27081da177e4SLinus Torvalds 27093928d4f5SLinus Torvalds new = vm_area_dup(vma); 27101da177e4SLinus Torvalds if (!new) 2711e3975891SChen Gang return -ENOMEM; 27121da177e4SLinus Torvalds 27131da177e4SLinus Torvalds if (new_below) 27141da177e4SLinus Torvalds new->vm_end = addr; 27151da177e4SLinus Torvalds else { 27161da177e4SLinus Torvalds new->vm_start = addr; 27171da177e4SLinus Torvalds new->vm_pgoff += ((addr - vma->vm_start) >> PAGE_SHIFT); 27181da177e4SLinus Torvalds } 27191da177e4SLinus Torvalds 2720ef0855d3SOleg Nesterov err = vma_dup_policy(vma, new); 2721ef0855d3SOleg Nesterov if (err) 27225beb4930SRik van Riel goto out_free_vma; 27231da177e4SLinus Torvalds 2724c4ea95d7SDaniel Forrest err = anon_vma_clone(new, vma); 2725c4ea95d7SDaniel Forrest if (err) 27265beb4930SRik van Riel goto out_free_mpol; 27275beb4930SRik van Riel 2728e9714acfSKonstantin Khlebnikov if (new->vm_file) 27291da177e4SLinus Torvalds get_file(new->vm_file); 27301da177e4SLinus Torvalds 27311da177e4SLinus Torvalds if (new->vm_ops && new->vm_ops->open) 27321da177e4SLinus Torvalds new->vm_ops->open(new); 27331da177e4SLinus Torvalds 27341da177e4SLinus Torvalds if (new_below) 27355beb4930SRik van Riel err = vma_adjust(vma, addr, vma->vm_end, vma->vm_pgoff + 27361da177e4SLinus Torvalds ((addr - new->vm_start) >> PAGE_SHIFT), new); 27371da177e4SLinus Torvalds else 27385beb4930SRik van Riel err = vma_adjust(vma, vma->vm_start, addr, vma->vm_pgoff, new); 27391da177e4SLinus Torvalds 27405beb4930SRik van Riel /* Success. */ 27415beb4930SRik van Riel if (!err) 27421da177e4SLinus Torvalds return 0; 27435beb4930SRik van Riel 27445beb4930SRik van Riel /* Clean everything up if vma_adjust failed. */ 274558927533SRik van Riel if (new->vm_ops && new->vm_ops->close) 27465beb4930SRik van Riel new->vm_ops->close(new); 2747e9714acfSKonstantin Khlebnikov if (new->vm_file) 27485beb4930SRik van Riel fput(new->vm_file); 27492aeadc30SAndrea Arcangeli unlink_anon_vmas(new); 27505beb4930SRik van Riel out_free_mpol: 2751ef0855d3SOleg Nesterov mpol_put(vma_policy(new)); 27525beb4930SRik van Riel out_free_vma: 27533928d4f5SLinus Torvalds vm_area_free(new); 27545beb4930SRik van Riel return err; 27551da177e4SLinus Torvalds } 27561da177e4SLinus Torvalds 2757659ace58SKOSAKI Motohiro /* 2758659ace58SKOSAKI Motohiro * Split a vma into two pieces at address 'addr', a new vma is allocated 2759659ace58SKOSAKI Motohiro * either for the first part or the tail. 2760659ace58SKOSAKI Motohiro */ 2761659ace58SKOSAKI Motohiro int split_vma(struct mm_struct *mm, struct vm_area_struct *vma, 2762659ace58SKOSAKI Motohiro unsigned long addr, int new_below) 2763659ace58SKOSAKI Motohiro { 2764659ace58SKOSAKI Motohiro if (mm->map_count >= sysctl_max_map_count) 2765659ace58SKOSAKI Motohiro return -ENOMEM; 2766659ace58SKOSAKI Motohiro 2767659ace58SKOSAKI Motohiro return __split_vma(mm, vma, addr, new_below); 2768659ace58SKOSAKI Motohiro } 2769659ace58SKOSAKI Motohiro 27701da177e4SLinus Torvalds /* Munmap is split into 2 main parts -- this part which finds 27711da177e4SLinus Torvalds * what needs doing, and the areas themselves, which do the 27721da177e4SLinus Torvalds * work. This now handles partial unmappings. 27731da177e4SLinus Torvalds * Jeremy Fitzhardinge <jeremy@goop.org> 27741da177e4SLinus Torvalds */ 277585a06835SYang Shi int __do_munmap(struct mm_struct *mm, unsigned long start, size_t len, 2776dd2283f2SYang Shi struct list_head *uf, bool downgrade) 27771da177e4SLinus Torvalds { 27781da177e4SLinus Torvalds unsigned long end; 2779146425a3SHugh Dickins struct vm_area_struct *vma, *prev, *last; 27801da177e4SLinus Torvalds 2781de1741a1SAlexander Kuleshov if ((offset_in_page(start)) || start > TASK_SIZE || len > TASK_SIZE-start) 27821da177e4SLinus Torvalds return -EINVAL; 27831da177e4SLinus Torvalds 2784cc71aba3Svishnu.ps len = PAGE_ALIGN(len); 27855a28fc94SDave Hansen end = start + len; 2786cc71aba3Svishnu.ps if (len == 0) 27871da177e4SLinus Torvalds return -EINVAL; 27881da177e4SLinus Torvalds 27895a28fc94SDave Hansen /* 27905a28fc94SDave Hansen * arch_unmap() might do unmaps itself. It must be called 27915a28fc94SDave Hansen * and finish any rbtree manipulation before this code 27925a28fc94SDave Hansen * runs and also starts to manipulate the rbtree. 27935a28fc94SDave Hansen */ 27945a28fc94SDave Hansen arch_unmap(mm, start, end); 27955a28fc94SDave Hansen 27961da177e4SLinus Torvalds /* Find the first overlapping VMA */ 27979be34c9dSLinus Torvalds vma = find_vma(mm, start); 2798146425a3SHugh Dickins if (!vma) 27991da177e4SLinus Torvalds return 0; 28009be34c9dSLinus Torvalds prev = vma->vm_prev; 2801146425a3SHugh Dickins /* we have start < vma->vm_end */ 28021da177e4SLinus Torvalds 28031da177e4SLinus Torvalds /* if it doesn't overlap, we have nothing.. */ 2804146425a3SHugh Dickins if (vma->vm_start >= end) 28051da177e4SLinus Torvalds return 0; 28061da177e4SLinus Torvalds 28071da177e4SLinus Torvalds /* 28081da177e4SLinus Torvalds * If we need to split any vma, do it now to save pain later. 28091da177e4SLinus Torvalds * 28101da177e4SLinus Torvalds * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially 28111da177e4SLinus Torvalds * unmapped vm_area_struct will remain in use: so lower split_vma 28121da177e4SLinus Torvalds * places tmp vma above, and higher split_vma places tmp vma below. 28131da177e4SLinus Torvalds */ 2814146425a3SHugh Dickins if (start > vma->vm_start) { 2815659ace58SKOSAKI Motohiro int error; 2816659ace58SKOSAKI Motohiro 2817659ace58SKOSAKI Motohiro /* 2818659ace58SKOSAKI Motohiro * Make sure that map_count on return from munmap() will 2819659ace58SKOSAKI Motohiro * not exceed its limit; but let map_count go just above 2820659ace58SKOSAKI Motohiro * its limit temporarily, to help free resources as expected. 2821659ace58SKOSAKI Motohiro */ 2822659ace58SKOSAKI Motohiro if (end < vma->vm_end && mm->map_count >= sysctl_max_map_count) 2823659ace58SKOSAKI Motohiro return -ENOMEM; 2824659ace58SKOSAKI Motohiro 2825659ace58SKOSAKI Motohiro error = __split_vma(mm, vma, start, 0); 28261da177e4SLinus Torvalds if (error) 28271da177e4SLinus Torvalds return error; 2828146425a3SHugh Dickins prev = vma; 28291da177e4SLinus Torvalds } 28301da177e4SLinus Torvalds 28311da177e4SLinus Torvalds /* Does it split the last one? */ 28321da177e4SLinus Torvalds last = find_vma(mm, end); 28331da177e4SLinus Torvalds if (last && end > last->vm_start) { 2834659ace58SKOSAKI Motohiro int error = __split_vma(mm, last, end, 1); 28351da177e4SLinus Torvalds if (error) 28361da177e4SLinus Torvalds return error; 28371da177e4SLinus Torvalds } 2838146425a3SHugh Dickins vma = prev ? prev->vm_next : mm->mmap; 28391da177e4SLinus Torvalds 28402376dd7cSAndrea Arcangeli if (unlikely(uf)) { 28412376dd7cSAndrea Arcangeli /* 28422376dd7cSAndrea Arcangeli * If userfaultfd_unmap_prep returns an error the vmas 28432376dd7cSAndrea Arcangeli * will remain splitted, but userland will get a 28442376dd7cSAndrea Arcangeli * highly unexpected error anyway. This is no 28452376dd7cSAndrea Arcangeli * different than the case where the first of the two 28462376dd7cSAndrea Arcangeli * __split_vma fails, but we don't undo the first 28472376dd7cSAndrea Arcangeli * split, despite we could. This is unlikely enough 28482376dd7cSAndrea Arcangeli * failure that it's not worth optimizing it for. 28492376dd7cSAndrea Arcangeli */ 28502376dd7cSAndrea Arcangeli int error = userfaultfd_unmap_prep(vma, start, end, uf); 28512376dd7cSAndrea Arcangeli if (error) 28522376dd7cSAndrea Arcangeli return error; 28532376dd7cSAndrea Arcangeli } 28542376dd7cSAndrea Arcangeli 28551da177e4SLinus Torvalds /* 2856ba470de4SRik van Riel * unlock any mlock()ed ranges before detaching vmas 2857ba470de4SRik van Riel */ 2858ba470de4SRik van Riel if (mm->locked_vm) { 2859ba470de4SRik van Riel struct vm_area_struct *tmp = vma; 2860ba470de4SRik van Riel while (tmp && tmp->vm_start < end) { 2861ba470de4SRik van Riel if (tmp->vm_flags & VM_LOCKED) { 2862ba470de4SRik van Riel mm->locked_vm -= vma_pages(tmp); 2863ba470de4SRik van Riel munlock_vma_pages_all(tmp); 2864ba470de4SRik van Riel } 2865dd2283f2SYang Shi 2866ba470de4SRik van Riel tmp = tmp->vm_next; 2867ba470de4SRik van Riel } 2868ba470de4SRik van Riel } 2869ba470de4SRik van Riel 2870dd2283f2SYang Shi /* Detach vmas from rbtree */ 2871246c320aSKirill A. Shutemov if (!detach_vmas_to_be_unmapped(mm, vma, prev, end)) 2872246c320aSKirill A. Shutemov downgrade = false; 28731da177e4SLinus Torvalds 2874dd2283f2SYang Shi if (downgrade) 2875d8ed45c5SMichel Lespinasse mmap_write_downgrade(mm); 2876dd2283f2SYang Shi 2877dd2283f2SYang Shi unmap_region(mm, vma, prev, start, end); 2878dd2283f2SYang Shi 28791da177e4SLinus Torvalds /* Fix up all other VM information */ 28802c0b3814SHugh Dickins remove_vma_list(mm, vma); 28811da177e4SLinus Torvalds 2882dd2283f2SYang Shi return downgrade ? 1 : 0; 28831da177e4SLinus Torvalds } 28841da177e4SLinus Torvalds 2885dd2283f2SYang Shi int do_munmap(struct mm_struct *mm, unsigned long start, size_t len, 2886dd2283f2SYang Shi struct list_head *uf) 2887dd2283f2SYang Shi { 2888dd2283f2SYang Shi return __do_munmap(mm, start, len, uf, false); 2889dd2283f2SYang Shi } 2890dd2283f2SYang Shi 2891dd2283f2SYang Shi static int __vm_munmap(unsigned long start, size_t len, bool downgrade) 2892a46ef99dSLinus Torvalds { 2893a46ef99dSLinus Torvalds int ret; 2894bfce281cSAl Viro struct mm_struct *mm = current->mm; 2895897ab3e0SMike Rapoport LIST_HEAD(uf); 2896a46ef99dSLinus Torvalds 2897d8ed45c5SMichel Lespinasse if (mmap_write_lock_killable(mm)) 2898ae798783SMichal Hocko return -EINTR; 2899ae798783SMichal Hocko 2900dd2283f2SYang Shi ret = __do_munmap(mm, start, len, &uf, downgrade); 2901dd2283f2SYang Shi /* 2902c1e8d7c6SMichel Lespinasse * Returning 1 indicates mmap_lock is downgraded. 2903dd2283f2SYang Shi * But 1 is not legal return value of vm_munmap() and munmap(), reset 2904dd2283f2SYang Shi * it to 0 before return. 2905dd2283f2SYang Shi */ 2906dd2283f2SYang Shi if (ret == 1) { 2907d8ed45c5SMichel Lespinasse mmap_read_unlock(mm); 2908dd2283f2SYang Shi ret = 0; 2909dd2283f2SYang Shi } else 2910d8ed45c5SMichel Lespinasse mmap_write_unlock(mm); 2911dd2283f2SYang Shi 2912897ab3e0SMike Rapoport userfaultfd_unmap_complete(mm, &uf); 2913a46ef99dSLinus Torvalds return ret; 2914a46ef99dSLinus Torvalds } 2915dd2283f2SYang Shi 2916dd2283f2SYang Shi int vm_munmap(unsigned long start, size_t len) 2917dd2283f2SYang Shi { 2918dd2283f2SYang Shi return __vm_munmap(start, len, false); 2919dd2283f2SYang Shi } 2920a46ef99dSLinus Torvalds EXPORT_SYMBOL(vm_munmap); 2921a46ef99dSLinus Torvalds 29226a6160a7SHeiko Carstens SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len) 29231da177e4SLinus Torvalds { 2924ce18d171SCatalin Marinas addr = untagged_addr(addr); 29251da177e4SLinus Torvalds profile_munmap(addr); 2926dd2283f2SYang Shi return __vm_munmap(addr, len, true); 29271da177e4SLinus Torvalds } 29281da177e4SLinus Torvalds 2929c8d78c18SKirill A. Shutemov 2930c8d78c18SKirill A. Shutemov /* 2931c8d78c18SKirill A. Shutemov * Emulation of deprecated remap_file_pages() syscall. 2932c8d78c18SKirill A. Shutemov */ 2933c8d78c18SKirill A. Shutemov SYSCALL_DEFINE5(remap_file_pages, unsigned long, start, unsigned long, size, 2934c8d78c18SKirill A. Shutemov unsigned long, prot, unsigned long, pgoff, unsigned long, flags) 2935c8d78c18SKirill A. Shutemov { 2936c8d78c18SKirill A. Shutemov 2937c8d78c18SKirill A. Shutemov struct mm_struct *mm = current->mm; 2938c8d78c18SKirill A. Shutemov struct vm_area_struct *vma; 2939c8d78c18SKirill A. Shutemov unsigned long populate = 0; 2940c8d78c18SKirill A. Shutemov unsigned long ret = -EINVAL; 2941c8d78c18SKirill A. Shutemov struct file *file; 2942c8d78c18SKirill A. Shutemov 2943ad56b738SMike Rapoport pr_warn_once("%s (%d) uses deprecated remap_file_pages() syscall. See Documentation/vm/remap_file_pages.rst.\n", 2944c8d78c18SKirill A. Shutemov current->comm, current->pid); 2945c8d78c18SKirill A. Shutemov 2946c8d78c18SKirill A. Shutemov if (prot) 2947c8d78c18SKirill A. Shutemov return ret; 2948c8d78c18SKirill A. Shutemov start = start & PAGE_MASK; 2949c8d78c18SKirill A. Shutemov size = size & PAGE_MASK; 2950c8d78c18SKirill A. Shutemov 2951c8d78c18SKirill A. Shutemov if (start + size <= start) 2952c8d78c18SKirill A. Shutemov return ret; 2953c8d78c18SKirill A. Shutemov 2954c8d78c18SKirill A. Shutemov /* Does pgoff wrap? */ 2955c8d78c18SKirill A. Shutemov if (pgoff + (size >> PAGE_SHIFT) < pgoff) 2956c8d78c18SKirill A. Shutemov return ret; 2957c8d78c18SKirill A. Shutemov 2958d8ed45c5SMichel Lespinasse if (mmap_write_lock_killable(mm)) 2959dc0ef0dfSMichal Hocko return -EINTR; 2960dc0ef0dfSMichal Hocko 2961c8d78c18SKirill A. Shutemov vma = find_vma(mm, start); 2962c8d78c18SKirill A. Shutemov 2963c8d78c18SKirill A. Shutemov if (!vma || !(vma->vm_flags & VM_SHARED)) 2964c8d78c18SKirill A. Shutemov goto out; 2965c8d78c18SKirill A. Shutemov 296648f7df32SKirill A. Shutemov if (start < vma->vm_start) 2967c8d78c18SKirill A. Shutemov goto out; 2968c8d78c18SKirill A. Shutemov 296948f7df32SKirill A. Shutemov if (start + size > vma->vm_end) { 297048f7df32SKirill A. Shutemov struct vm_area_struct *next; 297148f7df32SKirill A. Shutemov 297248f7df32SKirill A. Shutemov for (next = vma->vm_next; next; next = next->vm_next) { 297348f7df32SKirill A. Shutemov /* hole between vmas ? */ 297448f7df32SKirill A. Shutemov if (next->vm_start != next->vm_prev->vm_end) 297548f7df32SKirill A. Shutemov goto out; 297648f7df32SKirill A. Shutemov 297748f7df32SKirill A. Shutemov if (next->vm_file != vma->vm_file) 297848f7df32SKirill A. Shutemov goto out; 297948f7df32SKirill A. Shutemov 298048f7df32SKirill A. Shutemov if (next->vm_flags != vma->vm_flags) 298148f7df32SKirill A. Shutemov goto out; 298248f7df32SKirill A. Shutemov 298348f7df32SKirill A. Shutemov if (start + size <= next->vm_end) 298448f7df32SKirill A. Shutemov break; 298548f7df32SKirill A. Shutemov } 298648f7df32SKirill A. Shutemov 298748f7df32SKirill A. Shutemov if (!next) 2988c8d78c18SKirill A. Shutemov goto out; 2989c8d78c18SKirill A. Shutemov } 2990c8d78c18SKirill A. Shutemov 2991c8d78c18SKirill A. Shutemov prot |= vma->vm_flags & VM_READ ? PROT_READ : 0; 2992c8d78c18SKirill A. Shutemov prot |= vma->vm_flags & VM_WRITE ? PROT_WRITE : 0; 2993c8d78c18SKirill A. Shutemov prot |= vma->vm_flags & VM_EXEC ? PROT_EXEC : 0; 2994c8d78c18SKirill A. Shutemov 2995c8d78c18SKirill A. Shutemov flags &= MAP_NONBLOCK; 2996c8d78c18SKirill A. Shutemov flags |= MAP_SHARED | MAP_FIXED | MAP_POPULATE; 2997c8d78c18SKirill A. Shutemov if (vma->vm_flags & VM_LOCKED) { 299848f7df32SKirill A. Shutemov struct vm_area_struct *tmp; 2999c8d78c18SKirill A. Shutemov flags |= MAP_LOCKED; 300048f7df32SKirill A. Shutemov 3001c8d78c18SKirill A. Shutemov /* drop PG_Mlocked flag for over-mapped range */ 300248f7df32SKirill A. Shutemov for (tmp = vma; tmp->vm_start >= start + size; 300348f7df32SKirill A. Shutemov tmp = tmp->vm_next) { 30049a73f61bSKirill A. Shutemov /* 30059a73f61bSKirill A. Shutemov * Split pmd and munlock page on the border 30069a73f61bSKirill A. Shutemov * of the range. 30079a73f61bSKirill A. Shutemov */ 30089a73f61bSKirill A. Shutemov vma_adjust_trans_huge(tmp, start, start + size, 0); 30099a73f61bSKirill A. Shutemov 301048f7df32SKirill A. Shutemov munlock_vma_pages_range(tmp, 301148f7df32SKirill A. Shutemov max(tmp->vm_start, start), 301248f7df32SKirill A. Shutemov min(tmp->vm_end, start + size)); 301348f7df32SKirill A. Shutemov } 3014c8d78c18SKirill A. Shutemov } 3015c8d78c18SKirill A. Shutemov 3016c8d78c18SKirill A. Shutemov file = get_file(vma->vm_file); 301745e55300SPeter Collingbourne ret = do_mmap(vma->vm_file, start, size, 3018897ab3e0SMike Rapoport prot, flags, pgoff, &populate, NULL); 3019c8d78c18SKirill A. Shutemov fput(file); 3020c8d78c18SKirill A. Shutemov out: 3021d8ed45c5SMichel Lespinasse mmap_write_unlock(mm); 3022c8d78c18SKirill A. Shutemov if (populate) 3023c8d78c18SKirill A. Shutemov mm_populate(ret, populate); 3024c8d78c18SKirill A. Shutemov if (!IS_ERR_VALUE(ret)) 3025c8d78c18SKirill A. Shutemov ret = 0; 3026c8d78c18SKirill A. Shutemov return ret; 3027c8d78c18SKirill A. Shutemov } 3028c8d78c18SKirill A. Shutemov 30291da177e4SLinus Torvalds /* 30301da177e4SLinus Torvalds * this is really a simplified "do_mmap". it only handles 30311da177e4SLinus Torvalds * anonymous maps. eventually we may be able to do some 30321da177e4SLinus Torvalds * brk-specific accounting here. 30331da177e4SLinus Torvalds */ 3034bb177a73SMichal Hocko static int do_brk_flags(unsigned long addr, unsigned long len, unsigned long flags, struct list_head *uf) 30351da177e4SLinus Torvalds { 30361da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 30371da177e4SLinus Torvalds struct vm_area_struct *vma, *prev; 30381da177e4SLinus Torvalds struct rb_node **rb_link, *rb_parent; 30391da177e4SLinus Torvalds pgoff_t pgoff = addr >> PAGE_SHIFT; 30403a459756SKirill Korotaev int error; 3041ff68dac6SGaowei Pu unsigned long mapped_addr; 30421da177e4SLinus Torvalds 304316e72e9bSDenys Vlasenko /* Until we need other flags, refuse anything except VM_EXEC. */ 304416e72e9bSDenys Vlasenko if ((flags & (~VM_EXEC)) != 0) 304516e72e9bSDenys Vlasenko return -EINVAL; 304616e72e9bSDenys Vlasenko flags |= VM_DATA_DEFAULT_FLAGS | VM_ACCOUNT | mm->def_flags; 30473a459756SKirill Korotaev 3048ff68dac6SGaowei Pu mapped_addr = get_unmapped_area(NULL, addr, len, 0, MAP_FIXED); 3049ff68dac6SGaowei Pu if (IS_ERR_VALUE(mapped_addr)) 3050ff68dac6SGaowei Pu return mapped_addr; 30513a459756SKirill Korotaev 3052363ee17fSDavidlohr Bueso error = mlock_future_check(mm, mm->def_flags, len); 3053363ee17fSDavidlohr Bueso if (error) 3054363ee17fSDavidlohr Bueso return error; 30551da177e4SLinus Torvalds 30561da177e4SLinus Torvalds /* 30571da177e4SLinus Torvalds * Clear old maps. this also does some error checking for us 30581da177e4SLinus Torvalds */ 30599fcd1457SRasmus Villemoes while (find_vma_links(mm, addr, addr + len, &prev, &rb_link, 30609fcd1457SRasmus Villemoes &rb_parent)) { 3061897ab3e0SMike Rapoport if (do_munmap(mm, addr, len, uf)) 30621da177e4SLinus Torvalds return -ENOMEM; 30631da177e4SLinus Torvalds } 30641da177e4SLinus Torvalds 30651da177e4SLinus Torvalds /* Check against address space limits *after* clearing old maps... */ 306684638335SKonstantin Khlebnikov if (!may_expand_vm(mm, flags, len >> PAGE_SHIFT)) 30671da177e4SLinus Torvalds return -ENOMEM; 30681da177e4SLinus Torvalds 30691da177e4SLinus Torvalds if (mm->map_count > sysctl_max_map_count) 30701da177e4SLinus Torvalds return -ENOMEM; 30711da177e4SLinus Torvalds 3072191c5424SAl Viro if (security_vm_enough_memory_mm(mm, len >> PAGE_SHIFT)) 30731da177e4SLinus Torvalds return -ENOMEM; 30741da177e4SLinus Torvalds 30751da177e4SLinus Torvalds /* Can we just expand an old private anonymous mapping? */ 3076ba470de4SRik van Riel vma = vma_merge(mm, prev, addr, addr + len, flags, 307719a809afSAndrea Arcangeli NULL, NULL, pgoff, NULL, NULL_VM_UFFD_CTX); 3078ba470de4SRik van Riel if (vma) 30791da177e4SLinus Torvalds goto out; 30801da177e4SLinus Torvalds 30811da177e4SLinus Torvalds /* 30821da177e4SLinus Torvalds * create a vma struct for an anonymous mapping 30831da177e4SLinus Torvalds */ 3084490fc053SLinus Torvalds vma = vm_area_alloc(mm); 30851da177e4SLinus Torvalds if (!vma) { 30861da177e4SLinus Torvalds vm_unacct_memory(len >> PAGE_SHIFT); 30871da177e4SLinus Torvalds return -ENOMEM; 30881da177e4SLinus Torvalds } 30891da177e4SLinus Torvalds 3090bfd40eafSKirill A. Shutemov vma_set_anonymous(vma); 30911da177e4SLinus Torvalds vma->vm_start = addr; 30921da177e4SLinus Torvalds vma->vm_end = addr + len; 30931da177e4SLinus Torvalds vma->vm_pgoff = pgoff; 30941da177e4SLinus Torvalds vma->vm_flags = flags; 30953ed75eb8SColy Li vma->vm_page_prot = vm_get_page_prot(flags); 30961da177e4SLinus Torvalds vma_link(mm, vma, prev, rb_link, rb_parent); 30971da177e4SLinus Torvalds out: 30983af9e859SEric B Munson perf_event_mmap(vma); 30991da177e4SLinus Torvalds mm->total_vm += len >> PAGE_SHIFT; 310084638335SKonstantin Khlebnikov mm->data_vm += len >> PAGE_SHIFT; 3101128557ffSMichel Lespinasse if (flags & VM_LOCKED) 3102ba470de4SRik van Riel mm->locked_vm += (len >> PAGE_SHIFT); 3103d9104d1cSCyrill Gorcunov vma->vm_flags |= VM_SOFTDIRTY; 31045d22fc25SLinus Torvalds return 0; 31051da177e4SLinus Torvalds } 31061da177e4SLinus Torvalds 3107bb177a73SMichal Hocko int vm_brk_flags(unsigned long addr, unsigned long request, unsigned long flags) 3108e4eb1ff6SLinus Torvalds { 3109e4eb1ff6SLinus Torvalds struct mm_struct *mm = current->mm; 3110bb177a73SMichal Hocko unsigned long len; 31115d22fc25SLinus Torvalds int ret; 3112128557ffSMichel Lespinasse bool populate; 3113897ab3e0SMike Rapoport LIST_HEAD(uf); 3114e4eb1ff6SLinus Torvalds 3115bb177a73SMichal Hocko len = PAGE_ALIGN(request); 3116bb177a73SMichal Hocko if (len < request) 3117bb177a73SMichal Hocko return -ENOMEM; 3118bb177a73SMichal Hocko if (!len) 3119bb177a73SMichal Hocko return 0; 3120bb177a73SMichal Hocko 3121d8ed45c5SMichel Lespinasse if (mmap_write_lock_killable(mm)) 31222d6c9282SMichal Hocko return -EINTR; 31232d6c9282SMichal Hocko 3124897ab3e0SMike Rapoport ret = do_brk_flags(addr, len, flags, &uf); 3125128557ffSMichel Lespinasse populate = ((mm->def_flags & VM_LOCKED) != 0); 3126d8ed45c5SMichel Lespinasse mmap_write_unlock(mm); 3127897ab3e0SMike Rapoport userfaultfd_unmap_complete(mm, &uf); 31285d22fc25SLinus Torvalds if (populate && !ret) 3129128557ffSMichel Lespinasse mm_populate(addr, len); 3130e4eb1ff6SLinus Torvalds return ret; 3131e4eb1ff6SLinus Torvalds } 313216e72e9bSDenys Vlasenko EXPORT_SYMBOL(vm_brk_flags); 313316e72e9bSDenys Vlasenko 313416e72e9bSDenys Vlasenko int vm_brk(unsigned long addr, unsigned long len) 313516e72e9bSDenys Vlasenko { 313616e72e9bSDenys Vlasenko return vm_brk_flags(addr, len, 0); 313716e72e9bSDenys Vlasenko } 3138e4eb1ff6SLinus Torvalds EXPORT_SYMBOL(vm_brk); 31391da177e4SLinus Torvalds 31401da177e4SLinus Torvalds /* Release all mmaps. */ 31411da177e4SLinus Torvalds void exit_mmap(struct mm_struct *mm) 31421da177e4SLinus Torvalds { 3143d16dfc55SPeter Zijlstra struct mmu_gather tlb; 3144ba470de4SRik van Riel struct vm_area_struct *vma; 31451da177e4SLinus Torvalds unsigned long nr_accounted = 0; 31461da177e4SLinus Torvalds 3147d6dd61c8SJeremy Fitzhardinge /* mm's last user has gone, and its about to be pulled down */ 3148cddb8a5cSAndrea Arcangeli mmu_notifier_release(mm); 3149d6dd61c8SJeremy Fitzhardinge 315027ae357fSDavid Rientjes if (unlikely(mm_is_oom_victim(mm))) { 315127ae357fSDavid Rientjes /* 315227ae357fSDavid Rientjes * Manually reap the mm to free as much memory as possible. 315327ae357fSDavid Rientjes * Then, as the oom reaper does, set MMF_OOM_SKIP to disregard 3154c1e8d7c6SMichel Lespinasse * this mm from further consideration. Taking mm->mmap_lock for 315527ae357fSDavid Rientjes * write after setting MMF_OOM_SKIP will guarantee that the oom 3156c1e8d7c6SMichel Lespinasse * reaper will not run on this mm again after mmap_lock is 315727ae357fSDavid Rientjes * dropped. 315827ae357fSDavid Rientjes * 3159c1e8d7c6SMichel Lespinasse * Nothing can be holding mm->mmap_lock here and the above call 316027ae357fSDavid Rientjes * to mmu_notifier_release(mm) ensures mmu notifier callbacks in 316127ae357fSDavid Rientjes * __oom_reap_task_mm() will not block. 316227ae357fSDavid Rientjes * 316327ae357fSDavid Rientjes * This needs to be done before calling munlock_vma_pages_all(), 316427ae357fSDavid Rientjes * which clears VM_LOCKED, otherwise the oom reaper cannot 316527ae357fSDavid Rientjes * reliably test it. 316627ae357fSDavid Rientjes */ 316793065ac7SMichal Hocko (void)__oom_reap_task_mm(mm); 316827ae357fSDavid Rientjes 316927ae357fSDavid Rientjes set_bit(MMF_OOM_SKIP, &mm->flags); 3170d8ed45c5SMichel Lespinasse mmap_write_lock(mm); 3171d8ed45c5SMichel Lespinasse mmap_write_unlock(mm); 317227ae357fSDavid Rientjes } 317327ae357fSDavid Rientjes 3174ba470de4SRik van Riel if (mm->locked_vm) { 3175ba470de4SRik van Riel vma = mm->mmap; 3176ba470de4SRik van Riel while (vma) { 3177ba470de4SRik van Riel if (vma->vm_flags & VM_LOCKED) 3178ba470de4SRik van Riel munlock_vma_pages_all(vma); 3179ba470de4SRik van Riel vma = vma->vm_next; 3180ba470de4SRik van Riel } 3181ba470de4SRik van Riel } 31829480c53eSJeremy Fitzhardinge 31839480c53eSJeremy Fitzhardinge arch_exit_mmap(mm); 31849480c53eSJeremy Fitzhardinge 3185ba470de4SRik van Riel vma = mm->mmap; 31869480c53eSJeremy Fitzhardinge if (!vma) /* Can happen if dup_mmap() received an OOM */ 31879480c53eSJeremy Fitzhardinge return; 31889480c53eSJeremy Fitzhardinge 31891da177e4SLinus Torvalds lru_add_drain(); 31901da177e4SLinus Torvalds flush_cache_mm(mm); 31912b047252SLinus Torvalds tlb_gather_mmu(&tlb, mm, 0, -1); 3192901608d9SOleg Nesterov /* update_hiwater_rss(mm) here? but nobody should be looking */ 3193e0da382cSHugh Dickins /* Use -1 here to ensure all VMAs in the mm are unmapped */ 31944f74d2c8SLinus Torvalds unmap_vmas(&tlb, vma, 0, -1); 31956ee8630eSHugh Dickins free_pgtables(&tlb, vma, FIRST_USER_ADDRESS, USER_PGTABLES_CEILING); 3196853f5e26SAl Viro tlb_finish_mmu(&tlb, 0, -1); 31971da177e4SLinus Torvalds 31981da177e4SLinus Torvalds /* 31998f4f8c16SHugh Dickins * Walk the list again, actually closing and freeing it, 32008f4f8c16SHugh Dickins * with preemption enabled, without holding any MM locks. 32011da177e4SLinus Torvalds */ 32024f74d2c8SLinus Torvalds while (vma) { 32034f74d2c8SLinus Torvalds if (vma->vm_flags & VM_ACCOUNT) 32044f74d2c8SLinus Torvalds nr_accounted += vma_pages(vma); 3205a8fb5618SHugh Dickins vma = remove_vma(vma); 32060a3b3c25SPaul E. McKenney cond_resched(); 32074f74d2c8SLinus Torvalds } 32084f74d2c8SLinus Torvalds vm_unacct_memory(nr_accounted); 32091da177e4SLinus Torvalds } 32101da177e4SLinus Torvalds 32111da177e4SLinus Torvalds /* Insert vm structure into process list sorted by address 32121da177e4SLinus Torvalds * and into the inode's i_mmap tree. If vm_file is non-NULL 3213c8c06efaSDavidlohr Bueso * then i_mmap_rwsem is taken here. 32141da177e4SLinus Torvalds */ 32151da177e4SLinus Torvalds int insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma) 32161da177e4SLinus Torvalds { 32176597d783SHugh Dickins struct vm_area_struct *prev; 32181da177e4SLinus Torvalds struct rb_node **rb_link, *rb_parent; 32191da177e4SLinus Torvalds 3220c9d13f5fSChen Gang if (find_vma_links(mm, vma->vm_start, vma->vm_end, 3221c9d13f5fSChen Gang &prev, &rb_link, &rb_parent)) 3222c9d13f5fSChen Gang return -ENOMEM; 3223c9d13f5fSChen Gang if ((vma->vm_flags & VM_ACCOUNT) && 3224c9d13f5fSChen Gang security_vm_enough_memory_mm(mm, vma_pages(vma))) 3225c9d13f5fSChen Gang return -ENOMEM; 3226c9d13f5fSChen Gang 32271da177e4SLinus Torvalds /* 32281da177e4SLinus Torvalds * The vm_pgoff of a purely anonymous vma should be irrelevant 32291da177e4SLinus Torvalds * until its first write fault, when page's anon_vma and index 32301da177e4SLinus Torvalds * are set. But now set the vm_pgoff it will almost certainly 32311da177e4SLinus Torvalds * end up with (unless mremap moves it elsewhere before that 32321da177e4SLinus Torvalds * first wfault), so /proc/pid/maps tells a consistent story. 32331da177e4SLinus Torvalds * 32341da177e4SLinus Torvalds * By setting it to reflect the virtual start address of the 32351da177e4SLinus Torvalds * vma, merges and splits can happen in a seamless way, just 32361da177e4SLinus Torvalds * using the existing file pgoff checks and manipulations. 323745e55300SPeter Collingbourne * Similarly in do_mmap and in do_brk. 32381da177e4SLinus Torvalds */ 32398a9cc3b5SOleg Nesterov if (vma_is_anonymous(vma)) { 32401da177e4SLinus Torvalds BUG_ON(vma->anon_vma); 32411da177e4SLinus Torvalds vma->vm_pgoff = vma->vm_start >> PAGE_SHIFT; 32421da177e4SLinus Torvalds } 32432b144498SSrikar Dronamraju 32441da177e4SLinus Torvalds vma_link(mm, vma, prev, rb_link, rb_parent); 32451da177e4SLinus Torvalds return 0; 32461da177e4SLinus Torvalds } 32471da177e4SLinus Torvalds 32481da177e4SLinus Torvalds /* 32491da177e4SLinus Torvalds * Copy the vma structure to a new location in the same mm, 32501da177e4SLinus Torvalds * prior to moving page table entries, to effect an mremap move. 32511da177e4SLinus Torvalds */ 32521da177e4SLinus Torvalds struct vm_area_struct *copy_vma(struct vm_area_struct **vmap, 325338a76013SMichel Lespinasse unsigned long addr, unsigned long len, pgoff_t pgoff, 325438a76013SMichel Lespinasse bool *need_rmap_locks) 32551da177e4SLinus Torvalds { 32561da177e4SLinus Torvalds struct vm_area_struct *vma = *vmap; 32571da177e4SLinus Torvalds unsigned long vma_start = vma->vm_start; 32581da177e4SLinus Torvalds struct mm_struct *mm = vma->vm_mm; 32591da177e4SLinus Torvalds struct vm_area_struct *new_vma, *prev; 32601da177e4SLinus Torvalds struct rb_node **rb_link, *rb_parent; 3261948f017bSAndrea Arcangeli bool faulted_in_anon_vma = true; 32621da177e4SLinus Torvalds 32631da177e4SLinus Torvalds /* 32641da177e4SLinus Torvalds * If anonymous vma has not yet been faulted, update new pgoff 32651da177e4SLinus Torvalds * to match new location, to increase its chance of merging. 32661da177e4SLinus Torvalds */ 3267ce75799bSOleg Nesterov if (unlikely(vma_is_anonymous(vma) && !vma->anon_vma)) { 32681da177e4SLinus Torvalds pgoff = addr >> PAGE_SHIFT; 3269948f017bSAndrea Arcangeli faulted_in_anon_vma = false; 3270948f017bSAndrea Arcangeli } 32711da177e4SLinus Torvalds 32726597d783SHugh Dickins if (find_vma_links(mm, addr, addr + len, &prev, &rb_link, &rb_parent)) 32736597d783SHugh Dickins return NULL; /* should never get here */ 32741da177e4SLinus Torvalds new_vma = vma_merge(mm, prev, addr, addr + len, vma->vm_flags, 327519a809afSAndrea Arcangeli vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma), 327619a809afSAndrea Arcangeli vma->vm_userfaultfd_ctx); 32771da177e4SLinus Torvalds if (new_vma) { 32781da177e4SLinus Torvalds /* 32791da177e4SLinus Torvalds * Source vma may have been merged into new_vma 32801da177e4SLinus Torvalds */ 3281948f017bSAndrea Arcangeli if (unlikely(vma_start >= new_vma->vm_start && 3282948f017bSAndrea Arcangeli vma_start < new_vma->vm_end)) { 3283948f017bSAndrea Arcangeli /* 3284948f017bSAndrea Arcangeli * The only way we can get a vma_merge with 3285948f017bSAndrea Arcangeli * self during an mremap is if the vma hasn't 3286948f017bSAndrea Arcangeli * been faulted in yet and we were allowed to 3287948f017bSAndrea Arcangeli * reset the dst vma->vm_pgoff to the 3288948f017bSAndrea Arcangeli * destination address of the mremap to allow 3289948f017bSAndrea Arcangeli * the merge to happen. mremap must change the 3290948f017bSAndrea Arcangeli * vm_pgoff linearity between src and dst vmas 3291948f017bSAndrea Arcangeli * (in turn preventing a vma_merge) to be 3292948f017bSAndrea Arcangeli * safe. It is only safe to keep the vm_pgoff 3293948f017bSAndrea Arcangeli * linear if there are no pages mapped yet. 3294948f017bSAndrea Arcangeli */ 329581d1b09cSSasha Levin VM_BUG_ON_VMA(faulted_in_anon_vma, new_vma); 329638a76013SMichel Lespinasse *vmap = vma = new_vma; 3297108d6642SMichel Lespinasse } 329838a76013SMichel Lespinasse *need_rmap_locks = (new_vma->vm_pgoff <= vma->vm_pgoff); 32991da177e4SLinus Torvalds } else { 33003928d4f5SLinus Torvalds new_vma = vm_area_dup(vma); 3301e3975891SChen Gang if (!new_vma) 3302e3975891SChen Gang goto out; 33031da177e4SLinus Torvalds new_vma->vm_start = addr; 33041da177e4SLinus Torvalds new_vma->vm_end = addr + len; 33051da177e4SLinus Torvalds new_vma->vm_pgoff = pgoff; 3306ef0855d3SOleg Nesterov if (vma_dup_policy(vma, new_vma)) 3307523d4e20SMichel Lespinasse goto out_free_vma; 3308523d4e20SMichel Lespinasse if (anon_vma_clone(new_vma, vma)) 3309523d4e20SMichel Lespinasse goto out_free_mempol; 3310e9714acfSKonstantin Khlebnikov if (new_vma->vm_file) 33111da177e4SLinus Torvalds get_file(new_vma->vm_file); 33121da177e4SLinus Torvalds if (new_vma->vm_ops && new_vma->vm_ops->open) 33131da177e4SLinus Torvalds new_vma->vm_ops->open(new_vma); 33141da177e4SLinus Torvalds vma_link(mm, new_vma, prev, rb_link, rb_parent); 331538a76013SMichel Lespinasse *need_rmap_locks = false; 33161da177e4SLinus Torvalds } 33171da177e4SLinus Torvalds return new_vma; 33185beb4930SRik van Riel 33195beb4930SRik van Riel out_free_mempol: 3320ef0855d3SOleg Nesterov mpol_put(vma_policy(new_vma)); 33215beb4930SRik van Riel out_free_vma: 33223928d4f5SLinus Torvalds vm_area_free(new_vma); 3323e3975891SChen Gang out: 33245beb4930SRik van Riel return NULL; 33251da177e4SLinus Torvalds } 3326119f657cSakpm@osdl.org 3327119f657cSakpm@osdl.org /* 3328119f657cSakpm@osdl.org * Return true if the calling process may expand its vm space by the passed 3329119f657cSakpm@osdl.org * number of pages 3330119f657cSakpm@osdl.org */ 333184638335SKonstantin Khlebnikov bool may_expand_vm(struct mm_struct *mm, vm_flags_t flags, unsigned long npages) 3332119f657cSakpm@osdl.org { 333384638335SKonstantin Khlebnikov if (mm->total_vm + npages > rlimit(RLIMIT_AS) >> PAGE_SHIFT) 333484638335SKonstantin Khlebnikov return false; 3335119f657cSakpm@osdl.org 3336d977d56cSKonstantin Khlebnikov if (is_data_mapping(flags) && 3337d977d56cSKonstantin Khlebnikov mm->data_vm + npages > rlimit(RLIMIT_DATA) >> PAGE_SHIFT) { 3338f4fcd558SKonstantin Khlebnikov /* Workaround for Valgrind */ 3339f4fcd558SKonstantin Khlebnikov if (rlimit(RLIMIT_DATA) == 0 && 3340f4fcd558SKonstantin Khlebnikov mm->data_vm + npages <= rlimit_max(RLIMIT_DATA) >> PAGE_SHIFT) 3341f4fcd558SKonstantin Khlebnikov return true; 334257a7702bSDavid Woodhouse 334357a7702bSDavid Woodhouse pr_warn_once("%s (%d): VmData %lu exceed data ulimit %lu. Update limits%s.\n", 3344d977d56cSKonstantin Khlebnikov current->comm, current->pid, 3345d977d56cSKonstantin Khlebnikov (mm->data_vm + npages) << PAGE_SHIFT, 334657a7702bSDavid Woodhouse rlimit(RLIMIT_DATA), 334757a7702bSDavid Woodhouse ignore_rlimit_data ? "" : " or use boot option ignore_rlimit_data"); 334857a7702bSDavid Woodhouse 334957a7702bSDavid Woodhouse if (!ignore_rlimit_data) 3350d977d56cSKonstantin Khlebnikov return false; 3351d977d56cSKonstantin Khlebnikov } 3352119f657cSakpm@osdl.org 335384638335SKonstantin Khlebnikov return true; 335484638335SKonstantin Khlebnikov } 335584638335SKonstantin Khlebnikov 335684638335SKonstantin Khlebnikov void vm_stat_account(struct mm_struct *mm, vm_flags_t flags, long npages) 335784638335SKonstantin Khlebnikov { 335884638335SKonstantin Khlebnikov mm->total_vm += npages; 335984638335SKonstantin Khlebnikov 3360d977d56cSKonstantin Khlebnikov if (is_exec_mapping(flags)) 336184638335SKonstantin Khlebnikov mm->exec_vm += npages; 3362d977d56cSKonstantin Khlebnikov else if (is_stack_mapping(flags)) 336384638335SKonstantin Khlebnikov mm->stack_vm += npages; 3364d977d56cSKonstantin Khlebnikov else if (is_data_mapping(flags)) 336584638335SKonstantin Khlebnikov mm->data_vm += npages; 3366119f657cSakpm@osdl.org } 3367fa5dc22fSRoland McGrath 3368b3ec9f33SSouptick Joarder static vm_fault_t special_mapping_fault(struct vm_fault *vmf); 3369a62c34bdSAndy Lutomirski 3370a62c34bdSAndy Lutomirski /* 3371a62c34bdSAndy Lutomirski * Having a close hook prevents vma merging regardless of flags. 3372a62c34bdSAndy Lutomirski */ 3373a62c34bdSAndy Lutomirski static void special_mapping_close(struct vm_area_struct *vma) 3374a62c34bdSAndy Lutomirski { 3375a62c34bdSAndy Lutomirski } 3376a62c34bdSAndy Lutomirski 3377a62c34bdSAndy Lutomirski static const char *special_mapping_name(struct vm_area_struct *vma) 3378a62c34bdSAndy Lutomirski { 3379a62c34bdSAndy Lutomirski return ((struct vm_special_mapping *)vma->vm_private_data)->name; 3380a62c34bdSAndy Lutomirski } 3381a62c34bdSAndy Lutomirski 3382b059a453SDmitry Safonov static int special_mapping_mremap(struct vm_area_struct *new_vma) 3383b059a453SDmitry Safonov { 3384b059a453SDmitry Safonov struct vm_special_mapping *sm = new_vma->vm_private_data; 3385b059a453SDmitry Safonov 3386280e87e9SDmitry Safonov if (WARN_ON_ONCE(current->mm != new_vma->vm_mm)) 3387280e87e9SDmitry Safonov return -EFAULT; 3388280e87e9SDmitry Safonov 3389b059a453SDmitry Safonov if (sm->mremap) 3390b059a453SDmitry Safonov return sm->mremap(sm, new_vma); 3391280e87e9SDmitry Safonov 3392b059a453SDmitry Safonov return 0; 3393b059a453SDmitry Safonov } 3394b059a453SDmitry Safonov 3395a62c34bdSAndy Lutomirski static const struct vm_operations_struct special_mapping_vmops = { 3396a62c34bdSAndy Lutomirski .close = special_mapping_close, 3397a62c34bdSAndy Lutomirski .fault = special_mapping_fault, 3398b059a453SDmitry Safonov .mremap = special_mapping_mremap, 3399a62c34bdSAndy Lutomirski .name = special_mapping_name, 3400af34ebebSDmitry Safonov /* vDSO code relies that VVAR can't be accessed remotely */ 3401af34ebebSDmitry Safonov .access = NULL, 3402a62c34bdSAndy Lutomirski }; 3403a62c34bdSAndy Lutomirski 3404a62c34bdSAndy Lutomirski static const struct vm_operations_struct legacy_special_mapping_vmops = { 3405a62c34bdSAndy Lutomirski .close = special_mapping_close, 3406a62c34bdSAndy Lutomirski .fault = special_mapping_fault, 3407a62c34bdSAndy Lutomirski }; 3408fa5dc22fSRoland McGrath 3409b3ec9f33SSouptick Joarder static vm_fault_t special_mapping_fault(struct vm_fault *vmf) 3410fa5dc22fSRoland McGrath { 341111bac800SDave Jiang struct vm_area_struct *vma = vmf->vma; 3412b1d0e4f5SNick Piggin pgoff_t pgoff; 3413fa5dc22fSRoland McGrath struct page **pages; 3414fa5dc22fSRoland McGrath 3415f872f540SAndy Lutomirski if (vma->vm_ops == &legacy_special_mapping_vmops) { 3416a62c34bdSAndy Lutomirski pages = vma->vm_private_data; 3417f872f540SAndy Lutomirski } else { 3418f872f540SAndy Lutomirski struct vm_special_mapping *sm = vma->vm_private_data; 3419f872f540SAndy Lutomirski 3420f872f540SAndy Lutomirski if (sm->fault) 342111bac800SDave Jiang return sm->fault(sm, vmf->vma, vmf); 3422f872f540SAndy Lutomirski 3423f872f540SAndy Lutomirski pages = sm->pages; 3424f872f540SAndy Lutomirski } 3425a62c34bdSAndy Lutomirski 34268a9cc3b5SOleg Nesterov for (pgoff = vmf->pgoff; pgoff && *pages; ++pages) 3427b1d0e4f5SNick Piggin pgoff--; 3428fa5dc22fSRoland McGrath 3429fa5dc22fSRoland McGrath if (*pages) { 3430fa5dc22fSRoland McGrath struct page *page = *pages; 3431fa5dc22fSRoland McGrath get_page(page); 3432b1d0e4f5SNick Piggin vmf->page = page; 3433b1d0e4f5SNick Piggin return 0; 3434fa5dc22fSRoland McGrath } 3435fa5dc22fSRoland McGrath 3436b1d0e4f5SNick Piggin return VM_FAULT_SIGBUS; 3437fa5dc22fSRoland McGrath } 3438fa5dc22fSRoland McGrath 3439a62c34bdSAndy Lutomirski static struct vm_area_struct *__install_special_mapping( 3440a62c34bdSAndy Lutomirski struct mm_struct *mm, 3441fa5dc22fSRoland McGrath unsigned long addr, unsigned long len, 344227f28b97SChen Gang unsigned long vm_flags, void *priv, 344327f28b97SChen Gang const struct vm_operations_struct *ops) 3444fa5dc22fSRoland McGrath { 3445462e635eSTavis Ormandy int ret; 3446fa5dc22fSRoland McGrath struct vm_area_struct *vma; 3447fa5dc22fSRoland McGrath 3448490fc053SLinus Torvalds vma = vm_area_alloc(mm); 3449fa5dc22fSRoland McGrath if (unlikely(vma == NULL)) 34503935ed6aSStefani Seibold return ERR_PTR(-ENOMEM); 3451fa5dc22fSRoland McGrath 3452fa5dc22fSRoland McGrath vma->vm_start = addr; 3453fa5dc22fSRoland McGrath vma->vm_end = addr + len; 3454fa5dc22fSRoland McGrath 3455d9104d1cSCyrill Gorcunov vma->vm_flags = vm_flags | mm->def_flags | VM_DONTEXPAND | VM_SOFTDIRTY; 34563ed75eb8SColy Li vma->vm_page_prot = vm_get_page_prot(vma->vm_flags); 3457fa5dc22fSRoland McGrath 3458a62c34bdSAndy Lutomirski vma->vm_ops = ops; 3459a62c34bdSAndy Lutomirski vma->vm_private_data = priv; 3460fa5dc22fSRoland McGrath 3461462e635eSTavis Ormandy ret = insert_vm_struct(mm, vma); 3462462e635eSTavis Ormandy if (ret) 3463462e635eSTavis Ormandy goto out; 3464fa5dc22fSRoland McGrath 346584638335SKonstantin Khlebnikov vm_stat_account(mm, vma->vm_flags, len >> PAGE_SHIFT); 3466fa5dc22fSRoland McGrath 3467cdd6c482SIngo Molnar perf_event_mmap(vma); 3468089dd79dSPeter Zijlstra 34693935ed6aSStefani Seibold return vma; 3470462e635eSTavis Ormandy 3471462e635eSTavis Ormandy out: 34723928d4f5SLinus Torvalds vm_area_free(vma); 34733935ed6aSStefani Seibold return ERR_PTR(ret); 34743935ed6aSStefani Seibold } 34753935ed6aSStefani Seibold 34762eefd878SDmitry Safonov bool vma_is_special_mapping(const struct vm_area_struct *vma, 34772eefd878SDmitry Safonov const struct vm_special_mapping *sm) 34782eefd878SDmitry Safonov { 34792eefd878SDmitry Safonov return vma->vm_private_data == sm && 34802eefd878SDmitry Safonov (vma->vm_ops == &special_mapping_vmops || 34812eefd878SDmitry Safonov vma->vm_ops == &legacy_special_mapping_vmops); 34822eefd878SDmitry Safonov } 34832eefd878SDmitry Safonov 3484a62c34bdSAndy Lutomirski /* 3485c1e8d7c6SMichel Lespinasse * Called with mm->mmap_lock held for writing. 3486a62c34bdSAndy Lutomirski * Insert a new vma covering the given region, with the given flags. 3487a62c34bdSAndy Lutomirski * Its pages are supplied by the given array of struct page *. 3488a62c34bdSAndy Lutomirski * The array can be shorter than len >> PAGE_SHIFT if it's null-terminated. 3489a62c34bdSAndy Lutomirski * The region past the last page supplied will always produce SIGBUS. 3490a62c34bdSAndy Lutomirski * The array pointer and the pages it points to are assumed to stay alive 3491a62c34bdSAndy Lutomirski * for as long as this mapping might exist. 3492a62c34bdSAndy Lutomirski */ 3493a62c34bdSAndy Lutomirski struct vm_area_struct *_install_special_mapping( 3494a62c34bdSAndy Lutomirski struct mm_struct *mm, 3495a62c34bdSAndy Lutomirski unsigned long addr, unsigned long len, 3496a62c34bdSAndy Lutomirski unsigned long vm_flags, const struct vm_special_mapping *spec) 3497a62c34bdSAndy Lutomirski { 349827f28b97SChen Gang return __install_special_mapping(mm, addr, len, vm_flags, (void *)spec, 349927f28b97SChen Gang &special_mapping_vmops); 3500a62c34bdSAndy Lutomirski } 3501a62c34bdSAndy Lutomirski 35023935ed6aSStefani Seibold int install_special_mapping(struct mm_struct *mm, 35033935ed6aSStefani Seibold unsigned long addr, unsigned long len, 35043935ed6aSStefani Seibold unsigned long vm_flags, struct page **pages) 35053935ed6aSStefani Seibold { 3506a62c34bdSAndy Lutomirski struct vm_area_struct *vma = __install_special_mapping( 350727f28b97SChen Gang mm, addr, len, vm_flags, (void *)pages, 350827f28b97SChen Gang &legacy_special_mapping_vmops); 35093935ed6aSStefani Seibold 351014bd5b45SDuan Jiong return PTR_ERR_OR_ZERO(vma); 3511fa5dc22fSRoland McGrath } 35127906d00cSAndrea Arcangeli 35137906d00cSAndrea Arcangeli static DEFINE_MUTEX(mm_all_locks_mutex); 35147906d00cSAndrea Arcangeli 3515454ed842SPeter Zijlstra static void vm_lock_anon_vma(struct mm_struct *mm, struct anon_vma *anon_vma) 35167906d00cSAndrea Arcangeli { 3517f808c13fSDavidlohr Bueso if (!test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_root.rb_node)) { 35187906d00cSAndrea Arcangeli /* 35197906d00cSAndrea Arcangeli * The LSB of head.next can't change from under us 35207906d00cSAndrea Arcangeli * because we hold the mm_all_locks_mutex. 35217906d00cSAndrea Arcangeli */ 3522da1c55f1SMichel Lespinasse down_write_nest_lock(&anon_vma->root->rwsem, &mm->mmap_lock); 35237906d00cSAndrea Arcangeli /* 35247906d00cSAndrea Arcangeli * We can safely modify head.next after taking the 35255a505085SIngo Molnar * anon_vma->root->rwsem. If some other vma in this mm shares 35267906d00cSAndrea Arcangeli * the same anon_vma we won't take it again. 35277906d00cSAndrea Arcangeli * 35287906d00cSAndrea Arcangeli * No need of atomic instructions here, head.next 35297906d00cSAndrea Arcangeli * can't change from under us thanks to the 35305a505085SIngo Molnar * anon_vma->root->rwsem. 35317906d00cSAndrea Arcangeli */ 35327906d00cSAndrea Arcangeli if (__test_and_set_bit(0, (unsigned long *) 3533f808c13fSDavidlohr Bueso &anon_vma->root->rb_root.rb_root.rb_node)) 35347906d00cSAndrea Arcangeli BUG(); 35357906d00cSAndrea Arcangeli } 35367906d00cSAndrea Arcangeli } 35377906d00cSAndrea Arcangeli 3538454ed842SPeter Zijlstra static void vm_lock_mapping(struct mm_struct *mm, struct address_space *mapping) 35397906d00cSAndrea Arcangeli { 35407906d00cSAndrea Arcangeli if (!test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) { 35417906d00cSAndrea Arcangeli /* 35427906d00cSAndrea Arcangeli * AS_MM_ALL_LOCKS can't change from under us because 35437906d00cSAndrea Arcangeli * we hold the mm_all_locks_mutex. 35447906d00cSAndrea Arcangeli * 35457906d00cSAndrea Arcangeli * Operations on ->flags have to be atomic because 35467906d00cSAndrea Arcangeli * even if AS_MM_ALL_LOCKS is stable thanks to the 35477906d00cSAndrea Arcangeli * mm_all_locks_mutex, there may be other cpus 35487906d00cSAndrea Arcangeli * changing other bitflags in parallel to us. 35497906d00cSAndrea Arcangeli */ 35507906d00cSAndrea Arcangeli if (test_and_set_bit(AS_MM_ALL_LOCKS, &mapping->flags)) 35517906d00cSAndrea Arcangeli BUG(); 3552da1c55f1SMichel Lespinasse down_write_nest_lock(&mapping->i_mmap_rwsem, &mm->mmap_lock); 35537906d00cSAndrea Arcangeli } 35547906d00cSAndrea Arcangeli } 35557906d00cSAndrea Arcangeli 35567906d00cSAndrea Arcangeli /* 35577906d00cSAndrea Arcangeli * This operation locks against the VM for all pte/vma/mm related 35587906d00cSAndrea Arcangeli * operations that could ever happen on a certain mm. This includes 35597906d00cSAndrea Arcangeli * vmtruncate, try_to_unmap, and all page faults. 35607906d00cSAndrea Arcangeli * 3561c1e8d7c6SMichel Lespinasse * The caller must take the mmap_lock in write mode before calling 35627906d00cSAndrea Arcangeli * mm_take_all_locks(). The caller isn't allowed to release the 3563c1e8d7c6SMichel Lespinasse * mmap_lock until mm_drop_all_locks() returns. 35647906d00cSAndrea Arcangeli * 3565c1e8d7c6SMichel Lespinasse * mmap_lock in write mode is required in order to block all operations 35667906d00cSAndrea Arcangeli * that could modify pagetables and free pages without need of 356727ba0644SKirill A. Shutemov * altering the vma layout. It's also needed in write mode to avoid new 35687906d00cSAndrea Arcangeli * anon_vmas to be associated with existing vmas. 35697906d00cSAndrea Arcangeli * 35707906d00cSAndrea Arcangeli * A single task can't take more than one mm_take_all_locks() in a row 35717906d00cSAndrea Arcangeli * or it would deadlock. 35727906d00cSAndrea Arcangeli * 3573bf181b9fSMichel Lespinasse * The LSB in anon_vma->rb_root.rb_node and the AS_MM_ALL_LOCKS bitflag in 35747906d00cSAndrea Arcangeli * mapping->flags avoid to take the same lock twice, if more than one 35757906d00cSAndrea Arcangeli * vma in this mm is backed by the same anon_vma or address_space. 35767906d00cSAndrea Arcangeli * 357788f306b6SKirill A. Shutemov * We take locks in following order, accordingly to comment at beginning 357888f306b6SKirill A. Shutemov * of mm/rmap.c: 357988f306b6SKirill A. Shutemov * - all hugetlbfs_i_mmap_rwsem_key locks (aka mapping->i_mmap_rwsem for 358088f306b6SKirill A. Shutemov * hugetlb mapping); 358188f306b6SKirill A. Shutemov * - all i_mmap_rwsem locks; 358288f306b6SKirill A. Shutemov * - all anon_vma->rwseml 358388f306b6SKirill A. Shutemov * 358488f306b6SKirill A. Shutemov * We can take all locks within these types randomly because the VM code 358588f306b6SKirill A. Shutemov * doesn't nest them and we protected from parallel mm_take_all_locks() by 358688f306b6SKirill A. Shutemov * mm_all_locks_mutex. 35877906d00cSAndrea Arcangeli * 35887906d00cSAndrea Arcangeli * mm_take_all_locks() and mm_drop_all_locks are expensive operations 35897906d00cSAndrea Arcangeli * that may have to take thousand of locks. 35907906d00cSAndrea Arcangeli * 35917906d00cSAndrea Arcangeli * mm_take_all_locks() can fail if it's interrupted by signals. 35927906d00cSAndrea Arcangeli */ 35937906d00cSAndrea Arcangeli int mm_take_all_locks(struct mm_struct *mm) 35947906d00cSAndrea Arcangeli { 35957906d00cSAndrea Arcangeli struct vm_area_struct *vma; 35965beb4930SRik van Riel struct anon_vma_chain *avc; 35977906d00cSAndrea Arcangeli 3598d8ed45c5SMichel Lespinasse BUG_ON(mmap_read_trylock(mm)); 35997906d00cSAndrea Arcangeli 36007906d00cSAndrea Arcangeli mutex_lock(&mm_all_locks_mutex); 36017906d00cSAndrea Arcangeli 36027906d00cSAndrea Arcangeli for (vma = mm->mmap; vma; vma = vma->vm_next) { 36037906d00cSAndrea Arcangeli if (signal_pending(current)) 36047906d00cSAndrea Arcangeli goto out_unlock; 360588f306b6SKirill A. Shutemov if (vma->vm_file && vma->vm_file->f_mapping && 360688f306b6SKirill A. Shutemov is_vm_hugetlb_page(vma)) 360788f306b6SKirill A. Shutemov vm_lock_mapping(mm, vma->vm_file->f_mapping); 360888f306b6SKirill A. Shutemov } 360988f306b6SKirill A. Shutemov 361088f306b6SKirill A. Shutemov for (vma = mm->mmap; vma; vma = vma->vm_next) { 361188f306b6SKirill A. Shutemov if (signal_pending(current)) 361288f306b6SKirill A. Shutemov goto out_unlock; 361388f306b6SKirill A. Shutemov if (vma->vm_file && vma->vm_file->f_mapping && 361488f306b6SKirill A. Shutemov !is_vm_hugetlb_page(vma)) 3615454ed842SPeter Zijlstra vm_lock_mapping(mm, vma->vm_file->f_mapping); 36167906d00cSAndrea Arcangeli } 36177cd5a02fSPeter Zijlstra 36187cd5a02fSPeter Zijlstra for (vma = mm->mmap; vma; vma = vma->vm_next) { 36197cd5a02fSPeter Zijlstra if (signal_pending(current)) 36207cd5a02fSPeter Zijlstra goto out_unlock; 36217cd5a02fSPeter Zijlstra if (vma->anon_vma) 36225beb4930SRik van Riel list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) 36235beb4930SRik van Riel vm_lock_anon_vma(mm, avc->anon_vma); 36247cd5a02fSPeter Zijlstra } 36257cd5a02fSPeter Zijlstra 3626584cff54SKautuk Consul return 0; 36277906d00cSAndrea Arcangeli 36287906d00cSAndrea Arcangeli out_unlock: 36297906d00cSAndrea Arcangeli mm_drop_all_locks(mm); 3630584cff54SKautuk Consul return -EINTR; 36317906d00cSAndrea Arcangeli } 36327906d00cSAndrea Arcangeli 36337906d00cSAndrea Arcangeli static void vm_unlock_anon_vma(struct anon_vma *anon_vma) 36347906d00cSAndrea Arcangeli { 3635f808c13fSDavidlohr Bueso if (test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_root.rb_node)) { 36367906d00cSAndrea Arcangeli /* 36377906d00cSAndrea Arcangeli * The LSB of head.next can't change to 0 from under 36387906d00cSAndrea Arcangeli * us because we hold the mm_all_locks_mutex. 36397906d00cSAndrea Arcangeli * 36407906d00cSAndrea Arcangeli * We must however clear the bitflag before unlocking 3641bf181b9fSMichel Lespinasse * the vma so the users using the anon_vma->rb_root will 36427906d00cSAndrea Arcangeli * never see our bitflag. 36437906d00cSAndrea Arcangeli * 36447906d00cSAndrea Arcangeli * No need of atomic instructions here, head.next 36457906d00cSAndrea Arcangeli * can't change from under us until we release the 36465a505085SIngo Molnar * anon_vma->root->rwsem. 36477906d00cSAndrea Arcangeli */ 36487906d00cSAndrea Arcangeli if (!__test_and_clear_bit(0, (unsigned long *) 3649f808c13fSDavidlohr Bueso &anon_vma->root->rb_root.rb_root.rb_node)) 36507906d00cSAndrea Arcangeli BUG(); 365108b52706SKonstantin Khlebnikov anon_vma_unlock_write(anon_vma); 36527906d00cSAndrea Arcangeli } 36537906d00cSAndrea Arcangeli } 36547906d00cSAndrea Arcangeli 36557906d00cSAndrea Arcangeli static void vm_unlock_mapping(struct address_space *mapping) 36567906d00cSAndrea Arcangeli { 36577906d00cSAndrea Arcangeli if (test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) { 36587906d00cSAndrea Arcangeli /* 36597906d00cSAndrea Arcangeli * AS_MM_ALL_LOCKS can't change to 0 from under us 36607906d00cSAndrea Arcangeli * because we hold the mm_all_locks_mutex. 36617906d00cSAndrea Arcangeli */ 366283cde9e8SDavidlohr Bueso i_mmap_unlock_write(mapping); 36637906d00cSAndrea Arcangeli if (!test_and_clear_bit(AS_MM_ALL_LOCKS, 36647906d00cSAndrea Arcangeli &mapping->flags)) 36657906d00cSAndrea Arcangeli BUG(); 36667906d00cSAndrea Arcangeli } 36677906d00cSAndrea Arcangeli } 36687906d00cSAndrea Arcangeli 36697906d00cSAndrea Arcangeli /* 3670c1e8d7c6SMichel Lespinasse * The mmap_lock cannot be released by the caller until 36717906d00cSAndrea Arcangeli * mm_drop_all_locks() returns. 36727906d00cSAndrea Arcangeli */ 36737906d00cSAndrea Arcangeli void mm_drop_all_locks(struct mm_struct *mm) 36747906d00cSAndrea Arcangeli { 36757906d00cSAndrea Arcangeli struct vm_area_struct *vma; 36765beb4930SRik van Riel struct anon_vma_chain *avc; 36777906d00cSAndrea Arcangeli 3678d8ed45c5SMichel Lespinasse BUG_ON(mmap_read_trylock(mm)); 36797906d00cSAndrea Arcangeli BUG_ON(!mutex_is_locked(&mm_all_locks_mutex)); 36807906d00cSAndrea Arcangeli 36817906d00cSAndrea Arcangeli for (vma = mm->mmap; vma; vma = vma->vm_next) { 36827906d00cSAndrea Arcangeli if (vma->anon_vma) 36835beb4930SRik van Riel list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) 36845beb4930SRik van Riel vm_unlock_anon_vma(avc->anon_vma); 36857906d00cSAndrea Arcangeli if (vma->vm_file && vma->vm_file->f_mapping) 36867906d00cSAndrea Arcangeli vm_unlock_mapping(vma->vm_file->f_mapping); 36877906d00cSAndrea Arcangeli } 36887906d00cSAndrea Arcangeli 36897906d00cSAndrea Arcangeli mutex_unlock(&mm_all_locks_mutex); 36907906d00cSAndrea Arcangeli } 36918feae131SDavid Howells 36928feae131SDavid Howells /* 36933edf41d8Sseokhoon.yoon * initialise the percpu counter for VM 36948feae131SDavid Howells */ 36958feae131SDavid Howells void __init mmap_init(void) 36968feae131SDavid Howells { 369700a62ce9SKOSAKI Motohiro int ret; 369800a62ce9SKOSAKI Motohiro 3699908c7f19STejun Heo ret = percpu_counter_init(&vm_committed_as, 0, GFP_KERNEL); 370000a62ce9SKOSAKI Motohiro VM_BUG_ON(ret); 37018feae131SDavid Howells } 3702c9b1d098SAndrew Shewmaker 3703c9b1d098SAndrew Shewmaker /* 3704c9b1d098SAndrew Shewmaker * Initialise sysctl_user_reserve_kbytes. 3705c9b1d098SAndrew Shewmaker * 3706c9b1d098SAndrew Shewmaker * This is intended to prevent a user from starting a single memory hogging 3707c9b1d098SAndrew Shewmaker * process, such that they cannot recover (kill the hog) in OVERCOMMIT_NEVER 3708c9b1d098SAndrew Shewmaker * mode. 3709c9b1d098SAndrew Shewmaker * 3710c9b1d098SAndrew Shewmaker * The default value is min(3% of free memory, 128MB) 3711c9b1d098SAndrew Shewmaker * 128MB is enough to recover with sshd/login, bash, and top/kill. 3712c9b1d098SAndrew Shewmaker */ 37131640879aSAndrew Shewmaker static int init_user_reserve(void) 3714c9b1d098SAndrew Shewmaker { 3715c9b1d098SAndrew Shewmaker unsigned long free_kbytes; 3716c9b1d098SAndrew Shewmaker 3717c41f012aSMichal Hocko free_kbytes = global_zone_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10); 3718c9b1d098SAndrew Shewmaker 3719c9b1d098SAndrew Shewmaker sysctl_user_reserve_kbytes = min(free_kbytes / 32, 1UL << 17); 3720c9b1d098SAndrew Shewmaker return 0; 3721c9b1d098SAndrew Shewmaker } 3722a64fb3cdSPaul Gortmaker subsys_initcall(init_user_reserve); 37234eeab4f5SAndrew Shewmaker 37244eeab4f5SAndrew Shewmaker /* 37254eeab4f5SAndrew Shewmaker * Initialise sysctl_admin_reserve_kbytes. 37264eeab4f5SAndrew Shewmaker * 37274eeab4f5SAndrew Shewmaker * The purpose of sysctl_admin_reserve_kbytes is to allow the sys admin 37284eeab4f5SAndrew Shewmaker * to log in and kill a memory hogging process. 37294eeab4f5SAndrew Shewmaker * 37304eeab4f5SAndrew Shewmaker * Systems with more than 256MB will reserve 8MB, enough to recover 37314eeab4f5SAndrew Shewmaker * with sshd, bash, and top in OVERCOMMIT_GUESS. Smaller systems will 37324eeab4f5SAndrew Shewmaker * only reserve 3% of free pages by default. 37334eeab4f5SAndrew Shewmaker */ 37341640879aSAndrew Shewmaker static int init_admin_reserve(void) 37354eeab4f5SAndrew Shewmaker { 37364eeab4f5SAndrew Shewmaker unsigned long free_kbytes; 37374eeab4f5SAndrew Shewmaker 3738c41f012aSMichal Hocko free_kbytes = global_zone_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10); 37394eeab4f5SAndrew Shewmaker 37404eeab4f5SAndrew Shewmaker sysctl_admin_reserve_kbytes = min(free_kbytes / 32, 1UL << 13); 37414eeab4f5SAndrew Shewmaker return 0; 37424eeab4f5SAndrew Shewmaker } 3743a64fb3cdSPaul Gortmaker subsys_initcall(init_admin_reserve); 37441640879aSAndrew Shewmaker 37451640879aSAndrew Shewmaker /* 37461640879aSAndrew Shewmaker * Reinititalise user and admin reserves if memory is added or removed. 37471640879aSAndrew Shewmaker * 37481640879aSAndrew Shewmaker * The default user reserve max is 128MB, and the default max for the 37491640879aSAndrew Shewmaker * admin reserve is 8MB. These are usually, but not always, enough to 37501640879aSAndrew Shewmaker * enable recovery from a memory hogging process using login/sshd, a shell, 37511640879aSAndrew Shewmaker * and tools like top. It may make sense to increase or even disable the 37521640879aSAndrew Shewmaker * reserve depending on the existence of swap or variations in the recovery 37531640879aSAndrew Shewmaker * tools. So, the admin may have changed them. 37541640879aSAndrew Shewmaker * 37551640879aSAndrew Shewmaker * If memory is added and the reserves have been eliminated or increased above 37561640879aSAndrew Shewmaker * the default max, then we'll trust the admin. 37571640879aSAndrew Shewmaker * 37581640879aSAndrew Shewmaker * If memory is removed and there isn't enough free memory, then we 37591640879aSAndrew Shewmaker * need to reset the reserves. 37601640879aSAndrew Shewmaker * 37611640879aSAndrew Shewmaker * Otherwise keep the reserve set by the admin. 37621640879aSAndrew Shewmaker */ 37631640879aSAndrew Shewmaker static int reserve_mem_notifier(struct notifier_block *nb, 37641640879aSAndrew Shewmaker unsigned long action, void *data) 37651640879aSAndrew Shewmaker { 37661640879aSAndrew Shewmaker unsigned long tmp, free_kbytes; 37671640879aSAndrew Shewmaker 37681640879aSAndrew Shewmaker switch (action) { 37691640879aSAndrew Shewmaker case MEM_ONLINE: 37701640879aSAndrew Shewmaker /* Default max is 128MB. Leave alone if modified by operator. */ 37711640879aSAndrew Shewmaker tmp = sysctl_user_reserve_kbytes; 37721640879aSAndrew Shewmaker if (0 < tmp && tmp < (1UL << 17)) 37731640879aSAndrew Shewmaker init_user_reserve(); 37741640879aSAndrew Shewmaker 37751640879aSAndrew Shewmaker /* Default max is 8MB. Leave alone if modified by operator. */ 37761640879aSAndrew Shewmaker tmp = sysctl_admin_reserve_kbytes; 37771640879aSAndrew Shewmaker if (0 < tmp && tmp < (1UL << 13)) 37781640879aSAndrew Shewmaker init_admin_reserve(); 37791640879aSAndrew Shewmaker 37801640879aSAndrew Shewmaker break; 37811640879aSAndrew Shewmaker case MEM_OFFLINE: 3782c41f012aSMichal Hocko free_kbytes = global_zone_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10); 37831640879aSAndrew Shewmaker 37841640879aSAndrew Shewmaker if (sysctl_user_reserve_kbytes > free_kbytes) { 37851640879aSAndrew Shewmaker init_user_reserve(); 37861640879aSAndrew Shewmaker pr_info("vm.user_reserve_kbytes reset to %lu\n", 37871640879aSAndrew Shewmaker sysctl_user_reserve_kbytes); 37881640879aSAndrew Shewmaker } 37891640879aSAndrew Shewmaker 37901640879aSAndrew Shewmaker if (sysctl_admin_reserve_kbytes > free_kbytes) { 37911640879aSAndrew Shewmaker init_admin_reserve(); 37921640879aSAndrew Shewmaker pr_info("vm.admin_reserve_kbytes reset to %lu\n", 37931640879aSAndrew Shewmaker sysctl_admin_reserve_kbytes); 37941640879aSAndrew Shewmaker } 37951640879aSAndrew Shewmaker break; 37961640879aSAndrew Shewmaker default: 37971640879aSAndrew Shewmaker break; 37981640879aSAndrew Shewmaker } 37991640879aSAndrew Shewmaker return NOTIFY_OK; 38001640879aSAndrew Shewmaker } 38011640879aSAndrew Shewmaker 38021640879aSAndrew Shewmaker static struct notifier_block reserve_mem_nb = { 38031640879aSAndrew Shewmaker .notifier_call = reserve_mem_notifier, 38041640879aSAndrew Shewmaker }; 38051640879aSAndrew Shewmaker 38061640879aSAndrew Shewmaker static int __meminit init_reserve_notifier(void) 38071640879aSAndrew Shewmaker { 38081640879aSAndrew Shewmaker if (register_hotmemory_notifier(&reserve_mem_nb)) 3809b1de0d13SMitchel Humpherys pr_err("Failed registering memory add/remove notifier for admin reserve\n"); 38101640879aSAndrew Shewmaker 38111640879aSAndrew Shewmaker return 0; 38121640879aSAndrew Shewmaker } 3813a64fb3cdSPaul Gortmaker subsys_initcall(init_reserve_notifier); 3814