1457c8996SThomas Gleixner // SPDX-License-Identifier: GPL-2.0-only 21da177e4SLinus Torvalds /* 31da177e4SLinus Torvalds * mm/mmap.c 41da177e4SLinus Torvalds * 51da177e4SLinus Torvalds * Written by obz. 61da177e4SLinus Torvalds * 7046c6884SAlan Cox * Address space accounting code <alan@lxorguk.ukuu.org.uk> 81da177e4SLinus Torvalds */ 91da177e4SLinus Torvalds 10b1de0d13SMitchel Humpherys #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt 11b1de0d13SMitchel Humpherys 12e8420a8eSCyril Hrubis #include <linux/kernel.h> 131da177e4SLinus Torvalds #include <linux/slab.h> 144af3c9ccSAlexey Dobriyan #include <linux/backing-dev.h> 151da177e4SLinus Torvalds #include <linux/mm.h> 1617fca131SArnd Bergmann #include <linux/mm_inline.h> 17615d6e87SDavidlohr Bueso #include <linux/vmacache.h> 181da177e4SLinus Torvalds #include <linux/shm.h> 191da177e4SLinus Torvalds #include <linux/mman.h> 201da177e4SLinus Torvalds #include <linux/pagemap.h> 211da177e4SLinus Torvalds #include <linux/swap.h> 221da177e4SLinus Torvalds #include <linux/syscalls.h> 23c59ede7bSRandy.Dunlap #include <linux/capability.h> 241da177e4SLinus Torvalds #include <linux/init.h> 251da177e4SLinus Torvalds #include <linux/file.h> 261da177e4SLinus Torvalds #include <linux/fs.h> 271da177e4SLinus Torvalds #include <linux/personality.h> 281da177e4SLinus Torvalds #include <linux/security.h> 291da177e4SLinus Torvalds #include <linux/hugetlb.h> 30c01d5b30SHugh Dickins #include <linux/shmem_fs.h> 311da177e4SLinus Torvalds #include <linux/profile.h> 32b95f1b31SPaul Gortmaker #include <linux/export.h> 331da177e4SLinus Torvalds #include <linux/mount.h> 341da177e4SLinus Torvalds #include <linux/mempolicy.h> 351da177e4SLinus Torvalds #include <linux/rmap.h> 36cddb8a5cSAndrea Arcangeli #include <linux/mmu_notifier.h> 3782f71ae4SKonstantin Khlebnikov #include <linux/mmdebug.h> 38cdd6c482SIngo Molnar #include <linux/perf_event.h> 39120a795dSAl Viro #include <linux/audit.h> 40b15d00b6SAndrea Arcangeli #include <linux/khugepaged.h> 412b144498SSrikar Dronamraju #include <linux/uprobes.h> 42d3737187SMichel Lespinasse #include <linux/rbtree_augmented.h> 431640879aSAndrew Shewmaker #include <linux/notifier.h> 441640879aSAndrew Shewmaker #include <linux/memory.h> 45b1de0d13SMitchel Humpherys #include <linux/printk.h> 4619a809afSAndrea Arcangeli #include <linux/userfaultfd_k.h> 47d977d56cSKonstantin Khlebnikov #include <linux/moduleparam.h> 4862b5f7d0SDave Hansen #include <linux/pkeys.h> 4921292580SAndrea Arcangeli #include <linux/oom.h> 5004f5866eSAndrea Arcangeli #include <linux/sched/mm.h> 511da177e4SLinus Torvalds 527c0f6ba6SLinus Torvalds #include <linux/uaccess.h> 531da177e4SLinus Torvalds #include <asm/cacheflush.h> 541da177e4SLinus Torvalds #include <asm/tlb.h> 55d6dd61c8SJeremy Fitzhardinge #include <asm/mmu_context.h> 561da177e4SLinus Torvalds 57df529cabSJaewon Kim #define CREATE_TRACE_POINTS 58df529cabSJaewon Kim #include <trace/events/mmap.h> 59df529cabSJaewon Kim 6042b77728SJan Beulich #include "internal.h" 6142b77728SJan Beulich 623a459756SKirill Korotaev #ifndef arch_mmap_check 633a459756SKirill Korotaev #define arch_mmap_check(addr, len, flags) (0) 643a459756SKirill Korotaev #endif 653a459756SKirill Korotaev 66d07e2259SDaniel Cashman #ifdef CONFIG_HAVE_ARCH_MMAP_RND_BITS 67d07e2259SDaniel Cashman const int mmap_rnd_bits_min = CONFIG_ARCH_MMAP_RND_BITS_MIN; 68d07e2259SDaniel Cashman const int mmap_rnd_bits_max = CONFIG_ARCH_MMAP_RND_BITS_MAX; 69d07e2259SDaniel Cashman int mmap_rnd_bits __read_mostly = CONFIG_ARCH_MMAP_RND_BITS; 70d07e2259SDaniel Cashman #endif 71d07e2259SDaniel Cashman #ifdef CONFIG_HAVE_ARCH_MMAP_RND_COMPAT_BITS 72d07e2259SDaniel Cashman const int mmap_rnd_compat_bits_min = CONFIG_ARCH_MMAP_RND_COMPAT_BITS_MIN; 73d07e2259SDaniel Cashman const int mmap_rnd_compat_bits_max = CONFIG_ARCH_MMAP_RND_COMPAT_BITS_MAX; 74d07e2259SDaniel Cashman int mmap_rnd_compat_bits __read_mostly = CONFIG_ARCH_MMAP_RND_COMPAT_BITS; 75d07e2259SDaniel Cashman #endif 76d07e2259SDaniel Cashman 77f4fcd558SKonstantin Khlebnikov static bool ignore_rlimit_data; 78d977d56cSKonstantin Khlebnikov core_param(ignore_rlimit_data, ignore_rlimit_data, bool, 0644); 79d07e2259SDaniel Cashman 80e0da382cSHugh Dickins static void unmap_region(struct mm_struct *mm, 81e0da382cSHugh Dickins struct vm_area_struct *vma, struct vm_area_struct *prev, 82e0da382cSHugh Dickins unsigned long start, unsigned long end); 83e0da382cSHugh Dickins 841da177e4SLinus Torvalds /* description of effects of mapping type and prot in current implementation. 851da177e4SLinus Torvalds * this is due to the limited x86 page protection hardware. The expected 861da177e4SLinus Torvalds * behavior is in parens: 871da177e4SLinus Torvalds * 881da177e4SLinus Torvalds * map_type prot 891da177e4SLinus Torvalds * PROT_NONE PROT_READ PROT_WRITE PROT_EXEC 901da177e4SLinus Torvalds * MAP_SHARED r: (no) no r: (yes) yes r: (no) yes r: (no) yes 911da177e4SLinus Torvalds * w: (no) no w: (no) no w: (yes) yes w: (no) no 921da177e4SLinus Torvalds * x: (no) no x: (no) yes x: (no) yes x: (yes) yes 931da177e4SLinus Torvalds * 941da177e4SLinus Torvalds * MAP_PRIVATE r: (no) no r: (yes) yes r: (no) yes r: (no) yes 951da177e4SLinus Torvalds * w: (no) no w: (no) no w: (copy) copy w: (no) no 961da177e4SLinus Torvalds * x: (no) no x: (no) yes x: (no) yes x: (yes) yes 9718107f8aSVladimir Murzin * 9818107f8aSVladimir Murzin * On arm64, PROT_EXEC has the following behaviour for both MAP_SHARED and 9918107f8aSVladimir Murzin * MAP_PRIVATE (with Enhanced PAN supported): 10018107f8aSVladimir Murzin * r: (no) no 10118107f8aSVladimir Murzin * w: (no) no 10218107f8aSVladimir Murzin * x: (yes) yes 1031da177e4SLinus Torvalds */ 104ac34ceafSDaniel Micay pgprot_t protection_map[16] __ro_after_init = { 1056c862bd0SAnshuman Khandual [VM_NONE] = __P000, 1066c862bd0SAnshuman Khandual [VM_READ] = __P001, 1076c862bd0SAnshuman Khandual [VM_WRITE] = __P010, 1086c862bd0SAnshuman Khandual [VM_WRITE | VM_READ] = __P011, 1096c862bd0SAnshuman Khandual [VM_EXEC] = __P100, 1106c862bd0SAnshuman Khandual [VM_EXEC | VM_READ] = __P101, 1116c862bd0SAnshuman Khandual [VM_EXEC | VM_WRITE] = __P110, 1126c862bd0SAnshuman Khandual [VM_EXEC | VM_WRITE | VM_READ] = __P111, 1136c862bd0SAnshuman Khandual [VM_SHARED] = __S000, 1146c862bd0SAnshuman Khandual [VM_SHARED | VM_READ] = __S001, 1156c862bd0SAnshuman Khandual [VM_SHARED | VM_WRITE] = __S010, 1166c862bd0SAnshuman Khandual [VM_SHARED | VM_WRITE | VM_READ] = __S011, 1176c862bd0SAnshuman Khandual [VM_SHARED | VM_EXEC] = __S100, 1186c862bd0SAnshuman Khandual [VM_SHARED | VM_EXEC | VM_READ] = __S101, 1196c862bd0SAnshuman Khandual [VM_SHARED | VM_EXEC | VM_WRITE] = __S110, 1206c862bd0SAnshuman Khandual [VM_SHARED | VM_EXEC | VM_WRITE | VM_READ] = __S111 1211da177e4SLinus Torvalds }; 1221da177e4SLinus Torvalds 123316d097cSDave Hansen #ifndef CONFIG_ARCH_HAS_FILTER_PGPROT 124316d097cSDave Hansen static inline pgprot_t arch_filter_pgprot(pgprot_t prot) 125316d097cSDave Hansen { 126316d097cSDave Hansen return prot; 127316d097cSDave Hansen } 128316d097cSDave Hansen #endif 129316d097cSDave Hansen 130804af2cfSHugh Dickins pgprot_t vm_get_page_prot(unsigned long vm_flags) 131804af2cfSHugh Dickins { 132316d097cSDave Hansen pgprot_t ret = __pgprot(pgprot_val(protection_map[vm_flags & 133b845f313SDave Kleikamp (VM_READ|VM_WRITE|VM_EXEC|VM_SHARED)]) | 134b845f313SDave Kleikamp pgprot_val(arch_vm_get_page_prot(vm_flags))); 135316d097cSDave Hansen 136316d097cSDave Hansen return arch_filter_pgprot(ret); 137804af2cfSHugh Dickins } 138804af2cfSHugh Dickins EXPORT_SYMBOL(vm_get_page_prot); 139804af2cfSHugh Dickins 14064e45507SPeter Feiner static pgprot_t vm_pgprot_modify(pgprot_t oldprot, unsigned long vm_flags) 14164e45507SPeter Feiner { 14264e45507SPeter Feiner return pgprot_modify(oldprot, vm_get_page_prot(vm_flags)); 14364e45507SPeter Feiner } 14464e45507SPeter Feiner 14564e45507SPeter Feiner /* Update vma->vm_page_prot to reflect vma->vm_flags. */ 14664e45507SPeter Feiner void vma_set_page_prot(struct vm_area_struct *vma) 14764e45507SPeter Feiner { 14864e45507SPeter Feiner unsigned long vm_flags = vma->vm_flags; 1496d2329f8SAndrea Arcangeli pgprot_t vm_page_prot; 15064e45507SPeter Feiner 1516d2329f8SAndrea Arcangeli vm_page_prot = vm_pgprot_modify(vma->vm_page_prot, vm_flags); 1526d2329f8SAndrea Arcangeli if (vma_wants_writenotify(vma, vm_page_prot)) { 15364e45507SPeter Feiner vm_flags &= ~VM_SHARED; 1546d2329f8SAndrea Arcangeli vm_page_prot = vm_pgprot_modify(vm_page_prot, vm_flags); 15564e45507SPeter Feiner } 156c1e8d7c6SMichel Lespinasse /* remove_protection_ptes reads vma->vm_page_prot without mmap_lock */ 1576d2329f8SAndrea Arcangeli WRITE_ONCE(vma->vm_page_prot, vm_page_prot); 15864e45507SPeter Feiner } 15964e45507SPeter Feiner 1601da177e4SLinus Torvalds /* 161c8c06efaSDavidlohr Bueso * Requires inode->i_mapping->i_mmap_rwsem 1621da177e4SLinus Torvalds */ 1631da177e4SLinus Torvalds static void __remove_shared_vm_struct(struct vm_area_struct *vma, 1641da177e4SLinus Torvalds struct file *file, struct address_space *mapping) 1651da177e4SLinus Torvalds { 1661da177e4SLinus Torvalds if (vma->vm_flags & VM_SHARED) 1674bb5f5d9SDavid Herrmann mapping_unmap_writable(mapping); 1681da177e4SLinus Torvalds 1691da177e4SLinus Torvalds flush_dcache_mmap_lock(mapping); 1706b2dbba8SMichel Lespinasse vma_interval_tree_remove(vma, &mapping->i_mmap); 1711da177e4SLinus Torvalds flush_dcache_mmap_unlock(mapping); 1721da177e4SLinus Torvalds } 1731da177e4SLinus Torvalds 1741da177e4SLinus Torvalds /* 1756b2dbba8SMichel Lespinasse * Unlink a file-based vm structure from its interval tree, to hide 176a8fb5618SHugh Dickins * vma from rmap and vmtruncate before freeing its page tables. 1771da177e4SLinus Torvalds */ 178a8fb5618SHugh Dickins void unlink_file_vma(struct vm_area_struct *vma) 1791da177e4SLinus Torvalds { 1801da177e4SLinus Torvalds struct file *file = vma->vm_file; 1811da177e4SLinus Torvalds 1821da177e4SLinus Torvalds if (file) { 1831da177e4SLinus Torvalds struct address_space *mapping = file->f_mapping; 18483cde9e8SDavidlohr Bueso i_mmap_lock_write(mapping); 1851da177e4SLinus Torvalds __remove_shared_vm_struct(vma, file, mapping); 18683cde9e8SDavidlohr Bueso i_mmap_unlock_write(mapping); 1871da177e4SLinus Torvalds } 188a8fb5618SHugh Dickins } 189a8fb5618SHugh Dickins 190a8fb5618SHugh Dickins /* 191a8fb5618SHugh Dickins * Close a vm structure and free it, returning the next. 192a8fb5618SHugh Dickins */ 193a8fb5618SHugh Dickins static struct vm_area_struct *remove_vma(struct vm_area_struct *vma) 194a8fb5618SHugh Dickins { 195a8fb5618SHugh Dickins struct vm_area_struct *next = vma->vm_next; 196a8fb5618SHugh Dickins 197a8fb5618SHugh Dickins might_sleep(); 1981da177e4SLinus Torvalds if (vma->vm_ops && vma->vm_ops->close) 1991da177e4SLinus Torvalds vma->vm_ops->close(vma); 200e9714acfSKonstantin Khlebnikov if (vma->vm_file) 201a8fb5618SHugh Dickins fput(vma->vm_file); 202f0be3d32SLee Schermerhorn mpol_put(vma_policy(vma)); 2033928d4f5SLinus Torvalds vm_area_free(vma); 204a8fb5618SHugh Dickins return next; 2051da177e4SLinus Torvalds } 2061da177e4SLinus Torvalds 207bb177a73SMichal Hocko static int do_brk_flags(unsigned long addr, unsigned long request, unsigned long flags, 208bb177a73SMichal Hocko struct list_head *uf); 2096a6160a7SHeiko Carstens SYSCALL_DEFINE1(brk, unsigned long, brk) 2101da177e4SLinus Torvalds { 2119bc8039eSYang Shi unsigned long newbrk, oldbrk, origbrk; 2121da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 2131be7107fSHugh Dickins struct vm_area_struct *next; 214a5b4592cSJiri Kosina unsigned long min_brk; 215128557ffSMichel Lespinasse bool populate; 2169bc8039eSYang Shi bool downgraded = false; 217897ab3e0SMike Rapoport LIST_HEAD(uf); 2181da177e4SLinus Torvalds 219d8ed45c5SMichel Lespinasse if (mmap_write_lock_killable(mm)) 220dc0ef0dfSMichal Hocko return -EINTR; 2211da177e4SLinus Torvalds 2229bc8039eSYang Shi origbrk = mm->brk; 2239bc8039eSYang Shi 224a5b4592cSJiri Kosina #ifdef CONFIG_COMPAT_BRK 2255520e894SJiri Kosina /* 2265520e894SJiri Kosina * CONFIG_COMPAT_BRK can still be overridden by setting 2275520e894SJiri Kosina * randomize_va_space to 2, which will still cause mm->start_brk 2285520e894SJiri Kosina * to be arbitrarily shifted 2295520e894SJiri Kosina */ 2304471a675SJiri Kosina if (current->brk_randomized) 2315520e894SJiri Kosina min_brk = mm->start_brk; 2325520e894SJiri Kosina else 2335520e894SJiri Kosina min_brk = mm->end_data; 234a5b4592cSJiri Kosina #else 235a5b4592cSJiri Kosina min_brk = mm->start_brk; 236a5b4592cSJiri Kosina #endif 237a5b4592cSJiri Kosina if (brk < min_brk) 2381da177e4SLinus Torvalds goto out; 2391e624196SRam Gupta 2401e624196SRam Gupta /* 2411e624196SRam Gupta * Check against rlimit here. If this check is done later after the test 2421e624196SRam Gupta * of oldbrk with newbrk then it can escape the test and let the data 2431e624196SRam Gupta * segment grow beyond its set limit the in case where the limit is 2441e624196SRam Gupta * not page aligned -Ram Gupta 2451e624196SRam Gupta */ 2468764b338SCyrill Gorcunov if (check_data_rlimit(rlimit(RLIMIT_DATA), brk, mm->start_brk, 2478764b338SCyrill Gorcunov mm->end_data, mm->start_data)) 2481e624196SRam Gupta goto out; 2491e624196SRam Gupta 2501da177e4SLinus Torvalds newbrk = PAGE_ALIGN(brk); 2511da177e4SLinus Torvalds oldbrk = PAGE_ALIGN(mm->brk); 2529bc8039eSYang Shi if (oldbrk == newbrk) { 2539bc8039eSYang Shi mm->brk = brk; 2549bc8039eSYang Shi goto success; 2559bc8039eSYang Shi } 2561da177e4SLinus Torvalds 2579bc8039eSYang Shi /* 2589bc8039eSYang Shi * Always allow shrinking brk. 259c1e8d7c6SMichel Lespinasse * __do_munmap() may downgrade mmap_lock to read. 2609bc8039eSYang Shi */ 2611da177e4SLinus Torvalds if (brk <= mm->brk) { 2629bc8039eSYang Shi int ret; 2639bc8039eSYang Shi 2649bc8039eSYang Shi /* 265c1e8d7c6SMichel Lespinasse * mm->brk must to be protected by write mmap_lock so update it 266c1e8d7c6SMichel Lespinasse * before downgrading mmap_lock. When __do_munmap() fails, 2679bc8039eSYang Shi * mm->brk will be restored from origbrk. 2689bc8039eSYang Shi */ 2699bc8039eSYang Shi mm->brk = brk; 2709bc8039eSYang Shi ret = __do_munmap(mm, newbrk, oldbrk-newbrk, &uf, true); 2719bc8039eSYang Shi if (ret < 0) { 2729bc8039eSYang Shi mm->brk = origbrk; 2731da177e4SLinus Torvalds goto out; 2749bc8039eSYang Shi } else if (ret == 1) { 2759bc8039eSYang Shi downgraded = true; 2769bc8039eSYang Shi } 2779bc8039eSYang Shi goto success; 2781da177e4SLinus Torvalds } 2791da177e4SLinus Torvalds 2801da177e4SLinus Torvalds /* Check against existing mmap mappings. */ 2811be7107fSHugh Dickins next = find_vma(mm, oldbrk); 2821be7107fSHugh Dickins if (next && newbrk + PAGE_SIZE > vm_start_gap(next)) 2831da177e4SLinus Torvalds goto out; 2841da177e4SLinus Torvalds 2851da177e4SLinus Torvalds /* Ok, looks good - let it rip. */ 286bb177a73SMichal Hocko if (do_brk_flags(oldbrk, newbrk-oldbrk, 0, &uf) < 0) 2871da177e4SLinus Torvalds goto out; 2881da177e4SLinus Torvalds mm->brk = brk; 2899bc8039eSYang Shi 2909bc8039eSYang Shi success: 291128557ffSMichel Lespinasse populate = newbrk > oldbrk && (mm->def_flags & VM_LOCKED) != 0; 2929bc8039eSYang Shi if (downgraded) 293d8ed45c5SMichel Lespinasse mmap_read_unlock(mm); 2949bc8039eSYang Shi else 295d8ed45c5SMichel Lespinasse mmap_write_unlock(mm); 296897ab3e0SMike Rapoport userfaultfd_unmap_complete(mm, &uf); 297128557ffSMichel Lespinasse if (populate) 298128557ffSMichel Lespinasse mm_populate(oldbrk, newbrk - oldbrk); 299128557ffSMichel Lespinasse return brk; 300128557ffSMichel Lespinasse 3011da177e4SLinus Torvalds out: 302d8ed45c5SMichel Lespinasse mmap_write_unlock(mm); 303b7204006SAdrian Huang return origbrk; 3041da177e4SLinus Torvalds } 3051da177e4SLinus Torvalds 306315cc066SMichel Lespinasse static inline unsigned long vma_compute_gap(struct vm_area_struct *vma) 307d3737187SMichel Lespinasse { 308315cc066SMichel Lespinasse unsigned long gap, prev_end; 3091be7107fSHugh Dickins 3101be7107fSHugh Dickins /* 3111be7107fSHugh Dickins * Note: in the rare case of a VM_GROWSDOWN above a VM_GROWSUP, we 3121be7107fSHugh Dickins * allow two stack_guard_gaps between them here, and when choosing 3131be7107fSHugh Dickins * an unmapped area; whereas when expanding we only require one. 3141be7107fSHugh Dickins * That's a little inconsistent, but keeps the code here simpler. 3151be7107fSHugh Dickins */ 316315cc066SMichel Lespinasse gap = vm_start_gap(vma); 3171be7107fSHugh Dickins if (vma->vm_prev) { 3181be7107fSHugh Dickins prev_end = vm_end_gap(vma->vm_prev); 319315cc066SMichel Lespinasse if (gap > prev_end) 320315cc066SMichel Lespinasse gap -= prev_end; 3211be7107fSHugh Dickins else 322315cc066SMichel Lespinasse gap = 0; 3231be7107fSHugh Dickins } 324315cc066SMichel Lespinasse return gap; 325315cc066SMichel Lespinasse } 326315cc066SMichel Lespinasse 327315cc066SMichel Lespinasse #ifdef CONFIG_DEBUG_VM_RB 328315cc066SMichel Lespinasse static unsigned long vma_compute_subtree_gap(struct vm_area_struct *vma) 329315cc066SMichel Lespinasse { 330315cc066SMichel Lespinasse unsigned long max = vma_compute_gap(vma), subtree_gap; 331d3737187SMichel Lespinasse if (vma->vm_rb.rb_left) { 332d3737187SMichel Lespinasse subtree_gap = rb_entry(vma->vm_rb.rb_left, 333d3737187SMichel Lespinasse struct vm_area_struct, vm_rb)->rb_subtree_gap; 334d3737187SMichel Lespinasse if (subtree_gap > max) 335d3737187SMichel Lespinasse max = subtree_gap; 336d3737187SMichel Lespinasse } 337d3737187SMichel Lespinasse if (vma->vm_rb.rb_right) { 338d3737187SMichel Lespinasse subtree_gap = rb_entry(vma->vm_rb.rb_right, 339d3737187SMichel Lespinasse struct vm_area_struct, vm_rb)->rb_subtree_gap; 340d3737187SMichel Lespinasse if (subtree_gap > max) 341d3737187SMichel Lespinasse max = subtree_gap; 342d3737187SMichel Lespinasse } 343d3737187SMichel Lespinasse return max; 344d3737187SMichel Lespinasse } 345d3737187SMichel Lespinasse 346acf128d0SAndrea Arcangeli static int browse_rb(struct mm_struct *mm) 3471da177e4SLinus Torvalds { 348acf128d0SAndrea Arcangeli struct rb_root *root = &mm->mm_rb; 3495a0768f6SMichel Lespinasse int i = 0, j, bug = 0; 3501da177e4SLinus Torvalds struct rb_node *nd, *pn = NULL; 3511da177e4SLinus Torvalds unsigned long prev = 0, pend = 0; 3521da177e4SLinus Torvalds 3531da177e4SLinus Torvalds for (nd = rb_first(root); nd; nd = rb_next(nd)) { 3541da177e4SLinus Torvalds struct vm_area_struct *vma; 3551da177e4SLinus Torvalds vma = rb_entry(nd, struct vm_area_struct, vm_rb); 3565a0768f6SMichel Lespinasse if (vma->vm_start < prev) { 357ff26f70fSAndrew Morton pr_emerg("vm_start %lx < prev %lx\n", 358ff26f70fSAndrew Morton vma->vm_start, prev); 3595a0768f6SMichel Lespinasse bug = 1; 3605a0768f6SMichel Lespinasse } 3615a0768f6SMichel Lespinasse if (vma->vm_start < pend) { 362ff26f70fSAndrew Morton pr_emerg("vm_start %lx < pend %lx\n", 363ff26f70fSAndrew Morton vma->vm_start, pend); 3645a0768f6SMichel Lespinasse bug = 1; 3655a0768f6SMichel Lespinasse } 3665a0768f6SMichel Lespinasse if (vma->vm_start > vma->vm_end) { 367ff26f70fSAndrew Morton pr_emerg("vm_start %lx > vm_end %lx\n", 368ff26f70fSAndrew Morton vma->vm_start, vma->vm_end); 3695a0768f6SMichel Lespinasse bug = 1; 3705a0768f6SMichel Lespinasse } 371acf128d0SAndrea Arcangeli spin_lock(&mm->page_table_lock); 3725a0768f6SMichel Lespinasse if (vma->rb_subtree_gap != vma_compute_subtree_gap(vma)) { 3738542bdfcSSasha Levin pr_emerg("free gap %lx, correct %lx\n", 3745a0768f6SMichel Lespinasse vma->rb_subtree_gap, 3755a0768f6SMichel Lespinasse vma_compute_subtree_gap(vma)); 3765a0768f6SMichel Lespinasse bug = 1; 3775a0768f6SMichel Lespinasse } 378acf128d0SAndrea Arcangeli spin_unlock(&mm->page_table_lock); 3791da177e4SLinus Torvalds i++; 3801da177e4SLinus Torvalds pn = nd; 381d1af65d1SDavid Miller prev = vma->vm_start; 382d1af65d1SDavid Miller pend = vma->vm_end; 3831da177e4SLinus Torvalds } 3841da177e4SLinus Torvalds j = 0; 3855a0768f6SMichel Lespinasse for (nd = pn; nd; nd = rb_prev(nd)) 3861da177e4SLinus Torvalds j++; 3875a0768f6SMichel Lespinasse if (i != j) { 3888542bdfcSSasha Levin pr_emerg("backwards %d, forwards %d\n", j, i); 3895a0768f6SMichel Lespinasse bug = 1; 3901da177e4SLinus Torvalds } 3915a0768f6SMichel Lespinasse return bug ? -1 : i; 3921da177e4SLinus Torvalds } 3931da177e4SLinus Torvalds 394d3737187SMichel Lespinasse static void validate_mm_rb(struct rb_root *root, struct vm_area_struct *ignore) 395d3737187SMichel Lespinasse { 396d3737187SMichel Lespinasse struct rb_node *nd; 397d3737187SMichel Lespinasse 398d3737187SMichel Lespinasse for (nd = rb_first(root); nd; nd = rb_next(nd)) { 399d3737187SMichel Lespinasse struct vm_area_struct *vma; 400d3737187SMichel Lespinasse vma = rb_entry(nd, struct vm_area_struct, vm_rb); 40196dad67fSSasha Levin VM_BUG_ON_VMA(vma != ignore && 40296dad67fSSasha Levin vma->rb_subtree_gap != vma_compute_subtree_gap(vma), 40396dad67fSSasha Levin vma); 404d3737187SMichel Lespinasse } 4051da177e4SLinus Torvalds } 4061da177e4SLinus Torvalds 407eafd4dc4SRashika Kheria static void validate_mm(struct mm_struct *mm) 4081da177e4SLinus Torvalds { 4091da177e4SLinus Torvalds int bug = 0; 4101da177e4SLinus Torvalds int i = 0; 4115a0768f6SMichel Lespinasse unsigned long highest_address = 0; 412ed8ea815SMichel Lespinasse struct vm_area_struct *vma = mm->mmap; 413ff26f70fSAndrew Morton 414ed8ea815SMichel Lespinasse while (vma) { 41512352d3cSKonstantin Khlebnikov struct anon_vma *anon_vma = vma->anon_vma; 416ed8ea815SMichel Lespinasse struct anon_vma_chain *avc; 417ff26f70fSAndrew Morton 41812352d3cSKonstantin Khlebnikov if (anon_vma) { 41912352d3cSKonstantin Khlebnikov anon_vma_lock_read(anon_vma); 420ed8ea815SMichel Lespinasse list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) 421ed8ea815SMichel Lespinasse anon_vma_interval_tree_verify(avc); 42212352d3cSKonstantin Khlebnikov anon_vma_unlock_read(anon_vma); 42312352d3cSKonstantin Khlebnikov } 42412352d3cSKonstantin Khlebnikov 4251be7107fSHugh Dickins highest_address = vm_end_gap(vma); 426ed8ea815SMichel Lespinasse vma = vma->vm_next; 4271da177e4SLinus Torvalds i++; 4281da177e4SLinus Torvalds } 4295a0768f6SMichel Lespinasse if (i != mm->map_count) { 4308542bdfcSSasha Levin pr_emerg("map_count %d vm_next %d\n", mm->map_count, i); 4315a0768f6SMichel Lespinasse bug = 1; 4325a0768f6SMichel Lespinasse } 4335a0768f6SMichel Lespinasse if (highest_address != mm->highest_vm_end) { 4348542bdfcSSasha Levin pr_emerg("mm->highest_vm_end %lx, found %lx\n", 4355a0768f6SMichel Lespinasse mm->highest_vm_end, highest_address); 4365a0768f6SMichel Lespinasse bug = 1; 4375a0768f6SMichel Lespinasse } 438acf128d0SAndrea Arcangeli i = browse_rb(mm); 4395a0768f6SMichel Lespinasse if (i != mm->map_count) { 440ff26f70fSAndrew Morton if (i != -1) 4418542bdfcSSasha Levin pr_emerg("map_count %d rb %d\n", mm->map_count, i); 4425a0768f6SMichel Lespinasse bug = 1; 4435a0768f6SMichel Lespinasse } 44496dad67fSSasha Levin VM_BUG_ON_MM(bug, mm); 4451da177e4SLinus Torvalds } 4461da177e4SLinus Torvalds #else 447d3737187SMichel Lespinasse #define validate_mm_rb(root, ignore) do { } while (0) 4481da177e4SLinus Torvalds #define validate_mm(mm) do { } while (0) 4491da177e4SLinus Torvalds #endif 4501da177e4SLinus Torvalds 451315cc066SMichel Lespinasse RB_DECLARE_CALLBACKS_MAX(static, vma_gap_callbacks, 452315cc066SMichel Lespinasse struct vm_area_struct, vm_rb, 453315cc066SMichel Lespinasse unsigned long, rb_subtree_gap, vma_compute_gap) 454d3737187SMichel Lespinasse 455d3737187SMichel Lespinasse /* 456d3737187SMichel Lespinasse * Update augmented rbtree rb_subtree_gap values after vma->vm_start or 457d3737187SMichel Lespinasse * vma->vm_prev->vm_end values changed, without modifying the vma's position 458d3737187SMichel Lespinasse * in the rbtree. 459d3737187SMichel Lespinasse */ 460d3737187SMichel Lespinasse static void vma_gap_update(struct vm_area_struct *vma) 461d3737187SMichel Lespinasse { 462d3737187SMichel Lespinasse /* 463315cc066SMichel Lespinasse * As it turns out, RB_DECLARE_CALLBACKS_MAX() already created 464315cc066SMichel Lespinasse * a callback function that does exactly what we want. 465d3737187SMichel Lespinasse */ 466d3737187SMichel Lespinasse vma_gap_callbacks_propagate(&vma->vm_rb, NULL); 467d3737187SMichel Lespinasse } 468d3737187SMichel Lespinasse 469d3737187SMichel Lespinasse static inline void vma_rb_insert(struct vm_area_struct *vma, 470d3737187SMichel Lespinasse struct rb_root *root) 471d3737187SMichel Lespinasse { 472d3737187SMichel Lespinasse /* All rb_subtree_gap values must be consistent prior to insertion */ 473d3737187SMichel Lespinasse validate_mm_rb(root, NULL); 474d3737187SMichel Lespinasse 475d3737187SMichel Lespinasse rb_insert_augmented(&vma->vm_rb, root, &vma_gap_callbacks); 476d3737187SMichel Lespinasse } 477d3737187SMichel Lespinasse 4788f26e0b1SAndrea Arcangeli static void __vma_rb_erase(struct vm_area_struct *vma, struct rb_root *root) 4798f26e0b1SAndrea Arcangeli { 4808f26e0b1SAndrea Arcangeli /* 4818f26e0b1SAndrea Arcangeli * Note rb_erase_augmented is a fairly large inline function, 4828f26e0b1SAndrea Arcangeli * so make sure we instantiate it only once with our desired 4838f26e0b1SAndrea Arcangeli * augmented rbtree callbacks. 4848f26e0b1SAndrea Arcangeli */ 4858f26e0b1SAndrea Arcangeli rb_erase_augmented(&vma->vm_rb, root, &vma_gap_callbacks); 4868f26e0b1SAndrea Arcangeli } 4878f26e0b1SAndrea Arcangeli 4888f26e0b1SAndrea Arcangeli static __always_inline void vma_rb_erase_ignore(struct vm_area_struct *vma, 4898f26e0b1SAndrea Arcangeli struct rb_root *root, 4908f26e0b1SAndrea Arcangeli struct vm_area_struct *ignore) 4918f26e0b1SAndrea Arcangeli { 4928f26e0b1SAndrea Arcangeli /* 4938f26e0b1SAndrea Arcangeli * All rb_subtree_gap values must be consistent prior to erase, 4944d1e7243SWei Yang * with the possible exception of 4954d1e7243SWei Yang * 4964d1e7243SWei Yang * a. the "next" vma being erased if next->vm_start was reduced in 4974d1e7243SWei Yang * __vma_adjust() -> __vma_unlink() 4984d1e7243SWei Yang * b. the vma being erased in detach_vmas_to_be_unmapped() -> 4994d1e7243SWei Yang * vma_rb_erase() 5008f26e0b1SAndrea Arcangeli */ 5018f26e0b1SAndrea Arcangeli validate_mm_rb(root, ignore); 5028f26e0b1SAndrea Arcangeli 5038f26e0b1SAndrea Arcangeli __vma_rb_erase(vma, root); 5048f26e0b1SAndrea Arcangeli } 5058f26e0b1SAndrea Arcangeli 5068f26e0b1SAndrea Arcangeli static __always_inline void vma_rb_erase(struct vm_area_struct *vma, 5078f26e0b1SAndrea Arcangeli struct rb_root *root) 508d3737187SMichel Lespinasse { 5094d1e7243SWei Yang vma_rb_erase_ignore(vma, root, vma); 510d3737187SMichel Lespinasse } 511d3737187SMichel Lespinasse 512bf181b9fSMichel Lespinasse /* 513bf181b9fSMichel Lespinasse * vma has some anon_vma assigned, and is already inserted on that 514bf181b9fSMichel Lespinasse * anon_vma's interval trees. 515bf181b9fSMichel Lespinasse * 516bf181b9fSMichel Lespinasse * Before updating the vma's vm_start / vm_end / vm_pgoff fields, the 517bf181b9fSMichel Lespinasse * vma must be removed from the anon_vma's interval trees using 518bf181b9fSMichel Lespinasse * anon_vma_interval_tree_pre_update_vma(). 519bf181b9fSMichel Lespinasse * 520bf181b9fSMichel Lespinasse * After the update, the vma will be reinserted using 521bf181b9fSMichel Lespinasse * anon_vma_interval_tree_post_update_vma(). 522bf181b9fSMichel Lespinasse * 523c1e8d7c6SMichel Lespinasse * The entire update must be protected by exclusive mmap_lock and by 524bf181b9fSMichel Lespinasse * the root anon_vma's mutex. 525bf181b9fSMichel Lespinasse */ 526bf181b9fSMichel Lespinasse static inline void 527bf181b9fSMichel Lespinasse anon_vma_interval_tree_pre_update_vma(struct vm_area_struct *vma) 528bf181b9fSMichel Lespinasse { 529bf181b9fSMichel Lespinasse struct anon_vma_chain *avc; 530bf181b9fSMichel Lespinasse 531bf181b9fSMichel Lespinasse list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) 532bf181b9fSMichel Lespinasse anon_vma_interval_tree_remove(avc, &avc->anon_vma->rb_root); 533bf181b9fSMichel Lespinasse } 534bf181b9fSMichel Lespinasse 535bf181b9fSMichel Lespinasse static inline void 536bf181b9fSMichel Lespinasse anon_vma_interval_tree_post_update_vma(struct vm_area_struct *vma) 537bf181b9fSMichel Lespinasse { 538bf181b9fSMichel Lespinasse struct anon_vma_chain *avc; 539bf181b9fSMichel Lespinasse 540bf181b9fSMichel Lespinasse list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) 541bf181b9fSMichel Lespinasse anon_vma_interval_tree_insert(avc, &avc->anon_vma->rb_root); 542bf181b9fSMichel Lespinasse } 543bf181b9fSMichel Lespinasse 5446597d783SHugh Dickins static int find_vma_links(struct mm_struct *mm, unsigned long addr, 5456597d783SHugh Dickins unsigned long end, struct vm_area_struct **pprev, 5466597d783SHugh Dickins struct rb_node ***rb_link, struct rb_node **rb_parent) 5471da177e4SLinus Torvalds { 5481da177e4SLinus Torvalds struct rb_node **__rb_link, *__rb_parent, *rb_prev; 5491da177e4SLinus Torvalds 5505b78ed24SLuigi Rizzo mmap_assert_locked(mm); 5511da177e4SLinus Torvalds __rb_link = &mm->mm_rb.rb_node; 5521da177e4SLinus Torvalds rb_prev = __rb_parent = NULL; 5531da177e4SLinus Torvalds 5541da177e4SLinus Torvalds while (*__rb_link) { 5551da177e4SLinus Torvalds struct vm_area_struct *vma_tmp; 5561da177e4SLinus Torvalds 5571da177e4SLinus Torvalds __rb_parent = *__rb_link; 5581da177e4SLinus Torvalds vma_tmp = rb_entry(__rb_parent, struct vm_area_struct, vm_rb); 5591da177e4SLinus Torvalds 5601da177e4SLinus Torvalds if (vma_tmp->vm_end > addr) { 5616597d783SHugh Dickins /* Fail if an existing vma overlaps the area */ 5626597d783SHugh Dickins if (vma_tmp->vm_start < end) 5636597d783SHugh Dickins return -ENOMEM; 5641da177e4SLinus Torvalds __rb_link = &__rb_parent->rb_left; 5651da177e4SLinus Torvalds } else { 5661da177e4SLinus Torvalds rb_prev = __rb_parent; 5671da177e4SLinus Torvalds __rb_link = &__rb_parent->rb_right; 5681da177e4SLinus Torvalds } 5691da177e4SLinus Torvalds } 5701da177e4SLinus Torvalds 5711da177e4SLinus Torvalds *pprev = NULL; 5721da177e4SLinus Torvalds if (rb_prev) 5731da177e4SLinus Torvalds *pprev = rb_entry(rb_prev, struct vm_area_struct, vm_rb); 5741da177e4SLinus Torvalds *rb_link = __rb_link; 5751da177e4SLinus Torvalds *rb_parent = __rb_parent; 5766597d783SHugh Dickins return 0; 5771da177e4SLinus Torvalds } 5781da177e4SLinus Torvalds 5793903b55aSLiam R. Howlett /* 5803903b55aSLiam R. Howlett * vma_next() - Get the next VMA. 5813903b55aSLiam R. Howlett * @mm: The mm_struct. 5823903b55aSLiam R. Howlett * @vma: The current vma. 5833903b55aSLiam R. Howlett * 5843903b55aSLiam R. Howlett * If @vma is NULL, return the first vma in the mm. 5853903b55aSLiam R. Howlett * 5863903b55aSLiam R. Howlett * Returns: The next VMA after @vma. 5873903b55aSLiam R. Howlett */ 5883903b55aSLiam R. Howlett static inline struct vm_area_struct *vma_next(struct mm_struct *mm, 5893903b55aSLiam R. Howlett struct vm_area_struct *vma) 5903903b55aSLiam R. Howlett { 5913903b55aSLiam R. Howlett if (!vma) 5923903b55aSLiam R. Howlett return mm->mmap; 5933903b55aSLiam R. Howlett 5943903b55aSLiam R. Howlett return vma->vm_next; 5953903b55aSLiam R. Howlett } 596fb8090b6SLiam R. Howlett 597fb8090b6SLiam R. Howlett /* 598fb8090b6SLiam R. Howlett * munmap_vma_range() - munmap VMAs that overlap a range. 599fb8090b6SLiam R. Howlett * @mm: The mm struct 600fb8090b6SLiam R. Howlett * @start: The start of the range. 601fb8090b6SLiam R. Howlett * @len: The length of the range. 602fb8090b6SLiam R. Howlett * @pprev: pointer to the pointer that will be set to previous vm_area_struct 603fb8090b6SLiam R. Howlett * @rb_link: the rb_node 604fb8090b6SLiam R. Howlett * @rb_parent: the parent rb_node 605fb8090b6SLiam R. Howlett * 606fb8090b6SLiam R. Howlett * Find all the vm_area_struct that overlap from @start to 607fb8090b6SLiam R. Howlett * @end and munmap them. Set @pprev to the previous vm_area_struct. 608fb8090b6SLiam R. Howlett * 609fb8090b6SLiam R. Howlett * Returns: -ENOMEM on munmap failure or 0 on success. 610fb8090b6SLiam R. Howlett */ 611fb8090b6SLiam R. Howlett static inline int 612fb8090b6SLiam R. Howlett munmap_vma_range(struct mm_struct *mm, unsigned long start, unsigned long len, 613fb8090b6SLiam R. Howlett struct vm_area_struct **pprev, struct rb_node ***link, 614fb8090b6SLiam R. Howlett struct rb_node **parent, struct list_head *uf) 615fb8090b6SLiam R. Howlett { 616fb8090b6SLiam R. Howlett 617fb8090b6SLiam R. Howlett while (find_vma_links(mm, start, start + len, pprev, link, parent)) 618fb8090b6SLiam R. Howlett if (do_munmap(mm, start, len, uf)) 619fb8090b6SLiam R. Howlett return -ENOMEM; 620fb8090b6SLiam R. Howlett 621fb8090b6SLiam R. Howlett return 0; 622fb8090b6SLiam R. Howlett } 623e8420a8eSCyril Hrubis static unsigned long count_vma_pages_range(struct mm_struct *mm, 624e8420a8eSCyril Hrubis unsigned long addr, unsigned long end) 625e8420a8eSCyril Hrubis { 626e8420a8eSCyril Hrubis unsigned long nr_pages = 0; 627e8420a8eSCyril Hrubis struct vm_area_struct *vma; 628e8420a8eSCyril Hrubis 629f0953a1bSIngo Molnar /* Find first overlapping mapping */ 630e8420a8eSCyril Hrubis vma = find_vma_intersection(mm, addr, end); 631e8420a8eSCyril Hrubis if (!vma) 632e8420a8eSCyril Hrubis return 0; 633e8420a8eSCyril Hrubis 634e8420a8eSCyril Hrubis nr_pages = (min(end, vma->vm_end) - 635e8420a8eSCyril Hrubis max(addr, vma->vm_start)) >> PAGE_SHIFT; 636e8420a8eSCyril Hrubis 637e8420a8eSCyril Hrubis /* Iterate over the rest of the overlaps */ 638e8420a8eSCyril Hrubis for (vma = vma->vm_next; vma; vma = vma->vm_next) { 639e8420a8eSCyril Hrubis unsigned long overlap_len; 640e8420a8eSCyril Hrubis 641e8420a8eSCyril Hrubis if (vma->vm_start > end) 642e8420a8eSCyril Hrubis break; 643e8420a8eSCyril Hrubis 644e8420a8eSCyril Hrubis overlap_len = min(end, vma->vm_end) - vma->vm_start; 645e8420a8eSCyril Hrubis nr_pages += overlap_len >> PAGE_SHIFT; 646e8420a8eSCyril Hrubis } 647e8420a8eSCyril Hrubis 648e8420a8eSCyril Hrubis return nr_pages; 649e8420a8eSCyril Hrubis } 650e8420a8eSCyril Hrubis 6511da177e4SLinus Torvalds void __vma_link_rb(struct mm_struct *mm, struct vm_area_struct *vma, 6521da177e4SLinus Torvalds struct rb_node **rb_link, struct rb_node *rb_parent) 6531da177e4SLinus Torvalds { 654d3737187SMichel Lespinasse /* Update tracking information for the gap following the new vma. */ 655d3737187SMichel Lespinasse if (vma->vm_next) 656d3737187SMichel Lespinasse vma_gap_update(vma->vm_next); 657d3737187SMichel Lespinasse else 6581be7107fSHugh Dickins mm->highest_vm_end = vm_end_gap(vma); 659d3737187SMichel Lespinasse 660d3737187SMichel Lespinasse /* 661d3737187SMichel Lespinasse * vma->vm_prev wasn't known when we followed the rbtree to find the 662d3737187SMichel Lespinasse * correct insertion point for that vma. As a result, we could not 663d3737187SMichel Lespinasse * update the vma vm_rb parents rb_subtree_gap values on the way down. 664d3737187SMichel Lespinasse * So, we first insert the vma with a zero rb_subtree_gap value 665d3737187SMichel Lespinasse * (to be consistent with what we did on the way down), and then 666d3737187SMichel Lespinasse * immediately update the gap to the correct value. Finally we 667d3737187SMichel Lespinasse * rebalance the rbtree after all augmented values have been set. 668d3737187SMichel Lespinasse */ 6691da177e4SLinus Torvalds rb_link_node(&vma->vm_rb, rb_parent, rb_link); 670d3737187SMichel Lespinasse vma->rb_subtree_gap = 0; 671d3737187SMichel Lespinasse vma_gap_update(vma); 672d3737187SMichel Lespinasse vma_rb_insert(vma, &mm->mm_rb); 6731da177e4SLinus Torvalds } 6741da177e4SLinus Torvalds 675cb8f488cSDenys Vlasenko static void __vma_link_file(struct vm_area_struct *vma) 6761da177e4SLinus Torvalds { 6771da177e4SLinus Torvalds struct file *file; 6781da177e4SLinus Torvalds 6791da177e4SLinus Torvalds file = vma->vm_file; 6801da177e4SLinus Torvalds if (file) { 6811da177e4SLinus Torvalds struct address_space *mapping = file->f_mapping; 6821da177e4SLinus Torvalds 6831da177e4SLinus Torvalds if (vma->vm_flags & VM_SHARED) 684cf508b58SMiaohe Lin mapping_allow_writable(mapping); 6851da177e4SLinus Torvalds 6861da177e4SLinus Torvalds flush_dcache_mmap_lock(mapping); 6876b2dbba8SMichel Lespinasse vma_interval_tree_insert(vma, &mapping->i_mmap); 6881da177e4SLinus Torvalds flush_dcache_mmap_unlock(mapping); 6891da177e4SLinus Torvalds } 6901da177e4SLinus Torvalds } 6911da177e4SLinus Torvalds 6921da177e4SLinus Torvalds static void 6931da177e4SLinus Torvalds __vma_link(struct mm_struct *mm, struct vm_area_struct *vma, 6941da177e4SLinus Torvalds struct vm_area_struct *prev, struct rb_node **rb_link, 6951da177e4SLinus Torvalds struct rb_node *rb_parent) 6961da177e4SLinus Torvalds { 697aba6dfb7SWei Yang __vma_link_list(mm, vma, prev); 6981da177e4SLinus Torvalds __vma_link_rb(mm, vma, rb_link, rb_parent); 6991da177e4SLinus Torvalds } 7001da177e4SLinus Torvalds 7011da177e4SLinus Torvalds static void vma_link(struct mm_struct *mm, struct vm_area_struct *vma, 7021da177e4SLinus Torvalds struct vm_area_struct *prev, struct rb_node **rb_link, 7031da177e4SLinus Torvalds struct rb_node *rb_parent) 7041da177e4SLinus Torvalds { 7051da177e4SLinus Torvalds struct address_space *mapping = NULL; 7061da177e4SLinus Torvalds 70764ac4940SHuang Shijie if (vma->vm_file) { 7081da177e4SLinus Torvalds mapping = vma->vm_file->f_mapping; 70983cde9e8SDavidlohr Bueso i_mmap_lock_write(mapping); 71064ac4940SHuang Shijie } 7111da177e4SLinus Torvalds 7121da177e4SLinus Torvalds __vma_link(mm, vma, prev, rb_link, rb_parent); 7131da177e4SLinus Torvalds __vma_link_file(vma); 7141da177e4SLinus Torvalds 7151da177e4SLinus Torvalds if (mapping) 71683cde9e8SDavidlohr Bueso i_mmap_unlock_write(mapping); 7171da177e4SLinus Torvalds 7181da177e4SLinus Torvalds mm->map_count++; 7191da177e4SLinus Torvalds validate_mm(mm); 7201da177e4SLinus Torvalds } 7211da177e4SLinus Torvalds 7221da177e4SLinus Torvalds /* 72388f6b4c3SKautuk Consul * Helper for vma_adjust() in the split_vma insert case: insert a vma into the 7246b2dbba8SMichel Lespinasse * mm's list and rbtree. It has already been inserted into the interval tree. 7251da177e4SLinus Torvalds */ 72648aae425SZhenwenXu static void __insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma) 7271da177e4SLinus Torvalds { 7286597d783SHugh Dickins struct vm_area_struct *prev; 7291da177e4SLinus Torvalds struct rb_node **rb_link, *rb_parent; 7301da177e4SLinus Torvalds 7316597d783SHugh Dickins if (find_vma_links(mm, vma->vm_start, vma->vm_end, 7326597d783SHugh Dickins &prev, &rb_link, &rb_parent)) 7336597d783SHugh Dickins BUG(); 7341da177e4SLinus Torvalds __vma_link(mm, vma, prev, rb_link, rb_parent); 7351da177e4SLinus Torvalds mm->map_count++; 7361da177e4SLinus Torvalds } 7371da177e4SLinus Torvalds 7387c61f917SWei Yang static __always_inline void __vma_unlink(struct mm_struct *mm, 739e86f15eeSAndrea Arcangeli struct vm_area_struct *vma, 7408f26e0b1SAndrea Arcangeli struct vm_area_struct *ignore) 7411da177e4SLinus Torvalds { 7428f26e0b1SAndrea Arcangeli vma_rb_erase_ignore(vma, &mm->mm_rb, ignore); 7431b9fc5b2SWei Yang __vma_unlink_list(mm, vma); 744615d6e87SDavidlohr Bueso /* Kill the cache */ 745615d6e87SDavidlohr Bueso vmacache_invalidate(mm); 7461da177e4SLinus Torvalds } 7471da177e4SLinus Torvalds 7481da177e4SLinus Torvalds /* 7491da177e4SLinus Torvalds * We cannot adjust vm_start, vm_end, vm_pgoff fields of a vma that 7501da177e4SLinus Torvalds * is already present in an i_mmap tree without adjusting the tree. 7511da177e4SLinus Torvalds * The following helper function should be used when such adjustments 7521da177e4SLinus Torvalds * are necessary. The "insert" vma (if any) is to be inserted 7531da177e4SLinus Torvalds * before we drop the necessary locks. 7541da177e4SLinus Torvalds */ 755e86f15eeSAndrea Arcangeli int __vma_adjust(struct vm_area_struct *vma, unsigned long start, 756e86f15eeSAndrea Arcangeli unsigned long end, pgoff_t pgoff, struct vm_area_struct *insert, 757e86f15eeSAndrea Arcangeli struct vm_area_struct *expand) 7581da177e4SLinus Torvalds { 7591da177e4SLinus Torvalds struct mm_struct *mm = vma->vm_mm; 760e86f15eeSAndrea Arcangeli struct vm_area_struct *next = vma->vm_next, *orig_vma = vma; 7611da177e4SLinus Torvalds struct address_space *mapping = NULL; 762f808c13fSDavidlohr Bueso struct rb_root_cached *root = NULL; 763012f1800SRik van Riel struct anon_vma *anon_vma = NULL; 7641da177e4SLinus Torvalds struct file *file = vma->vm_file; 765d3737187SMichel Lespinasse bool start_changed = false, end_changed = false; 7661da177e4SLinus Torvalds long adjust_next = 0; 7671da177e4SLinus Torvalds int remove_next = 0; 7681da177e4SLinus Torvalds 7691da177e4SLinus Torvalds if (next && !insert) { 770734537c9SKirill A. Shutemov struct vm_area_struct *exporter = NULL, *importer = NULL; 771287d97acSLinus Torvalds 7721da177e4SLinus Torvalds if (end >= next->vm_end) { 7731da177e4SLinus Torvalds /* 7741da177e4SLinus Torvalds * vma expands, overlapping all the next, and 7751da177e4SLinus Torvalds * perhaps the one after too (mprotect case 6). 77686d12e47SAndrea Arcangeli * The only other cases that gets here are 777e86f15eeSAndrea Arcangeli * case 1, case 7 and case 8. 778e86f15eeSAndrea Arcangeli */ 779e86f15eeSAndrea Arcangeli if (next == expand) { 780e86f15eeSAndrea Arcangeli /* 781e86f15eeSAndrea Arcangeli * The only case where we don't expand "vma" 782e86f15eeSAndrea Arcangeli * and we expand "next" instead is case 8. 783e86f15eeSAndrea Arcangeli */ 784e86f15eeSAndrea Arcangeli VM_WARN_ON(end != next->vm_end); 785e86f15eeSAndrea Arcangeli /* 786e86f15eeSAndrea Arcangeli * remove_next == 3 means we're 787e86f15eeSAndrea Arcangeli * removing "vma" and that to do so we 788e86f15eeSAndrea Arcangeli * swapped "vma" and "next". 789e86f15eeSAndrea Arcangeli */ 790e86f15eeSAndrea Arcangeli remove_next = 3; 791e86f15eeSAndrea Arcangeli VM_WARN_ON(file != next->vm_file); 792e86f15eeSAndrea Arcangeli swap(vma, next); 793e86f15eeSAndrea Arcangeli } else { 794e86f15eeSAndrea Arcangeli VM_WARN_ON(expand != vma); 795e86f15eeSAndrea Arcangeli /* 796e86f15eeSAndrea Arcangeli * case 1, 6, 7, remove_next == 2 is case 6, 797e86f15eeSAndrea Arcangeli * remove_next == 1 is case 1 or 7. 7981da177e4SLinus Torvalds */ 799734537c9SKirill A. Shutemov remove_next = 1 + (end > next->vm_end); 800e86f15eeSAndrea Arcangeli VM_WARN_ON(remove_next == 2 && 801e86f15eeSAndrea Arcangeli end != next->vm_next->vm_end); 802e86f15eeSAndrea Arcangeli /* trim end to next, for case 6 first pass */ 8031da177e4SLinus Torvalds end = next->vm_end; 804e86f15eeSAndrea Arcangeli } 805e86f15eeSAndrea Arcangeli 806287d97acSLinus Torvalds exporter = next; 8071da177e4SLinus Torvalds importer = vma; 808734537c9SKirill A. Shutemov 809734537c9SKirill A. Shutemov /* 810734537c9SKirill A. Shutemov * If next doesn't have anon_vma, import from vma after 811734537c9SKirill A. Shutemov * next, if the vma overlaps with it. 812734537c9SKirill A. Shutemov */ 81397a42cd4SAndrea Arcangeli if (remove_next == 2 && !next->anon_vma) 814734537c9SKirill A. Shutemov exporter = next->vm_next; 815734537c9SKirill A. Shutemov 8161da177e4SLinus Torvalds } else if (end > next->vm_start) { 8171da177e4SLinus Torvalds /* 8181da177e4SLinus Torvalds * vma expands, overlapping part of the next: 8191da177e4SLinus Torvalds * mprotect case 5 shifting the boundary up. 8201da177e4SLinus Torvalds */ 821f9d86a60SWei Yang adjust_next = (end - next->vm_start); 822287d97acSLinus Torvalds exporter = next; 8231da177e4SLinus Torvalds importer = vma; 824e86f15eeSAndrea Arcangeli VM_WARN_ON(expand != importer); 8251da177e4SLinus Torvalds } else if (end < vma->vm_end) { 8261da177e4SLinus Torvalds /* 8271da177e4SLinus Torvalds * vma shrinks, and !insert tells it's not 8281da177e4SLinus Torvalds * split_vma inserting another: so it must be 8291da177e4SLinus Torvalds * mprotect case 4 shifting the boundary down. 8301da177e4SLinus Torvalds */ 831f9d86a60SWei Yang adjust_next = -(vma->vm_end - end); 832287d97acSLinus Torvalds exporter = vma; 8331da177e4SLinus Torvalds importer = next; 834e86f15eeSAndrea Arcangeli VM_WARN_ON(expand != importer); 8351da177e4SLinus Torvalds } 8361da177e4SLinus Torvalds 8375beb4930SRik van Riel /* 8385beb4930SRik van Riel * Easily overlooked: when mprotect shifts the boundary, 8395beb4930SRik van Riel * make sure the expanding vma has anon_vma set if the 8405beb4930SRik van Riel * shrinking vma had, to cover any anon pages imported. 8415beb4930SRik van Riel */ 842287d97acSLinus Torvalds if (exporter && exporter->anon_vma && !importer->anon_vma) { 843c4ea95d7SDaniel Forrest int error; 844c4ea95d7SDaniel Forrest 845287d97acSLinus Torvalds importer->anon_vma = exporter->anon_vma; 846b800c91aSKonstantin Khlebnikov error = anon_vma_clone(importer, exporter); 8473fe89b3eSLeon Yu if (error) 848b800c91aSKonstantin Khlebnikov return error; 849b800c91aSKonstantin Khlebnikov } 8505beb4930SRik van Riel } 851734537c9SKirill A. Shutemov again: 852e86f15eeSAndrea Arcangeli vma_adjust_trans_huge(orig_vma, start, end, adjust_next); 85337f9f559SKirill A. Shutemov 8541da177e4SLinus Torvalds if (file) { 8551da177e4SLinus Torvalds mapping = file->f_mapping; 8561da177e4SLinus Torvalds root = &mapping->i_mmap; 857cbc91f71SSrikar Dronamraju uprobe_munmap(vma, vma->vm_start, vma->vm_end); 858682968e0SSrikar Dronamraju 859682968e0SSrikar Dronamraju if (adjust_next) 86027ba0644SKirill A. Shutemov uprobe_munmap(next, next->vm_start, next->vm_end); 861682968e0SSrikar Dronamraju 86283cde9e8SDavidlohr Bueso i_mmap_lock_write(mapping); 8631da177e4SLinus Torvalds if (insert) { 8641da177e4SLinus Torvalds /* 8656b2dbba8SMichel Lespinasse * Put into interval tree now, so instantiated pages 8661da177e4SLinus Torvalds * are visible to arm/parisc __flush_dcache_page 8671da177e4SLinus Torvalds * throughout; but we cannot insert into address 8681da177e4SLinus Torvalds * space until vma start or end is updated. 8691da177e4SLinus Torvalds */ 8701da177e4SLinus Torvalds __vma_link_file(insert); 8711da177e4SLinus Torvalds } 8721da177e4SLinus Torvalds } 8731da177e4SLinus Torvalds 874012f1800SRik van Riel anon_vma = vma->anon_vma; 875bf181b9fSMichel Lespinasse if (!anon_vma && adjust_next) 876bf181b9fSMichel Lespinasse anon_vma = next->anon_vma; 877bf181b9fSMichel Lespinasse if (anon_vma) { 878e86f15eeSAndrea Arcangeli VM_WARN_ON(adjust_next && next->anon_vma && 879e86f15eeSAndrea Arcangeli anon_vma != next->anon_vma); 8804fc3f1d6SIngo Molnar anon_vma_lock_write(anon_vma); 881bf181b9fSMichel Lespinasse anon_vma_interval_tree_pre_update_vma(vma); 882bf181b9fSMichel Lespinasse if (adjust_next) 883bf181b9fSMichel Lespinasse anon_vma_interval_tree_pre_update_vma(next); 884bf181b9fSMichel Lespinasse } 885012f1800SRik van Riel 8860fc48a6eSWei Yang if (file) { 8871da177e4SLinus Torvalds flush_dcache_mmap_lock(mapping); 8886b2dbba8SMichel Lespinasse vma_interval_tree_remove(vma, root); 8891da177e4SLinus Torvalds if (adjust_next) 8906b2dbba8SMichel Lespinasse vma_interval_tree_remove(next, root); 8911da177e4SLinus Torvalds } 8921da177e4SLinus Torvalds 893d3737187SMichel Lespinasse if (start != vma->vm_start) { 8941da177e4SLinus Torvalds vma->vm_start = start; 895d3737187SMichel Lespinasse start_changed = true; 896d3737187SMichel Lespinasse } 897d3737187SMichel Lespinasse if (end != vma->vm_end) { 8981da177e4SLinus Torvalds vma->vm_end = end; 899d3737187SMichel Lespinasse end_changed = true; 900d3737187SMichel Lespinasse } 9011da177e4SLinus Torvalds vma->vm_pgoff = pgoff; 9021da177e4SLinus Torvalds if (adjust_next) { 903f9d86a60SWei Yang next->vm_start += adjust_next; 904f9d86a60SWei Yang next->vm_pgoff += adjust_next >> PAGE_SHIFT; 9051da177e4SLinus Torvalds } 9061da177e4SLinus Torvalds 9070fc48a6eSWei Yang if (file) { 9081da177e4SLinus Torvalds if (adjust_next) 9096b2dbba8SMichel Lespinasse vma_interval_tree_insert(next, root); 9106b2dbba8SMichel Lespinasse vma_interval_tree_insert(vma, root); 9111da177e4SLinus Torvalds flush_dcache_mmap_unlock(mapping); 9121da177e4SLinus Torvalds } 9131da177e4SLinus Torvalds 9141da177e4SLinus Torvalds if (remove_next) { 9151da177e4SLinus Torvalds /* 9161da177e4SLinus Torvalds * vma_merge has merged next into vma, and needs 9171da177e4SLinus Torvalds * us to remove next before dropping the locks. 9181da177e4SLinus Torvalds */ 919e86f15eeSAndrea Arcangeli if (remove_next != 3) 9207c61f917SWei Yang __vma_unlink(mm, next, next); 921e86f15eeSAndrea Arcangeli else 9228f26e0b1SAndrea Arcangeli /* 9238f26e0b1SAndrea Arcangeli * vma is not before next if they've been 9248f26e0b1SAndrea Arcangeli * swapped. 9258f26e0b1SAndrea Arcangeli * 9268f26e0b1SAndrea Arcangeli * pre-swap() next->vm_start was reduced so 9278f26e0b1SAndrea Arcangeli * tell validate_mm_rb to ignore pre-swap() 9288f26e0b1SAndrea Arcangeli * "next" (which is stored in post-swap() 9298f26e0b1SAndrea Arcangeli * "vma"). 9308f26e0b1SAndrea Arcangeli */ 9317c61f917SWei Yang __vma_unlink(mm, next, vma); 9321da177e4SLinus Torvalds if (file) 9331da177e4SLinus Torvalds __remove_shared_vm_struct(next, file, mapping); 9341da177e4SLinus Torvalds } else if (insert) { 9351da177e4SLinus Torvalds /* 9361da177e4SLinus Torvalds * split_vma has split insert from vma, and needs 9371da177e4SLinus Torvalds * us to insert it before dropping the locks 9381da177e4SLinus Torvalds * (it may either follow vma or precede it). 9391da177e4SLinus Torvalds */ 9401da177e4SLinus Torvalds __insert_vm_struct(mm, insert); 941d3737187SMichel Lespinasse } else { 942d3737187SMichel Lespinasse if (start_changed) 943d3737187SMichel Lespinasse vma_gap_update(vma); 944d3737187SMichel Lespinasse if (end_changed) { 945d3737187SMichel Lespinasse if (!next) 9461be7107fSHugh Dickins mm->highest_vm_end = vm_end_gap(vma); 947d3737187SMichel Lespinasse else if (!adjust_next) 948d3737187SMichel Lespinasse vma_gap_update(next); 949d3737187SMichel Lespinasse } 9501da177e4SLinus Torvalds } 9511da177e4SLinus Torvalds 952bf181b9fSMichel Lespinasse if (anon_vma) { 953bf181b9fSMichel Lespinasse anon_vma_interval_tree_post_update_vma(vma); 954bf181b9fSMichel Lespinasse if (adjust_next) 955bf181b9fSMichel Lespinasse anon_vma_interval_tree_post_update_vma(next); 95608b52706SKonstantin Khlebnikov anon_vma_unlock_write(anon_vma); 957bf181b9fSMichel Lespinasse } 9581da177e4SLinus Torvalds 9590fc48a6eSWei Yang if (file) { 960808fbdbeSWei Yang i_mmap_unlock_write(mapping); 9617b2d81d4SIngo Molnar uprobe_mmap(vma); 9622b144498SSrikar Dronamraju 9632b144498SSrikar Dronamraju if (adjust_next) 9647b2d81d4SIngo Molnar uprobe_mmap(next); 9652b144498SSrikar Dronamraju } 9662b144498SSrikar Dronamraju 9671da177e4SLinus Torvalds if (remove_next) { 968925d1c40SMatt Helsley if (file) { 969cbc91f71SSrikar Dronamraju uprobe_munmap(next, next->vm_start, next->vm_end); 9701da177e4SLinus Torvalds fput(file); 971925d1c40SMatt Helsley } 9725beb4930SRik van Riel if (next->anon_vma) 9735beb4930SRik van Riel anon_vma_merge(vma, next); 9741da177e4SLinus Torvalds mm->map_count--; 9753964acd0SOleg Nesterov mpol_put(vma_policy(next)); 9763928d4f5SLinus Torvalds vm_area_free(next); 9771da177e4SLinus Torvalds /* 9781da177e4SLinus Torvalds * In mprotect's case 6 (see comments on vma_merge), 9791da177e4SLinus Torvalds * we must remove another next too. It would clutter 9801da177e4SLinus Torvalds * up the code too much to do both in one go. 9811da177e4SLinus Torvalds */ 982e86f15eeSAndrea Arcangeli if (remove_next != 3) { 983e86f15eeSAndrea Arcangeli /* 984e86f15eeSAndrea Arcangeli * If "next" was removed and vma->vm_end was 985e86f15eeSAndrea Arcangeli * expanded (up) over it, in turn 986e86f15eeSAndrea Arcangeli * "next->vm_prev->vm_end" changed and the 987e86f15eeSAndrea Arcangeli * "vma->vm_next" gap must be updated. 988e86f15eeSAndrea Arcangeli */ 9891da177e4SLinus Torvalds next = vma->vm_next; 990e86f15eeSAndrea Arcangeli } else { 991e86f15eeSAndrea Arcangeli /* 992e86f15eeSAndrea Arcangeli * For the scope of the comment "next" and 993e86f15eeSAndrea Arcangeli * "vma" considered pre-swap(): if "vma" was 994e86f15eeSAndrea Arcangeli * removed, next->vm_start was expanded (down) 995e86f15eeSAndrea Arcangeli * over it and the "next" gap must be updated. 996e86f15eeSAndrea Arcangeli * Because of the swap() the post-swap() "vma" 997e86f15eeSAndrea Arcangeli * actually points to pre-swap() "next" 998e86f15eeSAndrea Arcangeli * (post-swap() "next" as opposed is now a 999e86f15eeSAndrea Arcangeli * dangling pointer). 1000e86f15eeSAndrea Arcangeli */ 1001e86f15eeSAndrea Arcangeli next = vma; 1002e86f15eeSAndrea Arcangeli } 1003734537c9SKirill A. Shutemov if (remove_next == 2) { 1004734537c9SKirill A. Shutemov remove_next = 1; 1005734537c9SKirill A. Shutemov end = next->vm_end; 10061da177e4SLinus Torvalds goto again; 1007734537c9SKirill A. Shutemov } 1008d3737187SMichel Lespinasse else if (next) 1009d3737187SMichel Lespinasse vma_gap_update(next); 1010fb8c41e9SAndrea Arcangeli else { 1011fb8c41e9SAndrea Arcangeli /* 1012fb8c41e9SAndrea Arcangeli * If remove_next == 2 we obviously can't 1013fb8c41e9SAndrea Arcangeli * reach this path. 1014fb8c41e9SAndrea Arcangeli * 1015fb8c41e9SAndrea Arcangeli * If remove_next == 3 we can't reach this 1016fb8c41e9SAndrea Arcangeli * path because pre-swap() next is always not 1017fb8c41e9SAndrea Arcangeli * NULL. pre-swap() "next" is not being 1018fb8c41e9SAndrea Arcangeli * removed and its next->vm_end is not altered 1019fb8c41e9SAndrea Arcangeli * (and furthermore "end" already matches 1020fb8c41e9SAndrea Arcangeli * next->vm_end in remove_next == 3). 1021fb8c41e9SAndrea Arcangeli * 1022fb8c41e9SAndrea Arcangeli * We reach this only in the remove_next == 1 1023fb8c41e9SAndrea Arcangeli * case if the "next" vma that was removed was 1024fb8c41e9SAndrea Arcangeli * the highest vma of the mm. However in such 1025fb8c41e9SAndrea Arcangeli * case next->vm_end == "end" and the extended 1026fb8c41e9SAndrea Arcangeli * "vma" has vma->vm_end == next->vm_end so 1027fb8c41e9SAndrea Arcangeli * mm->highest_vm_end doesn't need any update 1028fb8c41e9SAndrea Arcangeli * in remove_next == 1 case. 1029fb8c41e9SAndrea Arcangeli */ 10301be7107fSHugh Dickins VM_WARN_ON(mm->highest_vm_end != vm_end_gap(vma)); 1031fb8c41e9SAndrea Arcangeli } 10321da177e4SLinus Torvalds } 10332b144498SSrikar Dronamraju if (insert && file) 10347b2d81d4SIngo Molnar uprobe_mmap(insert); 10351da177e4SLinus Torvalds 10361da177e4SLinus Torvalds validate_mm(mm); 10375beb4930SRik van Riel 10385beb4930SRik van Riel return 0; 10391da177e4SLinus Torvalds } 10401da177e4SLinus Torvalds 10411da177e4SLinus Torvalds /* 10421da177e4SLinus Torvalds * If the vma has a ->close operation then the driver probably needs to release 10431da177e4SLinus Torvalds * per-vma resources, so we don't attempt to merge those. 10441da177e4SLinus Torvalds */ 10451da177e4SLinus Torvalds static inline int is_mergeable_vma(struct vm_area_struct *vma, 104619a809afSAndrea Arcangeli struct file *file, unsigned long vm_flags, 10479a10064fSColin Cross struct vm_userfaultfd_ctx vm_userfaultfd_ctx, 10485c26f6acSSuren Baghdasaryan struct anon_vma_name *anon_name) 10491da177e4SLinus Torvalds { 105034228d47SCyrill Gorcunov /* 105134228d47SCyrill Gorcunov * VM_SOFTDIRTY should not prevent from VMA merging, if we 105234228d47SCyrill Gorcunov * match the flags but dirty bit -- the caller should mark 105334228d47SCyrill Gorcunov * merged VMA as dirty. If dirty bit won't be excluded from 10548bb4e7a2SWei Yang * comparison, we increase pressure on the memory system forcing 105534228d47SCyrill Gorcunov * the kernel to generate new VMAs when old one could be 105634228d47SCyrill Gorcunov * extended instead. 105734228d47SCyrill Gorcunov */ 105834228d47SCyrill Gorcunov if ((vma->vm_flags ^ vm_flags) & ~VM_SOFTDIRTY) 10591da177e4SLinus Torvalds return 0; 10601da177e4SLinus Torvalds if (vma->vm_file != file) 10611da177e4SLinus Torvalds return 0; 10621da177e4SLinus Torvalds if (vma->vm_ops && vma->vm_ops->close) 10631da177e4SLinus Torvalds return 0; 106419a809afSAndrea Arcangeli if (!is_mergeable_vm_userfaultfd_ctx(vma, vm_userfaultfd_ctx)) 106519a809afSAndrea Arcangeli return 0; 10665c26f6acSSuren Baghdasaryan if (!anon_vma_name_eq(anon_vma_name(vma), anon_name)) 10679a10064fSColin Cross return 0; 10681da177e4SLinus Torvalds return 1; 10691da177e4SLinus Torvalds } 10701da177e4SLinus Torvalds 10711da177e4SLinus Torvalds static inline int is_mergeable_anon_vma(struct anon_vma *anon_vma1, 1072965f55deSShaohua Li struct anon_vma *anon_vma2, 1073965f55deSShaohua Li struct vm_area_struct *vma) 10741da177e4SLinus Torvalds { 1075965f55deSShaohua Li /* 1076965f55deSShaohua Li * The list_is_singular() test is to avoid merging VMA cloned from 1077965f55deSShaohua Li * parents. This can improve scalability caused by anon_vma lock. 1078965f55deSShaohua Li */ 1079965f55deSShaohua Li if ((!anon_vma1 || !anon_vma2) && (!vma || 1080965f55deSShaohua Li list_is_singular(&vma->anon_vma_chain))) 1081965f55deSShaohua Li return 1; 1082965f55deSShaohua Li return anon_vma1 == anon_vma2; 10831da177e4SLinus Torvalds } 10841da177e4SLinus Torvalds 10851da177e4SLinus Torvalds /* 10861da177e4SLinus Torvalds * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff) 10871da177e4SLinus Torvalds * in front of (at a lower virtual address and file offset than) the vma. 10881da177e4SLinus Torvalds * 10891da177e4SLinus Torvalds * We cannot merge two vmas if they have differently assigned (non-NULL) 10901da177e4SLinus Torvalds * anon_vmas, nor if same anon_vma is assigned but offsets incompatible. 10911da177e4SLinus Torvalds * 10921da177e4SLinus Torvalds * We don't check here for the merged mmap wrapping around the end of pagecache 109345e55300SPeter Collingbourne * indices (16TB on ia32) because do_mmap() does not permit mmap's which 10941da177e4SLinus Torvalds * wrap, nor mmaps which cover the final page at index -1UL. 10951da177e4SLinus Torvalds */ 10961da177e4SLinus Torvalds static int 10971da177e4SLinus Torvalds can_vma_merge_before(struct vm_area_struct *vma, unsigned long vm_flags, 109819a809afSAndrea Arcangeli struct anon_vma *anon_vma, struct file *file, 109919a809afSAndrea Arcangeli pgoff_t vm_pgoff, 11009a10064fSColin Cross struct vm_userfaultfd_ctx vm_userfaultfd_ctx, 11015c26f6acSSuren Baghdasaryan struct anon_vma_name *anon_name) 11021da177e4SLinus Torvalds { 11039a10064fSColin Cross if (is_mergeable_vma(vma, file, vm_flags, vm_userfaultfd_ctx, anon_name) && 1104965f55deSShaohua Li is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) { 11051da177e4SLinus Torvalds if (vma->vm_pgoff == vm_pgoff) 11061da177e4SLinus Torvalds return 1; 11071da177e4SLinus Torvalds } 11081da177e4SLinus Torvalds return 0; 11091da177e4SLinus Torvalds } 11101da177e4SLinus Torvalds 11111da177e4SLinus Torvalds /* 11121da177e4SLinus Torvalds * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff) 11131da177e4SLinus Torvalds * beyond (at a higher virtual address and file offset than) the vma. 11141da177e4SLinus Torvalds * 11151da177e4SLinus Torvalds * We cannot merge two vmas if they have differently assigned (non-NULL) 11161da177e4SLinus Torvalds * anon_vmas, nor if same anon_vma is assigned but offsets incompatible. 11171da177e4SLinus Torvalds */ 11181da177e4SLinus Torvalds static int 11191da177e4SLinus Torvalds can_vma_merge_after(struct vm_area_struct *vma, unsigned long vm_flags, 112019a809afSAndrea Arcangeli struct anon_vma *anon_vma, struct file *file, 112119a809afSAndrea Arcangeli pgoff_t vm_pgoff, 11229a10064fSColin Cross struct vm_userfaultfd_ctx vm_userfaultfd_ctx, 11235c26f6acSSuren Baghdasaryan struct anon_vma_name *anon_name) 11241da177e4SLinus Torvalds { 11259a10064fSColin Cross if (is_mergeable_vma(vma, file, vm_flags, vm_userfaultfd_ctx, anon_name) && 1126965f55deSShaohua Li is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) { 11271da177e4SLinus Torvalds pgoff_t vm_pglen; 1128d6e93217SLibin vm_pglen = vma_pages(vma); 11291da177e4SLinus Torvalds if (vma->vm_pgoff + vm_pglen == vm_pgoff) 11301da177e4SLinus Torvalds return 1; 11311da177e4SLinus Torvalds } 11321da177e4SLinus Torvalds return 0; 11331da177e4SLinus Torvalds } 11341da177e4SLinus Torvalds 11351da177e4SLinus Torvalds /* 11369a10064fSColin Cross * Given a mapping request (addr,end,vm_flags,file,pgoff,anon_name), 11379a10064fSColin Cross * figure out whether that can be merged with its predecessor or its 11389a10064fSColin Cross * successor. Or both (it neatly fills a hole). 11391da177e4SLinus Torvalds * 11401da177e4SLinus Torvalds * In most cases - when called for mmap, brk or mremap - [addr,end) is 11411da177e4SLinus Torvalds * certain not to be mapped by the time vma_merge is called; but when 11421da177e4SLinus Torvalds * called for mprotect, it is certain to be already mapped (either at 11431da177e4SLinus Torvalds * an offset within prev, or at the start of next), and the flags of 11441da177e4SLinus Torvalds * this area are about to be changed to vm_flags - and the no-change 11451da177e4SLinus Torvalds * case has already been eliminated. 11461da177e4SLinus Torvalds * 11471da177e4SLinus Torvalds * The following mprotect cases have to be considered, where AAAA is 11481da177e4SLinus Torvalds * the area passed down from mprotect_fixup, never extending beyond one 11491da177e4SLinus Torvalds * vma, PPPPPP is the prev vma specified, and NNNNNN the next vma after: 11501da177e4SLinus Torvalds * 11515d42ab29SWei Yang * AAAA AAAA AAAA 11525d42ab29SWei Yang * PPPPPPNNNNNN PPPPPPNNNNNN PPPPPPNNNNNN 11535d42ab29SWei Yang * cannot merge might become might become 11545d42ab29SWei Yang * PPNNNNNNNNNN PPPPPPPPPPNN 11555d42ab29SWei Yang * mmap, brk or case 4 below case 5 below 11565d42ab29SWei Yang * mremap move: 11575d42ab29SWei Yang * AAAA AAAA 11585d42ab29SWei Yang * PPPP NNNN PPPPNNNNXXXX 11595d42ab29SWei Yang * might become might become 11605d42ab29SWei Yang * PPPPPPPPPPPP 1 or PPPPPPPPPPPP 6 or 11615d42ab29SWei Yang * PPPPPPPPNNNN 2 or PPPPPPPPXXXX 7 or 11625d42ab29SWei Yang * PPPPNNNNNNNN 3 PPPPXXXXXXXX 8 11631da177e4SLinus Torvalds * 11648bb4e7a2SWei Yang * It is important for case 8 that the vma NNNN overlapping the 1165e86f15eeSAndrea Arcangeli * region AAAA is never going to extended over XXXX. Instead XXXX must 1166e86f15eeSAndrea Arcangeli * be extended in region AAAA and NNNN must be removed. This way in 1167e86f15eeSAndrea Arcangeli * all cases where vma_merge succeeds, the moment vma_adjust drops the 1168e86f15eeSAndrea Arcangeli * rmap_locks, the properties of the merged vma will be already 1169e86f15eeSAndrea Arcangeli * correct for the whole merged range. Some of those properties like 1170e86f15eeSAndrea Arcangeli * vm_page_prot/vm_flags may be accessed by rmap_walks and they must 1171e86f15eeSAndrea Arcangeli * be correct for the whole merged range immediately after the 1172e86f15eeSAndrea Arcangeli * rmap_locks are released. Otherwise if XXXX would be removed and 1173e86f15eeSAndrea Arcangeli * NNNN would be extended over the XXXX range, remove_migration_ptes 1174e86f15eeSAndrea Arcangeli * or other rmap walkers (if working on addresses beyond the "end" 1175e86f15eeSAndrea Arcangeli * parameter) may establish ptes with the wrong permissions of NNNN 1176e86f15eeSAndrea Arcangeli * instead of the right permissions of XXXX. 11771da177e4SLinus Torvalds */ 11781da177e4SLinus Torvalds struct vm_area_struct *vma_merge(struct mm_struct *mm, 11791da177e4SLinus Torvalds struct vm_area_struct *prev, unsigned long addr, 11801da177e4SLinus Torvalds unsigned long end, unsigned long vm_flags, 11811da177e4SLinus Torvalds struct anon_vma *anon_vma, struct file *file, 118219a809afSAndrea Arcangeli pgoff_t pgoff, struct mempolicy *policy, 11839a10064fSColin Cross struct vm_userfaultfd_ctx vm_userfaultfd_ctx, 11845c26f6acSSuren Baghdasaryan struct anon_vma_name *anon_name) 11851da177e4SLinus Torvalds { 11861da177e4SLinus Torvalds pgoff_t pglen = (end - addr) >> PAGE_SHIFT; 11871da177e4SLinus Torvalds struct vm_area_struct *area, *next; 11885beb4930SRik van Riel int err; 11891da177e4SLinus Torvalds 11901da177e4SLinus Torvalds /* 11911da177e4SLinus Torvalds * We later require that vma->vm_flags == vm_flags, 11921da177e4SLinus Torvalds * so this tests vma->vm_flags & VM_SPECIAL, too. 11931da177e4SLinus Torvalds */ 11941da177e4SLinus Torvalds if (vm_flags & VM_SPECIAL) 11951da177e4SLinus Torvalds return NULL; 11961da177e4SLinus Torvalds 11973903b55aSLiam R. Howlett next = vma_next(mm, prev); 11981da177e4SLinus Torvalds area = next; 1199e86f15eeSAndrea Arcangeli if (area && area->vm_end == end) /* cases 6, 7, 8 */ 12001da177e4SLinus Torvalds next = next->vm_next; 12011da177e4SLinus Torvalds 1202e86f15eeSAndrea Arcangeli /* verify some invariant that must be enforced by the caller */ 1203e86f15eeSAndrea Arcangeli VM_WARN_ON(prev && addr <= prev->vm_start); 1204e86f15eeSAndrea Arcangeli VM_WARN_ON(area && end > area->vm_end); 1205e86f15eeSAndrea Arcangeli VM_WARN_ON(addr >= end); 1206e86f15eeSAndrea Arcangeli 12071da177e4SLinus Torvalds /* 12081da177e4SLinus Torvalds * Can it merge with the predecessor? 12091da177e4SLinus Torvalds */ 12101da177e4SLinus Torvalds if (prev && prev->vm_end == addr && 12111da177e4SLinus Torvalds mpol_equal(vma_policy(prev), policy) && 12121da177e4SLinus Torvalds can_vma_merge_after(prev, vm_flags, 121319a809afSAndrea Arcangeli anon_vma, file, pgoff, 12149a10064fSColin Cross vm_userfaultfd_ctx, anon_name)) { 12151da177e4SLinus Torvalds /* 12161da177e4SLinus Torvalds * OK, it can. Can we now merge in the successor as well? 12171da177e4SLinus Torvalds */ 12181da177e4SLinus Torvalds if (next && end == next->vm_start && 12191da177e4SLinus Torvalds mpol_equal(policy, vma_policy(next)) && 12201da177e4SLinus Torvalds can_vma_merge_before(next, vm_flags, 122119a809afSAndrea Arcangeli anon_vma, file, 122219a809afSAndrea Arcangeli pgoff+pglen, 12239a10064fSColin Cross vm_userfaultfd_ctx, anon_name) && 12241da177e4SLinus Torvalds is_mergeable_anon_vma(prev->anon_vma, 1225965f55deSShaohua Li next->anon_vma, NULL)) { 12261da177e4SLinus Torvalds /* cases 1, 6 */ 1227e86f15eeSAndrea Arcangeli err = __vma_adjust(prev, prev->vm_start, 1228e86f15eeSAndrea Arcangeli next->vm_end, prev->vm_pgoff, NULL, 1229e86f15eeSAndrea Arcangeli prev); 12301da177e4SLinus Torvalds } else /* cases 2, 5, 7 */ 1231e86f15eeSAndrea Arcangeli err = __vma_adjust(prev, prev->vm_start, 1232e86f15eeSAndrea Arcangeli end, prev->vm_pgoff, NULL, prev); 12335beb4930SRik van Riel if (err) 12345beb4930SRik van Riel return NULL; 12356d50e60cSDavid Rientjes khugepaged_enter_vma_merge(prev, vm_flags); 12361da177e4SLinus Torvalds return prev; 12371da177e4SLinus Torvalds } 12381da177e4SLinus Torvalds 12391da177e4SLinus Torvalds /* 12401da177e4SLinus Torvalds * Can this new request be merged in front of next? 12411da177e4SLinus Torvalds */ 12421da177e4SLinus Torvalds if (next && end == next->vm_start && 12431da177e4SLinus Torvalds mpol_equal(policy, vma_policy(next)) && 12441da177e4SLinus Torvalds can_vma_merge_before(next, vm_flags, 124519a809afSAndrea Arcangeli anon_vma, file, pgoff+pglen, 12469a10064fSColin Cross vm_userfaultfd_ctx, anon_name)) { 12471da177e4SLinus Torvalds if (prev && addr < prev->vm_end) /* case 4 */ 1248e86f15eeSAndrea Arcangeli err = __vma_adjust(prev, prev->vm_start, 1249e86f15eeSAndrea Arcangeli addr, prev->vm_pgoff, NULL, next); 1250e86f15eeSAndrea Arcangeli else { /* cases 3, 8 */ 1251e86f15eeSAndrea Arcangeli err = __vma_adjust(area, addr, next->vm_end, 1252e86f15eeSAndrea Arcangeli next->vm_pgoff - pglen, NULL, next); 1253e86f15eeSAndrea Arcangeli /* 1254e86f15eeSAndrea Arcangeli * In case 3 area is already equal to next and 1255e86f15eeSAndrea Arcangeli * this is a noop, but in case 8 "area" has 1256e86f15eeSAndrea Arcangeli * been removed and next was expanded over it. 1257e86f15eeSAndrea Arcangeli */ 1258e86f15eeSAndrea Arcangeli area = next; 1259e86f15eeSAndrea Arcangeli } 12605beb4930SRik van Riel if (err) 12615beb4930SRik van Riel return NULL; 12626d50e60cSDavid Rientjes khugepaged_enter_vma_merge(area, vm_flags); 12631da177e4SLinus Torvalds return area; 12641da177e4SLinus Torvalds } 12651da177e4SLinus Torvalds 12661da177e4SLinus Torvalds return NULL; 12671da177e4SLinus Torvalds } 12681da177e4SLinus Torvalds 12691da177e4SLinus Torvalds /* 1270b4f315b4SEthon Paul * Rough compatibility check to quickly see if it's even worth looking 1271d0e9fe17SLinus Torvalds * at sharing an anon_vma. 1272d0e9fe17SLinus Torvalds * 1273d0e9fe17SLinus Torvalds * They need to have the same vm_file, and the flags can only differ 1274d0e9fe17SLinus Torvalds * in things that mprotect may change. 1275d0e9fe17SLinus Torvalds * 1276d0e9fe17SLinus Torvalds * NOTE! The fact that we share an anon_vma doesn't _have_ to mean that 1277d0e9fe17SLinus Torvalds * we can merge the two vma's. For example, we refuse to merge a vma if 1278d0e9fe17SLinus Torvalds * there is a vm_ops->close() function, because that indicates that the 1279d0e9fe17SLinus Torvalds * driver is doing some kind of reference counting. But that doesn't 1280d0e9fe17SLinus Torvalds * really matter for the anon_vma sharing case. 1281d0e9fe17SLinus Torvalds */ 1282d0e9fe17SLinus Torvalds static int anon_vma_compatible(struct vm_area_struct *a, struct vm_area_struct *b) 1283d0e9fe17SLinus Torvalds { 1284d0e9fe17SLinus Torvalds return a->vm_end == b->vm_start && 1285d0e9fe17SLinus Torvalds mpol_equal(vma_policy(a), vma_policy(b)) && 1286d0e9fe17SLinus Torvalds a->vm_file == b->vm_file && 12876cb4d9a2SAnshuman Khandual !((a->vm_flags ^ b->vm_flags) & ~(VM_ACCESS_FLAGS | VM_SOFTDIRTY)) && 1288d0e9fe17SLinus Torvalds b->vm_pgoff == a->vm_pgoff + ((b->vm_start - a->vm_start) >> PAGE_SHIFT); 1289d0e9fe17SLinus Torvalds } 1290d0e9fe17SLinus Torvalds 1291d0e9fe17SLinus Torvalds /* 1292d0e9fe17SLinus Torvalds * Do some basic sanity checking to see if we can re-use the anon_vma 1293d0e9fe17SLinus Torvalds * from 'old'. The 'a'/'b' vma's are in VM order - one of them will be 1294d0e9fe17SLinus Torvalds * the same as 'old', the other will be the new one that is trying 1295d0e9fe17SLinus Torvalds * to share the anon_vma. 1296d0e9fe17SLinus Torvalds * 1297d0e9fe17SLinus Torvalds * NOTE! This runs with mm_sem held for reading, so it is possible that 1298d0e9fe17SLinus Torvalds * the anon_vma of 'old' is concurrently in the process of being set up 1299d0e9fe17SLinus Torvalds * by another page fault trying to merge _that_. But that's ok: if it 1300d0e9fe17SLinus Torvalds * is being set up, that automatically means that it will be a singleton 1301d0e9fe17SLinus Torvalds * acceptable for merging, so we can do all of this optimistically. But 13024db0c3c2SJason Low * we do that READ_ONCE() to make sure that we never re-load the pointer. 1303d0e9fe17SLinus Torvalds * 1304d0e9fe17SLinus Torvalds * IOW: that the "list_is_singular()" test on the anon_vma_chain only 1305d0e9fe17SLinus Torvalds * matters for the 'stable anon_vma' case (ie the thing we want to avoid 1306d0e9fe17SLinus Torvalds * is to return an anon_vma that is "complex" due to having gone through 1307d0e9fe17SLinus Torvalds * a fork). 1308d0e9fe17SLinus Torvalds * 1309d0e9fe17SLinus Torvalds * We also make sure that the two vma's are compatible (adjacent, 1310d0e9fe17SLinus Torvalds * and with the same memory policies). That's all stable, even with just 1311d0e9fe17SLinus Torvalds * a read lock on the mm_sem. 1312d0e9fe17SLinus Torvalds */ 1313d0e9fe17SLinus Torvalds static struct anon_vma *reusable_anon_vma(struct vm_area_struct *old, struct vm_area_struct *a, struct vm_area_struct *b) 1314d0e9fe17SLinus Torvalds { 1315d0e9fe17SLinus Torvalds if (anon_vma_compatible(a, b)) { 13164db0c3c2SJason Low struct anon_vma *anon_vma = READ_ONCE(old->anon_vma); 1317d0e9fe17SLinus Torvalds 1318d0e9fe17SLinus Torvalds if (anon_vma && list_is_singular(&old->anon_vma_chain)) 1319d0e9fe17SLinus Torvalds return anon_vma; 1320d0e9fe17SLinus Torvalds } 1321d0e9fe17SLinus Torvalds return NULL; 1322d0e9fe17SLinus Torvalds } 1323d0e9fe17SLinus Torvalds 1324d0e9fe17SLinus Torvalds /* 13251da177e4SLinus Torvalds * find_mergeable_anon_vma is used by anon_vma_prepare, to check 13261da177e4SLinus Torvalds * neighbouring vmas for a suitable anon_vma, before it goes off 13271da177e4SLinus Torvalds * to allocate a new anon_vma. It checks because a repetitive 13281da177e4SLinus Torvalds * sequence of mprotects and faults may otherwise lead to distinct 13291da177e4SLinus Torvalds * anon_vmas being allocated, preventing vma merge in subsequent 13301da177e4SLinus Torvalds * mprotect. 13311da177e4SLinus Torvalds */ 13321da177e4SLinus Torvalds struct anon_vma *find_mergeable_anon_vma(struct vm_area_struct *vma) 13331da177e4SLinus Torvalds { 1334a67c8caaSMiaohe Lin struct anon_vma *anon_vma = NULL; 13351da177e4SLinus Torvalds 1336a67c8caaSMiaohe Lin /* Try next first. */ 1337a67c8caaSMiaohe Lin if (vma->vm_next) { 1338a67c8caaSMiaohe Lin anon_vma = reusable_anon_vma(vma->vm_next, vma, vma->vm_next); 1339d0e9fe17SLinus Torvalds if (anon_vma) 1340d0e9fe17SLinus Torvalds return anon_vma; 1341a67c8caaSMiaohe Lin } 13421da177e4SLinus Torvalds 1343a67c8caaSMiaohe Lin /* Try prev next. */ 1344a67c8caaSMiaohe Lin if (vma->vm_prev) 1345a67c8caaSMiaohe Lin anon_vma = reusable_anon_vma(vma->vm_prev, vma->vm_prev, vma); 1346a67c8caaSMiaohe Lin 13471da177e4SLinus Torvalds /* 1348a67c8caaSMiaohe Lin * We might reach here with anon_vma == NULL if we can't find 1349a67c8caaSMiaohe Lin * any reusable anon_vma. 13501da177e4SLinus Torvalds * There's no absolute need to look only at touching neighbours: 13511da177e4SLinus Torvalds * we could search further afield for "compatible" anon_vmas. 13521da177e4SLinus Torvalds * But it would probably just be a waste of time searching, 13531da177e4SLinus Torvalds * or lead to too many vmas hanging off the same anon_vma. 13541da177e4SLinus Torvalds * We're trying to allow mprotect remerging later on, 13551da177e4SLinus Torvalds * not trying to minimize memory used for anon_vmas. 13561da177e4SLinus Torvalds */ 1357a67c8caaSMiaohe Lin return anon_vma; 13581da177e4SLinus Torvalds } 13591da177e4SLinus Torvalds 13601da177e4SLinus Torvalds /* 136140401530SAl Viro * If a hint addr is less than mmap_min_addr change hint to be as 136240401530SAl Viro * low as possible but still greater than mmap_min_addr 136340401530SAl Viro */ 136440401530SAl Viro static inline unsigned long round_hint_to_min(unsigned long hint) 136540401530SAl Viro { 136640401530SAl Viro hint &= PAGE_MASK; 136740401530SAl Viro if (((void *)hint != NULL) && 136840401530SAl Viro (hint < mmap_min_addr)) 136940401530SAl Viro return PAGE_ALIGN(mmap_min_addr); 137040401530SAl Viro return hint; 137140401530SAl Viro } 137240401530SAl Viro 13736aeb2542SMike Rapoport int mlock_future_check(struct mm_struct *mm, unsigned long flags, 1374363ee17fSDavidlohr Bueso unsigned long len) 1375363ee17fSDavidlohr Bueso { 1376363ee17fSDavidlohr Bueso unsigned long locked, lock_limit; 1377363ee17fSDavidlohr Bueso 1378363ee17fSDavidlohr Bueso /* mlock MCL_FUTURE? */ 1379363ee17fSDavidlohr Bueso if (flags & VM_LOCKED) { 1380363ee17fSDavidlohr Bueso locked = len >> PAGE_SHIFT; 1381363ee17fSDavidlohr Bueso locked += mm->locked_vm; 1382363ee17fSDavidlohr Bueso lock_limit = rlimit(RLIMIT_MEMLOCK); 1383363ee17fSDavidlohr Bueso lock_limit >>= PAGE_SHIFT; 1384363ee17fSDavidlohr Bueso if (locked > lock_limit && !capable(CAP_IPC_LOCK)) 1385363ee17fSDavidlohr Bueso return -EAGAIN; 1386363ee17fSDavidlohr Bueso } 1387363ee17fSDavidlohr Bueso return 0; 1388363ee17fSDavidlohr Bueso } 1389363ee17fSDavidlohr Bueso 1390be83bbf8SLinus Torvalds static inline u64 file_mmap_size_max(struct file *file, struct inode *inode) 1391be83bbf8SLinus Torvalds { 1392be83bbf8SLinus Torvalds if (S_ISREG(inode->i_mode)) 1393423913adSLinus Torvalds return MAX_LFS_FILESIZE; 1394be83bbf8SLinus Torvalds 1395be83bbf8SLinus Torvalds if (S_ISBLK(inode->i_mode)) 1396be83bbf8SLinus Torvalds return MAX_LFS_FILESIZE; 1397be83bbf8SLinus Torvalds 139876f34950SIvan Khoronzhuk if (S_ISSOCK(inode->i_mode)) 139976f34950SIvan Khoronzhuk return MAX_LFS_FILESIZE; 140076f34950SIvan Khoronzhuk 1401be83bbf8SLinus Torvalds /* Special "we do even unsigned file positions" case */ 1402be83bbf8SLinus Torvalds if (file->f_mode & FMODE_UNSIGNED_OFFSET) 1403be83bbf8SLinus Torvalds return 0; 1404be83bbf8SLinus Torvalds 1405be83bbf8SLinus Torvalds /* Yes, random drivers might want more. But I'm tired of buggy drivers */ 1406be83bbf8SLinus Torvalds return ULONG_MAX; 1407be83bbf8SLinus Torvalds } 1408be83bbf8SLinus Torvalds 1409be83bbf8SLinus Torvalds static inline bool file_mmap_ok(struct file *file, struct inode *inode, 1410be83bbf8SLinus Torvalds unsigned long pgoff, unsigned long len) 1411be83bbf8SLinus Torvalds { 1412be83bbf8SLinus Torvalds u64 maxsize = file_mmap_size_max(file, inode); 1413be83bbf8SLinus Torvalds 1414be83bbf8SLinus Torvalds if (maxsize && len > maxsize) 1415be83bbf8SLinus Torvalds return false; 1416be83bbf8SLinus Torvalds maxsize -= len; 1417be83bbf8SLinus Torvalds if (pgoff > maxsize >> PAGE_SHIFT) 1418be83bbf8SLinus Torvalds return false; 1419be83bbf8SLinus Torvalds return true; 1420be83bbf8SLinus Torvalds } 1421be83bbf8SLinus Torvalds 142240401530SAl Viro /* 14233e4e28c5SMichel Lespinasse * The caller must write-lock current->mm->mmap_lock. 14241da177e4SLinus Torvalds */ 14251fcfd8dbSOleg Nesterov unsigned long do_mmap(struct file *file, unsigned long addr, 14261da177e4SLinus Torvalds unsigned long len, unsigned long prot, 142745e55300SPeter Collingbourne unsigned long flags, unsigned long pgoff, 142845e55300SPeter Collingbourne unsigned long *populate, struct list_head *uf) 14291da177e4SLinus Torvalds { 14301da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 143145e55300SPeter Collingbourne vm_flags_t vm_flags; 143262b5f7d0SDave Hansen int pkey = 0; 14331da177e4SLinus Torvalds 143441badc15SMichel Lespinasse *populate = 0; 1435bebeb3d6SMichel Lespinasse 1436e37609bbSPiotr Kwapulinski if (!len) 1437e37609bbSPiotr Kwapulinski return -EINVAL; 1438e37609bbSPiotr Kwapulinski 14391da177e4SLinus Torvalds /* 14401da177e4SLinus Torvalds * Does the application expect PROT_READ to imply PROT_EXEC? 14411da177e4SLinus Torvalds * 14421da177e4SLinus Torvalds * (the exception is when the underlying filesystem is noexec 14431da177e4SLinus Torvalds * mounted, in which case we dont add PROT_EXEC.) 14441da177e4SLinus Torvalds */ 14451da177e4SLinus Torvalds if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC)) 144690f8572bSEric W. Biederman if (!(file && path_noexec(&file->f_path))) 14471da177e4SLinus Torvalds prot |= PROT_EXEC; 14481da177e4SLinus Torvalds 1449a4ff8e86SMichal Hocko /* force arch specific MAP_FIXED handling in get_unmapped_area */ 1450a4ff8e86SMichal Hocko if (flags & MAP_FIXED_NOREPLACE) 1451a4ff8e86SMichal Hocko flags |= MAP_FIXED; 1452a4ff8e86SMichal Hocko 14537cd94146SEric Paris if (!(flags & MAP_FIXED)) 14547cd94146SEric Paris addr = round_hint_to_min(addr); 14557cd94146SEric Paris 14561da177e4SLinus Torvalds /* Careful about overflows.. */ 14571da177e4SLinus Torvalds len = PAGE_ALIGN(len); 14589206de95SAl Viro if (!len) 14591da177e4SLinus Torvalds return -ENOMEM; 14601da177e4SLinus Torvalds 14611da177e4SLinus Torvalds /* offset overflow? */ 14621da177e4SLinus Torvalds if ((pgoff + (len >> PAGE_SHIFT)) < pgoff) 14631da177e4SLinus Torvalds return -EOVERFLOW; 14641da177e4SLinus Torvalds 14651da177e4SLinus Torvalds /* Too many mappings? */ 14661da177e4SLinus Torvalds if (mm->map_count > sysctl_max_map_count) 14671da177e4SLinus Torvalds return -ENOMEM; 14681da177e4SLinus Torvalds 14691da177e4SLinus Torvalds /* Obtain the address to map to. we verify (or select) it and ensure 14701da177e4SLinus Torvalds * that it represents a valid section of the address space. 14711da177e4SLinus Torvalds */ 14721da177e4SLinus Torvalds addr = get_unmapped_area(file, addr, len, pgoff, flags); 1473ff68dac6SGaowei Pu if (IS_ERR_VALUE(addr)) 14741da177e4SLinus Torvalds return addr; 14751da177e4SLinus Torvalds 1476a4ff8e86SMichal Hocko if (flags & MAP_FIXED_NOREPLACE) { 147735e43c5fSLiam Howlett if (find_vma_intersection(mm, addr, addr + len)) 1478a4ff8e86SMichal Hocko return -EEXIST; 1479a4ff8e86SMichal Hocko } 1480a4ff8e86SMichal Hocko 148162b5f7d0SDave Hansen if (prot == PROT_EXEC) { 148262b5f7d0SDave Hansen pkey = execute_only_pkey(mm); 148362b5f7d0SDave Hansen if (pkey < 0) 148462b5f7d0SDave Hansen pkey = 0; 148562b5f7d0SDave Hansen } 148662b5f7d0SDave Hansen 14871da177e4SLinus Torvalds /* Do simple checking here so the lower-level routines won't have 14881da177e4SLinus Torvalds * to. we assume access permissions have been handled by the open 14891da177e4SLinus Torvalds * of the memory object, so we don't do any here. 14901da177e4SLinus Torvalds */ 149145e55300SPeter Collingbourne vm_flags = calc_vm_prot_bits(prot, pkey) | calc_vm_flag_bits(flags) | 14921da177e4SLinus Torvalds mm->def_flags | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC; 14931da177e4SLinus Torvalds 1494cdf7b341SHuang Shijie if (flags & MAP_LOCKED) 14951da177e4SLinus Torvalds if (!can_do_mlock()) 14961da177e4SLinus Torvalds return -EPERM; 1497ba470de4SRik van Riel 1498363ee17fSDavidlohr Bueso if (mlock_future_check(mm, vm_flags, len)) 14991da177e4SLinus Torvalds return -EAGAIN; 15001da177e4SLinus Torvalds 15011da177e4SLinus Torvalds if (file) { 1502077bf22bSOleg Nesterov struct inode *inode = file_inode(file); 15031c972597SDan Williams unsigned long flags_mask; 15041c972597SDan Williams 1505be83bbf8SLinus Torvalds if (!file_mmap_ok(file, inode, pgoff, len)) 1506be83bbf8SLinus Torvalds return -EOVERFLOW; 1507be83bbf8SLinus Torvalds 15081c972597SDan Williams flags_mask = LEGACY_MAP_MASK | file->f_op->mmap_supported_flags; 1509077bf22bSOleg Nesterov 15101da177e4SLinus Torvalds switch (flags & MAP_TYPE) { 15111da177e4SLinus Torvalds case MAP_SHARED: 15121c972597SDan Williams /* 15131c972597SDan Williams * Force use of MAP_SHARED_VALIDATE with non-legacy 15141c972597SDan Williams * flags. E.g. MAP_SYNC is dangerous to use with 15151c972597SDan Williams * MAP_SHARED as you don't know which consistency model 15161c972597SDan Williams * you will get. We silently ignore unsupported flags 15171c972597SDan Williams * with MAP_SHARED to preserve backward compatibility. 15181c972597SDan Williams */ 15191c972597SDan Williams flags &= LEGACY_MAP_MASK; 1520e4a9bc58SJoe Perches fallthrough; 15211c972597SDan Williams case MAP_SHARED_VALIDATE: 15221c972597SDan Williams if (flags & ~flags_mask) 15231c972597SDan Williams return -EOPNOTSUPP; 1524dc617f29SDarrick J. Wong if (prot & PROT_WRITE) { 1525dc617f29SDarrick J. Wong if (!(file->f_mode & FMODE_WRITE)) 15261da177e4SLinus Torvalds return -EACCES; 1527dc617f29SDarrick J. Wong if (IS_SWAPFILE(file->f_mapping->host)) 1528dc617f29SDarrick J. Wong return -ETXTBSY; 1529dc617f29SDarrick J. Wong } 15301da177e4SLinus Torvalds 15311da177e4SLinus Torvalds /* 15321da177e4SLinus Torvalds * Make sure we don't allow writing to an append-only 15331da177e4SLinus Torvalds * file.. 15341da177e4SLinus Torvalds */ 15351da177e4SLinus Torvalds if (IS_APPEND(inode) && (file->f_mode & FMODE_WRITE)) 15361da177e4SLinus Torvalds return -EACCES; 15371da177e4SLinus Torvalds 15381da177e4SLinus Torvalds vm_flags |= VM_SHARED | VM_MAYSHARE; 15391da177e4SLinus Torvalds if (!(file->f_mode & FMODE_WRITE)) 15401da177e4SLinus Torvalds vm_flags &= ~(VM_MAYWRITE | VM_SHARED); 1541e4a9bc58SJoe Perches fallthrough; 15421da177e4SLinus Torvalds case MAP_PRIVATE: 15431da177e4SLinus Torvalds if (!(file->f_mode & FMODE_READ)) 15441da177e4SLinus Torvalds return -EACCES; 154590f8572bSEric W. Biederman if (path_noexec(&file->f_path)) { 154680c5606cSLinus Torvalds if (vm_flags & VM_EXEC) 154780c5606cSLinus Torvalds return -EPERM; 154880c5606cSLinus Torvalds vm_flags &= ~VM_MAYEXEC; 154980c5606cSLinus Torvalds } 155080c5606cSLinus Torvalds 155172c2d531SAl Viro if (!file->f_op->mmap) 155280c5606cSLinus Torvalds return -ENODEV; 1553b2c56e4fSOleg Nesterov if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP)) 1554b2c56e4fSOleg Nesterov return -EINVAL; 15551da177e4SLinus Torvalds break; 15561da177e4SLinus Torvalds 15571da177e4SLinus Torvalds default: 15581da177e4SLinus Torvalds return -EINVAL; 15591da177e4SLinus Torvalds } 15601da177e4SLinus Torvalds } else { 15611da177e4SLinus Torvalds switch (flags & MAP_TYPE) { 15621da177e4SLinus Torvalds case MAP_SHARED: 1563b2c56e4fSOleg Nesterov if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP)) 1564b2c56e4fSOleg Nesterov return -EINVAL; 1565ce363942STejun Heo /* 1566ce363942STejun Heo * Ignore pgoff. 1567ce363942STejun Heo */ 1568ce363942STejun Heo pgoff = 0; 15691da177e4SLinus Torvalds vm_flags |= VM_SHARED | VM_MAYSHARE; 15701da177e4SLinus Torvalds break; 15711da177e4SLinus Torvalds case MAP_PRIVATE: 15721da177e4SLinus Torvalds /* 15731da177e4SLinus Torvalds * Set pgoff according to addr for anon_vma. 15741da177e4SLinus Torvalds */ 15751da177e4SLinus Torvalds pgoff = addr >> PAGE_SHIFT; 15761da177e4SLinus Torvalds break; 15771da177e4SLinus Torvalds default: 15781da177e4SLinus Torvalds return -EINVAL; 15791da177e4SLinus Torvalds } 15801da177e4SLinus Torvalds } 15811da177e4SLinus Torvalds 1582c22c0d63SMichel Lespinasse /* 1583c22c0d63SMichel Lespinasse * Set 'VM_NORESERVE' if we should not account for the 1584c22c0d63SMichel Lespinasse * memory use of this mapping. 1585c22c0d63SMichel Lespinasse */ 1586c22c0d63SMichel Lespinasse if (flags & MAP_NORESERVE) { 1587c22c0d63SMichel Lespinasse /* We honor MAP_NORESERVE if allowed to overcommit */ 1588c22c0d63SMichel Lespinasse if (sysctl_overcommit_memory != OVERCOMMIT_NEVER) 1589c22c0d63SMichel Lespinasse vm_flags |= VM_NORESERVE; 1590c22c0d63SMichel Lespinasse 1591c22c0d63SMichel Lespinasse /* hugetlb applies strict overcommit unless MAP_NORESERVE */ 1592c22c0d63SMichel Lespinasse if (file && is_file_hugepages(file)) 1593c22c0d63SMichel Lespinasse vm_flags |= VM_NORESERVE; 1594c22c0d63SMichel Lespinasse } 1595c22c0d63SMichel Lespinasse 1596897ab3e0SMike Rapoport addr = mmap_region(file, addr, len, vm_flags, pgoff, uf); 159709a9f1d2SMichel Lespinasse if (!IS_ERR_VALUE(addr) && 159809a9f1d2SMichel Lespinasse ((vm_flags & VM_LOCKED) || 159909a9f1d2SMichel Lespinasse (flags & (MAP_POPULATE | MAP_NONBLOCK)) == MAP_POPULATE)) 160041badc15SMichel Lespinasse *populate = len; 1601bebeb3d6SMichel Lespinasse return addr; 16020165ab44SMiklos Szeredi } 16036be5ceb0SLinus Torvalds 1604a90f590aSDominik Brodowski unsigned long ksys_mmap_pgoff(unsigned long addr, unsigned long len, 1605a90f590aSDominik Brodowski unsigned long prot, unsigned long flags, 1606a90f590aSDominik Brodowski unsigned long fd, unsigned long pgoff) 160766f0dc48SHugh Dickins { 160866f0dc48SHugh Dickins struct file *file = NULL; 16091e3ee14bSChen Gang unsigned long retval; 161066f0dc48SHugh Dickins 161166f0dc48SHugh Dickins if (!(flags & MAP_ANONYMOUS)) { 1612120a795dSAl Viro audit_mmap_fd(fd, flags); 161366f0dc48SHugh Dickins file = fget(fd); 161466f0dc48SHugh Dickins if (!file) 16151e3ee14bSChen Gang return -EBADF; 16167bba8f0eSZhen Lei if (is_file_hugepages(file)) { 1617af73e4d9SNaoya Horiguchi len = ALIGN(len, huge_page_size(hstate_file(file))); 16187bba8f0eSZhen Lei } else if (unlikely(flags & MAP_HUGETLB)) { 1619493af578SJörn Engel retval = -EINVAL; 1620493af578SJörn Engel goto out_fput; 16217bba8f0eSZhen Lei } 162266f0dc48SHugh Dickins } else if (flags & MAP_HUGETLB) { 1623c103a4dcSAndrew Morton struct hstate *hs; 1624af73e4d9SNaoya Horiguchi 162520ac2893SAnshuman Khandual hs = hstate_sizelog((flags >> MAP_HUGE_SHIFT) & MAP_HUGE_MASK); 1626091d0d55SLi Zefan if (!hs) 1627091d0d55SLi Zefan return -EINVAL; 1628091d0d55SLi Zefan 1629091d0d55SLi Zefan len = ALIGN(len, huge_page_size(hs)); 163066f0dc48SHugh Dickins /* 163166f0dc48SHugh Dickins * VM_NORESERVE is used because the reservations will be 163266f0dc48SHugh Dickins * taken when vm_ops->mmap() is called 163366f0dc48SHugh Dickins */ 1634af73e4d9SNaoya Horiguchi file = hugetlb_file_setup(HUGETLB_ANON_FILE, len, 163542d7395fSAndi Kleen VM_NORESERVE, 163683c1fd76Szhangyiru HUGETLB_ANONHUGE_INODE, 163742d7395fSAndi Kleen (flags >> MAP_HUGE_SHIFT) & MAP_HUGE_MASK); 163866f0dc48SHugh Dickins if (IS_ERR(file)) 163966f0dc48SHugh Dickins return PTR_ERR(file); 164066f0dc48SHugh Dickins } 164166f0dc48SHugh Dickins 16429fbeb5abSMichal Hocko retval = vm_mmap_pgoff(file, addr, len, prot, flags, pgoff); 1643493af578SJörn Engel out_fput: 164466f0dc48SHugh Dickins if (file) 164566f0dc48SHugh Dickins fput(file); 164666f0dc48SHugh Dickins return retval; 164766f0dc48SHugh Dickins } 164866f0dc48SHugh Dickins 1649a90f590aSDominik Brodowski SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len, 1650a90f590aSDominik Brodowski unsigned long, prot, unsigned long, flags, 1651a90f590aSDominik Brodowski unsigned long, fd, unsigned long, pgoff) 1652a90f590aSDominik Brodowski { 1653a90f590aSDominik Brodowski return ksys_mmap_pgoff(addr, len, prot, flags, fd, pgoff); 1654a90f590aSDominik Brodowski } 1655a90f590aSDominik Brodowski 1656a4679373SChristoph Hellwig #ifdef __ARCH_WANT_SYS_OLD_MMAP 1657a4679373SChristoph Hellwig struct mmap_arg_struct { 1658a4679373SChristoph Hellwig unsigned long addr; 1659a4679373SChristoph Hellwig unsigned long len; 1660a4679373SChristoph Hellwig unsigned long prot; 1661a4679373SChristoph Hellwig unsigned long flags; 1662a4679373SChristoph Hellwig unsigned long fd; 1663a4679373SChristoph Hellwig unsigned long offset; 1664a4679373SChristoph Hellwig }; 1665a4679373SChristoph Hellwig 1666a4679373SChristoph Hellwig SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg) 1667a4679373SChristoph Hellwig { 1668a4679373SChristoph Hellwig struct mmap_arg_struct a; 1669a4679373SChristoph Hellwig 1670a4679373SChristoph Hellwig if (copy_from_user(&a, arg, sizeof(a))) 1671a4679373SChristoph Hellwig return -EFAULT; 1672de1741a1SAlexander Kuleshov if (offset_in_page(a.offset)) 1673a4679373SChristoph Hellwig return -EINVAL; 1674a4679373SChristoph Hellwig 1675a90f590aSDominik Brodowski return ksys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd, 1676a4679373SChristoph Hellwig a.offset >> PAGE_SHIFT); 1677a4679373SChristoph Hellwig } 1678a4679373SChristoph Hellwig #endif /* __ARCH_WANT_SYS_OLD_MMAP */ 1679a4679373SChristoph Hellwig 16804e950f6fSAlexey Dobriyan /* 16818bb4e7a2SWei Yang * Some shared mappings will want the pages marked read-only 16824e950f6fSAlexey Dobriyan * to track write events. If so, we'll downgrade vm_page_prot 16834e950f6fSAlexey Dobriyan * to the private version (using protection_map[] without the 16844e950f6fSAlexey Dobriyan * VM_SHARED bit). 16854e950f6fSAlexey Dobriyan */ 16866d2329f8SAndrea Arcangeli int vma_wants_writenotify(struct vm_area_struct *vma, pgprot_t vm_page_prot) 16874e950f6fSAlexey Dobriyan { 1688ca16d140SKOSAKI Motohiro vm_flags_t vm_flags = vma->vm_flags; 16898a04446aSKirill A. Shutemov const struct vm_operations_struct *vm_ops = vma->vm_ops; 16904e950f6fSAlexey Dobriyan 16914e950f6fSAlexey Dobriyan /* If it was private or non-writable, the write bit is already clear */ 16924e950f6fSAlexey Dobriyan if ((vm_flags & (VM_WRITE|VM_SHARED)) != ((VM_WRITE|VM_SHARED))) 16934e950f6fSAlexey Dobriyan return 0; 16944e950f6fSAlexey Dobriyan 16954e950f6fSAlexey Dobriyan /* The backer wishes to know when pages are first written to? */ 16968a04446aSKirill A. Shutemov if (vm_ops && (vm_ops->page_mkwrite || vm_ops->pfn_mkwrite)) 16974e950f6fSAlexey Dobriyan return 1; 16984e950f6fSAlexey Dobriyan 169964e45507SPeter Feiner /* The open routine did something to the protections that pgprot_modify 170064e45507SPeter Feiner * won't preserve? */ 17016d2329f8SAndrea Arcangeli if (pgprot_val(vm_page_prot) != 17026d2329f8SAndrea Arcangeli pgprot_val(vm_pgprot_modify(vm_page_prot, vm_flags))) 17034e950f6fSAlexey Dobriyan return 0; 17044e950f6fSAlexey Dobriyan 170564e45507SPeter Feiner /* Do we need to track softdirty? */ 170664e45507SPeter Feiner if (IS_ENABLED(CONFIG_MEM_SOFT_DIRTY) && !(vm_flags & VM_SOFTDIRTY)) 170764e45507SPeter Feiner return 1; 170864e45507SPeter Feiner 17094e950f6fSAlexey Dobriyan /* Specialty mapping? */ 17104b6e1e37SKonstantin Khlebnikov if (vm_flags & VM_PFNMAP) 17114e950f6fSAlexey Dobriyan return 0; 17124e950f6fSAlexey Dobriyan 17134e950f6fSAlexey Dobriyan /* Can the mapping track the dirty pages? */ 17144e950f6fSAlexey Dobriyan return vma->vm_file && vma->vm_file->f_mapping && 1715f56753acSChristoph Hellwig mapping_can_writeback(vma->vm_file->f_mapping); 17164e950f6fSAlexey Dobriyan } 17174e950f6fSAlexey Dobriyan 1718fc8744adSLinus Torvalds /* 1719fc8744adSLinus Torvalds * We account for memory if it's a private writeable mapping, 17205a6fe125SMel Gorman * not hugepages and VM_NORESERVE wasn't set. 1721fc8744adSLinus Torvalds */ 1722ca16d140SKOSAKI Motohiro static inline int accountable_mapping(struct file *file, vm_flags_t vm_flags) 1723fc8744adSLinus Torvalds { 17245a6fe125SMel Gorman /* 17255a6fe125SMel Gorman * hugetlb has its own accounting separate from the core VM 17265a6fe125SMel Gorman * VM_HUGETLB may not be set yet so we cannot check for that flag. 17275a6fe125SMel Gorman */ 17285a6fe125SMel Gorman if (file && is_file_hugepages(file)) 17295a6fe125SMel Gorman return 0; 17305a6fe125SMel Gorman 1731fc8744adSLinus Torvalds return (vm_flags & (VM_NORESERVE | VM_SHARED | VM_WRITE)) == VM_WRITE; 1732fc8744adSLinus Torvalds } 1733fc8744adSLinus Torvalds 17340165ab44SMiklos Szeredi unsigned long mmap_region(struct file *file, unsigned long addr, 1735897ab3e0SMike Rapoport unsigned long len, vm_flags_t vm_flags, unsigned long pgoff, 1736897ab3e0SMike Rapoport struct list_head *uf) 17370165ab44SMiklos Szeredi { 17380165ab44SMiklos Szeredi struct mm_struct *mm = current->mm; 1739d70cec89SMiaohe Lin struct vm_area_struct *vma, *prev, *merge; 17400165ab44SMiklos Szeredi int error; 17410165ab44SMiklos Szeredi struct rb_node **rb_link, *rb_parent; 17420165ab44SMiklos Szeredi unsigned long charged = 0; 17430165ab44SMiklos Szeredi 1744e8420a8eSCyril Hrubis /* Check against address space limit. */ 174584638335SKonstantin Khlebnikov if (!may_expand_vm(mm, vm_flags, len >> PAGE_SHIFT)) { 1746e8420a8eSCyril Hrubis unsigned long nr_pages; 1747e8420a8eSCyril Hrubis 1748e8420a8eSCyril Hrubis /* 1749e8420a8eSCyril Hrubis * MAP_FIXED may remove pages of mappings that intersects with 1750e8420a8eSCyril Hrubis * requested mapping. Account for the pages it would unmap. 1751e8420a8eSCyril Hrubis */ 1752e8420a8eSCyril Hrubis nr_pages = count_vma_pages_range(mm, addr, addr + len); 1753e8420a8eSCyril Hrubis 175484638335SKonstantin Khlebnikov if (!may_expand_vm(mm, vm_flags, 175584638335SKonstantin Khlebnikov (len >> PAGE_SHIFT) - nr_pages)) 1756e8420a8eSCyril Hrubis return -ENOMEM; 1757e8420a8eSCyril Hrubis } 1758e8420a8eSCyril Hrubis 1759fb8090b6SLiam R. Howlett /* Clear old maps, set up prev, rb_link, rb_parent, and uf */ 1760fb8090b6SLiam R. Howlett if (munmap_vma_range(mm, addr, len, &prev, &rb_link, &rb_parent, uf)) 17611da177e4SLinus Torvalds return -ENOMEM; 1762fc8744adSLinus Torvalds /* 17631da177e4SLinus Torvalds * Private writable mapping: check memory availability 17641da177e4SLinus Torvalds */ 17655a6fe125SMel Gorman if (accountable_mapping(file, vm_flags)) { 17661da177e4SLinus Torvalds charged = len >> PAGE_SHIFT; 1767191c5424SAl Viro if (security_vm_enough_memory_mm(mm, charged)) 17681da177e4SLinus Torvalds return -ENOMEM; 17691da177e4SLinus Torvalds vm_flags |= VM_ACCOUNT; 17701da177e4SLinus Torvalds } 17711da177e4SLinus Torvalds 17721da177e4SLinus Torvalds /* 1773de33c8dbSLinus Torvalds * Can we just expand an old mapping? 17741da177e4SLinus Torvalds */ 177519a809afSAndrea Arcangeli vma = vma_merge(mm, prev, addr, addr + len, vm_flags, 17769a10064fSColin Cross NULL, file, pgoff, NULL, NULL_VM_UFFD_CTX, NULL); 1777ba470de4SRik van Riel if (vma) 17781da177e4SLinus Torvalds goto out; 17791da177e4SLinus Torvalds 17801da177e4SLinus Torvalds /* 17811da177e4SLinus Torvalds * Determine the object being mapped and call the appropriate 17821da177e4SLinus Torvalds * specific mapper. the address has already been validated, but 17831da177e4SLinus Torvalds * not unmapped, but the maps are removed from the list. 17841da177e4SLinus Torvalds */ 1785490fc053SLinus Torvalds vma = vm_area_alloc(mm); 17861da177e4SLinus Torvalds if (!vma) { 17871da177e4SLinus Torvalds error = -ENOMEM; 17881da177e4SLinus Torvalds goto unacct_error; 17891da177e4SLinus Torvalds } 17901da177e4SLinus Torvalds 17911da177e4SLinus Torvalds vma->vm_start = addr; 17921da177e4SLinus Torvalds vma->vm_end = addr + len; 17931da177e4SLinus Torvalds vma->vm_flags = vm_flags; 17943ed75eb8SColy Li vma->vm_page_prot = vm_get_page_prot(vm_flags); 17951da177e4SLinus Torvalds vma->vm_pgoff = pgoff; 17961da177e4SLinus Torvalds 17971da177e4SLinus Torvalds if (file) { 17984bb5f5d9SDavid Herrmann if (vm_flags & VM_SHARED) { 17994bb5f5d9SDavid Herrmann error = mapping_map_writable(file->f_mapping); 18004bb5f5d9SDavid Herrmann if (error) 18018d0920bdSDavid Hildenbrand goto free_vma; 18024bb5f5d9SDavid Herrmann } 18034bb5f5d9SDavid Herrmann 1804cb0942b8SAl Viro vma->vm_file = get_file(file); 1805f74ac015SMiklos Szeredi error = call_mmap(file, vma); 18061da177e4SLinus Torvalds if (error) 18071da177e4SLinus Torvalds goto unmap_and_free_vma; 18081da177e4SLinus Torvalds 1809309d08d9SLiu Zixian /* Can addr have changed?? 1810309d08d9SLiu Zixian * 1811309d08d9SLiu Zixian * Answer: Yes, several device drivers can do it in their 1812309d08d9SLiu Zixian * f_op->mmap method. -DaveM 1813309d08d9SLiu Zixian * Bug: If addr is changed, prev, rb_link, rb_parent should 1814309d08d9SLiu Zixian * be updated for vma_link() 1815309d08d9SLiu Zixian */ 1816309d08d9SLiu Zixian WARN_ON_ONCE(addr != vma->vm_start); 1817309d08d9SLiu Zixian 1818309d08d9SLiu Zixian addr = vma->vm_start; 1819309d08d9SLiu Zixian 1820d70cec89SMiaohe Lin /* If vm_flags changed after call_mmap(), we should try merge vma again 1821d70cec89SMiaohe Lin * as we may succeed this time. 1822d70cec89SMiaohe Lin */ 1823d70cec89SMiaohe Lin if (unlikely(vm_flags != vma->vm_flags && prev)) { 1824d70cec89SMiaohe Lin merge = vma_merge(mm, prev, vma->vm_start, vma->vm_end, vma->vm_flags, 18259a10064fSColin Cross NULL, vma->vm_file, vma->vm_pgoff, NULL, NULL_VM_UFFD_CTX, NULL); 1826d70cec89SMiaohe Lin if (merge) { 1827bc4fe4cdSMiaohe Lin /* ->mmap() can change vma->vm_file and fput the original file. So 1828bc4fe4cdSMiaohe Lin * fput the vma->vm_file here or we would add an extra fput for file 1829bc4fe4cdSMiaohe Lin * and cause general protection fault ultimately. 1830bc4fe4cdSMiaohe Lin */ 1831bc4fe4cdSMiaohe Lin fput(vma->vm_file); 1832d70cec89SMiaohe Lin vm_area_free(vma); 1833d70cec89SMiaohe Lin vma = merge; 1834309d08d9SLiu Zixian /* Update vm_flags to pick up the change. */ 1835d70cec89SMiaohe Lin vm_flags = vma->vm_flags; 1836d70cec89SMiaohe Lin goto unmap_writable; 1837d70cec89SMiaohe Lin } 1838d70cec89SMiaohe Lin } 1839d70cec89SMiaohe Lin 18401da177e4SLinus Torvalds vm_flags = vma->vm_flags; 1841f8dbf0a7SHuang Shijie } else if (vm_flags & VM_SHARED) { 1842f8dbf0a7SHuang Shijie error = shmem_zero_setup(vma); 1843f8dbf0a7SHuang Shijie if (error) 1844f8dbf0a7SHuang Shijie goto free_vma; 1845bfd40eafSKirill A. Shutemov } else { 1846bfd40eafSKirill A. Shutemov vma_set_anonymous(vma); 1847f8dbf0a7SHuang Shijie } 18481da177e4SLinus Torvalds 1849c462ac28SCatalin Marinas /* Allow architectures to sanity-check the vm_flags */ 1850c462ac28SCatalin Marinas if (!arch_validate_flags(vma->vm_flags)) { 1851c462ac28SCatalin Marinas error = -EINVAL; 1852c462ac28SCatalin Marinas if (file) 1853c462ac28SCatalin Marinas goto unmap_and_free_vma; 1854c462ac28SCatalin Marinas else 1855c462ac28SCatalin Marinas goto free_vma; 1856c462ac28SCatalin Marinas } 1857c462ac28SCatalin Marinas 18584d3d5b41SOleg Nesterov vma_link(mm, vma, prev, rb_link, rb_parent); 18594d3d5b41SOleg Nesterov /* Once vma denies write, undo our temporary denial count */ 1860d70cec89SMiaohe Lin unmap_writable: 18618d0920bdSDavid Hildenbrand if (file && vm_flags & VM_SHARED) 18624bb5f5d9SDavid Herrmann mapping_unmap_writable(file->f_mapping); 1863e8686772SOleg Nesterov file = vma->vm_file; 18641da177e4SLinus Torvalds out: 1865cdd6c482SIngo Molnar perf_event_mmap(vma); 18660a4a9391SPeter Zijlstra 186784638335SKonstantin Khlebnikov vm_stat_account(mm, vm_flags, len >> PAGE_SHIFT); 18681da177e4SLinus Torvalds if (vm_flags & VM_LOCKED) { 1869e1fb4a08SDave Jiang if ((vm_flags & VM_SPECIAL) || vma_is_dax(vma) || 1870e1fb4a08SDave Jiang is_vm_hugetlb_page(vma) || 1871e1fb4a08SDave Jiang vma == get_gate_vma(current->mm)) 1872de60f5f1SEric B Munson vma->vm_flags &= VM_LOCKED_CLEAR_MASK; 1873e1fb4a08SDave Jiang else 1874e1fb4a08SDave Jiang mm->locked_vm += (len >> PAGE_SHIFT); 1875bebeb3d6SMichel Lespinasse } 18762b144498SSrikar Dronamraju 1877c7a3a88cSOleg Nesterov if (file) 1878c7a3a88cSOleg Nesterov uprobe_mmap(vma); 18792b144498SSrikar Dronamraju 1880d9104d1cSCyrill Gorcunov /* 1881d9104d1cSCyrill Gorcunov * New (or expanded) vma always get soft dirty status. 1882d9104d1cSCyrill Gorcunov * Otherwise user-space soft-dirty page tracker won't 1883d9104d1cSCyrill Gorcunov * be able to distinguish situation when vma area unmapped, 1884d9104d1cSCyrill Gorcunov * then new mapped in-place (which must be aimed as 1885d9104d1cSCyrill Gorcunov * a completely new data area). 1886d9104d1cSCyrill Gorcunov */ 1887d9104d1cSCyrill Gorcunov vma->vm_flags |= VM_SOFTDIRTY; 1888d9104d1cSCyrill Gorcunov 188964e45507SPeter Feiner vma_set_page_prot(vma); 189064e45507SPeter Feiner 18911da177e4SLinus Torvalds return addr; 18921da177e4SLinus Torvalds 18931da177e4SLinus Torvalds unmap_and_free_vma: 18941527f926SChristian König fput(vma->vm_file); 18951da177e4SLinus Torvalds vma->vm_file = NULL; 18961da177e4SLinus Torvalds 18971da177e4SLinus Torvalds /* Undo any partial mapping done by a device driver. */ 1898e0da382cSHugh Dickins unmap_region(mm, vma, prev, vma->vm_start, vma->vm_end); 1899e0da382cSHugh Dickins charged = 0; 19004bb5f5d9SDavid Herrmann if (vm_flags & VM_SHARED) 19014bb5f5d9SDavid Herrmann mapping_unmap_writable(file->f_mapping); 19021da177e4SLinus Torvalds free_vma: 19033928d4f5SLinus Torvalds vm_area_free(vma); 19041da177e4SLinus Torvalds unacct_error: 19051da177e4SLinus Torvalds if (charged) 19061da177e4SLinus Torvalds vm_unacct_memory(charged); 19071da177e4SLinus Torvalds return error; 19081da177e4SLinus Torvalds } 19091da177e4SLinus Torvalds 1910baceaf1cSJaewon Kim static unsigned long unmapped_area(struct vm_unmapped_area_info *info) 1911db4fbfb9SMichel Lespinasse { 1912db4fbfb9SMichel Lespinasse /* 1913db4fbfb9SMichel Lespinasse * We implement the search by looking for an rbtree node that 1914db4fbfb9SMichel Lespinasse * immediately follows a suitable gap. That is, 1915db4fbfb9SMichel Lespinasse * - gap_start = vma->vm_prev->vm_end <= info->high_limit - length; 1916db4fbfb9SMichel Lespinasse * - gap_end = vma->vm_start >= info->low_limit + length; 1917db4fbfb9SMichel Lespinasse * - gap_end - gap_start >= length 1918db4fbfb9SMichel Lespinasse */ 1919db4fbfb9SMichel Lespinasse 1920db4fbfb9SMichel Lespinasse struct mm_struct *mm = current->mm; 1921db4fbfb9SMichel Lespinasse struct vm_area_struct *vma; 1922db4fbfb9SMichel Lespinasse unsigned long length, low_limit, high_limit, gap_start, gap_end; 1923db4fbfb9SMichel Lespinasse 1924db4fbfb9SMichel Lespinasse /* Adjust search length to account for worst case alignment overhead */ 1925db4fbfb9SMichel Lespinasse length = info->length + info->align_mask; 1926db4fbfb9SMichel Lespinasse if (length < info->length) 1927db4fbfb9SMichel Lespinasse return -ENOMEM; 1928db4fbfb9SMichel Lespinasse 1929db4fbfb9SMichel Lespinasse /* Adjust search limits by the desired length */ 1930db4fbfb9SMichel Lespinasse if (info->high_limit < length) 1931db4fbfb9SMichel Lespinasse return -ENOMEM; 1932db4fbfb9SMichel Lespinasse high_limit = info->high_limit - length; 1933db4fbfb9SMichel Lespinasse 1934db4fbfb9SMichel Lespinasse if (info->low_limit > high_limit) 1935db4fbfb9SMichel Lespinasse return -ENOMEM; 1936db4fbfb9SMichel Lespinasse low_limit = info->low_limit + length; 1937db4fbfb9SMichel Lespinasse 1938db4fbfb9SMichel Lespinasse /* Check if rbtree root looks promising */ 1939db4fbfb9SMichel Lespinasse if (RB_EMPTY_ROOT(&mm->mm_rb)) 1940db4fbfb9SMichel Lespinasse goto check_highest; 1941db4fbfb9SMichel Lespinasse vma = rb_entry(mm->mm_rb.rb_node, struct vm_area_struct, vm_rb); 1942db4fbfb9SMichel Lespinasse if (vma->rb_subtree_gap < length) 1943db4fbfb9SMichel Lespinasse goto check_highest; 1944db4fbfb9SMichel Lespinasse 1945db4fbfb9SMichel Lespinasse while (true) { 1946db4fbfb9SMichel Lespinasse /* Visit left subtree if it looks promising */ 19471be7107fSHugh Dickins gap_end = vm_start_gap(vma); 1948db4fbfb9SMichel Lespinasse if (gap_end >= low_limit && vma->vm_rb.rb_left) { 1949db4fbfb9SMichel Lespinasse struct vm_area_struct *left = 1950db4fbfb9SMichel Lespinasse rb_entry(vma->vm_rb.rb_left, 1951db4fbfb9SMichel Lespinasse struct vm_area_struct, vm_rb); 1952db4fbfb9SMichel Lespinasse if (left->rb_subtree_gap >= length) { 1953db4fbfb9SMichel Lespinasse vma = left; 1954db4fbfb9SMichel Lespinasse continue; 1955db4fbfb9SMichel Lespinasse } 1956db4fbfb9SMichel Lespinasse } 1957db4fbfb9SMichel Lespinasse 19581be7107fSHugh Dickins gap_start = vma->vm_prev ? vm_end_gap(vma->vm_prev) : 0; 1959db4fbfb9SMichel Lespinasse check_current: 1960db4fbfb9SMichel Lespinasse /* Check if current node has a suitable gap */ 1961db4fbfb9SMichel Lespinasse if (gap_start > high_limit) 1962db4fbfb9SMichel Lespinasse return -ENOMEM; 1963f4cb767dSHugh Dickins if (gap_end >= low_limit && 1964f4cb767dSHugh Dickins gap_end > gap_start && gap_end - gap_start >= length) 1965db4fbfb9SMichel Lespinasse goto found; 1966db4fbfb9SMichel Lespinasse 1967db4fbfb9SMichel Lespinasse /* Visit right subtree if it looks promising */ 1968db4fbfb9SMichel Lespinasse if (vma->vm_rb.rb_right) { 1969db4fbfb9SMichel Lespinasse struct vm_area_struct *right = 1970db4fbfb9SMichel Lespinasse rb_entry(vma->vm_rb.rb_right, 1971db4fbfb9SMichel Lespinasse struct vm_area_struct, vm_rb); 1972db4fbfb9SMichel Lespinasse if (right->rb_subtree_gap >= length) { 1973db4fbfb9SMichel Lespinasse vma = right; 1974db4fbfb9SMichel Lespinasse continue; 1975db4fbfb9SMichel Lespinasse } 1976db4fbfb9SMichel Lespinasse } 1977db4fbfb9SMichel Lespinasse 1978db4fbfb9SMichel Lespinasse /* Go back up the rbtree to find next candidate node */ 1979db4fbfb9SMichel Lespinasse while (true) { 1980db4fbfb9SMichel Lespinasse struct rb_node *prev = &vma->vm_rb; 1981db4fbfb9SMichel Lespinasse if (!rb_parent(prev)) 1982db4fbfb9SMichel Lespinasse goto check_highest; 1983db4fbfb9SMichel Lespinasse vma = rb_entry(rb_parent(prev), 1984db4fbfb9SMichel Lespinasse struct vm_area_struct, vm_rb); 1985db4fbfb9SMichel Lespinasse if (prev == vma->vm_rb.rb_left) { 19861be7107fSHugh Dickins gap_start = vm_end_gap(vma->vm_prev); 19871be7107fSHugh Dickins gap_end = vm_start_gap(vma); 1988db4fbfb9SMichel Lespinasse goto check_current; 1989db4fbfb9SMichel Lespinasse } 1990db4fbfb9SMichel Lespinasse } 1991db4fbfb9SMichel Lespinasse } 1992db4fbfb9SMichel Lespinasse 1993db4fbfb9SMichel Lespinasse check_highest: 1994db4fbfb9SMichel Lespinasse /* Check highest gap, which does not precede any rbtree node */ 1995db4fbfb9SMichel Lespinasse gap_start = mm->highest_vm_end; 1996db4fbfb9SMichel Lespinasse gap_end = ULONG_MAX; /* Only for VM_BUG_ON below */ 1997db4fbfb9SMichel Lespinasse if (gap_start > high_limit) 1998db4fbfb9SMichel Lespinasse return -ENOMEM; 1999db4fbfb9SMichel Lespinasse 2000db4fbfb9SMichel Lespinasse found: 2001db4fbfb9SMichel Lespinasse /* We found a suitable gap. Clip it with the original low_limit. */ 2002db4fbfb9SMichel Lespinasse if (gap_start < info->low_limit) 2003db4fbfb9SMichel Lespinasse gap_start = info->low_limit; 2004db4fbfb9SMichel Lespinasse 2005db4fbfb9SMichel Lespinasse /* Adjust gap address to the desired alignment */ 2006db4fbfb9SMichel Lespinasse gap_start += (info->align_offset - gap_start) & info->align_mask; 2007db4fbfb9SMichel Lespinasse 2008db4fbfb9SMichel Lespinasse VM_BUG_ON(gap_start + info->length > info->high_limit); 2009db4fbfb9SMichel Lespinasse VM_BUG_ON(gap_start + info->length > gap_end); 2010db4fbfb9SMichel Lespinasse return gap_start; 2011db4fbfb9SMichel Lespinasse } 2012db4fbfb9SMichel Lespinasse 2013baceaf1cSJaewon Kim static unsigned long unmapped_area_topdown(struct vm_unmapped_area_info *info) 2014db4fbfb9SMichel Lespinasse { 2015db4fbfb9SMichel Lespinasse struct mm_struct *mm = current->mm; 2016db4fbfb9SMichel Lespinasse struct vm_area_struct *vma; 2017db4fbfb9SMichel Lespinasse unsigned long length, low_limit, high_limit, gap_start, gap_end; 2018db4fbfb9SMichel Lespinasse 2019db4fbfb9SMichel Lespinasse /* Adjust search length to account for worst case alignment overhead */ 2020db4fbfb9SMichel Lespinasse length = info->length + info->align_mask; 2021db4fbfb9SMichel Lespinasse if (length < info->length) 2022db4fbfb9SMichel Lespinasse return -ENOMEM; 2023db4fbfb9SMichel Lespinasse 2024db4fbfb9SMichel Lespinasse /* 2025db4fbfb9SMichel Lespinasse * Adjust search limits by the desired length. 2026db4fbfb9SMichel Lespinasse * See implementation comment at top of unmapped_area(). 2027db4fbfb9SMichel Lespinasse */ 2028db4fbfb9SMichel Lespinasse gap_end = info->high_limit; 2029db4fbfb9SMichel Lespinasse if (gap_end < length) 2030db4fbfb9SMichel Lespinasse return -ENOMEM; 2031db4fbfb9SMichel Lespinasse high_limit = gap_end - length; 2032db4fbfb9SMichel Lespinasse 2033db4fbfb9SMichel Lespinasse if (info->low_limit > high_limit) 2034db4fbfb9SMichel Lespinasse return -ENOMEM; 2035db4fbfb9SMichel Lespinasse low_limit = info->low_limit + length; 2036db4fbfb9SMichel Lespinasse 2037db4fbfb9SMichel Lespinasse /* Check highest gap, which does not precede any rbtree node */ 2038db4fbfb9SMichel Lespinasse gap_start = mm->highest_vm_end; 2039db4fbfb9SMichel Lespinasse if (gap_start <= high_limit) 2040db4fbfb9SMichel Lespinasse goto found_highest; 2041db4fbfb9SMichel Lespinasse 2042db4fbfb9SMichel Lespinasse /* Check if rbtree root looks promising */ 2043db4fbfb9SMichel Lespinasse if (RB_EMPTY_ROOT(&mm->mm_rb)) 2044db4fbfb9SMichel Lespinasse return -ENOMEM; 2045db4fbfb9SMichel Lespinasse vma = rb_entry(mm->mm_rb.rb_node, struct vm_area_struct, vm_rb); 2046db4fbfb9SMichel Lespinasse if (vma->rb_subtree_gap < length) 2047db4fbfb9SMichel Lespinasse return -ENOMEM; 2048db4fbfb9SMichel Lespinasse 2049db4fbfb9SMichel Lespinasse while (true) { 2050db4fbfb9SMichel Lespinasse /* Visit right subtree if it looks promising */ 20511be7107fSHugh Dickins gap_start = vma->vm_prev ? vm_end_gap(vma->vm_prev) : 0; 2052db4fbfb9SMichel Lespinasse if (gap_start <= high_limit && vma->vm_rb.rb_right) { 2053db4fbfb9SMichel Lespinasse struct vm_area_struct *right = 2054db4fbfb9SMichel Lespinasse rb_entry(vma->vm_rb.rb_right, 2055db4fbfb9SMichel Lespinasse struct vm_area_struct, vm_rb); 2056db4fbfb9SMichel Lespinasse if (right->rb_subtree_gap >= length) { 2057db4fbfb9SMichel Lespinasse vma = right; 2058db4fbfb9SMichel Lespinasse continue; 2059db4fbfb9SMichel Lespinasse } 2060db4fbfb9SMichel Lespinasse } 2061db4fbfb9SMichel Lespinasse 2062db4fbfb9SMichel Lespinasse check_current: 2063db4fbfb9SMichel Lespinasse /* Check if current node has a suitable gap */ 20641be7107fSHugh Dickins gap_end = vm_start_gap(vma); 2065db4fbfb9SMichel Lespinasse if (gap_end < low_limit) 2066db4fbfb9SMichel Lespinasse return -ENOMEM; 2067f4cb767dSHugh Dickins if (gap_start <= high_limit && 2068f4cb767dSHugh Dickins gap_end > gap_start && gap_end - gap_start >= length) 2069db4fbfb9SMichel Lespinasse goto found; 2070db4fbfb9SMichel Lespinasse 2071db4fbfb9SMichel Lespinasse /* Visit left subtree if it looks promising */ 2072db4fbfb9SMichel Lespinasse if (vma->vm_rb.rb_left) { 2073db4fbfb9SMichel Lespinasse struct vm_area_struct *left = 2074db4fbfb9SMichel Lespinasse rb_entry(vma->vm_rb.rb_left, 2075db4fbfb9SMichel Lespinasse struct vm_area_struct, vm_rb); 2076db4fbfb9SMichel Lespinasse if (left->rb_subtree_gap >= length) { 2077db4fbfb9SMichel Lespinasse vma = left; 2078db4fbfb9SMichel Lespinasse continue; 2079db4fbfb9SMichel Lespinasse } 2080db4fbfb9SMichel Lespinasse } 2081db4fbfb9SMichel Lespinasse 2082db4fbfb9SMichel Lespinasse /* Go back up the rbtree to find next candidate node */ 2083db4fbfb9SMichel Lespinasse while (true) { 2084db4fbfb9SMichel Lespinasse struct rb_node *prev = &vma->vm_rb; 2085db4fbfb9SMichel Lespinasse if (!rb_parent(prev)) 2086db4fbfb9SMichel Lespinasse return -ENOMEM; 2087db4fbfb9SMichel Lespinasse vma = rb_entry(rb_parent(prev), 2088db4fbfb9SMichel Lespinasse struct vm_area_struct, vm_rb); 2089db4fbfb9SMichel Lespinasse if (prev == vma->vm_rb.rb_right) { 2090db4fbfb9SMichel Lespinasse gap_start = vma->vm_prev ? 20911be7107fSHugh Dickins vm_end_gap(vma->vm_prev) : 0; 2092db4fbfb9SMichel Lespinasse goto check_current; 2093db4fbfb9SMichel Lespinasse } 2094db4fbfb9SMichel Lespinasse } 2095db4fbfb9SMichel Lespinasse } 2096db4fbfb9SMichel Lespinasse 2097db4fbfb9SMichel Lespinasse found: 2098db4fbfb9SMichel Lespinasse /* We found a suitable gap. Clip it with the original high_limit. */ 2099db4fbfb9SMichel Lespinasse if (gap_end > info->high_limit) 2100db4fbfb9SMichel Lespinasse gap_end = info->high_limit; 2101db4fbfb9SMichel Lespinasse 2102db4fbfb9SMichel Lespinasse found_highest: 2103db4fbfb9SMichel Lespinasse /* Compute highest gap address at the desired alignment */ 2104db4fbfb9SMichel Lespinasse gap_end -= info->length; 2105db4fbfb9SMichel Lespinasse gap_end -= (gap_end - info->align_offset) & info->align_mask; 2106db4fbfb9SMichel Lespinasse 2107db4fbfb9SMichel Lespinasse VM_BUG_ON(gap_end < info->low_limit); 2108db4fbfb9SMichel Lespinasse VM_BUG_ON(gap_end < gap_start); 2109db4fbfb9SMichel Lespinasse return gap_end; 2110db4fbfb9SMichel Lespinasse } 2111db4fbfb9SMichel Lespinasse 2112baceaf1cSJaewon Kim /* 2113baceaf1cSJaewon Kim * Search for an unmapped address range. 2114baceaf1cSJaewon Kim * 2115baceaf1cSJaewon Kim * We are looking for a range that: 2116baceaf1cSJaewon Kim * - does not intersect with any VMA; 2117baceaf1cSJaewon Kim * - is contained within the [low_limit, high_limit) interval; 2118baceaf1cSJaewon Kim * - is at least the desired size. 2119baceaf1cSJaewon Kim * - satisfies (begin_addr & align_mask) == (align_offset & align_mask) 2120baceaf1cSJaewon Kim */ 2121baceaf1cSJaewon Kim unsigned long vm_unmapped_area(struct vm_unmapped_area_info *info) 2122baceaf1cSJaewon Kim { 2123df529cabSJaewon Kim unsigned long addr; 2124df529cabSJaewon Kim 2125baceaf1cSJaewon Kim if (info->flags & VM_UNMAPPED_AREA_TOPDOWN) 2126df529cabSJaewon Kim addr = unmapped_area_topdown(info); 2127baceaf1cSJaewon Kim else 2128df529cabSJaewon Kim addr = unmapped_area(info); 2129df529cabSJaewon Kim 2130df529cabSJaewon Kim trace_vm_unmapped_area(addr, info); 2131df529cabSJaewon Kim return addr; 2132baceaf1cSJaewon Kim } 2133f6795053SSteve Capper 21341da177e4SLinus Torvalds /* Get an address range which is currently unmapped. 21351da177e4SLinus Torvalds * For shmat() with addr=0. 21361da177e4SLinus Torvalds * 21371da177e4SLinus Torvalds * Ugly calling convention alert: 21381da177e4SLinus Torvalds * Return value with the low bits set means error value, 21391da177e4SLinus Torvalds * ie 21401da177e4SLinus Torvalds * if (ret & ~PAGE_MASK) 21411da177e4SLinus Torvalds * error = ret; 21421da177e4SLinus Torvalds * 21431da177e4SLinus Torvalds * This function "knows" that -ENOMEM has the bits set. 21441da177e4SLinus Torvalds */ 21451da177e4SLinus Torvalds #ifndef HAVE_ARCH_UNMAPPED_AREA 21461da177e4SLinus Torvalds unsigned long 21471da177e4SLinus Torvalds arch_get_unmapped_area(struct file *filp, unsigned long addr, 21481da177e4SLinus Torvalds unsigned long len, unsigned long pgoff, unsigned long flags) 21491da177e4SLinus Torvalds { 21501da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 21511be7107fSHugh Dickins struct vm_area_struct *vma, *prev; 2152db4fbfb9SMichel Lespinasse struct vm_unmapped_area_info info; 2153f6795053SSteve Capper const unsigned long mmap_end = arch_get_mmap_end(addr); 21541da177e4SLinus Torvalds 2155f6795053SSteve Capper if (len > mmap_end - mmap_min_addr) 21561da177e4SLinus Torvalds return -ENOMEM; 21571da177e4SLinus Torvalds 215806abdfb4SBenjamin Herrenschmidt if (flags & MAP_FIXED) 215906abdfb4SBenjamin Herrenschmidt return addr; 216006abdfb4SBenjamin Herrenschmidt 21611da177e4SLinus Torvalds if (addr) { 21621da177e4SLinus Torvalds addr = PAGE_ALIGN(addr); 21631be7107fSHugh Dickins vma = find_vma_prev(mm, addr, &prev); 2164f6795053SSteve Capper if (mmap_end - len >= addr && addr >= mmap_min_addr && 21651be7107fSHugh Dickins (!vma || addr + len <= vm_start_gap(vma)) && 21661be7107fSHugh Dickins (!prev || addr >= vm_end_gap(prev))) 21671da177e4SLinus Torvalds return addr; 21681da177e4SLinus Torvalds } 21691da177e4SLinus Torvalds 2170db4fbfb9SMichel Lespinasse info.flags = 0; 2171db4fbfb9SMichel Lespinasse info.length = len; 21724e99b021SHeiko Carstens info.low_limit = mm->mmap_base; 2173f6795053SSteve Capper info.high_limit = mmap_end; 2174db4fbfb9SMichel Lespinasse info.align_mask = 0; 217509ef5283SJaewon Kim info.align_offset = 0; 2176db4fbfb9SMichel Lespinasse return vm_unmapped_area(&info); 21771da177e4SLinus Torvalds } 21781da177e4SLinus Torvalds #endif 21791da177e4SLinus Torvalds 21801da177e4SLinus Torvalds /* 21811da177e4SLinus Torvalds * This mmap-allocator allocates new areas top-down from below the 21821da177e4SLinus Torvalds * stack's low limit (the base): 21831da177e4SLinus Torvalds */ 21841da177e4SLinus Torvalds #ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN 21851da177e4SLinus Torvalds unsigned long 218643cca0b1SYang Fan arch_get_unmapped_area_topdown(struct file *filp, unsigned long addr, 218743cca0b1SYang Fan unsigned long len, unsigned long pgoff, 218843cca0b1SYang Fan unsigned long flags) 21891da177e4SLinus Torvalds { 21901be7107fSHugh Dickins struct vm_area_struct *vma, *prev; 21911da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 2192db4fbfb9SMichel Lespinasse struct vm_unmapped_area_info info; 2193f6795053SSteve Capper const unsigned long mmap_end = arch_get_mmap_end(addr); 21941da177e4SLinus Torvalds 21951da177e4SLinus Torvalds /* requested length too big for entire address space */ 2196f6795053SSteve Capper if (len > mmap_end - mmap_min_addr) 21971da177e4SLinus Torvalds return -ENOMEM; 21981da177e4SLinus Torvalds 219906abdfb4SBenjamin Herrenschmidt if (flags & MAP_FIXED) 220006abdfb4SBenjamin Herrenschmidt return addr; 220106abdfb4SBenjamin Herrenschmidt 22021da177e4SLinus Torvalds /* requesting a specific address */ 22031da177e4SLinus Torvalds if (addr) { 22041da177e4SLinus Torvalds addr = PAGE_ALIGN(addr); 22051be7107fSHugh Dickins vma = find_vma_prev(mm, addr, &prev); 2206f6795053SSteve Capper if (mmap_end - len >= addr && addr >= mmap_min_addr && 22071be7107fSHugh Dickins (!vma || addr + len <= vm_start_gap(vma)) && 22081be7107fSHugh Dickins (!prev || addr >= vm_end_gap(prev))) 22091da177e4SLinus Torvalds return addr; 22101da177e4SLinus Torvalds } 22111da177e4SLinus Torvalds 2212db4fbfb9SMichel Lespinasse info.flags = VM_UNMAPPED_AREA_TOPDOWN; 2213db4fbfb9SMichel Lespinasse info.length = len; 22142afc745fSAkira Takeuchi info.low_limit = max(PAGE_SIZE, mmap_min_addr); 2215f6795053SSteve Capper info.high_limit = arch_get_mmap_base(addr, mm->mmap_base); 2216db4fbfb9SMichel Lespinasse info.align_mask = 0; 221709ef5283SJaewon Kim info.align_offset = 0; 2218db4fbfb9SMichel Lespinasse addr = vm_unmapped_area(&info); 2219b716ad95SXiao Guangrong 22201da177e4SLinus Torvalds /* 22211da177e4SLinus Torvalds * A failed mmap() very likely causes application failure, 22221da177e4SLinus Torvalds * so fall back to the bottom-up function here. This scenario 22231da177e4SLinus Torvalds * can happen with large stack limits and large mmap() 22241da177e4SLinus Torvalds * allocations. 22251da177e4SLinus Torvalds */ 2226de1741a1SAlexander Kuleshov if (offset_in_page(addr)) { 2227db4fbfb9SMichel Lespinasse VM_BUG_ON(addr != -ENOMEM); 2228db4fbfb9SMichel Lespinasse info.flags = 0; 2229db4fbfb9SMichel Lespinasse info.low_limit = TASK_UNMAPPED_BASE; 2230f6795053SSteve Capper info.high_limit = mmap_end; 2231db4fbfb9SMichel Lespinasse addr = vm_unmapped_area(&info); 2232db4fbfb9SMichel Lespinasse } 22331da177e4SLinus Torvalds 22341da177e4SLinus Torvalds return addr; 22351da177e4SLinus Torvalds } 22361da177e4SLinus Torvalds #endif 22371da177e4SLinus Torvalds 22381da177e4SLinus Torvalds unsigned long 22391da177e4SLinus Torvalds get_unmapped_area(struct file *file, unsigned long addr, unsigned long len, 22401da177e4SLinus Torvalds unsigned long pgoff, unsigned long flags) 22411da177e4SLinus Torvalds { 224206abdfb4SBenjamin Herrenschmidt unsigned long (*get_area)(struct file *, unsigned long, 224306abdfb4SBenjamin Herrenschmidt unsigned long, unsigned long, unsigned long); 224407ab67c8SLinus Torvalds 22459206de95SAl Viro unsigned long error = arch_mmap_check(addr, len, flags); 22469206de95SAl Viro if (error) 22479206de95SAl Viro return error; 22489206de95SAl Viro 22499206de95SAl Viro /* Careful about overflows.. */ 22509206de95SAl Viro if (len > TASK_SIZE) 22519206de95SAl Viro return -ENOMEM; 22529206de95SAl Viro 225307ab67c8SLinus Torvalds get_area = current->mm->get_unmapped_area; 2254c01d5b30SHugh Dickins if (file) { 2255c01d5b30SHugh Dickins if (file->f_op->get_unmapped_area) 225607ab67c8SLinus Torvalds get_area = file->f_op->get_unmapped_area; 2257c01d5b30SHugh Dickins } else if (flags & MAP_SHARED) { 2258c01d5b30SHugh Dickins /* 2259c01d5b30SHugh Dickins * mmap_region() will call shmem_zero_setup() to create a file, 2260c01d5b30SHugh Dickins * so use shmem's get_unmapped_area in case it can be huge. 226145e55300SPeter Collingbourne * do_mmap() will clear pgoff, so match alignment. 2262c01d5b30SHugh Dickins */ 2263c01d5b30SHugh Dickins pgoff = 0; 2264c01d5b30SHugh Dickins get_area = shmem_get_unmapped_area; 2265c01d5b30SHugh Dickins } 2266c01d5b30SHugh Dickins 226707ab67c8SLinus Torvalds addr = get_area(file, addr, len, pgoff, flags); 226807ab67c8SLinus Torvalds if (IS_ERR_VALUE(addr)) 226907ab67c8SLinus Torvalds return addr; 227007ab67c8SLinus Torvalds 22711da177e4SLinus Torvalds if (addr > TASK_SIZE - len) 22721da177e4SLinus Torvalds return -ENOMEM; 2273de1741a1SAlexander Kuleshov if (offset_in_page(addr)) 22741da177e4SLinus Torvalds return -EINVAL; 227506abdfb4SBenjamin Herrenschmidt 22769ac4ed4bSAl Viro error = security_mmap_addr(addr); 22779ac4ed4bSAl Viro return error ? error : addr; 22781da177e4SLinus Torvalds } 22791da177e4SLinus Torvalds 22801da177e4SLinus Torvalds EXPORT_SYMBOL(get_unmapped_area); 22811da177e4SLinus Torvalds 22821da177e4SLinus Torvalds /* Look up the first VMA which satisfies addr < vm_end, NULL if none. */ 22831da177e4SLinus Torvalds struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr) 22841da177e4SLinus Torvalds { 2285615d6e87SDavidlohr Bueso struct rb_node *rb_node; 2286615d6e87SDavidlohr Bueso struct vm_area_struct *vma; 22871da177e4SLinus Torvalds 22885b78ed24SLuigi Rizzo mmap_assert_locked(mm); 22891da177e4SLinus Torvalds /* Check the cache first. */ 2290615d6e87SDavidlohr Bueso vma = vmacache_find(mm, addr); 2291615d6e87SDavidlohr Bueso if (likely(vma)) 2292615d6e87SDavidlohr Bueso return vma; 22931da177e4SLinus Torvalds 22941da177e4SLinus Torvalds rb_node = mm->mm_rb.rb_node; 22951da177e4SLinus Torvalds 22961da177e4SLinus Torvalds while (rb_node) { 2297615d6e87SDavidlohr Bueso struct vm_area_struct *tmp; 22981da177e4SLinus Torvalds 2299615d6e87SDavidlohr Bueso tmp = rb_entry(rb_node, struct vm_area_struct, vm_rb); 23001da177e4SLinus Torvalds 2301615d6e87SDavidlohr Bueso if (tmp->vm_end > addr) { 2302615d6e87SDavidlohr Bueso vma = tmp; 2303615d6e87SDavidlohr Bueso if (tmp->vm_start <= addr) 23041da177e4SLinus Torvalds break; 23051da177e4SLinus Torvalds rb_node = rb_node->rb_left; 23061da177e4SLinus Torvalds } else 23071da177e4SLinus Torvalds rb_node = rb_node->rb_right; 23081da177e4SLinus Torvalds } 2309615d6e87SDavidlohr Bueso 23101da177e4SLinus Torvalds if (vma) 2311615d6e87SDavidlohr Bueso vmacache_update(addr, vma); 23121da177e4SLinus Torvalds return vma; 23131da177e4SLinus Torvalds } 23141da177e4SLinus Torvalds 23151da177e4SLinus Torvalds EXPORT_SYMBOL(find_vma); 23161da177e4SLinus Torvalds 23176bd4837dSKOSAKI Motohiro /* 23186bd4837dSKOSAKI Motohiro * Same as find_vma, but also return a pointer to the previous VMA in *pprev. 23196bd4837dSKOSAKI Motohiro */ 23201da177e4SLinus Torvalds struct vm_area_struct * 23211da177e4SLinus Torvalds find_vma_prev(struct mm_struct *mm, unsigned long addr, 23221da177e4SLinus Torvalds struct vm_area_struct **pprev) 23231da177e4SLinus Torvalds { 23246bd4837dSKOSAKI Motohiro struct vm_area_struct *vma; 23251da177e4SLinus Torvalds 23266bd4837dSKOSAKI Motohiro vma = find_vma(mm, addr); 232783cd904dSMikulas Patocka if (vma) { 232883cd904dSMikulas Patocka *pprev = vma->vm_prev; 232983cd904dSMikulas Patocka } else { 233073848a97SWei Yang struct rb_node *rb_node = rb_last(&mm->mm_rb); 233173848a97SWei Yang 233273848a97SWei Yang *pprev = rb_node ? rb_entry(rb_node, struct vm_area_struct, vm_rb) : NULL; 233383cd904dSMikulas Patocka } 23346bd4837dSKOSAKI Motohiro return vma; 23351da177e4SLinus Torvalds } 23361da177e4SLinus Torvalds 23371da177e4SLinus Torvalds /* 23381da177e4SLinus Torvalds * Verify that the stack growth is acceptable and 23391da177e4SLinus Torvalds * update accounting. This is shared with both the 23401da177e4SLinus Torvalds * grow-up and grow-down cases. 23411da177e4SLinus Torvalds */ 23421be7107fSHugh Dickins static int acct_stack_growth(struct vm_area_struct *vma, 23431be7107fSHugh Dickins unsigned long size, unsigned long grow) 23441da177e4SLinus Torvalds { 23451da177e4SLinus Torvalds struct mm_struct *mm = vma->vm_mm; 23461be7107fSHugh Dickins unsigned long new_start; 23471da177e4SLinus Torvalds 23481da177e4SLinus Torvalds /* address space limit tests */ 234984638335SKonstantin Khlebnikov if (!may_expand_vm(mm, vma->vm_flags, grow)) 23501da177e4SLinus Torvalds return -ENOMEM; 23511da177e4SLinus Torvalds 23521da177e4SLinus Torvalds /* Stack limit test */ 235324c79d8eSKrzysztof Opasiak if (size > rlimit(RLIMIT_STACK)) 23541da177e4SLinus Torvalds return -ENOMEM; 23551da177e4SLinus Torvalds 23561da177e4SLinus Torvalds /* mlock limit tests */ 2357*c5d8a364SMiaohe Lin if (mlock_future_check(mm, vma->vm_flags, grow << PAGE_SHIFT)) 23581da177e4SLinus Torvalds return -ENOMEM; 23591da177e4SLinus Torvalds 23600d59a01bSAdam Litke /* Check to ensure the stack will not grow into a hugetlb-only region */ 23610d59a01bSAdam Litke new_start = (vma->vm_flags & VM_GROWSUP) ? vma->vm_start : 23620d59a01bSAdam Litke vma->vm_end - size; 23630d59a01bSAdam Litke if (is_hugepage_only_range(vma->vm_mm, new_start, size)) 23640d59a01bSAdam Litke return -EFAULT; 23650d59a01bSAdam Litke 23661da177e4SLinus Torvalds /* 23671da177e4SLinus Torvalds * Overcommit.. This must be the final test, as it will 23681da177e4SLinus Torvalds * update security statistics. 23691da177e4SLinus Torvalds */ 237005fa199dSHugh Dickins if (security_vm_enough_memory_mm(mm, grow)) 23711da177e4SLinus Torvalds return -ENOMEM; 23721da177e4SLinus Torvalds 23731da177e4SLinus Torvalds return 0; 23741da177e4SLinus Torvalds } 23751da177e4SLinus Torvalds 237646dea3d0SHugh Dickins #if defined(CONFIG_STACK_GROWSUP) || defined(CONFIG_IA64) 23771da177e4SLinus Torvalds /* 237846dea3d0SHugh Dickins * PA-RISC uses this for its stack; IA64 for its Register Backing Store. 237946dea3d0SHugh Dickins * vma is the last one with address > vma->vm_end. Have to extend vma. 23801da177e4SLinus Torvalds */ 238146dea3d0SHugh Dickins int expand_upwards(struct vm_area_struct *vma, unsigned long address) 23821da177e4SLinus Torvalds { 238309357814SOleg Nesterov struct mm_struct *mm = vma->vm_mm; 23841be7107fSHugh Dickins struct vm_area_struct *next; 23851be7107fSHugh Dickins unsigned long gap_addr; 238612352d3cSKonstantin Khlebnikov int error = 0; 23871da177e4SLinus Torvalds 23881da177e4SLinus Torvalds if (!(vma->vm_flags & VM_GROWSUP)) 23891da177e4SLinus Torvalds return -EFAULT; 23901da177e4SLinus Torvalds 2391bd726c90SHelge Deller /* Guard against exceeding limits of the address space. */ 23921be7107fSHugh Dickins address &= PAGE_MASK; 239337511fb5SHelge Deller if (address >= (TASK_SIZE & PAGE_MASK)) 239412352d3cSKonstantin Khlebnikov return -ENOMEM; 2395bd726c90SHelge Deller address += PAGE_SIZE; 239612352d3cSKonstantin Khlebnikov 23971be7107fSHugh Dickins /* Enforce stack_guard_gap */ 23981be7107fSHugh Dickins gap_addr = address + stack_guard_gap; 2399bd726c90SHelge Deller 2400bd726c90SHelge Deller /* Guard against overflow */ 2401bd726c90SHelge Deller if (gap_addr < address || gap_addr > TASK_SIZE) 2402bd726c90SHelge Deller gap_addr = TASK_SIZE; 2403bd726c90SHelge Deller 24041be7107fSHugh Dickins next = vma->vm_next; 24053122e80eSAnshuman Khandual if (next && next->vm_start < gap_addr && vma_is_accessible(next)) { 24061be7107fSHugh Dickins if (!(next->vm_flags & VM_GROWSUP)) 24071be7107fSHugh Dickins return -ENOMEM; 24081be7107fSHugh Dickins /* Check that both stack segments have the same anon_vma? */ 24091be7107fSHugh Dickins } 24101be7107fSHugh Dickins 241112352d3cSKonstantin Khlebnikov /* We must make sure the anon_vma is allocated. */ 24121da177e4SLinus Torvalds if (unlikely(anon_vma_prepare(vma))) 24131da177e4SLinus Torvalds return -ENOMEM; 24141da177e4SLinus Torvalds 24151da177e4SLinus Torvalds /* 24161da177e4SLinus Torvalds * vma->vm_start/vm_end cannot change under us because the caller 2417c1e8d7c6SMichel Lespinasse * is required to hold the mmap_lock in read mode. We need the 24181da177e4SLinus Torvalds * anon_vma lock to serialize against concurrent expand_stacks. 24191da177e4SLinus Torvalds */ 242012352d3cSKonstantin Khlebnikov anon_vma_lock_write(vma->anon_vma); 24211da177e4SLinus Torvalds 24221da177e4SLinus Torvalds /* Somebody else might have raced and expanded it already */ 24231da177e4SLinus Torvalds if (address > vma->vm_end) { 24241da177e4SLinus Torvalds unsigned long size, grow; 24251da177e4SLinus Torvalds 24261da177e4SLinus Torvalds size = address - vma->vm_start; 24271da177e4SLinus Torvalds grow = (address - vma->vm_end) >> PAGE_SHIFT; 24281da177e4SLinus Torvalds 242942c36f63SHugh Dickins error = -ENOMEM; 243042c36f63SHugh Dickins if (vma->vm_pgoff + (size >> PAGE_SHIFT) >= vma->vm_pgoff) { 24311da177e4SLinus Torvalds error = acct_stack_growth(vma, size, grow); 24323af9e859SEric B Munson if (!error) { 24334128997bSMichel Lespinasse /* 24344128997bSMichel Lespinasse * vma_gap_update() doesn't support concurrent 2435c1e8d7c6SMichel Lespinasse * updates, but we only hold a shared mmap_lock 24364128997bSMichel Lespinasse * lock here, so we need to protect against 24374128997bSMichel Lespinasse * concurrent vma expansions. 243812352d3cSKonstantin Khlebnikov * anon_vma_lock_write() doesn't help here, as 24394128997bSMichel Lespinasse * we don't guarantee that all growable vmas 24404128997bSMichel Lespinasse * in a mm share the same root anon vma. 24414128997bSMichel Lespinasse * So, we reuse mm->page_table_lock to guard 24424128997bSMichel Lespinasse * against concurrent vma expansions. 24434128997bSMichel Lespinasse */ 244409357814SOleg Nesterov spin_lock(&mm->page_table_lock); 244587e8827bSOleg Nesterov if (vma->vm_flags & VM_LOCKED) 244609357814SOleg Nesterov mm->locked_vm += grow; 244784638335SKonstantin Khlebnikov vm_stat_account(mm, vma->vm_flags, grow); 2448bf181b9fSMichel Lespinasse anon_vma_interval_tree_pre_update_vma(vma); 24491da177e4SLinus Torvalds vma->vm_end = address; 2450bf181b9fSMichel Lespinasse anon_vma_interval_tree_post_update_vma(vma); 2451d3737187SMichel Lespinasse if (vma->vm_next) 2452d3737187SMichel Lespinasse vma_gap_update(vma->vm_next); 2453d3737187SMichel Lespinasse else 24541be7107fSHugh Dickins mm->highest_vm_end = vm_end_gap(vma); 245509357814SOleg Nesterov spin_unlock(&mm->page_table_lock); 24564128997bSMichel Lespinasse 24573af9e859SEric B Munson perf_event_mmap(vma); 24583af9e859SEric B Munson } 24591da177e4SLinus Torvalds } 246042c36f63SHugh Dickins } 246112352d3cSKonstantin Khlebnikov anon_vma_unlock_write(vma->anon_vma); 24626d50e60cSDavid Rientjes khugepaged_enter_vma_merge(vma, vma->vm_flags); 246309357814SOleg Nesterov validate_mm(mm); 24641da177e4SLinus Torvalds return error; 24651da177e4SLinus Torvalds } 246646dea3d0SHugh Dickins #endif /* CONFIG_STACK_GROWSUP || CONFIG_IA64 */ 246746dea3d0SHugh Dickins 24681da177e4SLinus Torvalds /* 24691da177e4SLinus Torvalds * vma is the first one with address < vma->vm_start. Have to extend vma. 24701da177e4SLinus Torvalds */ 2471d05f3169SMichal Hocko int expand_downwards(struct vm_area_struct *vma, 2472b6a2fea3SOllie Wild unsigned long address) 24731da177e4SLinus Torvalds { 247409357814SOleg Nesterov struct mm_struct *mm = vma->vm_mm; 24751be7107fSHugh Dickins struct vm_area_struct *prev; 24760a1d5299SJann Horn int error = 0; 24771da177e4SLinus Torvalds 24788869477aSEric Paris address &= PAGE_MASK; 24790a1d5299SJann Horn if (address < mmap_min_addr) 24800a1d5299SJann Horn return -EPERM; 24818869477aSEric Paris 24821be7107fSHugh Dickins /* Enforce stack_guard_gap */ 24831be7107fSHugh Dickins prev = vma->vm_prev; 24841be7107fSHugh Dickins /* Check that both stack segments have the same anon_vma? */ 248532e4e6d5SOleg Nesterov if (prev && !(prev->vm_flags & VM_GROWSDOWN) && 24863122e80eSAnshuman Khandual vma_is_accessible(prev)) { 248732e4e6d5SOleg Nesterov if (address - prev->vm_end < stack_guard_gap) 248832e4e6d5SOleg Nesterov return -ENOMEM; 24891be7107fSHugh Dickins } 24901be7107fSHugh Dickins 249112352d3cSKonstantin Khlebnikov /* We must make sure the anon_vma is allocated. */ 249212352d3cSKonstantin Khlebnikov if (unlikely(anon_vma_prepare(vma))) 249312352d3cSKonstantin Khlebnikov return -ENOMEM; 24941da177e4SLinus Torvalds 24951da177e4SLinus Torvalds /* 24961da177e4SLinus Torvalds * vma->vm_start/vm_end cannot change under us because the caller 2497c1e8d7c6SMichel Lespinasse * is required to hold the mmap_lock in read mode. We need the 24981da177e4SLinus Torvalds * anon_vma lock to serialize against concurrent expand_stacks. 24991da177e4SLinus Torvalds */ 250012352d3cSKonstantin Khlebnikov anon_vma_lock_write(vma->anon_vma); 25011da177e4SLinus Torvalds 25021da177e4SLinus Torvalds /* Somebody else might have raced and expanded it already */ 25031da177e4SLinus Torvalds if (address < vma->vm_start) { 25041da177e4SLinus Torvalds unsigned long size, grow; 25051da177e4SLinus Torvalds 25061da177e4SLinus Torvalds size = vma->vm_end - address; 25071da177e4SLinus Torvalds grow = (vma->vm_start - address) >> PAGE_SHIFT; 25081da177e4SLinus Torvalds 2509a626ca6aSLinus Torvalds error = -ENOMEM; 2510a626ca6aSLinus Torvalds if (grow <= vma->vm_pgoff) { 25111da177e4SLinus Torvalds error = acct_stack_growth(vma, size, grow); 25121da177e4SLinus Torvalds if (!error) { 25134128997bSMichel Lespinasse /* 25144128997bSMichel Lespinasse * vma_gap_update() doesn't support concurrent 2515c1e8d7c6SMichel Lespinasse * updates, but we only hold a shared mmap_lock 25164128997bSMichel Lespinasse * lock here, so we need to protect against 25174128997bSMichel Lespinasse * concurrent vma expansions. 251812352d3cSKonstantin Khlebnikov * anon_vma_lock_write() doesn't help here, as 25194128997bSMichel Lespinasse * we don't guarantee that all growable vmas 25204128997bSMichel Lespinasse * in a mm share the same root anon vma. 25214128997bSMichel Lespinasse * So, we reuse mm->page_table_lock to guard 25224128997bSMichel Lespinasse * against concurrent vma expansions. 25234128997bSMichel Lespinasse */ 252409357814SOleg Nesterov spin_lock(&mm->page_table_lock); 252587e8827bSOleg Nesterov if (vma->vm_flags & VM_LOCKED) 252609357814SOleg Nesterov mm->locked_vm += grow; 252784638335SKonstantin Khlebnikov vm_stat_account(mm, vma->vm_flags, grow); 2528bf181b9fSMichel Lespinasse anon_vma_interval_tree_pre_update_vma(vma); 25291da177e4SLinus Torvalds vma->vm_start = address; 25301da177e4SLinus Torvalds vma->vm_pgoff -= grow; 2531bf181b9fSMichel Lespinasse anon_vma_interval_tree_post_update_vma(vma); 2532d3737187SMichel Lespinasse vma_gap_update(vma); 253309357814SOleg Nesterov spin_unlock(&mm->page_table_lock); 25344128997bSMichel Lespinasse 25353af9e859SEric B Munson perf_event_mmap(vma); 25361da177e4SLinus Torvalds } 25371da177e4SLinus Torvalds } 2538a626ca6aSLinus Torvalds } 253912352d3cSKonstantin Khlebnikov anon_vma_unlock_write(vma->anon_vma); 25406d50e60cSDavid Rientjes khugepaged_enter_vma_merge(vma, vma->vm_flags); 254109357814SOleg Nesterov validate_mm(mm); 25421da177e4SLinus Torvalds return error; 25431da177e4SLinus Torvalds } 25441da177e4SLinus Torvalds 25451be7107fSHugh Dickins /* enforced gap between the expanding stack and other mappings. */ 25461be7107fSHugh Dickins unsigned long stack_guard_gap = 256UL<<PAGE_SHIFT; 25471be7107fSHugh Dickins 25481be7107fSHugh Dickins static int __init cmdline_parse_stack_guard_gap(char *p) 25491be7107fSHugh Dickins { 25501be7107fSHugh Dickins unsigned long val; 25511be7107fSHugh Dickins char *endptr; 25521be7107fSHugh Dickins 25531be7107fSHugh Dickins val = simple_strtoul(p, &endptr, 10); 25541be7107fSHugh Dickins if (!*endptr) 25551be7107fSHugh Dickins stack_guard_gap = val << PAGE_SHIFT; 25561be7107fSHugh Dickins 2557e6d09493SRandy Dunlap return 1; 25581be7107fSHugh Dickins } 25591be7107fSHugh Dickins __setup("stack_guard_gap=", cmdline_parse_stack_guard_gap); 25601be7107fSHugh Dickins 2561b6a2fea3SOllie Wild #ifdef CONFIG_STACK_GROWSUP 2562b6a2fea3SOllie Wild int expand_stack(struct vm_area_struct *vma, unsigned long address) 2563b6a2fea3SOllie Wild { 2564b6a2fea3SOllie Wild return expand_upwards(vma, address); 2565b6a2fea3SOllie Wild } 2566b6a2fea3SOllie Wild 2567b6a2fea3SOllie Wild struct vm_area_struct * 2568b6a2fea3SOllie Wild find_extend_vma(struct mm_struct *mm, unsigned long addr) 2569b6a2fea3SOllie Wild { 2570b6a2fea3SOllie Wild struct vm_area_struct *vma, *prev; 2571b6a2fea3SOllie Wild 2572b6a2fea3SOllie Wild addr &= PAGE_MASK; 2573b6a2fea3SOllie Wild vma = find_vma_prev(mm, addr, &prev); 2574b6a2fea3SOllie Wild if (vma && (vma->vm_start <= addr)) 2575b6a2fea3SOllie Wild return vma; 257604f5866eSAndrea Arcangeli /* don't alter vm_end if the coredump is running */ 25774d45e75aSJann Horn if (!prev || expand_stack(prev, addr)) 2578b6a2fea3SOllie Wild return NULL; 2579cea10a19SMichel Lespinasse if (prev->vm_flags & VM_LOCKED) 2580fc05f566SKirill A. Shutemov populate_vma_page_range(prev, addr, prev->vm_end, NULL); 2581b6a2fea3SOllie Wild return prev; 2582b6a2fea3SOllie Wild } 2583b6a2fea3SOllie Wild #else 2584b6a2fea3SOllie Wild int expand_stack(struct vm_area_struct *vma, unsigned long address) 2585b6a2fea3SOllie Wild { 2586b6a2fea3SOllie Wild return expand_downwards(vma, address); 2587b6a2fea3SOllie Wild } 2588b6a2fea3SOllie Wild 25891da177e4SLinus Torvalds struct vm_area_struct * 25901da177e4SLinus Torvalds find_extend_vma(struct mm_struct *mm, unsigned long addr) 25911da177e4SLinus Torvalds { 25921da177e4SLinus Torvalds struct vm_area_struct *vma; 25931da177e4SLinus Torvalds unsigned long start; 25941da177e4SLinus Torvalds 25951da177e4SLinus Torvalds addr &= PAGE_MASK; 25961da177e4SLinus Torvalds vma = find_vma(mm, addr); 25971da177e4SLinus Torvalds if (!vma) 25981da177e4SLinus Torvalds return NULL; 25991da177e4SLinus Torvalds if (vma->vm_start <= addr) 26001da177e4SLinus Torvalds return vma; 26011da177e4SLinus Torvalds if (!(vma->vm_flags & VM_GROWSDOWN)) 26021da177e4SLinus Torvalds return NULL; 26031da177e4SLinus Torvalds start = vma->vm_start; 26041da177e4SLinus Torvalds if (expand_stack(vma, addr)) 26051da177e4SLinus Torvalds return NULL; 2606cea10a19SMichel Lespinasse if (vma->vm_flags & VM_LOCKED) 2607fc05f566SKirill A. Shutemov populate_vma_page_range(vma, addr, start, NULL); 26081da177e4SLinus Torvalds return vma; 26091da177e4SLinus Torvalds } 26101da177e4SLinus Torvalds #endif 26111da177e4SLinus Torvalds 2612e1d6d01aSJesse Barnes EXPORT_SYMBOL_GPL(find_extend_vma); 2613e1d6d01aSJesse Barnes 26142c0b3814SHugh Dickins /* 26152c0b3814SHugh Dickins * Ok - we have the memory areas we should free on the vma list, 26162c0b3814SHugh Dickins * so release them, and do the vma updates. 26171da177e4SLinus Torvalds * 26182c0b3814SHugh Dickins * Called with the mm semaphore held. 26191da177e4SLinus Torvalds */ 26202c0b3814SHugh Dickins static void remove_vma_list(struct mm_struct *mm, struct vm_area_struct *vma) 26211da177e4SLinus Torvalds { 26224f74d2c8SLinus Torvalds unsigned long nr_accounted = 0; 26234f74d2c8SLinus Torvalds 2624365e9c87SHugh Dickins /* Update high watermark before we lower total_vm */ 2625365e9c87SHugh Dickins update_hiwater_vm(mm); 26262c0b3814SHugh Dickins do { 2627ab50b8edSHugh Dickins long nrpages = vma_pages(vma); 26281da177e4SLinus Torvalds 26294f74d2c8SLinus Torvalds if (vma->vm_flags & VM_ACCOUNT) 26304f74d2c8SLinus Torvalds nr_accounted += nrpages; 263184638335SKonstantin Khlebnikov vm_stat_account(mm, vma->vm_flags, -nrpages); 2632a8fb5618SHugh Dickins vma = remove_vma(vma); 2633146425a3SHugh Dickins } while (vma); 26344f74d2c8SLinus Torvalds vm_unacct_memory(nr_accounted); 26351da177e4SLinus Torvalds validate_mm(mm); 26361da177e4SLinus Torvalds } 26371da177e4SLinus Torvalds 26381da177e4SLinus Torvalds /* 26391da177e4SLinus Torvalds * Get rid of page table information in the indicated region. 26401da177e4SLinus Torvalds * 2641f10df686SPaolo 'Blaisorblade' Giarrusso * Called with the mm semaphore held. 26421da177e4SLinus Torvalds */ 26431da177e4SLinus Torvalds static void unmap_region(struct mm_struct *mm, 2644e0da382cSHugh Dickins struct vm_area_struct *vma, struct vm_area_struct *prev, 2645e0da382cSHugh Dickins unsigned long start, unsigned long end) 26461da177e4SLinus Torvalds { 26473903b55aSLiam R. Howlett struct vm_area_struct *next = vma_next(mm, prev); 2648d16dfc55SPeter Zijlstra struct mmu_gather tlb; 26491da177e4SLinus Torvalds 26501da177e4SLinus Torvalds lru_add_drain(); 2651a72afd87SWill Deacon tlb_gather_mmu(&tlb, mm); 2652365e9c87SHugh Dickins update_hiwater_rss(mm); 26534f74d2c8SLinus Torvalds unmap_vmas(&tlb, vma, start, end); 2654d16dfc55SPeter Zijlstra free_pgtables(&tlb, vma, prev ? prev->vm_end : FIRST_USER_ADDRESS, 26556ee8630eSHugh Dickins next ? next->vm_start : USER_PGTABLES_CEILING); 2656ae8eba8bSWill Deacon tlb_finish_mmu(&tlb); 26571da177e4SLinus Torvalds } 26581da177e4SLinus Torvalds 26591da177e4SLinus Torvalds /* 26601da177e4SLinus Torvalds * Create a list of vma's touched by the unmap, removing them from the mm's 26611da177e4SLinus Torvalds * vma list as we go.. 26621da177e4SLinus Torvalds */ 2663246c320aSKirill A. Shutemov static bool 26641da177e4SLinus Torvalds detach_vmas_to_be_unmapped(struct mm_struct *mm, struct vm_area_struct *vma, 26651da177e4SLinus Torvalds struct vm_area_struct *prev, unsigned long end) 26661da177e4SLinus Torvalds { 26671da177e4SLinus Torvalds struct vm_area_struct **insertion_point; 26681da177e4SLinus Torvalds struct vm_area_struct *tail_vma = NULL; 26691da177e4SLinus Torvalds 26701da177e4SLinus Torvalds insertion_point = (prev ? &prev->vm_next : &mm->mmap); 2671297c5eeeSLinus Torvalds vma->vm_prev = NULL; 26721da177e4SLinus Torvalds do { 2673d3737187SMichel Lespinasse vma_rb_erase(vma, &mm->mm_rb); 2674a213e5cfSHugh Dickins if (vma->vm_flags & VM_LOCKED) 2675a213e5cfSHugh Dickins mm->locked_vm -= vma_pages(vma); 26761da177e4SLinus Torvalds mm->map_count--; 26771da177e4SLinus Torvalds tail_vma = vma; 26781da177e4SLinus Torvalds vma = vma->vm_next; 26791da177e4SLinus Torvalds } while (vma && vma->vm_start < end); 26801da177e4SLinus Torvalds *insertion_point = vma; 2681d3737187SMichel Lespinasse if (vma) { 2682297c5eeeSLinus Torvalds vma->vm_prev = prev; 2683d3737187SMichel Lespinasse vma_gap_update(vma); 2684d3737187SMichel Lespinasse } else 26851be7107fSHugh Dickins mm->highest_vm_end = prev ? vm_end_gap(prev) : 0; 26861da177e4SLinus Torvalds tail_vma->vm_next = NULL; 2687615d6e87SDavidlohr Bueso 2688615d6e87SDavidlohr Bueso /* Kill the cache */ 2689615d6e87SDavidlohr Bueso vmacache_invalidate(mm); 2690246c320aSKirill A. Shutemov 2691246c320aSKirill A. Shutemov /* 2692246c320aSKirill A. Shutemov * Do not downgrade mmap_lock if we are next to VM_GROWSDOWN or 2693246c320aSKirill A. Shutemov * VM_GROWSUP VMA. Such VMAs can change their size under 2694246c320aSKirill A. Shutemov * down_read(mmap_lock) and collide with the VMA we are about to unmap. 2695246c320aSKirill A. Shutemov */ 2696246c320aSKirill A. Shutemov if (vma && (vma->vm_flags & VM_GROWSDOWN)) 2697246c320aSKirill A. Shutemov return false; 2698246c320aSKirill A. Shutemov if (prev && (prev->vm_flags & VM_GROWSUP)) 2699246c320aSKirill A. Shutemov return false; 2700246c320aSKirill A. Shutemov return true; 27011da177e4SLinus Torvalds } 27021da177e4SLinus Torvalds 27031da177e4SLinus Torvalds /* 2704def5efe0SDavid Rientjes * __split_vma() bypasses sysctl_max_map_count checking. We use this where it 2705def5efe0SDavid Rientjes * has already been checked or doesn't make sense to fail. 27061da177e4SLinus Torvalds */ 2707def5efe0SDavid Rientjes int __split_vma(struct mm_struct *mm, struct vm_area_struct *vma, 27081da177e4SLinus Torvalds unsigned long addr, int new_below) 27091da177e4SLinus Torvalds { 27101da177e4SLinus Torvalds struct vm_area_struct *new; 2711e3975891SChen Gang int err; 27121da177e4SLinus Torvalds 2713dd3b614fSDmitry Safonov if (vma->vm_ops && vma->vm_ops->may_split) { 2714dd3b614fSDmitry Safonov err = vma->vm_ops->may_split(vma, addr); 271531383c68SDan Williams if (err) 271631383c68SDan Williams return err; 271731383c68SDan Williams } 27181da177e4SLinus Torvalds 27193928d4f5SLinus Torvalds new = vm_area_dup(vma); 27201da177e4SLinus Torvalds if (!new) 2721e3975891SChen Gang return -ENOMEM; 27221da177e4SLinus Torvalds 27231da177e4SLinus Torvalds if (new_below) 27241da177e4SLinus Torvalds new->vm_end = addr; 27251da177e4SLinus Torvalds else { 27261da177e4SLinus Torvalds new->vm_start = addr; 27271da177e4SLinus Torvalds new->vm_pgoff += ((addr - vma->vm_start) >> PAGE_SHIFT); 27281da177e4SLinus Torvalds } 27291da177e4SLinus Torvalds 2730ef0855d3SOleg Nesterov err = vma_dup_policy(vma, new); 2731ef0855d3SOleg Nesterov if (err) 27325beb4930SRik van Riel goto out_free_vma; 27331da177e4SLinus Torvalds 2734c4ea95d7SDaniel Forrest err = anon_vma_clone(new, vma); 2735c4ea95d7SDaniel Forrest if (err) 27365beb4930SRik van Riel goto out_free_mpol; 27375beb4930SRik van Riel 2738e9714acfSKonstantin Khlebnikov if (new->vm_file) 27391da177e4SLinus Torvalds get_file(new->vm_file); 27401da177e4SLinus Torvalds 27411da177e4SLinus Torvalds if (new->vm_ops && new->vm_ops->open) 27421da177e4SLinus Torvalds new->vm_ops->open(new); 27431da177e4SLinus Torvalds 27441da177e4SLinus Torvalds if (new_below) 27455beb4930SRik van Riel err = vma_adjust(vma, addr, vma->vm_end, vma->vm_pgoff + 27461da177e4SLinus Torvalds ((addr - new->vm_start) >> PAGE_SHIFT), new); 27471da177e4SLinus Torvalds else 27485beb4930SRik van Riel err = vma_adjust(vma, vma->vm_start, addr, vma->vm_pgoff, new); 27491da177e4SLinus Torvalds 27505beb4930SRik van Riel /* Success. */ 27515beb4930SRik van Riel if (!err) 27521da177e4SLinus Torvalds return 0; 27535beb4930SRik van Riel 27545beb4930SRik van Riel /* Clean everything up if vma_adjust failed. */ 275558927533SRik van Riel if (new->vm_ops && new->vm_ops->close) 27565beb4930SRik van Riel new->vm_ops->close(new); 2757e9714acfSKonstantin Khlebnikov if (new->vm_file) 27585beb4930SRik van Riel fput(new->vm_file); 27592aeadc30SAndrea Arcangeli unlink_anon_vmas(new); 27605beb4930SRik van Riel out_free_mpol: 2761ef0855d3SOleg Nesterov mpol_put(vma_policy(new)); 27625beb4930SRik van Riel out_free_vma: 27633928d4f5SLinus Torvalds vm_area_free(new); 27645beb4930SRik van Riel return err; 27651da177e4SLinus Torvalds } 27661da177e4SLinus Torvalds 2767659ace58SKOSAKI Motohiro /* 2768659ace58SKOSAKI Motohiro * Split a vma into two pieces at address 'addr', a new vma is allocated 2769659ace58SKOSAKI Motohiro * either for the first part or the tail. 2770659ace58SKOSAKI Motohiro */ 2771659ace58SKOSAKI Motohiro int split_vma(struct mm_struct *mm, struct vm_area_struct *vma, 2772659ace58SKOSAKI Motohiro unsigned long addr, int new_below) 2773659ace58SKOSAKI Motohiro { 2774659ace58SKOSAKI Motohiro if (mm->map_count >= sysctl_max_map_count) 2775659ace58SKOSAKI Motohiro return -ENOMEM; 2776659ace58SKOSAKI Motohiro 2777659ace58SKOSAKI Motohiro return __split_vma(mm, vma, addr, new_below); 2778659ace58SKOSAKI Motohiro } 2779659ace58SKOSAKI Motohiro 27801da177e4SLinus Torvalds /* Munmap is split into 2 main parts -- this part which finds 27811da177e4SLinus Torvalds * what needs doing, and the areas themselves, which do the 27821da177e4SLinus Torvalds * work. This now handles partial unmappings. 27831da177e4SLinus Torvalds * Jeremy Fitzhardinge <jeremy@goop.org> 27841da177e4SLinus Torvalds */ 278585a06835SYang Shi int __do_munmap(struct mm_struct *mm, unsigned long start, size_t len, 2786dd2283f2SYang Shi struct list_head *uf, bool downgrade) 27871da177e4SLinus Torvalds { 27881da177e4SLinus Torvalds unsigned long end; 2789146425a3SHugh Dickins struct vm_area_struct *vma, *prev, *last; 27901da177e4SLinus Torvalds 2791de1741a1SAlexander Kuleshov if ((offset_in_page(start)) || start > TASK_SIZE || len > TASK_SIZE-start) 27921da177e4SLinus Torvalds return -EINVAL; 27931da177e4SLinus Torvalds 2794cc71aba3Svishnu.ps len = PAGE_ALIGN(len); 27955a28fc94SDave Hansen end = start + len; 2796cc71aba3Svishnu.ps if (len == 0) 27971da177e4SLinus Torvalds return -EINVAL; 27981da177e4SLinus Torvalds 27995a28fc94SDave Hansen /* 28005a28fc94SDave Hansen * arch_unmap() might do unmaps itself. It must be called 28015a28fc94SDave Hansen * and finish any rbtree manipulation before this code 28025a28fc94SDave Hansen * runs and also starts to manipulate the rbtree. 28035a28fc94SDave Hansen */ 28045a28fc94SDave Hansen arch_unmap(mm, start, end); 28055a28fc94SDave Hansen 280678d9cf60SGonzalo Matias Juarez Tello /* Find the first overlapping VMA where start < vma->vm_end */ 280778d9cf60SGonzalo Matias Juarez Tello vma = find_vma_intersection(mm, start, end); 2808146425a3SHugh Dickins if (!vma) 28091da177e4SLinus Torvalds return 0; 28109be34c9dSLinus Torvalds prev = vma->vm_prev; 28111da177e4SLinus Torvalds 28121da177e4SLinus Torvalds /* 28131da177e4SLinus Torvalds * If we need to split any vma, do it now to save pain later. 28141da177e4SLinus Torvalds * 28151da177e4SLinus Torvalds * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially 28161da177e4SLinus Torvalds * unmapped vm_area_struct will remain in use: so lower split_vma 28171da177e4SLinus Torvalds * places tmp vma above, and higher split_vma places tmp vma below. 28181da177e4SLinus Torvalds */ 2819146425a3SHugh Dickins if (start > vma->vm_start) { 2820659ace58SKOSAKI Motohiro int error; 2821659ace58SKOSAKI Motohiro 2822659ace58SKOSAKI Motohiro /* 2823659ace58SKOSAKI Motohiro * Make sure that map_count on return from munmap() will 2824659ace58SKOSAKI Motohiro * not exceed its limit; but let map_count go just above 2825659ace58SKOSAKI Motohiro * its limit temporarily, to help free resources as expected. 2826659ace58SKOSAKI Motohiro */ 2827659ace58SKOSAKI Motohiro if (end < vma->vm_end && mm->map_count >= sysctl_max_map_count) 2828659ace58SKOSAKI Motohiro return -ENOMEM; 2829659ace58SKOSAKI Motohiro 2830659ace58SKOSAKI Motohiro error = __split_vma(mm, vma, start, 0); 28311da177e4SLinus Torvalds if (error) 28321da177e4SLinus Torvalds return error; 2833146425a3SHugh Dickins prev = vma; 28341da177e4SLinus Torvalds } 28351da177e4SLinus Torvalds 28361da177e4SLinus Torvalds /* Does it split the last one? */ 28371da177e4SLinus Torvalds last = find_vma(mm, end); 28381da177e4SLinus Torvalds if (last && end > last->vm_start) { 2839659ace58SKOSAKI Motohiro int error = __split_vma(mm, last, end, 1); 28401da177e4SLinus Torvalds if (error) 28411da177e4SLinus Torvalds return error; 28421da177e4SLinus Torvalds } 28433903b55aSLiam R. Howlett vma = vma_next(mm, prev); 28441da177e4SLinus Torvalds 28452376dd7cSAndrea Arcangeli if (unlikely(uf)) { 28462376dd7cSAndrea Arcangeli /* 28472376dd7cSAndrea Arcangeli * If userfaultfd_unmap_prep returns an error the vmas 2848f0953a1bSIngo Molnar * will remain split, but userland will get a 28492376dd7cSAndrea Arcangeli * highly unexpected error anyway. This is no 28502376dd7cSAndrea Arcangeli * different than the case where the first of the two 28512376dd7cSAndrea Arcangeli * __split_vma fails, but we don't undo the first 28522376dd7cSAndrea Arcangeli * split, despite we could. This is unlikely enough 28532376dd7cSAndrea Arcangeli * failure that it's not worth optimizing it for. 28542376dd7cSAndrea Arcangeli */ 28552376dd7cSAndrea Arcangeli int error = userfaultfd_unmap_prep(vma, start, end, uf); 28562376dd7cSAndrea Arcangeli if (error) 28572376dd7cSAndrea Arcangeli return error; 28582376dd7cSAndrea Arcangeli } 28592376dd7cSAndrea Arcangeli 2860dd2283f2SYang Shi /* Detach vmas from rbtree */ 2861246c320aSKirill A. Shutemov if (!detach_vmas_to_be_unmapped(mm, vma, prev, end)) 2862246c320aSKirill A. Shutemov downgrade = false; 28631da177e4SLinus Torvalds 2864dd2283f2SYang Shi if (downgrade) 2865d8ed45c5SMichel Lespinasse mmap_write_downgrade(mm); 2866dd2283f2SYang Shi 2867dd2283f2SYang Shi unmap_region(mm, vma, prev, start, end); 2868dd2283f2SYang Shi 28691da177e4SLinus Torvalds /* Fix up all other VM information */ 28702c0b3814SHugh Dickins remove_vma_list(mm, vma); 28711da177e4SLinus Torvalds 2872dd2283f2SYang Shi return downgrade ? 1 : 0; 28731da177e4SLinus Torvalds } 28741da177e4SLinus Torvalds 2875dd2283f2SYang Shi int do_munmap(struct mm_struct *mm, unsigned long start, size_t len, 2876dd2283f2SYang Shi struct list_head *uf) 2877dd2283f2SYang Shi { 2878dd2283f2SYang Shi return __do_munmap(mm, start, len, uf, false); 2879dd2283f2SYang Shi } 2880dd2283f2SYang Shi 2881dd2283f2SYang Shi static int __vm_munmap(unsigned long start, size_t len, bool downgrade) 2882a46ef99dSLinus Torvalds { 2883a46ef99dSLinus Torvalds int ret; 2884bfce281cSAl Viro struct mm_struct *mm = current->mm; 2885897ab3e0SMike Rapoport LIST_HEAD(uf); 2886a46ef99dSLinus Torvalds 2887d8ed45c5SMichel Lespinasse if (mmap_write_lock_killable(mm)) 2888ae798783SMichal Hocko return -EINTR; 2889ae798783SMichal Hocko 2890dd2283f2SYang Shi ret = __do_munmap(mm, start, len, &uf, downgrade); 2891dd2283f2SYang Shi /* 2892c1e8d7c6SMichel Lespinasse * Returning 1 indicates mmap_lock is downgraded. 2893dd2283f2SYang Shi * But 1 is not legal return value of vm_munmap() and munmap(), reset 2894dd2283f2SYang Shi * it to 0 before return. 2895dd2283f2SYang Shi */ 2896dd2283f2SYang Shi if (ret == 1) { 2897d8ed45c5SMichel Lespinasse mmap_read_unlock(mm); 2898dd2283f2SYang Shi ret = 0; 2899dd2283f2SYang Shi } else 2900d8ed45c5SMichel Lespinasse mmap_write_unlock(mm); 2901dd2283f2SYang Shi 2902897ab3e0SMike Rapoport userfaultfd_unmap_complete(mm, &uf); 2903a46ef99dSLinus Torvalds return ret; 2904a46ef99dSLinus Torvalds } 2905dd2283f2SYang Shi 2906dd2283f2SYang Shi int vm_munmap(unsigned long start, size_t len) 2907dd2283f2SYang Shi { 2908dd2283f2SYang Shi return __vm_munmap(start, len, false); 2909dd2283f2SYang Shi } 2910a46ef99dSLinus Torvalds EXPORT_SYMBOL(vm_munmap); 2911a46ef99dSLinus Torvalds 29126a6160a7SHeiko Carstens SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len) 29131da177e4SLinus Torvalds { 2914ce18d171SCatalin Marinas addr = untagged_addr(addr); 2915dd2283f2SYang Shi return __vm_munmap(addr, len, true); 29161da177e4SLinus Torvalds } 29171da177e4SLinus Torvalds 2918c8d78c18SKirill A. Shutemov 2919c8d78c18SKirill A. Shutemov /* 2920c8d78c18SKirill A. Shutemov * Emulation of deprecated remap_file_pages() syscall. 2921c8d78c18SKirill A. Shutemov */ 2922c8d78c18SKirill A. Shutemov SYSCALL_DEFINE5(remap_file_pages, unsigned long, start, unsigned long, size, 2923c8d78c18SKirill A. Shutemov unsigned long, prot, unsigned long, pgoff, unsigned long, flags) 2924c8d78c18SKirill A. Shutemov { 2925c8d78c18SKirill A. Shutemov 2926c8d78c18SKirill A. Shutemov struct mm_struct *mm = current->mm; 2927c8d78c18SKirill A. Shutemov struct vm_area_struct *vma; 2928c8d78c18SKirill A. Shutemov unsigned long populate = 0; 2929c8d78c18SKirill A. Shutemov unsigned long ret = -EINVAL; 2930c8d78c18SKirill A. Shutemov struct file *file; 2931c8d78c18SKirill A. Shutemov 2932ad56b738SMike Rapoport pr_warn_once("%s (%d) uses deprecated remap_file_pages() syscall. See Documentation/vm/remap_file_pages.rst.\n", 2933c8d78c18SKirill A. Shutemov current->comm, current->pid); 2934c8d78c18SKirill A. Shutemov 2935c8d78c18SKirill A. Shutemov if (prot) 2936c8d78c18SKirill A. Shutemov return ret; 2937c8d78c18SKirill A. Shutemov start = start & PAGE_MASK; 2938c8d78c18SKirill A. Shutemov size = size & PAGE_MASK; 2939c8d78c18SKirill A. Shutemov 2940c8d78c18SKirill A. Shutemov if (start + size <= start) 2941c8d78c18SKirill A. Shutemov return ret; 2942c8d78c18SKirill A. Shutemov 2943c8d78c18SKirill A. Shutemov /* Does pgoff wrap? */ 2944c8d78c18SKirill A. Shutemov if (pgoff + (size >> PAGE_SHIFT) < pgoff) 2945c8d78c18SKirill A. Shutemov return ret; 2946c8d78c18SKirill A. Shutemov 2947d8ed45c5SMichel Lespinasse if (mmap_write_lock_killable(mm)) 2948dc0ef0dfSMichal Hocko return -EINTR; 2949dc0ef0dfSMichal Hocko 29509b593cb2SLiam R. Howlett vma = vma_lookup(mm, start); 2951c8d78c18SKirill A. Shutemov 2952c8d78c18SKirill A. Shutemov if (!vma || !(vma->vm_flags & VM_SHARED)) 2953c8d78c18SKirill A. Shutemov goto out; 2954c8d78c18SKirill A. Shutemov 295548f7df32SKirill A. Shutemov if (start + size > vma->vm_end) { 295648f7df32SKirill A. Shutemov struct vm_area_struct *next; 295748f7df32SKirill A. Shutemov 295848f7df32SKirill A. Shutemov for (next = vma->vm_next; next; next = next->vm_next) { 295948f7df32SKirill A. Shutemov /* hole between vmas ? */ 296048f7df32SKirill A. Shutemov if (next->vm_start != next->vm_prev->vm_end) 296148f7df32SKirill A. Shutemov goto out; 296248f7df32SKirill A. Shutemov 296348f7df32SKirill A. Shutemov if (next->vm_file != vma->vm_file) 296448f7df32SKirill A. Shutemov goto out; 296548f7df32SKirill A. Shutemov 296648f7df32SKirill A. Shutemov if (next->vm_flags != vma->vm_flags) 296748f7df32SKirill A. Shutemov goto out; 296848f7df32SKirill A. Shutemov 296948f7df32SKirill A. Shutemov if (start + size <= next->vm_end) 297048f7df32SKirill A. Shutemov break; 297148f7df32SKirill A. Shutemov } 297248f7df32SKirill A. Shutemov 297348f7df32SKirill A. Shutemov if (!next) 2974c8d78c18SKirill A. Shutemov goto out; 2975c8d78c18SKirill A. Shutemov } 2976c8d78c18SKirill A. Shutemov 2977c8d78c18SKirill A. Shutemov prot |= vma->vm_flags & VM_READ ? PROT_READ : 0; 2978c8d78c18SKirill A. Shutemov prot |= vma->vm_flags & VM_WRITE ? PROT_WRITE : 0; 2979c8d78c18SKirill A. Shutemov prot |= vma->vm_flags & VM_EXEC ? PROT_EXEC : 0; 2980c8d78c18SKirill A. Shutemov 2981c8d78c18SKirill A. Shutemov flags &= MAP_NONBLOCK; 2982c8d78c18SKirill A. Shutemov flags |= MAP_SHARED | MAP_FIXED | MAP_POPULATE; 2983fce000b1SLiam Howlett if (vma->vm_flags & VM_LOCKED) 2984c8d78c18SKirill A. Shutemov flags |= MAP_LOCKED; 298548f7df32SKirill A. Shutemov 2986c8d78c18SKirill A. Shutemov file = get_file(vma->vm_file); 298745e55300SPeter Collingbourne ret = do_mmap(vma->vm_file, start, size, 2988897ab3e0SMike Rapoport prot, flags, pgoff, &populate, NULL); 2989c8d78c18SKirill A. Shutemov fput(file); 2990c8d78c18SKirill A. Shutemov out: 2991d8ed45c5SMichel Lespinasse mmap_write_unlock(mm); 2992c8d78c18SKirill A. Shutemov if (populate) 2993c8d78c18SKirill A. Shutemov mm_populate(ret, populate); 2994c8d78c18SKirill A. Shutemov if (!IS_ERR_VALUE(ret)) 2995c8d78c18SKirill A. Shutemov ret = 0; 2996c8d78c18SKirill A. Shutemov return ret; 2997c8d78c18SKirill A. Shutemov } 2998c8d78c18SKirill A. Shutemov 29991da177e4SLinus Torvalds /* 30001da177e4SLinus Torvalds * this is really a simplified "do_mmap". it only handles 30011da177e4SLinus Torvalds * anonymous maps. eventually we may be able to do some 30021da177e4SLinus Torvalds * brk-specific accounting here. 30031da177e4SLinus Torvalds */ 3004bb177a73SMichal Hocko static int do_brk_flags(unsigned long addr, unsigned long len, unsigned long flags, struct list_head *uf) 30051da177e4SLinus Torvalds { 30061da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 30071da177e4SLinus Torvalds struct vm_area_struct *vma, *prev; 30081da177e4SLinus Torvalds struct rb_node **rb_link, *rb_parent; 30091da177e4SLinus Torvalds pgoff_t pgoff = addr >> PAGE_SHIFT; 30103a459756SKirill Korotaev int error; 3011ff68dac6SGaowei Pu unsigned long mapped_addr; 30121da177e4SLinus Torvalds 301316e72e9bSDenys Vlasenko /* Until we need other flags, refuse anything except VM_EXEC. */ 301416e72e9bSDenys Vlasenko if ((flags & (~VM_EXEC)) != 0) 301516e72e9bSDenys Vlasenko return -EINVAL; 301616e72e9bSDenys Vlasenko flags |= VM_DATA_DEFAULT_FLAGS | VM_ACCOUNT | mm->def_flags; 30173a459756SKirill Korotaev 3018ff68dac6SGaowei Pu mapped_addr = get_unmapped_area(NULL, addr, len, 0, MAP_FIXED); 3019ff68dac6SGaowei Pu if (IS_ERR_VALUE(mapped_addr)) 3020ff68dac6SGaowei Pu return mapped_addr; 30213a459756SKirill Korotaev 3022363ee17fSDavidlohr Bueso error = mlock_future_check(mm, mm->def_flags, len); 3023363ee17fSDavidlohr Bueso if (error) 3024363ee17fSDavidlohr Bueso return error; 30251da177e4SLinus Torvalds 3026fb8090b6SLiam R. Howlett /* Clear old maps, set up prev, rb_link, rb_parent, and uf */ 3027fb8090b6SLiam R. Howlett if (munmap_vma_range(mm, addr, len, &prev, &rb_link, &rb_parent, uf)) 30281da177e4SLinus Torvalds return -ENOMEM; 30291da177e4SLinus Torvalds 30301da177e4SLinus Torvalds /* Check against address space limits *after* clearing old maps... */ 303184638335SKonstantin Khlebnikov if (!may_expand_vm(mm, flags, len >> PAGE_SHIFT)) 30321da177e4SLinus Torvalds return -ENOMEM; 30331da177e4SLinus Torvalds 30341da177e4SLinus Torvalds if (mm->map_count > sysctl_max_map_count) 30351da177e4SLinus Torvalds return -ENOMEM; 30361da177e4SLinus Torvalds 3037191c5424SAl Viro if (security_vm_enough_memory_mm(mm, len >> PAGE_SHIFT)) 30381da177e4SLinus Torvalds return -ENOMEM; 30391da177e4SLinus Torvalds 30401da177e4SLinus Torvalds /* Can we just expand an old private anonymous mapping? */ 3041ba470de4SRik van Riel vma = vma_merge(mm, prev, addr, addr + len, flags, 30429a10064fSColin Cross NULL, NULL, pgoff, NULL, NULL_VM_UFFD_CTX, NULL); 3043ba470de4SRik van Riel if (vma) 30441da177e4SLinus Torvalds goto out; 30451da177e4SLinus Torvalds 30461da177e4SLinus Torvalds /* 30471da177e4SLinus Torvalds * create a vma struct for an anonymous mapping 30481da177e4SLinus Torvalds */ 3049490fc053SLinus Torvalds vma = vm_area_alloc(mm); 30501da177e4SLinus Torvalds if (!vma) { 30511da177e4SLinus Torvalds vm_unacct_memory(len >> PAGE_SHIFT); 30521da177e4SLinus Torvalds return -ENOMEM; 30531da177e4SLinus Torvalds } 30541da177e4SLinus Torvalds 3055bfd40eafSKirill A. Shutemov vma_set_anonymous(vma); 30561da177e4SLinus Torvalds vma->vm_start = addr; 30571da177e4SLinus Torvalds vma->vm_end = addr + len; 30581da177e4SLinus Torvalds vma->vm_pgoff = pgoff; 30591da177e4SLinus Torvalds vma->vm_flags = flags; 30603ed75eb8SColy Li vma->vm_page_prot = vm_get_page_prot(flags); 30611da177e4SLinus Torvalds vma_link(mm, vma, prev, rb_link, rb_parent); 30621da177e4SLinus Torvalds out: 30633af9e859SEric B Munson perf_event_mmap(vma); 30641da177e4SLinus Torvalds mm->total_vm += len >> PAGE_SHIFT; 306584638335SKonstantin Khlebnikov mm->data_vm += len >> PAGE_SHIFT; 3066128557ffSMichel Lespinasse if (flags & VM_LOCKED) 3067ba470de4SRik van Riel mm->locked_vm += (len >> PAGE_SHIFT); 3068d9104d1cSCyrill Gorcunov vma->vm_flags |= VM_SOFTDIRTY; 30695d22fc25SLinus Torvalds return 0; 30701da177e4SLinus Torvalds } 30711da177e4SLinus Torvalds 3072bb177a73SMichal Hocko int vm_brk_flags(unsigned long addr, unsigned long request, unsigned long flags) 3073e4eb1ff6SLinus Torvalds { 3074e4eb1ff6SLinus Torvalds struct mm_struct *mm = current->mm; 3075bb177a73SMichal Hocko unsigned long len; 30765d22fc25SLinus Torvalds int ret; 3077128557ffSMichel Lespinasse bool populate; 3078897ab3e0SMike Rapoport LIST_HEAD(uf); 3079e4eb1ff6SLinus Torvalds 3080bb177a73SMichal Hocko len = PAGE_ALIGN(request); 3081bb177a73SMichal Hocko if (len < request) 3082bb177a73SMichal Hocko return -ENOMEM; 3083bb177a73SMichal Hocko if (!len) 3084bb177a73SMichal Hocko return 0; 3085bb177a73SMichal Hocko 3086d8ed45c5SMichel Lespinasse if (mmap_write_lock_killable(mm)) 30872d6c9282SMichal Hocko return -EINTR; 30882d6c9282SMichal Hocko 3089897ab3e0SMike Rapoport ret = do_brk_flags(addr, len, flags, &uf); 3090128557ffSMichel Lespinasse populate = ((mm->def_flags & VM_LOCKED) != 0); 3091d8ed45c5SMichel Lespinasse mmap_write_unlock(mm); 3092897ab3e0SMike Rapoport userfaultfd_unmap_complete(mm, &uf); 30935d22fc25SLinus Torvalds if (populate && !ret) 3094128557ffSMichel Lespinasse mm_populate(addr, len); 3095e4eb1ff6SLinus Torvalds return ret; 3096e4eb1ff6SLinus Torvalds } 309716e72e9bSDenys Vlasenko EXPORT_SYMBOL(vm_brk_flags); 309816e72e9bSDenys Vlasenko 309916e72e9bSDenys Vlasenko int vm_brk(unsigned long addr, unsigned long len) 310016e72e9bSDenys Vlasenko { 310116e72e9bSDenys Vlasenko return vm_brk_flags(addr, len, 0); 310216e72e9bSDenys Vlasenko } 3103e4eb1ff6SLinus Torvalds EXPORT_SYMBOL(vm_brk); 31041da177e4SLinus Torvalds 31051da177e4SLinus Torvalds /* Release all mmaps. */ 31061da177e4SLinus Torvalds void exit_mmap(struct mm_struct *mm) 31071da177e4SLinus Torvalds { 3108d16dfc55SPeter Zijlstra struct mmu_gather tlb; 3109ba470de4SRik van Riel struct vm_area_struct *vma; 31101da177e4SLinus Torvalds unsigned long nr_accounted = 0; 31111da177e4SLinus Torvalds 3112d6dd61c8SJeremy Fitzhardinge /* mm's last user has gone, and its about to be pulled down */ 3113cddb8a5cSAndrea Arcangeli mmu_notifier_release(mm); 3114d6dd61c8SJeremy Fitzhardinge 311527ae357fSDavid Rientjes if (unlikely(mm_is_oom_victim(mm))) { 311627ae357fSDavid Rientjes /* 311727ae357fSDavid Rientjes * Manually reap the mm to free as much memory as possible. 311827ae357fSDavid Rientjes * Then, as the oom reaper does, set MMF_OOM_SKIP to disregard 3119c1e8d7c6SMichel Lespinasse * this mm from further consideration. Taking mm->mmap_lock for 312027ae357fSDavid Rientjes * write after setting MMF_OOM_SKIP will guarantee that the oom 3121c1e8d7c6SMichel Lespinasse * reaper will not run on this mm again after mmap_lock is 312227ae357fSDavid Rientjes * dropped. 312327ae357fSDavid Rientjes * 3124c1e8d7c6SMichel Lespinasse * Nothing can be holding mm->mmap_lock here and the above call 312527ae357fSDavid Rientjes * to mmu_notifier_release(mm) ensures mmu notifier callbacks in 312627ae357fSDavid Rientjes * __oom_reap_task_mm() will not block. 312727ae357fSDavid Rientjes */ 312893065ac7SMichal Hocko (void)__oom_reap_task_mm(mm); 312927ae357fSDavid Rientjes set_bit(MMF_OOM_SKIP, &mm->flags); 313027ae357fSDavid Rientjes } 313127ae357fSDavid Rientjes 313264591e86SSuren Baghdasaryan mmap_write_lock(mm); 31339480c53eSJeremy Fitzhardinge arch_exit_mmap(mm); 31349480c53eSJeremy Fitzhardinge 3135ba470de4SRik van Riel vma = mm->mmap; 313664591e86SSuren Baghdasaryan if (!vma) { 313764591e86SSuren Baghdasaryan /* Can happen if dup_mmap() received an OOM */ 313864591e86SSuren Baghdasaryan mmap_write_unlock(mm); 31399480c53eSJeremy Fitzhardinge return; 314064591e86SSuren Baghdasaryan } 31419480c53eSJeremy Fitzhardinge 31421da177e4SLinus Torvalds lru_add_drain(); 31431da177e4SLinus Torvalds flush_cache_mm(mm); 3144d8b45053SWill Deacon tlb_gather_mmu_fullmm(&tlb, mm); 3145901608d9SOleg Nesterov /* update_hiwater_rss(mm) here? but nobody should be looking */ 3146e0da382cSHugh Dickins /* Use -1 here to ensure all VMAs in the mm are unmapped */ 31474f74d2c8SLinus Torvalds unmap_vmas(&tlb, vma, 0, -1); 31486ee8630eSHugh Dickins free_pgtables(&tlb, vma, FIRST_USER_ADDRESS, USER_PGTABLES_CEILING); 3149ae8eba8bSWill Deacon tlb_finish_mmu(&tlb); 31501da177e4SLinus Torvalds 315164591e86SSuren Baghdasaryan /* Walk the list again, actually closing and freeing it. */ 31524f74d2c8SLinus Torvalds while (vma) { 31534f74d2c8SLinus Torvalds if (vma->vm_flags & VM_ACCOUNT) 31544f74d2c8SLinus Torvalds nr_accounted += vma_pages(vma); 3155a8fb5618SHugh Dickins vma = remove_vma(vma); 31560a3b3c25SPaul E. McKenney cond_resched(); 31574f74d2c8SLinus Torvalds } 3158f798a1d4SSuren Baghdasaryan mm->mmap = NULL; 315964591e86SSuren Baghdasaryan mmap_write_unlock(mm); 31604f74d2c8SLinus Torvalds vm_unacct_memory(nr_accounted); 31611da177e4SLinus Torvalds } 31621da177e4SLinus Torvalds 31631da177e4SLinus Torvalds /* Insert vm structure into process list sorted by address 31641da177e4SLinus Torvalds * and into the inode's i_mmap tree. If vm_file is non-NULL 3165c8c06efaSDavidlohr Bueso * then i_mmap_rwsem is taken here. 31661da177e4SLinus Torvalds */ 31671da177e4SLinus Torvalds int insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma) 31681da177e4SLinus Torvalds { 31696597d783SHugh Dickins struct vm_area_struct *prev; 31701da177e4SLinus Torvalds struct rb_node **rb_link, *rb_parent; 31711da177e4SLinus Torvalds 3172c9d13f5fSChen Gang if (find_vma_links(mm, vma->vm_start, vma->vm_end, 3173c9d13f5fSChen Gang &prev, &rb_link, &rb_parent)) 3174c9d13f5fSChen Gang return -ENOMEM; 3175c9d13f5fSChen Gang if ((vma->vm_flags & VM_ACCOUNT) && 3176c9d13f5fSChen Gang security_vm_enough_memory_mm(mm, vma_pages(vma))) 3177c9d13f5fSChen Gang return -ENOMEM; 3178c9d13f5fSChen Gang 31791da177e4SLinus Torvalds /* 31801da177e4SLinus Torvalds * The vm_pgoff of a purely anonymous vma should be irrelevant 31811da177e4SLinus Torvalds * until its first write fault, when page's anon_vma and index 31821da177e4SLinus Torvalds * are set. But now set the vm_pgoff it will almost certainly 31831da177e4SLinus Torvalds * end up with (unless mremap moves it elsewhere before that 31841da177e4SLinus Torvalds * first wfault), so /proc/pid/maps tells a consistent story. 31851da177e4SLinus Torvalds * 31861da177e4SLinus Torvalds * By setting it to reflect the virtual start address of the 31871da177e4SLinus Torvalds * vma, merges and splits can happen in a seamless way, just 31881da177e4SLinus Torvalds * using the existing file pgoff checks and manipulations. 31898332326eSLiao Pingfang * Similarly in do_mmap and in do_brk_flags. 31901da177e4SLinus Torvalds */ 31918a9cc3b5SOleg Nesterov if (vma_is_anonymous(vma)) { 31921da177e4SLinus Torvalds BUG_ON(vma->anon_vma); 31931da177e4SLinus Torvalds vma->vm_pgoff = vma->vm_start >> PAGE_SHIFT; 31941da177e4SLinus Torvalds } 31952b144498SSrikar Dronamraju 31961da177e4SLinus Torvalds vma_link(mm, vma, prev, rb_link, rb_parent); 31971da177e4SLinus Torvalds return 0; 31981da177e4SLinus Torvalds } 31991da177e4SLinus Torvalds 32001da177e4SLinus Torvalds /* 32011da177e4SLinus Torvalds * Copy the vma structure to a new location in the same mm, 32021da177e4SLinus Torvalds * prior to moving page table entries, to effect an mremap move. 32031da177e4SLinus Torvalds */ 32041da177e4SLinus Torvalds struct vm_area_struct *copy_vma(struct vm_area_struct **vmap, 320538a76013SMichel Lespinasse unsigned long addr, unsigned long len, pgoff_t pgoff, 320638a76013SMichel Lespinasse bool *need_rmap_locks) 32071da177e4SLinus Torvalds { 32081da177e4SLinus Torvalds struct vm_area_struct *vma = *vmap; 32091da177e4SLinus Torvalds unsigned long vma_start = vma->vm_start; 32101da177e4SLinus Torvalds struct mm_struct *mm = vma->vm_mm; 32111da177e4SLinus Torvalds struct vm_area_struct *new_vma, *prev; 32121da177e4SLinus Torvalds struct rb_node **rb_link, *rb_parent; 3213948f017bSAndrea Arcangeli bool faulted_in_anon_vma = true; 32141da177e4SLinus Torvalds 32151da177e4SLinus Torvalds /* 32161da177e4SLinus Torvalds * If anonymous vma has not yet been faulted, update new pgoff 32171da177e4SLinus Torvalds * to match new location, to increase its chance of merging. 32181da177e4SLinus Torvalds */ 3219ce75799bSOleg Nesterov if (unlikely(vma_is_anonymous(vma) && !vma->anon_vma)) { 32201da177e4SLinus Torvalds pgoff = addr >> PAGE_SHIFT; 3221948f017bSAndrea Arcangeli faulted_in_anon_vma = false; 3222948f017bSAndrea Arcangeli } 32231da177e4SLinus Torvalds 32246597d783SHugh Dickins if (find_vma_links(mm, addr, addr + len, &prev, &rb_link, &rb_parent)) 32256597d783SHugh Dickins return NULL; /* should never get here */ 32261da177e4SLinus Torvalds new_vma = vma_merge(mm, prev, addr, addr + len, vma->vm_flags, 322719a809afSAndrea Arcangeli vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma), 32285c26f6acSSuren Baghdasaryan vma->vm_userfaultfd_ctx, anon_vma_name(vma)); 32291da177e4SLinus Torvalds if (new_vma) { 32301da177e4SLinus Torvalds /* 32311da177e4SLinus Torvalds * Source vma may have been merged into new_vma 32321da177e4SLinus Torvalds */ 3233948f017bSAndrea Arcangeli if (unlikely(vma_start >= new_vma->vm_start && 3234948f017bSAndrea Arcangeli vma_start < new_vma->vm_end)) { 3235948f017bSAndrea Arcangeli /* 3236948f017bSAndrea Arcangeli * The only way we can get a vma_merge with 3237948f017bSAndrea Arcangeli * self during an mremap is if the vma hasn't 3238948f017bSAndrea Arcangeli * been faulted in yet and we were allowed to 3239948f017bSAndrea Arcangeli * reset the dst vma->vm_pgoff to the 3240948f017bSAndrea Arcangeli * destination address of the mremap to allow 3241948f017bSAndrea Arcangeli * the merge to happen. mremap must change the 3242948f017bSAndrea Arcangeli * vm_pgoff linearity between src and dst vmas 3243948f017bSAndrea Arcangeli * (in turn preventing a vma_merge) to be 3244948f017bSAndrea Arcangeli * safe. It is only safe to keep the vm_pgoff 3245948f017bSAndrea Arcangeli * linear if there are no pages mapped yet. 3246948f017bSAndrea Arcangeli */ 324781d1b09cSSasha Levin VM_BUG_ON_VMA(faulted_in_anon_vma, new_vma); 324838a76013SMichel Lespinasse *vmap = vma = new_vma; 3249108d6642SMichel Lespinasse } 325038a76013SMichel Lespinasse *need_rmap_locks = (new_vma->vm_pgoff <= vma->vm_pgoff); 32511da177e4SLinus Torvalds } else { 32523928d4f5SLinus Torvalds new_vma = vm_area_dup(vma); 3253e3975891SChen Gang if (!new_vma) 3254e3975891SChen Gang goto out; 32551da177e4SLinus Torvalds new_vma->vm_start = addr; 32561da177e4SLinus Torvalds new_vma->vm_end = addr + len; 32571da177e4SLinus Torvalds new_vma->vm_pgoff = pgoff; 3258ef0855d3SOleg Nesterov if (vma_dup_policy(vma, new_vma)) 3259523d4e20SMichel Lespinasse goto out_free_vma; 3260523d4e20SMichel Lespinasse if (anon_vma_clone(new_vma, vma)) 3261523d4e20SMichel Lespinasse goto out_free_mempol; 3262e9714acfSKonstantin Khlebnikov if (new_vma->vm_file) 32631da177e4SLinus Torvalds get_file(new_vma->vm_file); 32641da177e4SLinus Torvalds if (new_vma->vm_ops && new_vma->vm_ops->open) 32651da177e4SLinus Torvalds new_vma->vm_ops->open(new_vma); 32661da177e4SLinus Torvalds vma_link(mm, new_vma, prev, rb_link, rb_parent); 326738a76013SMichel Lespinasse *need_rmap_locks = false; 32681da177e4SLinus Torvalds } 32691da177e4SLinus Torvalds return new_vma; 32705beb4930SRik van Riel 32715beb4930SRik van Riel out_free_mempol: 3272ef0855d3SOleg Nesterov mpol_put(vma_policy(new_vma)); 32735beb4930SRik van Riel out_free_vma: 32743928d4f5SLinus Torvalds vm_area_free(new_vma); 3275e3975891SChen Gang out: 32765beb4930SRik van Riel return NULL; 32771da177e4SLinus Torvalds } 3278119f657cSakpm@osdl.org 3279119f657cSakpm@osdl.org /* 3280119f657cSakpm@osdl.org * Return true if the calling process may expand its vm space by the passed 3281119f657cSakpm@osdl.org * number of pages 3282119f657cSakpm@osdl.org */ 328384638335SKonstantin Khlebnikov bool may_expand_vm(struct mm_struct *mm, vm_flags_t flags, unsigned long npages) 3284119f657cSakpm@osdl.org { 328584638335SKonstantin Khlebnikov if (mm->total_vm + npages > rlimit(RLIMIT_AS) >> PAGE_SHIFT) 328684638335SKonstantin Khlebnikov return false; 3287119f657cSakpm@osdl.org 3288d977d56cSKonstantin Khlebnikov if (is_data_mapping(flags) && 3289d977d56cSKonstantin Khlebnikov mm->data_vm + npages > rlimit(RLIMIT_DATA) >> PAGE_SHIFT) { 3290f4fcd558SKonstantin Khlebnikov /* Workaround for Valgrind */ 3291f4fcd558SKonstantin Khlebnikov if (rlimit(RLIMIT_DATA) == 0 && 3292f4fcd558SKonstantin Khlebnikov mm->data_vm + npages <= rlimit_max(RLIMIT_DATA) >> PAGE_SHIFT) 3293f4fcd558SKonstantin Khlebnikov return true; 329457a7702bSDavid Woodhouse 329557a7702bSDavid Woodhouse pr_warn_once("%s (%d): VmData %lu exceed data ulimit %lu. Update limits%s.\n", 3296d977d56cSKonstantin Khlebnikov current->comm, current->pid, 3297d977d56cSKonstantin Khlebnikov (mm->data_vm + npages) << PAGE_SHIFT, 329857a7702bSDavid Woodhouse rlimit(RLIMIT_DATA), 329957a7702bSDavid Woodhouse ignore_rlimit_data ? "" : " or use boot option ignore_rlimit_data"); 330057a7702bSDavid Woodhouse 330157a7702bSDavid Woodhouse if (!ignore_rlimit_data) 3302d977d56cSKonstantin Khlebnikov return false; 3303d977d56cSKonstantin Khlebnikov } 3304119f657cSakpm@osdl.org 330584638335SKonstantin Khlebnikov return true; 330684638335SKonstantin Khlebnikov } 330784638335SKonstantin Khlebnikov 330884638335SKonstantin Khlebnikov void vm_stat_account(struct mm_struct *mm, vm_flags_t flags, long npages) 330984638335SKonstantin Khlebnikov { 33107866076bSPeng Liu WRITE_ONCE(mm->total_vm, READ_ONCE(mm->total_vm)+npages); 331184638335SKonstantin Khlebnikov 3312d977d56cSKonstantin Khlebnikov if (is_exec_mapping(flags)) 331384638335SKonstantin Khlebnikov mm->exec_vm += npages; 3314d977d56cSKonstantin Khlebnikov else if (is_stack_mapping(flags)) 331584638335SKonstantin Khlebnikov mm->stack_vm += npages; 3316d977d56cSKonstantin Khlebnikov else if (is_data_mapping(flags)) 331784638335SKonstantin Khlebnikov mm->data_vm += npages; 3318119f657cSakpm@osdl.org } 3319fa5dc22fSRoland McGrath 3320b3ec9f33SSouptick Joarder static vm_fault_t special_mapping_fault(struct vm_fault *vmf); 3321a62c34bdSAndy Lutomirski 3322a62c34bdSAndy Lutomirski /* 3323a62c34bdSAndy Lutomirski * Having a close hook prevents vma merging regardless of flags. 3324a62c34bdSAndy Lutomirski */ 3325a62c34bdSAndy Lutomirski static void special_mapping_close(struct vm_area_struct *vma) 3326a62c34bdSAndy Lutomirski { 3327a62c34bdSAndy Lutomirski } 3328a62c34bdSAndy Lutomirski 3329a62c34bdSAndy Lutomirski static const char *special_mapping_name(struct vm_area_struct *vma) 3330a62c34bdSAndy Lutomirski { 3331a62c34bdSAndy Lutomirski return ((struct vm_special_mapping *)vma->vm_private_data)->name; 3332a62c34bdSAndy Lutomirski } 3333a62c34bdSAndy Lutomirski 333414d07113SBrian Geffon static int special_mapping_mremap(struct vm_area_struct *new_vma) 3335b059a453SDmitry Safonov { 3336b059a453SDmitry Safonov struct vm_special_mapping *sm = new_vma->vm_private_data; 3337b059a453SDmitry Safonov 3338280e87e9SDmitry Safonov if (WARN_ON_ONCE(current->mm != new_vma->vm_mm)) 3339280e87e9SDmitry Safonov return -EFAULT; 3340280e87e9SDmitry Safonov 3341b059a453SDmitry Safonov if (sm->mremap) 3342b059a453SDmitry Safonov return sm->mremap(sm, new_vma); 3343280e87e9SDmitry Safonov 3344b059a453SDmitry Safonov return 0; 3345b059a453SDmitry Safonov } 3346b059a453SDmitry Safonov 3347871402e0SDmitry Safonov static int special_mapping_split(struct vm_area_struct *vma, unsigned long addr) 3348871402e0SDmitry Safonov { 3349871402e0SDmitry Safonov /* 3350871402e0SDmitry Safonov * Forbid splitting special mappings - kernel has expectations over 3351871402e0SDmitry Safonov * the number of pages in mapping. Together with VM_DONTEXPAND 3352871402e0SDmitry Safonov * the size of vma should stay the same over the special mapping's 3353871402e0SDmitry Safonov * lifetime. 3354871402e0SDmitry Safonov */ 3355871402e0SDmitry Safonov return -EINVAL; 3356871402e0SDmitry Safonov } 3357871402e0SDmitry Safonov 3358a62c34bdSAndy Lutomirski static const struct vm_operations_struct special_mapping_vmops = { 3359a62c34bdSAndy Lutomirski .close = special_mapping_close, 3360a62c34bdSAndy Lutomirski .fault = special_mapping_fault, 3361b059a453SDmitry Safonov .mremap = special_mapping_mremap, 3362a62c34bdSAndy Lutomirski .name = special_mapping_name, 3363af34ebebSDmitry Safonov /* vDSO code relies that VVAR can't be accessed remotely */ 3364af34ebebSDmitry Safonov .access = NULL, 3365871402e0SDmitry Safonov .may_split = special_mapping_split, 3366a62c34bdSAndy Lutomirski }; 3367a62c34bdSAndy Lutomirski 3368a62c34bdSAndy Lutomirski static const struct vm_operations_struct legacy_special_mapping_vmops = { 3369a62c34bdSAndy Lutomirski .close = special_mapping_close, 3370a62c34bdSAndy Lutomirski .fault = special_mapping_fault, 3371a62c34bdSAndy Lutomirski }; 3372fa5dc22fSRoland McGrath 3373b3ec9f33SSouptick Joarder static vm_fault_t special_mapping_fault(struct vm_fault *vmf) 3374fa5dc22fSRoland McGrath { 337511bac800SDave Jiang struct vm_area_struct *vma = vmf->vma; 3376b1d0e4f5SNick Piggin pgoff_t pgoff; 3377fa5dc22fSRoland McGrath struct page **pages; 3378fa5dc22fSRoland McGrath 3379f872f540SAndy Lutomirski if (vma->vm_ops == &legacy_special_mapping_vmops) { 3380a62c34bdSAndy Lutomirski pages = vma->vm_private_data; 3381f872f540SAndy Lutomirski } else { 3382f872f540SAndy Lutomirski struct vm_special_mapping *sm = vma->vm_private_data; 3383f872f540SAndy Lutomirski 3384f872f540SAndy Lutomirski if (sm->fault) 338511bac800SDave Jiang return sm->fault(sm, vmf->vma, vmf); 3386f872f540SAndy Lutomirski 3387f872f540SAndy Lutomirski pages = sm->pages; 3388f872f540SAndy Lutomirski } 3389a62c34bdSAndy Lutomirski 33908a9cc3b5SOleg Nesterov for (pgoff = vmf->pgoff; pgoff && *pages; ++pages) 3391b1d0e4f5SNick Piggin pgoff--; 3392fa5dc22fSRoland McGrath 3393fa5dc22fSRoland McGrath if (*pages) { 3394fa5dc22fSRoland McGrath struct page *page = *pages; 3395fa5dc22fSRoland McGrath get_page(page); 3396b1d0e4f5SNick Piggin vmf->page = page; 3397b1d0e4f5SNick Piggin return 0; 3398fa5dc22fSRoland McGrath } 3399fa5dc22fSRoland McGrath 3400b1d0e4f5SNick Piggin return VM_FAULT_SIGBUS; 3401fa5dc22fSRoland McGrath } 3402fa5dc22fSRoland McGrath 3403a62c34bdSAndy Lutomirski static struct vm_area_struct *__install_special_mapping( 3404a62c34bdSAndy Lutomirski struct mm_struct *mm, 3405fa5dc22fSRoland McGrath unsigned long addr, unsigned long len, 340627f28b97SChen Gang unsigned long vm_flags, void *priv, 340727f28b97SChen Gang const struct vm_operations_struct *ops) 3408fa5dc22fSRoland McGrath { 3409462e635eSTavis Ormandy int ret; 3410fa5dc22fSRoland McGrath struct vm_area_struct *vma; 3411fa5dc22fSRoland McGrath 3412490fc053SLinus Torvalds vma = vm_area_alloc(mm); 3413fa5dc22fSRoland McGrath if (unlikely(vma == NULL)) 34143935ed6aSStefani Seibold return ERR_PTR(-ENOMEM); 3415fa5dc22fSRoland McGrath 3416fa5dc22fSRoland McGrath vma->vm_start = addr; 3417fa5dc22fSRoland McGrath vma->vm_end = addr + len; 3418fa5dc22fSRoland McGrath 3419d9104d1cSCyrill Gorcunov vma->vm_flags = vm_flags | mm->def_flags | VM_DONTEXPAND | VM_SOFTDIRTY; 34201fc09228SHugh Dickins vma->vm_flags &= VM_LOCKED_CLEAR_MASK; 34213ed75eb8SColy Li vma->vm_page_prot = vm_get_page_prot(vma->vm_flags); 3422fa5dc22fSRoland McGrath 3423a62c34bdSAndy Lutomirski vma->vm_ops = ops; 3424a62c34bdSAndy Lutomirski vma->vm_private_data = priv; 3425fa5dc22fSRoland McGrath 3426462e635eSTavis Ormandy ret = insert_vm_struct(mm, vma); 3427462e635eSTavis Ormandy if (ret) 3428462e635eSTavis Ormandy goto out; 3429fa5dc22fSRoland McGrath 343084638335SKonstantin Khlebnikov vm_stat_account(mm, vma->vm_flags, len >> PAGE_SHIFT); 3431fa5dc22fSRoland McGrath 3432cdd6c482SIngo Molnar perf_event_mmap(vma); 3433089dd79dSPeter Zijlstra 34343935ed6aSStefani Seibold return vma; 3435462e635eSTavis Ormandy 3436462e635eSTavis Ormandy out: 34373928d4f5SLinus Torvalds vm_area_free(vma); 34383935ed6aSStefani Seibold return ERR_PTR(ret); 34393935ed6aSStefani Seibold } 34403935ed6aSStefani Seibold 34412eefd878SDmitry Safonov bool vma_is_special_mapping(const struct vm_area_struct *vma, 34422eefd878SDmitry Safonov const struct vm_special_mapping *sm) 34432eefd878SDmitry Safonov { 34442eefd878SDmitry Safonov return vma->vm_private_data == sm && 34452eefd878SDmitry Safonov (vma->vm_ops == &special_mapping_vmops || 34462eefd878SDmitry Safonov vma->vm_ops == &legacy_special_mapping_vmops); 34472eefd878SDmitry Safonov } 34482eefd878SDmitry Safonov 3449a62c34bdSAndy Lutomirski /* 3450c1e8d7c6SMichel Lespinasse * Called with mm->mmap_lock held for writing. 3451a62c34bdSAndy Lutomirski * Insert a new vma covering the given region, with the given flags. 3452a62c34bdSAndy Lutomirski * Its pages are supplied by the given array of struct page *. 3453a62c34bdSAndy Lutomirski * The array can be shorter than len >> PAGE_SHIFT if it's null-terminated. 3454a62c34bdSAndy Lutomirski * The region past the last page supplied will always produce SIGBUS. 3455a62c34bdSAndy Lutomirski * The array pointer and the pages it points to are assumed to stay alive 3456a62c34bdSAndy Lutomirski * for as long as this mapping might exist. 3457a62c34bdSAndy Lutomirski */ 3458a62c34bdSAndy Lutomirski struct vm_area_struct *_install_special_mapping( 3459a62c34bdSAndy Lutomirski struct mm_struct *mm, 3460a62c34bdSAndy Lutomirski unsigned long addr, unsigned long len, 3461a62c34bdSAndy Lutomirski unsigned long vm_flags, const struct vm_special_mapping *spec) 3462a62c34bdSAndy Lutomirski { 346327f28b97SChen Gang return __install_special_mapping(mm, addr, len, vm_flags, (void *)spec, 346427f28b97SChen Gang &special_mapping_vmops); 3465a62c34bdSAndy Lutomirski } 3466a62c34bdSAndy Lutomirski 34673935ed6aSStefani Seibold int install_special_mapping(struct mm_struct *mm, 34683935ed6aSStefani Seibold unsigned long addr, unsigned long len, 34693935ed6aSStefani Seibold unsigned long vm_flags, struct page **pages) 34703935ed6aSStefani Seibold { 3471a62c34bdSAndy Lutomirski struct vm_area_struct *vma = __install_special_mapping( 347227f28b97SChen Gang mm, addr, len, vm_flags, (void *)pages, 347327f28b97SChen Gang &legacy_special_mapping_vmops); 34743935ed6aSStefani Seibold 347514bd5b45SDuan Jiong return PTR_ERR_OR_ZERO(vma); 3476fa5dc22fSRoland McGrath } 34777906d00cSAndrea Arcangeli 34787906d00cSAndrea Arcangeli static DEFINE_MUTEX(mm_all_locks_mutex); 34797906d00cSAndrea Arcangeli 3480454ed842SPeter Zijlstra static void vm_lock_anon_vma(struct mm_struct *mm, struct anon_vma *anon_vma) 34817906d00cSAndrea Arcangeli { 3482f808c13fSDavidlohr Bueso if (!test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_root.rb_node)) { 34837906d00cSAndrea Arcangeli /* 34847906d00cSAndrea Arcangeli * The LSB of head.next can't change from under us 34857906d00cSAndrea Arcangeli * because we hold the mm_all_locks_mutex. 34867906d00cSAndrea Arcangeli */ 3487da1c55f1SMichel Lespinasse down_write_nest_lock(&anon_vma->root->rwsem, &mm->mmap_lock); 34887906d00cSAndrea Arcangeli /* 34897906d00cSAndrea Arcangeli * We can safely modify head.next after taking the 34905a505085SIngo Molnar * anon_vma->root->rwsem. If some other vma in this mm shares 34917906d00cSAndrea Arcangeli * the same anon_vma we won't take it again. 34927906d00cSAndrea Arcangeli * 34937906d00cSAndrea Arcangeli * No need of atomic instructions here, head.next 34947906d00cSAndrea Arcangeli * can't change from under us thanks to the 34955a505085SIngo Molnar * anon_vma->root->rwsem. 34967906d00cSAndrea Arcangeli */ 34977906d00cSAndrea Arcangeli if (__test_and_set_bit(0, (unsigned long *) 3498f808c13fSDavidlohr Bueso &anon_vma->root->rb_root.rb_root.rb_node)) 34997906d00cSAndrea Arcangeli BUG(); 35007906d00cSAndrea Arcangeli } 35017906d00cSAndrea Arcangeli } 35027906d00cSAndrea Arcangeli 3503454ed842SPeter Zijlstra static void vm_lock_mapping(struct mm_struct *mm, struct address_space *mapping) 35047906d00cSAndrea Arcangeli { 35057906d00cSAndrea Arcangeli if (!test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) { 35067906d00cSAndrea Arcangeli /* 35077906d00cSAndrea Arcangeli * AS_MM_ALL_LOCKS can't change from under us because 35087906d00cSAndrea Arcangeli * we hold the mm_all_locks_mutex. 35097906d00cSAndrea Arcangeli * 35107906d00cSAndrea Arcangeli * Operations on ->flags have to be atomic because 35117906d00cSAndrea Arcangeli * even if AS_MM_ALL_LOCKS is stable thanks to the 35127906d00cSAndrea Arcangeli * mm_all_locks_mutex, there may be other cpus 35137906d00cSAndrea Arcangeli * changing other bitflags in parallel to us. 35147906d00cSAndrea Arcangeli */ 35157906d00cSAndrea Arcangeli if (test_and_set_bit(AS_MM_ALL_LOCKS, &mapping->flags)) 35167906d00cSAndrea Arcangeli BUG(); 3517da1c55f1SMichel Lespinasse down_write_nest_lock(&mapping->i_mmap_rwsem, &mm->mmap_lock); 35187906d00cSAndrea Arcangeli } 35197906d00cSAndrea Arcangeli } 35207906d00cSAndrea Arcangeli 35217906d00cSAndrea Arcangeli /* 35227906d00cSAndrea Arcangeli * This operation locks against the VM for all pte/vma/mm related 35237906d00cSAndrea Arcangeli * operations that could ever happen on a certain mm. This includes 35247906d00cSAndrea Arcangeli * vmtruncate, try_to_unmap, and all page faults. 35257906d00cSAndrea Arcangeli * 3526c1e8d7c6SMichel Lespinasse * The caller must take the mmap_lock in write mode before calling 35277906d00cSAndrea Arcangeli * mm_take_all_locks(). The caller isn't allowed to release the 3528c1e8d7c6SMichel Lespinasse * mmap_lock until mm_drop_all_locks() returns. 35297906d00cSAndrea Arcangeli * 3530c1e8d7c6SMichel Lespinasse * mmap_lock in write mode is required in order to block all operations 35317906d00cSAndrea Arcangeli * that could modify pagetables and free pages without need of 353227ba0644SKirill A. Shutemov * altering the vma layout. It's also needed in write mode to avoid new 35337906d00cSAndrea Arcangeli * anon_vmas to be associated with existing vmas. 35347906d00cSAndrea Arcangeli * 35357906d00cSAndrea Arcangeli * A single task can't take more than one mm_take_all_locks() in a row 35367906d00cSAndrea Arcangeli * or it would deadlock. 35377906d00cSAndrea Arcangeli * 3538bf181b9fSMichel Lespinasse * The LSB in anon_vma->rb_root.rb_node and the AS_MM_ALL_LOCKS bitflag in 35397906d00cSAndrea Arcangeli * mapping->flags avoid to take the same lock twice, if more than one 35407906d00cSAndrea Arcangeli * vma in this mm is backed by the same anon_vma or address_space. 35417906d00cSAndrea Arcangeli * 354288f306b6SKirill A. Shutemov * We take locks in following order, accordingly to comment at beginning 354388f306b6SKirill A. Shutemov * of mm/rmap.c: 354488f306b6SKirill A. Shutemov * - all hugetlbfs_i_mmap_rwsem_key locks (aka mapping->i_mmap_rwsem for 354588f306b6SKirill A. Shutemov * hugetlb mapping); 354688f306b6SKirill A. Shutemov * - all i_mmap_rwsem locks; 354788f306b6SKirill A. Shutemov * - all anon_vma->rwseml 354888f306b6SKirill A. Shutemov * 354988f306b6SKirill A. Shutemov * We can take all locks within these types randomly because the VM code 355088f306b6SKirill A. Shutemov * doesn't nest them and we protected from parallel mm_take_all_locks() by 355188f306b6SKirill A. Shutemov * mm_all_locks_mutex. 35527906d00cSAndrea Arcangeli * 35537906d00cSAndrea Arcangeli * mm_take_all_locks() and mm_drop_all_locks are expensive operations 35547906d00cSAndrea Arcangeli * that may have to take thousand of locks. 35557906d00cSAndrea Arcangeli * 35567906d00cSAndrea Arcangeli * mm_take_all_locks() can fail if it's interrupted by signals. 35577906d00cSAndrea Arcangeli */ 35587906d00cSAndrea Arcangeli int mm_take_all_locks(struct mm_struct *mm) 35597906d00cSAndrea Arcangeli { 35607906d00cSAndrea Arcangeli struct vm_area_struct *vma; 35615beb4930SRik van Riel struct anon_vma_chain *avc; 35627906d00cSAndrea Arcangeli 3563325bca1fSRolf Eike Beer mmap_assert_write_locked(mm); 35647906d00cSAndrea Arcangeli 35657906d00cSAndrea Arcangeli mutex_lock(&mm_all_locks_mutex); 35667906d00cSAndrea Arcangeli 35677906d00cSAndrea Arcangeli for (vma = mm->mmap; vma; vma = vma->vm_next) { 35687906d00cSAndrea Arcangeli if (signal_pending(current)) 35697906d00cSAndrea Arcangeli goto out_unlock; 357088f306b6SKirill A. Shutemov if (vma->vm_file && vma->vm_file->f_mapping && 357188f306b6SKirill A. Shutemov is_vm_hugetlb_page(vma)) 357288f306b6SKirill A. Shutemov vm_lock_mapping(mm, vma->vm_file->f_mapping); 357388f306b6SKirill A. Shutemov } 357488f306b6SKirill A. Shutemov 357588f306b6SKirill A. Shutemov for (vma = mm->mmap; vma; vma = vma->vm_next) { 357688f306b6SKirill A. Shutemov if (signal_pending(current)) 357788f306b6SKirill A. Shutemov goto out_unlock; 357888f306b6SKirill A. Shutemov if (vma->vm_file && vma->vm_file->f_mapping && 357988f306b6SKirill A. Shutemov !is_vm_hugetlb_page(vma)) 3580454ed842SPeter Zijlstra vm_lock_mapping(mm, vma->vm_file->f_mapping); 35817906d00cSAndrea Arcangeli } 35827cd5a02fSPeter Zijlstra 35837cd5a02fSPeter Zijlstra for (vma = mm->mmap; vma; vma = vma->vm_next) { 35847cd5a02fSPeter Zijlstra if (signal_pending(current)) 35857cd5a02fSPeter Zijlstra goto out_unlock; 35867cd5a02fSPeter Zijlstra if (vma->anon_vma) 35875beb4930SRik van Riel list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) 35885beb4930SRik van Riel vm_lock_anon_vma(mm, avc->anon_vma); 35897cd5a02fSPeter Zijlstra } 35907cd5a02fSPeter Zijlstra 3591584cff54SKautuk Consul return 0; 35927906d00cSAndrea Arcangeli 35937906d00cSAndrea Arcangeli out_unlock: 35947906d00cSAndrea Arcangeli mm_drop_all_locks(mm); 3595584cff54SKautuk Consul return -EINTR; 35967906d00cSAndrea Arcangeli } 35977906d00cSAndrea Arcangeli 35987906d00cSAndrea Arcangeli static void vm_unlock_anon_vma(struct anon_vma *anon_vma) 35997906d00cSAndrea Arcangeli { 3600f808c13fSDavidlohr Bueso if (test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_root.rb_node)) { 36017906d00cSAndrea Arcangeli /* 36027906d00cSAndrea Arcangeli * The LSB of head.next can't change to 0 from under 36037906d00cSAndrea Arcangeli * us because we hold the mm_all_locks_mutex. 36047906d00cSAndrea Arcangeli * 36057906d00cSAndrea Arcangeli * We must however clear the bitflag before unlocking 3606bf181b9fSMichel Lespinasse * the vma so the users using the anon_vma->rb_root will 36077906d00cSAndrea Arcangeli * never see our bitflag. 36087906d00cSAndrea Arcangeli * 36097906d00cSAndrea Arcangeli * No need of atomic instructions here, head.next 36107906d00cSAndrea Arcangeli * can't change from under us until we release the 36115a505085SIngo Molnar * anon_vma->root->rwsem. 36127906d00cSAndrea Arcangeli */ 36137906d00cSAndrea Arcangeli if (!__test_and_clear_bit(0, (unsigned long *) 3614f808c13fSDavidlohr Bueso &anon_vma->root->rb_root.rb_root.rb_node)) 36157906d00cSAndrea Arcangeli BUG(); 361608b52706SKonstantin Khlebnikov anon_vma_unlock_write(anon_vma); 36177906d00cSAndrea Arcangeli } 36187906d00cSAndrea Arcangeli } 36197906d00cSAndrea Arcangeli 36207906d00cSAndrea Arcangeli static void vm_unlock_mapping(struct address_space *mapping) 36217906d00cSAndrea Arcangeli { 36227906d00cSAndrea Arcangeli if (test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) { 36237906d00cSAndrea Arcangeli /* 36247906d00cSAndrea Arcangeli * AS_MM_ALL_LOCKS can't change to 0 from under us 36257906d00cSAndrea Arcangeli * because we hold the mm_all_locks_mutex. 36267906d00cSAndrea Arcangeli */ 362783cde9e8SDavidlohr Bueso i_mmap_unlock_write(mapping); 36287906d00cSAndrea Arcangeli if (!test_and_clear_bit(AS_MM_ALL_LOCKS, 36297906d00cSAndrea Arcangeli &mapping->flags)) 36307906d00cSAndrea Arcangeli BUG(); 36317906d00cSAndrea Arcangeli } 36327906d00cSAndrea Arcangeli } 36337906d00cSAndrea Arcangeli 36347906d00cSAndrea Arcangeli /* 3635c1e8d7c6SMichel Lespinasse * The mmap_lock cannot be released by the caller until 36367906d00cSAndrea Arcangeli * mm_drop_all_locks() returns. 36377906d00cSAndrea Arcangeli */ 36387906d00cSAndrea Arcangeli void mm_drop_all_locks(struct mm_struct *mm) 36397906d00cSAndrea Arcangeli { 36407906d00cSAndrea Arcangeli struct vm_area_struct *vma; 36415beb4930SRik van Riel struct anon_vma_chain *avc; 36427906d00cSAndrea Arcangeli 3643325bca1fSRolf Eike Beer mmap_assert_write_locked(mm); 36447906d00cSAndrea Arcangeli BUG_ON(!mutex_is_locked(&mm_all_locks_mutex)); 36457906d00cSAndrea Arcangeli 36467906d00cSAndrea Arcangeli for (vma = mm->mmap; vma; vma = vma->vm_next) { 36477906d00cSAndrea Arcangeli if (vma->anon_vma) 36485beb4930SRik van Riel list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) 36495beb4930SRik van Riel vm_unlock_anon_vma(avc->anon_vma); 36507906d00cSAndrea Arcangeli if (vma->vm_file && vma->vm_file->f_mapping) 36517906d00cSAndrea Arcangeli vm_unlock_mapping(vma->vm_file->f_mapping); 36527906d00cSAndrea Arcangeli } 36537906d00cSAndrea Arcangeli 36547906d00cSAndrea Arcangeli mutex_unlock(&mm_all_locks_mutex); 36557906d00cSAndrea Arcangeli } 36568feae131SDavid Howells 36578feae131SDavid Howells /* 36583edf41d8Sseokhoon.yoon * initialise the percpu counter for VM 36598feae131SDavid Howells */ 36608feae131SDavid Howells void __init mmap_init(void) 36618feae131SDavid Howells { 366200a62ce9SKOSAKI Motohiro int ret; 366300a62ce9SKOSAKI Motohiro 3664908c7f19STejun Heo ret = percpu_counter_init(&vm_committed_as, 0, GFP_KERNEL); 366500a62ce9SKOSAKI Motohiro VM_BUG_ON(ret); 36668feae131SDavid Howells } 3667c9b1d098SAndrew Shewmaker 3668c9b1d098SAndrew Shewmaker /* 3669c9b1d098SAndrew Shewmaker * Initialise sysctl_user_reserve_kbytes. 3670c9b1d098SAndrew Shewmaker * 3671c9b1d098SAndrew Shewmaker * This is intended to prevent a user from starting a single memory hogging 3672c9b1d098SAndrew Shewmaker * process, such that they cannot recover (kill the hog) in OVERCOMMIT_NEVER 3673c9b1d098SAndrew Shewmaker * mode. 3674c9b1d098SAndrew Shewmaker * 3675c9b1d098SAndrew Shewmaker * The default value is min(3% of free memory, 128MB) 3676c9b1d098SAndrew Shewmaker * 128MB is enough to recover with sshd/login, bash, and top/kill. 3677c9b1d098SAndrew Shewmaker */ 36781640879aSAndrew Shewmaker static int init_user_reserve(void) 3679c9b1d098SAndrew Shewmaker { 3680c9b1d098SAndrew Shewmaker unsigned long free_kbytes; 3681c9b1d098SAndrew Shewmaker 3682c41f012aSMichal Hocko free_kbytes = global_zone_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10); 3683c9b1d098SAndrew Shewmaker 3684c9b1d098SAndrew Shewmaker sysctl_user_reserve_kbytes = min(free_kbytes / 32, 1UL << 17); 3685c9b1d098SAndrew Shewmaker return 0; 3686c9b1d098SAndrew Shewmaker } 3687a64fb3cdSPaul Gortmaker subsys_initcall(init_user_reserve); 36884eeab4f5SAndrew Shewmaker 36894eeab4f5SAndrew Shewmaker /* 36904eeab4f5SAndrew Shewmaker * Initialise sysctl_admin_reserve_kbytes. 36914eeab4f5SAndrew Shewmaker * 36924eeab4f5SAndrew Shewmaker * The purpose of sysctl_admin_reserve_kbytes is to allow the sys admin 36934eeab4f5SAndrew Shewmaker * to log in and kill a memory hogging process. 36944eeab4f5SAndrew Shewmaker * 36954eeab4f5SAndrew Shewmaker * Systems with more than 256MB will reserve 8MB, enough to recover 36964eeab4f5SAndrew Shewmaker * with sshd, bash, and top in OVERCOMMIT_GUESS. Smaller systems will 36974eeab4f5SAndrew Shewmaker * only reserve 3% of free pages by default. 36984eeab4f5SAndrew Shewmaker */ 36991640879aSAndrew Shewmaker static int init_admin_reserve(void) 37004eeab4f5SAndrew Shewmaker { 37014eeab4f5SAndrew Shewmaker unsigned long free_kbytes; 37024eeab4f5SAndrew Shewmaker 3703c41f012aSMichal Hocko free_kbytes = global_zone_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10); 37044eeab4f5SAndrew Shewmaker 37054eeab4f5SAndrew Shewmaker sysctl_admin_reserve_kbytes = min(free_kbytes / 32, 1UL << 13); 37064eeab4f5SAndrew Shewmaker return 0; 37074eeab4f5SAndrew Shewmaker } 3708a64fb3cdSPaul Gortmaker subsys_initcall(init_admin_reserve); 37091640879aSAndrew Shewmaker 37101640879aSAndrew Shewmaker /* 37111640879aSAndrew Shewmaker * Reinititalise user and admin reserves if memory is added or removed. 37121640879aSAndrew Shewmaker * 37131640879aSAndrew Shewmaker * The default user reserve max is 128MB, and the default max for the 37141640879aSAndrew Shewmaker * admin reserve is 8MB. These are usually, but not always, enough to 37151640879aSAndrew Shewmaker * enable recovery from a memory hogging process using login/sshd, a shell, 37161640879aSAndrew Shewmaker * and tools like top. It may make sense to increase or even disable the 37171640879aSAndrew Shewmaker * reserve depending on the existence of swap or variations in the recovery 37181640879aSAndrew Shewmaker * tools. So, the admin may have changed them. 37191640879aSAndrew Shewmaker * 37201640879aSAndrew Shewmaker * If memory is added and the reserves have been eliminated or increased above 37211640879aSAndrew Shewmaker * the default max, then we'll trust the admin. 37221640879aSAndrew Shewmaker * 37231640879aSAndrew Shewmaker * If memory is removed and there isn't enough free memory, then we 37241640879aSAndrew Shewmaker * need to reset the reserves. 37251640879aSAndrew Shewmaker * 37261640879aSAndrew Shewmaker * Otherwise keep the reserve set by the admin. 37271640879aSAndrew Shewmaker */ 37281640879aSAndrew Shewmaker static int reserve_mem_notifier(struct notifier_block *nb, 37291640879aSAndrew Shewmaker unsigned long action, void *data) 37301640879aSAndrew Shewmaker { 37311640879aSAndrew Shewmaker unsigned long tmp, free_kbytes; 37321640879aSAndrew Shewmaker 37331640879aSAndrew Shewmaker switch (action) { 37341640879aSAndrew Shewmaker case MEM_ONLINE: 37351640879aSAndrew Shewmaker /* Default max is 128MB. Leave alone if modified by operator. */ 37361640879aSAndrew Shewmaker tmp = sysctl_user_reserve_kbytes; 37371640879aSAndrew Shewmaker if (0 < tmp && tmp < (1UL << 17)) 37381640879aSAndrew Shewmaker init_user_reserve(); 37391640879aSAndrew Shewmaker 37401640879aSAndrew Shewmaker /* Default max is 8MB. Leave alone if modified by operator. */ 37411640879aSAndrew Shewmaker tmp = sysctl_admin_reserve_kbytes; 37421640879aSAndrew Shewmaker if (0 < tmp && tmp < (1UL << 13)) 37431640879aSAndrew Shewmaker init_admin_reserve(); 37441640879aSAndrew Shewmaker 37451640879aSAndrew Shewmaker break; 37461640879aSAndrew Shewmaker case MEM_OFFLINE: 3747c41f012aSMichal Hocko free_kbytes = global_zone_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10); 37481640879aSAndrew Shewmaker 37491640879aSAndrew Shewmaker if (sysctl_user_reserve_kbytes > free_kbytes) { 37501640879aSAndrew Shewmaker init_user_reserve(); 37511640879aSAndrew Shewmaker pr_info("vm.user_reserve_kbytes reset to %lu\n", 37521640879aSAndrew Shewmaker sysctl_user_reserve_kbytes); 37531640879aSAndrew Shewmaker } 37541640879aSAndrew Shewmaker 37551640879aSAndrew Shewmaker if (sysctl_admin_reserve_kbytes > free_kbytes) { 37561640879aSAndrew Shewmaker init_admin_reserve(); 37571640879aSAndrew Shewmaker pr_info("vm.admin_reserve_kbytes reset to %lu\n", 37581640879aSAndrew Shewmaker sysctl_admin_reserve_kbytes); 37591640879aSAndrew Shewmaker } 37601640879aSAndrew Shewmaker break; 37611640879aSAndrew Shewmaker default: 37621640879aSAndrew Shewmaker break; 37631640879aSAndrew Shewmaker } 37641640879aSAndrew Shewmaker return NOTIFY_OK; 37651640879aSAndrew Shewmaker } 37661640879aSAndrew Shewmaker 37671640879aSAndrew Shewmaker static struct notifier_block reserve_mem_nb = { 37681640879aSAndrew Shewmaker .notifier_call = reserve_mem_notifier, 37691640879aSAndrew Shewmaker }; 37701640879aSAndrew Shewmaker 37711640879aSAndrew Shewmaker static int __meminit init_reserve_notifier(void) 37721640879aSAndrew Shewmaker { 37731640879aSAndrew Shewmaker if (register_hotmemory_notifier(&reserve_mem_nb)) 3774b1de0d13SMitchel Humpherys pr_err("Failed registering memory add/remove notifier for admin reserve\n"); 37751640879aSAndrew Shewmaker 37761640879aSAndrew Shewmaker return 0; 37771640879aSAndrew Shewmaker } 3778a64fb3cdSPaul Gortmaker subsys_initcall(init_reserve_notifier); 3779