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 8464e45507SPeter Feiner static pgprot_t vm_pgprot_modify(pgprot_t oldprot, unsigned long vm_flags) 8564e45507SPeter Feiner { 8664e45507SPeter Feiner return pgprot_modify(oldprot, vm_get_page_prot(vm_flags)); 8764e45507SPeter Feiner } 8864e45507SPeter Feiner 8964e45507SPeter Feiner /* Update vma->vm_page_prot to reflect vma->vm_flags. */ 9064e45507SPeter Feiner void vma_set_page_prot(struct vm_area_struct *vma) 9164e45507SPeter Feiner { 9264e45507SPeter Feiner unsigned long vm_flags = vma->vm_flags; 936d2329f8SAndrea Arcangeli pgprot_t vm_page_prot; 9464e45507SPeter Feiner 956d2329f8SAndrea Arcangeli vm_page_prot = vm_pgprot_modify(vma->vm_page_prot, vm_flags); 966d2329f8SAndrea Arcangeli if (vma_wants_writenotify(vma, vm_page_prot)) { 9764e45507SPeter Feiner vm_flags &= ~VM_SHARED; 986d2329f8SAndrea Arcangeli vm_page_prot = vm_pgprot_modify(vm_page_prot, vm_flags); 9964e45507SPeter Feiner } 100c1e8d7c6SMichel Lespinasse /* remove_protection_ptes reads vma->vm_page_prot without mmap_lock */ 1016d2329f8SAndrea Arcangeli WRITE_ONCE(vma->vm_page_prot, vm_page_prot); 10264e45507SPeter Feiner } 10364e45507SPeter Feiner 1041da177e4SLinus Torvalds /* 105c8c06efaSDavidlohr Bueso * Requires inode->i_mapping->i_mmap_rwsem 1061da177e4SLinus Torvalds */ 1071da177e4SLinus Torvalds static void __remove_shared_vm_struct(struct vm_area_struct *vma, 1081da177e4SLinus Torvalds struct file *file, struct address_space *mapping) 1091da177e4SLinus Torvalds { 1101da177e4SLinus Torvalds if (vma->vm_flags & VM_SHARED) 1114bb5f5d9SDavid Herrmann mapping_unmap_writable(mapping); 1121da177e4SLinus Torvalds 1131da177e4SLinus Torvalds flush_dcache_mmap_lock(mapping); 1146b2dbba8SMichel Lespinasse vma_interval_tree_remove(vma, &mapping->i_mmap); 1151da177e4SLinus Torvalds flush_dcache_mmap_unlock(mapping); 1161da177e4SLinus Torvalds } 1171da177e4SLinus Torvalds 1181da177e4SLinus Torvalds /* 1196b2dbba8SMichel Lespinasse * Unlink a file-based vm structure from its interval tree, to hide 120a8fb5618SHugh Dickins * vma from rmap and vmtruncate before freeing its page tables. 1211da177e4SLinus Torvalds */ 122a8fb5618SHugh Dickins void unlink_file_vma(struct vm_area_struct *vma) 1231da177e4SLinus Torvalds { 1241da177e4SLinus Torvalds struct file *file = vma->vm_file; 1251da177e4SLinus Torvalds 1261da177e4SLinus Torvalds if (file) { 1271da177e4SLinus Torvalds struct address_space *mapping = file->f_mapping; 12883cde9e8SDavidlohr Bueso i_mmap_lock_write(mapping); 1291da177e4SLinus Torvalds __remove_shared_vm_struct(vma, file, mapping); 13083cde9e8SDavidlohr Bueso i_mmap_unlock_write(mapping); 1311da177e4SLinus Torvalds } 132a8fb5618SHugh Dickins } 133a8fb5618SHugh Dickins 134a8fb5618SHugh Dickins /* 135a8fb5618SHugh Dickins * Close a vm structure and free it, returning the next. 136a8fb5618SHugh Dickins */ 137a8fb5618SHugh Dickins static struct vm_area_struct *remove_vma(struct vm_area_struct *vma) 138a8fb5618SHugh Dickins { 139a8fb5618SHugh Dickins struct vm_area_struct *next = vma->vm_next; 140a8fb5618SHugh Dickins 141a8fb5618SHugh Dickins might_sleep(); 1421da177e4SLinus Torvalds if (vma->vm_ops && vma->vm_ops->close) 1431da177e4SLinus Torvalds vma->vm_ops->close(vma); 144e9714acfSKonstantin Khlebnikov if (vma->vm_file) 145a8fb5618SHugh Dickins fput(vma->vm_file); 146f0be3d32SLee Schermerhorn mpol_put(vma_policy(vma)); 1473928d4f5SLinus Torvalds vm_area_free(vma); 148a8fb5618SHugh Dickins return next; 1491da177e4SLinus Torvalds } 1501da177e4SLinus Torvalds 151bb177a73SMichal Hocko static int do_brk_flags(unsigned long addr, unsigned long request, unsigned long flags, 152bb177a73SMichal Hocko struct list_head *uf); 1536a6160a7SHeiko Carstens SYSCALL_DEFINE1(brk, unsigned long, brk) 1541da177e4SLinus Torvalds { 1559bc8039eSYang Shi unsigned long newbrk, oldbrk, origbrk; 1561da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 1571be7107fSHugh Dickins struct vm_area_struct *next; 158a5b4592cSJiri Kosina unsigned long min_brk; 159128557ffSMichel Lespinasse bool populate; 1609bc8039eSYang Shi bool downgraded = false; 161897ab3e0SMike Rapoport LIST_HEAD(uf); 1621da177e4SLinus Torvalds 163d8ed45c5SMichel Lespinasse if (mmap_write_lock_killable(mm)) 164dc0ef0dfSMichal Hocko return -EINTR; 1651da177e4SLinus Torvalds 1669bc8039eSYang Shi origbrk = mm->brk; 1679bc8039eSYang Shi 168a5b4592cSJiri Kosina #ifdef CONFIG_COMPAT_BRK 1695520e894SJiri Kosina /* 1705520e894SJiri Kosina * CONFIG_COMPAT_BRK can still be overridden by setting 1715520e894SJiri Kosina * randomize_va_space to 2, which will still cause mm->start_brk 1725520e894SJiri Kosina * to be arbitrarily shifted 1735520e894SJiri Kosina */ 1744471a675SJiri Kosina if (current->brk_randomized) 1755520e894SJiri Kosina min_brk = mm->start_brk; 1765520e894SJiri Kosina else 1775520e894SJiri Kosina min_brk = mm->end_data; 178a5b4592cSJiri Kosina #else 179a5b4592cSJiri Kosina min_brk = mm->start_brk; 180a5b4592cSJiri Kosina #endif 181a5b4592cSJiri Kosina if (brk < min_brk) 1821da177e4SLinus Torvalds goto out; 1831e624196SRam Gupta 1841e624196SRam Gupta /* 1851e624196SRam Gupta * Check against rlimit here. If this check is done later after the test 1861e624196SRam Gupta * of oldbrk with newbrk then it can escape the test and let the data 1871e624196SRam Gupta * segment grow beyond its set limit the in case where the limit is 1881e624196SRam Gupta * not page aligned -Ram Gupta 1891e624196SRam Gupta */ 1908764b338SCyrill Gorcunov if (check_data_rlimit(rlimit(RLIMIT_DATA), brk, mm->start_brk, 1918764b338SCyrill Gorcunov mm->end_data, mm->start_data)) 1921e624196SRam Gupta goto out; 1931e624196SRam Gupta 1941da177e4SLinus Torvalds newbrk = PAGE_ALIGN(brk); 1951da177e4SLinus Torvalds oldbrk = PAGE_ALIGN(mm->brk); 1969bc8039eSYang Shi if (oldbrk == newbrk) { 1979bc8039eSYang Shi mm->brk = brk; 1989bc8039eSYang Shi goto success; 1999bc8039eSYang Shi } 2001da177e4SLinus Torvalds 2019bc8039eSYang Shi /* 2029bc8039eSYang Shi * Always allow shrinking brk. 203c1e8d7c6SMichel Lespinasse * __do_munmap() may downgrade mmap_lock to read. 2049bc8039eSYang Shi */ 2051da177e4SLinus Torvalds if (brk <= mm->brk) { 2069bc8039eSYang Shi int ret; 2079bc8039eSYang Shi 2089bc8039eSYang Shi /* 209c1e8d7c6SMichel Lespinasse * mm->brk must to be protected by write mmap_lock so update it 210c1e8d7c6SMichel Lespinasse * before downgrading mmap_lock. When __do_munmap() fails, 2119bc8039eSYang Shi * mm->brk will be restored from origbrk. 2129bc8039eSYang Shi */ 2139bc8039eSYang Shi mm->brk = brk; 2149bc8039eSYang Shi ret = __do_munmap(mm, newbrk, oldbrk-newbrk, &uf, true); 2159bc8039eSYang Shi if (ret < 0) { 2169bc8039eSYang Shi mm->brk = origbrk; 2171da177e4SLinus Torvalds goto out; 2189bc8039eSYang Shi } else if (ret == 1) { 2199bc8039eSYang Shi downgraded = true; 2209bc8039eSYang Shi } 2219bc8039eSYang Shi goto success; 2221da177e4SLinus Torvalds } 2231da177e4SLinus Torvalds 2241da177e4SLinus Torvalds /* Check against existing mmap mappings. */ 2251be7107fSHugh Dickins next = find_vma(mm, oldbrk); 2261be7107fSHugh Dickins if (next && newbrk + PAGE_SIZE > vm_start_gap(next)) 2271da177e4SLinus Torvalds goto out; 2281da177e4SLinus Torvalds 2291da177e4SLinus Torvalds /* Ok, looks good - let it rip. */ 230bb177a73SMichal Hocko if (do_brk_flags(oldbrk, newbrk-oldbrk, 0, &uf) < 0) 2311da177e4SLinus Torvalds goto out; 2321da177e4SLinus Torvalds mm->brk = brk; 2339bc8039eSYang Shi 2349bc8039eSYang Shi success: 235128557ffSMichel Lespinasse populate = newbrk > oldbrk && (mm->def_flags & VM_LOCKED) != 0; 2369bc8039eSYang Shi if (downgraded) 237d8ed45c5SMichel Lespinasse mmap_read_unlock(mm); 2389bc8039eSYang Shi else 239d8ed45c5SMichel Lespinasse mmap_write_unlock(mm); 240897ab3e0SMike Rapoport userfaultfd_unmap_complete(mm, &uf); 241128557ffSMichel Lespinasse if (populate) 242128557ffSMichel Lespinasse mm_populate(oldbrk, newbrk - oldbrk); 243128557ffSMichel Lespinasse return brk; 244128557ffSMichel Lespinasse 2451da177e4SLinus Torvalds out: 246d8ed45c5SMichel Lespinasse mmap_write_unlock(mm); 247b7204006SAdrian Huang return origbrk; 2481da177e4SLinus Torvalds } 2491da177e4SLinus Torvalds 250315cc066SMichel Lespinasse static inline unsigned long vma_compute_gap(struct vm_area_struct *vma) 251d3737187SMichel Lespinasse { 252315cc066SMichel Lespinasse unsigned long gap, prev_end; 2531be7107fSHugh Dickins 2541be7107fSHugh Dickins /* 2551be7107fSHugh Dickins * Note: in the rare case of a VM_GROWSDOWN above a VM_GROWSUP, we 2561be7107fSHugh Dickins * allow two stack_guard_gaps between them here, and when choosing 2571be7107fSHugh Dickins * an unmapped area; whereas when expanding we only require one. 2581be7107fSHugh Dickins * That's a little inconsistent, but keeps the code here simpler. 2591be7107fSHugh Dickins */ 260315cc066SMichel Lespinasse gap = vm_start_gap(vma); 2611be7107fSHugh Dickins if (vma->vm_prev) { 2621be7107fSHugh Dickins prev_end = vm_end_gap(vma->vm_prev); 263315cc066SMichel Lespinasse if (gap > prev_end) 264315cc066SMichel Lespinasse gap -= prev_end; 2651be7107fSHugh Dickins else 266315cc066SMichel Lespinasse gap = 0; 2671be7107fSHugh Dickins } 268315cc066SMichel Lespinasse return gap; 269315cc066SMichel Lespinasse } 270315cc066SMichel Lespinasse 271315cc066SMichel Lespinasse #ifdef CONFIG_DEBUG_VM_RB 272315cc066SMichel Lespinasse static unsigned long vma_compute_subtree_gap(struct vm_area_struct *vma) 273315cc066SMichel Lespinasse { 274315cc066SMichel Lespinasse unsigned long max = vma_compute_gap(vma), subtree_gap; 275d3737187SMichel Lespinasse if (vma->vm_rb.rb_left) { 276d3737187SMichel Lespinasse subtree_gap = rb_entry(vma->vm_rb.rb_left, 277d3737187SMichel Lespinasse struct vm_area_struct, vm_rb)->rb_subtree_gap; 278d3737187SMichel Lespinasse if (subtree_gap > max) 279d3737187SMichel Lespinasse max = subtree_gap; 280d3737187SMichel Lespinasse } 281d3737187SMichel Lespinasse if (vma->vm_rb.rb_right) { 282d3737187SMichel Lespinasse subtree_gap = rb_entry(vma->vm_rb.rb_right, 283d3737187SMichel Lespinasse struct vm_area_struct, vm_rb)->rb_subtree_gap; 284d3737187SMichel Lespinasse if (subtree_gap > max) 285d3737187SMichel Lespinasse max = subtree_gap; 286d3737187SMichel Lespinasse } 287d3737187SMichel Lespinasse return max; 288d3737187SMichel Lespinasse } 289d3737187SMichel Lespinasse 290acf128d0SAndrea Arcangeli static int browse_rb(struct mm_struct *mm) 2911da177e4SLinus Torvalds { 292acf128d0SAndrea Arcangeli struct rb_root *root = &mm->mm_rb; 2935a0768f6SMichel Lespinasse int i = 0, j, bug = 0; 2941da177e4SLinus Torvalds struct rb_node *nd, *pn = NULL; 2951da177e4SLinus Torvalds unsigned long prev = 0, pend = 0; 2961da177e4SLinus Torvalds 2971da177e4SLinus Torvalds for (nd = rb_first(root); nd; nd = rb_next(nd)) { 2981da177e4SLinus Torvalds struct vm_area_struct *vma; 2991da177e4SLinus Torvalds vma = rb_entry(nd, struct vm_area_struct, vm_rb); 3005a0768f6SMichel Lespinasse if (vma->vm_start < prev) { 301ff26f70fSAndrew Morton pr_emerg("vm_start %lx < prev %lx\n", 302ff26f70fSAndrew Morton vma->vm_start, prev); 3035a0768f6SMichel Lespinasse bug = 1; 3045a0768f6SMichel Lespinasse } 3055a0768f6SMichel Lespinasse if (vma->vm_start < pend) { 306ff26f70fSAndrew Morton pr_emerg("vm_start %lx < pend %lx\n", 307ff26f70fSAndrew Morton vma->vm_start, pend); 3085a0768f6SMichel Lespinasse bug = 1; 3095a0768f6SMichel Lespinasse } 3105a0768f6SMichel Lespinasse if (vma->vm_start > vma->vm_end) { 311ff26f70fSAndrew Morton pr_emerg("vm_start %lx > vm_end %lx\n", 312ff26f70fSAndrew Morton vma->vm_start, vma->vm_end); 3135a0768f6SMichel Lespinasse bug = 1; 3145a0768f6SMichel Lespinasse } 315acf128d0SAndrea Arcangeli spin_lock(&mm->page_table_lock); 3165a0768f6SMichel Lespinasse if (vma->rb_subtree_gap != vma_compute_subtree_gap(vma)) { 3178542bdfcSSasha Levin pr_emerg("free gap %lx, correct %lx\n", 3185a0768f6SMichel Lespinasse vma->rb_subtree_gap, 3195a0768f6SMichel Lespinasse vma_compute_subtree_gap(vma)); 3205a0768f6SMichel Lespinasse bug = 1; 3215a0768f6SMichel Lespinasse } 322acf128d0SAndrea Arcangeli spin_unlock(&mm->page_table_lock); 3231da177e4SLinus Torvalds i++; 3241da177e4SLinus Torvalds pn = nd; 325d1af65d1SDavid Miller prev = vma->vm_start; 326d1af65d1SDavid Miller pend = vma->vm_end; 3271da177e4SLinus Torvalds } 3281da177e4SLinus Torvalds j = 0; 3295a0768f6SMichel Lespinasse for (nd = pn; nd; nd = rb_prev(nd)) 3301da177e4SLinus Torvalds j++; 3315a0768f6SMichel Lespinasse if (i != j) { 3328542bdfcSSasha Levin pr_emerg("backwards %d, forwards %d\n", j, i); 3335a0768f6SMichel Lespinasse bug = 1; 3341da177e4SLinus Torvalds } 3355a0768f6SMichel Lespinasse return bug ? -1 : i; 3361da177e4SLinus Torvalds } 3371da177e4SLinus Torvalds 338d3737187SMichel Lespinasse static void validate_mm_rb(struct rb_root *root, struct vm_area_struct *ignore) 339d3737187SMichel Lespinasse { 340d3737187SMichel Lespinasse struct rb_node *nd; 341d3737187SMichel Lespinasse 342d3737187SMichel Lespinasse for (nd = rb_first(root); nd; nd = rb_next(nd)) { 343d3737187SMichel Lespinasse struct vm_area_struct *vma; 344d3737187SMichel Lespinasse vma = rb_entry(nd, struct vm_area_struct, vm_rb); 34596dad67fSSasha Levin VM_BUG_ON_VMA(vma != ignore && 34696dad67fSSasha Levin vma->rb_subtree_gap != vma_compute_subtree_gap(vma), 34796dad67fSSasha Levin vma); 348d3737187SMichel Lespinasse } 3491da177e4SLinus Torvalds } 3501da177e4SLinus Torvalds 351eafd4dc4SRashika Kheria static void validate_mm(struct mm_struct *mm) 3521da177e4SLinus Torvalds { 3531da177e4SLinus Torvalds int bug = 0; 3541da177e4SLinus Torvalds int i = 0; 3555a0768f6SMichel Lespinasse unsigned long highest_address = 0; 356ed8ea815SMichel Lespinasse struct vm_area_struct *vma = mm->mmap; 357ff26f70fSAndrew Morton 358ed8ea815SMichel Lespinasse while (vma) { 35912352d3cSKonstantin Khlebnikov struct anon_vma *anon_vma = vma->anon_vma; 360ed8ea815SMichel Lespinasse struct anon_vma_chain *avc; 361ff26f70fSAndrew Morton 36212352d3cSKonstantin Khlebnikov if (anon_vma) { 36312352d3cSKonstantin Khlebnikov anon_vma_lock_read(anon_vma); 364ed8ea815SMichel Lespinasse list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) 365ed8ea815SMichel Lespinasse anon_vma_interval_tree_verify(avc); 36612352d3cSKonstantin Khlebnikov anon_vma_unlock_read(anon_vma); 36712352d3cSKonstantin Khlebnikov } 36812352d3cSKonstantin Khlebnikov 3691be7107fSHugh Dickins highest_address = vm_end_gap(vma); 370ed8ea815SMichel Lespinasse vma = vma->vm_next; 3711da177e4SLinus Torvalds i++; 3721da177e4SLinus Torvalds } 3735a0768f6SMichel Lespinasse if (i != mm->map_count) { 3748542bdfcSSasha Levin pr_emerg("map_count %d vm_next %d\n", mm->map_count, i); 3755a0768f6SMichel Lespinasse bug = 1; 3765a0768f6SMichel Lespinasse } 3775a0768f6SMichel Lespinasse if (highest_address != mm->highest_vm_end) { 3788542bdfcSSasha Levin pr_emerg("mm->highest_vm_end %lx, found %lx\n", 3795a0768f6SMichel Lespinasse mm->highest_vm_end, highest_address); 3805a0768f6SMichel Lespinasse bug = 1; 3815a0768f6SMichel Lespinasse } 382acf128d0SAndrea Arcangeli i = browse_rb(mm); 3835a0768f6SMichel Lespinasse if (i != mm->map_count) { 384ff26f70fSAndrew Morton if (i != -1) 3858542bdfcSSasha Levin pr_emerg("map_count %d rb %d\n", mm->map_count, i); 3865a0768f6SMichel Lespinasse bug = 1; 3875a0768f6SMichel Lespinasse } 38896dad67fSSasha Levin VM_BUG_ON_MM(bug, mm); 3891da177e4SLinus Torvalds } 3901da177e4SLinus Torvalds #else 391d3737187SMichel Lespinasse #define validate_mm_rb(root, ignore) do { } while (0) 3921da177e4SLinus Torvalds #define validate_mm(mm) do { } while (0) 3931da177e4SLinus Torvalds #endif 3941da177e4SLinus Torvalds 395315cc066SMichel Lespinasse RB_DECLARE_CALLBACKS_MAX(static, vma_gap_callbacks, 396315cc066SMichel Lespinasse struct vm_area_struct, vm_rb, 397315cc066SMichel Lespinasse unsigned long, rb_subtree_gap, vma_compute_gap) 398d3737187SMichel Lespinasse 399d3737187SMichel Lespinasse /* 400d3737187SMichel Lespinasse * Update augmented rbtree rb_subtree_gap values after vma->vm_start or 401d3737187SMichel Lespinasse * vma->vm_prev->vm_end values changed, without modifying the vma's position 402d3737187SMichel Lespinasse * in the rbtree. 403d3737187SMichel Lespinasse */ 404d3737187SMichel Lespinasse static void vma_gap_update(struct vm_area_struct *vma) 405d3737187SMichel Lespinasse { 406d3737187SMichel Lespinasse /* 407315cc066SMichel Lespinasse * As it turns out, RB_DECLARE_CALLBACKS_MAX() already created 408315cc066SMichel Lespinasse * a callback function that does exactly what we want. 409d3737187SMichel Lespinasse */ 410d3737187SMichel Lespinasse vma_gap_callbacks_propagate(&vma->vm_rb, NULL); 411d3737187SMichel Lespinasse } 412d3737187SMichel Lespinasse 413d3737187SMichel Lespinasse static inline void vma_rb_insert(struct vm_area_struct *vma, 414d3737187SMichel Lespinasse struct rb_root *root) 415d3737187SMichel Lespinasse { 416d3737187SMichel Lespinasse /* All rb_subtree_gap values must be consistent prior to insertion */ 417d3737187SMichel Lespinasse validate_mm_rb(root, NULL); 418d3737187SMichel Lespinasse 419d3737187SMichel Lespinasse rb_insert_augmented(&vma->vm_rb, root, &vma_gap_callbacks); 420d3737187SMichel Lespinasse } 421d3737187SMichel Lespinasse 4228f26e0b1SAndrea Arcangeli static void __vma_rb_erase(struct vm_area_struct *vma, struct rb_root *root) 4238f26e0b1SAndrea Arcangeli { 4248f26e0b1SAndrea Arcangeli /* 4258f26e0b1SAndrea Arcangeli * Note rb_erase_augmented is a fairly large inline function, 4268f26e0b1SAndrea Arcangeli * so make sure we instantiate it only once with our desired 4278f26e0b1SAndrea Arcangeli * augmented rbtree callbacks. 4288f26e0b1SAndrea Arcangeli */ 4298f26e0b1SAndrea Arcangeli rb_erase_augmented(&vma->vm_rb, root, &vma_gap_callbacks); 4308f26e0b1SAndrea Arcangeli } 4318f26e0b1SAndrea Arcangeli 4328f26e0b1SAndrea Arcangeli static __always_inline void vma_rb_erase_ignore(struct vm_area_struct *vma, 4338f26e0b1SAndrea Arcangeli struct rb_root *root, 4348f26e0b1SAndrea Arcangeli struct vm_area_struct *ignore) 4358f26e0b1SAndrea Arcangeli { 4368f26e0b1SAndrea Arcangeli /* 4378f26e0b1SAndrea Arcangeli * All rb_subtree_gap values must be consistent prior to erase, 4384d1e7243SWei Yang * with the possible exception of 4394d1e7243SWei Yang * 4404d1e7243SWei Yang * a. the "next" vma being erased if next->vm_start was reduced in 4414d1e7243SWei Yang * __vma_adjust() -> __vma_unlink() 4424d1e7243SWei Yang * b. the vma being erased in detach_vmas_to_be_unmapped() -> 4434d1e7243SWei Yang * vma_rb_erase() 4448f26e0b1SAndrea Arcangeli */ 4458f26e0b1SAndrea Arcangeli validate_mm_rb(root, ignore); 4468f26e0b1SAndrea Arcangeli 4478f26e0b1SAndrea Arcangeli __vma_rb_erase(vma, root); 4488f26e0b1SAndrea Arcangeli } 4498f26e0b1SAndrea Arcangeli 4508f26e0b1SAndrea Arcangeli static __always_inline void vma_rb_erase(struct vm_area_struct *vma, 4518f26e0b1SAndrea Arcangeli struct rb_root *root) 452d3737187SMichel Lespinasse { 4534d1e7243SWei Yang vma_rb_erase_ignore(vma, root, vma); 454d3737187SMichel Lespinasse } 455d3737187SMichel Lespinasse 456bf181b9fSMichel Lespinasse /* 457bf181b9fSMichel Lespinasse * vma has some anon_vma assigned, and is already inserted on that 458bf181b9fSMichel Lespinasse * anon_vma's interval trees. 459bf181b9fSMichel Lespinasse * 460bf181b9fSMichel Lespinasse * Before updating the vma's vm_start / vm_end / vm_pgoff fields, the 461bf181b9fSMichel Lespinasse * vma must be removed from the anon_vma's interval trees using 462bf181b9fSMichel Lespinasse * anon_vma_interval_tree_pre_update_vma(). 463bf181b9fSMichel Lespinasse * 464bf181b9fSMichel Lespinasse * After the update, the vma will be reinserted using 465bf181b9fSMichel Lespinasse * anon_vma_interval_tree_post_update_vma(). 466bf181b9fSMichel Lespinasse * 467c1e8d7c6SMichel Lespinasse * The entire update must be protected by exclusive mmap_lock and by 468bf181b9fSMichel Lespinasse * the root anon_vma's mutex. 469bf181b9fSMichel Lespinasse */ 470bf181b9fSMichel Lespinasse static inline void 471bf181b9fSMichel Lespinasse anon_vma_interval_tree_pre_update_vma(struct vm_area_struct *vma) 472bf181b9fSMichel Lespinasse { 473bf181b9fSMichel Lespinasse struct anon_vma_chain *avc; 474bf181b9fSMichel Lespinasse 475bf181b9fSMichel Lespinasse list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) 476bf181b9fSMichel Lespinasse anon_vma_interval_tree_remove(avc, &avc->anon_vma->rb_root); 477bf181b9fSMichel Lespinasse } 478bf181b9fSMichel Lespinasse 479bf181b9fSMichel Lespinasse static inline void 480bf181b9fSMichel Lespinasse anon_vma_interval_tree_post_update_vma(struct vm_area_struct *vma) 481bf181b9fSMichel Lespinasse { 482bf181b9fSMichel Lespinasse struct anon_vma_chain *avc; 483bf181b9fSMichel Lespinasse 484bf181b9fSMichel Lespinasse list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) 485bf181b9fSMichel Lespinasse anon_vma_interval_tree_insert(avc, &avc->anon_vma->rb_root); 486bf181b9fSMichel Lespinasse } 487bf181b9fSMichel Lespinasse 4886597d783SHugh Dickins static int find_vma_links(struct mm_struct *mm, unsigned long addr, 4896597d783SHugh Dickins unsigned long end, struct vm_area_struct **pprev, 4906597d783SHugh Dickins struct rb_node ***rb_link, struct rb_node **rb_parent) 4911da177e4SLinus Torvalds { 4921da177e4SLinus Torvalds struct rb_node **__rb_link, *__rb_parent, *rb_prev; 4931da177e4SLinus Torvalds 4945b78ed24SLuigi Rizzo mmap_assert_locked(mm); 4951da177e4SLinus Torvalds __rb_link = &mm->mm_rb.rb_node; 4961da177e4SLinus Torvalds rb_prev = __rb_parent = NULL; 4971da177e4SLinus Torvalds 4981da177e4SLinus Torvalds while (*__rb_link) { 4991da177e4SLinus Torvalds struct vm_area_struct *vma_tmp; 5001da177e4SLinus Torvalds 5011da177e4SLinus Torvalds __rb_parent = *__rb_link; 5021da177e4SLinus Torvalds vma_tmp = rb_entry(__rb_parent, struct vm_area_struct, vm_rb); 5031da177e4SLinus Torvalds 5041da177e4SLinus Torvalds if (vma_tmp->vm_end > addr) { 5056597d783SHugh Dickins /* Fail if an existing vma overlaps the area */ 5066597d783SHugh Dickins if (vma_tmp->vm_start < end) 5076597d783SHugh Dickins return -ENOMEM; 5081da177e4SLinus Torvalds __rb_link = &__rb_parent->rb_left; 5091da177e4SLinus Torvalds } else { 5101da177e4SLinus Torvalds rb_prev = __rb_parent; 5111da177e4SLinus Torvalds __rb_link = &__rb_parent->rb_right; 5121da177e4SLinus Torvalds } 5131da177e4SLinus Torvalds } 5141da177e4SLinus Torvalds 5151da177e4SLinus Torvalds *pprev = NULL; 5161da177e4SLinus Torvalds if (rb_prev) 5171da177e4SLinus Torvalds *pprev = rb_entry(rb_prev, struct vm_area_struct, vm_rb); 5181da177e4SLinus Torvalds *rb_link = __rb_link; 5191da177e4SLinus Torvalds *rb_parent = __rb_parent; 5206597d783SHugh Dickins return 0; 5211da177e4SLinus Torvalds } 5221da177e4SLinus Torvalds 5233903b55aSLiam R. Howlett /* 5243903b55aSLiam R. Howlett * vma_next() - Get the next VMA. 5253903b55aSLiam R. Howlett * @mm: The mm_struct. 5263903b55aSLiam R. Howlett * @vma: The current vma. 5273903b55aSLiam R. Howlett * 5283903b55aSLiam R. Howlett * If @vma is NULL, return the first vma in the mm. 5293903b55aSLiam R. Howlett * 5303903b55aSLiam R. Howlett * Returns: The next VMA after @vma. 5313903b55aSLiam R. Howlett */ 5323903b55aSLiam R. Howlett static inline struct vm_area_struct *vma_next(struct mm_struct *mm, 5333903b55aSLiam R. Howlett struct vm_area_struct *vma) 5343903b55aSLiam R. Howlett { 5353903b55aSLiam R. Howlett if (!vma) 5363903b55aSLiam R. Howlett return mm->mmap; 5373903b55aSLiam R. Howlett 5383903b55aSLiam R. Howlett return vma->vm_next; 5393903b55aSLiam R. Howlett } 540fb8090b6SLiam R. Howlett 541fb8090b6SLiam R. Howlett /* 542fb8090b6SLiam R. Howlett * munmap_vma_range() - munmap VMAs that overlap a range. 543fb8090b6SLiam R. Howlett * @mm: The mm struct 544fb8090b6SLiam R. Howlett * @start: The start of the range. 545fb8090b6SLiam R. Howlett * @len: The length of the range. 546fb8090b6SLiam R. Howlett * @pprev: pointer to the pointer that will be set to previous vm_area_struct 547fb8090b6SLiam R. Howlett * @rb_link: the rb_node 548fb8090b6SLiam R. Howlett * @rb_parent: the parent rb_node 549fb8090b6SLiam R. Howlett * 550fb8090b6SLiam R. Howlett * Find all the vm_area_struct that overlap from @start to 551fb8090b6SLiam R. Howlett * @end and munmap them. Set @pprev to the previous vm_area_struct. 552fb8090b6SLiam R. Howlett * 553fb8090b6SLiam R. Howlett * Returns: -ENOMEM on munmap failure or 0 on success. 554fb8090b6SLiam R. Howlett */ 555fb8090b6SLiam R. Howlett static inline int 556fb8090b6SLiam R. Howlett munmap_vma_range(struct mm_struct *mm, unsigned long start, unsigned long len, 557fb8090b6SLiam R. Howlett struct vm_area_struct **pprev, struct rb_node ***link, 558fb8090b6SLiam R. Howlett struct rb_node **parent, struct list_head *uf) 559fb8090b6SLiam R. Howlett { 560fb8090b6SLiam R. Howlett 561fb8090b6SLiam R. Howlett while (find_vma_links(mm, start, start + len, pprev, link, parent)) 562fb8090b6SLiam R. Howlett if (do_munmap(mm, start, len, uf)) 563fb8090b6SLiam R. Howlett return -ENOMEM; 564fb8090b6SLiam R. Howlett 565fb8090b6SLiam R. Howlett return 0; 566fb8090b6SLiam R. Howlett } 567e8420a8eSCyril Hrubis static unsigned long count_vma_pages_range(struct mm_struct *mm, 568e8420a8eSCyril Hrubis unsigned long addr, unsigned long end) 569e8420a8eSCyril Hrubis { 570e8420a8eSCyril Hrubis unsigned long nr_pages = 0; 571e8420a8eSCyril Hrubis struct vm_area_struct *vma; 572e8420a8eSCyril Hrubis 573f0953a1bSIngo Molnar /* Find first overlapping mapping */ 574e8420a8eSCyril Hrubis vma = find_vma_intersection(mm, addr, end); 575e8420a8eSCyril Hrubis if (!vma) 576e8420a8eSCyril Hrubis return 0; 577e8420a8eSCyril Hrubis 578e8420a8eSCyril Hrubis nr_pages = (min(end, vma->vm_end) - 579e8420a8eSCyril Hrubis max(addr, vma->vm_start)) >> PAGE_SHIFT; 580e8420a8eSCyril Hrubis 581e8420a8eSCyril Hrubis /* Iterate over the rest of the overlaps */ 582e8420a8eSCyril Hrubis for (vma = vma->vm_next; vma; vma = vma->vm_next) { 583e8420a8eSCyril Hrubis unsigned long overlap_len; 584e8420a8eSCyril Hrubis 585e8420a8eSCyril Hrubis if (vma->vm_start > end) 586e8420a8eSCyril Hrubis break; 587e8420a8eSCyril Hrubis 588e8420a8eSCyril Hrubis overlap_len = min(end, vma->vm_end) - vma->vm_start; 589e8420a8eSCyril Hrubis nr_pages += overlap_len >> PAGE_SHIFT; 590e8420a8eSCyril Hrubis } 591e8420a8eSCyril Hrubis 592e8420a8eSCyril Hrubis return nr_pages; 593e8420a8eSCyril Hrubis } 594e8420a8eSCyril Hrubis 5951da177e4SLinus Torvalds void __vma_link_rb(struct mm_struct *mm, struct vm_area_struct *vma, 5961da177e4SLinus Torvalds struct rb_node **rb_link, struct rb_node *rb_parent) 5971da177e4SLinus Torvalds { 598d3737187SMichel Lespinasse /* Update tracking information for the gap following the new vma. */ 599d3737187SMichel Lespinasse if (vma->vm_next) 600d3737187SMichel Lespinasse vma_gap_update(vma->vm_next); 601d3737187SMichel Lespinasse else 6021be7107fSHugh Dickins mm->highest_vm_end = vm_end_gap(vma); 603d3737187SMichel Lespinasse 604d3737187SMichel Lespinasse /* 605d3737187SMichel Lespinasse * vma->vm_prev wasn't known when we followed the rbtree to find the 606d3737187SMichel Lespinasse * correct insertion point for that vma. As a result, we could not 607d3737187SMichel Lespinasse * update the vma vm_rb parents rb_subtree_gap values on the way down. 608d3737187SMichel Lespinasse * So, we first insert the vma with a zero rb_subtree_gap value 609d3737187SMichel Lespinasse * (to be consistent with what we did on the way down), and then 610d3737187SMichel Lespinasse * immediately update the gap to the correct value. Finally we 611d3737187SMichel Lespinasse * rebalance the rbtree after all augmented values have been set. 612d3737187SMichel Lespinasse */ 6131da177e4SLinus Torvalds rb_link_node(&vma->vm_rb, rb_parent, rb_link); 614d3737187SMichel Lespinasse vma->rb_subtree_gap = 0; 615d3737187SMichel Lespinasse vma_gap_update(vma); 616d3737187SMichel Lespinasse vma_rb_insert(vma, &mm->mm_rb); 6171da177e4SLinus Torvalds } 6181da177e4SLinus Torvalds 619cb8f488cSDenys Vlasenko static void __vma_link_file(struct vm_area_struct *vma) 6201da177e4SLinus Torvalds { 6211da177e4SLinus Torvalds struct file *file; 6221da177e4SLinus Torvalds 6231da177e4SLinus Torvalds file = vma->vm_file; 6241da177e4SLinus Torvalds if (file) { 6251da177e4SLinus Torvalds struct address_space *mapping = file->f_mapping; 6261da177e4SLinus Torvalds 6271da177e4SLinus Torvalds if (vma->vm_flags & VM_SHARED) 628cf508b58SMiaohe Lin mapping_allow_writable(mapping); 6291da177e4SLinus Torvalds 6301da177e4SLinus Torvalds flush_dcache_mmap_lock(mapping); 6316b2dbba8SMichel Lespinasse vma_interval_tree_insert(vma, &mapping->i_mmap); 6321da177e4SLinus Torvalds flush_dcache_mmap_unlock(mapping); 6331da177e4SLinus Torvalds } 6341da177e4SLinus Torvalds } 6351da177e4SLinus Torvalds 6361da177e4SLinus Torvalds static void 6371da177e4SLinus Torvalds __vma_link(struct mm_struct *mm, struct vm_area_struct *vma, 6381da177e4SLinus Torvalds struct vm_area_struct *prev, struct rb_node **rb_link, 6391da177e4SLinus Torvalds struct rb_node *rb_parent) 6401da177e4SLinus Torvalds { 641aba6dfb7SWei Yang __vma_link_list(mm, vma, prev); 6421da177e4SLinus Torvalds __vma_link_rb(mm, vma, rb_link, rb_parent); 6431da177e4SLinus Torvalds } 6441da177e4SLinus Torvalds 6451da177e4SLinus Torvalds static void vma_link(struct mm_struct *mm, struct vm_area_struct *vma, 6461da177e4SLinus Torvalds struct vm_area_struct *prev, struct rb_node **rb_link, 6471da177e4SLinus Torvalds struct rb_node *rb_parent) 6481da177e4SLinus Torvalds { 6491da177e4SLinus Torvalds struct address_space *mapping = NULL; 6501da177e4SLinus Torvalds 65164ac4940SHuang Shijie if (vma->vm_file) { 6521da177e4SLinus Torvalds mapping = vma->vm_file->f_mapping; 65383cde9e8SDavidlohr Bueso i_mmap_lock_write(mapping); 65464ac4940SHuang Shijie } 6551da177e4SLinus Torvalds 6561da177e4SLinus Torvalds __vma_link(mm, vma, prev, rb_link, rb_parent); 6571da177e4SLinus Torvalds __vma_link_file(vma); 6581da177e4SLinus Torvalds 6591da177e4SLinus Torvalds if (mapping) 66083cde9e8SDavidlohr Bueso i_mmap_unlock_write(mapping); 6611da177e4SLinus Torvalds 6621da177e4SLinus Torvalds mm->map_count++; 6631da177e4SLinus Torvalds validate_mm(mm); 6641da177e4SLinus Torvalds } 6651da177e4SLinus Torvalds 6661da177e4SLinus Torvalds /* 66788f6b4c3SKautuk Consul * Helper for vma_adjust() in the split_vma insert case: insert a vma into the 6686b2dbba8SMichel Lespinasse * mm's list and rbtree. It has already been inserted into the interval tree. 6691da177e4SLinus Torvalds */ 67048aae425SZhenwenXu static void __insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma) 6711da177e4SLinus Torvalds { 6726597d783SHugh Dickins struct vm_area_struct *prev; 6731da177e4SLinus Torvalds struct rb_node **rb_link, *rb_parent; 6741da177e4SLinus Torvalds 6756597d783SHugh Dickins if (find_vma_links(mm, vma->vm_start, vma->vm_end, 6766597d783SHugh Dickins &prev, &rb_link, &rb_parent)) 6776597d783SHugh Dickins BUG(); 6781da177e4SLinus Torvalds __vma_link(mm, vma, prev, rb_link, rb_parent); 6791da177e4SLinus Torvalds mm->map_count++; 6801da177e4SLinus Torvalds } 6811da177e4SLinus Torvalds 6827c61f917SWei Yang static __always_inline void __vma_unlink(struct mm_struct *mm, 683e86f15eeSAndrea Arcangeli struct vm_area_struct *vma, 6848f26e0b1SAndrea Arcangeli struct vm_area_struct *ignore) 6851da177e4SLinus Torvalds { 6868f26e0b1SAndrea Arcangeli vma_rb_erase_ignore(vma, &mm->mm_rb, ignore); 6871b9fc5b2SWei Yang __vma_unlink_list(mm, vma); 688615d6e87SDavidlohr Bueso /* Kill the cache */ 689615d6e87SDavidlohr Bueso vmacache_invalidate(mm); 6901da177e4SLinus Torvalds } 6911da177e4SLinus Torvalds 6921da177e4SLinus Torvalds /* 6931da177e4SLinus Torvalds * We cannot adjust vm_start, vm_end, vm_pgoff fields of a vma that 6941da177e4SLinus Torvalds * is already present in an i_mmap tree without adjusting the tree. 6951da177e4SLinus Torvalds * The following helper function should be used when such adjustments 6961da177e4SLinus Torvalds * are necessary. The "insert" vma (if any) is to be inserted 6971da177e4SLinus Torvalds * before we drop the necessary locks. 6981da177e4SLinus Torvalds */ 699e86f15eeSAndrea Arcangeli int __vma_adjust(struct vm_area_struct *vma, unsigned long start, 700e86f15eeSAndrea Arcangeli unsigned long end, pgoff_t pgoff, struct vm_area_struct *insert, 701e86f15eeSAndrea Arcangeli struct vm_area_struct *expand) 7021da177e4SLinus Torvalds { 7031da177e4SLinus Torvalds struct mm_struct *mm = vma->vm_mm; 704e86f15eeSAndrea Arcangeli struct vm_area_struct *next = vma->vm_next, *orig_vma = vma; 7051da177e4SLinus Torvalds struct address_space *mapping = NULL; 706f808c13fSDavidlohr Bueso struct rb_root_cached *root = NULL; 707012f1800SRik van Riel struct anon_vma *anon_vma = NULL; 7081da177e4SLinus Torvalds struct file *file = vma->vm_file; 709d3737187SMichel Lespinasse bool start_changed = false, end_changed = false; 7101da177e4SLinus Torvalds long adjust_next = 0; 7111da177e4SLinus Torvalds int remove_next = 0; 7121da177e4SLinus Torvalds 7131da177e4SLinus Torvalds if (next && !insert) { 714734537c9SKirill A. Shutemov struct vm_area_struct *exporter = NULL, *importer = NULL; 715287d97acSLinus Torvalds 7161da177e4SLinus Torvalds if (end >= next->vm_end) { 7171da177e4SLinus Torvalds /* 7181da177e4SLinus Torvalds * vma expands, overlapping all the next, and 7191da177e4SLinus Torvalds * perhaps the one after too (mprotect case 6). 72086d12e47SAndrea Arcangeli * The only other cases that gets here are 721e86f15eeSAndrea Arcangeli * case 1, case 7 and case 8. 722e86f15eeSAndrea Arcangeli */ 723e86f15eeSAndrea Arcangeli if (next == expand) { 724e86f15eeSAndrea Arcangeli /* 725e86f15eeSAndrea Arcangeli * The only case where we don't expand "vma" 726e86f15eeSAndrea Arcangeli * and we expand "next" instead is case 8. 727e86f15eeSAndrea Arcangeli */ 728e86f15eeSAndrea Arcangeli VM_WARN_ON(end != next->vm_end); 729e86f15eeSAndrea Arcangeli /* 730e86f15eeSAndrea Arcangeli * remove_next == 3 means we're 731e86f15eeSAndrea Arcangeli * removing "vma" and that to do so we 732e86f15eeSAndrea Arcangeli * swapped "vma" and "next". 733e86f15eeSAndrea Arcangeli */ 734e86f15eeSAndrea Arcangeli remove_next = 3; 735e86f15eeSAndrea Arcangeli VM_WARN_ON(file != next->vm_file); 736e86f15eeSAndrea Arcangeli swap(vma, next); 737e86f15eeSAndrea Arcangeli } else { 738e86f15eeSAndrea Arcangeli VM_WARN_ON(expand != vma); 739e86f15eeSAndrea Arcangeli /* 740e86f15eeSAndrea Arcangeli * case 1, 6, 7, remove_next == 2 is case 6, 741e86f15eeSAndrea Arcangeli * remove_next == 1 is case 1 or 7. 7421da177e4SLinus Torvalds */ 743734537c9SKirill A. Shutemov remove_next = 1 + (end > next->vm_end); 744e86f15eeSAndrea Arcangeli VM_WARN_ON(remove_next == 2 && 745e86f15eeSAndrea Arcangeli end != next->vm_next->vm_end); 746e86f15eeSAndrea Arcangeli /* trim end to next, for case 6 first pass */ 7471da177e4SLinus Torvalds end = next->vm_end; 748e86f15eeSAndrea Arcangeli } 749e86f15eeSAndrea Arcangeli 750287d97acSLinus Torvalds exporter = next; 7511da177e4SLinus Torvalds importer = vma; 752734537c9SKirill A. Shutemov 753734537c9SKirill A. Shutemov /* 754734537c9SKirill A. Shutemov * If next doesn't have anon_vma, import from vma after 755734537c9SKirill A. Shutemov * next, if the vma overlaps with it. 756734537c9SKirill A. Shutemov */ 75797a42cd4SAndrea Arcangeli if (remove_next == 2 && !next->anon_vma) 758734537c9SKirill A. Shutemov exporter = next->vm_next; 759734537c9SKirill A. Shutemov 7601da177e4SLinus Torvalds } else if (end > next->vm_start) { 7611da177e4SLinus Torvalds /* 7621da177e4SLinus Torvalds * vma expands, overlapping part of the next: 7631da177e4SLinus Torvalds * mprotect case 5 shifting the boundary up. 7641da177e4SLinus Torvalds */ 765f9d86a60SWei Yang adjust_next = (end - next->vm_start); 766287d97acSLinus Torvalds exporter = next; 7671da177e4SLinus Torvalds importer = vma; 768e86f15eeSAndrea Arcangeli VM_WARN_ON(expand != importer); 7691da177e4SLinus Torvalds } else if (end < vma->vm_end) { 7701da177e4SLinus Torvalds /* 7711da177e4SLinus Torvalds * vma shrinks, and !insert tells it's not 7721da177e4SLinus Torvalds * split_vma inserting another: so it must be 7731da177e4SLinus Torvalds * mprotect case 4 shifting the boundary down. 7741da177e4SLinus Torvalds */ 775f9d86a60SWei Yang adjust_next = -(vma->vm_end - end); 776287d97acSLinus Torvalds exporter = vma; 7771da177e4SLinus Torvalds importer = next; 778e86f15eeSAndrea Arcangeli VM_WARN_ON(expand != importer); 7791da177e4SLinus Torvalds } 7801da177e4SLinus Torvalds 7815beb4930SRik van Riel /* 7825beb4930SRik van Riel * Easily overlooked: when mprotect shifts the boundary, 7835beb4930SRik van Riel * make sure the expanding vma has anon_vma set if the 7845beb4930SRik van Riel * shrinking vma had, to cover any anon pages imported. 7855beb4930SRik van Riel */ 786287d97acSLinus Torvalds if (exporter && exporter->anon_vma && !importer->anon_vma) { 787c4ea95d7SDaniel Forrest int error; 788c4ea95d7SDaniel Forrest 789287d97acSLinus Torvalds importer->anon_vma = exporter->anon_vma; 790b800c91aSKonstantin Khlebnikov error = anon_vma_clone(importer, exporter); 7913fe89b3eSLeon Yu if (error) 792b800c91aSKonstantin Khlebnikov return error; 793b800c91aSKonstantin Khlebnikov } 7945beb4930SRik van Riel } 795734537c9SKirill A. Shutemov again: 796e86f15eeSAndrea Arcangeli vma_adjust_trans_huge(orig_vma, start, end, adjust_next); 79737f9f559SKirill A. Shutemov 7981da177e4SLinus Torvalds if (file) { 7991da177e4SLinus Torvalds mapping = file->f_mapping; 8001da177e4SLinus Torvalds root = &mapping->i_mmap; 801cbc91f71SSrikar Dronamraju uprobe_munmap(vma, vma->vm_start, vma->vm_end); 802682968e0SSrikar Dronamraju 803682968e0SSrikar Dronamraju if (adjust_next) 80427ba0644SKirill A. Shutemov uprobe_munmap(next, next->vm_start, next->vm_end); 805682968e0SSrikar Dronamraju 80683cde9e8SDavidlohr Bueso i_mmap_lock_write(mapping); 8071da177e4SLinus Torvalds if (insert) { 8081da177e4SLinus Torvalds /* 8096b2dbba8SMichel Lespinasse * Put into interval tree now, so instantiated pages 8101da177e4SLinus Torvalds * are visible to arm/parisc __flush_dcache_page 8111da177e4SLinus Torvalds * throughout; but we cannot insert into address 8121da177e4SLinus Torvalds * space until vma start or end is updated. 8131da177e4SLinus Torvalds */ 8141da177e4SLinus Torvalds __vma_link_file(insert); 8151da177e4SLinus Torvalds } 8161da177e4SLinus Torvalds } 8171da177e4SLinus Torvalds 818012f1800SRik van Riel anon_vma = vma->anon_vma; 819bf181b9fSMichel Lespinasse if (!anon_vma && adjust_next) 820bf181b9fSMichel Lespinasse anon_vma = next->anon_vma; 821bf181b9fSMichel Lespinasse if (anon_vma) { 822e86f15eeSAndrea Arcangeli VM_WARN_ON(adjust_next && next->anon_vma && 823e86f15eeSAndrea Arcangeli anon_vma != next->anon_vma); 8244fc3f1d6SIngo Molnar anon_vma_lock_write(anon_vma); 825bf181b9fSMichel Lespinasse anon_vma_interval_tree_pre_update_vma(vma); 826bf181b9fSMichel Lespinasse if (adjust_next) 827bf181b9fSMichel Lespinasse anon_vma_interval_tree_pre_update_vma(next); 828bf181b9fSMichel Lespinasse } 829012f1800SRik van Riel 8300fc48a6eSWei Yang if (file) { 8311da177e4SLinus Torvalds flush_dcache_mmap_lock(mapping); 8326b2dbba8SMichel Lespinasse vma_interval_tree_remove(vma, root); 8331da177e4SLinus Torvalds if (adjust_next) 8346b2dbba8SMichel Lespinasse vma_interval_tree_remove(next, root); 8351da177e4SLinus Torvalds } 8361da177e4SLinus Torvalds 837d3737187SMichel Lespinasse if (start != vma->vm_start) { 8381da177e4SLinus Torvalds vma->vm_start = start; 839d3737187SMichel Lespinasse start_changed = true; 840d3737187SMichel Lespinasse } 841d3737187SMichel Lespinasse if (end != vma->vm_end) { 8421da177e4SLinus Torvalds vma->vm_end = end; 843d3737187SMichel Lespinasse end_changed = true; 844d3737187SMichel Lespinasse } 8451da177e4SLinus Torvalds vma->vm_pgoff = pgoff; 8461da177e4SLinus Torvalds if (adjust_next) { 847f9d86a60SWei Yang next->vm_start += adjust_next; 848f9d86a60SWei Yang next->vm_pgoff += adjust_next >> PAGE_SHIFT; 8491da177e4SLinus Torvalds } 8501da177e4SLinus Torvalds 8510fc48a6eSWei Yang if (file) { 8521da177e4SLinus Torvalds if (adjust_next) 8536b2dbba8SMichel Lespinasse vma_interval_tree_insert(next, root); 8546b2dbba8SMichel Lespinasse vma_interval_tree_insert(vma, root); 8551da177e4SLinus Torvalds flush_dcache_mmap_unlock(mapping); 8561da177e4SLinus Torvalds } 8571da177e4SLinus Torvalds 8581da177e4SLinus Torvalds if (remove_next) { 8591da177e4SLinus Torvalds /* 8601da177e4SLinus Torvalds * vma_merge has merged next into vma, and needs 8611da177e4SLinus Torvalds * us to remove next before dropping the locks. 8621da177e4SLinus Torvalds */ 863e86f15eeSAndrea Arcangeli if (remove_next != 3) 8647c61f917SWei Yang __vma_unlink(mm, next, next); 865e86f15eeSAndrea Arcangeli else 8668f26e0b1SAndrea Arcangeli /* 8678f26e0b1SAndrea Arcangeli * vma is not before next if they've been 8688f26e0b1SAndrea Arcangeli * swapped. 8698f26e0b1SAndrea Arcangeli * 8708f26e0b1SAndrea Arcangeli * pre-swap() next->vm_start was reduced so 8718f26e0b1SAndrea Arcangeli * tell validate_mm_rb to ignore pre-swap() 8728f26e0b1SAndrea Arcangeli * "next" (which is stored in post-swap() 8738f26e0b1SAndrea Arcangeli * "vma"). 8748f26e0b1SAndrea Arcangeli */ 8757c61f917SWei Yang __vma_unlink(mm, next, vma); 8761da177e4SLinus Torvalds if (file) 8771da177e4SLinus Torvalds __remove_shared_vm_struct(next, file, mapping); 8781da177e4SLinus Torvalds } else if (insert) { 8791da177e4SLinus Torvalds /* 8801da177e4SLinus Torvalds * split_vma has split insert from vma, and needs 8811da177e4SLinus Torvalds * us to insert it before dropping the locks 8821da177e4SLinus Torvalds * (it may either follow vma or precede it). 8831da177e4SLinus Torvalds */ 8841da177e4SLinus Torvalds __insert_vm_struct(mm, insert); 885d3737187SMichel Lespinasse } else { 886d3737187SMichel Lespinasse if (start_changed) 887d3737187SMichel Lespinasse vma_gap_update(vma); 888d3737187SMichel Lespinasse if (end_changed) { 889d3737187SMichel Lespinasse if (!next) 8901be7107fSHugh Dickins mm->highest_vm_end = vm_end_gap(vma); 891d3737187SMichel Lespinasse else if (!adjust_next) 892d3737187SMichel Lespinasse vma_gap_update(next); 893d3737187SMichel Lespinasse } 8941da177e4SLinus Torvalds } 8951da177e4SLinus Torvalds 896bf181b9fSMichel Lespinasse if (anon_vma) { 897bf181b9fSMichel Lespinasse anon_vma_interval_tree_post_update_vma(vma); 898bf181b9fSMichel Lespinasse if (adjust_next) 899bf181b9fSMichel Lespinasse anon_vma_interval_tree_post_update_vma(next); 90008b52706SKonstantin Khlebnikov anon_vma_unlock_write(anon_vma); 901bf181b9fSMichel Lespinasse } 9021da177e4SLinus Torvalds 9030fc48a6eSWei Yang if (file) { 904808fbdbeSWei Yang i_mmap_unlock_write(mapping); 9057b2d81d4SIngo Molnar uprobe_mmap(vma); 9062b144498SSrikar Dronamraju 9072b144498SSrikar Dronamraju if (adjust_next) 9087b2d81d4SIngo Molnar uprobe_mmap(next); 9092b144498SSrikar Dronamraju } 9102b144498SSrikar Dronamraju 9111da177e4SLinus Torvalds if (remove_next) { 912925d1c40SMatt Helsley if (file) { 913cbc91f71SSrikar Dronamraju uprobe_munmap(next, next->vm_start, next->vm_end); 9141da177e4SLinus Torvalds fput(file); 915925d1c40SMatt Helsley } 9165beb4930SRik van Riel if (next->anon_vma) 9175beb4930SRik van Riel anon_vma_merge(vma, next); 9181da177e4SLinus Torvalds mm->map_count--; 9193964acd0SOleg Nesterov mpol_put(vma_policy(next)); 9203928d4f5SLinus Torvalds vm_area_free(next); 9211da177e4SLinus Torvalds /* 9221da177e4SLinus Torvalds * In mprotect's case 6 (see comments on vma_merge), 9231da177e4SLinus Torvalds * we must remove another next too. It would clutter 9241da177e4SLinus Torvalds * up the code too much to do both in one go. 9251da177e4SLinus Torvalds */ 926e86f15eeSAndrea Arcangeli if (remove_next != 3) { 927e86f15eeSAndrea Arcangeli /* 928e86f15eeSAndrea Arcangeli * If "next" was removed and vma->vm_end was 929e86f15eeSAndrea Arcangeli * expanded (up) over it, in turn 930e86f15eeSAndrea Arcangeli * "next->vm_prev->vm_end" changed and the 931e86f15eeSAndrea Arcangeli * "vma->vm_next" gap must be updated. 932e86f15eeSAndrea Arcangeli */ 9331da177e4SLinus Torvalds next = vma->vm_next; 934e86f15eeSAndrea Arcangeli } else { 935e86f15eeSAndrea Arcangeli /* 936e86f15eeSAndrea Arcangeli * For the scope of the comment "next" and 937e86f15eeSAndrea Arcangeli * "vma" considered pre-swap(): if "vma" was 938e86f15eeSAndrea Arcangeli * removed, next->vm_start was expanded (down) 939e86f15eeSAndrea Arcangeli * over it and the "next" gap must be updated. 940e86f15eeSAndrea Arcangeli * Because of the swap() the post-swap() "vma" 941e86f15eeSAndrea Arcangeli * actually points to pre-swap() "next" 942e86f15eeSAndrea Arcangeli * (post-swap() "next" as opposed is now a 943e86f15eeSAndrea Arcangeli * dangling pointer). 944e86f15eeSAndrea Arcangeli */ 945e86f15eeSAndrea Arcangeli next = vma; 946e86f15eeSAndrea Arcangeli } 947734537c9SKirill A. Shutemov if (remove_next == 2) { 948734537c9SKirill A. Shutemov remove_next = 1; 949734537c9SKirill A. Shutemov end = next->vm_end; 9501da177e4SLinus Torvalds goto again; 951734537c9SKirill A. Shutemov } 952d3737187SMichel Lespinasse else if (next) 953d3737187SMichel Lespinasse vma_gap_update(next); 954fb8c41e9SAndrea Arcangeli else { 955fb8c41e9SAndrea Arcangeli /* 956fb8c41e9SAndrea Arcangeli * If remove_next == 2 we obviously can't 957fb8c41e9SAndrea Arcangeli * reach this path. 958fb8c41e9SAndrea Arcangeli * 959fb8c41e9SAndrea Arcangeli * If remove_next == 3 we can't reach this 960fb8c41e9SAndrea Arcangeli * path because pre-swap() next is always not 961fb8c41e9SAndrea Arcangeli * NULL. pre-swap() "next" is not being 962fb8c41e9SAndrea Arcangeli * removed and its next->vm_end is not altered 963fb8c41e9SAndrea Arcangeli * (and furthermore "end" already matches 964fb8c41e9SAndrea Arcangeli * next->vm_end in remove_next == 3). 965fb8c41e9SAndrea Arcangeli * 966fb8c41e9SAndrea Arcangeli * We reach this only in the remove_next == 1 967fb8c41e9SAndrea Arcangeli * case if the "next" vma that was removed was 968fb8c41e9SAndrea Arcangeli * the highest vma of the mm. However in such 969fb8c41e9SAndrea Arcangeli * case next->vm_end == "end" and the extended 970fb8c41e9SAndrea Arcangeli * "vma" has vma->vm_end == next->vm_end so 971fb8c41e9SAndrea Arcangeli * mm->highest_vm_end doesn't need any update 972fb8c41e9SAndrea Arcangeli * in remove_next == 1 case. 973fb8c41e9SAndrea Arcangeli */ 9741be7107fSHugh Dickins VM_WARN_ON(mm->highest_vm_end != vm_end_gap(vma)); 975fb8c41e9SAndrea Arcangeli } 9761da177e4SLinus Torvalds } 9772b144498SSrikar Dronamraju if (insert && file) 9787b2d81d4SIngo Molnar uprobe_mmap(insert); 9791da177e4SLinus Torvalds 9801da177e4SLinus Torvalds validate_mm(mm); 9815beb4930SRik van Riel 9825beb4930SRik van Riel return 0; 9831da177e4SLinus Torvalds } 9841da177e4SLinus Torvalds 9851da177e4SLinus Torvalds /* 9861da177e4SLinus Torvalds * If the vma has a ->close operation then the driver probably needs to release 9871da177e4SLinus Torvalds * per-vma resources, so we don't attempt to merge those. 9881da177e4SLinus Torvalds */ 9891da177e4SLinus Torvalds static inline int is_mergeable_vma(struct vm_area_struct *vma, 99019a809afSAndrea Arcangeli struct file *file, unsigned long vm_flags, 9919a10064fSColin Cross struct vm_userfaultfd_ctx vm_userfaultfd_ctx, 9925c26f6acSSuren Baghdasaryan struct anon_vma_name *anon_name) 9931da177e4SLinus Torvalds { 99434228d47SCyrill Gorcunov /* 99534228d47SCyrill Gorcunov * VM_SOFTDIRTY should not prevent from VMA merging, if we 99634228d47SCyrill Gorcunov * match the flags but dirty bit -- the caller should mark 99734228d47SCyrill Gorcunov * merged VMA as dirty. If dirty bit won't be excluded from 9988bb4e7a2SWei Yang * comparison, we increase pressure on the memory system forcing 99934228d47SCyrill Gorcunov * the kernel to generate new VMAs when old one could be 100034228d47SCyrill Gorcunov * extended instead. 100134228d47SCyrill Gorcunov */ 100234228d47SCyrill Gorcunov if ((vma->vm_flags ^ vm_flags) & ~VM_SOFTDIRTY) 10031da177e4SLinus Torvalds return 0; 10041da177e4SLinus Torvalds if (vma->vm_file != file) 10051da177e4SLinus Torvalds return 0; 10061da177e4SLinus Torvalds if (vma->vm_ops && vma->vm_ops->close) 10071da177e4SLinus Torvalds return 0; 100819a809afSAndrea Arcangeli if (!is_mergeable_vm_userfaultfd_ctx(vma, vm_userfaultfd_ctx)) 100919a809afSAndrea Arcangeli return 0; 10105c26f6acSSuren Baghdasaryan if (!anon_vma_name_eq(anon_vma_name(vma), anon_name)) 10119a10064fSColin Cross return 0; 10121da177e4SLinus Torvalds return 1; 10131da177e4SLinus Torvalds } 10141da177e4SLinus Torvalds 10151da177e4SLinus Torvalds static inline int is_mergeable_anon_vma(struct anon_vma *anon_vma1, 1016965f55deSShaohua Li struct anon_vma *anon_vma2, 1017965f55deSShaohua Li struct vm_area_struct *vma) 10181da177e4SLinus Torvalds { 1019965f55deSShaohua Li /* 1020965f55deSShaohua Li * The list_is_singular() test is to avoid merging VMA cloned from 1021965f55deSShaohua Li * parents. This can improve scalability caused by anon_vma lock. 1022965f55deSShaohua Li */ 1023965f55deSShaohua Li if ((!anon_vma1 || !anon_vma2) && (!vma || 1024965f55deSShaohua Li list_is_singular(&vma->anon_vma_chain))) 1025965f55deSShaohua Li return 1; 1026965f55deSShaohua Li return anon_vma1 == anon_vma2; 10271da177e4SLinus Torvalds } 10281da177e4SLinus Torvalds 10291da177e4SLinus Torvalds /* 10301da177e4SLinus Torvalds * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff) 10311da177e4SLinus Torvalds * in front of (at a lower virtual address and file offset than) the vma. 10321da177e4SLinus Torvalds * 10331da177e4SLinus Torvalds * We cannot merge two vmas if they have differently assigned (non-NULL) 10341da177e4SLinus Torvalds * anon_vmas, nor if same anon_vma is assigned but offsets incompatible. 10351da177e4SLinus Torvalds * 10361da177e4SLinus Torvalds * We don't check here for the merged mmap wrapping around the end of pagecache 103745e55300SPeter Collingbourne * indices (16TB on ia32) because do_mmap() does not permit mmap's which 10381da177e4SLinus Torvalds * wrap, nor mmaps which cover the final page at index -1UL. 10391da177e4SLinus Torvalds */ 10401da177e4SLinus Torvalds static int 10411da177e4SLinus Torvalds can_vma_merge_before(struct vm_area_struct *vma, unsigned long vm_flags, 104219a809afSAndrea Arcangeli struct anon_vma *anon_vma, struct file *file, 104319a809afSAndrea Arcangeli pgoff_t vm_pgoff, 10449a10064fSColin Cross struct vm_userfaultfd_ctx vm_userfaultfd_ctx, 10455c26f6acSSuren Baghdasaryan struct anon_vma_name *anon_name) 10461da177e4SLinus Torvalds { 10479a10064fSColin Cross if (is_mergeable_vma(vma, file, vm_flags, vm_userfaultfd_ctx, anon_name) && 1048965f55deSShaohua Li is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) { 10491da177e4SLinus Torvalds if (vma->vm_pgoff == vm_pgoff) 10501da177e4SLinus Torvalds return 1; 10511da177e4SLinus Torvalds } 10521da177e4SLinus Torvalds return 0; 10531da177e4SLinus Torvalds } 10541da177e4SLinus Torvalds 10551da177e4SLinus Torvalds /* 10561da177e4SLinus Torvalds * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff) 10571da177e4SLinus Torvalds * beyond (at a higher virtual address and file offset than) the vma. 10581da177e4SLinus Torvalds * 10591da177e4SLinus Torvalds * We cannot merge two vmas if they have differently assigned (non-NULL) 10601da177e4SLinus Torvalds * anon_vmas, nor if same anon_vma is assigned but offsets incompatible. 10611da177e4SLinus Torvalds */ 10621da177e4SLinus Torvalds static int 10631da177e4SLinus Torvalds can_vma_merge_after(struct vm_area_struct *vma, unsigned long vm_flags, 106419a809afSAndrea Arcangeli struct anon_vma *anon_vma, struct file *file, 106519a809afSAndrea Arcangeli pgoff_t vm_pgoff, 10669a10064fSColin Cross struct vm_userfaultfd_ctx vm_userfaultfd_ctx, 10675c26f6acSSuren Baghdasaryan struct anon_vma_name *anon_name) 10681da177e4SLinus Torvalds { 10699a10064fSColin Cross if (is_mergeable_vma(vma, file, vm_flags, vm_userfaultfd_ctx, anon_name) && 1070965f55deSShaohua Li is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) { 10711da177e4SLinus Torvalds pgoff_t vm_pglen; 1072d6e93217SLibin vm_pglen = vma_pages(vma); 10731da177e4SLinus Torvalds if (vma->vm_pgoff + vm_pglen == vm_pgoff) 10741da177e4SLinus Torvalds return 1; 10751da177e4SLinus Torvalds } 10761da177e4SLinus Torvalds return 0; 10771da177e4SLinus Torvalds } 10781da177e4SLinus Torvalds 10791da177e4SLinus Torvalds /* 10809a10064fSColin Cross * Given a mapping request (addr,end,vm_flags,file,pgoff,anon_name), 10819a10064fSColin Cross * figure out whether that can be merged with its predecessor or its 10829a10064fSColin Cross * successor. Or both (it neatly fills a hole). 10831da177e4SLinus Torvalds * 10841da177e4SLinus Torvalds * In most cases - when called for mmap, brk or mremap - [addr,end) is 10851da177e4SLinus Torvalds * certain not to be mapped by the time vma_merge is called; but when 10861da177e4SLinus Torvalds * called for mprotect, it is certain to be already mapped (either at 10871da177e4SLinus Torvalds * an offset within prev, or at the start of next), and the flags of 10881da177e4SLinus Torvalds * this area are about to be changed to vm_flags - and the no-change 10891da177e4SLinus Torvalds * case has already been eliminated. 10901da177e4SLinus Torvalds * 10911da177e4SLinus Torvalds * The following mprotect cases have to be considered, where AAAA is 10921da177e4SLinus Torvalds * the area passed down from mprotect_fixup, never extending beyond one 10931da177e4SLinus Torvalds * vma, PPPPPP is the prev vma specified, and NNNNNN the next vma after: 10941da177e4SLinus Torvalds * 10955d42ab29SWei Yang * AAAA AAAA AAAA 10965d42ab29SWei Yang * PPPPPPNNNNNN PPPPPPNNNNNN PPPPPPNNNNNN 10975d42ab29SWei Yang * cannot merge might become might become 10985d42ab29SWei Yang * PPNNNNNNNNNN PPPPPPPPPPNN 10995d42ab29SWei Yang * mmap, brk or case 4 below case 5 below 11005d42ab29SWei Yang * mremap move: 11015d42ab29SWei Yang * AAAA AAAA 11025d42ab29SWei Yang * PPPP NNNN PPPPNNNNXXXX 11035d42ab29SWei Yang * might become might become 11045d42ab29SWei Yang * PPPPPPPPPPPP 1 or PPPPPPPPPPPP 6 or 11055d42ab29SWei Yang * PPPPPPPPNNNN 2 or PPPPPPPPXXXX 7 or 11065d42ab29SWei Yang * PPPPNNNNNNNN 3 PPPPXXXXXXXX 8 11071da177e4SLinus Torvalds * 11088bb4e7a2SWei Yang * It is important for case 8 that the vma NNNN overlapping the 1109e86f15eeSAndrea Arcangeli * region AAAA is never going to extended over XXXX. Instead XXXX must 1110e86f15eeSAndrea Arcangeli * be extended in region AAAA and NNNN must be removed. This way in 1111e86f15eeSAndrea Arcangeli * all cases where vma_merge succeeds, the moment vma_adjust drops the 1112e86f15eeSAndrea Arcangeli * rmap_locks, the properties of the merged vma will be already 1113e86f15eeSAndrea Arcangeli * correct for the whole merged range. Some of those properties like 1114e86f15eeSAndrea Arcangeli * vm_page_prot/vm_flags may be accessed by rmap_walks and they must 1115e86f15eeSAndrea Arcangeli * be correct for the whole merged range immediately after the 1116e86f15eeSAndrea Arcangeli * rmap_locks are released. Otherwise if XXXX would be removed and 1117e86f15eeSAndrea Arcangeli * NNNN would be extended over the XXXX range, remove_migration_ptes 1118e86f15eeSAndrea Arcangeli * or other rmap walkers (if working on addresses beyond the "end" 1119e86f15eeSAndrea Arcangeli * parameter) may establish ptes with the wrong permissions of NNNN 1120e86f15eeSAndrea Arcangeli * instead of the right permissions of XXXX. 11211da177e4SLinus Torvalds */ 11221da177e4SLinus Torvalds struct vm_area_struct *vma_merge(struct mm_struct *mm, 11231da177e4SLinus Torvalds struct vm_area_struct *prev, unsigned long addr, 11241da177e4SLinus Torvalds unsigned long end, unsigned long vm_flags, 11251da177e4SLinus Torvalds struct anon_vma *anon_vma, struct file *file, 112619a809afSAndrea Arcangeli pgoff_t pgoff, struct mempolicy *policy, 11279a10064fSColin Cross struct vm_userfaultfd_ctx vm_userfaultfd_ctx, 11285c26f6acSSuren Baghdasaryan struct anon_vma_name *anon_name) 11291da177e4SLinus Torvalds { 11301da177e4SLinus Torvalds pgoff_t pglen = (end - addr) >> PAGE_SHIFT; 11311da177e4SLinus Torvalds struct vm_area_struct *area, *next; 11325beb4930SRik van Riel int err; 11331da177e4SLinus Torvalds 11341da177e4SLinus Torvalds /* 11351da177e4SLinus Torvalds * We later require that vma->vm_flags == vm_flags, 11361da177e4SLinus Torvalds * so this tests vma->vm_flags & VM_SPECIAL, too. 11371da177e4SLinus Torvalds */ 11381da177e4SLinus Torvalds if (vm_flags & VM_SPECIAL) 11391da177e4SLinus Torvalds return NULL; 11401da177e4SLinus Torvalds 11413903b55aSLiam R. Howlett next = vma_next(mm, prev); 11421da177e4SLinus Torvalds area = next; 1143e86f15eeSAndrea Arcangeli if (area && area->vm_end == end) /* cases 6, 7, 8 */ 11441da177e4SLinus Torvalds next = next->vm_next; 11451da177e4SLinus Torvalds 1146e86f15eeSAndrea Arcangeli /* verify some invariant that must be enforced by the caller */ 1147e86f15eeSAndrea Arcangeli VM_WARN_ON(prev && addr <= prev->vm_start); 1148e86f15eeSAndrea Arcangeli VM_WARN_ON(area && end > area->vm_end); 1149e86f15eeSAndrea Arcangeli VM_WARN_ON(addr >= end); 1150e86f15eeSAndrea Arcangeli 11511da177e4SLinus Torvalds /* 11521da177e4SLinus Torvalds * Can it merge with the predecessor? 11531da177e4SLinus Torvalds */ 11541da177e4SLinus Torvalds if (prev && prev->vm_end == addr && 11551da177e4SLinus Torvalds mpol_equal(vma_policy(prev), policy) && 11561da177e4SLinus Torvalds can_vma_merge_after(prev, vm_flags, 115719a809afSAndrea Arcangeli anon_vma, file, pgoff, 11589a10064fSColin Cross vm_userfaultfd_ctx, anon_name)) { 11591da177e4SLinus Torvalds /* 11601da177e4SLinus Torvalds * OK, it can. Can we now merge in the successor as well? 11611da177e4SLinus Torvalds */ 11621da177e4SLinus Torvalds if (next && end == next->vm_start && 11631da177e4SLinus Torvalds mpol_equal(policy, vma_policy(next)) && 11641da177e4SLinus Torvalds can_vma_merge_before(next, vm_flags, 116519a809afSAndrea Arcangeli anon_vma, file, 116619a809afSAndrea Arcangeli pgoff+pglen, 11679a10064fSColin Cross vm_userfaultfd_ctx, anon_name) && 11681da177e4SLinus Torvalds is_mergeable_anon_vma(prev->anon_vma, 1169965f55deSShaohua Li next->anon_vma, NULL)) { 11701da177e4SLinus Torvalds /* cases 1, 6 */ 1171e86f15eeSAndrea Arcangeli err = __vma_adjust(prev, prev->vm_start, 1172e86f15eeSAndrea Arcangeli next->vm_end, prev->vm_pgoff, NULL, 1173e86f15eeSAndrea Arcangeli prev); 11741da177e4SLinus Torvalds } else /* cases 2, 5, 7 */ 1175e86f15eeSAndrea Arcangeli err = __vma_adjust(prev, prev->vm_start, 1176e86f15eeSAndrea Arcangeli end, prev->vm_pgoff, NULL, prev); 11775beb4930SRik van Riel if (err) 11785beb4930SRik van Riel return NULL; 1179c791576cSYang Shi khugepaged_enter_vma(prev, vm_flags); 11801da177e4SLinus Torvalds return prev; 11811da177e4SLinus Torvalds } 11821da177e4SLinus Torvalds 11831da177e4SLinus Torvalds /* 11841da177e4SLinus Torvalds * Can this new request be merged in front of next? 11851da177e4SLinus Torvalds */ 11861da177e4SLinus Torvalds if (next && end == next->vm_start && 11871da177e4SLinus Torvalds mpol_equal(policy, vma_policy(next)) && 11881da177e4SLinus Torvalds can_vma_merge_before(next, vm_flags, 118919a809afSAndrea Arcangeli anon_vma, file, pgoff+pglen, 11909a10064fSColin Cross vm_userfaultfd_ctx, anon_name)) { 11911da177e4SLinus Torvalds if (prev && addr < prev->vm_end) /* case 4 */ 1192e86f15eeSAndrea Arcangeli err = __vma_adjust(prev, prev->vm_start, 1193e86f15eeSAndrea Arcangeli addr, prev->vm_pgoff, NULL, next); 1194e86f15eeSAndrea Arcangeli else { /* cases 3, 8 */ 1195e86f15eeSAndrea Arcangeli err = __vma_adjust(area, addr, next->vm_end, 1196e86f15eeSAndrea Arcangeli next->vm_pgoff - pglen, NULL, next); 1197e86f15eeSAndrea Arcangeli /* 1198e86f15eeSAndrea Arcangeli * In case 3 area is already equal to next and 1199e86f15eeSAndrea Arcangeli * this is a noop, but in case 8 "area" has 1200e86f15eeSAndrea Arcangeli * been removed and next was expanded over it. 1201e86f15eeSAndrea Arcangeli */ 1202e86f15eeSAndrea Arcangeli area = next; 1203e86f15eeSAndrea Arcangeli } 12045beb4930SRik van Riel if (err) 12055beb4930SRik van Riel return NULL; 1206c791576cSYang Shi khugepaged_enter_vma(area, vm_flags); 12071da177e4SLinus Torvalds return area; 12081da177e4SLinus Torvalds } 12091da177e4SLinus Torvalds 12101da177e4SLinus Torvalds return NULL; 12111da177e4SLinus Torvalds } 12121da177e4SLinus Torvalds 12131da177e4SLinus Torvalds /* 1214b4f315b4SEthon Paul * Rough compatibility check to quickly see if it's even worth looking 1215d0e9fe17SLinus Torvalds * at sharing an anon_vma. 1216d0e9fe17SLinus Torvalds * 1217d0e9fe17SLinus Torvalds * They need to have the same vm_file, and the flags can only differ 1218d0e9fe17SLinus Torvalds * in things that mprotect may change. 1219d0e9fe17SLinus Torvalds * 1220d0e9fe17SLinus Torvalds * NOTE! The fact that we share an anon_vma doesn't _have_ to mean that 1221d0e9fe17SLinus Torvalds * we can merge the two vma's. For example, we refuse to merge a vma if 1222d0e9fe17SLinus Torvalds * there is a vm_ops->close() function, because that indicates that the 1223d0e9fe17SLinus Torvalds * driver is doing some kind of reference counting. But that doesn't 1224d0e9fe17SLinus Torvalds * really matter for the anon_vma sharing case. 1225d0e9fe17SLinus Torvalds */ 1226d0e9fe17SLinus Torvalds static int anon_vma_compatible(struct vm_area_struct *a, struct vm_area_struct *b) 1227d0e9fe17SLinus Torvalds { 1228d0e9fe17SLinus Torvalds return a->vm_end == b->vm_start && 1229d0e9fe17SLinus Torvalds mpol_equal(vma_policy(a), vma_policy(b)) && 1230d0e9fe17SLinus Torvalds a->vm_file == b->vm_file && 12316cb4d9a2SAnshuman Khandual !((a->vm_flags ^ b->vm_flags) & ~(VM_ACCESS_FLAGS | VM_SOFTDIRTY)) && 1232d0e9fe17SLinus Torvalds b->vm_pgoff == a->vm_pgoff + ((b->vm_start - a->vm_start) >> PAGE_SHIFT); 1233d0e9fe17SLinus Torvalds } 1234d0e9fe17SLinus Torvalds 1235d0e9fe17SLinus Torvalds /* 1236d0e9fe17SLinus Torvalds * Do some basic sanity checking to see if we can re-use the anon_vma 1237d0e9fe17SLinus Torvalds * from 'old'. The 'a'/'b' vma's are in VM order - one of them will be 1238d0e9fe17SLinus Torvalds * the same as 'old', the other will be the new one that is trying 1239d0e9fe17SLinus Torvalds * to share the anon_vma. 1240d0e9fe17SLinus Torvalds * 12415b449489SFlorian Rommel * NOTE! This runs with mmap_lock held for reading, so it is possible that 1242d0e9fe17SLinus Torvalds * the anon_vma of 'old' is concurrently in the process of being set up 1243d0e9fe17SLinus Torvalds * by another page fault trying to merge _that_. But that's ok: if it 1244d0e9fe17SLinus Torvalds * is being set up, that automatically means that it will be a singleton 1245d0e9fe17SLinus Torvalds * acceptable for merging, so we can do all of this optimistically. But 12464db0c3c2SJason Low * we do that READ_ONCE() to make sure that we never re-load the pointer. 1247d0e9fe17SLinus Torvalds * 1248d0e9fe17SLinus Torvalds * IOW: that the "list_is_singular()" test on the anon_vma_chain only 1249d0e9fe17SLinus Torvalds * matters for the 'stable anon_vma' case (ie the thing we want to avoid 1250d0e9fe17SLinus Torvalds * is to return an anon_vma that is "complex" due to having gone through 1251d0e9fe17SLinus Torvalds * a fork). 1252d0e9fe17SLinus Torvalds * 1253d0e9fe17SLinus Torvalds * We also make sure that the two vma's are compatible (adjacent, 1254d0e9fe17SLinus Torvalds * and with the same memory policies). That's all stable, even with just 12555b449489SFlorian Rommel * a read lock on the mmap_lock. 1256d0e9fe17SLinus Torvalds */ 1257d0e9fe17SLinus Torvalds static struct anon_vma *reusable_anon_vma(struct vm_area_struct *old, struct vm_area_struct *a, struct vm_area_struct *b) 1258d0e9fe17SLinus Torvalds { 1259d0e9fe17SLinus Torvalds if (anon_vma_compatible(a, b)) { 12604db0c3c2SJason Low struct anon_vma *anon_vma = READ_ONCE(old->anon_vma); 1261d0e9fe17SLinus Torvalds 1262d0e9fe17SLinus Torvalds if (anon_vma && list_is_singular(&old->anon_vma_chain)) 1263d0e9fe17SLinus Torvalds return anon_vma; 1264d0e9fe17SLinus Torvalds } 1265d0e9fe17SLinus Torvalds return NULL; 1266d0e9fe17SLinus Torvalds } 1267d0e9fe17SLinus Torvalds 1268d0e9fe17SLinus Torvalds /* 12691da177e4SLinus Torvalds * find_mergeable_anon_vma is used by anon_vma_prepare, to check 12701da177e4SLinus Torvalds * neighbouring vmas for a suitable anon_vma, before it goes off 12711da177e4SLinus Torvalds * to allocate a new anon_vma. It checks because a repetitive 12721da177e4SLinus Torvalds * sequence of mprotects and faults may otherwise lead to distinct 12731da177e4SLinus Torvalds * anon_vmas being allocated, preventing vma merge in subsequent 12741da177e4SLinus Torvalds * mprotect. 12751da177e4SLinus Torvalds */ 12761da177e4SLinus Torvalds struct anon_vma *find_mergeable_anon_vma(struct vm_area_struct *vma) 12771da177e4SLinus Torvalds { 1278a67c8caaSMiaohe Lin struct anon_vma *anon_vma = NULL; 12791da177e4SLinus Torvalds 1280a67c8caaSMiaohe Lin /* Try next first. */ 1281a67c8caaSMiaohe Lin if (vma->vm_next) { 1282a67c8caaSMiaohe Lin anon_vma = reusable_anon_vma(vma->vm_next, vma, vma->vm_next); 1283d0e9fe17SLinus Torvalds if (anon_vma) 1284d0e9fe17SLinus Torvalds return anon_vma; 1285a67c8caaSMiaohe Lin } 12861da177e4SLinus Torvalds 1287a67c8caaSMiaohe Lin /* Try prev next. */ 1288a67c8caaSMiaohe Lin if (vma->vm_prev) 1289a67c8caaSMiaohe Lin anon_vma = reusable_anon_vma(vma->vm_prev, vma->vm_prev, vma); 1290a67c8caaSMiaohe Lin 12911da177e4SLinus Torvalds /* 1292a67c8caaSMiaohe Lin * We might reach here with anon_vma == NULL if we can't find 1293a67c8caaSMiaohe Lin * any reusable anon_vma. 12941da177e4SLinus Torvalds * There's no absolute need to look only at touching neighbours: 12951da177e4SLinus Torvalds * we could search further afield for "compatible" anon_vmas. 12961da177e4SLinus Torvalds * But it would probably just be a waste of time searching, 12971da177e4SLinus Torvalds * or lead to too many vmas hanging off the same anon_vma. 12981da177e4SLinus Torvalds * We're trying to allow mprotect remerging later on, 12991da177e4SLinus Torvalds * not trying to minimize memory used for anon_vmas. 13001da177e4SLinus Torvalds */ 1301a67c8caaSMiaohe Lin return anon_vma; 13021da177e4SLinus Torvalds } 13031da177e4SLinus Torvalds 13041da177e4SLinus Torvalds /* 130540401530SAl Viro * If a hint addr is less than mmap_min_addr change hint to be as 130640401530SAl Viro * low as possible but still greater than mmap_min_addr 130740401530SAl Viro */ 130840401530SAl Viro static inline unsigned long round_hint_to_min(unsigned long hint) 130940401530SAl Viro { 131040401530SAl Viro hint &= PAGE_MASK; 131140401530SAl Viro if (((void *)hint != NULL) && 131240401530SAl Viro (hint < mmap_min_addr)) 131340401530SAl Viro return PAGE_ALIGN(mmap_min_addr); 131440401530SAl Viro return hint; 131540401530SAl Viro } 131640401530SAl Viro 13176aeb2542SMike Rapoport int mlock_future_check(struct mm_struct *mm, unsigned long flags, 1318363ee17fSDavidlohr Bueso unsigned long len) 1319363ee17fSDavidlohr Bueso { 1320363ee17fSDavidlohr Bueso unsigned long locked, lock_limit; 1321363ee17fSDavidlohr Bueso 1322363ee17fSDavidlohr Bueso /* mlock MCL_FUTURE? */ 1323363ee17fSDavidlohr Bueso if (flags & VM_LOCKED) { 1324363ee17fSDavidlohr Bueso locked = len >> PAGE_SHIFT; 1325363ee17fSDavidlohr Bueso locked += mm->locked_vm; 1326363ee17fSDavidlohr Bueso lock_limit = rlimit(RLIMIT_MEMLOCK); 1327363ee17fSDavidlohr Bueso lock_limit >>= PAGE_SHIFT; 1328363ee17fSDavidlohr Bueso if (locked > lock_limit && !capable(CAP_IPC_LOCK)) 1329363ee17fSDavidlohr Bueso return -EAGAIN; 1330363ee17fSDavidlohr Bueso } 1331363ee17fSDavidlohr Bueso return 0; 1332363ee17fSDavidlohr Bueso } 1333363ee17fSDavidlohr Bueso 1334be83bbf8SLinus Torvalds static inline u64 file_mmap_size_max(struct file *file, struct inode *inode) 1335be83bbf8SLinus Torvalds { 1336be83bbf8SLinus Torvalds if (S_ISREG(inode->i_mode)) 1337423913adSLinus Torvalds return MAX_LFS_FILESIZE; 1338be83bbf8SLinus Torvalds 1339be83bbf8SLinus Torvalds if (S_ISBLK(inode->i_mode)) 1340be83bbf8SLinus Torvalds return MAX_LFS_FILESIZE; 1341be83bbf8SLinus Torvalds 134276f34950SIvan Khoronzhuk if (S_ISSOCK(inode->i_mode)) 134376f34950SIvan Khoronzhuk return MAX_LFS_FILESIZE; 134476f34950SIvan Khoronzhuk 1345be83bbf8SLinus Torvalds /* Special "we do even unsigned file positions" case */ 1346be83bbf8SLinus Torvalds if (file->f_mode & FMODE_UNSIGNED_OFFSET) 1347be83bbf8SLinus Torvalds return 0; 1348be83bbf8SLinus Torvalds 1349be83bbf8SLinus Torvalds /* Yes, random drivers might want more. But I'm tired of buggy drivers */ 1350be83bbf8SLinus Torvalds return ULONG_MAX; 1351be83bbf8SLinus Torvalds } 1352be83bbf8SLinus Torvalds 1353be83bbf8SLinus Torvalds static inline bool file_mmap_ok(struct file *file, struct inode *inode, 1354be83bbf8SLinus Torvalds unsigned long pgoff, unsigned long len) 1355be83bbf8SLinus Torvalds { 1356be83bbf8SLinus Torvalds u64 maxsize = file_mmap_size_max(file, inode); 1357be83bbf8SLinus Torvalds 1358be83bbf8SLinus Torvalds if (maxsize && len > maxsize) 1359be83bbf8SLinus Torvalds return false; 1360be83bbf8SLinus Torvalds maxsize -= len; 1361be83bbf8SLinus Torvalds if (pgoff > maxsize >> PAGE_SHIFT) 1362be83bbf8SLinus Torvalds return false; 1363be83bbf8SLinus Torvalds return true; 1364be83bbf8SLinus Torvalds } 1365be83bbf8SLinus Torvalds 136640401530SAl Viro /* 13673e4e28c5SMichel Lespinasse * The caller must write-lock current->mm->mmap_lock. 13681da177e4SLinus Torvalds */ 13691fcfd8dbSOleg Nesterov unsigned long do_mmap(struct file *file, unsigned long addr, 13701da177e4SLinus Torvalds unsigned long len, unsigned long prot, 137145e55300SPeter Collingbourne unsigned long flags, unsigned long pgoff, 137245e55300SPeter Collingbourne unsigned long *populate, struct list_head *uf) 13731da177e4SLinus Torvalds { 13741da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 137545e55300SPeter Collingbourne vm_flags_t vm_flags; 137662b5f7d0SDave Hansen int pkey = 0; 13771da177e4SLinus Torvalds 137841badc15SMichel Lespinasse *populate = 0; 1379bebeb3d6SMichel Lespinasse 1380e37609bbSPiotr Kwapulinski if (!len) 1381e37609bbSPiotr Kwapulinski return -EINVAL; 1382e37609bbSPiotr Kwapulinski 13831da177e4SLinus Torvalds /* 13841da177e4SLinus Torvalds * Does the application expect PROT_READ to imply PROT_EXEC? 13851da177e4SLinus Torvalds * 13861da177e4SLinus Torvalds * (the exception is when the underlying filesystem is noexec 13871da177e4SLinus Torvalds * mounted, in which case we dont add PROT_EXEC.) 13881da177e4SLinus Torvalds */ 13891da177e4SLinus Torvalds if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC)) 139090f8572bSEric W. Biederman if (!(file && path_noexec(&file->f_path))) 13911da177e4SLinus Torvalds prot |= PROT_EXEC; 13921da177e4SLinus Torvalds 1393a4ff8e86SMichal Hocko /* force arch specific MAP_FIXED handling in get_unmapped_area */ 1394a4ff8e86SMichal Hocko if (flags & MAP_FIXED_NOREPLACE) 1395a4ff8e86SMichal Hocko flags |= MAP_FIXED; 1396a4ff8e86SMichal Hocko 13977cd94146SEric Paris if (!(flags & MAP_FIXED)) 13987cd94146SEric Paris addr = round_hint_to_min(addr); 13997cd94146SEric Paris 14001da177e4SLinus Torvalds /* Careful about overflows.. */ 14011da177e4SLinus Torvalds len = PAGE_ALIGN(len); 14029206de95SAl Viro if (!len) 14031da177e4SLinus Torvalds return -ENOMEM; 14041da177e4SLinus Torvalds 14051da177e4SLinus Torvalds /* offset overflow? */ 14061da177e4SLinus Torvalds if ((pgoff + (len >> PAGE_SHIFT)) < pgoff) 14071da177e4SLinus Torvalds return -EOVERFLOW; 14081da177e4SLinus Torvalds 14091da177e4SLinus Torvalds /* Too many mappings? */ 14101da177e4SLinus Torvalds if (mm->map_count > sysctl_max_map_count) 14111da177e4SLinus Torvalds return -ENOMEM; 14121da177e4SLinus Torvalds 14131da177e4SLinus Torvalds /* Obtain the address to map to. we verify (or select) it and ensure 14141da177e4SLinus Torvalds * that it represents a valid section of the address space. 14151da177e4SLinus Torvalds */ 14161da177e4SLinus Torvalds addr = get_unmapped_area(file, addr, len, pgoff, flags); 1417ff68dac6SGaowei Pu if (IS_ERR_VALUE(addr)) 14181da177e4SLinus Torvalds return addr; 14191da177e4SLinus Torvalds 1420a4ff8e86SMichal Hocko if (flags & MAP_FIXED_NOREPLACE) { 142135e43c5fSLiam Howlett if (find_vma_intersection(mm, addr, addr + len)) 1422a4ff8e86SMichal Hocko return -EEXIST; 1423a4ff8e86SMichal Hocko } 1424a4ff8e86SMichal Hocko 142562b5f7d0SDave Hansen if (prot == PROT_EXEC) { 142662b5f7d0SDave Hansen pkey = execute_only_pkey(mm); 142762b5f7d0SDave Hansen if (pkey < 0) 142862b5f7d0SDave Hansen pkey = 0; 142962b5f7d0SDave Hansen } 143062b5f7d0SDave Hansen 14311da177e4SLinus Torvalds /* Do simple checking here so the lower-level routines won't have 14321da177e4SLinus Torvalds * to. we assume access permissions have been handled by the open 14331da177e4SLinus Torvalds * of the memory object, so we don't do any here. 14341da177e4SLinus Torvalds */ 143545e55300SPeter Collingbourne vm_flags = calc_vm_prot_bits(prot, pkey) | calc_vm_flag_bits(flags) | 14361da177e4SLinus Torvalds mm->def_flags | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC; 14371da177e4SLinus Torvalds 1438cdf7b341SHuang Shijie if (flags & MAP_LOCKED) 14391da177e4SLinus Torvalds if (!can_do_mlock()) 14401da177e4SLinus Torvalds return -EPERM; 1441ba470de4SRik van Riel 1442363ee17fSDavidlohr Bueso if (mlock_future_check(mm, vm_flags, len)) 14431da177e4SLinus Torvalds return -EAGAIN; 14441da177e4SLinus Torvalds 14451da177e4SLinus Torvalds if (file) { 1446077bf22bSOleg Nesterov struct inode *inode = file_inode(file); 14471c972597SDan Williams unsigned long flags_mask; 14481c972597SDan Williams 1449be83bbf8SLinus Torvalds if (!file_mmap_ok(file, inode, pgoff, len)) 1450be83bbf8SLinus Torvalds return -EOVERFLOW; 1451be83bbf8SLinus Torvalds 14521c972597SDan Williams flags_mask = LEGACY_MAP_MASK | file->f_op->mmap_supported_flags; 1453077bf22bSOleg Nesterov 14541da177e4SLinus Torvalds switch (flags & MAP_TYPE) { 14551da177e4SLinus Torvalds case MAP_SHARED: 14561c972597SDan Williams /* 14571c972597SDan Williams * Force use of MAP_SHARED_VALIDATE with non-legacy 14581c972597SDan Williams * flags. E.g. MAP_SYNC is dangerous to use with 14591c972597SDan Williams * MAP_SHARED as you don't know which consistency model 14601c972597SDan Williams * you will get. We silently ignore unsupported flags 14611c972597SDan Williams * with MAP_SHARED to preserve backward compatibility. 14621c972597SDan Williams */ 14631c972597SDan Williams flags &= LEGACY_MAP_MASK; 1464e4a9bc58SJoe Perches fallthrough; 14651c972597SDan Williams case MAP_SHARED_VALIDATE: 14661c972597SDan Williams if (flags & ~flags_mask) 14671c972597SDan Williams return -EOPNOTSUPP; 1468dc617f29SDarrick J. Wong if (prot & PROT_WRITE) { 1469dc617f29SDarrick J. Wong if (!(file->f_mode & FMODE_WRITE)) 14701da177e4SLinus Torvalds return -EACCES; 1471dc617f29SDarrick J. Wong if (IS_SWAPFILE(file->f_mapping->host)) 1472dc617f29SDarrick J. Wong return -ETXTBSY; 1473dc617f29SDarrick J. Wong } 14741da177e4SLinus Torvalds 14751da177e4SLinus Torvalds /* 14761da177e4SLinus Torvalds * Make sure we don't allow writing to an append-only 14771da177e4SLinus Torvalds * file.. 14781da177e4SLinus Torvalds */ 14791da177e4SLinus Torvalds if (IS_APPEND(inode) && (file->f_mode & FMODE_WRITE)) 14801da177e4SLinus Torvalds return -EACCES; 14811da177e4SLinus Torvalds 14821da177e4SLinus Torvalds vm_flags |= VM_SHARED | VM_MAYSHARE; 14831da177e4SLinus Torvalds if (!(file->f_mode & FMODE_WRITE)) 14841da177e4SLinus Torvalds vm_flags &= ~(VM_MAYWRITE | VM_SHARED); 1485e4a9bc58SJoe Perches fallthrough; 14861da177e4SLinus Torvalds case MAP_PRIVATE: 14871da177e4SLinus Torvalds if (!(file->f_mode & FMODE_READ)) 14881da177e4SLinus Torvalds return -EACCES; 148990f8572bSEric W. Biederman if (path_noexec(&file->f_path)) { 149080c5606cSLinus Torvalds if (vm_flags & VM_EXEC) 149180c5606cSLinus Torvalds return -EPERM; 149280c5606cSLinus Torvalds vm_flags &= ~VM_MAYEXEC; 149380c5606cSLinus Torvalds } 149480c5606cSLinus Torvalds 149572c2d531SAl Viro if (!file->f_op->mmap) 149680c5606cSLinus Torvalds return -ENODEV; 1497b2c56e4fSOleg Nesterov if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP)) 1498b2c56e4fSOleg Nesterov return -EINVAL; 14991da177e4SLinus Torvalds break; 15001da177e4SLinus Torvalds 15011da177e4SLinus Torvalds default: 15021da177e4SLinus Torvalds return -EINVAL; 15031da177e4SLinus Torvalds } 15041da177e4SLinus Torvalds } else { 15051da177e4SLinus Torvalds switch (flags & MAP_TYPE) { 15061da177e4SLinus Torvalds case MAP_SHARED: 1507b2c56e4fSOleg Nesterov if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP)) 1508b2c56e4fSOleg Nesterov return -EINVAL; 1509ce363942STejun Heo /* 1510ce363942STejun Heo * Ignore pgoff. 1511ce363942STejun Heo */ 1512ce363942STejun Heo pgoff = 0; 15131da177e4SLinus Torvalds vm_flags |= VM_SHARED | VM_MAYSHARE; 15141da177e4SLinus Torvalds break; 15151da177e4SLinus Torvalds case MAP_PRIVATE: 15161da177e4SLinus Torvalds /* 15171da177e4SLinus Torvalds * Set pgoff according to addr for anon_vma. 15181da177e4SLinus Torvalds */ 15191da177e4SLinus Torvalds pgoff = addr >> PAGE_SHIFT; 15201da177e4SLinus Torvalds break; 15211da177e4SLinus Torvalds default: 15221da177e4SLinus Torvalds return -EINVAL; 15231da177e4SLinus Torvalds } 15241da177e4SLinus Torvalds } 15251da177e4SLinus Torvalds 1526c22c0d63SMichel Lespinasse /* 1527c22c0d63SMichel Lespinasse * Set 'VM_NORESERVE' if we should not account for the 1528c22c0d63SMichel Lespinasse * memory use of this mapping. 1529c22c0d63SMichel Lespinasse */ 1530c22c0d63SMichel Lespinasse if (flags & MAP_NORESERVE) { 1531c22c0d63SMichel Lespinasse /* We honor MAP_NORESERVE if allowed to overcommit */ 1532c22c0d63SMichel Lespinasse if (sysctl_overcommit_memory != OVERCOMMIT_NEVER) 1533c22c0d63SMichel Lespinasse vm_flags |= VM_NORESERVE; 1534c22c0d63SMichel Lespinasse 1535c22c0d63SMichel Lespinasse /* hugetlb applies strict overcommit unless MAP_NORESERVE */ 1536c22c0d63SMichel Lespinasse if (file && is_file_hugepages(file)) 1537c22c0d63SMichel Lespinasse vm_flags |= VM_NORESERVE; 1538c22c0d63SMichel Lespinasse } 1539c22c0d63SMichel Lespinasse 1540897ab3e0SMike Rapoport addr = mmap_region(file, addr, len, vm_flags, pgoff, uf); 154109a9f1d2SMichel Lespinasse if (!IS_ERR_VALUE(addr) && 154209a9f1d2SMichel Lespinasse ((vm_flags & VM_LOCKED) || 154309a9f1d2SMichel Lespinasse (flags & (MAP_POPULATE | MAP_NONBLOCK)) == MAP_POPULATE)) 154441badc15SMichel Lespinasse *populate = len; 1545bebeb3d6SMichel Lespinasse return addr; 15460165ab44SMiklos Szeredi } 15476be5ceb0SLinus Torvalds 1548a90f590aSDominik Brodowski unsigned long ksys_mmap_pgoff(unsigned long addr, unsigned long len, 1549a90f590aSDominik Brodowski unsigned long prot, unsigned long flags, 1550a90f590aSDominik Brodowski unsigned long fd, unsigned long pgoff) 155166f0dc48SHugh Dickins { 155266f0dc48SHugh Dickins struct file *file = NULL; 15531e3ee14bSChen Gang unsigned long retval; 155466f0dc48SHugh Dickins 155566f0dc48SHugh Dickins if (!(flags & MAP_ANONYMOUS)) { 1556120a795dSAl Viro audit_mmap_fd(fd, flags); 155766f0dc48SHugh Dickins file = fget(fd); 155866f0dc48SHugh Dickins if (!file) 15591e3ee14bSChen Gang return -EBADF; 15607bba8f0eSZhen Lei if (is_file_hugepages(file)) { 1561af73e4d9SNaoya Horiguchi len = ALIGN(len, huge_page_size(hstate_file(file))); 15627bba8f0eSZhen Lei } else if (unlikely(flags & MAP_HUGETLB)) { 1563493af578SJörn Engel retval = -EINVAL; 1564493af578SJörn Engel goto out_fput; 15657bba8f0eSZhen Lei } 156666f0dc48SHugh Dickins } else if (flags & MAP_HUGETLB) { 1567c103a4dcSAndrew Morton struct hstate *hs; 1568af73e4d9SNaoya Horiguchi 156920ac2893SAnshuman Khandual hs = hstate_sizelog((flags >> MAP_HUGE_SHIFT) & MAP_HUGE_MASK); 1570091d0d55SLi Zefan if (!hs) 1571091d0d55SLi Zefan return -EINVAL; 1572091d0d55SLi Zefan 1573091d0d55SLi Zefan len = ALIGN(len, huge_page_size(hs)); 157466f0dc48SHugh Dickins /* 157566f0dc48SHugh Dickins * VM_NORESERVE is used because the reservations will be 157666f0dc48SHugh Dickins * taken when vm_ops->mmap() is called 157766f0dc48SHugh Dickins */ 1578af73e4d9SNaoya Horiguchi file = hugetlb_file_setup(HUGETLB_ANON_FILE, len, 157942d7395fSAndi Kleen VM_NORESERVE, 158083c1fd76Szhangyiru HUGETLB_ANONHUGE_INODE, 158142d7395fSAndi Kleen (flags >> MAP_HUGE_SHIFT) & MAP_HUGE_MASK); 158266f0dc48SHugh Dickins if (IS_ERR(file)) 158366f0dc48SHugh Dickins return PTR_ERR(file); 158466f0dc48SHugh Dickins } 158566f0dc48SHugh Dickins 15869fbeb5abSMichal Hocko retval = vm_mmap_pgoff(file, addr, len, prot, flags, pgoff); 1587493af578SJörn Engel out_fput: 158866f0dc48SHugh Dickins if (file) 158966f0dc48SHugh Dickins fput(file); 159066f0dc48SHugh Dickins return retval; 159166f0dc48SHugh Dickins } 159266f0dc48SHugh Dickins 1593a90f590aSDominik Brodowski SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len, 1594a90f590aSDominik Brodowski unsigned long, prot, unsigned long, flags, 1595a90f590aSDominik Brodowski unsigned long, fd, unsigned long, pgoff) 1596a90f590aSDominik Brodowski { 1597a90f590aSDominik Brodowski return ksys_mmap_pgoff(addr, len, prot, flags, fd, pgoff); 1598a90f590aSDominik Brodowski } 1599a90f590aSDominik Brodowski 1600a4679373SChristoph Hellwig #ifdef __ARCH_WANT_SYS_OLD_MMAP 1601a4679373SChristoph Hellwig struct mmap_arg_struct { 1602a4679373SChristoph Hellwig unsigned long addr; 1603a4679373SChristoph Hellwig unsigned long len; 1604a4679373SChristoph Hellwig unsigned long prot; 1605a4679373SChristoph Hellwig unsigned long flags; 1606a4679373SChristoph Hellwig unsigned long fd; 1607a4679373SChristoph Hellwig unsigned long offset; 1608a4679373SChristoph Hellwig }; 1609a4679373SChristoph Hellwig 1610a4679373SChristoph Hellwig SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg) 1611a4679373SChristoph Hellwig { 1612a4679373SChristoph Hellwig struct mmap_arg_struct a; 1613a4679373SChristoph Hellwig 1614a4679373SChristoph Hellwig if (copy_from_user(&a, arg, sizeof(a))) 1615a4679373SChristoph Hellwig return -EFAULT; 1616de1741a1SAlexander Kuleshov if (offset_in_page(a.offset)) 1617a4679373SChristoph Hellwig return -EINVAL; 1618a4679373SChristoph Hellwig 1619a90f590aSDominik Brodowski return ksys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd, 1620a4679373SChristoph Hellwig a.offset >> PAGE_SHIFT); 1621a4679373SChristoph Hellwig } 1622a4679373SChristoph Hellwig #endif /* __ARCH_WANT_SYS_OLD_MMAP */ 1623a4679373SChristoph Hellwig 16244e950f6fSAlexey Dobriyan /* 16258bb4e7a2SWei Yang * Some shared mappings will want the pages marked read-only 16264e950f6fSAlexey Dobriyan * to track write events. If so, we'll downgrade vm_page_prot 16274e950f6fSAlexey Dobriyan * to the private version (using protection_map[] without the 16284e950f6fSAlexey Dobriyan * VM_SHARED bit). 16294e950f6fSAlexey Dobriyan */ 16306d2329f8SAndrea Arcangeli int vma_wants_writenotify(struct vm_area_struct *vma, pgprot_t vm_page_prot) 16314e950f6fSAlexey Dobriyan { 1632ca16d140SKOSAKI Motohiro vm_flags_t vm_flags = vma->vm_flags; 16338a04446aSKirill A. Shutemov const struct vm_operations_struct *vm_ops = vma->vm_ops; 16344e950f6fSAlexey Dobriyan 16354e950f6fSAlexey Dobriyan /* If it was private or non-writable, the write bit is already clear */ 16364e950f6fSAlexey Dobriyan if ((vm_flags & (VM_WRITE|VM_SHARED)) != ((VM_WRITE|VM_SHARED))) 16374e950f6fSAlexey Dobriyan return 0; 16384e950f6fSAlexey Dobriyan 16394e950f6fSAlexey Dobriyan /* The backer wishes to know when pages are first written to? */ 16408a04446aSKirill A. Shutemov if (vm_ops && (vm_ops->page_mkwrite || vm_ops->pfn_mkwrite)) 16414e950f6fSAlexey Dobriyan return 1; 16424e950f6fSAlexey Dobriyan 164364e45507SPeter Feiner /* The open routine did something to the protections that pgprot_modify 164464e45507SPeter Feiner * won't preserve? */ 16456d2329f8SAndrea Arcangeli if (pgprot_val(vm_page_prot) != 16466d2329f8SAndrea Arcangeli pgprot_val(vm_pgprot_modify(vm_page_prot, vm_flags))) 16474e950f6fSAlexey Dobriyan return 0; 16484e950f6fSAlexey Dobriyan 1649*f96f7a40SDavid Hildenbrand /* 1650*f96f7a40SDavid Hildenbrand * Do we need to track softdirty? hugetlb does not support softdirty 1651*f96f7a40SDavid Hildenbrand * tracking yet. 1652*f96f7a40SDavid Hildenbrand */ 1653*f96f7a40SDavid Hildenbrand if (vma_soft_dirty_enabled(vma) && !is_vm_hugetlb_page(vma)) 165464e45507SPeter Feiner return 1; 165564e45507SPeter Feiner 16564e950f6fSAlexey Dobriyan /* Specialty mapping? */ 16574b6e1e37SKonstantin Khlebnikov if (vm_flags & VM_PFNMAP) 16584e950f6fSAlexey Dobriyan return 0; 16594e950f6fSAlexey Dobriyan 16604e950f6fSAlexey Dobriyan /* Can the mapping track the dirty pages? */ 16614e950f6fSAlexey Dobriyan return vma->vm_file && vma->vm_file->f_mapping && 1662f56753acSChristoph Hellwig mapping_can_writeback(vma->vm_file->f_mapping); 16634e950f6fSAlexey Dobriyan } 16644e950f6fSAlexey Dobriyan 1665fc8744adSLinus Torvalds /* 1666fc8744adSLinus Torvalds * We account for memory if it's a private writeable mapping, 16675a6fe125SMel Gorman * not hugepages and VM_NORESERVE wasn't set. 1668fc8744adSLinus Torvalds */ 1669ca16d140SKOSAKI Motohiro static inline int accountable_mapping(struct file *file, vm_flags_t vm_flags) 1670fc8744adSLinus Torvalds { 16715a6fe125SMel Gorman /* 16725a6fe125SMel Gorman * hugetlb has its own accounting separate from the core VM 16735a6fe125SMel Gorman * VM_HUGETLB may not be set yet so we cannot check for that flag. 16745a6fe125SMel Gorman */ 16755a6fe125SMel Gorman if (file && is_file_hugepages(file)) 16765a6fe125SMel Gorman return 0; 16775a6fe125SMel Gorman 1678fc8744adSLinus Torvalds return (vm_flags & (VM_NORESERVE | VM_SHARED | VM_WRITE)) == VM_WRITE; 1679fc8744adSLinus Torvalds } 1680fc8744adSLinus Torvalds 16810165ab44SMiklos Szeredi unsigned long mmap_region(struct file *file, unsigned long addr, 1682897ab3e0SMike Rapoport unsigned long len, vm_flags_t vm_flags, unsigned long pgoff, 1683897ab3e0SMike Rapoport struct list_head *uf) 16840165ab44SMiklos Szeredi { 16850165ab44SMiklos Szeredi struct mm_struct *mm = current->mm; 1686d70cec89SMiaohe Lin struct vm_area_struct *vma, *prev, *merge; 16870165ab44SMiklos Szeredi int error; 16880165ab44SMiklos Szeredi struct rb_node **rb_link, *rb_parent; 16890165ab44SMiklos Szeredi unsigned long charged = 0; 16900165ab44SMiklos Szeredi 1691e8420a8eSCyril Hrubis /* Check against address space limit. */ 169284638335SKonstantin Khlebnikov if (!may_expand_vm(mm, vm_flags, len >> PAGE_SHIFT)) { 1693e8420a8eSCyril Hrubis unsigned long nr_pages; 1694e8420a8eSCyril Hrubis 1695e8420a8eSCyril Hrubis /* 1696e8420a8eSCyril Hrubis * MAP_FIXED may remove pages of mappings that intersects with 1697e8420a8eSCyril Hrubis * requested mapping. Account for the pages it would unmap. 1698e8420a8eSCyril Hrubis */ 1699e8420a8eSCyril Hrubis nr_pages = count_vma_pages_range(mm, addr, addr + len); 1700e8420a8eSCyril Hrubis 170184638335SKonstantin Khlebnikov if (!may_expand_vm(mm, vm_flags, 170284638335SKonstantin Khlebnikov (len >> PAGE_SHIFT) - nr_pages)) 1703e8420a8eSCyril Hrubis return -ENOMEM; 1704e8420a8eSCyril Hrubis } 1705e8420a8eSCyril Hrubis 1706fb8090b6SLiam R. Howlett /* Clear old maps, set up prev, rb_link, rb_parent, and uf */ 1707fb8090b6SLiam R. Howlett if (munmap_vma_range(mm, addr, len, &prev, &rb_link, &rb_parent, uf)) 17081da177e4SLinus Torvalds return -ENOMEM; 1709fc8744adSLinus Torvalds /* 17101da177e4SLinus Torvalds * Private writable mapping: check memory availability 17111da177e4SLinus Torvalds */ 17125a6fe125SMel Gorman if (accountable_mapping(file, vm_flags)) { 17131da177e4SLinus Torvalds charged = len >> PAGE_SHIFT; 1714191c5424SAl Viro if (security_vm_enough_memory_mm(mm, charged)) 17151da177e4SLinus Torvalds return -ENOMEM; 17161da177e4SLinus Torvalds vm_flags |= VM_ACCOUNT; 17171da177e4SLinus Torvalds } 17181da177e4SLinus Torvalds 17191da177e4SLinus Torvalds /* 1720de33c8dbSLinus Torvalds * Can we just expand an old mapping? 17211da177e4SLinus Torvalds */ 172219a809afSAndrea Arcangeli vma = vma_merge(mm, prev, addr, addr + len, vm_flags, 17239a10064fSColin Cross NULL, file, pgoff, NULL, NULL_VM_UFFD_CTX, NULL); 1724ba470de4SRik van Riel if (vma) 17251da177e4SLinus Torvalds goto out; 17261da177e4SLinus Torvalds 17271da177e4SLinus Torvalds /* 17281da177e4SLinus Torvalds * Determine the object being mapped and call the appropriate 17291da177e4SLinus Torvalds * specific mapper. the address has already been validated, but 17301da177e4SLinus Torvalds * not unmapped, but the maps are removed from the list. 17311da177e4SLinus Torvalds */ 1732490fc053SLinus Torvalds vma = vm_area_alloc(mm); 17331da177e4SLinus Torvalds if (!vma) { 17341da177e4SLinus Torvalds error = -ENOMEM; 17351da177e4SLinus Torvalds goto unacct_error; 17361da177e4SLinus Torvalds } 17371da177e4SLinus Torvalds 17381da177e4SLinus Torvalds vma->vm_start = addr; 17391da177e4SLinus Torvalds vma->vm_end = addr + len; 17401da177e4SLinus Torvalds vma->vm_flags = vm_flags; 17413ed75eb8SColy Li vma->vm_page_prot = vm_get_page_prot(vm_flags); 17421da177e4SLinus Torvalds vma->vm_pgoff = pgoff; 17431da177e4SLinus Torvalds 17441da177e4SLinus Torvalds if (file) { 17454bb5f5d9SDavid Herrmann if (vm_flags & VM_SHARED) { 17464bb5f5d9SDavid Herrmann error = mapping_map_writable(file->f_mapping); 17474bb5f5d9SDavid Herrmann if (error) 17488d0920bdSDavid Hildenbrand goto free_vma; 17494bb5f5d9SDavid Herrmann } 17504bb5f5d9SDavid Herrmann 1751cb0942b8SAl Viro vma->vm_file = get_file(file); 1752f74ac015SMiklos Szeredi error = call_mmap(file, vma); 17531da177e4SLinus Torvalds if (error) 17541da177e4SLinus Torvalds goto unmap_and_free_vma; 17551da177e4SLinus Torvalds 1756309d08d9SLiu Zixian /* Can addr have changed?? 1757309d08d9SLiu Zixian * 1758309d08d9SLiu Zixian * Answer: Yes, several device drivers can do it in their 1759309d08d9SLiu Zixian * f_op->mmap method. -DaveM 1760309d08d9SLiu Zixian * Bug: If addr is changed, prev, rb_link, rb_parent should 1761309d08d9SLiu Zixian * be updated for vma_link() 1762309d08d9SLiu Zixian */ 1763309d08d9SLiu Zixian WARN_ON_ONCE(addr != vma->vm_start); 1764309d08d9SLiu Zixian 1765309d08d9SLiu Zixian addr = vma->vm_start; 1766309d08d9SLiu Zixian 1767d70cec89SMiaohe Lin /* If vm_flags changed after call_mmap(), we should try merge vma again 1768d70cec89SMiaohe Lin * as we may succeed this time. 1769d70cec89SMiaohe Lin */ 1770d70cec89SMiaohe Lin if (unlikely(vm_flags != vma->vm_flags && prev)) { 1771d70cec89SMiaohe Lin merge = vma_merge(mm, prev, vma->vm_start, vma->vm_end, vma->vm_flags, 17729a10064fSColin Cross NULL, vma->vm_file, vma->vm_pgoff, NULL, NULL_VM_UFFD_CTX, NULL); 1773d70cec89SMiaohe Lin if (merge) { 1774bc4fe4cdSMiaohe Lin /* ->mmap() can change vma->vm_file and fput the original file. So 1775bc4fe4cdSMiaohe Lin * fput the vma->vm_file here or we would add an extra fput for file 1776bc4fe4cdSMiaohe Lin * and cause general protection fault ultimately. 1777bc4fe4cdSMiaohe Lin */ 1778bc4fe4cdSMiaohe Lin fput(vma->vm_file); 1779d70cec89SMiaohe Lin vm_area_free(vma); 1780d70cec89SMiaohe Lin vma = merge; 1781309d08d9SLiu Zixian /* Update vm_flags to pick up the change. */ 1782d70cec89SMiaohe Lin vm_flags = vma->vm_flags; 1783d70cec89SMiaohe Lin goto unmap_writable; 1784d70cec89SMiaohe Lin } 1785d70cec89SMiaohe Lin } 1786d70cec89SMiaohe Lin 17871da177e4SLinus Torvalds vm_flags = vma->vm_flags; 1788f8dbf0a7SHuang Shijie } else if (vm_flags & VM_SHARED) { 1789f8dbf0a7SHuang Shijie error = shmem_zero_setup(vma); 1790f8dbf0a7SHuang Shijie if (error) 1791f8dbf0a7SHuang Shijie goto free_vma; 1792bfd40eafSKirill A. Shutemov } else { 1793bfd40eafSKirill A. Shutemov vma_set_anonymous(vma); 1794f8dbf0a7SHuang Shijie } 17951da177e4SLinus Torvalds 1796c462ac28SCatalin Marinas /* Allow architectures to sanity-check the vm_flags */ 1797c462ac28SCatalin Marinas if (!arch_validate_flags(vma->vm_flags)) { 1798c462ac28SCatalin Marinas error = -EINVAL; 1799c462ac28SCatalin Marinas if (file) 1800c462ac28SCatalin Marinas goto unmap_and_free_vma; 1801c462ac28SCatalin Marinas else 1802c462ac28SCatalin Marinas goto free_vma; 1803c462ac28SCatalin Marinas } 1804c462ac28SCatalin Marinas 18054d3d5b41SOleg Nesterov vma_link(mm, vma, prev, rb_link, rb_parent); 1806613bec09SYang Shi 1807613bec09SYang Shi /* 1808613bec09SYang Shi * vma_merge() calls khugepaged_enter_vma() either, the below 1809613bec09SYang Shi * call covers the non-merge case. 1810613bec09SYang Shi */ 1811613bec09SYang Shi khugepaged_enter_vma(vma, vma->vm_flags); 1812613bec09SYang Shi 18134d3d5b41SOleg Nesterov /* Once vma denies write, undo our temporary denial count */ 1814d70cec89SMiaohe Lin unmap_writable: 18158d0920bdSDavid Hildenbrand if (file && vm_flags & VM_SHARED) 18164bb5f5d9SDavid Herrmann mapping_unmap_writable(file->f_mapping); 1817e8686772SOleg Nesterov file = vma->vm_file; 18181da177e4SLinus Torvalds out: 1819cdd6c482SIngo Molnar perf_event_mmap(vma); 18200a4a9391SPeter Zijlstra 182184638335SKonstantin Khlebnikov vm_stat_account(mm, vm_flags, len >> PAGE_SHIFT); 18221da177e4SLinus Torvalds if (vm_flags & VM_LOCKED) { 1823e1fb4a08SDave Jiang if ((vm_flags & VM_SPECIAL) || vma_is_dax(vma) || 1824e1fb4a08SDave Jiang is_vm_hugetlb_page(vma) || 1825e1fb4a08SDave Jiang vma == get_gate_vma(current->mm)) 1826de60f5f1SEric B Munson vma->vm_flags &= VM_LOCKED_CLEAR_MASK; 1827e1fb4a08SDave Jiang else 1828e1fb4a08SDave Jiang mm->locked_vm += (len >> PAGE_SHIFT); 1829bebeb3d6SMichel Lespinasse } 18302b144498SSrikar Dronamraju 1831c7a3a88cSOleg Nesterov if (file) 1832c7a3a88cSOleg Nesterov uprobe_mmap(vma); 18332b144498SSrikar Dronamraju 1834d9104d1cSCyrill Gorcunov /* 1835d9104d1cSCyrill Gorcunov * New (or expanded) vma always get soft dirty status. 1836d9104d1cSCyrill Gorcunov * Otherwise user-space soft-dirty page tracker won't 1837d9104d1cSCyrill Gorcunov * be able to distinguish situation when vma area unmapped, 1838d9104d1cSCyrill Gorcunov * then new mapped in-place (which must be aimed as 1839d9104d1cSCyrill Gorcunov * a completely new data area). 1840d9104d1cSCyrill Gorcunov */ 1841d9104d1cSCyrill Gorcunov vma->vm_flags |= VM_SOFTDIRTY; 1842d9104d1cSCyrill Gorcunov 184364e45507SPeter Feiner vma_set_page_prot(vma); 184464e45507SPeter Feiner 18451da177e4SLinus Torvalds return addr; 18461da177e4SLinus Torvalds 18471da177e4SLinus Torvalds unmap_and_free_vma: 18481527f926SChristian König fput(vma->vm_file); 18491da177e4SLinus Torvalds vma->vm_file = NULL; 18501da177e4SLinus Torvalds 18511da177e4SLinus Torvalds /* Undo any partial mapping done by a device driver. */ 1852e0da382cSHugh Dickins unmap_region(mm, vma, prev, vma->vm_start, vma->vm_end); 18534bb5f5d9SDavid Herrmann if (vm_flags & VM_SHARED) 18544bb5f5d9SDavid Herrmann mapping_unmap_writable(file->f_mapping); 18551da177e4SLinus Torvalds free_vma: 18563928d4f5SLinus Torvalds vm_area_free(vma); 18571da177e4SLinus Torvalds unacct_error: 18581da177e4SLinus Torvalds if (charged) 18591da177e4SLinus Torvalds vm_unacct_memory(charged); 18601da177e4SLinus Torvalds return error; 18611da177e4SLinus Torvalds } 18621da177e4SLinus Torvalds 1863baceaf1cSJaewon Kim static unsigned long unmapped_area(struct vm_unmapped_area_info *info) 1864db4fbfb9SMichel Lespinasse { 1865db4fbfb9SMichel Lespinasse /* 1866db4fbfb9SMichel Lespinasse * We implement the search by looking for an rbtree node that 1867db4fbfb9SMichel Lespinasse * immediately follows a suitable gap. That is, 1868db4fbfb9SMichel Lespinasse * - gap_start = vma->vm_prev->vm_end <= info->high_limit - length; 1869db4fbfb9SMichel Lespinasse * - gap_end = vma->vm_start >= info->low_limit + length; 1870db4fbfb9SMichel Lespinasse * - gap_end - gap_start >= length 1871db4fbfb9SMichel Lespinasse */ 1872db4fbfb9SMichel Lespinasse 1873db4fbfb9SMichel Lespinasse struct mm_struct *mm = current->mm; 1874db4fbfb9SMichel Lespinasse struct vm_area_struct *vma; 1875db4fbfb9SMichel Lespinasse unsigned long length, low_limit, high_limit, gap_start, gap_end; 1876db4fbfb9SMichel Lespinasse 1877db4fbfb9SMichel Lespinasse /* Adjust search length to account for worst case alignment overhead */ 1878db4fbfb9SMichel Lespinasse length = info->length + info->align_mask; 1879db4fbfb9SMichel Lespinasse if (length < info->length) 1880db4fbfb9SMichel Lespinasse return -ENOMEM; 1881db4fbfb9SMichel Lespinasse 1882db4fbfb9SMichel Lespinasse /* Adjust search limits by the desired length */ 1883db4fbfb9SMichel Lespinasse if (info->high_limit < length) 1884db4fbfb9SMichel Lespinasse return -ENOMEM; 1885db4fbfb9SMichel Lespinasse high_limit = info->high_limit - length; 1886db4fbfb9SMichel Lespinasse 1887db4fbfb9SMichel Lespinasse if (info->low_limit > high_limit) 1888db4fbfb9SMichel Lespinasse return -ENOMEM; 1889db4fbfb9SMichel Lespinasse low_limit = info->low_limit + length; 1890db4fbfb9SMichel Lespinasse 1891db4fbfb9SMichel Lespinasse /* Check if rbtree root looks promising */ 1892db4fbfb9SMichel Lespinasse if (RB_EMPTY_ROOT(&mm->mm_rb)) 1893db4fbfb9SMichel Lespinasse goto check_highest; 1894db4fbfb9SMichel Lespinasse vma = rb_entry(mm->mm_rb.rb_node, struct vm_area_struct, vm_rb); 1895db4fbfb9SMichel Lespinasse if (vma->rb_subtree_gap < length) 1896db4fbfb9SMichel Lespinasse goto check_highest; 1897db4fbfb9SMichel Lespinasse 1898db4fbfb9SMichel Lespinasse while (true) { 1899db4fbfb9SMichel Lespinasse /* Visit left subtree if it looks promising */ 19001be7107fSHugh Dickins gap_end = vm_start_gap(vma); 1901db4fbfb9SMichel Lespinasse if (gap_end >= low_limit && vma->vm_rb.rb_left) { 1902db4fbfb9SMichel Lespinasse struct vm_area_struct *left = 1903db4fbfb9SMichel Lespinasse rb_entry(vma->vm_rb.rb_left, 1904db4fbfb9SMichel Lespinasse struct vm_area_struct, vm_rb); 1905db4fbfb9SMichel Lespinasse if (left->rb_subtree_gap >= length) { 1906db4fbfb9SMichel Lespinasse vma = left; 1907db4fbfb9SMichel Lespinasse continue; 1908db4fbfb9SMichel Lespinasse } 1909db4fbfb9SMichel Lespinasse } 1910db4fbfb9SMichel Lespinasse 19111be7107fSHugh Dickins gap_start = vma->vm_prev ? vm_end_gap(vma->vm_prev) : 0; 1912db4fbfb9SMichel Lespinasse check_current: 1913db4fbfb9SMichel Lespinasse /* Check if current node has a suitable gap */ 1914db4fbfb9SMichel Lespinasse if (gap_start > high_limit) 1915db4fbfb9SMichel Lespinasse return -ENOMEM; 1916f4cb767dSHugh Dickins if (gap_end >= low_limit && 1917f4cb767dSHugh Dickins gap_end > gap_start && gap_end - gap_start >= length) 1918db4fbfb9SMichel Lespinasse goto found; 1919db4fbfb9SMichel Lespinasse 1920db4fbfb9SMichel Lespinasse /* Visit right subtree if it looks promising */ 1921db4fbfb9SMichel Lespinasse if (vma->vm_rb.rb_right) { 1922db4fbfb9SMichel Lespinasse struct vm_area_struct *right = 1923db4fbfb9SMichel Lespinasse rb_entry(vma->vm_rb.rb_right, 1924db4fbfb9SMichel Lespinasse struct vm_area_struct, vm_rb); 1925db4fbfb9SMichel Lespinasse if (right->rb_subtree_gap >= length) { 1926db4fbfb9SMichel Lespinasse vma = right; 1927db4fbfb9SMichel Lespinasse continue; 1928db4fbfb9SMichel Lespinasse } 1929db4fbfb9SMichel Lespinasse } 1930db4fbfb9SMichel Lespinasse 1931db4fbfb9SMichel Lespinasse /* Go back up the rbtree to find next candidate node */ 1932db4fbfb9SMichel Lespinasse while (true) { 1933db4fbfb9SMichel Lespinasse struct rb_node *prev = &vma->vm_rb; 1934db4fbfb9SMichel Lespinasse if (!rb_parent(prev)) 1935db4fbfb9SMichel Lespinasse goto check_highest; 1936db4fbfb9SMichel Lespinasse vma = rb_entry(rb_parent(prev), 1937db4fbfb9SMichel Lespinasse struct vm_area_struct, vm_rb); 1938db4fbfb9SMichel Lespinasse if (prev == vma->vm_rb.rb_left) { 19391be7107fSHugh Dickins gap_start = vm_end_gap(vma->vm_prev); 19401be7107fSHugh Dickins gap_end = vm_start_gap(vma); 1941db4fbfb9SMichel Lespinasse goto check_current; 1942db4fbfb9SMichel Lespinasse } 1943db4fbfb9SMichel Lespinasse } 1944db4fbfb9SMichel Lespinasse } 1945db4fbfb9SMichel Lespinasse 1946db4fbfb9SMichel Lespinasse check_highest: 1947db4fbfb9SMichel Lespinasse /* Check highest gap, which does not precede any rbtree node */ 1948db4fbfb9SMichel Lespinasse gap_start = mm->highest_vm_end; 1949db4fbfb9SMichel Lespinasse gap_end = ULONG_MAX; /* Only for VM_BUG_ON below */ 1950db4fbfb9SMichel Lespinasse if (gap_start > high_limit) 1951db4fbfb9SMichel Lespinasse return -ENOMEM; 1952db4fbfb9SMichel Lespinasse 1953db4fbfb9SMichel Lespinasse found: 1954db4fbfb9SMichel Lespinasse /* We found a suitable gap. Clip it with the original low_limit. */ 1955db4fbfb9SMichel Lespinasse if (gap_start < info->low_limit) 1956db4fbfb9SMichel Lespinasse gap_start = info->low_limit; 1957db4fbfb9SMichel Lespinasse 1958db4fbfb9SMichel Lespinasse /* Adjust gap address to the desired alignment */ 1959db4fbfb9SMichel Lespinasse gap_start += (info->align_offset - gap_start) & info->align_mask; 1960db4fbfb9SMichel Lespinasse 1961db4fbfb9SMichel Lespinasse VM_BUG_ON(gap_start + info->length > info->high_limit); 1962db4fbfb9SMichel Lespinasse VM_BUG_ON(gap_start + info->length > gap_end); 1963db4fbfb9SMichel Lespinasse return gap_start; 1964db4fbfb9SMichel Lespinasse } 1965db4fbfb9SMichel Lespinasse 1966baceaf1cSJaewon Kim static unsigned long unmapped_area_topdown(struct vm_unmapped_area_info *info) 1967db4fbfb9SMichel Lespinasse { 1968db4fbfb9SMichel Lespinasse struct mm_struct *mm = current->mm; 1969db4fbfb9SMichel Lespinasse struct vm_area_struct *vma; 1970db4fbfb9SMichel Lespinasse unsigned long length, low_limit, high_limit, gap_start, gap_end; 1971db4fbfb9SMichel Lespinasse 1972db4fbfb9SMichel Lespinasse /* Adjust search length to account for worst case alignment overhead */ 1973db4fbfb9SMichel Lespinasse length = info->length + info->align_mask; 1974db4fbfb9SMichel Lespinasse if (length < info->length) 1975db4fbfb9SMichel Lespinasse return -ENOMEM; 1976db4fbfb9SMichel Lespinasse 1977db4fbfb9SMichel Lespinasse /* 1978db4fbfb9SMichel Lespinasse * Adjust search limits by the desired length. 1979db4fbfb9SMichel Lespinasse * See implementation comment at top of unmapped_area(). 1980db4fbfb9SMichel Lespinasse */ 1981db4fbfb9SMichel Lespinasse gap_end = info->high_limit; 1982db4fbfb9SMichel Lespinasse if (gap_end < length) 1983db4fbfb9SMichel Lespinasse return -ENOMEM; 1984db4fbfb9SMichel Lespinasse high_limit = gap_end - length; 1985db4fbfb9SMichel Lespinasse 1986db4fbfb9SMichel Lespinasse if (info->low_limit > high_limit) 1987db4fbfb9SMichel Lespinasse return -ENOMEM; 1988db4fbfb9SMichel Lespinasse low_limit = info->low_limit + length; 1989db4fbfb9SMichel Lespinasse 1990db4fbfb9SMichel Lespinasse /* Check highest gap, which does not precede any rbtree node */ 1991db4fbfb9SMichel Lespinasse gap_start = mm->highest_vm_end; 1992db4fbfb9SMichel Lespinasse if (gap_start <= high_limit) 1993db4fbfb9SMichel Lespinasse goto found_highest; 1994db4fbfb9SMichel Lespinasse 1995db4fbfb9SMichel Lespinasse /* Check if rbtree root looks promising */ 1996db4fbfb9SMichel Lespinasse if (RB_EMPTY_ROOT(&mm->mm_rb)) 1997db4fbfb9SMichel Lespinasse return -ENOMEM; 1998db4fbfb9SMichel Lespinasse vma = rb_entry(mm->mm_rb.rb_node, struct vm_area_struct, vm_rb); 1999db4fbfb9SMichel Lespinasse if (vma->rb_subtree_gap < length) 2000db4fbfb9SMichel Lespinasse return -ENOMEM; 2001db4fbfb9SMichel Lespinasse 2002db4fbfb9SMichel Lespinasse while (true) { 2003db4fbfb9SMichel Lespinasse /* Visit right subtree if it looks promising */ 20041be7107fSHugh Dickins gap_start = vma->vm_prev ? vm_end_gap(vma->vm_prev) : 0; 2005db4fbfb9SMichel Lespinasse if (gap_start <= high_limit && vma->vm_rb.rb_right) { 2006db4fbfb9SMichel Lespinasse struct vm_area_struct *right = 2007db4fbfb9SMichel Lespinasse rb_entry(vma->vm_rb.rb_right, 2008db4fbfb9SMichel Lespinasse struct vm_area_struct, vm_rb); 2009db4fbfb9SMichel Lespinasse if (right->rb_subtree_gap >= length) { 2010db4fbfb9SMichel Lespinasse vma = right; 2011db4fbfb9SMichel Lespinasse continue; 2012db4fbfb9SMichel Lespinasse } 2013db4fbfb9SMichel Lespinasse } 2014db4fbfb9SMichel Lespinasse 2015db4fbfb9SMichel Lespinasse check_current: 2016db4fbfb9SMichel Lespinasse /* Check if current node has a suitable gap */ 20171be7107fSHugh Dickins gap_end = vm_start_gap(vma); 2018db4fbfb9SMichel Lespinasse if (gap_end < low_limit) 2019db4fbfb9SMichel Lespinasse return -ENOMEM; 2020f4cb767dSHugh Dickins if (gap_start <= high_limit && 2021f4cb767dSHugh Dickins gap_end > gap_start && gap_end - gap_start >= length) 2022db4fbfb9SMichel Lespinasse goto found; 2023db4fbfb9SMichel Lespinasse 2024db4fbfb9SMichel Lespinasse /* Visit left subtree if it looks promising */ 2025db4fbfb9SMichel Lespinasse if (vma->vm_rb.rb_left) { 2026db4fbfb9SMichel Lespinasse struct vm_area_struct *left = 2027db4fbfb9SMichel Lespinasse rb_entry(vma->vm_rb.rb_left, 2028db4fbfb9SMichel Lespinasse struct vm_area_struct, vm_rb); 2029db4fbfb9SMichel Lespinasse if (left->rb_subtree_gap >= length) { 2030db4fbfb9SMichel Lespinasse vma = left; 2031db4fbfb9SMichel Lespinasse continue; 2032db4fbfb9SMichel Lespinasse } 2033db4fbfb9SMichel Lespinasse } 2034db4fbfb9SMichel Lespinasse 2035db4fbfb9SMichel Lespinasse /* Go back up the rbtree to find next candidate node */ 2036db4fbfb9SMichel Lespinasse while (true) { 2037db4fbfb9SMichel Lespinasse struct rb_node *prev = &vma->vm_rb; 2038db4fbfb9SMichel Lespinasse if (!rb_parent(prev)) 2039db4fbfb9SMichel Lespinasse return -ENOMEM; 2040db4fbfb9SMichel Lespinasse vma = rb_entry(rb_parent(prev), 2041db4fbfb9SMichel Lespinasse struct vm_area_struct, vm_rb); 2042db4fbfb9SMichel Lespinasse if (prev == vma->vm_rb.rb_right) { 2043db4fbfb9SMichel Lespinasse gap_start = vma->vm_prev ? 20441be7107fSHugh Dickins vm_end_gap(vma->vm_prev) : 0; 2045db4fbfb9SMichel Lespinasse goto check_current; 2046db4fbfb9SMichel Lespinasse } 2047db4fbfb9SMichel Lespinasse } 2048db4fbfb9SMichel Lespinasse } 2049db4fbfb9SMichel Lespinasse 2050db4fbfb9SMichel Lespinasse found: 2051db4fbfb9SMichel Lespinasse /* We found a suitable gap. Clip it with the original high_limit. */ 2052db4fbfb9SMichel Lespinasse if (gap_end > info->high_limit) 2053db4fbfb9SMichel Lespinasse gap_end = info->high_limit; 2054db4fbfb9SMichel Lespinasse 2055db4fbfb9SMichel Lespinasse found_highest: 2056db4fbfb9SMichel Lespinasse /* Compute highest gap address at the desired alignment */ 2057db4fbfb9SMichel Lespinasse gap_end -= info->length; 2058db4fbfb9SMichel Lespinasse gap_end -= (gap_end - info->align_offset) & info->align_mask; 2059db4fbfb9SMichel Lespinasse 2060db4fbfb9SMichel Lespinasse VM_BUG_ON(gap_end < info->low_limit); 2061db4fbfb9SMichel Lespinasse VM_BUG_ON(gap_end < gap_start); 2062db4fbfb9SMichel Lespinasse return gap_end; 2063db4fbfb9SMichel Lespinasse } 2064db4fbfb9SMichel Lespinasse 2065baceaf1cSJaewon Kim /* 2066baceaf1cSJaewon Kim * Search for an unmapped address range. 2067baceaf1cSJaewon Kim * 2068baceaf1cSJaewon Kim * We are looking for a range that: 2069baceaf1cSJaewon Kim * - does not intersect with any VMA; 2070baceaf1cSJaewon Kim * - is contained within the [low_limit, high_limit) interval; 2071baceaf1cSJaewon Kim * - is at least the desired size. 2072baceaf1cSJaewon Kim * - satisfies (begin_addr & align_mask) == (align_offset & align_mask) 2073baceaf1cSJaewon Kim */ 2074baceaf1cSJaewon Kim unsigned long vm_unmapped_area(struct vm_unmapped_area_info *info) 2075baceaf1cSJaewon Kim { 2076df529cabSJaewon Kim unsigned long addr; 2077df529cabSJaewon Kim 2078baceaf1cSJaewon Kim if (info->flags & VM_UNMAPPED_AREA_TOPDOWN) 2079df529cabSJaewon Kim addr = unmapped_area_topdown(info); 2080baceaf1cSJaewon Kim else 2081df529cabSJaewon Kim addr = unmapped_area(info); 2082df529cabSJaewon Kim 2083df529cabSJaewon Kim trace_vm_unmapped_area(addr, info); 2084df529cabSJaewon Kim return addr; 2085baceaf1cSJaewon Kim } 2086f6795053SSteve Capper 20871da177e4SLinus Torvalds /* Get an address range which is currently unmapped. 20881da177e4SLinus Torvalds * For shmat() with addr=0. 20891da177e4SLinus Torvalds * 20901da177e4SLinus Torvalds * Ugly calling convention alert: 20911da177e4SLinus Torvalds * Return value with the low bits set means error value, 20921da177e4SLinus Torvalds * ie 20931da177e4SLinus Torvalds * if (ret & ~PAGE_MASK) 20941da177e4SLinus Torvalds * error = ret; 20951da177e4SLinus Torvalds * 20961da177e4SLinus Torvalds * This function "knows" that -ENOMEM has the bits set. 20971da177e4SLinus Torvalds */ 20981da177e4SLinus Torvalds unsigned long 20994b439e25SChristophe Leroy generic_get_unmapped_area(struct file *filp, unsigned long addr, 21004b439e25SChristophe Leroy unsigned long len, unsigned long pgoff, 21014b439e25SChristophe Leroy unsigned long flags) 21021da177e4SLinus Torvalds { 21031da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 21041be7107fSHugh Dickins struct vm_area_struct *vma, *prev; 2105db4fbfb9SMichel Lespinasse struct vm_unmapped_area_info info; 21062cb4de08SChristophe Leroy const unsigned long mmap_end = arch_get_mmap_end(addr, len, flags); 21071da177e4SLinus Torvalds 2108f6795053SSteve Capper if (len > mmap_end - mmap_min_addr) 21091da177e4SLinus Torvalds return -ENOMEM; 21101da177e4SLinus Torvalds 211106abdfb4SBenjamin Herrenschmidt if (flags & MAP_FIXED) 211206abdfb4SBenjamin Herrenschmidt return addr; 211306abdfb4SBenjamin Herrenschmidt 21141da177e4SLinus Torvalds if (addr) { 21151da177e4SLinus Torvalds addr = PAGE_ALIGN(addr); 21161be7107fSHugh Dickins vma = find_vma_prev(mm, addr, &prev); 2117f6795053SSteve Capper if (mmap_end - len >= addr && addr >= mmap_min_addr && 21181be7107fSHugh Dickins (!vma || addr + len <= vm_start_gap(vma)) && 21191be7107fSHugh Dickins (!prev || addr >= vm_end_gap(prev))) 21201da177e4SLinus Torvalds return addr; 21211da177e4SLinus Torvalds } 21221da177e4SLinus Torvalds 2123db4fbfb9SMichel Lespinasse info.flags = 0; 2124db4fbfb9SMichel Lespinasse info.length = len; 21254e99b021SHeiko Carstens info.low_limit = mm->mmap_base; 2126f6795053SSteve Capper info.high_limit = mmap_end; 2127db4fbfb9SMichel Lespinasse info.align_mask = 0; 212809ef5283SJaewon Kim info.align_offset = 0; 2129db4fbfb9SMichel Lespinasse return vm_unmapped_area(&info); 21301da177e4SLinus Torvalds } 21314b439e25SChristophe Leroy 21324b439e25SChristophe Leroy #ifndef HAVE_ARCH_UNMAPPED_AREA 21334b439e25SChristophe Leroy unsigned long 21344b439e25SChristophe Leroy arch_get_unmapped_area(struct file *filp, unsigned long addr, 21354b439e25SChristophe Leroy unsigned long len, unsigned long pgoff, 21364b439e25SChristophe Leroy unsigned long flags) 21374b439e25SChristophe Leroy { 21384b439e25SChristophe Leroy return generic_get_unmapped_area(filp, addr, len, pgoff, flags); 21394b439e25SChristophe Leroy } 21401da177e4SLinus Torvalds #endif 21411da177e4SLinus Torvalds 21421da177e4SLinus Torvalds /* 21431da177e4SLinus Torvalds * This mmap-allocator allocates new areas top-down from below the 21441da177e4SLinus Torvalds * stack's low limit (the base): 21451da177e4SLinus Torvalds */ 21461da177e4SLinus Torvalds unsigned long 21474b439e25SChristophe Leroy generic_get_unmapped_area_topdown(struct file *filp, unsigned long addr, 214843cca0b1SYang Fan unsigned long len, unsigned long pgoff, 214943cca0b1SYang Fan unsigned long flags) 21501da177e4SLinus Torvalds { 21511be7107fSHugh Dickins struct vm_area_struct *vma, *prev; 21521da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 2153db4fbfb9SMichel Lespinasse struct vm_unmapped_area_info info; 21542cb4de08SChristophe Leroy const unsigned long mmap_end = arch_get_mmap_end(addr, len, flags); 21551da177e4SLinus Torvalds 21561da177e4SLinus Torvalds /* requested length too big for entire address space */ 2157f6795053SSteve Capper if (len > mmap_end - mmap_min_addr) 21581da177e4SLinus Torvalds return -ENOMEM; 21591da177e4SLinus Torvalds 216006abdfb4SBenjamin Herrenschmidt if (flags & MAP_FIXED) 216106abdfb4SBenjamin Herrenschmidt return addr; 216206abdfb4SBenjamin Herrenschmidt 21631da177e4SLinus Torvalds /* requesting a specific address */ 21641da177e4SLinus Torvalds if (addr) { 21651da177e4SLinus Torvalds addr = PAGE_ALIGN(addr); 21661be7107fSHugh Dickins vma = find_vma_prev(mm, addr, &prev); 2167f6795053SSteve Capper if (mmap_end - len >= addr && addr >= mmap_min_addr && 21681be7107fSHugh Dickins (!vma || addr + len <= vm_start_gap(vma)) && 21691be7107fSHugh Dickins (!prev || addr >= vm_end_gap(prev))) 21701da177e4SLinus Torvalds return addr; 21711da177e4SLinus Torvalds } 21721da177e4SLinus Torvalds 2173db4fbfb9SMichel Lespinasse info.flags = VM_UNMAPPED_AREA_TOPDOWN; 2174db4fbfb9SMichel Lespinasse info.length = len; 21752afc745fSAkira Takeuchi info.low_limit = max(PAGE_SIZE, mmap_min_addr); 2176f6795053SSteve Capper info.high_limit = arch_get_mmap_base(addr, mm->mmap_base); 2177db4fbfb9SMichel Lespinasse info.align_mask = 0; 217809ef5283SJaewon Kim info.align_offset = 0; 2179db4fbfb9SMichel Lespinasse addr = vm_unmapped_area(&info); 2180b716ad95SXiao Guangrong 21811da177e4SLinus Torvalds /* 21821da177e4SLinus Torvalds * A failed mmap() very likely causes application failure, 21831da177e4SLinus Torvalds * so fall back to the bottom-up function here. This scenario 21841da177e4SLinus Torvalds * can happen with large stack limits and large mmap() 21851da177e4SLinus Torvalds * allocations. 21861da177e4SLinus Torvalds */ 2187de1741a1SAlexander Kuleshov if (offset_in_page(addr)) { 2188db4fbfb9SMichel Lespinasse VM_BUG_ON(addr != -ENOMEM); 2189db4fbfb9SMichel Lespinasse info.flags = 0; 2190db4fbfb9SMichel Lespinasse info.low_limit = TASK_UNMAPPED_BASE; 2191f6795053SSteve Capper info.high_limit = mmap_end; 2192db4fbfb9SMichel Lespinasse addr = vm_unmapped_area(&info); 2193db4fbfb9SMichel Lespinasse } 21941da177e4SLinus Torvalds 21951da177e4SLinus Torvalds return addr; 21961da177e4SLinus Torvalds } 21974b439e25SChristophe Leroy 21984b439e25SChristophe Leroy #ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN 21994b439e25SChristophe Leroy unsigned long 22004b439e25SChristophe Leroy arch_get_unmapped_area_topdown(struct file *filp, unsigned long addr, 22014b439e25SChristophe Leroy unsigned long len, unsigned long pgoff, 22024b439e25SChristophe Leroy unsigned long flags) 22034b439e25SChristophe Leroy { 22044b439e25SChristophe Leroy return generic_get_unmapped_area_topdown(filp, addr, len, pgoff, flags); 22054b439e25SChristophe Leroy } 22061da177e4SLinus Torvalds #endif 22071da177e4SLinus Torvalds 22081da177e4SLinus Torvalds unsigned long 22091da177e4SLinus Torvalds get_unmapped_area(struct file *file, unsigned long addr, unsigned long len, 22101da177e4SLinus Torvalds unsigned long pgoff, unsigned long flags) 22111da177e4SLinus Torvalds { 221206abdfb4SBenjamin Herrenschmidt unsigned long (*get_area)(struct file *, unsigned long, 221306abdfb4SBenjamin Herrenschmidt unsigned long, unsigned long, unsigned long); 221407ab67c8SLinus Torvalds 22159206de95SAl Viro unsigned long error = arch_mmap_check(addr, len, flags); 22169206de95SAl Viro if (error) 22179206de95SAl Viro return error; 22189206de95SAl Viro 22199206de95SAl Viro /* Careful about overflows.. */ 22209206de95SAl Viro if (len > TASK_SIZE) 22219206de95SAl Viro return -ENOMEM; 22229206de95SAl Viro 222307ab67c8SLinus Torvalds get_area = current->mm->get_unmapped_area; 2224c01d5b30SHugh Dickins if (file) { 2225c01d5b30SHugh Dickins if (file->f_op->get_unmapped_area) 222607ab67c8SLinus Torvalds get_area = file->f_op->get_unmapped_area; 2227c01d5b30SHugh Dickins } else if (flags & MAP_SHARED) { 2228c01d5b30SHugh Dickins /* 2229c01d5b30SHugh Dickins * mmap_region() will call shmem_zero_setup() to create a file, 2230c01d5b30SHugh Dickins * so use shmem's get_unmapped_area in case it can be huge. 223145e55300SPeter Collingbourne * do_mmap() will clear pgoff, so match alignment. 2232c01d5b30SHugh Dickins */ 2233c01d5b30SHugh Dickins pgoff = 0; 2234c01d5b30SHugh Dickins get_area = shmem_get_unmapped_area; 2235c01d5b30SHugh Dickins } 2236c01d5b30SHugh Dickins 223707ab67c8SLinus Torvalds addr = get_area(file, addr, len, pgoff, flags); 223807ab67c8SLinus Torvalds if (IS_ERR_VALUE(addr)) 223907ab67c8SLinus Torvalds return addr; 224007ab67c8SLinus Torvalds 22411da177e4SLinus Torvalds if (addr > TASK_SIZE - len) 22421da177e4SLinus Torvalds return -ENOMEM; 2243de1741a1SAlexander Kuleshov if (offset_in_page(addr)) 22441da177e4SLinus Torvalds return -EINVAL; 224506abdfb4SBenjamin Herrenschmidt 22469ac4ed4bSAl Viro error = security_mmap_addr(addr); 22479ac4ed4bSAl Viro return error ? error : addr; 22481da177e4SLinus Torvalds } 22491da177e4SLinus Torvalds 22501da177e4SLinus Torvalds EXPORT_SYMBOL(get_unmapped_area); 22511da177e4SLinus Torvalds 22521da177e4SLinus Torvalds /* Look up the first VMA which satisfies addr < vm_end, NULL if none. */ 22531da177e4SLinus Torvalds struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr) 22541da177e4SLinus Torvalds { 2255615d6e87SDavidlohr Bueso struct rb_node *rb_node; 2256615d6e87SDavidlohr Bueso struct vm_area_struct *vma; 22571da177e4SLinus Torvalds 22585b78ed24SLuigi Rizzo mmap_assert_locked(mm); 22591da177e4SLinus Torvalds /* Check the cache first. */ 2260615d6e87SDavidlohr Bueso vma = vmacache_find(mm, addr); 2261615d6e87SDavidlohr Bueso if (likely(vma)) 2262615d6e87SDavidlohr Bueso return vma; 22631da177e4SLinus Torvalds 22641da177e4SLinus Torvalds rb_node = mm->mm_rb.rb_node; 22651da177e4SLinus Torvalds 22661da177e4SLinus Torvalds while (rb_node) { 2267615d6e87SDavidlohr Bueso struct vm_area_struct *tmp; 22681da177e4SLinus Torvalds 2269615d6e87SDavidlohr Bueso tmp = rb_entry(rb_node, struct vm_area_struct, vm_rb); 22701da177e4SLinus Torvalds 2271615d6e87SDavidlohr Bueso if (tmp->vm_end > addr) { 2272615d6e87SDavidlohr Bueso vma = tmp; 2273615d6e87SDavidlohr Bueso if (tmp->vm_start <= addr) 22741da177e4SLinus Torvalds break; 22751da177e4SLinus Torvalds rb_node = rb_node->rb_left; 22761da177e4SLinus Torvalds } else 22771da177e4SLinus Torvalds rb_node = rb_node->rb_right; 22781da177e4SLinus Torvalds } 2279615d6e87SDavidlohr Bueso 22801da177e4SLinus Torvalds if (vma) 2281615d6e87SDavidlohr Bueso vmacache_update(addr, vma); 22821da177e4SLinus Torvalds return vma; 22831da177e4SLinus Torvalds } 22841da177e4SLinus Torvalds 22851da177e4SLinus Torvalds EXPORT_SYMBOL(find_vma); 22861da177e4SLinus Torvalds 22876bd4837dSKOSAKI Motohiro /* 22886bd4837dSKOSAKI Motohiro * Same as find_vma, but also return a pointer to the previous VMA in *pprev. 22896bd4837dSKOSAKI Motohiro */ 22901da177e4SLinus Torvalds struct vm_area_struct * 22911da177e4SLinus Torvalds find_vma_prev(struct mm_struct *mm, unsigned long addr, 22921da177e4SLinus Torvalds struct vm_area_struct **pprev) 22931da177e4SLinus Torvalds { 22946bd4837dSKOSAKI Motohiro struct vm_area_struct *vma; 22951da177e4SLinus Torvalds 22966bd4837dSKOSAKI Motohiro vma = find_vma(mm, addr); 229783cd904dSMikulas Patocka if (vma) { 229883cd904dSMikulas Patocka *pprev = vma->vm_prev; 229983cd904dSMikulas Patocka } else { 230073848a97SWei Yang struct rb_node *rb_node = rb_last(&mm->mm_rb); 230173848a97SWei Yang 230273848a97SWei Yang *pprev = rb_node ? rb_entry(rb_node, struct vm_area_struct, vm_rb) : NULL; 230383cd904dSMikulas Patocka } 23046bd4837dSKOSAKI Motohiro return vma; 23051da177e4SLinus Torvalds } 23061da177e4SLinus Torvalds 23071da177e4SLinus Torvalds /* 23081da177e4SLinus Torvalds * Verify that the stack growth is acceptable and 23091da177e4SLinus Torvalds * update accounting. This is shared with both the 23101da177e4SLinus Torvalds * grow-up and grow-down cases. 23111da177e4SLinus Torvalds */ 23121be7107fSHugh Dickins static int acct_stack_growth(struct vm_area_struct *vma, 23131be7107fSHugh Dickins unsigned long size, unsigned long grow) 23141da177e4SLinus Torvalds { 23151da177e4SLinus Torvalds struct mm_struct *mm = vma->vm_mm; 23161be7107fSHugh Dickins unsigned long new_start; 23171da177e4SLinus Torvalds 23181da177e4SLinus Torvalds /* address space limit tests */ 231984638335SKonstantin Khlebnikov if (!may_expand_vm(mm, vma->vm_flags, grow)) 23201da177e4SLinus Torvalds return -ENOMEM; 23211da177e4SLinus Torvalds 23221da177e4SLinus Torvalds /* Stack limit test */ 232324c79d8eSKrzysztof Opasiak if (size > rlimit(RLIMIT_STACK)) 23241da177e4SLinus Torvalds return -ENOMEM; 23251da177e4SLinus Torvalds 23261da177e4SLinus Torvalds /* mlock limit tests */ 2327c5d8a364SMiaohe Lin if (mlock_future_check(mm, vma->vm_flags, grow << PAGE_SHIFT)) 23281da177e4SLinus Torvalds return -ENOMEM; 23291da177e4SLinus Torvalds 23300d59a01bSAdam Litke /* Check to ensure the stack will not grow into a hugetlb-only region */ 23310d59a01bSAdam Litke new_start = (vma->vm_flags & VM_GROWSUP) ? vma->vm_start : 23320d59a01bSAdam Litke vma->vm_end - size; 23330d59a01bSAdam Litke if (is_hugepage_only_range(vma->vm_mm, new_start, size)) 23340d59a01bSAdam Litke return -EFAULT; 23350d59a01bSAdam Litke 23361da177e4SLinus Torvalds /* 23371da177e4SLinus Torvalds * Overcommit.. This must be the final test, as it will 23381da177e4SLinus Torvalds * update security statistics. 23391da177e4SLinus Torvalds */ 234005fa199dSHugh Dickins if (security_vm_enough_memory_mm(mm, grow)) 23411da177e4SLinus Torvalds return -ENOMEM; 23421da177e4SLinus Torvalds 23431da177e4SLinus Torvalds return 0; 23441da177e4SLinus Torvalds } 23451da177e4SLinus Torvalds 234646dea3d0SHugh Dickins #if defined(CONFIG_STACK_GROWSUP) || defined(CONFIG_IA64) 23471da177e4SLinus Torvalds /* 234846dea3d0SHugh Dickins * PA-RISC uses this for its stack; IA64 for its Register Backing Store. 234946dea3d0SHugh Dickins * vma is the last one with address > vma->vm_end. Have to extend vma. 23501da177e4SLinus Torvalds */ 235146dea3d0SHugh Dickins int expand_upwards(struct vm_area_struct *vma, unsigned long address) 23521da177e4SLinus Torvalds { 235309357814SOleg Nesterov struct mm_struct *mm = vma->vm_mm; 23541be7107fSHugh Dickins struct vm_area_struct *next; 23551be7107fSHugh Dickins unsigned long gap_addr; 235612352d3cSKonstantin Khlebnikov int error = 0; 23571da177e4SLinus Torvalds 23581da177e4SLinus Torvalds if (!(vma->vm_flags & VM_GROWSUP)) 23591da177e4SLinus Torvalds return -EFAULT; 23601da177e4SLinus Torvalds 2361bd726c90SHelge Deller /* Guard against exceeding limits of the address space. */ 23621be7107fSHugh Dickins address &= PAGE_MASK; 236337511fb5SHelge Deller if (address >= (TASK_SIZE & PAGE_MASK)) 236412352d3cSKonstantin Khlebnikov return -ENOMEM; 2365bd726c90SHelge Deller address += PAGE_SIZE; 236612352d3cSKonstantin Khlebnikov 23671be7107fSHugh Dickins /* Enforce stack_guard_gap */ 23681be7107fSHugh Dickins gap_addr = address + stack_guard_gap; 2369bd726c90SHelge Deller 2370bd726c90SHelge Deller /* Guard against overflow */ 2371bd726c90SHelge Deller if (gap_addr < address || gap_addr > TASK_SIZE) 2372bd726c90SHelge Deller gap_addr = TASK_SIZE; 2373bd726c90SHelge Deller 23741be7107fSHugh Dickins next = vma->vm_next; 23753122e80eSAnshuman Khandual if (next && next->vm_start < gap_addr && vma_is_accessible(next)) { 23761be7107fSHugh Dickins if (!(next->vm_flags & VM_GROWSUP)) 23771be7107fSHugh Dickins return -ENOMEM; 23781be7107fSHugh Dickins /* Check that both stack segments have the same anon_vma? */ 23791be7107fSHugh Dickins } 23801be7107fSHugh Dickins 238112352d3cSKonstantin Khlebnikov /* We must make sure the anon_vma is allocated. */ 23821da177e4SLinus Torvalds if (unlikely(anon_vma_prepare(vma))) 23831da177e4SLinus Torvalds return -ENOMEM; 23841da177e4SLinus Torvalds 23851da177e4SLinus Torvalds /* 23861da177e4SLinus Torvalds * vma->vm_start/vm_end cannot change under us because the caller 2387c1e8d7c6SMichel Lespinasse * is required to hold the mmap_lock in read mode. We need the 23881da177e4SLinus Torvalds * anon_vma lock to serialize against concurrent expand_stacks. 23891da177e4SLinus Torvalds */ 239012352d3cSKonstantin Khlebnikov anon_vma_lock_write(vma->anon_vma); 23911da177e4SLinus Torvalds 23921da177e4SLinus Torvalds /* Somebody else might have raced and expanded it already */ 23931da177e4SLinus Torvalds if (address > vma->vm_end) { 23941da177e4SLinus Torvalds unsigned long size, grow; 23951da177e4SLinus Torvalds 23961da177e4SLinus Torvalds size = address - vma->vm_start; 23971da177e4SLinus Torvalds grow = (address - vma->vm_end) >> PAGE_SHIFT; 23981da177e4SLinus Torvalds 239942c36f63SHugh Dickins error = -ENOMEM; 240042c36f63SHugh Dickins if (vma->vm_pgoff + (size >> PAGE_SHIFT) >= vma->vm_pgoff) { 24011da177e4SLinus Torvalds error = acct_stack_growth(vma, size, grow); 24023af9e859SEric B Munson if (!error) { 24034128997bSMichel Lespinasse /* 24044128997bSMichel Lespinasse * vma_gap_update() doesn't support concurrent 2405c1e8d7c6SMichel Lespinasse * updates, but we only hold a shared mmap_lock 24064128997bSMichel Lespinasse * lock here, so we need to protect against 24074128997bSMichel Lespinasse * concurrent vma expansions. 240812352d3cSKonstantin Khlebnikov * anon_vma_lock_write() doesn't help here, as 24094128997bSMichel Lespinasse * we don't guarantee that all growable vmas 24104128997bSMichel Lespinasse * in a mm share the same root anon vma. 24114128997bSMichel Lespinasse * So, we reuse mm->page_table_lock to guard 24124128997bSMichel Lespinasse * against concurrent vma expansions. 24134128997bSMichel Lespinasse */ 241409357814SOleg Nesterov spin_lock(&mm->page_table_lock); 241587e8827bSOleg Nesterov if (vma->vm_flags & VM_LOCKED) 241609357814SOleg Nesterov mm->locked_vm += grow; 241784638335SKonstantin Khlebnikov vm_stat_account(mm, vma->vm_flags, grow); 2418bf181b9fSMichel Lespinasse anon_vma_interval_tree_pre_update_vma(vma); 24191da177e4SLinus Torvalds vma->vm_end = address; 2420bf181b9fSMichel Lespinasse anon_vma_interval_tree_post_update_vma(vma); 2421d3737187SMichel Lespinasse if (vma->vm_next) 2422d3737187SMichel Lespinasse vma_gap_update(vma->vm_next); 2423d3737187SMichel Lespinasse else 24241be7107fSHugh Dickins mm->highest_vm_end = vm_end_gap(vma); 242509357814SOleg Nesterov spin_unlock(&mm->page_table_lock); 24264128997bSMichel Lespinasse 24273af9e859SEric B Munson perf_event_mmap(vma); 24283af9e859SEric B Munson } 24291da177e4SLinus Torvalds } 243042c36f63SHugh Dickins } 243112352d3cSKonstantin Khlebnikov anon_vma_unlock_write(vma->anon_vma); 2432c791576cSYang Shi khugepaged_enter_vma(vma, vma->vm_flags); 243309357814SOleg Nesterov validate_mm(mm); 24341da177e4SLinus Torvalds return error; 24351da177e4SLinus Torvalds } 243646dea3d0SHugh Dickins #endif /* CONFIG_STACK_GROWSUP || CONFIG_IA64 */ 243746dea3d0SHugh Dickins 24381da177e4SLinus Torvalds /* 24391da177e4SLinus Torvalds * vma is the first one with address < vma->vm_start. Have to extend vma. 24401da177e4SLinus Torvalds */ 2441d05f3169SMichal Hocko int expand_downwards(struct vm_area_struct *vma, 2442b6a2fea3SOllie Wild unsigned long address) 24431da177e4SLinus Torvalds { 244409357814SOleg Nesterov struct mm_struct *mm = vma->vm_mm; 24451be7107fSHugh Dickins struct vm_area_struct *prev; 24460a1d5299SJann Horn int error = 0; 24471da177e4SLinus Torvalds 24488869477aSEric Paris address &= PAGE_MASK; 24490a1d5299SJann Horn if (address < mmap_min_addr) 24500a1d5299SJann Horn return -EPERM; 24518869477aSEric Paris 24521be7107fSHugh Dickins /* Enforce stack_guard_gap */ 24531be7107fSHugh Dickins prev = vma->vm_prev; 24541be7107fSHugh Dickins /* Check that both stack segments have the same anon_vma? */ 245532e4e6d5SOleg Nesterov if (prev && !(prev->vm_flags & VM_GROWSDOWN) && 24563122e80eSAnshuman Khandual vma_is_accessible(prev)) { 245732e4e6d5SOleg Nesterov if (address - prev->vm_end < stack_guard_gap) 245832e4e6d5SOleg Nesterov return -ENOMEM; 24591be7107fSHugh Dickins } 24601be7107fSHugh Dickins 246112352d3cSKonstantin Khlebnikov /* We must make sure the anon_vma is allocated. */ 246212352d3cSKonstantin Khlebnikov if (unlikely(anon_vma_prepare(vma))) 246312352d3cSKonstantin Khlebnikov return -ENOMEM; 24641da177e4SLinus Torvalds 24651da177e4SLinus Torvalds /* 24661da177e4SLinus Torvalds * vma->vm_start/vm_end cannot change under us because the caller 2467c1e8d7c6SMichel Lespinasse * is required to hold the mmap_lock in read mode. We need the 24681da177e4SLinus Torvalds * anon_vma lock to serialize against concurrent expand_stacks. 24691da177e4SLinus Torvalds */ 247012352d3cSKonstantin Khlebnikov anon_vma_lock_write(vma->anon_vma); 24711da177e4SLinus Torvalds 24721da177e4SLinus Torvalds /* Somebody else might have raced and expanded it already */ 24731da177e4SLinus Torvalds if (address < vma->vm_start) { 24741da177e4SLinus Torvalds unsigned long size, grow; 24751da177e4SLinus Torvalds 24761da177e4SLinus Torvalds size = vma->vm_end - address; 24771da177e4SLinus Torvalds grow = (vma->vm_start - address) >> PAGE_SHIFT; 24781da177e4SLinus Torvalds 2479a626ca6aSLinus Torvalds error = -ENOMEM; 2480a626ca6aSLinus Torvalds if (grow <= vma->vm_pgoff) { 24811da177e4SLinus Torvalds error = acct_stack_growth(vma, size, grow); 24821da177e4SLinus Torvalds if (!error) { 24834128997bSMichel Lespinasse /* 24844128997bSMichel Lespinasse * vma_gap_update() doesn't support concurrent 2485c1e8d7c6SMichel Lespinasse * updates, but we only hold a shared mmap_lock 24864128997bSMichel Lespinasse * lock here, so we need to protect against 24874128997bSMichel Lespinasse * concurrent vma expansions. 248812352d3cSKonstantin Khlebnikov * anon_vma_lock_write() doesn't help here, as 24894128997bSMichel Lespinasse * we don't guarantee that all growable vmas 24904128997bSMichel Lespinasse * in a mm share the same root anon vma. 24914128997bSMichel Lespinasse * So, we reuse mm->page_table_lock to guard 24924128997bSMichel Lespinasse * against concurrent vma expansions. 24934128997bSMichel Lespinasse */ 249409357814SOleg Nesterov spin_lock(&mm->page_table_lock); 249587e8827bSOleg Nesterov if (vma->vm_flags & VM_LOCKED) 249609357814SOleg Nesterov mm->locked_vm += grow; 249784638335SKonstantin Khlebnikov vm_stat_account(mm, vma->vm_flags, grow); 2498bf181b9fSMichel Lespinasse anon_vma_interval_tree_pre_update_vma(vma); 24991da177e4SLinus Torvalds vma->vm_start = address; 25001da177e4SLinus Torvalds vma->vm_pgoff -= grow; 2501bf181b9fSMichel Lespinasse anon_vma_interval_tree_post_update_vma(vma); 2502d3737187SMichel Lespinasse vma_gap_update(vma); 250309357814SOleg Nesterov spin_unlock(&mm->page_table_lock); 25044128997bSMichel Lespinasse 25053af9e859SEric B Munson perf_event_mmap(vma); 25061da177e4SLinus Torvalds } 25071da177e4SLinus Torvalds } 2508a626ca6aSLinus Torvalds } 250912352d3cSKonstantin Khlebnikov anon_vma_unlock_write(vma->anon_vma); 2510c791576cSYang Shi khugepaged_enter_vma(vma, vma->vm_flags); 251109357814SOleg Nesterov validate_mm(mm); 25121da177e4SLinus Torvalds return error; 25131da177e4SLinus Torvalds } 25141da177e4SLinus Torvalds 25151be7107fSHugh Dickins /* enforced gap between the expanding stack and other mappings. */ 25161be7107fSHugh Dickins unsigned long stack_guard_gap = 256UL<<PAGE_SHIFT; 25171be7107fSHugh Dickins 25181be7107fSHugh Dickins static int __init cmdline_parse_stack_guard_gap(char *p) 25191be7107fSHugh Dickins { 25201be7107fSHugh Dickins unsigned long val; 25211be7107fSHugh Dickins char *endptr; 25221be7107fSHugh Dickins 25231be7107fSHugh Dickins val = simple_strtoul(p, &endptr, 10); 25241be7107fSHugh Dickins if (!*endptr) 25251be7107fSHugh Dickins stack_guard_gap = val << PAGE_SHIFT; 25261be7107fSHugh Dickins 2527e6d09493SRandy Dunlap return 1; 25281be7107fSHugh Dickins } 25291be7107fSHugh Dickins __setup("stack_guard_gap=", cmdline_parse_stack_guard_gap); 25301be7107fSHugh Dickins 2531b6a2fea3SOllie Wild #ifdef CONFIG_STACK_GROWSUP 2532b6a2fea3SOllie Wild int expand_stack(struct vm_area_struct *vma, unsigned long address) 2533b6a2fea3SOllie Wild { 2534b6a2fea3SOllie Wild return expand_upwards(vma, address); 2535b6a2fea3SOllie Wild } 2536b6a2fea3SOllie Wild 2537b6a2fea3SOllie Wild struct vm_area_struct * 2538b6a2fea3SOllie Wild find_extend_vma(struct mm_struct *mm, unsigned long addr) 2539b6a2fea3SOllie Wild { 2540b6a2fea3SOllie Wild struct vm_area_struct *vma, *prev; 2541b6a2fea3SOllie Wild 2542b6a2fea3SOllie Wild addr &= PAGE_MASK; 2543b6a2fea3SOllie Wild vma = find_vma_prev(mm, addr, &prev); 2544b6a2fea3SOllie Wild if (vma && (vma->vm_start <= addr)) 2545b6a2fea3SOllie Wild return vma; 25464d45e75aSJann Horn if (!prev || expand_stack(prev, addr)) 2547b6a2fea3SOllie Wild return NULL; 2548cea10a19SMichel Lespinasse if (prev->vm_flags & VM_LOCKED) 2549fc05f566SKirill A. Shutemov populate_vma_page_range(prev, addr, prev->vm_end, NULL); 2550b6a2fea3SOllie Wild return prev; 2551b6a2fea3SOllie Wild } 2552b6a2fea3SOllie Wild #else 2553b6a2fea3SOllie Wild int expand_stack(struct vm_area_struct *vma, unsigned long address) 2554b6a2fea3SOllie Wild { 2555b6a2fea3SOllie Wild return expand_downwards(vma, address); 2556b6a2fea3SOllie Wild } 2557b6a2fea3SOllie Wild 25581da177e4SLinus Torvalds struct vm_area_struct * 25591da177e4SLinus Torvalds find_extend_vma(struct mm_struct *mm, unsigned long addr) 25601da177e4SLinus Torvalds { 25611da177e4SLinus Torvalds struct vm_area_struct *vma; 25621da177e4SLinus Torvalds unsigned long start; 25631da177e4SLinus Torvalds 25641da177e4SLinus Torvalds addr &= PAGE_MASK; 25651da177e4SLinus Torvalds vma = find_vma(mm, addr); 25661da177e4SLinus Torvalds if (!vma) 25671da177e4SLinus Torvalds return NULL; 25681da177e4SLinus Torvalds if (vma->vm_start <= addr) 25691da177e4SLinus Torvalds return vma; 25701da177e4SLinus Torvalds if (!(vma->vm_flags & VM_GROWSDOWN)) 25711da177e4SLinus Torvalds return NULL; 25721da177e4SLinus Torvalds start = vma->vm_start; 25731da177e4SLinus Torvalds if (expand_stack(vma, addr)) 25741da177e4SLinus Torvalds return NULL; 2575cea10a19SMichel Lespinasse if (vma->vm_flags & VM_LOCKED) 2576fc05f566SKirill A. Shutemov populate_vma_page_range(vma, addr, start, NULL); 25771da177e4SLinus Torvalds return vma; 25781da177e4SLinus Torvalds } 25791da177e4SLinus Torvalds #endif 25801da177e4SLinus Torvalds 2581e1d6d01aSJesse Barnes EXPORT_SYMBOL_GPL(find_extend_vma); 2582e1d6d01aSJesse Barnes 25832c0b3814SHugh Dickins /* 25842c0b3814SHugh Dickins * Ok - we have the memory areas we should free on the vma list, 25852c0b3814SHugh Dickins * so release them, and do the vma updates. 25861da177e4SLinus Torvalds * 25872c0b3814SHugh Dickins * Called with the mm semaphore held. 25881da177e4SLinus Torvalds */ 25892c0b3814SHugh Dickins static void remove_vma_list(struct mm_struct *mm, struct vm_area_struct *vma) 25901da177e4SLinus Torvalds { 25914f74d2c8SLinus Torvalds unsigned long nr_accounted = 0; 25924f74d2c8SLinus Torvalds 2593365e9c87SHugh Dickins /* Update high watermark before we lower total_vm */ 2594365e9c87SHugh Dickins update_hiwater_vm(mm); 25952c0b3814SHugh Dickins do { 2596ab50b8edSHugh Dickins long nrpages = vma_pages(vma); 25971da177e4SLinus Torvalds 25984f74d2c8SLinus Torvalds if (vma->vm_flags & VM_ACCOUNT) 25994f74d2c8SLinus Torvalds nr_accounted += nrpages; 260084638335SKonstantin Khlebnikov vm_stat_account(mm, vma->vm_flags, -nrpages); 2601a8fb5618SHugh Dickins vma = remove_vma(vma); 2602146425a3SHugh Dickins } while (vma); 26034f74d2c8SLinus Torvalds vm_unacct_memory(nr_accounted); 26041da177e4SLinus Torvalds validate_mm(mm); 26051da177e4SLinus Torvalds } 26061da177e4SLinus Torvalds 26071da177e4SLinus Torvalds /* 26081da177e4SLinus Torvalds * Get rid of page table information in the indicated region. 26091da177e4SLinus Torvalds * 2610f10df686SPaolo 'Blaisorblade' Giarrusso * Called with the mm semaphore held. 26111da177e4SLinus Torvalds */ 26121da177e4SLinus Torvalds static void unmap_region(struct mm_struct *mm, 2613e0da382cSHugh Dickins struct vm_area_struct *vma, struct vm_area_struct *prev, 2614e0da382cSHugh Dickins unsigned long start, unsigned long end) 26151da177e4SLinus Torvalds { 26163903b55aSLiam R. Howlett struct vm_area_struct *next = vma_next(mm, prev); 2617d16dfc55SPeter Zijlstra struct mmu_gather tlb; 26181da177e4SLinus Torvalds 26191da177e4SLinus Torvalds lru_add_drain(); 2620a72afd87SWill Deacon tlb_gather_mmu(&tlb, mm); 2621365e9c87SHugh Dickins update_hiwater_rss(mm); 26224f74d2c8SLinus Torvalds unmap_vmas(&tlb, vma, start, end); 2623d16dfc55SPeter Zijlstra free_pgtables(&tlb, vma, prev ? prev->vm_end : FIRST_USER_ADDRESS, 26246ee8630eSHugh Dickins next ? next->vm_start : USER_PGTABLES_CEILING); 2625ae8eba8bSWill Deacon tlb_finish_mmu(&tlb); 26261da177e4SLinus Torvalds } 26271da177e4SLinus Torvalds 26281da177e4SLinus Torvalds /* 26291da177e4SLinus Torvalds * Create a list of vma's touched by the unmap, removing them from the mm's 26301da177e4SLinus Torvalds * vma list as we go.. 26311da177e4SLinus Torvalds */ 2632246c320aSKirill A. Shutemov static bool 26331da177e4SLinus Torvalds detach_vmas_to_be_unmapped(struct mm_struct *mm, struct vm_area_struct *vma, 26341da177e4SLinus Torvalds struct vm_area_struct *prev, unsigned long end) 26351da177e4SLinus Torvalds { 26361da177e4SLinus Torvalds struct vm_area_struct **insertion_point; 26371da177e4SLinus Torvalds struct vm_area_struct *tail_vma = NULL; 26381da177e4SLinus Torvalds 26391da177e4SLinus Torvalds insertion_point = (prev ? &prev->vm_next : &mm->mmap); 2640297c5eeeSLinus Torvalds vma->vm_prev = NULL; 26411da177e4SLinus Torvalds do { 2642d3737187SMichel Lespinasse vma_rb_erase(vma, &mm->mm_rb); 2643a213e5cfSHugh Dickins if (vma->vm_flags & VM_LOCKED) 2644a213e5cfSHugh Dickins mm->locked_vm -= vma_pages(vma); 26451da177e4SLinus Torvalds mm->map_count--; 26461da177e4SLinus Torvalds tail_vma = vma; 26471da177e4SLinus Torvalds vma = vma->vm_next; 26481da177e4SLinus Torvalds } while (vma && vma->vm_start < end); 26491da177e4SLinus Torvalds *insertion_point = vma; 2650d3737187SMichel Lespinasse if (vma) { 2651297c5eeeSLinus Torvalds vma->vm_prev = prev; 2652d3737187SMichel Lespinasse vma_gap_update(vma); 2653d3737187SMichel Lespinasse } else 26541be7107fSHugh Dickins mm->highest_vm_end = prev ? vm_end_gap(prev) : 0; 26551da177e4SLinus Torvalds tail_vma->vm_next = NULL; 2656615d6e87SDavidlohr Bueso 2657615d6e87SDavidlohr Bueso /* Kill the cache */ 2658615d6e87SDavidlohr Bueso vmacache_invalidate(mm); 2659246c320aSKirill A. Shutemov 2660246c320aSKirill A. Shutemov /* 2661246c320aSKirill A. Shutemov * Do not downgrade mmap_lock if we are next to VM_GROWSDOWN or 2662246c320aSKirill A. Shutemov * VM_GROWSUP VMA. Such VMAs can change their size under 2663246c320aSKirill A. Shutemov * down_read(mmap_lock) and collide with the VMA we are about to unmap. 2664246c320aSKirill A. Shutemov */ 2665246c320aSKirill A. Shutemov if (vma && (vma->vm_flags & VM_GROWSDOWN)) 2666246c320aSKirill A. Shutemov return false; 2667246c320aSKirill A. Shutemov if (prev && (prev->vm_flags & VM_GROWSUP)) 2668246c320aSKirill A. Shutemov return false; 2669246c320aSKirill A. Shutemov return true; 26701da177e4SLinus Torvalds } 26711da177e4SLinus Torvalds 26721da177e4SLinus Torvalds /* 2673def5efe0SDavid Rientjes * __split_vma() bypasses sysctl_max_map_count checking. We use this where it 2674def5efe0SDavid Rientjes * has already been checked or doesn't make sense to fail. 26751da177e4SLinus Torvalds */ 2676def5efe0SDavid Rientjes int __split_vma(struct mm_struct *mm, struct vm_area_struct *vma, 26771da177e4SLinus Torvalds unsigned long addr, int new_below) 26781da177e4SLinus Torvalds { 26791da177e4SLinus Torvalds struct vm_area_struct *new; 2680e3975891SChen Gang int err; 26811da177e4SLinus Torvalds 2682dd3b614fSDmitry Safonov if (vma->vm_ops && vma->vm_ops->may_split) { 2683dd3b614fSDmitry Safonov err = vma->vm_ops->may_split(vma, addr); 268431383c68SDan Williams if (err) 268531383c68SDan Williams return err; 268631383c68SDan Williams } 26871da177e4SLinus Torvalds 26883928d4f5SLinus Torvalds new = vm_area_dup(vma); 26891da177e4SLinus Torvalds if (!new) 2690e3975891SChen Gang return -ENOMEM; 26911da177e4SLinus Torvalds 26921da177e4SLinus Torvalds if (new_below) 26931da177e4SLinus Torvalds new->vm_end = addr; 26941da177e4SLinus Torvalds else { 26951da177e4SLinus Torvalds new->vm_start = addr; 26961da177e4SLinus Torvalds new->vm_pgoff += ((addr - vma->vm_start) >> PAGE_SHIFT); 26971da177e4SLinus Torvalds } 26981da177e4SLinus Torvalds 2699ef0855d3SOleg Nesterov err = vma_dup_policy(vma, new); 2700ef0855d3SOleg Nesterov if (err) 27015beb4930SRik van Riel goto out_free_vma; 27021da177e4SLinus Torvalds 2703c4ea95d7SDaniel Forrest err = anon_vma_clone(new, vma); 2704c4ea95d7SDaniel Forrest if (err) 27055beb4930SRik van Riel goto out_free_mpol; 27065beb4930SRik van Riel 2707e9714acfSKonstantin Khlebnikov if (new->vm_file) 27081da177e4SLinus Torvalds get_file(new->vm_file); 27091da177e4SLinus Torvalds 27101da177e4SLinus Torvalds if (new->vm_ops && new->vm_ops->open) 27111da177e4SLinus Torvalds new->vm_ops->open(new); 27121da177e4SLinus Torvalds 27131da177e4SLinus Torvalds if (new_below) 27145beb4930SRik van Riel err = vma_adjust(vma, addr, vma->vm_end, vma->vm_pgoff + 27151da177e4SLinus Torvalds ((addr - new->vm_start) >> PAGE_SHIFT), new); 27161da177e4SLinus Torvalds else 27175beb4930SRik van Riel err = vma_adjust(vma, vma->vm_start, addr, vma->vm_pgoff, new); 27181da177e4SLinus Torvalds 27195beb4930SRik van Riel /* Success. */ 27205beb4930SRik van Riel if (!err) 27211da177e4SLinus Torvalds return 0; 27225beb4930SRik van Riel 27235beb4930SRik van Riel /* Clean everything up if vma_adjust failed. */ 272458927533SRik van Riel if (new->vm_ops && new->vm_ops->close) 27255beb4930SRik van Riel new->vm_ops->close(new); 2726e9714acfSKonstantin Khlebnikov if (new->vm_file) 27275beb4930SRik van Riel fput(new->vm_file); 27282aeadc30SAndrea Arcangeli unlink_anon_vmas(new); 27295beb4930SRik van Riel out_free_mpol: 2730ef0855d3SOleg Nesterov mpol_put(vma_policy(new)); 27315beb4930SRik van Riel out_free_vma: 27323928d4f5SLinus Torvalds vm_area_free(new); 27335beb4930SRik van Riel return err; 27341da177e4SLinus Torvalds } 27351da177e4SLinus Torvalds 2736659ace58SKOSAKI Motohiro /* 2737659ace58SKOSAKI Motohiro * Split a vma into two pieces at address 'addr', a new vma is allocated 2738659ace58SKOSAKI Motohiro * either for the first part or the tail. 2739659ace58SKOSAKI Motohiro */ 2740659ace58SKOSAKI Motohiro int split_vma(struct mm_struct *mm, struct vm_area_struct *vma, 2741659ace58SKOSAKI Motohiro unsigned long addr, int new_below) 2742659ace58SKOSAKI Motohiro { 2743659ace58SKOSAKI Motohiro if (mm->map_count >= sysctl_max_map_count) 2744659ace58SKOSAKI Motohiro return -ENOMEM; 2745659ace58SKOSAKI Motohiro 2746659ace58SKOSAKI Motohiro return __split_vma(mm, vma, addr, new_below); 2747659ace58SKOSAKI Motohiro } 2748659ace58SKOSAKI Motohiro 27491da177e4SLinus Torvalds /* Munmap is split into 2 main parts -- this part which finds 27501da177e4SLinus Torvalds * what needs doing, and the areas themselves, which do the 27511da177e4SLinus Torvalds * work. This now handles partial unmappings. 27521da177e4SLinus Torvalds * Jeremy Fitzhardinge <jeremy@goop.org> 27531da177e4SLinus Torvalds */ 275485a06835SYang Shi int __do_munmap(struct mm_struct *mm, unsigned long start, size_t len, 2755dd2283f2SYang Shi struct list_head *uf, bool downgrade) 27561da177e4SLinus Torvalds { 27571da177e4SLinus Torvalds unsigned long end; 2758146425a3SHugh Dickins struct vm_area_struct *vma, *prev, *last; 27591da177e4SLinus Torvalds 2760de1741a1SAlexander Kuleshov if ((offset_in_page(start)) || start > TASK_SIZE || len > TASK_SIZE-start) 27611da177e4SLinus Torvalds return -EINVAL; 27621da177e4SLinus Torvalds 2763cc71aba3Svishnu.ps len = PAGE_ALIGN(len); 27645a28fc94SDave Hansen end = start + len; 2765cc71aba3Svishnu.ps if (len == 0) 27661da177e4SLinus Torvalds return -EINVAL; 27671da177e4SLinus Torvalds 27685a28fc94SDave Hansen /* 27695a28fc94SDave Hansen * arch_unmap() might do unmaps itself. It must be called 27705a28fc94SDave Hansen * and finish any rbtree manipulation before this code 27715a28fc94SDave Hansen * runs and also starts to manipulate the rbtree. 27725a28fc94SDave Hansen */ 27735a28fc94SDave Hansen arch_unmap(mm, start, end); 27745a28fc94SDave Hansen 277578d9cf60SGonzalo Matias Juarez Tello /* Find the first overlapping VMA where start < vma->vm_end */ 277678d9cf60SGonzalo Matias Juarez Tello vma = find_vma_intersection(mm, start, end); 2777146425a3SHugh Dickins if (!vma) 27781da177e4SLinus Torvalds return 0; 27799be34c9dSLinus Torvalds prev = vma->vm_prev; 27801da177e4SLinus Torvalds 27811da177e4SLinus Torvalds /* 27821da177e4SLinus Torvalds * If we need to split any vma, do it now to save pain later. 27831da177e4SLinus Torvalds * 27841da177e4SLinus Torvalds * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially 27851da177e4SLinus Torvalds * unmapped vm_area_struct will remain in use: so lower split_vma 27861da177e4SLinus Torvalds * places tmp vma above, and higher split_vma places tmp vma below. 27871da177e4SLinus Torvalds */ 2788146425a3SHugh Dickins if (start > vma->vm_start) { 2789659ace58SKOSAKI Motohiro int error; 2790659ace58SKOSAKI Motohiro 2791659ace58SKOSAKI Motohiro /* 2792659ace58SKOSAKI Motohiro * Make sure that map_count on return from munmap() will 2793659ace58SKOSAKI Motohiro * not exceed its limit; but let map_count go just above 2794659ace58SKOSAKI Motohiro * its limit temporarily, to help free resources as expected. 2795659ace58SKOSAKI Motohiro */ 2796659ace58SKOSAKI Motohiro if (end < vma->vm_end && mm->map_count >= sysctl_max_map_count) 2797659ace58SKOSAKI Motohiro return -ENOMEM; 2798659ace58SKOSAKI Motohiro 2799659ace58SKOSAKI Motohiro error = __split_vma(mm, vma, start, 0); 28001da177e4SLinus Torvalds if (error) 28011da177e4SLinus Torvalds return error; 2802146425a3SHugh Dickins prev = vma; 28031da177e4SLinus Torvalds } 28041da177e4SLinus Torvalds 28051da177e4SLinus Torvalds /* Does it split the last one? */ 28061da177e4SLinus Torvalds last = find_vma(mm, end); 28071da177e4SLinus Torvalds if (last && end > last->vm_start) { 2808659ace58SKOSAKI Motohiro int error = __split_vma(mm, last, end, 1); 28091da177e4SLinus Torvalds if (error) 28101da177e4SLinus Torvalds return error; 28111da177e4SLinus Torvalds } 28123903b55aSLiam R. Howlett vma = vma_next(mm, prev); 28131da177e4SLinus Torvalds 28142376dd7cSAndrea Arcangeli if (unlikely(uf)) { 28152376dd7cSAndrea Arcangeli /* 28162376dd7cSAndrea Arcangeli * If userfaultfd_unmap_prep returns an error the vmas 2817f0953a1bSIngo Molnar * will remain split, but userland will get a 28182376dd7cSAndrea Arcangeli * highly unexpected error anyway. This is no 28192376dd7cSAndrea Arcangeli * different than the case where the first of the two 28202376dd7cSAndrea Arcangeli * __split_vma fails, but we don't undo the first 28212376dd7cSAndrea Arcangeli * split, despite we could. This is unlikely enough 28222376dd7cSAndrea Arcangeli * failure that it's not worth optimizing it for. 28232376dd7cSAndrea Arcangeli */ 28242376dd7cSAndrea Arcangeli int error = userfaultfd_unmap_prep(vma, start, end, uf); 28252376dd7cSAndrea Arcangeli if (error) 28262376dd7cSAndrea Arcangeli return error; 28272376dd7cSAndrea Arcangeli } 28282376dd7cSAndrea Arcangeli 2829dd2283f2SYang Shi /* Detach vmas from rbtree */ 2830246c320aSKirill A. Shutemov if (!detach_vmas_to_be_unmapped(mm, vma, prev, end)) 2831246c320aSKirill A. Shutemov downgrade = false; 28321da177e4SLinus Torvalds 2833dd2283f2SYang Shi if (downgrade) 2834d8ed45c5SMichel Lespinasse mmap_write_downgrade(mm); 2835dd2283f2SYang Shi 2836dd2283f2SYang Shi unmap_region(mm, vma, prev, start, end); 2837dd2283f2SYang Shi 28381da177e4SLinus Torvalds /* Fix up all other VM information */ 28392c0b3814SHugh Dickins remove_vma_list(mm, vma); 28401da177e4SLinus Torvalds 2841dd2283f2SYang Shi return downgrade ? 1 : 0; 28421da177e4SLinus Torvalds } 28431da177e4SLinus Torvalds 2844dd2283f2SYang Shi int do_munmap(struct mm_struct *mm, unsigned long start, size_t len, 2845dd2283f2SYang Shi struct list_head *uf) 2846dd2283f2SYang Shi { 2847dd2283f2SYang Shi return __do_munmap(mm, start, len, uf, false); 2848dd2283f2SYang Shi } 2849dd2283f2SYang Shi 2850dd2283f2SYang Shi static int __vm_munmap(unsigned long start, size_t len, bool downgrade) 2851a46ef99dSLinus Torvalds { 2852a46ef99dSLinus Torvalds int ret; 2853bfce281cSAl Viro struct mm_struct *mm = current->mm; 2854897ab3e0SMike Rapoport LIST_HEAD(uf); 2855a46ef99dSLinus Torvalds 2856d8ed45c5SMichel Lespinasse if (mmap_write_lock_killable(mm)) 2857ae798783SMichal Hocko return -EINTR; 2858ae798783SMichal Hocko 2859dd2283f2SYang Shi ret = __do_munmap(mm, start, len, &uf, downgrade); 2860dd2283f2SYang Shi /* 2861c1e8d7c6SMichel Lespinasse * Returning 1 indicates mmap_lock is downgraded. 2862dd2283f2SYang Shi * But 1 is not legal return value of vm_munmap() and munmap(), reset 2863dd2283f2SYang Shi * it to 0 before return. 2864dd2283f2SYang Shi */ 2865dd2283f2SYang Shi if (ret == 1) { 2866d8ed45c5SMichel Lespinasse mmap_read_unlock(mm); 2867dd2283f2SYang Shi ret = 0; 2868dd2283f2SYang Shi } else 2869d8ed45c5SMichel Lespinasse mmap_write_unlock(mm); 2870dd2283f2SYang Shi 2871897ab3e0SMike Rapoport userfaultfd_unmap_complete(mm, &uf); 2872a46ef99dSLinus Torvalds return ret; 2873a46ef99dSLinus Torvalds } 2874dd2283f2SYang Shi 2875dd2283f2SYang Shi int vm_munmap(unsigned long start, size_t len) 2876dd2283f2SYang Shi { 2877dd2283f2SYang Shi return __vm_munmap(start, len, false); 2878dd2283f2SYang Shi } 2879a46ef99dSLinus Torvalds EXPORT_SYMBOL(vm_munmap); 2880a46ef99dSLinus Torvalds 28816a6160a7SHeiko Carstens SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len) 28821da177e4SLinus Torvalds { 2883ce18d171SCatalin Marinas addr = untagged_addr(addr); 2884dd2283f2SYang Shi return __vm_munmap(addr, len, true); 28851da177e4SLinus Torvalds } 28861da177e4SLinus Torvalds 2887c8d78c18SKirill A. Shutemov 2888c8d78c18SKirill A. Shutemov /* 2889c8d78c18SKirill A. Shutemov * Emulation of deprecated remap_file_pages() syscall. 2890c8d78c18SKirill A. Shutemov */ 2891c8d78c18SKirill A. Shutemov SYSCALL_DEFINE5(remap_file_pages, unsigned long, start, unsigned long, size, 2892c8d78c18SKirill A. Shutemov unsigned long, prot, unsigned long, pgoff, unsigned long, flags) 2893c8d78c18SKirill A. Shutemov { 2894c8d78c18SKirill A. Shutemov 2895c8d78c18SKirill A. Shutemov struct mm_struct *mm = current->mm; 2896c8d78c18SKirill A. Shutemov struct vm_area_struct *vma; 2897c8d78c18SKirill A. Shutemov unsigned long populate = 0; 2898c8d78c18SKirill A. Shutemov unsigned long ret = -EINVAL; 2899c8d78c18SKirill A. Shutemov struct file *file; 2900c8d78c18SKirill A. Shutemov 2901ee65728eSMike Rapoport pr_warn_once("%s (%d) uses deprecated remap_file_pages() syscall. See Documentation/mm/remap_file_pages.rst.\n", 2902c8d78c18SKirill A. Shutemov current->comm, current->pid); 2903c8d78c18SKirill A. Shutemov 2904c8d78c18SKirill A. Shutemov if (prot) 2905c8d78c18SKirill A. Shutemov return ret; 2906c8d78c18SKirill A. Shutemov start = start & PAGE_MASK; 2907c8d78c18SKirill A. Shutemov size = size & PAGE_MASK; 2908c8d78c18SKirill A. Shutemov 2909c8d78c18SKirill A. Shutemov if (start + size <= start) 2910c8d78c18SKirill A. Shutemov return ret; 2911c8d78c18SKirill A. Shutemov 2912c8d78c18SKirill A. Shutemov /* Does pgoff wrap? */ 2913c8d78c18SKirill A. Shutemov if (pgoff + (size >> PAGE_SHIFT) < pgoff) 2914c8d78c18SKirill A. Shutemov return ret; 2915c8d78c18SKirill A. Shutemov 2916d8ed45c5SMichel Lespinasse if (mmap_write_lock_killable(mm)) 2917dc0ef0dfSMichal Hocko return -EINTR; 2918dc0ef0dfSMichal Hocko 29199b593cb2SLiam R. Howlett vma = vma_lookup(mm, start); 2920c8d78c18SKirill A. Shutemov 2921c8d78c18SKirill A. Shutemov if (!vma || !(vma->vm_flags & VM_SHARED)) 2922c8d78c18SKirill A. Shutemov goto out; 2923c8d78c18SKirill A. Shutemov 292448f7df32SKirill A. Shutemov if (start + size > vma->vm_end) { 292548f7df32SKirill A. Shutemov struct vm_area_struct *next; 292648f7df32SKirill A. Shutemov 292748f7df32SKirill A. Shutemov for (next = vma->vm_next; next; next = next->vm_next) { 292848f7df32SKirill A. Shutemov /* hole between vmas ? */ 292948f7df32SKirill A. Shutemov if (next->vm_start != next->vm_prev->vm_end) 293048f7df32SKirill A. Shutemov goto out; 293148f7df32SKirill A. Shutemov 293248f7df32SKirill A. Shutemov if (next->vm_file != vma->vm_file) 293348f7df32SKirill A. Shutemov goto out; 293448f7df32SKirill A. Shutemov 293548f7df32SKirill A. Shutemov if (next->vm_flags != vma->vm_flags) 293648f7df32SKirill A. Shutemov goto out; 293748f7df32SKirill A. Shutemov 293848f7df32SKirill A. Shutemov if (start + size <= next->vm_end) 293948f7df32SKirill A. Shutemov break; 294048f7df32SKirill A. Shutemov } 294148f7df32SKirill A. Shutemov 294248f7df32SKirill A. Shutemov if (!next) 2943c8d78c18SKirill A. Shutemov goto out; 2944c8d78c18SKirill A. Shutemov } 2945c8d78c18SKirill A. Shutemov 2946c8d78c18SKirill A. Shutemov prot |= vma->vm_flags & VM_READ ? PROT_READ : 0; 2947c8d78c18SKirill A. Shutemov prot |= vma->vm_flags & VM_WRITE ? PROT_WRITE : 0; 2948c8d78c18SKirill A. Shutemov prot |= vma->vm_flags & VM_EXEC ? PROT_EXEC : 0; 2949c8d78c18SKirill A. Shutemov 2950c8d78c18SKirill A. Shutemov flags &= MAP_NONBLOCK; 2951c8d78c18SKirill A. Shutemov flags |= MAP_SHARED | MAP_FIXED | MAP_POPULATE; 2952fce000b1SLiam Howlett if (vma->vm_flags & VM_LOCKED) 2953c8d78c18SKirill A. Shutemov flags |= MAP_LOCKED; 295448f7df32SKirill A. Shutemov 2955c8d78c18SKirill A. Shutemov file = get_file(vma->vm_file); 295645e55300SPeter Collingbourne ret = do_mmap(vma->vm_file, start, size, 2957897ab3e0SMike Rapoport prot, flags, pgoff, &populate, NULL); 2958c8d78c18SKirill A. Shutemov fput(file); 2959c8d78c18SKirill A. Shutemov out: 2960d8ed45c5SMichel Lespinasse mmap_write_unlock(mm); 2961c8d78c18SKirill A. Shutemov if (populate) 2962c8d78c18SKirill A. Shutemov mm_populate(ret, populate); 2963c8d78c18SKirill A. Shutemov if (!IS_ERR_VALUE(ret)) 2964c8d78c18SKirill A. Shutemov ret = 0; 2965c8d78c18SKirill A. Shutemov return ret; 2966c8d78c18SKirill A. Shutemov } 2967c8d78c18SKirill A. Shutemov 29681da177e4SLinus Torvalds /* 29691da177e4SLinus Torvalds * this is really a simplified "do_mmap". it only handles 29701da177e4SLinus Torvalds * anonymous maps. eventually we may be able to do some 29711da177e4SLinus Torvalds * brk-specific accounting here. 29721da177e4SLinus Torvalds */ 2973bb177a73SMichal Hocko static int do_brk_flags(unsigned long addr, unsigned long len, unsigned long flags, struct list_head *uf) 29741da177e4SLinus Torvalds { 29751da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 29761da177e4SLinus Torvalds struct vm_area_struct *vma, *prev; 29771da177e4SLinus Torvalds struct rb_node **rb_link, *rb_parent; 29781da177e4SLinus Torvalds pgoff_t pgoff = addr >> PAGE_SHIFT; 29793a459756SKirill Korotaev int error; 2980ff68dac6SGaowei Pu unsigned long mapped_addr; 29811da177e4SLinus Torvalds 298216e72e9bSDenys Vlasenko /* Until we need other flags, refuse anything except VM_EXEC. */ 298316e72e9bSDenys Vlasenko if ((flags & (~VM_EXEC)) != 0) 298416e72e9bSDenys Vlasenko return -EINVAL; 298516e72e9bSDenys Vlasenko flags |= VM_DATA_DEFAULT_FLAGS | VM_ACCOUNT | mm->def_flags; 29863a459756SKirill Korotaev 2987ff68dac6SGaowei Pu mapped_addr = get_unmapped_area(NULL, addr, len, 0, MAP_FIXED); 2988ff68dac6SGaowei Pu if (IS_ERR_VALUE(mapped_addr)) 2989ff68dac6SGaowei Pu return mapped_addr; 29903a459756SKirill Korotaev 2991363ee17fSDavidlohr Bueso error = mlock_future_check(mm, mm->def_flags, len); 2992363ee17fSDavidlohr Bueso if (error) 2993363ee17fSDavidlohr Bueso return error; 29941da177e4SLinus Torvalds 2995fb8090b6SLiam R. Howlett /* Clear old maps, set up prev, rb_link, rb_parent, and uf */ 2996fb8090b6SLiam R. Howlett if (munmap_vma_range(mm, addr, len, &prev, &rb_link, &rb_parent, uf)) 29971da177e4SLinus Torvalds return -ENOMEM; 29981da177e4SLinus Torvalds 29991da177e4SLinus Torvalds /* Check against address space limits *after* clearing old maps... */ 300084638335SKonstantin Khlebnikov if (!may_expand_vm(mm, flags, len >> PAGE_SHIFT)) 30011da177e4SLinus Torvalds return -ENOMEM; 30021da177e4SLinus Torvalds 30031da177e4SLinus Torvalds if (mm->map_count > sysctl_max_map_count) 30041da177e4SLinus Torvalds return -ENOMEM; 30051da177e4SLinus Torvalds 3006191c5424SAl Viro if (security_vm_enough_memory_mm(mm, len >> PAGE_SHIFT)) 30071da177e4SLinus Torvalds return -ENOMEM; 30081da177e4SLinus Torvalds 30091da177e4SLinus Torvalds /* Can we just expand an old private anonymous mapping? */ 3010ba470de4SRik van Riel vma = vma_merge(mm, prev, addr, addr + len, flags, 30119a10064fSColin Cross NULL, NULL, pgoff, NULL, NULL_VM_UFFD_CTX, NULL); 3012ba470de4SRik van Riel if (vma) 30131da177e4SLinus Torvalds goto out; 30141da177e4SLinus Torvalds 30151da177e4SLinus Torvalds /* 30161da177e4SLinus Torvalds * create a vma struct for an anonymous mapping 30171da177e4SLinus Torvalds */ 3018490fc053SLinus Torvalds vma = vm_area_alloc(mm); 30191da177e4SLinus Torvalds if (!vma) { 30201da177e4SLinus Torvalds vm_unacct_memory(len >> PAGE_SHIFT); 30211da177e4SLinus Torvalds return -ENOMEM; 30221da177e4SLinus Torvalds } 30231da177e4SLinus Torvalds 3024bfd40eafSKirill A. Shutemov vma_set_anonymous(vma); 30251da177e4SLinus Torvalds vma->vm_start = addr; 30261da177e4SLinus Torvalds vma->vm_end = addr + len; 30271da177e4SLinus Torvalds vma->vm_pgoff = pgoff; 30281da177e4SLinus Torvalds vma->vm_flags = flags; 30293ed75eb8SColy Li vma->vm_page_prot = vm_get_page_prot(flags); 30301da177e4SLinus Torvalds vma_link(mm, vma, prev, rb_link, rb_parent); 30311da177e4SLinus Torvalds out: 30323af9e859SEric B Munson perf_event_mmap(vma); 30331da177e4SLinus Torvalds mm->total_vm += len >> PAGE_SHIFT; 303484638335SKonstantin Khlebnikov mm->data_vm += len >> PAGE_SHIFT; 3035128557ffSMichel Lespinasse if (flags & VM_LOCKED) 3036ba470de4SRik van Riel mm->locked_vm += (len >> PAGE_SHIFT); 3037d9104d1cSCyrill Gorcunov vma->vm_flags |= VM_SOFTDIRTY; 30385d22fc25SLinus Torvalds return 0; 30391da177e4SLinus Torvalds } 30401da177e4SLinus Torvalds 3041bb177a73SMichal Hocko int vm_brk_flags(unsigned long addr, unsigned long request, unsigned long flags) 3042e4eb1ff6SLinus Torvalds { 3043e4eb1ff6SLinus Torvalds struct mm_struct *mm = current->mm; 3044bb177a73SMichal Hocko unsigned long len; 30455d22fc25SLinus Torvalds int ret; 3046128557ffSMichel Lespinasse bool populate; 3047897ab3e0SMike Rapoport LIST_HEAD(uf); 3048e4eb1ff6SLinus Torvalds 3049bb177a73SMichal Hocko len = PAGE_ALIGN(request); 3050bb177a73SMichal Hocko if (len < request) 3051bb177a73SMichal Hocko return -ENOMEM; 3052bb177a73SMichal Hocko if (!len) 3053bb177a73SMichal Hocko return 0; 3054bb177a73SMichal Hocko 3055d8ed45c5SMichel Lespinasse if (mmap_write_lock_killable(mm)) 30562d6c9282SMichal Hocko return -EINTR; 30572d6c9282SMichal Hocko 3058897ab3e0SMike Rapoport ret = do_brk_flags(addr, len, flags, &uf); 3059128557ffSMichel Lespinasse populate = ((mm->def_flags & VM_LOCKED) != 0); 3060d8ed45c5SMichel Lespinasse mmap_write_unlock(mm); 3061897ab3e0SMike Rapoport userfaultfd_unmap_complete(mm, &uf); 30625d22fc25SLinus Torvalds if (populate && !ret) 3063128557ffSMichel Lespinasse mm_populate(addr, len); 3064e4eb1ff6SLinus Torvalds return ret; 3065e4eb1ff6SLinus Torvalds } 306616e72e9bSDenys Vlasenko EXPORT_SYMBOL(vm_brk_flags); 306716e72e9bSDenys Vlasenko 306816e72e9bSDenys Vlasenko int vm_brk(unsigned long addr, unsigned long len) 306916e72e9bSDenys Vlasenko { 307016e72e9bSDenys Vlasenko return vm_brk_flags(addr, len, 0); 307116e72e9bSDenys Vlasenko } 3072e4eb1ff6SLinus Torvalds EXPORT_SYMBOL(vm_brk); 30731da177e4SLinus Torvalds 30741da177e4SLinus Torvalds /* Release all mmaps. */ 30751da177e4SLinus Torvalds void exit_mmap(struct mm_struct *mm) 30761da177e4SLinus Torvalds { 3077d16dfc55SPeter Zijlstra struct mmu_gather tlb; 3078ba470de4SRik van Riel struct vm_area_struct *vma; 30791da177e4SLinus Torvalds unsigned long nr_accounted = 0; 30801da177e4SLinus Torvalds 3081d6dd61c8SJeremy Fitzhardinge /* mm's last user has gone, and its about to be pulled down */ 3082cddb8a5cSAndrea Arcangeli mmu_notifier_release(mm); 3083d6dd61c8SJeremy Fitzhardinge 308427ae357fSDavid Rientjes if (unlikely(mm_is_oom_victim(mm))) { 308527ae357fSDavid Rientjes /* 308627ae357fSDavid Rientjes * Manually reap the mm to free as much memory as possible. 308727ae357fSDavid Rientjes * Then, as the oom reaper does, set MMF_OOM_SKIP to disregard 3088c1e8d7c6SMichel Lespinasse * this mm from further consideration. Taking mm->mmap_lock for 308927ae357fSDavid Rientjes * write after setting MMF_OOM_SKIP will guarantee that the oom 3090c1e8d7c6SMichel Lespinasse * reaper will not run on this mm again after mmap_lock is 309127ae357fSDavid Rientjes * dropped. 309227ae357fSDavid Rientjes * 3093c1e8d7c6SMichel Lespinasse * Nothing can be holding mm->mmap_lock here and the above call 309427ae357fSDavid Rientjes * to mmu_notifier_release(mm) ensures mmu notifier callbacks in 309527ae357fSDavid Rientjes * __oom_reap_task_mm() will not block. 309627ae357fSDavid Rientjes */ 309793065ac7SMichal Hocko (void)__oom_reap_task_mm(mm); 309827ae357fSDavid Rientjes set_bit(MMF_OOM_SKIP, &mm->flags); 309927ae357fSDavid Rientjes } 310027ae357fSDavid Rientjes 310164591e86SSuren Baghdasaryan mmap_write_lock(mm); 31029480c53eSJeremy Fitzhardinge arch_exit_mmap(mm); 31039480c53eSJeremy Fitzhardinge 3104ba470de4SRik van Riel vma = mm->mmap; 310564591e86SSuren Baghdasaryan if (!vma) { 310664591e86SSuren Baghdasaryan /* Can happen if dup_mmap() received an OOM */ 310764591e86SSuren Baghdasaryan mmap_write_unlock(mm); 31089480c53eSJeremy Fitzhardinge return; 310964591e86SSuren Baghdasaryan } 31109480c53eSJeremy Fitzhardinge 31111da177e4SLinus Torvalds lru_add_drain(); 31121da177e4SLinus Torvalds flush_cache_mm(mm); 3113d8b45053SWill Deacon tlb_gather_mmu_fullmm(&tlb, mm); 3114901608d9SOleg Nesterov /* update_hiwater_rss(mm) here? but nobody should be looking */ 3115e0da382cSHugh Dickins /* Use -1 here to ensure all VMAs in the mm are unmapped */ 31164f74d2c8SLinus Torvalds unmap_vmas(&tlb, vma, 0, -1); 31176ee8630eSHugh Dickins free_pgtables(&tlb, vma, FIRST_USER_ADDRESS, USER_PGTABLES_CEILING); 3118ae8eba8bSWill Deacon tlb_finish_mmu(&tlb); 31191da177e4SLinus Torvalds 312064591e86SSuren Baghdasaryan /* Walk the list again, actually closing and freeing it. */ 31214f74d2c8SLinus Torvalds while (vma) { 31224f74d2c8SLinus Torvalds if (vma->vm_flags & VM_ACCOUNT) 31234f74d2c8SLinus Torvalds nr_accounted += vma_pages(vma); 3124a8fb5618SHugh Dickins vma = remove_vma(vma); 31250a3b3c25SPaul E. McKenney cond_resched(); 31264f74d2c8SLinus Torvalds } 3127f798a1d4SSuren Baghdasaryan mm->mmap = NULL; 312864591e86SSuren Baghdasaryan mmap_write_unlock(mm); 31294f74d2c8SLinus Torvalds vm_unacct_memory(nr_accounted); 31301da177e4SLinus Torvalds } 31311da177e4SLinus Torvalds 31321da177e4SLinus Torvalds /* Insert vm structure into process list sorted by address 31331da177e4SLinus Torvalds * and into the inode's i_mmap tree. If vm_file is non-NULL 3134c8c06efaSDavidlohr Bueso * then i_mmap_rwsem is taken here. 31351da177e4SLinus Torvalds */ 31361da177e4SLinus Torvalds int insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma) 31371da177e4SLinus Torvalds { 31386597d783SHugh Dickins struct vm_area_struct *prev; 31391da177e4SLinus Torvalds struct rb_node **rb_link, *rb_parent; 31401da177e4SLinus Torvalds 3141c9d13f5fSChen Gang if (find_vma_links(mm, vma->vm_start, vma->vm_end, 3142c9d13f5fSChen Gang &prev, &rb_link, &rb_parent)) 3143c9d13f5fSChen Gang return -ENOMEM; 3144c9d13f5fSChen Gang if ((vma->vm_flags & VM_ACCOUNT) && 3145c9d13f5fSChen Gang security_vm_enough_memory_mm(mm, vma_pages(vma))) 3146c9d13f5fSChen Gang return -ENOMEM; 3147c9d13f5fSChen Gang 31481da177e4SLinus Torvalds /* 31491da177e4SLinus Torvalds * The vm_pgoff of a purely anonymous vma should be irrelevant 31501da177e4SLinus Torvalds * until its first write fault, when page's anon_vma and index 31511da177e4SLinus Torvalds * are set. But now set the vm_pgoff it will almost certainly 31521da177e4SLinus Torvalds * end up with (unless mremap moves it elsewhere before that 31531da177e4SLinus Torvalds * first wfault), so /proc/pid/maps tells a consistent story. 31541da177e4SLinus Torvalds * 31551da177e4SLinus Torvalds * By setting it to reflect the virtual start address of the 31561da177e4SLinus Torvalds * vma, merges and splits can happen in a seamless way, just 31571da177e4SLinus Torvalds * using the existing file pgoff checks and manipulations. 31588332326eSLiao Pingfang * Similarly in do_mmap and in do_brk_flags. 31591da177e4SLinus Torvalds */ 31608a9cc3b5SOleg Nesterov if (vma_is_anonymous(vma)) { 31611da177e4SLinus Torvalds BUG_ON(vma->anon_vma); 31621da177e4SLinus Torvalds vma->vm_pgoff = vma->vm_start >> PAGE_SHIFT; 31631da177e4SLinus Torvalds } 31642b144498SSrikar Dronamraju 31651da177e4SLinus Torvalds vma_link(mm, vma, prev, rb_link, rb_parent); 31661da177e4SLinus Torvalds return 0; 31671da177e4SLinus Torvalds } 31681da177e4SLinus Torvalds 31691da177e4SLinus Torvalds /* 31701da177e4SLinus Torvalds * Copy the vma structure to a new location in the same mm, 31711da177e4SLinus Torvalds * prior to moving page table entries, to effect an mremap move. 31721da177e4SLinus Torvalds */ 31731da177e4SLinus Torvalds struct vm_area_struct *copy_vma(struct vm_area_struct **vmap, 317438a76013SMichel Lespinasse unsigned long addr, unsigned long len, pgoff_t pgoff, 317538a76013SMichel Lespinasse bool *need_rmap_locks) 31761da177e4SLinus Torvalds { 31771da177e4SLinus Torvalds struct vm_area_struct *vma = *vmap; 31781da177e4SLinus Torvalds unsigned long vma_start = vma->vm_start; 31791da177e4SLinus Torvalds struct mm_struct *mm = vma->vm_mm; 31801da177e4SLinus Torvalds struct vm_area_struct *new_vma, *prev; 31811da177e4SLinus Torvalds struct rb_node **rb_link, *rb_parent; 3182948f017bSAndrea Arcangeli bool faulted_in_anon_vma = true; 31831da177e4SLinus Torvalds 31841da177e4SLinus Torvalds /* 31851da177e4SLinus Torvalds * If anonymous vma has not yet been faulted, update new pgoff 31861da177e4SLinus Torvalds * to match new location, to increase its chance of merging. 31871da177e4SLinus Torvalds */ 3188ce75799bSOleg Nesterov if (unlikely(vma_is_anonymous(vma) && !vma->anon_vma)) { 31891da177e4SLinus Torvalds pgoff = addr >> PAGE_SHIFT; 3190948f017bSAndrea Arcangeli faulted_in_anon_vma = false; 3191948f017bSAndrea Arcangeli } 31921da177e4SLinus Torvalds 31936597d783SHugh Dickins if (find_vma_links(mm, addr, addr + len, &prev, &rb_link, &rb_parent)) 31946597d783SHugh Dickins return NULL; /* should never get here */ 31951da177e4SLinus Torvalds new_vma = vma_merge(mm, prev, addr, addr + len, vma->vm_flags, 319619a809afSAndrea Arcangeli vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma), 31975c26f6acSSuren Baghdasaryan vma->vm_userfaultfd_ctx, anon_vma_name(vma)); 31981da177e4SLinus Torvalds if (new_vma) { 31991da177e4SLinus Torvalds /* 32001da177e4SLinus Torvalds * Source vma may have been merged into new_vma 32011da177e4SLinus Torvalds */ 3202948f017bSAndrea Arcangeli if (unlikely(vma_start >= new_vma->vm_start && 3203948f017bSAndrea Arcangeli vma_start < new_vma->vm_end)) { 3204948f017bSAndrea Arcangeli /* 3205948f017bSAndrea Arcangeli * The only way we can get a vma_merge with 3206948f017bSAndrea Arcangeli * self during an mremap is if the vma hasn't 3207948f017bSAndrea Arcangeli * been faulted in yet and we were allowed to 3208948f017bSAndrea Arcangeli * reset the dst vma->vm_pgoff to the 3209948f017bSAndrea Arcangeli * destination address of the mremap to allow 3210948f017bSAndrea Arcangeli * the merge to happen. mremap must change the 3211948f017bSAndrea Arcangeli * vm_pgoff linearity between src and dst vmas 3212948f017bSAndrea Arcangeli * (in turn preventing a vma_merge) to be 3213948f017bSAndrea Arcangeli * safe. It is only safe to keep the vm_pgoff 3214948f017bSAndrea Arcangeli * linear if there are no pages mapped yet. 3215948f017bSAndrea Arcangeli */ 321681d1b09cSSasha Levin VM_BUG_ON_VMA(faulted_in_anon_vma, new_vma); 321738a76013SMichel Lespinasse *vmap = vma = new_vma; 3218108d6642SMichel Lespinasse } 321938a76013SMichel Lespinasse *need_rmap_locks = (new_vma->vm_pgoff <= vma->vm_pgoff); 32201da177e4SLinus Torvalds } else { 32213928d4f5SLinus Torvalds new_vma = vm_area_dup(vma); 3222e3975891SChen Gang if (!new_vma) 3223e3975891SChen Gang goto out; 32241da177e4SLinus Torvalds new_vma->vm_start = addr; 32251da177e4SLinus Torvalds new_vma->vm_end = addr + len; 32261da177e4SLinus Torvalds new_vma->vm_pgoff = pgoff; 3227ef0855d3SOleg Nesterov if (vma_dup_policy(vma, new_vma)) 3228523d4e20SMichel Lespinasse goto out_free_vma; 3229523d4e20SMichel Lespinasse if (anon_vma_clone(new_vma, vma)) 3230523d4e20SMichel Lespinasse goto out_free_mempol; 3231e9714acfSKonstantin Khlebnikov if (new_vma->vm_file) 32321da177e4SLinus Torvalds get_file(new_vma->vm_file); 32331da177e4SLinus Torvalds if (new_vma->vm_ops && new_vma->vm_ops->open) 32341da177e4SLinus Torvalds new_vma->vm_ops->open(new_vma); 32351da177e4SLinus Torvalds vma_link(mm, new_vma, prev, rb_link, rb_parent); 323638a76013SMichel Lespinasse *need_rmap_locks = false; 32371da177e4SLinus Torvalds } 32381da177e4SLinus Torvalds return new_vma; 32395beb4930SRik van Riel 32405beb4930SRik van Riel out_free_mempol: 3241ef0855d3SOleg Nesterov mpol_put(vma_policy(new_vma)); 32425beb4930SRik van Riel out_free_vma: 32433928d4f5SLinus Torvalds vm_area_free(new_vma); 3244e3975891SChen Gang out: 32455beb4930SRik van Riel return NULL; 32461da177e4SLinus Torvalds } 3247119f657cSakpm@osdl.org 3248119f657cSakpm@osdl.org /* 3249119f657cSakpm@osdl.org * Return true if the calling process may expand its vm space by the passed 3250119f657cSakpm@osdl.org * number of pages 3251119f657cSakpm@osdl.org */ 325284638335SKonstantin Khlebnikov bool may_expand_vm(struct mm_struct *mm, vm_flags_t flags, unsigned long npages) 3253119f657cSakpm@osdl.org { 325484638335SKonstantin Khlebnikov if (mm->total_vm + npages > rlimit(RLIMIT_AS) >> PAGE_SHIFT) 325584638335SKonstantin Khlebnikov return false; 3256119f657cSakpm@osdl.org 3257d977d56cSKonstantin Khlebnikov if (is_data_mapping(flags) && 3258d977d56cSKonstantin Khlebnikov mm->data_vm + npages > rlimit(RLIMIT_DATA) >> PAGE_SHIFT) { 3259f4fcd558SKonstantin Khlebnikov /* Workaround for Valgrind */ 3260f4fcd558SKonstantin Khlebnikov if (rlimit(RLIMIT_DATA) == 0 && 3261f4fcd558SKonstantin Khlebnikov mm->data_vm + npages <= rlimit_max(RLIMIT_DATA) >> PAGE_SHIFT) 3262f4fcd558SKonstantin Khlebnikov return true; 326357a7702bSDavid Woodhouse 326457a7702bSDavid Woodhouse pr_warn_once("%s (%d): VmData %lu exceed data ulimit %lu. Update limits%s.\n", 3265d977d56cSKonstantin Khlebnikov current->comm, current->pid, 3266d977d56cSKonstantin Khlebnikov (mm->data_vm + npages) << PAGE_SHIFT, 326757a7702bSDavid Woodhouse rlimit(RLIMIT_DATA), 326857a7702bSDavid Woodhouse ignore_rlimit_data ? "" : " or use boot option ignore_rlimit_data"); 326957a7702bSDavid Woodhouse 327057a7702bSDavid Woodhouse if (!ignore_rlimit_data) 3271d977d56cSKonstantin Khlebnikov return false; 3272d977d56cSKonstantin Khlebnikov } 3273119f657cSakpm@osdl.org 327484638335SKonstantin Khlebnikov return true; 327584638335SKonstantin Khlebnikov } 327684638335SKonstantin Khlebnikov 327784638335SKonstantin Khlebnikov void vm_stat_account(struct mm_struct *mm, vm_flags_t flags, long npages) 327884638335SKonstantin Khlebnikov { 32797866076bSPeng Liu WRITE_ONCE(mm->total_vm, READ_ONCE(mm->total_vm)+npages); 328084638335SKonstantin Khlebnikov 3281d977d56cSKonstantin Khlebnikov if (is_exec_mapping(flags)) 328284638335SKonstantin Khlebnikov mm->exec_vm += npages; 3283d977d56cSKonstantin Khlebnikov else if (is_stack_mapping(flags)) 328484638335SKonstantin Khlebnikov mm->stack_vm += npages; 3285d977d56cSKonstantin Khlebnikov else if (is_data_mapping(flags)) 328684638335SKonstantin Khlebnikov mm->data_vm += npages; 3287119f657cSakpm@osdl.org } 3288fa5dc22fSRoland McGrath 3289b3ec9f33SSouptick Joarder static vm_fault_t special_mapping_fault(struct vm_fault *vmf); 3290a62c34bdSAndy Lutomirski 3291a62c34bdSAndy Lutomirski /* 3292a62c34bdSAndy Lutomirski * Having a close hook prevents vma merging regardless of flags. 3293a62c34bdSAndy Lutomirski */ 3294a62c34bdSAndy Lutomirski static void special_mapping_close(struct vm_area_struct *vma) 3295a62c34bdSAndy Lutomirski { 3296a62c34bdSAndy Lutomirski } 3297a62c34bdSAndy Lutomirski 3298a62c34bdSAndy Lutomirski static const char *special_mapping_name(struct vm_area_struct *vma) 3299a62c34bdSAndy Lutomirski { 3300a62c34bdSAndy Lutomirski return ((struct vm_special_mapping *)vma->vm_private_data)->name; 3301a62c34bdSAndy Lutomirski } 3302a62c34bdSAndy Lutomirski 330314d07113SBrian Geffon static int special_mapping_mremap(struct vm_area_struct *new_vma) 3304b059a453SDmitry Safonov { 3305b059a453SDmitry Safonov struct vm_special_mapping *sm = new_vma->vm_private_data; 3306b059a453SDmitry Safonov 3307280e87e9SDmitry Safonov if (WARN_ON_ONCE(current->mm != new_vma->vm_mm)) 3308280e87e9SDmitry Safonov return -EFAULT; 3309280e87e9SDmitry Safonov 3310b059a453SDmitry Safonov if (sm->mremap) 3311b059a453SDmitry Safonov return sm->mremap(sm, new_vma); 3312280e87e9SDmitry Safonov 3313b059a453SDmitry Safonov return 0; 3314b059a453SDmitry Safonov } 3315b059a453SDmitry Safonov 3316871402e0SDmitry Safonov static int special_mapping_split(struct vm_area_struct *vma, unsigned long addr) 3317871402e0SDmitry Safonov { 3318871402e0SDmitry Safonov /* 3319871402e0SDmitry Safonov * Forbid splitting special mappings - kernel has expectations over 3320871402e0SDmitry Safonov * the number of pages in mapping. Together with VM_DONTEXPAND 3321871402e0SDmitry Safonov * the size of vma should stay the same over the special mapping's 3322871402e0SDmitry Safonov * lifetime. 3323871402e0SDmitry Safonov */ 3324871402e0SDmitry Safonov return -EINVAL; 3325871402e0SDmitry Safonov } 3326871402e0SDmitry Safonov 3327a62c34bdSAndy Lutomirski static const struct vm_operations_struct special_mapping_vmops = { 3328a62c34bdSAndy Lutomirski .close = special_mapping_close, 3329a62c34bdSAndy Lutomirski .fault = special_mapping_fault, 3330b059a453SDmitry Safonov .mremap = special_mapping_mremap, 3331a62c34bdSAndy Lutomirski .name = special_mapping_name, 3332af34ebebSDmitry Safonov /* vDSO code relies that VVAR can't be accessed remotely */ 3333af34ebebSDmitry Safonov .access = NULL, 3334871402e0SDmitry Safonov .may_split = special_mapping_split, 3335a62c34bdSAndy Lutomirski }; 3336a62c34bdSAndy Lutomirski 3337a62c34bdSAndy Lutomirski static const struct vm_operations_struct legacy_special_mapping_vmops = { 3338a62c34bdSAndy Lutomirski .close = special_mapping_close, 3339a62c34bdSAndy Lutomirski .fault = special_mapping_fault, 3340a62c34bdSAndy Lutomirski }; 3341fa5dc22fSRoland McGrath 3342b3ec9f33SSouptick Joarder static vm_fault_t special_mapping_fault(struct vm_fault *vmf) 3343fa5dc22fSRoland McGrath { 334411bac800SDave Jiang struct vm_area_struct *vma = vmf->vma; 3345b1d0e4f5SNick Piggin pgoff_t pgoff; 3346fa5dc22fSRoland McGrath struct page **pages; 3347fa5dc22fSRoland McGrath 3348f872f540SAndy Lutomirski if (vma->vm_ops == &legacy_special_mapping_vmops) { 3349a62c34bdSAndy Lutomirski pages = vma->vm_private_data; 3350f872f540SAndy Lutomirski } else { 3351f872f540SAndy Lutomirski struct vm_special_mapping *sm = vma->vm_private_data; 3352f872f540SAndy Lutomirski 3353f872f540SAndy Lutomirski if (sm->fault) 335411bac800SDave Jiang return sm->fault(sm, vmf->vma, vmf); 3355f872f540SAndy Lutomirski 3356f872f540SAndy Lutomirski pages = sm->pages; 3357f872f540SAndy Lutomirski } 3358a62c34bdSAndy Lutomirski 33598a9cc3b5SOleg Nesterov for (pgoff = vmf->pgoff; pgoff && *pages; ++pages) 3360b1d0e4f5SNick Piggin pgoff--; 3361fa5dc22fSRoland McGrath 3362fa5dc22fSRoland McGrath if (*pages) { 3363fa5dc22fSRoland McGrath struct page *page = *pages; 3364fa5dc22fSRoland McGrath get_page(page); 3365b1d0e4f5SNick Piggin vmf->page = page; 3366b1d0e4f5SNick Piggin return 0; 3367fa5dc22fSRoland McGrath } 3368fa5dc22fSRoland McGrath 3369b1d0e4f5SNick Piggin return VM_FAULT_SIGBUS; 3370fa5dc22fSRoland McGrath } 3371fa5dc22fSRoland McGrath 3372a62c34bdSAndy Lutomirski static struct vm_area_struct *__install_special_mapping( 3373a62c34bdSAndy Lutomirski struct mm_struct *mm, 3374fa5dc22fSRoland McGrath unsigned long addr, unsigned long len, 337527f28b97SChen Gang unsigned long vm_flags, void *priv, 337627f28b97SChen Gang const struct vm_operations_struct *ops) 3377fa5dc22fSRoland McGrath { 3378462e635eSTavis Ormandy int ret; 3379fa5dc22fSRoland McGrath struct vm_area_struct *vma; 3380fa5dc22fSRoland McGrath 3381490fc053SLinus Torvalds vma = vm_area_alloc(mm); 3382fa5dc22fSRoland McGrath if (unlikely(vma == NULL)) 33833935ed6aSStefani Seibold return ERR_PTR(-ENOMEM); 3384fa5dc22fSRoland McGrath 3385fa5dc22fSRoland McGrath vma->vm_start = addr; 3386fa5dc22fSRoland McGrath vma->vm_end = addr + len; 3387fa5dc22fSRoland McGrath 3388d9104d1cSCyrill Gorcunov vma->vm_flags = vm_flags | mm->def_flags | VM_DONTEXPAND | VM_SOFTDIRTY; 33891fc09228SHugh Dickins vma->vm_flags &= VM_LOCKED_CLEAR_MASK; 33903ed75eb8SColy Li vma->vm_page_prot = vm_get_page_prot(vma->vm_flags); 3391fa5dc22fSRoland McGrath 3392a62c34bdSAndy Lutomirski vma->vm_ops = ops; 3393a62c34bdSAndy Lutomirski vma->vm_private_data = priv; 3394fa5dc22fSRoland McGrath 3395462e635eSTavis Ormandy ret = insert_vm_struct(mm, vma); 3396462e635eSTavis Ormandy if (ret) 3397462e635eSTavis Ormandy goto out; 3398fa5dc22fSRoland McGrath 339984638335SKonstantin Khlebnikov vm_stat_account(mm, vma->vm_flags, len >> PAGE_SHIFT); 3400fa5dc22fSRoland McGrath 3401cdd6c482SIngo Molnar perf_event_mmap(vma); 3402089dd79dSPeter Zijlstra 34033935ed6aSStefani Seibold return vma; 3404462e635eSTavis Ormandy 3405462e635eSTavis Ormandy out: 34063928d4f5SLinus Torvalds vm_area_free(vma); 34073935ed6aSStefani Seibold return ERR_PTR(ret); 34083935ed6aSStefani Seibold } 34093935ed6aSStefani Seibold 34102eefd878SDmitry Safonov bool vma_is_special_mapping(const struct vm_area_struct *vma, 34112eefd878SDmitry Safonov const struct vm_special_mapping *sm) 34122eefd878SDmitry Safonov { 34132eefd878SDmitry Safonov return vma->vm_private_data == sm && 34142eefd878SDmitry Safonov (vma->vm_ops == &special_mapping_vmops || 34152eefd878SDmitry Safonov vma->vm_ops == &legacy_special_mapping_vmops); 34162eefd878SDmitry Safonov } 34172eefd878SDmitry Safonov 3418a62c34bdSAndy Lutomirski /* 3419c1e8d7c6SMichel Lespinasse * Called with mm->mmap_lock held for writing. 3420a62c34bdSAndy Lutomirski * Insert a new vma covering the given region, with the given flags. 3421a62c34bdSAndy Lutomirski * Its pages are supplied by the given array of struct page *. 3422a62c34bdSAndy Lutomirski * The array can be shorter than len >> PAGE_SHIFT if it's null-terminated. 3423a62c34bdSAndy Lutomirski * The region past the last page supplied will always produce SIGBUS. 3424a62c34bdSAndy Lutomirski * The array pointer and the pages it points to are assumed to stay alive 3425a62c34bdSAndy Lutomirski * for as long as this mapping might exist. 3426a62c34bdSAndy Lutomirski */ 3427a62c34bdSAndy Lutomirski struct vm_area_struct *_install_special_mapping( 3428a62c34bdSAndy Lutomirski struct mm_struct *mm, 3429a62c34bdSAndy Lutomirski unsigned long addr, unsigned long len, 3430a62c34bdSAndy Lutomirski unsigned long vm_flags, const struct vm_special_mapping *spec) 3431a62c34bdSAndy Lutomirski { 343227f28b97SChen Gang return __install_special_mapping(mm, addr, len, vm_flags, (void *)spec, 343327f28b97SChen Gang &special_mapping_vmops); 3434a62c34bdSAndy Lutomirski } 3435a62c34bdSAndy Lutomirski 34363935ed6aSStefani Seibold int install_special_mapping(struct mm_struct *mm, 34373935ed6aSStefani Seibold unsigned long addr, unsigned long len, 34383935ed6aSStefani Seibold unsigned long vm_flags, struct page **pages) 34393935ed6aSStefani Seibold { 3440a62c34bdSAndy Lutomirski struct vm_area_struct *vma = __install_special_mapping( 344127f28b97SChen Gang mm, addr, len, vm_flags, (void *)pages, 344227f28b97SChen Gang &legacy_special_mapping_vmops); 34433935ed6aSStefani Seibold 344414bd5b45SDuan Jiong return PTR_ERR_OR_ZERO(vma); 3445fa5dc22fSRoland McGrath } 34467906d00cSAndrea Arcangeli 34477906d00cSAndrea Arcangeli static DEFINE_MUTEX(mm_all_locks_mutex); 34487906d00cSAndrea Arcangeli 3449454ed842SPeter Zijlstra static void vm_lock_anon_vma(struct mm_struct *mm, struct anon_vma *anon_vma) 34507906d00cSAndrea Arcangeli { 3451f808c13fSDavidlohr Bueso if (!test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_root.rb_node)) { 34527906d00cSAndrea Arcangeli /* 34537906d00cSAndrea Arcangeli * The LSB of head.next can't change from under us 34547906d00cSAndrea Arcangeli * because we hold the mm_all_locks_mutex. 34557906d00cSAndrea Arcangeli */ 3456da1c55f1SMichel Lespinasse down_write_nest_lock(&anon_vma->root->rwsem, &mm->mmap_lock); 34577906d00cSAndrea Arcangeli /* 34587906d00cSAndrea Arcangeli * We can safely modify head.next after taking the 34595a505085SIngo Molnar * anon_vma->root->rwsem. If some other vma in this mm shares 34607906d00cSAndrea Arcangeli * the same anon_vma we won't take it again. 34617906d00cSAndrea Arcangeli * 34627906d00cSAndrea Arcangeli * No need of atomic instructions here, head.next 34637906d00cSAndrea Arcangeli * can't change from under us thanks to the 34645a505085SIngo Molnar * anon_vma->root->rwsem. 34657906d00cSAndrea Arcangeli */ 34667906d00cSAndrea Arcangeli if (__test_and_set_bit(0, (unsigned long *) 3467f808c13fSDavidlohr Bueso &anon_vma->root->rb_root.rb_root.rb_node)) 34687906d00cSAndrea Arcangeli BUG(); 34697906d00cSAndrea Arcangeli } 34707906d00cSAndrea Arcangeli } 34717906d00cSAndrea Arcangeli 3472454ed842SPeter Zijlstra static void vm_lock_mapping(struct mm_struct *mm, struct address_space *mapping) 34737906d00cSAndrea Arcangeli { 34747906d00cSAndrea Arcangeli if (!test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) { 34757906d00cSAndrea Arcangeli /* 34767906d00cSAndrea Arcangeli * AS_MM_ALL_LOCKS can't change from under us because 34777906d00cSAndrea Arcangeli * we hold the mm_all_locks_mutex. 34787906d00cSAndrea Arcangeli * 34797906d00cSAndrea Arcangeli * Operations on ->flags have to be atomic because 34807906d00cSAndrea Arcangeli * even if AS_MM_ALL_LOCKS is stable thanks to the 34817906d00cSAndrea Arcangeli * mm_all_locks_mutex, there may be other cpus 34827906d00cSAndrea Arcangeli * changing other bitflags in parallel to us. 34837906d00cSAndrea Arcangeli */ 34847906d00cSAndrea Arcangeli if (test_and_set_bit(AS_MM_ALL_LOCKS, &mapping->flags)) 34857906d00cSAndrea Arcangeli BUG(); 3486da1c55f1SMichel Lespinasse down_write_nest_lock(&mapping->i_mmap_rwsem, &mm->mmap_lock); 34877906d00cSAndrea Arcangeli } 34887906d00cSAndrea Arcangeli } 34897906d00cSAndrea Arcangeli 34907906d00cSAndrea Arcangeli /* 34917906d00cSAndrea Arcangeli * This operation locks against the VM for all pte/vma/mm related 34927906d00cSAndrea Arcangeli * operations that could ever happen on a certain mm. This includes 34937906d00cSAndrea Arcangeli * vmtruncate, try_to_unmap, and all page faults. 34947906d00cSAndrea Arcangeli * 3495c1e8d7c6SMichel Lespinasse * The caller must take the mmap_lock in write mode before calling 34967906d00cSAndrea Arcangeli * mm_take_all_locks(). The caller isn't allowed to release the 3497c1e8d7c6SMichel Lespinasse * mmap_lock until mm_drop_all_locks() returns. 34987906d00cSAndrea Arcangeli * 3499c1e8d7c6SMichel Lespinasse * mmap_lock in write mode is required in order to block all operations 35007906d00cSAndrea Arcangeli * that could modify pagetables and free pages without need of 350127ba0644SKirill A. Shutemov * altering the vma layout. It's also needed in write mode to avoid new 35027906d00cSAndrea Arcangeli * anon_vmas to be associated with existing vmas. 35037906d00cSAndrea Arcangeli * 35047906d00cSAndrea Arcangeli * A single task can't take more than one mm_take_all_locks() in a row 35057906d00cSAndrea Arcangeli * or it would deadlock. 35067906d00cSAndrea Arcangeli * 3507bf181b9fSMichel Lespinasse * The LSB in anon_vma->rb_root.rb_node and the AS_MM_ALL_LOCKS bitflag in 35087906d00cSAndrea Arcangeli * mapping->flags avoid to take the same lock twice, if more than one 35097906d00cSAndrea Arcangeli * vma in this mm is backed by the same anon_vma or address_space. 35107906d00cSAndrea Arcangeli * 351188f306b6SKirill A. Shutemov * We take locks in following order, accordingly to comment at beginning 351288f306b6SKirill A. Shutemov * of mm/rmap.c: 351388f306b6SKirill A. Shutemov * - all hugetlbfs_i_mmap_rwsem_key locks (aka mapping->i_mmap_rwsem for 351488f306b6SKirill A. Shutemov * hugetlb mapping); 351588f306b6SKirill A. Shutemov * - all i_mmap_rwsem locks; 351688f306b6SKirill A. Shutemov * - all anon_vma->rwseml 351788f306b6SKirill A. Shutemov * 351888f306b6SKirill A. Shutemov * We can take all locks within these types randomly because the VM code 351988f306b6SKirill A. Shutemov * doesn't nest them and we protected from parallel mm_take_all_locks() by 352088f306b6SKirill A. Shutemov * mm_all_locks_mutex. 35217906d00cSAndrea Arcangeli * 35227906d00cSAndrea Arcangeli * mm_take_all_locks() and mm_drop_all_locks are expensive operations 35237906d00cSAndrea Arcangeli * that may have to take thousand of locks. 35247906d00cSAndrea Arcangeli * 35257906d00cSAndrea Arcangeli * mm_take_all_locks() can fail if it's interrupted by signals. 35267906d00cSAndrea Arcangeli */ 35277906d00cSAndrea Arcangeli int mm_take_all_locks(struct mm_struct *mm) 35287906d00cSAndrea Arcangeli { 35297906d00cSAndrea Arcangeli struct vm_area_struct *vma; 35305beb4930SRik van Riel struct anon_vma_chain *avc; 35317906d00cSAndrea Arcangeli 3532325bca1fSRolf Eike Beer mmap_assert_write_locked(mm); 35337906d00cSAndrea Arcangeli 35347906d00cSAndrea Arcangeli mutex_lock(&mm_all_locks_mutex); 35357906d00cSAndrea Arcangeli 35367906d00cSAndrea Arcangeli for (vma = mm->mmap; vma; vma = vma->vm_next) { 35377906d00cSAndrea Arcangeli if (signal_pending(current)) 35387906d00cSAndrea Arcangeli goto out_unlock; 353988f306b6SKirill A. Shutemov if (vma->vm_file && vma->vm_file->f_mapping && 354088f306b6SKirill A. Shutemov is_vm_hugetlb_page(vma)) 354188f306b6SKirill A. Shutemov vm_lock_mapping(mm, vma->vm_file->f_mapping); 354288f306b6SKirill A. Shutemov } 354388f306b6SKirill A. Shutemov 354488f306b6SKirill A. Shutemov for (vma = mm->mmap; vma; vma = vma->vm_next) { 354588f306b6SKirill A. Shutemov if (signal_pending(current)) 354688f306b6SKirill A. Shutemov goto out_unlock; 354788f306b6SKirill A. Shutemov if (vma->vm_file && vma->vm_file->f_mapping && 354888f306b6SKirill A. Shutemov !is_vm_hugetlb_page(vma)) 3549454ed842SPeter Zijlstra vm_lock_mapping(mm, vma->vm_file->f_mapping); 35507906d00cSAndrea Arcangeli } 35517cd5a02fSPeter Zijlstra 35527cd5a02fSPeter Zijlstra for (vma = mm->mmap; vma; vma = vma->vm_next) { 35537cd5a02fSPeter Zijlstra if (signal_pending(current)) 35547cd5a02fSPeter Zijlstra goto out_unlock; 35557cd5a02fSPeter Zijlstra if (vma->anon_vma) 35565beb4930SRik van Riel list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) 35575beb4930SRik van Riel vm_lock_anon_vma(mm, avc->anon_vma); 35587cd5a02fSPeter Zijlstra } 35597cd5a02fSPeter Zijlstra 3560584cff54SKautuk Consul return 0; 35617906d00cSAndrea Arcangeli 35627906d00cSAndrea Arcangeli out_unlock: 35637906d00cSAndrea Arcangeli mm_drop_all_locks(mm); 3564584cff54SKautuk Consul return -EINTR; 35657906d00cSAndrea Arcangeli } 35667906d00cSAndrea Arcangeli 35677906d00cSAndrea Arcangeli static void vm_unlock_anon_vma(struct anon_vma *anon_vma) 35687906d00cSAndrea Arcangeli { 3569f808c13fSDavidlohr Bueso if (test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_root.rb_node)) { 35707906d00cSAndrea Arcangeli /* 35717906d00cSAndrea Arcangeli * The LSB of head.next can't change to 0 from under 35727906d00cSAndrea Arcangeli * us because we hold the mm_all_locks_mutex. 35737906d00cSAndrea Arcangeli * 35747906d00cSAndrea Arcangeli * We must however clear the bitflag before unlocking 3575bf181b9fSMichel Lespinasse * the vma so the users using the anon_vma->rb_root will 35767906d00cSAndrea Arcangeli * never see our bitflag. 35777906d00cSAndrea Arcangeli * 35787906d00cSAndrea Arcangeli * No need of atomic instructions here, head.next 35797906d00cSAndrea Arcangeli * can't change from under us until we release the 35805a505085SIngo Molnar * anon_vma->root->rwsem. 35817906d00cSAndrea Arcangeli */ 35827906d00cSAndrea Arcangeli if (!__test_and_clear_bit(0, (unsigned long *) 3583f808c13fSDavidlohr Bueso &anon_vma->root->rb_root.rb_root.rb_node)) 35847906d00cSAndrea Arcangeli BUG(); 358508b52706SKonstantin Khlebnikov anon_vma_unlock_write(anon_vma); 35867906d00cSAndrea Arcangeli } 35877906d00cSAndrea Arcangeli } 35887906d00cSAndrea Arcangeli 35897906d00cSAndrea Arcangeli static void vm_unlock_mapping(struct address_space *mapping) 35907906d00cSAndrea Arcangeli { 35917906d00cSAndrea Arcangeli if (test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) { 35927906d00cSAndrea Arcangeli /* 35937906d00cSAndrea Arcangeli * AS_MM_ALL_LOCKS can't change to 0 from under us 35947906d00cSAndrea Arcangeli * because we hold the mm_all_locks_mutex. 35957906d00cSAndrea Arcangeli */ 359683cde9e8SDavidlohr Bueso i_mmap_unlock_write(mapping); 35977906d00cSAndrea Arcangeli if (!test_and_clear_bit(AS_MM_ALL_LOCKS, 35987906d00cSAndrea Arcangeli &mapping->flags)) 35997906d00cSAndrea Arcangeli BUG(); 36007906d00cSAndrea Arcangeli } 36017906d00cSAndrea Arcangeli } 36027906d00cSAndrea Arcangeli 36037906d00cSAndrea Arcangeli /* 3604c1e8d7c6SMichel Lespinasse * The mmap_lock cannot be released by the caller until 36057906d00cSAndrea Arcangeli * mm_drop_all_locks() returns. 36067906d00cSAndrea Arcangeli */ 36077906d00cSAndrea Arcangeli void mm_drop_all_locks(struct mm_struct *mm) 36087906d00cSAndrea Arcangeli { 36097906d00cSAndrea Arcangeli struct vm_area_struct *vma; 36105beb4930SRik van Riel struct anon_vma_chain *avc; 36117906d00cSAndrea Arcangeli 3612325bca1fSRolf Eike Beer mmap_assert_write_locked(mm); 36137906d00cSAndrea Arcangeli BUG_ON(!mutex_is_locked(&mm_all_locks_mutex)); 36147906d00cSAndrea Arcangeli 36157906d00cSAndrea Arcangeli for (vma = mm->mmap; vma; vma = vma->vm_next) { 36167906d00cSAndrea Arcangeli if (vma->anon_vma) 36175beb4930SRik van Riel list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) 36185beb4930SRik van Riel vm_unlock_anon_vma(avc->anon_vma); 36197906d00cSAndrea Arcangeli if (vma->vm_file && vma->vm_file->f_mapping) 36207906d00cSAndrea Arcangeli vm_unlock_mapping(vma->vm_file->f_mapping); 36217906d00cSAndrea Arcangeli } 36227906d00cSAndrea Arcangeli 36237906d00cSAndrea Arcangeli mutex_unlock(&mm_all_locks_mutex); 36247906d00cSAndrea Arcangeli } 36258feae131SDavid Howells 36268feae131SDavid Howells /* 36273edf41d8Sseokhoon.yoon * initialise the percpu counter for VM 36288feae131SDavid Howells */ 36298feae131SDavid Howells void __init mmap_init(void) 36308feae131SDavid Howells { 363100a62ce9SKOSAKI Motohiro int ret; 363200a62ce9SKOSAKI Motohiro 3633908c7f19STejun Heo ret = percpu_counter_init(&vm_committed_as, 0, GFP_KERNEL); 363400a62ce9SKOSAKI Motohiro VM_BUG_ON(ret); 36358feae131SDavid Howells } 3636c9b1d098SAndrew Shewmaker 3637c9b1d098SAndrew Shewmaker /* 3638c9b1d098SAndrew Shewmaker * Initialise sysctl_user_reserve_kbytes. 3639c9b1d098SAndrew Shewmaker * 3640c9b1d098SAndrew Shewmaker * This is intended to prevent a user from starting a single memory hogging 3641c9b1d098SAndrew Shewmaker * process, such that they cannot recover (kill the hog) in OVERCOMMIT_NEVER 3642c9b1d098SAndrew Shewmaker * mode. 3643c9b1d098SAndrew Shewmaker * 3644c9b1d098SAndrew Shewmaker * The default value is min(3% of free memory, 128MB) 3645c9b1d098SAndrew Shewmaker * 128MB is enough to recover with sshd/login, bash, and top/kill. 3646c9b1d098SAndrew Shewmaker */ 36471640879aSAndrew Shewmaker static int init_user_reserve(void) 3648c9b1d098SAndrew Shewmaker { 3649c9b1d098SAndrew Shewmaker unsigned long free_kbytes; 3650c9b1d098SAndrew Shewmaker 3651c41f012aSMichal Hocko free_kbytes = global_zone_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10); 3652c9b1d098SAndrew Shewmaker 3653c9b1d098SAndrew Shewmaker sysctl_user_reserve_kbytes = min(free_kbytes / 32, 1UL << 17); 3654c9b1d098SAndrew Shewmaker return 0; 3655c9b1d098SAndrew Shewmaker } 3656a64fb3cdSPaul Gortmaker subsys_initcall(init_user_reserve); 36574eeab4f5SAndrew Shewmaker 36584eeab4f5SAndrew Shewmaker /* 36594eeab4f5SAndrew Shewmaker * Initialise sysctl_admin_reserve_kbytes. 36604eeab4f5SAndrew Shewmaker * 36614eeab4f5SAndrew Shewmaker * The purpose of sysctl_admin_reserve_kbytes is to allow the sys admin 36624eeab4f5SAndrew Shewmaker * to log in and kill a memory hogging process. 36634eeab4f5SAndrew Shewmaker * 36644eeab4f5SAndrew Shewmaker * Systems with more than 256MB will reserve 8MB, enough to recover 36654eeab4f5SAndrew Shewmaker * with sshd, bash, and top in OVERCOMMIT_GUESS. Smaller systems will 36664eeab4f5SAndrew Shewmaker * only reserve 3% of free pages by default. 36674eeab4f5SAndrew Shewmaker */ 36681640879aSAndrew Shewmaker static int init_admin_reserve(void) 36694eeab4f5SAndrew Shewmaker { 36704eeab4f5SAndrew Shewmaker unsigned long free_kbytes; 36714eeab4f5SAndrew Shewmaker 3672c41f012aSMichal Hocko free_kbytes = global_zone_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10); 36734eeab4f5SAndrew Shewmaker 36744eeab4f5SAndrew Shewmaker sysctl_admin_reserve_kbytes = min(free_kbytes / 32, 1UL << 13); 36754eeab4f5SAndrew Shewmaker return 0; 36764eeab4f5SAndrew Shewmaker } 3677a64fb3cdSPaul Gortmaker subsys_initcall(init_admin_reserve); 36781640879aSAndrew Shewmaker 36791640879aSAndrew Shewmaker /* 36801640879aSAndrew Shewmaker * Reinititalise user and admin reserves if memory is added or removed. 36811640879aSAndrew Shewmaker * 36821640879aSAndrew Shewmaker * The default user reserve max is 128MB, and the default max for the 36831640879aSAndrew Shewmaker * admin reserve is 8MB. These are usually, but not always, enough to 36841640879aSAndrew Shewmaker * enable recovery from a memory hogging process using login/sshd, a shell, 36851640879aSAndrew Shewmaker * and tools like top. It may make sense to increase or even disable the 36861640879aSAndrew Shewmaker * reserve depending on the existence of swap or variations in the recovery 36871640879aSAndrew Shewmaker * tools. So, the admin may have changed them. 36881640879aSAndrew Shewmaker * 36891640879aSAndrew Shewmaker * If memory is added and the reserves have been eliminated or increased above 36901640879aSAndrew Shewmaker * the default max, then we'll trust the admin. 36911640879aSAndrew Shewmaker * 36921640879aSAndrew Shewmaker * If memory is removed and there isn't enough free memory, then we 36931640879aSAndrew Shewmaker * need to reset the reserves. 36941640879aSAndrew Shewmaker * 36951640879aSAndrew Shewmaker * Otherwise keep the reserve set by the admin. 36961640879aSAndrew Shewmaker */ 36971640879aSAndrew Shewmaker static int reserve_mem_notifier(struct notifier_block *nb, 36981640879aSAndrew Shewmaker unsigned long action, void *data) 36991640879aSAndrew Shewmaker { 37001640879aSAndrew Shewmaker unsigned long tmp, free_kbytes; 37011640879aSAndrew Shewmaker 37021640879aSAndrew Shewmaker switch (action) { 37031640879aSAndrew Shewmaker case MEM_ONLINE: 37041640879aSAndrew Shewmaker /* Default max is 128MB. Leave alone if modified by operator. */ 37051640879aSAndrew Shewmaker tmp = sysctl_user_reserve_kbytes; 37061640879aSAndrew Shewmaker if (0 < tmp && tmp < (1UL << 17)) 37071640879aSAndrew Shewmaker init_user_reserve(); 37081640879aSAndrew Shewmaker 37091640879aSAndrew Shewmaker /* Default max is 8MB. Leave alone if modified by operator. */ 37101640879aSAndrew Shewmaker tmp = sysctl_admin_reserve_kbytes; 37111640879aSAndrew Shewmaker if (0 < tmp && tmp < (1UL << 13)) 37121640879aSAndrew Shewmaker init_admin_reserve(); 37131640879aSAndrew Shewmaker 37141640879aSAndrew Shewmaker break; 37151640879aSAndrew Shewmaker case MEM_OFFLINE: 3716c41f012aSMichal Hocko free_kbytes = global_zone_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10); 37171640879aSAndrew Shewmaker 37181640879aSAndrew Shewmaker if (sysctl_user_reserve_kbytes > free_kbytes) { 37191640879aSAndrew Shewmaker init_user_reserve(); 37201640879aSAndrew Shewmaker pr_info("vm.user_reserve_kbytes reset to %lu\n", 37211640879aSAndrew Shewmaker sysctl_user_reserve_kbytes); 37221640879aSAndrew Shewmaker } 37231640879aSAndrew Shewmaker 37241640879aSAndrew Shewmaker if (sysctl_admin_reserve_kbytes > free_kbytes) { 37251640879aSAndrew Shewmaker init_admin_reserve(); 37261640879aSAndrew Shewmaker pr_info("vm.admin_reserve_kbytes reset to %lu\n", 37271640879aSAndrew Shewmaker sysctl_admin_reserve_kbytes); 37281640879aSAndrew Shewmaker } 37291640879aSAndrew Shewmaker break; 37301640879aSAndrew Shewmaker default: 37311640879aSAndrew Shewmaker break; 37321640879aSAndrew Shewmaker } 37331640879aSAndrew Shewmaker return NOTIFY_OK; 37341640879aSAndrew Shewmaker } 37351640879aSAndrew Shewmaker 37361640879aSAndrew Shewmaker static struct notifier_block reserve_mem_nb = { 37371640879aSAndrew Shewmaker .notifier_call = reserve_mem_notifier, 37381640879aSAndrew Shewmaker }; 37391640879aSAndrew Shewmaker 37401640879aSAndrew Shewmaker static int __meminit init_reserve_notifier(void) 37411640879aSAndrew Shewmaker { 37421640879aSAndrew Shewmaker if (register_hotmemory_notifier(&reserve_mem_nb)) 3743b1de0d13SMitchel Humpherys pr_err("Failed registering memory add/remove notifier for admin reserve\n"); 37441640879aSAndrew Shewmaker 37451640879aSAndrew Shewmaker return 0; 37461640879aSAndrew Shewmaker } 3747a64fb3cdSPaul Gortmaker subsys_initcall(init_reserve_notifier); 3748