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> 171da177e4SLinus Torvalds #include <linux/shm.h> 181da177e4SLinus Torvalds #include <linux/mman.h> 191da177e4SLinus Torvalds #include <linux/pagemap.h> 201da177e4SLinus Torvalds #include <linux/swap.h> 211da177e4SLinus Torvalds #include <linux/syscalls.h> 22c59ede7bSRandy.Dunlap #include <linux/capability.h> 231da177e4SLinus Torvalds #include <linux/init.h> 241da177e4SLinus Torvalds #include <linux/file.h> 251da177e4SLinus Torvalds #include <linux/fs.h> 261da177e4SLinus Torvalds #include <linux/personality.h> 271da177e4SLinus Torvalds #include <linux/security.h> 281da177e4SLinus Torvalds #include <linux/hugetlb.h> 29c01d5b30SHugh Dickins #include <linux/shmem_fs.h> 301da177e4SLinus Torvalds #include <linux/profile.h> 31b95f1b31SPaul Gortmaker #include <linux/export.h> 321da177e4SLinus Torvalds #include <linux/mount.h> 331da177e4SLinus Torvalds #include <linux/mempolicy.h> 341da177e4SLinus Torvalds #include <linux/rmap.h> 35cddb8a5cSAndrea Arcangeli #include <linux/mmu_notifier.h> 3682f71ae4SKonstantin Khlebnikov #include <linux/mmdebug.h> 37cdd6c482SIngo Molnar #include <linux/perf_event.h> 38120a795dSAl Viro #include <linux/audit.h> 39b15d00b6SAndrea Arcangeli #include <linux/khugepaged.h> 402b144498SSrikar Dronamraju #include <linux/uprobes.h> 411640879aSAndrew Shewmaker #include <linux/notifier.h> 421640879aSAndrew Shewmaker #include <linux/memory.h> 43b1de0d13SMitchel Humpherys #include <linux/printk.h> 4419a809afSAndrea Arcangeli #include <linux/userfaultfd_k.h> 45d977d56cSKonstantin Khlebnikov #include <linux/moduleparam.h> 4662b5f7d0SDave Hansen #include <linux/pkeys.h> 4721292580SAndrea Arcangeli #include <linux/oom.h> 4804f5866eSAndrea Arcangeli #include <linux/sched/mm.h> 49d7597f59SStefan Roesch #include <linux/ksm.h> 501da177e4SLinus Torvalds 517c0f6ba6SLinus Torvalds #include <linux/uaccess.h> 521da177e4SLinus Torvalds #include <asm/cacheflush.h> 531da177e4SLinus Torvalds #include <asm/tlb.h> 54d6dd61c8SJeremy Fitzhardinge #include <asm/mmu_context.h> 551da177e4SLinus Torvalds 56df529cabSJaewon Kim #define CREATE_TRACE_POINTS 57df529cabSJaewon Kim #include <trace/events/mmap.h> 58df529cabSJaewon Kim 5942b77728SJan Beulich #include "internal.h" 6042b77728SJan Beulich 613a459756SKirill Korotaev #ifndef arch_mmap_check 623a459756SKirill Korotaev #define arch_mmap_check(addr, len, flags) (0) 633a459756SKirill Korotaev #endif 643a459756SKirill Korotaev 65d07e2259SDaniel Cashman #ifdef CONFIG_HAVE_ARCH_MMAP_RND_BITS 66d07e2259SDaniel Cashman const int mmap_rnd_bits_min = CONFIG_ARCH_MMAP_RND_BITS_MIN; 67d07e2259SDaniel Cashman const int mmap_rnd_bits_max = CONFIG_ARCH_MMAP_RND_BITS_MAX; 68d07e2259SDaniel Cashman int mmap_rnd_bits __read_mostly = CONFIG_ARCH_MMAP_RND_BITS; 69d07e2259SDaniel Cashman #endif 70d07e2259SDaniel Cashman #ifdef CONFIG_HAVE_ARCH_MMAP_RND_COMPAT_BITS 71d07e2259SDaniel Cashman const int mmap_rnd_compat_bits_min = CONFIG_ARCH_MMAP_RND_COMPAT_BITS_MIN; 72d07e2259SDaniel Cashman const int mmap_rnd_compat_bits_max = CONFIG_ARCH_MMAP_RND_COMPAT_BITS_MAX; 73d07e2259SDaniel Cashman int mmap_rnd_compat_bits __read_mostly = CONFIG_ARCH_MMAP_RND_COMPAT_BITS; 74d07e2259SDaniel Cashman #endif 75d07e2259SDaniel Cashman 76f4fcd558SKonstantin Khlebnikov static bool ignore_rlimit_data; 77d977d56cSKonstantin Khlebnikov core_param(ignore_rlimit_data, ignore_rlimit_data, bool, 0644); 78d07e2259SDaniel Cashman 79763ecb03SLiam R. Howlett static void unmap_region(struct mm_struct *mm, struct maple_tree *mt, 80e0da382cSHugh Dickins struct vm_area_struct *vma, struct vm_area_struct *prev, 81763ecb03SLiam R. Howlett struct vm_area_struct *next, unsigned long start, 8268f48381SSuren Baghdasaryan unsigned long end, bool mm_wr_locked); 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 /* 135763ecb03SLiam R. Howlett * Close a vm structure and free it. 136a8fb5618SHugh Dickins */ 1370d2ebf9cSSuren Baghdasaryan static void remove_vma(struct vm_area_struct *vma, bool unreachable) 138a8fb5618SHugh Dickins { 139a8fb5618SHugh Dickins might_sleep(); 1401da177e4SLinus Torvalds if (vma->vm_ops && vma->vm_ops->close) 1411da177e4SLinus Torvalds vma->vm_ops->close(vma); 142e9714acfSKonstantin Khlebnikov if (vma->vm_file) 143a8fb5618SHugh Dickins fput(vma->vm_file); 144f0be3d32SLee Schermerhorn mpol_put(vma_policy(vma)); 1450d2ebf9cSSuren Baghdasaryan if (unreachable) 1460d2ebf9cSSuren Baghdasaryan __vm_area_free(vma); 1470d2ebf9cSSuren Baghdasaryan else 1483928d4f5SLinus Torvalds vm_area_free(vma); 1491da177e4SLinus Torvalds } 1501da177e4SLinus Torvalds 151b62b633eSLiam R. Howlett static inline struct vm_area_struct *vma_prev_limit(struct vma_iterator *vmi, 152b62b633eSLiam R. Howlett unsigned long min) 153b62b633eSLiam R. Howlett { 154b62b633eSLiam R. Howlett return mas_prev(&vmi->mas, min); 155b62b633eSLiam R. Howlett } 156b62b633eSLiam R. Howlett 157b62b633eSLiam R. Howlett static inline int vma_iter_clear_gfp(struct vma_iterator *vmi, 158b62b633eSLiam R. Howlett unsigned long start, unsigned long end, gfp_t gfp) 159b62b633eSLiam R. Howlett { 160b62b633eSLiam R. Howlett vmi->mas.index = start; 161b62b633eSLiam R. Howlett vmi->mas.last = end - 1; 162b62b633eSLiam R. Howlett mas_store_gfp(&vmi->mas, NULL, gfp); 163b62b633eSLiam R. Howlett if (unlikely(mas_is_err(&vmi->mas))) 164b62b633eSLiam R. Howlett return -ENOMEM; 165b62b633eSLiam R. Howlett 166b62b633eSLiam R. Howlett return 0; 167b62b633eSLiam R. Howlett } 168b62b633eSLiam R. Howlett 1692e7ce7d3SLiam R. Howlett /* 1702e7ce7d3SLiam R. Howlett * check_brk_limits() - Use platform specific check of range & verify mlock 1712e7ce7d3SLiam R. Howlett * limits. 1722e7ce7d3SLiam R. Howlett * @addr: The address to check 1732e7ce7d3SLiam R. Howlett * @len: The size of increase. 1742e7ce7d3SLiam R. Howlett * 1752e7ce7d3SLiam R. Howlett * Return: 0 on success. 1762e7ce7d3SLiam R. Howlett */ 1772e7ce7d3SLiam R. Howlett static int check_brk_limits(unsigned long addr, unsigned long len) 1782e7ce7d3SLiam R. Howlett { 1792e7ce7d3SLiam R. Howlett unsigned long mapped_addr; 1802e7ce7d3SLiam R. Howlett 1812e7ce7d3SLiam R. Howlett mapped_addr = get_unmapped_area(NULL, addr, len, 0, MAP_FIXED); 1822e7ce7d3SLiam R. Howlett if (IS_ERR_VALUE(mapped_addr)) 1832e7ce7d3SLiam R. Howlett return mapped_addr; 1842e7ce7d3SLiam R. Howlett 185b0cc5e89SAndrew Morton return mlock_future_ok(current->mm, current->mm->def_flags, len) 1863c54a298SLorenzo Stoakes ? 0 : -EAGAIN; 1872e7ce7d3SLiam R. Howlett } 18892fed820SLiam R. Howlett static int do_brk_flags(struct vma_iterator *vmi, struct vm_area_struct *brkvma, 189763ecb03SLiam R. Howlett unsigned long addr, unsigned long request, unsigned long flags); 1906a6160a7SHeiko Carstens SYSCALL_DEFINE1(brk, unsigned long, brk) 1911da177e4SLinus Torvalds { 1929bc8039eSYang Shi unsigned long newbrk, oldbrk, origbrk; 1931da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 1942e7ce7d3SLiam R. Howlett struct vm_area_struct *brkvma, *next = NULL; 195a5b4592cSJiri Kosina unsigned long min_brk; 196408579cdSLiam R. Howlett bool populate = false; 197897ab3e0SMike Rapoport LIST_HEAD(uf); 19892fed820SLiam R. Howlett struct vma_iterator vmi; 1991da177e4SLinus Torvalds 200d8ed45c5SMichel Lespinasse if (mmap_write_lock_killable(mm)) 201dc0ef0dfSMichal Hocko return -EINTR; 2021da177e4SLinus Torvalds 2039bc8039eSYang Shi origbrk = mm->brk; 2049bc8039eSYang Shi 205a5b4592cSJiri Kosina #ifdef CONFIG_COMPAT_BRK 2065520e894SJiri Kosina /* 2075520e894SJiri Kosina * CONFIG_COMPAT_BRK can still be overridden by setting 2085520e894SJiri Kosina * randomize_va_space to 2, which will still cause mm->start_brk 2095520e894SJiri Kosina * to be arbitrarily shifted 2105520e894SJiri Kosina */ 2114471a675SJiri Kosina if (current->brk_randomized) 2125520e894SJiri Kosina min_brk = mm->start_brk; 2135520e894SJiri Kosina else 2145520e894SJiri Kosina min_brk = mm->end_data; 215a5b4592cSJiri Kosina #else 216a5b4592cSJiri Kosina min_brk = mm->start_brk; 217a5b4592cSJiri Kosina #endif 218a5b4592cSJiri Kosina if (brk < min_brk) 2191da177e4SLinus Torvalds goto out; 2201e624196SRam Gupta 2211e624196SRam Gupta /* 2221e624196SRam Gupta * Check against rlimit here. If this check is done later after the test 2231e624196SRam Gupta * of oldbrk with newbrk then it can escape the test and let the data 2241e624196SRam Gupta * segment grow beyond its set limit the in case where the limit is 2251e624196SRam Gupta * not page aligned -Ram Gupta 2261e624196SRam Gupta */ 2278764b338SCyrill Gorcunov if (check_data_rlimit(rlimit(RLIMIT_DATA), brk, mm->start_brk, 2288764b338SCyrill Gorcunov mm->end_data, mm->start_data)) 2291e624196SRam Gupta goto out; 2301e624196SRam Gupta 2311da177e4SLinus Torvalds newbrk = PAGE_ALIGN(brk); 2321da177e4SLinus Torvalds oldbrk = PAGE_ALIGN(mm->brk); 2339bc8039eSYang Shi if (oldbrk == newbrk) { 2349bc8039eSYang Shi mm->brk = brk; 2359bc8039eSYang Shi goto success; 2369bc8039eSYang Shi } 2371da177e4SLinus Torvalds 238408579cdSLiam R. Howlett /* Always allow shrinking brk. */ 2391da177e4SLinus Torvalds if (brk <= mm->brk) { 2402e7ce7d3SLiam R. Howlett /* Search one past newbrk */ 24192fed820SLiam R. Howlett vma_iter_init(&vmi, mm, newbrk); 24292fed820SLiam R. Howlett brkvma = vma_find(&vmi, oldbrk); 243f5ad5083SJason A. Donenfeld if (!brkvma || brkvma->vm_start >= oldbrk) 2442e7ce7d3SLiam R. Howlett goto out; /* mapping intersects with an existing non-brk vma. */ 2459bc8039eSYang Shi /* 2462e7ce7d3SLiam R. Howlett * mm->brk must be protected by write mmap_lock. 247408579cdSLiam R. Howlett * do_vma_munmap() will drop the lock on success, so update it 24827b26701SLiam R. Howlett * before calling do_vma_munmap(). 2499bc8039eSYang Shi */ 2509bc8039eSYang Shi mm->brk = brk; 251408579cdSLiam R. Howlett if (do_vma_munmap(&vmi, brkvma, newbrk, oldbrk, &uf, true)) 2521da177e4SLinus Torvalds goto out; 253408579cdSLiam R. Howlett 254408579cdSLiam R. Howlett goto success_unlocked; 2551da177e4SLinus Torvalds } 2561da177e4SLinus Torvalds 2572e7ce7d3SLiam R. Howlett if (check_brk_limits(oldbrk, newbrk - oldbrk)) 2582e7ce7d3SLiam R. Howlett goto out; 2592e7ce7d3SLiam R. Howlett 2602e7ce7d3SLiam R. Howlett /* 2612e7ce7d3SLiam R. Howlett * Only check if the next VMA is within the stack_guard_gap of the 2622e7ce7d3SLiam R. Howlett * expansion area 2632e7ce7d3SLiam R. Howlett */ 26492fed820SLiam R. Howlett vma_iter_init(&vmi, mm, oldbrk); 26592fed820SLiam R. Howlett next = vma_find(&vmi, newbrk + PAGE_SIZE + stack_guard_gap); 2661be7107fSHugh Dickins if (next && newbrk + PAGE_SIZE > vm_start_gap(next)) 2671da177e4SLinus Torvalds goto out; 2681da177e4SLinus Torvalds 26992fed820SLiam R. Howlett brkvma = vma_prev_limit(&vmi, mm->start_brk); 2701da177e4SLinus Torvalds /* Ok, looks good - let it rip. */ 27192fed820SLiam R. Howlett if (do_brk_flags(&vmi, brkvma, oldbrk, newbrk - oldbrk, 0) < 0) 2721da177e4SLinus Torvalds goto out; 2732e7ce7d3SLiam R. Howlett 2741da177e4SLinus Torvalds mm->brk = brk; 275408579cdSLiam R. Howlett if (mm->def_flags & VM_LOCKED) 276408579cdSLiam R. Howlett populate = true; 2779bc8039eSYang Shi 2789bc8039eSYang Shi success: 279d8ed45c5SMichel Lespinasse mmap_write_unlock(mm); 280408579cdSLiam R. Howlett success_unlocked: 281897ab3e0SMike Rapoport userfaultfd_unmap_complete(mm, &uf); 282128557ffSMichel Lespinasse if (populate) 283128557ffSMichel Lespinasse mm_populate(oldbrk, newbrk - oldbrk); 284128557ffSMichel Lespinasse return brk; 285128557ffSMichel Lespinasse 2861da177e4SLinus Torvalds out: 287408579cdSLiam R. Howlett mm->brk = origbrk; 288d8ed45c5SMichel Lespinasse mmap_write_unlock(mm); 289b7204006SAdrian Huang return origbrk; 2901da177e4SLinus Torvalds } 2911da177e4SLinus Torvalds 292d4af56c5SLiam R. Howlett #if defined(CONFIG_DEBUG_VM_MAPLE_TREE) 293eafd4dc4SRashika Kheria static void validate_mm(struct mm_struct *mm) 2941da177e4SLinus Torvalds { 2951da177e4SLinus Torvalds int bug = 0; 2961da177e4SLinus Torvalds int i = 0; 297763ecb03SLiam R. Howlett struct vm_area_struct *vma; 298b50e195fSLiam R. Howlett VMA_ITERATOR(vmi, mm, 0); 299ff26f70fSAndrew Morton 300b50e195fSLiam R. Howlett mt_validate(&mm->mm_mt); 301b50e195fSLiam R. Howlett for_each_vma(vmi, vma) { 302524e00b3SLiam R. Howlett #ifdef CONFIG_DEBUG_VM_RB 30312352d3cSKonstantin Khlebnikov struct anon_vma *anon_vma = vma->anon_vma; 304ed8ea815SMichel Lespinasse struct anon_vma_chain *avc; 305b50e195fSLiam R. Howlett #endif 306b50e195fSLiam R. Howlett unsigned long vmi_start, vmi_end; 307b50e195fSLiam R. Howlett bool warn = 0; 308ff26f70fSAndrew Morton 309b50e195fSLiam R. Howlett vmi_start = vma_iter_addr(&vmi); 310b50e195fSLiam R. Howlett vmi_end = vma_iter_end(&vmi); 311b50e195fSLiam R. Howlett if (VM_WARN_ON_ONCE_MM(vma->vm_end != vmi_end, mm)) 312b50e195fSLiam R. Howlett warn = 1; 313b50e195fSLiam R. Howlett 314b50e195fSLiam R. Howlett if (VM_WARN_ON_ONCE_MM(vma->vm_start != vmi_start, mm)) 315b50e195fSLiam R. Howlett warn = 1; 316b50e195fSLiam R. Howlett 317b50e195fSLiam R. Howlett if (warn) { 318b50e195fSLiam R. Howlett pr_emerg("issue in %s\n", current->comm); 319b50e195fSLiam R. Howlett dump_stack(); 320b50e195fSLiam R. Howlett dump_vma(vma); 321b50e195fSLiam R. Howlett pr_emerg("tree range: %px start %lx end %lx\n", vma, 322b50e195fSLiam R. Howlett vmi_start, vmi_end - 1); 323b50e195fSLiam R. Howlett vma_iter_dump_tree(&vmi); 324b50e195fSLiam R. Howlett } 325b50e195fSLiam R. Howlett 326b50e195fSLiam R. Howlett #ifdef CONFIG_DEBUG_VM_RB 32712352d3cSKonstantin Khlebnikov if (anon_vma) { 32812352d3cSKonstantin Khlebnikov anon_vma_lock_read(anon_vma); 329ed8ea815SMichel Lespinasse list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) 330ed8ea815SMichel Lespinasse anon_vma_interval_tree_verify(avc); 33112352d3cSKonstantin Khlebnikov anon_vma_unlock_read(anon_vma); 33212352d3cSKonstantin Khlebnikov } 333524e00b3SLiam R. Howlett #endif 3341da177e4SLinus Torvalds i++; 3351da177e4SLinus Torvalds } 3365a0768f6SMichel Lespinasse if (i != mm->map_count) { 337b50e195fSLiam R. Howlett pr_emerg("map_count %d vma iterator %d\n", mm->map_count, i); 3385a0768f6SMichel Lespinasse bug = 1; 3395a0768f6SMichel Lespinasse } 34096dad67fSSasha Levin VM_BUG_ON_MM(bug, mm); 3411da177e4SLinus Torvalds } 342524e00b3SLiam R. Howlett 343524e00b3SLiam R. Howlett #else /* !CONFIG_DEBUG_VM_MAPLE_TREE */ 3441da177e4SLinus Torvalds #define validate_mm(mm) do { } while (0) 345524e00b3SLiam R. Howlett #endif /* CONFIG_DEBUG_VM_MAPLE_TREE */ 346d3737187SMichel Lespinasse 347bf181b9fSMichel Lespinasse /* 348bf181b9fSMichel Lespinasse * vma has some anon_vma assigned, and is already inserted on that 349bf181b9fSMichel Lespinasse * anon_vma's interval trees. 350bf181b9fSMichel Lespinasse * 351bf181b9fSMichel Lespinasse * Before updating the vma's vm_start / vm_end / vm_pgoff fields, the 352bf181b9fSMichel Lespinasse * vma must be removed from the anon_vma's interval trees using 353bf181b9fSMichel Lespinasse * anon_vma_interval_tree_pre_update_vma(). 354bf181b9fSMichel Lespinasse * 355bf181b9fSMichel Lespinasse * After the update, the vma will be reinserted using 356bf181b9fSMichel Lespinasse * anon_vma_interval_tree_post_update_vma(). 357bf181b9fSMichel Lespinasse * 358c1e8d7c6SMichel Lespinasse * The entire update must be protected by exclusive mmap_lock and by 359bf181b9fSMichel Lespinasse * the root anon_vma's mutex. 360bf181b9fSMichel Lespinasse */ 361bf181b9fSMichel Lespinasse static inline void 362bf181b9fSMichel Lespinasse anon_vma_interval_tree_pre_update_vma(struct vm_area_struct *vma) 363bf181b9fSMichel Lespinasse { 364bf181b9fSMichel Lespinasse struct anon_vma_chain *avc; 365bf181b9fSMichel Lespinasse 366bf181b9fSMichel Lespinasse list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) 367bf181b9fSMichel Lespinasse anon_vma_interval_tree_remove(avc, &avc->anon_vma->rb_root); 368bf181b9fSMichel Lespinasse } 369bf181b9fSMichel Lespinasse 370bf181b9fSMichel Lespinasse static inline void 371bf181b9fSMichel Lespinasse anon_vma_interval_tree_post_update_vma(struct vm_area_struct *vma) 372bf181b9fSMichel Lespinasse { 373bf181b9fSMichel Lespinasse struct anon_vma_chain *avc; 374bf181b9fSMichel Lespinasse 375bf181b9fSMichel Lespinasse list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) 376bf181b9fSMichel Lespinasse anon_vma_interval_tree_insert(avc, &avc->anon_vma->rb_root); 377bf181b9fSMichel Lespinasse } 378bf181b9fSMichel Lespinasse 379e8420a8eSCyril Hrubis static unsigned long count_vma_pages_range(struct mm_struct *mm, 380e8420a8eSCyril Hrubis unsigned long addr, unsigned long end) 381e8420a8eSCyril Hrubis { 3822e3af1dbSMatthew Wilcox (Oracle) VMA_ITERATOR(vmi, mm, addr); 383e8420a8eSCyril Hrubis struct vm_area_struct *vma; 3842e3af1dbSMatthew Wilcox (Oracle) unsigned long nr_pages = 0; 385e8420a8eSCyril Hrubis 3862e3af1dbSMatthew Wilcox (Oracle) for_each_vma_range(vmi, vma, end) { 3872e3af1dbSMatthew Wilcox (Oracle) unsigned long vm_start = max(addr, vma->vm_start); 3882e3af1dbSMatthew Wilcox (Oracle) unsigned long vm_end = min(end, vma->vm_end); 389e8420a8eSCyril Hrubis 3902e3af1dbSMatthew Wilcox (Oracle) nr_pages += PHYS_PFN(vm_end - vm_start); 391e8420a8eSCyril Hrubis } 392e8420a8eSCyril Hrubis 393e8420a8eSCyril Hrubis return nr_pages; 394e8420a8eSCyril Hrubis } 395e8420a8eSCyril Hrubis 396c154124fSLiam R. Howlett static void __vma_link_file(struct vm_area_struct *vma, 397c154124fSLiam R. Howlett struct address_space *mapping) 3981da177e4SLinus Torvalds { 3991da177e4SLinus Torvalds if (vma->vm_flags & VM_SHARED) 400cf508b58SMiaohe Lin mapping_allow_writable(mapping); 4011da177e4SLinus Torvalds 4021da177e4SLinus Torvalds flush_dcache_mmap_lock(mapping); 4036b2dbba8SMichel Lespinasse vma_interval_tree_insert(vma, &mapping->i_mmap); 4041da177e4SLinus Torvalds flush_dcache_mmap_unlock(mapping); 4051da177e4SLinus Torvalds } 4061da177e4SLinus Torvalds 407763ecb03SLiam R. Howlett static int vma_link(struct mm_struct *mm, struct vm_area_struct *vma) 4081da177e4SLinus Torvalds { 40979e4f2caSLiam R. Howlett VMA_ITERATOR(vmi, mm, 0); 4101da177e4SLinus Torvalds struct address_space *mapping = NULL; 4111da177e4SLinus Torvalds 41279e4f2caSLiam R. Howlett if (vma_iter_prealloc(&vmi)) 413d4af56c5SLiam R. Howlett return -ENOMEM; 414d4af56c5SLiam R. Howlett 41564ac4940SHuang Shijie if (vma->vm_file) { 4161da177e4SLinus Torvalds mapping = vma->vm_file->f_mapping; 41783cde9e8SDavidlohr Bueso i_mmap_lock_write(mapping); 41864ac4940SHuang Shijie } 4191da177e4SLinus Torvalds 42079e4f2caSLiam R. Howlett vma_iter_store(&vmi, vma); 4211da177e4SLinus Torvalds 422c154124fSLiam R. Howlett if (mapping) { 423c154124fSLiam R. Howlett __vma_link_file(vma, mapping); 42483cde9e8SDavidlohr Bueso i_mmap_unlock_write(mapping); 425c154124fSLiam R. Howlett } 4261da177e4SLinus Torvalds 4271da177e4SLinus Torvalds mm->map_count++; 4281da177e4SLinus Torvalds validate_mm(mm); 429d4af56c5SLiam R. Howlett return 0; 4301da177e4SLinus Torvalds } 4311da177e4SLinus Torvalds 4321da177e4SLinus Torvalds /* 43368cefec5SLiam R. Howlett * init_multi_vma_prep() - Initializer for struct vma_prepare 43468cefec5SLiam R. Howlett * @vp: The vma_prepare struct 43568cefec5SLiam R. Howlett * @vma: The vma that will be altered once locked 43668cefec5SLiam R. Howlett * @next: The next vma if it is to be adjusted 43768cefec5SLiam R. Howlett * @remove: The first vma to be removed 43868cefec5SLiam R. Howlett * @remove2: The second vma to be removed 43968cefec5SLiam R. Howlett */ 44068cefec5SLiam R. Howlett static inline void init_multi_vma_prep(struct vma_prepare *vp, 44168cefec5SLiam R. Howlett struct vm_area_struct *vma, struct vm_area_struct *next, 44268cefec5SLiam R. Howlett struct vm_area_struct *remove, struct vm_area_struct *remove2) 44368cefec5SLiam R. Howlett { 44468cefec5SLiam R. Howlett memset(vp, 0, sizeof(struct vma_prepare)); 44568cefec5SLiam R. Howlett vp->vma = vma; 44668cefec5SLiam R. Howlett vp->anon_vma = vma->anon_vma; 44768cefec5SLiam R. Howlett vp->remove = remove; 44868cefec5SLiam R. Howlett vp->remove2 = remove2; 44968cefec5SLiam R. Howlett vp->adj_next = next; 45068cefec5SLiam R. Howlett if (!vp->anon_vma && next) 45168cefec5SLiam R. Howlett vp->anon_vma = next->anon_vma; 45268cefec5SLiam R. Howlett 45368cefec5SLiam R. Howlett vp->file = vma->vm_file; 45468cefec5SLiam R. Howlett if (vp->file) 45568cefec5SLiam R. Howlett vp->mapping = vma->vm_file->f_mapping; 45668cefec5SLiam R. Howlett 45768cefec5SLiam R. Howlett } 45868cefec5SLiam R. Howlett 45968cefec5SLiam R. Howlett /* 46068cefec5SLiam R. Howlett * init_vma_prep() - Initializer wrapper for vma_prepare struct 46168cefec5SLiam R. Howlett * @vp: The vma_prepare struct 46268cefec5SLiam R. Howlett * @vma: The vma that will be altered once locked 46368cefec5SLiam R. Howlett */ 46468cefec5SLiam R. Howlett static inline void init_vma_prep(struct vma_prepare *vp, 46568cefec5SLiam R. Howlett struct vm_area_struct *vma) 46668cefec5SLiam R. Howlett { 46768cefec5SLiam R. Howlett init_multi_vma_prep(vp, vma, NULL, NULL, NULL); 46868cefec5SLiam R. Howlett } 46968cefec5SLiam R. Howlett 47068cefec5SLiam R. Howlett 47168cefec5SLiam R. Howlett /* 472440703e0SLiam R. Howlett * vma_prepare() - Helper function for handling locking VMAs prior to altering 473440703e0SLiam R. Howlett * @vp: The initialized vma_prepare struct 474440703e0SLiam R. Howlett */ 475440703e0SLiam R. Howlett static inline void vma_prepare(struct vma_prepare *vp) 476440703e0SLiam R. Howlett { 47710fca64aSSuren Baghdasaryan vma_start_write(vp->vma); 47810fca64aSSuren Baghdasaryan if (vp->adj_next) 47910fca64aSSuren Baghdasaryan vma_start_write(vp->adj_next); 48010fca64aSSuren Baghdasaryan /* vp->insert is always a newly created VMA, no need for locking */ 48110fca64aSSuren Baghdasaryan if (vp->remove) 48210fca64aSSuren Baghdasaryan vma_start_write(vp->remove); 48310fca64aSSuren Baghdasaryan if (vp->remove2) 48410fca64aSSuren Baghdasaryan vma_start_write(vp->remove2); 48510fca64aSSuren Baghdasaryan 486440703e0SLiam R. Howlett if (vp->file) { 487440703e0SLiam R. Howlett uprobe_munmap(vp->vma, vp->vma->vm_start, vp->vma->vm_end); 488440703e0SLiam R. Howlett 489440703e0SLiam R. Howlett if (vp->adj_next) 490440703e0SLiam R. Howlett uprobe_munmap(vp->adj_next, vp->adj_next->vm_start, 491440703e0SLiam R. Howlett vp->adj_next->vm_end); 492440703e0SLiam R. Howlett 493440703e0SLiam R. Howlett i_mmap_lock_write(vp->mapping); 494440703e0SLiam R. Howlett if (vp->insert && vp->insert->vm_file) { 495440703e0SLiam R. Howlett /* 496440703e0SLiam R. Howlett * Put into interval tree now, so instantiated pages 497440703e0SLiam R. Howlett * are visible to arm/parisc __flush_dcache_page 498440703e0SLiam R. Howlett * throughout; but we cannot insert into address 499440703e0SLiam R. Howlett * space until vma start or end is updated. 500440703e0SLiam R. Howlett */ 501440703e0SLiam R. Howlett __vma_link_file(vp->insert, 502440703e0SLiam R. Howlett vp->insert->vm_file->f_mapping); 503440703e0SLiam R. Howlett } 504440703e0SLiam R. Howlett } 505440703e0SLiam R. Howlett 506440703e0SLiam R. Howlett if (vp->anon_vma) { 507440703e0SLiam R. Howlett anon_vma_lock_write(vp->anon_vma); 508440703e0SLiam R. Howlett anon_vma_interval_tree_pre_update_vma(vp->vma); 509440703e0SLiam R. Howlett if (vp->adj_next) 510440703e0SLiam R. Howlett anon_vma_interval_tree_pre_update_vma(vp->adj_next); 511440703e0SLiam R. Howlett } 512440703e0SLiam R. Howlett 513440703e0SLiam R. Howlett if (vp->file) { 514440703e0SLiam R. Howlett flush_dcache_mmap_lock(vp->mapping); 515440703e0SLiam R. Howlett vma_interval_tree_remove(vp->vma, &vp->mapping->i_mmap); 516440703e0SLiam R. Howlett if (vp->adj_next) 517440703e0SLiam R. Howlett vma_interval_tree_remove(vp->adj_next, 518440703e0SLiam R. Howlett &vp->mapping->i_mmap); 519440703e0SLiam R. Howlett } 520440703e0SLiam R. Howlett 521440703e0SLiam R. Howlett } 522440703e0SLiam R. Howlett 523440703e0SLiam R. Howlett /* 524440703e0SLiam R. Howlett * vma_complete- Helper function for handling the unlocking after altering VMAs, 525440703e0SLiam R. Howlett * or for inserting a VMA. 526440703e0SLiam R. Howlett * 527440703e0SLiam R. Howlett * @vp: The vma_prepare struct 528440703e0SLiam R. Howlett * @vmi: The vma iterator 529440703e0SLiam R. Howlett * @mm: The mm_struct 530440703e0SLiam R. Howlett */ 531440703e0SLiam R. Howlett static inline void vma_complete(struct vma_prepare *vp, 532440703e0SLiam R. Howlett struct vma_iterator *vmi, struct mm_struct *mm) 533440703e0SLiam R. Howlett { 534440703e0SLiam R. Howlett if (vp->file) { 535440703e0SLiam R. Howlett if (vp->adj_next) 536440703e0SLiam R. Howlett vma_interval_tree_insert(vp->adj_next, 537440703e0SLiam R. Howlett &vp->mapping->i_mmap); 538440703e0SLiam R. Howlett vma_interval_tree_insert(vp->vma, &vp->mapping->i_mmap); 539440703e0SLiam R. Howlett flush_dcache_mmap_unlock(vp->mapping); 540440703e0SLiam R. Howlett } 541440703e0SLiam R. Howlett 542440703e0SLiam R. Howlett if (vp->remove && vp->file) { 543440703e0SLiam R. Howlett __remove_shared_vm_struct(vp->remove, vp->file, vp->mapping); 544440703e0SLiam R. Howlett if (vp->remove2) 545440703e0SLiam R. Howlett __remove_shared_vm_struct(vp->remove2, vp->file, 546440703e0SLiam R. Howlett vp->mapping); 547440703e0SLiam R. Howlett } else if (vp->insert) { 548440703e0SLiam R. Howlett /* 549440703e0SLiam R. Howlett * split_vma has split insert from vma, and needs 550440703e0SLiam R. Howlett * us to insert it before dropping the locks 551440703e0SLiam R. Howlett * (it may either follow vma or precede it). 552440703e0SLiam R. Howlett */ 553440703e0SLiam R. Howlett vma_iter_store(vmi, vp->insert); 554440703e0SLiam R. Howlett mm->map_count++; 555440703e0SLiam R. Howlett } 556440703e0SLiam R. Howlett 557440703e0SLiam R. Howlett if (vp->anon_vma) { 558440703e0SLiam R. Howlett anon_vma_interval_tree_post_update_vma(vp->vma); 559440703e0SLiam R. Howlett if (vp->adj_next) 560440703e0SLiam R. Howlett anon_vma_interval_tree_post_update_vma(vp->adj_next); 561440703e0SLiam R. Howlett anon_vma_unlock_write(vp->anon_vma); 562440703e0SLiam R. Howlett } 563440703e0SLiam R. Howlett 564440703e0SLiam R. Howlett if (vp->file) { 565440703e0SLiam R. Howlett i_mmap_unlock_write(vp->mapping); 566440703e0SLiam R. Howlett uprobe_mmap(vp->vma); 567440703e0SLiam R. Howlett 568440703e0SLiam R. Howlett if (vp->adj_next) 569440703e0SLiam R. Howlett uprobe_mmap(vp->adj_next); 570440703e0SLiam R. Howlett } 571440703e0SLiam R. Howlett 572440703e0SLiam R. Howlett if (vp->remove) { 573440703e0SLiam R. Howlett again: 574457f67beSSuren Baghdasaryan vma_mark_detached(vp->remove, true); 575440703e0SLiam R. Howlett if (vp->file) { 576440703e0SLiam R. Howlett uprobe_munmap(vp->remove, vp->remove->vm_start, 577440703e0SLiam R. Howlett vp->remove->vm_end); 578440703e0SLiam R. Howlett fput(vp->file); 579440703e0SLiam R. Howlett } 580440703e0SLiam R. Howlett if (vp->remove->anon_vma) 581440703e0SLiam R. Howlett anon_vma_merge(vp->vma, vp->remove); 582440703e0SLiam R. Howlett mm->map_count--; 583440703e0SLiam R. Howlett mpol_put(vma_policy(vp->remove)); 584440703e0SLiam R. Howlett if (!vp->remove2) 585440703e0SLiam R. Howlett WARN_ON_ONCE(vp->vma->vm_end < vp->remove->vm_end); 586440703e0SLiam R. Howlett vm_area_free(vp->remove); 587440703e0SLiam R. Howlett 588440703e0SLiam R. Howlett /* 589440703e0SLiam R. Howlett * In mprotect's case 6 (see comments on vma_merge), 5905ff783f1SVlastimil Babka * we are removing both mid and next vmas 591440703e0SLiam R. Howlett */ 592440703e0SLiam R. Howlett if (vp->remove2) { 593440703e0SLiam R. Howlett vp->remove = vp->remove2; 594440703e0SLiam R. Howlett vp->remove2 = NULL; 595440703e0SLiam R. Howlett goto again; 596440703e0SLiam R. Howlett } 597440703e0SLiam R. Howlett } 598440703e0SLiam R. Howlett if (vp->insert && vp->file) 599440703e0SLiam R. Howlett uprobe_mmap(vp->insert); 600440703e0SLiam R. Howlett } 601440703e0SLiam R. Howlett 602440703e0SLiam R. Howlett /* 60304241ffeSLiam R. Howlett * dup_anon_vma() - Helper function to duplicate anon_vma 60404241ffeSLiam R. Howlett * @dst: The destination VMA 60504241ffeSLiam R. Howlett * @src: The source VMA 60604241ffeSLiam R. Howlett * 60704241ffeSLiam R. Howlett * Returns: 0 on success. 60804241ffeSLiam R. Howlett */ 60904241ffeSLiam R. Howlett static inline int dup_anon_vma(struct vm_area_struct *dst, 61004241ffeSLiam R. Howlett struct vm_area_struct *src) 61104241ffeSLiam R. Howlett { 61204241ffeSLiam R. Howlett /* 61304241ffeSLiam R. Howlett * Easily overlooked: when mprotect shifts the boundary, make sure the 61404241ffeSLiam R. Howlett * expanding vma has anon_vma set if the shrinking vma had, to cover any 61504241ffeSLiam R. Howlett * anon pages imported. 61604241ffeSLiam R. Howlett */ 61704241ffeSLiam R. Howlett if (src->anon_vma && !dst->anon_vma) { 61804241ffeSLiam R. Howlett dst->anon_vma = src->anon_vma; 61904241ffeSLiam R. Howlett return anon_vma_clone(dst, src); 62004241ffeSLiam R. Howlett } 62104241ffeSLiam R. Howlett 62204241ffeSLiam R. Howlett return 0; 62304241ffeSLiam R. Howlett } 62404241ffeSLiam R. Howlett 62504241ffeSLiam R. Howlett /* 6269303d3e1SLiam R. Howlett * vma_expand - Expand an existing VMA 6279303d3e1SLiam R. Howlett * 6289303d3e1SLiam R. Howlett * @vmi: The vma iterator 6299303d3e1SLiam R. Howlett * @vma: The vma to expand 6309303d3e1SLiam R. Howlett * @start: The start of the vma 6319303d3e1SLiam R. Howlett * @end: The exclusive end of the vma 6329303d3e1SLiam R. Howlett * @pgoff: The page offset of vma 6339303d3e1SLiam R. Howlett * @next: The current of next vma. 6349303d3e1SLiam R. Howlett * 6359303d3e1SLiam R. Howlett * Expand @vma to @start and @end. Can expand off the start and end. Will 6369303d3e1SLiam R. Howlett * expand over @next if it's different from @vma and @end == @next->vm_end. 6379303d3e1SLiam R. Howlett * Checking if the @vma can expand and merge with @next needs to be handled by 6389303d3e1SLiam R. Howlett * the caller. 6399303d3e1SLiam R. Howlett * 6409303d3e1SLiam R. Howlett * Returns: 0 on success 6419303d3e1SLiam R. Howlett */ 6427c9813e8SLiam R. Howlett int vma_expand(struct vma_iterator *vmi, struct vm_area_struct *vma, 6439303d3e1SLiam R. Howlett unsigned long start, unsigned long end, pgoff_t pgoff, 6449303d3e1SLiam R. Howlett struct vm_area_struct *next) 6459303d3e1SLiam R. Howlett { 64668cefec5SLiam R. Howlett bool remove_next = false; 6479303d3e1SLiam R. Howlett struct vma_prepare vp; 6489303d3e1SLiam R. Howlett 6499303d3e1SLiam R. Howlett if (next && (vma != next) && (end == next->vm_end)) { 65004241ffeSLiam R. Howlett int ret; 6519303d3e1SLiam R. Howlett 65204241ffeSLiam R. Howlett remove_next = true; 65304241ffeSLiam R. Howlett ret = dup_anon_vma(vma, next); 65404241ffeSLiam R. Howlett if (ret) 65504241ffeSLiam R. Howlett return ret; 6569303d3e1SLiam R. Howlett } 6579303d3e1SLiam R. Howlett 65868cefec5SLiam R. Howlett init_multi_vma_prep(&vp, vma, NULL, remove_next ? next : NULL, NULL); 6599303d3e1SLiam R. Howlett /* Not merging but overwriting any part of next is not handled. */ 6609303d3e1SLiam R. Howlett VM_WARN_ON(next && !vp.remove && 6619303d3e1SLiam R. Howlett next != vma && end > next->vm_start); 6629303d3e1SLiam R. Howlett /* Only handles expanding */ 6639303d3e1SLiam R. Howlett VM_WARN_ON(vma->vm_start < start || vma->vm_end > end); 6649303d3e1SLiam R. Howlett 6659303d3e1SLiam R. Howlett if (vma_iter_prealloc(vmi)) 6669303d3e1SLiam R. Howlett goto nomem; 6679303d3e1SLiam R. Howlett 668ccf1d78dSSuren Baghdasaryan vma_prepare(&vp); 6699303d3e1SLiam R. Howlett vma_adjust_trans_huge(vma, start, end, 0); 6709303d3e1SLiam R. Howlett /* VMA iterator points to previous, so set to start if necessary */ 6719303d3e1SLiam R. Howlett if (vma_iter_addr(vmi) != start) 6729303d3e1SLiam R. Howlett vma_iter_set(vmi, start); 6739303d3e1SLiam R. Howlett 6749303d3e1SLiam R. Howlett vma->vm_start = start; 6759303d3e1SLiam R. Howlett vma->vm_end = end; 6769303d3e1SLiam R. Howlett vma->vm_pgoff = pgoff; 6779303d3e1SLiam R. Howlett /* Note: mas must be pointing to the expanding VMA */ 6789303d3e1SLiam R. Howlett vma_iter_store(vmi, vma); 6799303d3e1SLiam R. Howlett 6809303d3e1SLiam R. Howlett vma_complete(&vp, vmi, vma->vm_mm); 6819303d3e1SLiam R. Howlett validate_mm(vma->vm_mm); 6829303d3e1SLiam R. Howlett return 0; 6839303d3e1SLiam R. Howlett 6849303d3e1SLiam R. Howlett nomem: 6859303d3e1SLiam R. Howlett return -ENOMEM; 6869303d3e1SLiam R. Howlett } 687cf51e86dSLiam R. Howlett 688cf51e86dSLiam R. Howlett /* 689cf51e86dSLiam R. Howlett * vma_shrink() - Reduce an existing VMAs memory area 690cf51e86dSLiam R. Howlett * @vmi: The vma iterator 691cf51e86dSLiam R. Howlett * @vma: The VMA to modify 692cf51e86dSLiam R. Howlett * @start: The new start 693cf51e86dSLiam R. Howlett * @end: The new end 694cf51e86dSLiam R. Howlett * 695cf51e86dSLiam R. Howlett * Returns: 0 on success, -ENOMEM otherwise 696cf51e86dSLiam R. Howlett */ 697cf51e86dSLiam R. Howlett int vma_shrink(struct vma_iterator *vmi, struct vm_area_struct *vma, 698cf51e86dSLiam R. Howlett unsigned long start, unsigned long end, pgoff_t pgoff) 699cf51e86dSLiam R. Howlett { 700cf51e86dSLiam R. Howlett struct vma_prepare vp; 701cf51e86dSLiam R. Howlett 702cf51e86dSLiam R. Howlett WARN_ON((vma->vm_start != start) && (vma->vm_end != end)); 703cf51e86dSLiam R. Howlett 704cf51e86dSLiam R. Howlett if (vma_iter_prealloc(vmi)) 705cf51e86dSLiam R. Howlett return -ENOMEM; 706cf51e86dSLiam R. Howlett 707cf51e86dSLiam R. Howlett init_vma_prep(&vp, vma); 708cf51e86dSLiam R. Howlett vma_prepare(&vp); 709ccf1d78dSSuren Baghdasaryan vma_adjust_trans_huge(vma, start, end, 0); 710cf51e86dSLiam R. Howlett 711cf51e86dSLiam R. Howlett if (vma->vm_start < start) 712cf51e86dSLiam R. Howlett vma_iter_clear(vmi, vma->vm_start, start); 713cf51e86dSLiam R. Howlett 714cf51e86dSLiam R. Howlett if (vma->vm_end > end) 715cf51e86dSLiam R. Howlett vma_iter_clear(vmi, end, vma->vm_end); 716cf51e86dSLiam R. Howlett 717cf51e86dSLiam R. Howlett vma->vm_start = start; 718cf51e86dSLiam R. Howlett vma->vm_end = end; 719cf51e86dSLiam R. Howlett vma->vm_pgoff = pgoff; 720cf51e86dSLiam R. Howlett vma_complete(&vp, vmi, vma->vm_mm); 721cf51e86dSLiam R. Howlett validate_mm(vma->vm_mm); 722cf51e86dSLiam R. Howlett return 0; 723cf51e86dSLiam R. Howlett } 724cf51e86dSLiam R. Howlett 7259303d3e1SLiam R. Howlett /* 7261da177e4SLinus Torvalds * If the vma has a ->close operation then the driver probably needs to release 727714965caSVlastimil Babka * per-vma resources, so we don't attempt to merge those if the caller indicates 728714965caSVlastimil Babka * the current vma may be removed as part of the merge. 7291da177e4SLinus Torvalds */ 7302dbf4010SVlastimil Babka static inline bool is_mergeable_vma(struct vm_area_struct *vma, 73119a809afSAndrea Arcangeli struct file *file, unsigned long vm_flags, 7329a10064fSColin Cross struct vm_userfaultfd_ctx vm_userfaultfd_ctx, 733714965caSVlastimil Babka struct anon_vma_name *anon_name, bool may_remove_vma) 7341da177e4SLinus Torvalds { 73534228d47SCyrill Gorcunov /* 73634228d47SCyrill Gorcunov * VM_SOFTDIRTY should not prevent from VMA merging, if we 73734228d47SCyrill Gorcunov * match the flags but dirty bit -- the caller should mark 73834228d47SCyrill Gorcunov * merged VMA as dirty. If dirty bit won't be excluded from 7398bb4e7a2SWei Yang * comparison, we increase pressure on the memory system forcing 74034228d47SCyrill Gorcunov * the kernel to generate new VMAs when old one could be 74134228d47SCyrill Gorcunov * extended instead. 74234228d47SCyrill Gorcunov */ 74334228d47SCyrill Gorcunov if ((vma->vm_flags ^ vm_flags) & ~VM_SOFTDIRTY) 7442dbf4010SVlastimil Babka return false; 7451da177e4SLinus Torvalds if (vma->vm_file != file) 7462dbf4010SVlastimil Babka return false; 747714965caSVlastimil Babka if (may_remove_vma && vma->vm_ops && vma->vm_ops->close) 7482dbf4010SVlastimil Babka return false; 74919a809afSAndrea Arcangeli if (!is_mergeable_vm_userfaultfd_ctx(vma, vm_userfaultfd_ctx)) 7502dbf4010SVlastimil Babka return false; 7515c26f6acSSuren Baghdasaryan if (!anon_vma_name_eq(anon_vma_name(vma), anon_name)) 7522dbf4010SVlastimil Babka return false; 7532dbf4010SVlastimil Babka return true; 7541da177e4SLinus Torvalds } 7551da177e4SLinus Torvalds 7562dbf4010SVlastimil Babka static inline bool is_mergeable_anon_vma(struct anon_vma *anon_vma1, 7572dbf4010SVlastimil Babka struct anon_vma *anon_vma2, struct vm_area_struct *vma) 7581da177e4SLinus Torvalds { 759965f55deSShaohua Li /* 760965f55deSShaohua Li * The list_is_singular() test is to avoid merging VMA cloned from 761965f55deSShaohua Li * parents. This can improve scalability caused by anon_vma lock. 762965f55deSShaohua Li */ 763965f55deSShaohua Li if ((!anon_vma1 || !anon_vma2) && (!vma || 764965f55deSShaohua Li list_is_singular(&vma->anon_vma_chain))) 7652dbf4010SVlastimil Babka return true; 766965f55deSShaohua Li return anon_vma1 == anon_vma2; 7671da177e4SLinus Torvalds } 7681da177e4SLinus Torvalds 7691da177e4SLinus Torvalds /* 7701da177e4SLinus Torvalds * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff) 7711da177e4SLinus Torvalds * in front of (at a lower virtual address and file offset than) the vma. 7721da177e4SLinus Torvalds * 7731da177e4SLinus Torvalds * We cannot merge two vmas if they have differently assigned (non-NULL) 7741da177e4SLinus Torvalds * anon_vmas, nor if same anon_vma is assigned but offsets incompatible. 7751da177e4SLinus Torvalds * 7761da177e4SLinus Torvalds * We don't check here for the merged mmap wrapping around the end of pagecache 77745e55300SPeter Collingbourne * indices (16TB on ia32) because do_mmap() does not permit mmap's which 7781da177e4SLinus Torvalds * wrap, nor mmaps which cover the final page at index -1UL. 779714965caSVlastimil Babka * 780714965caSVlastimil Babka * We assume the vma may be removed as part of the merge. 7811da177e4SLinus Torvalds */ 7822dbf4010SVlastimil Babka static bool 7831da177e4SLinus Torvalds can_vma_merge_before(struct vm_area_struct *vma, unsigned long vm_flags, 78419a809afSAndrea Arcangeli struct anon_vma *anon_vma, struct file *file, 7852dbf4010SVlastimil Babka pgoff_t vm_pgoff, struct vm_userfaultfd_ctx vm_userfaultfd_ctx, 7865c26f6acSSuren Baghdasaryan struct anon_vma_name *anon_name) 7871da177e4SLinus Torvalds { 788714965caSVlastimil Babka if (is_mergeable_vma(vma, file, vm_flags, vm_userfaultfd_ctx, anon_name, true) && 789965f55deSShaohua Li is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) { 7901da177e4SLinus Torvalds if (vma->vm_pgoff == vm_pgoff) 7912dbf4010SVlastimil Babka return true; 7921da177e4SLinus Torvalds } 7932dbf4010SVlastimil Babka return false; 7941da177e4SLinus Torvalds } 7951da177e4SLinus Torvalds 7961da177e4SLinus Torvalds /* 7971da177e4SLinus Torvalds * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff) 7981da177e4SLinus Torvalds * beyond (at a higher virtual address and file offset than) the vma. 7991da177e4SLinus Torvalds * 8001da177e4SLinus Torvalds * We cannot merge two vmas if they have differently assigned (non-NULL) 8011da177e4SLinus Torvalds * anon_vmas, nor if same anon_vma is assigned but offsets incompatible. 802714965caSVlastimil Babka * 803714965caSVlastimil Babka * We assume that vma is not removed as part of the merge. 8041da177e4SLinus Torvalds */ 8052dbf4010SVlastimil Babka static bool 8061da177e4SLinus Torvalds can_vma_merge_after(struct vm_area_struct *vma, unsigned long vm_flags, 80719a809afSAndrea Arcangeli struct anon_vma *anon_vma, struct file *file, 8082dbf4010SVlastimil Babka pgoff_t vm_pgoff, struct vm_userfaultfd_ctx vm_userfaultfd_ctx, 8095c26f6acSSuren Baghdasaryan struct anon_vma_name *anon_name) 8101da177e4SLinus Torvalds { 811714965caSVlastimil Babka if (is_mergeable_vma(vma, file, vm_flags, vm_userfaultfd_ctx, anon_name, false) && 812965f55deSShaohua Li is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) { 8131da177e4SLinus Torvalds pgoff_t vm_pglen; 814d6e93217SLibin vm_pglen = vma_pages(vma); 8151da177e4SLinus Torvalds if (vma->vm_pgoff + vm_pglen == vm_pgoff) 8162dbf4010SVlastimil Babka return true; 8171da177e4SLinus Torvalds } 8182dbf4010SVlastimil Babka return false; 8191da177e4SLinus Torvalds } 8201da177e4SLinus Torvalds 8211da177e4SLinus Torvalds /* 8229a10064fSColin Cross * Given a mapping request (addr,end,vm_flags,file,pgoff,anon_name), 8239a10064fSColin Cross * figure out whether that can be merged with its predecessor or its 8249a10064fSColin Cross * successor. Or both (it neatly fills a hole). 8251da177e4SLinus Torvalds * 8261da177e4SLinus Torvalds * In most cases - when called for mmap, brk or mremap - [addr,end) is 8271da177e4SLinus Torvalds * certain not to be mapped by the time vma_merge is called; but when 8281da177e4SLinus Torvalds * called for mprotect, it is certain to be already mapped (either at 8291da177e4SLinus Torvalds * an offset within prev, or at the start of next), and the flags of 8301da177e4SLinus Torvalds * this area are about to be changed to vm_flags - and the no-change 8311da177e4SLinus Torvalds * case has already been eliminated. 8321da177e4SLinus Torvalds * 833fcfccd91SLorenzo Stoakes * The following mprotect cases have to be considered, where **** is 8341da177e4SLinus Torvalds * the area passed down from mprotect_fixup, never extending beyond one 835fcfccd91SLorenzo Stoakes * vma, PPPP is the previous vma, CCCC is a concurrent vma that starts 836fcfccd91SLorenzo Stoakes * at the same address as **** and is of the same or larger span, and 837fcfccd91SLorenzo Stoakes * NNNN the next vma after ****: 8381da177e4SLinus Torvalds * 839fcfccd91SLorenzo Stoakes * **** **** **** 840fcfccd91SLorenzo Stoakes * PPPPPPNNNNNN PPPPPPNNNNNN PPPPPPCCCCCC 8415d42ab29SWei Yang * cannot merge might become might become 842fcfccd91SLorenzo Stoakes * PPNNNNNNNNNN PPPPPPPPPPCC 8435d42ab29SWei Yang * mmap, brk or case 4 below case 5 below 8445d42ab29SWei Yang * mremap move: 845fcfccd91SLorenzo Stoakes * **** **** 846fcfccd91SLorenzo Stoakes * PPPP NNNN PPPPCCCCNNNN 8475d42ab29SWei Yang * might become might become 8485d42ab29SWei Yang * PPPPPPPPPPPP 1 or PPPPPPPPPPPP 6 or 849fcfccd91SLorenzo Stoakes * PPPPPPPPNNNN 2 or PPPPPPPPNNNN 7 or 850fcfccd91SLorenzo Stoakes * PPPPNNNNNNNN 3 PPPPNNNNNNNN 8 8511da177e4SLinus Torvalds * 852fcfccd91SLorenzo Stoakes * It is important for case 8 that the vma CCCC overlapping the 853fcfccd91SLorenzo Stoakes * region **** is never going to extended over NNNN. Instead NNNN must 854fcfccd91SLorenzo Stoakes * be extended in region **** and CCCC must be removed. This way in 8550503ea8fSLiam R. Howlett * all cases where vma_merge succeeds, the moment vma_merge drops the 856e86f15eeSAndrea Arcangeli * rmap_locks, the properties of the merged vma will be already 857e86f15eeSAndrea Arcangeli * correct for the whole merged range. Some of those properties like 858e86f15eeSAndrea Arcangeli * vm_page_prot/vm_flags may be accessed by rmap_walks and they must 859e86f15eeSAndrea Arcangeli * be correct for the whole merged range immediately after the 860fcfccd91SLorenzo Stoakes * rmap_locks are released. Otherwise if NNNN would be removed and 861fcfccd91SLorenzo Stoakes * CCCC would be extended over the NNNN range, remove_migration_ptes 862e86f15eeSAndrea Arcangeli * or other rmap walkers (if working on addresses beyond the "end" 863fcfccd91SLorenzo Stoakes * parameter) may establish ptes with the wrong permissions of CCCC 864fcfccd91SLorenzo Stoakes * instead of the right permissions of NNNN. 8650503ea8fSLiam R. Howlett * 8660503ea8fSLiam R. Howlett * In the code below: 8670503ea8fSLiam R. Howlett * PPPP is represented by *prev 868fcfccd91SLorenzo Stoakes * CCCC is represented by *curr or not represented at all (NULL) 869fcfccd91SLorenzo Stoakes * NNNN is represented by *next or not represented at all (NULL) 870fcfccd91SLorenzo Stoakes * **** is not represented - it will be merged and the vma containing the 8719e8a39d2SVlastimil Babka * area is returned, or the function will return NULL 8721da177e4SLinus Torvalds */ 8739760ebffSLiam R. Howlett struct vm_area_struct *vma_merge(struct vma_iterator *vmi, struct mm_struct *mm, 8741da177e4SLinus Torvalds struct vm_area_struct *prev, unsigned long addr, 8751da177e4SLinus Torvalds unsigned long end, unsigned long vm_flags, 8761da177e4SLinus Torvalds struct anon_vma *anon_vma, struct file *file, 87719a809afSAndrea Arcangeli pgoff_t pgoff, struct mempolicy *policy, 8789a10064fSColin Cross struct vm_userfaultfd_ctx vm_userfaultfd_ctx, 8795c26f6acSSuren Baghdasaryan struct anon_vma_name *anon_name) 8801da177e4SLinus Torvalds { 881b0729ae0SLorenzo Stoakes struct vm_area_struct *curr, *next, *res; 8820503ea8fSLiam R. Howlett struct vm_area_struct *vma, *adjust, *remove, *remove2; 8830173db4fSLorenzo Stoakes struct vma_prepare vp; 8840173db4fSLorenzo Stoakes pgoff_t vma_pgoff; 8850173db4fSLorenzo Stoakes int err = 0; 886eef19944SJakub Matěna bool merge_prev = false; 887eef19944SJakub Matěna bool merge_next = false; 8880503ea8fSLiam R. Howlett bool vma_expanded = false; 8890503ea8fSLiam R. Howlett unsigned long vma_start = addr; 8900173db4fSLorenzo Stoakes unsigned long vma_end = end; 8910173db4fSLorenzo Stoakes pgoff_t pglen = (end - addr) >> PAGE_SHIFT; 8920173db4fSLorenzo Stoakes long adj_start = 0; 8931da177e4SLinus Torvalds 8940503ea8fSLiam R. Howlett validate_mm(mm); 8951da177e4SLinus Torvalds /* 8961da177e4SLinus Torvalds * We later require that vma->vm_flags == vm_flags, 8971da177e4SLinus Torvalds * so this tests vma->vm_flags & VM_SPECIAL, too. 8981da177e4SLinus Torvalds */ 8991da177e4SLinus Torvalds if (vm_flags & VM_SPECIAL) 9001da177e4SLinus Torvalds return NULL; 9011da177e4SLinus Torvalds 90200cd00a6SLorenzo Stoakes /* Does the input range span an existing VMA? (cases 5 - 8) */ 90300cd00a6SLorenzo Stoakes curr = find_vma_intersection(mm, prev ? prev->vm_end : 0, end); 9041da177e4SLinus Torvalds 90500cd00a6SLorenzo Stoakes if (!curr || /* cases 1 - 4 */ 90600cd00a6SLorenzo Stoakes end == curr->vm_end) /* cases 6 - 8, adjacent VMA */ 90700cd00a6SLorenzo Stoakes next = vma_lookup(mm, end); 90800cd00a6SLorenzo Stoakes else 90900cd00a6SLorenzo Stoakes next = NULL; /* case 5 */ 910e86f15eeSAndrea Arcangeli 9110503ea8fSLiam R. Howlett if (prev) { 9120503ea8fSLiam R. Howlett vma_start = prev->vm_start; 9130503ea8fSLiam R. Howlett vma_pgoff = prev->vm_pgoff; 9140173db4fSLorenzo Stoakes 915eef19944SJakub Matěna /* Can we merge the predecessor? */ 9160173db4fSLorenzo Stoakes if (addr == prev->vm_end && mpol_equal(vma_policy(prev), policy) 9170503ea8fSLiam R. Howlett && can_vma_merge_after(prev, vm_flags, anon_vma, file, 9180503ea8fSLiam R. Howlett pgoff, vm_userfaultfd_ctx, anon_name)) { 919eef19944SJakub Matěna merge_prev = true; 92018b098afSLiam R. Howlett vma_prev(vmi); 9211da177e4SLinus Torvalds } 9220503ea8fSLiam R. Howlett } 923b0729ae0SLorenzo Stoakes 924eef19944SJakub Matěna /* Can we merge the successor? */ 92500cd00a6SLorenzo Stoakes if (next && mpol_equal(policy, vma_policy(next)) && 9260173db4fSLorenzo Stoakes can_vma_merge_before(next, vm_flags, anon_vma, file, pgoff+pglen, 9279a10064fSColin Cross vm_userfaultfd_ctx, anon_name)) { 928eef19944SJakub Matěna merge_next = true; 929eef19944SJakub Matěna } 9300503ea8fSLiam R. Howlett 93129417d29SLorenzo Stoakes /* Verify some invariant that must be enforced by the caller. */ 93229417d29SLorenzo Stoakes VM_WARN_ON(prev && addr <= prev->vm_start); 93329417d29SLorenzo Stoakes VM_WARN_ON(curr && (addr != curr->vm_start || end > curr->vm_end)); 93429417d29SLorenzo Stoakes VM_WARN_ON(addr >= end); 93529417d29SLorenzo Stoakes 9360173db4fSLorenzo Stoakes if (!merge_prev && !merge_next) 9370173db4fSLorenzo Stoakes return NULL; /* Not mergeable. */ 9380173db4fSLorenzo Stoakes 9390173db4fSLorenzo Stoakes res = vma = prev; 9400503ea8fSLiam R. Howlett remove = remove2 = adjust = NULL; 9410173db4fSLorenzo Stoakes 942eef19944SJakub Matěna /* Can we merge both the predecessor and the successor? */ 943eef19944SJakub Matěna if (merge_prev && merge_next && 9440503ea8fSLiam R. Howlett is_mergeable_anon_vma(prev->anon_vma, next->anon_vma, NULL)) { 9455ff783f1SVlastimil Babka remove = next; /* case 1 */ 9460503ea8fSLiam R. Howlett vma_end = next->vm_end; 94750dac011SVlastimil Babka err = dup_anon_vma(prev, next); 948fcfccd91SLorenzo Stoakes if (curr) { /* case 6 */ 949fcfccd91SLorenzo Stoakes remove = curr; 9500503ea8fSLiam R. Howlett remove2 = next; 9515ff783f1SVlastimil Babka if (!next->anon_vma) 952fcfccd91SLorenzo Stoakes err = dup_anon_vma(prev, curr); 9530503ea8fSLiam R. Howlett } 9540173db4fSLorenzo Stoakes } else if (merge_prev) { /* case 2 */ 955fcfccd91SLorenzo Stoakes if (curr) { 956fcfccd91SLorenzo Stoakes err = dup_anon_vma(prev, curr); 957fcfccd91SLorenzo Stoakes if (end == curr->vm_end) { /* case 7 */ 958fcfccd91SLorenzo Stoakes remove = curr; 9590503ea8fSLiam R. Howlett } else { /* case 5 */ 960fcfccd91SLorenzo Stoakes adjust = curr; 961fcfccd91SLorenzo Stoakes adj_start = (end - curr->vm_start); 9620503ea8fSLiam R. Howlett } 9630503ea8fSLiam R. Howlett } 9640173db4fSLorenzo Stoakes } else { /* merge_next */ 965eef19944SJakub Matěna res = next; 9660503ea8fSLiam R. Howlett if (prev && addr < prev->vm_end) { /* case 4 */ 9670503ea8fSLiam R. Howlett vma_end = addr; 968183b7a60SVlastimil Babka adjust = next; 9691e76454fSVlastimil Babka adj_start = -(prev->vm_end - addr); 970183b7a60SVlastimil Babka err = dup_anon_vma(next, prev); 9710503ea8fSLiam R. Howlett } else { 972b0729ae0SLorenzo Stoakes /* 973b0729ae0SLorenzo Stoakes * Note that cases 3 and 8 are the ONLY ones where prev 974b0729ae0SLorenzo Stoakes * is permitted to be (but is not necessarily) NULL. 975b0729ae0SLorenzo Stoakes */ 9760503ea8fSLiam R. Howlett vma = next; /* case 3 */ 9770503ea8fSLiam R. Howlett vma_start = addr; 9780503ea8fSLiam R. Howlett vma_end = next->vm_end; 9797e775787SVlastimil Babka vma_pgoff = next->vm_pgoff - pglen; 980fcfccd91SLorenzo Stoakes if (curr) { /* case 8 */ 981fcfccd91SLorenzo Stoakes vma_pgoff = curr->vm_pgoff; 982fcfccd91SLorenzo Stoakes remove = curr; 983fcfccd91SLorenzo Stoakes err = dup_anon_vma(next, curr); 9840503ea8fSLiam R. Howlett } 9850503ea8fSLiam R. Howlett } 9861da177e4SLinus Torvalds } 9871da177e4SLinus Torvalds 9880173db4fSLorenzo Stoakes /* Error in anon_vma clone. */ 989eef19944SJakub Matěna if (err) 9901da177e4SLinus Torvalds return NULL; 9910503ea8fSLiam R. Howlett 9920503ea8fSLiam R. Howlett if (vma_iter_prealloc(vmi)) 9930503ea8fSLiam R. Howlett return NULL; 9940503ea8fSLiam R. Howlett 9950503ea8fSLiam R. Howlett init_multi_vma_prep(&vp, vma, adjust, remove, remove2); 9960503ea8fSLiam R. Howlett VM_WARN_ON(vp.anon_vma && adjust && adjust->anon_vma && 9970503ea8fSLiam R. Howlett vp.anon_vma != adjust->anon_vma); 9980503ea8fSLiam R. Howlett 9990503ea8fSLiam R. Howlett vma_prepare(&vp); 1000ccf1d78dSSuren Baghdasaryan vma_adjust_trans_huge(vma, vma_start, vma_end, adj_start); 10010503ea8fSLiam R. Howlett if (vma_start < vma->vm_start || vma_end > vma->vm_end) 10020503ea8fSLiam R. Howlett vma_expanded = true; 10030503ea8fSLiam R. Howlett 10040503ea8fSLiam R. Howlett vma->vm_start = vma_start; 10050503ea8fSLiam R. Howlett vma->vm_end = vma_end; 10060503ea8fSLiam R. Howlett vma->vm_pgoff = vma_pgoff; 10070503ea8fSLiam R. Howlett 10080503ea8fSLiam R. Howlett if (vma_expanded) 10090503ea8fSLiam R. Howlett vma_iter_store(vmi, vma); 10100503ea8fSLiam R. Howlett 10111e76454fSVlastimil Babka if (adj_start) { 10121e76454fSVlastimil Babka adjust->vm_start += adj_start; 10131e76454fSVlastimil Babka adjust->vm_pgoff += adj_start >> PAGE_SHIFT; 10141e76454fSVlastimil Babka if (adj_start < 0) { 10150503ea8fSLiam R. Howlett WARN_ON(vma_expanded); 10160503ea8fSLiam R. Howlett vma_iter_store(vmi, next); 10170503ea8fSLiam R. Howlett } 10180503ea8fSLiam R. Howlett } 10190503ea8fSLiam R. Howlett 10200503ea8fSLiam R. Howlett vma_complete(&vp, vmi, mm); 10210503ea8fSLiam R. Howlett vma_iter_free(vmi); 10220503ea8fSLiam R. Howlett validate_mm(mm); 1023eef19944SJakub Matěna khugepaged_enter_vma(res, vm_flags); 10241da177e4SLinus Torvalds 10259760ebffSLiam R. Howlett return res; 1026f2ebfe43SLiam R. Howlett } 1027f2ebfe43SLiam R. Howlett 10281da177e4SLinus Torvalds /* 1029b4f315b4SEthon Paul * Rough compatibility check to quickly see if it's even worth looking 1030d0e9fe17SLinus Torvalds * at sharing an anon_vma. 1031d0e9fe17SLinus Torvalds * 1032d0e9fe17SLinus Torvalds * They need to have the same vm_file, and the flags can only differ 1033d0e9fe17SLinus Torvalds * in things that mprotect may change. 1034d0e9fe17SLinus Torvalds * 1035d0e9fe17SLinus Torvalds * NOTE! The fact that we share an anon_vma doesn't _have_ to mean that 1036d0e9fe17SLinus Torvalds * we can merge the two vma's. For example, we refuse to merge a vma if 1037d0e9fe17SLinus Torvalds * there is a vm_ops->close() function, because that indicates that the 1038d0e9fe17SLinus Torvalds * driver is doing some kind of reference counting. But that doesn't 1039d0e9fe17SLinus Torvalds * really matter for the anon_vma sharing case. 1040d0e9fe17SLinus Torvalds */ 1041d0e9fe17SLinus Torvalds static int anon_vma_compatible(struct vm_area_struct *a, struct vm_area_struct *b) 1042d0e9fe17SLinus Torvalds { 1043d0e9fe17SLinus Torvalds return a->vm_end == b->vm_start && 1044d0e9fe17SLinus Torvalds mpol_equal(vma_policy(a), vma_policy(b)) && 1045d0e9fe17SLinus Torvalds a->vm_file == b->vm_file && 10466cb4d9a2SAnshuman Khandual !((a->vm_flags ^ b->vm_flags) & ~(VM_ACCESS_FLAGS | VM_SOFTDIRTY)) && 1047d0e9fe17SLinus Torvalds b->vm_pgoff == a->vm_pgoff + ((b->vm_start - a->vm_start) >> PAGE_SHIFT); 1048d0e9fe17SLinus Torvalds } 1049d0e9fe17SLinus Torvalds 1050d0e9fe17SLinus Torvalds /* 1051d0e9fe17SLinus Torvalds * Do some basic sanity checking to see if we can re-use the anon_vma 1052d0e9fe17SLinus Torvalds * from 'old'. The 'a'/'b' vma's are in VM order - one of them will be 1053d0e9fe17SLinus Torvalds * the same as 'old', the other will be the new one that is trying 1054d0e9fe17SLinus Torvalds * to share the anon_vma. 1055d0e9fe17SLinus Torvalds * 10565b449489SFlorian Rommel * NOTE! This runs with mmap_lock held for reading, so it is possible that 1057d0e9fe17SLinus Torvalds * the anon_vma of 'old' is concurrently in the process of being set up 1058d0e9fe17SLinus Torvalds * by another page fault trying to merge _that_. But that's ok: if it 1059d0e9fe17SLinus Torvalds * is being set up, that automatically means that it will be a singleton 1060d0e9fe17SLinus Torvalds * acceptable for merging, so we can do all of this optimistically. But 10614db0c3c2SJason Low * we do that READ_ONCE() to make sure that we never re-load the pointer. 1062d0e9fe17SLinus Torvalds * 1063d0e9fe17SLinus Torvalds * IOW: that the "list_is_singular()" test on the anon_vma_chain only 1064d0e9fe17SLinus Torvalds * matters for the 'stable anon_vma' case (ie the thing we want to avoid 1065d0e9fe17SLinus Torvalds * is to return an anon_vma that is "complex" due to having gone through 1066d0e9fe17SLinus Torvalds * a fork). 1067d0e9fe17SLinus Torvalds * 1068d0e9fe17SLinus Torvalds * We also make sure that the two vma's are compatible (adjacent, 1069d0e9fe17SLinus Torvalds * and with the same memory policies). That's all stable, even with just 10705b449489SFlorian Rommel * a read lock on the mmap_lock. 1071d0e9fe17SLinus Torvalds */ 1072d0e9fe17SLinus Torvalds static struct anon_vma *reusable_anon_vma(struct vm_area_struct *old, struct vm_area_struct *a, struct vm_area_struct *b) 1073d0e9fe17SLinus Torvalds { 1074d0e9fe17SLinus Torvalds if (anon_vma_compatible(a, b)) { 10754db0c3c2SJason Low struct anon_vma *anon_vma = READ_ONCE(old->anon_vma); 1076d0e9fe17SLinus Torvalds 1077d0e9fe17SLinus Torvalds if (anon_vma && list_is_singular(&old->anon_vma_chain)) 1078d0e9fe17SLinus Torvalds return anon_vma; 1079d0e9fe17SLinus Torvalds } 1080d0e9fe17SLinus Torvalds return NULL; 1081d0e9fe17SLinus Torvalds } 1082d0e9fe17SLinus Torvalds 1083d0e9fe17SLinus Torvalds /* 10841da177e4SLinus Torvalds * find_mergeable_anon_vma is used by anon_vma_prepare, to check 10851da177e4SLinus Torvalds * neighbouring vmas for a suitable anon_vma, before it goes off 10861da177e4SLinus Torvalds * to allocate a new anon_vma. It checks because a repetitive 10871da177e4SLinus Torvalds * sequence of mprotects and faults may otherwise lead to distinct 10881da177e4SLinus Torvalds * anon_vmas being allocated, preventing vma merge in subsequent 10891da177e4SLinus Torvalds * mprotect. 10901da177e4SLinus Torvalds */ 10911da177e4SLinus Torvalds struct anon_vma *find_mergeable_anon_vma(struct vm_area_struct *vma) 10921da177e4SLinus Torvalds { 1093763ecb03SLiam R. Howlett MA_STATE(mas, &vma->vm_mm->mm_mt, vma->vm_end, vma->vm_end); 1094a67c8caaSMiaohe Lin struct anon_vma *anon_vma = NULL; 1095763ecb03SLiam R. Howlett struct vm_area_struct *prev, *next; 10961da177e4SLinus Torvalds 1097a67c8caaSMiaohe Lin /* Try next first. */ 1098763ecb03SLiam R. Howlett next = mas_walk(&mas); 1099763ecb03SLiam R. Howlett if (next) { 1100763ecb03SLiam R. Howlett anon_vma = reusable_anon_vma(next, vma, next); 1101d0e9fe17SLinus Torvalds if (anon_vma) 1102d0e9fe17SLinus Torvalds return anon_vma; 1103a67c8caaSMiaohe Lin } 11041da177e4SLinus Torvalds 1105763ecb03SLiam R. Howlett prev = mas_prev(&mas, 0); 1106763ecb03SLiam R. Howlett VM_BUG_ON_VMA(prev != vma, vma); 1107763ecb03SLiam R. Howlett prev = mas_prev(&mas, 0); 1108a67c8caaSMiaohe Lin /* Try prev next. */ 1109763ecb03SLiam R. Howlett if (prev) 1110763ecb03SLiam R. Howlett anon_vma = reusable_anon_vma(prev, prev, vma); 1111a67c8caaSMiaohe Lin 11121da177e4SLinus Torvalds /* 1113a67c8caaSMiaohe Lin * We might reach here with anon_vma == NULL if we can't find 1114a67c8caaSMiaohe Lin * any reusable anon_vma. 11151da177e4SLinus Torvalds * There's no absolute need to look only at touching neighbours: 11161da177e4SLinus Torvalds * we could search further afield for "compatible" anon_vmas. 11171da177e4SLinus Torvalds * But it would probably just be a waste of time searching, 11181da177e4SLinus Torvalds * or lead to too many vmas hanging off the same anon_vma. 11191da177e4SLinus Torvalds * We're trying to allow mprotect remerging later on, 11201da177e4SLinus Torvalds * not trying to minimize memory used for anon_vmas. 11211da177e4SLinus Torvalds */ 1122a67c8caaSMiaohe Lin return anon_vma; 11231da177e4SLinus Torvalds } 11241da177e4SLinus Torvalds 11251da177e4SLinus Torvalds /* 112640401530SAl Viro * If a hint addr is less than mmap_min_addr change hint to be as 112740401530SAl Viro * low as possible but still greater than mmap_min_addr 112840401530SAl Viro */ 112940401530SAl Viro static inline unsigned long round_hint_to_min(unsigned long hint) 113040401530SAl Viro { 113140401530SAl Viro hint &= PAGE_MASK; 113240401530SAl Viro if (((void *)hint != NULL) && 113340401530SAl Viro (hint < mmap_min_addr)) 113440401530SAl Viro return PAGE_ALIGN(mmap_min_addr); 113540401530SAl Viro return hint; 113640401530SAl Viro } 113740401530SAl Viro 1138b0cc5e89SAndrew Morton bool mlock_future_ok(struct mm_struct *mm, unsigned long flags, 11393c54a298SLorenzo Stoakes unsigned long bytes) 1140363ee17fSDavidlohr Bueso { 11413c54a298SLorenzo Stoakes unsigned long locked_pages, limit_pages; 1142363ee17fSDavidlohr Bueso 11433c54a298SLorenzo Stoakes if (!(flags & VM_LOCKED) || capable(CAP_IPC_LOCK)) 11443c54a298SLorenzo Stoakes return true; 11453c54a298SLorenzo Stoakes 11463c54a298SLorenzo Stoakes locked_pages = bytes >> PAGE_SHIFT; 11473c54a298SLorenzo Stoakes locked_pages += mm->locked_vm; 11483c54a298SLorenzo Stoakes 11493c54a298SLorenzo Stoakes limit_pages = rlimit(RLIMIT_MEMLOCK); 11503c54a298SLorenzo Stoakes limit_pages >>= PAGE_SHIFT; 11513c54a298SLorenzo Stoakes 11523c54a298SLorenzo Stoakes return locked_pages <= limit_pages; 1153363ee17fSDavidlohr Bueso } 1154363ee17fSDavidlohr Bueso 1155be83bbf8SLinus Torvalds static inline u64 file_mmap_size_max(struct file *file, struct inode *inode) 1156be83bbf8SLinus Torvalds { 1157be83bbf8SLinus Torvalds if (S_ISREG(inode->i_mode)) 1158423913adSLinus Torvalds return MAX_LFS_FILESIZE; 1159be83bbf8SLinus Torvalds 1160be83bbf8SLinus Torvalds if (S_ISBLK(inode->i_mode)) 1161be83bbf8SLinus Torvalds return MAX_LFS_FILESIZE; 1162be83bbf8SLinus Torvalds 116376f34950SIvan Khoronzhuk if (S_ISSOCK(inode->i_mode)) 116476f34950SIvan Khoronzhuk return MAX_LFS_FILESIZE; 116576f34950SIvan Khoronzhuk 1166be83bbf8SLinus Torvalds /* Special "we do even unsigned file positions" case */ 1167be83bbf8SLinus Torvalds if (file->f_mode & FMODE_UNSIGNED_OFFSET) 1168be83bbf8SLinus Torvalds return 0; 1169be83bbf8SLinus Torvalds 1170be83bbf8SLinus Torvalds /* Yes, random drivers might want more. But I'm tired of buggy drivers */ 1171be83bbf8SLinus Torvalds return ULONG_MAX; 1172be83bbf8SLinus Torvalds } 1173be83bbf8SLinus Torvalds 1174be83bbf8SLinus Torvalds static inline bool file_mmap_ok(struct file *file, struct inode *inode, 1175be83bbf8SLinus Torvalds unsigned long pgoff, unsigned long len) 1176be83bbf8SLinus Torvalds { 1177be83bbf8SLinus Torvalds u64 maxsize = file_mmap_size_max(file, inode); 1178be83bbf8SLinus Torvalds 1179be83bbf8SLinus Torvalds if (maxsize && len > maxsize) 1180be83bbf8SLinus Torvalds return false; 1181be83bbf8SLinus Torvalds maxsize -= len; 1182be83bbf8SLinus Torvalds if (pgoff > maxsize >> PAGE_SHIFT) 1183be83bbf8SLinus Torvalds return false; 1184be83bbf8SLinus Torvalds return true; 1185be83bbf8SLinus Torvalds } 1186be83bbf8SLinus Torvalds 118740401530SAl Viro /* 11883e4e28c5SMichel Lespinasse * The caller must write-lock current->mm->mmap_lock. 11891da177e4SLinus Torvalds */ 11901fcfd8dbSOleg Nesterov unsigned long do_mmap(struct file *file, unsigned long addr, 11911da177e4SLinus Torvalds unsigned long len, unsigned long prot, 1192*592b5fadSYu-cheng Yu unsigned long flags, vm_flags_t vm_flags, 1193*592b5fadSYu-cheng Yu unsigned long pgoff, unsigned long *populate, 1194*592b5fadSYu-cheng Yu struct list_head *uf) 11951da177e4SLinus Torvalds { 11961da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 119762b5f7d0SDave Hansen int pkey = 0; 11981da177e4SLinus Torvalds 1199524e00b3SLiam R. Howlett validate_mm(mm); 120041badc15SMichel Lespinasse *populate = 0; 1201bebeb3d6SMichel Lespinasse 1202e37609bbSPiotr Kwapulinski if (!len) 1203e37609bbSPiotr Kwapulinski return -EINVAL; 1204e37609bbSPiotr Kwapulinski 12051da177e4SLinus Torvalds /* 12061da177e4SLinus Torvalds * Does the application expect PROT_READ to imply PROT_EXEC? 12071da177e4SLinus Torvalds * 12081da177e4SLinus Torvalds * (the exception is when the underlying filesystem is noexec 12091da177e4SLinus Torvalds * mounted, in which case we dont add PROT_EXEC.) 12101da177e4SLinus Torvalds */ 12111da177e4SLinus Torvalds if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC)) 121290f8572bSEric W. Biederman if (!(file && path_noexec(&file->f_path))) 12131da177e4SLinus Torvalds prot |= PROT_EXEC; 12141da177e4SLinus Torvalds 1215a4ff8e86SMichal Hocko /* force arch specific MAP_FIXED handling in get_unmapped_area */ 1216a4ff8e86SMichal Hocko if (flags & MAP_FIXED_NOREPLACE) 1217a4ff8e86SMichal Hocko flags |= MAP_FIXED; 1218a4ff8e86SMichal Hocko 12197cd94146SEric Paris if (!(flags & MAP_FIXED)) 12207cd94146SEric Paris addr = round_hint_to_min(addr); 12217cd94146SEric Paris 12221da177e4SLinus Torvalds /* Careful about overflows.. */ 12231da177e4SLinus Torvalds len = PAGE_ALIGN(len); 12249206de95SAl Viro if (!len) 12251da177e4SLinus Torvalds return -ENOMEM; 12261da177e4SLinus Torvalds 12271da177e4SLinus Torvalds /* offset overflow? */ 12281da177e4SLinus Torvalds if ((pgoff + (len >> PAGE_SHIFT)) < pgoff) 12291da177e4SLinus Torvalds return -EOVERFLOW; 12301da177e4SLinus Torvalds 12311da177e4SLinus Torvalds /* Too many mappings? */ 12321da177e4SLinus Torvalds if (mm->map_count > sysctl_max_map_count) 12331da177e4SLinus Torvalds return -ENOMEM; 12341da177e4SLinus Torvalds 12351da177e4SLinus Torvalds /* Obtain the address to map to. we verify (or select) it and ensure 12361da177e4SLinus Torvalds * that it represents a valid section of the address space. 12371da177e4SLinus Torvalds */ 12381da177e4SLinus Torvalds addr = get_unmapped_area(file, addr, len, pgoff, flags); 1239ff68dac6SGaowei Pu if (IS_ERR_VALUE(addr)) 12401da177e4SLinus Torvalds return addr; 12411da177e4SLinus Torvalds 1242a4ff8e86SMichal Hocko if (flags & MAP_FIXED_NOREPLACE) { 124335e43c5fSLiam Howlett if (find_vma_intersection(mm, addr, addr + len)) 1244a4ff8e86SMichal Hocko return -EEXIST; 1245a4ff8e86SMichal Hocko } 1246a4ff8e86SMichal Hocko 124762b5f7d0SDave Hansen if (prot == PROT_EXEC) { 124862b5f7d0SDave Hansen pkey = execute_only_pkey(mm); 124962b5f7d0SDave Hansen if (pkey < 0) 125062b5f7d0SDave Hansen pkey = 0; 125162b5f7d0SDave Hansen } 125262b5f7d0SDave Hansen 12531da177e4SLinus Torvalds /* Do simple checking here so the lower-level routines won't have 12541da177e4SLinus Torvalds * to. we assume access permissions have been handled by the open 12551da177e4SLinus Torvalds * of the memory object, so we don't do any here. 12561da177e4SLinus Torvalds */ 1257*592b5fadSYu-cheng Yu vm_flags |= calc_vm_prot_bits(prot, pkey) | calc_vm_flag_bits(flags) | 12581da177e4SLinus Torvalds mm->def_flags | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC; 12591da177e4SLinus Torvalds 1260cdf7b341SHuang Shijie if (flags & MAP_LOCKED) 12611da177e4SLinus Torvalds if (!can_do_mlock()) 12621da177e4SLinus Torvalds return -EPERM; 1263ba470de4SRik van Riel 1264b0cc5e89SAndrew Morton if (!mlock_future_ok(mm, vm_flags, len)) 12651da177e4SLinus Torvalds return -EAGAIN; 12661da177e4SLinus Torvalds 12671da177e4SLinus Torvalds if (file) { 1268077bf22bSOleg Nesterov struct inode *inode = file_inode(file); 12691c972597SDan Williams unsigned long flags_mask; 12701c972597SDan Williams 1271be83bbf8SLinus Torvalds if (!file_mmap_ok(file, inode, pgoff, len)) 1272be83bbf8SLinus Torvalds return -EOVERFLOW; 1273be83bbf8SLinus Torvalds 12741c972597SDan Williams flags_mask = LEGACY_MAP_MASK | file->f_op->mmap_supported_flags; 1275077bf22bSOleg Nesterov 12761da177e4SLinus Torvalds switch (flags & MAP_TYPE) { 12771da177e4SLinus Torvalds case MAP_SHARED: 12781c972597SDan Williams /* 12791c972597SDan Williams * Force use of MAP_SHARED_VALIDATE with non-legacy 12801c972597SDan Williams * flags. E.g. MAP_SYNC is dangerous to use with 12811c972597SDan Williams * MAP_SHARED as you don't know which consistency model 12821c972597SDan Williams * you will get. We silently ignore unsupported flags 12831c972597SDan Williams * with MAP_SHARED to preserve backward compatibility. 12841c972597SDan Williams */ 12851c972597SDan Williams flags &= LEGACY_MAP_MASK; 1286e4a9bc58SJoe Perches fallthrough; 12871c972597SDan Williams case MAP_SHARED_VALIDATE: 12881c972597SDan Williams if (flags & ~flags_mask) 12891c972597SDan Williams return -EOPNOTSUPP; 1290dc617f29SDarrick J. Wong if (prot & PROT_WRITE) { 1291dc617f29SDarrick J. Wong if (!(file->f_mode & FMODE_WRITE)) 12921da177e4SLinus Torvalds return -EACCES; 1293dc617f29SDarrick J. Wong if (IS_SWAPFILE(file->f_mapping->host)) 1294dc617f29SDarrick J. Wong return -ETXTBSY; 1295dc617f29SDarrick J. Wong } 12961da177e4SLinus Torvalds 12971da177e4SLinus Torvalds /* 12981da177e4SLinus Torvalds * Make sure we don't allow writing to an append-only 12991da177e4SLinus Torvalds * file.. 13001da177e4SLinus Torvalds */ 13011da177e4SLinus Torvalds if (IS_APPEND(inode) && (file->f_mode & FMODE_WRITE)) 13021da177e4SLinus Torvalds return -EACCES; 13031da177e4SLinus Torvalds 13041da177e4SLinus Torvalds vm_flags |= VM_SHARED | VM_MAYSHARE; 13051da177e4SLinus Torvalds if (!(file->f_mode & FMODE_WRITE)) 13061da177e4SLinus Torvalds vm_flags &= ~(VM_MAYWRITE | VM_SHARED); 1307e4a9bc58SJoe Perches fallthrough; 13081da177e4SLinus Torvalds case MAP_PRIVATE: 13091da177e4SLinus Torvalds if (!(file->f_mode & FMODE_READ)) 13101da177e4SLinus Torvalds return -EACCES; 131190f8572bSEric W. Biederman if (path_noexec(&file->f_path)) { 131280c5606cSLinus Torvalds if (vm_flags & VM_EXEC) 131380c5606cSLinus Torvalds return -EPERM; 131480c5606cSLinus Torvalds vm_flags &= ~VM_MAYEXEC; 131580c5606cSLinus Torvalds } 131680c5606cSLinus Torvalds 131772c2d531SAl Viro if (!file->f_op->mmap) 131880c5606cSLinus Torvalds return -ENODEV; 1319b2c56e4fSOleg Nesterov if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP)) 1320b2c56e4fSOleg Nesterov return -EINVAL; 13211da177e4SLinus Torvalds break; 13221da177e4SLinus Torvalds 13231da177e4SLinus Torvalds default: 13241da177e4SLinus Torvalds return -EINVAL; 13251da177e4SLinus Torvalds } 13261da177e4SLinus Torvalds } else { 13271da177e4SLinus Torvalds switch (flags & MAP_TYPE) { 13281da177e4SLinus Torvalds case MAP_SHARED: 1329b2c56e4fSOleg Nesterov if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP)) 1330b2c56e4fSOleg Nesterov return -EINVAL; 1331ce363942STejun Heo /* 1332ce363942STejun Heo * Ignore pgoff. 1333ce363942STejun Heo */ 1334ce363942STejun Heo pgoff = 0; 13351da177e4SLinus Torvalds vm_flags |= VM_SHARED | VM_MAYSHARE; 13361da177e4SLinus Torvalds break; 13371da177e4SLinus Torvalds case MAP_PRIVATE: 13381da177e4SLinus Torvalds /* 13391da177e4SLinus Torvalds * Set pgoff according to addr for anon_vma. 13401da177e4SLinus Torvalds */ 13411da177e4SLinus Torvalds pgoff = addr >> PAGE_SHIFT; 13421da177e4SLinus Torvalds break; 13431da177e4SLinus Torvalds default: 13441da177e4SLinus Torvalds return -EINVAL; 13451da177e4SLinus Torvalds } 13461da177e4SLinus Torvalds } 13471da177e4SLinus Torvalds 1348c22c0d63SMichel Lespinasse /* 1349c22c0d63SMichel Lespinasse * Set 'VM_NORESERVE' if we should not account for the 1350c22c0d63SMichel Lespinasse * memory use of this mapping. 1351c22c0d63SMichel Lespinasse */ 1352c22c0d63SMichel Lespinasse if (flags & MAP_NORESERVE) { 1353c22c0d63SMichel Lespinasse /* We honor MAP_NORESERVE if allowed to overcommit */ 1354c22c0d63SMichel Lespinasse if (sysctl_overcommit_memory != OVERCOMMIT_NEVER) 1355c22c0d63SMichel Lespinasse vm_flags |= VM_NORESERVE; 1356c22c0d63SMichel Lespinasse 1357c22c0d63SMichel Lespinasse /* hugetlb applies strict overcommit unless MAP_NORESERVE */ 1358c22c0d63SMichel Lespinasse if (file && is_file_hugepages(file)) 1359c22c0d63SMichel Lespinasse vm_flags |= VM_NORESERVE; 1360c22c0d63SMichel Lespinasse } 1361c22c0d63SMichel Lespinasse 1362897ab3e0SMike Rapoport addr = mmap_region(file, addr, len, vm_flags, pgoff, uf); 136309a9f1d2SMichel Lespinasse if (!IS_ERR_VALUE(addr) && 136409a9f1d2SMichel Lespinasse ((vm_flags & VM_LOCKED) || 136509a9f1d2SMichel Lespinasse (flags & (MAP_POPULATE | MAP_NONBLOCK)) == MAP_POPULATE)) 136641badc15SMichel Lespinasse *populate = len; 1367bebeb3d6SMichel Lespinasse return addr; 13680165ab44SMiklos Szeredi } 13696be5ceb0SLinus Torvalds 1370a90f590aSDominik Brodowski unsigned long ksys_mmap_pgoff(unsigned long addr, unsigned long len, 1371a90f590aSDominik Brodowski unsigned long prot, unsigned long flags, 1372a90f590aSDominik Brodowski unsigned long fd, unsigned long pgoff) 137366f0dc48SHugh Dickins { 137466f0dc48SHugh Dickins struct file *file = NULL; 13751e3ee14bSChen Gang unsigned long retval; 137666f0dc48SHugh Dickins 137766f0dc48SHugh Dickins if (!(flags & MAP_ANONYMOUS)) { 1378120a795dSAl Viro audit_mmap_fd(fd, flags); 137966f0dc48SHugh Dickins file = fget(fd); 138066f0dc48SHugh Dickins if (!file) 13811e3ee14bSChen Gang return -EBADF; 13827bba8f0eSZhen Lei if (is_file_hugepages(file)) { 1383af73e4d9SNaoya Horiguchi len = ALIGN(len, huge_page_size(hstate_file(file))); 13847bba8f0eSZhen Lei } else if (unlikely(flags & MAP_HUGETLB)) { 1385493af578SJörn Engel retval = -EINVAL; 1386493af578SJörn Engel goto out_fput; 13877bba8f0eSZhen Lei } 138866f0dc48SHugh Dickins } else if (flags & MAP_HUGETLB) { 1389c103a4dcSAndrew Morton struct hstate *hs; 1390af73e4d9SNaoya Horiguchi 139120ac2893SAnshuman Khandual hs = hstate_sizelog((flags >> MAP_HUGE_SHIFT) & MAP_HUGE_MASK); 1392091d0d55SLi Zefan if (!hs) 1393091d0d55SLi Zefan return -EINVAL; 1394091d0d55SLi Zefan 1395091d0d55SLi Zefan len = ALIGN(len, huge_page_size(hs)); 139666f0dc48SHugh Dickins /* 139766f0dc48SHugh Dickins * VM_NORESERVE is used because the reservations will be 139866f0dc48SHugh Dickins * taken when vm_ops->mmap() is called 139966f0dc48SHugh Dickins */ 1400af73e4d9SNaoya Horiguchi file = hugetlb_file_setup(HUGETLB_ANON_FILE, len, 140142d7395fSAndi Kleen VM_NORESERVE, 140283c1fd76Szhangyiru HUGETLB_ANONHUGE_INODE, 140342d7395fSAndi Kleen (flags >> MAP_HUGE_SHIFT) & MAP_HUGE_MASK); 140466f0dc48SHugh Dickins if (IS_ERR(file)) 140566f0dc48SHugh Dickins return PTR_ERR(file); 140666f0dc48SHugh Dickins } 140766f0dc48SHugh Dickins 14089fbeb5abSMichal Hocko retval = vm_mmap_pgoff(file, addr, len, prot, flags, pgoff); 1409493af578SJörn Engel out_fput: 141066f0dc48SHugh Dickins if (file) 141166f0dc48SHugh Dickins fput(file); 141266f0dc48SHugh Dickins return retval; 141366f0dc48SHugh Dickins } 141466f0dc48SHugh Dickins 1415a90f590aSDominik Brodowski SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len, 1416a90f590aSDominik Brodowski unsigned long, prot, unsigned long, flags, 1417a90f590aSDominik Brodowski unsigned long, fd, unsigned long, pgoff) 1418a90f590aSDominik Brodowski { 1419a90f590aSDominik Brodowski return ksys_mmap_pgoff(addr, len, prot, flags, fd, pgoff); 1420a90f590aSDominik Brodowski } 1421a90f590aSDominik Brodowski 1422a4679373SChristoph Hellwig #ifdef __ARCH_WANT_SYS_OLD_MMAP 1423a4679373SChristoph Hellwig struct mmap_arg_struct { 1424a4679373SChristoph Hellwig unsigned long addr; 1425a4679373SChristoph Hellwig unsigned long len; 1426a4679373SChristoph Hellwig unsigned long prot; 1427a4679373SChristoph Hellwig unsigned long flags; 1428a4679373SChristoph Hellwig unsigned long fd; 1429a4679373SChristoph Hellwig unsigned long offset; 1430a4679373SChristoph Hellwig }; 1431a4679373SChristoph Hellwig 1432a4679373SChristoph Hellwig SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg) 1433a4679373SChristoph Hellwig { 1434a4679373SChristoph Hellwig struct mmap_arg_struct a; 1435a4679373SChristoph Hellwig 1436a4679373SChristoph Hellwig if (copy_from_user(&a, arg, sizeof(a))) 1437a4679373SChristoph Hellwig return -EFAULT; 1438de1741a1SAlexander Kuleshov if (offset_in_page(a.offset)) 1439a4679373SChristoph Hellwig return -EINVAL; 1440a4679373SChristoph Hellwig 1441a90f590aSDominik Brodowski return ksys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd, 1442a4679373SChristoph Hellwig a.offset >> PAGE_SHIFT); 1443a4679373SChristoph Hellwig } 1444a4679373SChristoph Hellwig #endif /* __ARCH_WANT_SYS_OLD_MMAP */ 1445a4679373SChristoph Hellwig 144654cbbbf3SLorenzo Stoakes static bool vm_ops_needs_writenotify(const struct vm_operations_struct *vm_ops) 144754cbbbf3SLorenzo Stoakes { 144854cbbbf3SLorenzo Stoakes return vm_ops && (vm_ops->page_mkwrite || vm_ops->pfn_mkwrite); 144954cbbbf3SLorenzo Stoakes } 145054cbbbf3SLorenzo Stoakes 145154cbbbf3SLorenzo Stoakes static bool vma_is_shared_writable(struct vm_area_struct *vma) 145254cbbbf3SLorenzo Stoakes { 145354cbbbf3SLorenzo Stoakes return (vma->vm_flags & (VM_WRITE | VM_SHARED)) == 145454cbbbf3SLorenzo Stoakes (VM_WRITE | VM_SHARED); 145554cbbbf3SLorenzo Stoakes } 145654cbbbf3SLorenzo Stoakes 145754cbbbf3SLorenzo Stoakes static bool vma_fs_can_writeback(struct vm_area_struct *vma) 145854cbbbf3SLorenzo Stoakes { 145954cbbbf3SLorenzo Stoakes /* No managed pages to writeback. */ 146054cbbbf3SLorenzo Stoakes if (vma->vm_flags & VM_PFNMAP) 146154cbbbf3SLorenzo Stoakes return false; 146254cbbbf3SLorenzo Stoakes 146354cbbbf3SLorenzo Stoakes return vma->vm_file && vma->vm_file->f_mapping && 146454cbbbf3SLorenzo Stoakes mapping_can_writeback(vma->vm_file->f_mapping); 146554cbbbf3SLorenzo Stoakes } 146654cbbbf3SLorenzo Stoakes 146754cbbbf3SLorenzo Stoakes /* 146854cbbbf3SLorenzo Stoakes * Does this VMA require the underlying folios to have their dirty state 146954cbbbf3SLorenzo Stoakes * tracked? 147054cbbbf3SLorenzo Stoakes */ 147154cbbbf3SLorenzo Stoakes bool vma_needs_dirty_tracking(struct vm_area_struct *vma) 147254cbbbf3SLorenzo Stoakes { 147354cbbbf3SLorenzo Stoakes /* Only shared, writable VMAs require dirty tracking. */ 147454cbbbf3SLorenzo Stoakes if (!vma_is_shared_writable(vma)) 147554cbbbf3SLorenzo Stoakes return false; 147654cbbbf3SLorenzo Stoakes 147754cbbbf3SLorenzo Stoakes /* Does the filesystem need to be notified? */ 147854cbbbf3SLorenzo Stoakes if (vm_ops_needs_writenotify(vma->vm_ops)) 147954cbbbf3SLorenzo Stoakes return true; 148054cbbbf3SLorenzo Stoakes 148154cbbbf3SLorenzo Stoakes /* 148254cbbbf3SLorenzo Stoakes * Even if the filesystem doesn't indicate a need for writenotify, if it 148354cbbbf3SLorenzo Stoakes * can writeback, dirty tracking is still required. 148454cbbbf3SLorenzo Stoakes */ 148554cbbbf3SLorenzo Stoakes return vma_fs_can_writeback(vma); 148654cbbbf3SLorenzo Stoakes } 148754cbbbf3SLorenzo Stoakes 14884e950f6fSAlexey Dobriyan /* 14898bb4e7a2SWei Yang * Some shared mappings will want the pages marked read-only 14904e950f6fSAlexey Dobriyan * to track write events. If so, we'll downgrade vm_page_prot 14914e950f6fSAlexey Dobriyan * to the private version (using protection_map[] without the 14924e950f6fSAlexey Dobriyan * VM_SHARED bit). 14934e950f6fSAlexey Dobriyan */ 14946d2329f8SAndrea Arcangeli int vma_wants_writenotify(struct vm_area_struct *vma, pgprot_t vm_page_prot) 14954e950f6fSAlexey Dobriyan { 14964e950f6fSAlexey Dobriyan /* If it was private or non-writable, the write bit is already clear */ 149754cbbbf3SLorenzo Stoakes if (!vma_is_shared_writable(vma)) 14984e950f6fSAlexey Dobriyan return 0; 14994e950f6fSAlexey Dobriyan 15004e950f6fSAlexey Dobriyan /* The backer wishes to know when pages are first written to? */ 150154cbbbf3SLorenzo Stoakes if (vm_ops_needs_writenotify(vma->vm_ops)) 15024e950f6fSAlexey Dobriyan return 1; 15034e950f6fSAlexey Dobriyan 150464e45507SPeter Feiner /* The open routine did something to the protections that pgprot_modify 150564e45507SPeter Feiner * won't preserve? */ 15066d2329f8SAndrea Arcangeli if (pgprot_val(vm_page_prot) != 150754cbbbf3SLorenzo Stoakes pgprot_val(vm_pgprot_modify(vm_page_prot, vma->vm_flags))) 15084e950f6fSAlexey Dobriyan return 0; 15094e950f6fSAlexey Dobriyan 1510f96f7a40SDavid Hildenbrand /* 1511f96f7a40SDavid Hildenbrand * Do we need to track softdirty? hugetlb does not support softdirty 1512f96f7a40SDavid Hildenbrand * tracking yet. 1513f96f7a40SDavid Hildenbrand */ 1514f96f7a40SDavid Hildenbrand if (vma_soft_dirty_enabled(vma) && !is_vm_hugetlb_page(vma)) 151564e45507SPeter Feiner return 1; 151664e45507SPeter Feiner 151751d3d5ebSDavid Hildenbrand /* Do we need write faults for uffd-wp tracking? */ 151851d3d5ebSDavid Hildenbrand if (userfaultfd_wp(vma)) 151951d3d5ebSDavid Hildenbrand return 1; 152051d3d5ebSDavid Hildenbrand 15214e950f6fSAlexey Dobriyan /* Can the mapping track the dirty pages? */ 152254cbbbf3SLorenzo Stoakes return vma_fs_can_writeback(vma); 15234e950f6fSAlexey Dobriyan } 15244e950f6fSAlexey Dobriyan 1525fc8744adSLinus Torvalds /* 1526fc8744adSLinus Torvalds * We account for memory if it's a private writeable mapping, 15275a6fe125SMel Gorman * not hugepages and VM_NORESERVE wasn't set. 1528fc8744adSLinus Torvalds */ 1529ca16d140SKOSAKI Motohiro static inline int accountable_mapping(struct file *file, vm_flags_t vm_flags) 1530fc8744adSLinus Torvalds { 15315a6fe125SMel Gorman /* 15325a6fe125SMel Gorman * hugetlb has its own accounting separate from the core VM 15335a6fe125SMel Gorman * VM_HUGETLB may not be set yet so we cannot check for that flag. 15345a6fe125SMel Gorman */ 15355a6fe125SMel Gorman if (file && is_file_hugepages(file)) 15365a6fe125SMel Gorman return 0; 15375a6fe125SMel Gorman 1538fc8744adSLinus Torvalds return (vm_flags & (VM_NORESERVE | VM_SHARED | VM_WRITE)) == VM_WRITE; 1539fc8744adSLinus Torvalds } 1540fc8744adSLinus Torvalds 15413499a131SLiam R. Howlett /** 15423499a131SLiam R. Howlett * unmapped_area() - Find an area between the low_limit and the high_limit with 15433499a131SLiam R. Howlett * the correct alignment and offset, all from @info. Note: current->mm is used 15443499a131SLiam R. Howlett * for the search. 15453499a131SLiam R. Howlett * 154682b24936SVernon Yang * @info: The unmapped area information including the range [low_limit - 154782b24936SVernon Yang * high_limit), the alignment offset and mask. 15483499a131SLiam R. Howlett * 15493499a131SLiam R. Howlett * Return: A memory address or -ENOMEM. 15503499a131SLiam R. Howlett */ 1551baceaf1cSJaewon Kim static unsigned long unmapped_area(struct vm_unmapped_area_info *info) 1552db4fbfb9SMichel Lespinasse { 15536b008640SLinus Torvalds unsigned long length, gap; 15546b008640SLinus Torvalds unsigned long low_limit, high_limit; 155558c5d0d6SLiam R. Howlett struct vm_area_struct *tmp; 1556db4fbfb9SMichel Lespinasse 15573499a131SLiam R. Howlett MA_STATE(mas, ¤t->mm->mm_mt, 0, 0); 1558db4fbfb9SMichel Lespinasse 1559db4fbfb9SMichel Lespinasse /* Adjust search length to account for worst case alignment overhead */ 1560db4fbfb9SMichel Lespinasse length = info->length + info->align_mask; 1561db4fbfb9SMichel Lespinasse if (length < info->length) 1562db4fbfb9SMichel Lespinasse return -ENOMEM; 1563db4fbfb9SMichel Lespinasse 156458c5d0d6SLiam R. Howlett low_limit = info->low_limit; 15656b008640SLinus Torvalds if (low_limit < mmap_min_addr) 15666b008640SLinus Torvalds low_limit = mmap_min_addr; 15676b008640SLinus Torvalds high_limit = info->high_limit; 156858c5d0d6SLiam R. Howlett retry: 15696b008640SLinus Torvalds if (mas_empty_area(&mas, low_limit, high_limit - 1, length)) 15703499a131SLiam R. Howlett return -ENOMEM; 15713499a131SLiam R. Howlett 1572d4af56c5SLiam R. Howlett gap = mas.index; 1573d4af56c5SLiam R. Howlett gap += (info->align_offset - gap) & info->align_mask; 157458c5d0d6SLiam R. Howlett tmp = mas_next(&mas, ULONG_MAX); 157558c5d0d6SLiam R. Howlett if (tmp && (tmp->vm_flags & VM_GROWSDOWN)) { /* Avoid prev check if possible */ 157658c5d0d6SLiam R. Howlett if (vm_start_gap(tmp) < gap + length - 1) { 157758c5d0d6SLiam R. Howlett low_limit = tmp->vm_end; 157858c5d0d6SLiam R. Howlett mas_reset(&mas); 157958c5d0d6SLiam R. Howlett goto retry; 158058c5d0d6SLiam R. Howlett } 158158c5d0d6SLiam R. Howlett } else { 158258c5d0d6SLiam R. Howlett tmp = mas_prev(&mas, 0); 158358c5d0d6SLiam R. Howlett if (tmp && vm_end_gap(tmp) > gap) { 158458c5d0d6SLiam R. Howlett low_limit = vm_end_gap(tmp); 158558c5d0d6SLiam R. Howlett mas_reset(&mas); 158658c5d0d6SLiam R. Howlett goto retry; 158758c5d0d6SLiam R. Howlett } 158858c5d0d6SLiam R. Howlett } 158958c5d0d6SLiam R. Howlett 15903499a131SLiam R. Howlett return gap; 1591db4fbfb9SMichel Lespinasse } 1592db4fbfb9SMichel Lespinasse 15933499a131SLiam R. Howlett /** 15943499a131SLiam R. Howlett * unmapped_area_topdown() - Find an area between the low_limit and the 159582b24936SVernon Yang * high_limit with the correct alignment and offset at the highest available 15963499a131SLiam R. Howlett * address, all from @info. Note: current->mm is used for the search. 15973499a131SLiam R. Howlett * 159882b24936SVernon Yang * @info: The unmapped area information including the range [low_limit - 159982b24936SVernon Yang * high_limit), the alignment offset and mask. 16003499a131SLiam R. Howlett * 16013499a131SLiam R. Howlett * Return: A memory address or -ENOMEM. 16023499a131SLiam R. Howlett */ 1603baceaf1cSJaewon Kim static unsigned long unmapped_area_topdown(struct vm_unmapped_area_info *info) 1604db4fbfb9SMichel Lespinasse { 16056b008640SLinus Torvalds unsigned long length, gap, gap_end; 16066b008640SLinus Torvalds unsigned long low_limit, high_limit; 160758c5d0d6SLiam R. Howlett struct vm_area_struct *tmp; 1608d4af56c5SLiam R. Howlett 16093499a131SLiam R. Howlett MA_STATE(mas, ¤t->mm->mm_mt, 0, 0); 1610db4fbfb9SMichel Lespinasse /* Adjust search length to account for worst case alignment overhead */ 1611db4fbfb9SMichel Lespinasse length = info->length + info->align_mask; 1612db4fbfb9SMichel Lespinasse if (length < info->length) 1613db4fbfb9SMichel Lespinasse return -ENOMEM; 1614db4fbfb9SMichel Lespinasse 16156b008640SLinus Torvalds low_limit = info->low_limit; 16166b008640SLinus Torvalds if (low_limit < mmap_min_addr) 16176b008640SLinus Torvalds low_limit = mmap_min_addr; 161858c5d0d6SLiam R. Howlett high_limit = info->high_limit; 161958c5d0d6SLiam R. Howlett retry: 16206b008640SLinus Torvalds if (mas_empty_area_rev(&mas, low_limit, high_limit - 1, length)) 16213499a131SLiam R. Howlett return -ENOMEM; 16223499a131SLiam R. Howlett 1623d4af56c5SLiam R. Howlett gap = mas.last + 1 - info->length; 1624d4af56c5SLiam R. Howlett gap -= (gap - info->align_offset) & info->align_mask; 162558c5d0d6SLiam R. Howlett gap_end = mas.last; 162658c5d0d6SLiam R. Howlett tmp = mas_next(&mas, ULONG_MAX); 162758c5d0d6SLiam R. Howlett if (tmp && (tmp->vm_flags & VM_GROWSDOWN)) { /* Avoid prev check if possible */ 162858c5d0d6SLiam R. Howlett if (vm_start_gap(tmp) <= gap_end) { 162958c5d0d6SLiam R. Howlett high_limit = vm_start_gap(tmp); 163058c5d0d6SLiam R. Howlett mas_reset(&mas); 163158c5d0d6SLiam R. Howlett goto retry; 163258c5d0d6SLiam R. Howlett } 163358c5d0d6SLiam R. Howlett } else { 163458c5d0d6SLiam R. Howlett tmp = mas_prev(&mas, 0); 163558c5d0d6SLiam R. Howlett if (tmp && vm_end_gap(tmp) > gap) { 163658c5d0d6SLiam R. Howlett high_limit = tmp->vm_start; 163758c5d0d6SLiam R. Howlett mas_reset(&mas); 163858c5d0d6SLiam R. Howlett goto retry; 163958c5d0d6SLiam R. Howlett } 164058c5d0d6SLiam R. Howlett } 164158c5d0d6SLiam R. Howlett 16423499a131SLiam R. Howlett return gap; 1643db4fbfb9SMichel Lespinasse } 1644db4fbfb9SMichel Lespinasse 1645baceaf1cSJaewon Kim /* 1646baceaf1cSJaewon Kim * Search for an unmapped address range. 1647baceaf1cSJaewon Kim * 1648baceaf1cSJaewon Kim * We are looking for a range that: 1649baceaf1cSJaewon Kim * - does not intersect with any VMA; 1650baceaf1cSJaewon Kim * - is contained within the [low_limit, high_limit) interval; 1651baceaf1cSJaewon Kim * - is at least the desired size. 1652baceaf1cSJaewon Kim * - satisfies (begin_addr & align_mask) == (align_offset & align_mask) 1653baceaf1cSJaewon Kim */ 1654baceaf1cSJaewon Kim unsigned long vm_unmapped_area(struct vm_unmapped_area_info *info) 1655baceaf1cSJaewon Kim { 1656df529cabSJaewon Kim unsigned long addr; 1657df529cabSJaewon Kim 1658baceaf1cSJaewon Kim if (info->flags & VM_UNMAPPED_AREA_TOPDOWN) 1659df529cabSJaewon Kim addr = unmapped_area_topdown(info); 1660baceaf1cSJaewon Kim else 1661df529cabSJaewon Kim addr = unmapped_area(info); 1662df529cabSJaewon Kim 1663df529cabSJaewon Kim trace_vm_unmapped_area(addr, info); 1664df529cabSJaewon Kim return addr; 1665baceaf1cSJaewon Kim } 1666f6795053SSteve Capper 16671da177e4SLinus Torvalds /* Get an address range which is currently unmapped. 16681da177e4SLinus Torvalds * For shmat() with addr=0. 16691da177e4SLinus Torvalds * 16701da177e4SLinus Torvalds * Ugly calling convention alert: 16711da177e4SLinus Torvalds * Return value with the low bits set means error value, 16721da177e4SLinus Torvalds * ie 16731da177e4SLinus Torvalds * if (ret & ~PAGE_MASK) 16741da177e4SLinus Torvalds * error = ret; 16751da177e4SLinus Torvalds * 16761da177e4SLinus Torvalds * This function "knows" that -ENOMEM has the bits set. 16771da177e4SLinus Torvalds */ 16781da177e4SLinus Torvalds unsigned long 16794b439e25SChristophe Leroy generic_get_unmapped_area(struct file *filp, unsigned long addr, 16804b439e25SChristophe Leroy unsigned long len, unsigned long pgoff, 16814b439e25SChristophe Leroy unsigned long flags) 16821da177e4SLinus Torvalds { 16831da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 16841be7107fSHugh Dickins struct vm_area_struct *vma, *prev; 1685db4fbfb9SMichel Lespinasse struct vm_unmapped_area_info info; 16862cb4de08SChristophe Leroy const unsigned long mmap_end = arch_get_mmap_end(addr, len, flags); 16871da177e4SLinus Torvalds 1688f6795053SSteve Capper if (len > mmap_end - mmap_min_addr) 16891da177e4SLinus Torvalds return -ENOMEM; 16901da177e4SLinus Torvalds 169106abdfb4SBenjamin Herrenschmidt if (flags & MAP_FIXED) 169206abdfb4SBenjamin Herrenschmidt return addr; 169306abdfb4SBenjamin Herrenschmidt 16941da177e4SLinus Torvalds if (addr) { 16951da177e4SLinus Torvalds addr = PAGE_ALIGN(addr); 16961be7107fSHugh Dickins vma = find_vma_prev(mm, addr, &prev); 1697f6795053SSteve Capper if (mmap_end - len >= addr && addr >= mmap_min_addr && 16981be7107fSHugh Dickins (!vma || addr + len <= vm_start_gap(vma)) && 16991be7107fSHugh Dickins (!prev || addr >= vm_end_gap(prev))) 17001da177e4SLinus Torvalds return addr; 17011da177e4SLinus Torvalds } 17021da177e4SLinus Torvalds 1703db4fbfb9SMichel Lespinasse info.flags = 0; 1704db4fbfb9SMichel Lespinasse info.length = len; 17054e99b021SHeiko Carstens info.low_limit = mm->mmap_base; 1706f6795053SSteve Capper info.high_limit = mmap_end; 1707db4fbfb9SMichel Lespinasse info.align_mask = 0; 170809ef5283SJaewon Kim info.align_offset = 0; 1709db4fbfb9SMichel Lespinasse return vm_unmapped_area(&info); 17101da177e4SLinus Torvalds } 17114b439e25SChristophe Leroy 17124b439e25SChristophe Leroy #ifndef HAVE_ARCH_UNMAPPED_AREA 17134b439e25SChristophe Leroy unsigned long 17144b439e25SChristophe Leroy arch_get_unmapped_area(struct file *filp, unsigned long addr, 17154b439e25SChristophe Leroy unsigned long len, unsigned long pgoff, 17164b439e25SChristophe Leroy unsigned long flags) 17174b439e25SChristophe Leroy { 17184b439e25SChristophe Leroy return generic_get_unmapped_area(filp, addr, len, pgoff, flags); 17194b439e25SChristophe Leroy } 17201da177e4SLinus Torvalds #endif 17211da177e4SLinus Torvalds 17221da177e4SLinus Torvalds /* 17231da177e4SLinus Torvalds * This mmap-allocator allocates new areas top-down from below the 17241da177e4SLinus Torvalds * stack's low limit (the base): 17251da177e4SLinus Torvalds */ 17261da177e4SLinus Torvalds unsigned long 17274b439e25SChristophe Leroy generic_get_unmapped_area_topdown(struct file *filp, unsigned long addr, 172843cca0b1SYang Fan unsigned long len, unsigned long pgoff, 172943cca0b1SYang Fan unsigned long flags) 17301da177e4SLinus Torvalds { 17311be7107fSHugh Dickins struct vm_area_struct *vma, *prev; 17321da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 1733db4fbfb9SMichel Lespinasse struct vm_unmapped_area_info info; 17342cb4de08SChristophe Leroy const unsigned long mmap_end = arch_get_mmap_end(addr, len, flags); 17351da177e4SLinus Torvalds 17361da177e4SLinus Torvalds /* requested length too big for entire address space */ 1737f6795053SSteve Capper if (len > mmap_end - mmap_min_addr) 17381da177e4SLinus Torvalds return -ENOMEM; 17391da177e4SLinus Torvalds 174006abdfb4SBenjamin Herrenschmidt if (flags & MAP_FIXED) 174106abdfb4SBenjamin Herrenschmidt return addr; 174206abdfb4SBenjamin Herrenschmidt 17431da177e4SLinus Torvalds /* requesting a specific address */ 17441da177e4SLinus Torvalds if (addr) { 17451da177e4SLinus Torvalds addr = PAGE_ALIGN(addr); 17461be7107fSHugh Dickins vma = find_vma_prev(mm, addr, &prev); 1747f6795053SSteve Capper if (mmap_end - len >= addr && addr >= mmap_min_addr && 17481be7107fSHugh Dickins (!vma || addr + len <= vm_start_gap(vma)) && 17491be7107fSHugh Dickins (!prev || addr >= vm_end_gap(prev))) 17501da177e4SLinus Torvalds return addr; 17511da177e4SLinus Torvalds } 17521da177e4SLinus Torvalds 1753db4fbfb9SMichel Lespinasse info.flags = VM_UNMAPPED_AREA_TOPDOWN; 1754db4fbfb9SMichel Lespinasse info.length = len; 17556b008640SLinus Torvalds info.low_limit = PAGE_SIZE; 1756f6795053SSteve Capper info.high_limit = arch_get_mmap_base(addr, mm->mmap_base); 1757db4fbfb9SMichel Lespinasse info.align_mask = 0; 175809ef5283SJaewon Kim info.align_offset = 0; 1759db4fbfb9SMichel Lespinasse addr = vm_unmapped_area(&info); 1760b716ad95SXiao Guangrong 17611da177e4SLinus Torvalds /* 17621da177e4SLinus Torvalds * A failed mmap() very likely causes application failure, 17631da177e4SLinus Torvalds * so fall back to the bottom-up function here. This scenario 17641da177e4SLinus Torvalds * can happen with large stack limits and large mmap() 17651da177e4SLinus Torvalds * allocations. 17661da177e4SLinus Torvalds */ 1767de1741a1SAlexander Kuleshov if (offset_in_page(addr)) { 1768db4fbfb9SMichel Lespinasse VM_BUG_ON(addr != -ENOMEM); 1769db4fbfb9SMichel Lespinasse info.flags = 0; 1770db4fbfb9SMichel Lespinasse info.low_limit = TASK_UNMAPPED_BASE; 1771f6795053SSteve Capper info.high_limit = mmap_end; 1772db4fbfb9SMichel Lespinasse addr = vm_unmapped_area(&info); 1773db4fbfb9SMichel Lespinasse } 17741da177e4SLinus Torvalds 17751da177e4SLinus Torvalds return addr; 17761da177e4SLinus Torvalds } 17774b439e25SChristophe Leroy 17784b439e25SChristophe Leroy #ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN 17794b439e25SChristophe Leroy unsigned long 17804b439e25SChristophe Leroy arch_get_unmapped_area_topdown(struct file *filp, unsigned long addr, 17814b439e25SChristophe Leroy unsigned long len, unsigned long pgoff, 17824b439e25SChristophe Leroy unsigned long flags) 17834b439e25SChristophe Leroy { 17844b439e25SChristophe Leroy return generic_get_unmapped_area_topdown(filp, addr, len, pgoff, flags); 17854b439e25SChristophe Leroy } 17861da177e4SLinus Torvalds #endif 17871da177e4SLinus Torvalds 17881da177e4SLinus Torvalds unsigned long 17891da177e4SLinus Torvalds get_unmapped_area(struct file *file, unsigned long addr, unsigned long len, 17901da177e4SLinus Torvalds unsigned long pgoff, unsigned long flags) 17911da177e4SLinus Torvalds { 179206abdfb4SBenjamin Herrenschmidt unsigned long (*get_area)(struct file *, unsigned long, 179306abdfb4SBenjamin Herrenschmidt unsigned long, unsigned long, unsigned long); 179407ab67c8SLinus Torvalds 17959206de95SAl Viro unsigned long error = arch_mmap_check(addr, len, flags); 17969206de95SAl Viro if (error) 17979206de95SAl Viro return error; 17989206de95SAl Viro 17999206de95SAl Viro /* Careful about overflows.. */ 18009206de95SAl Viro if (len > TASK_SIZE) 18019206de95SAl Viro return -ENOMEM; 18029206de95SAl Viro 180307ab67c8SLinus Torvalds get_area = current->mm->get_unmapped_area; 1804c01d5b30SHugh Dickins if (file) { 1805c01d5b30SHugh Dickins if (file->f_op->get_unmapped_area) 180607ab67c8SLinus Torvalds get_area = file->f_op->get_unmapped_area; 1807c01d5b30SHugh Dickins } else if (flags & MAP_SHARED) { 1808c01d5b30SHugh Dickins /* 1809c01d5b30SHugh Dickins * mmap_region() will call shmem_zero_setup() to create a file, 1810c01d5b30SHugh Dickins * so use shmem's get_unmapped_area in case it can be huge. 181145e55300SPeter Collingbourne * do_mmap() will clear pgoff, so match alignment. 1812c01d5b30SHugh Dickins */ 1813c01d5b30SHugh Dickins pgoff = 0; 1814c01d5b30SHugh Dickins get_area = shmem_get_unmapped_area; 1815c01d5b30SHugh Dickins } 1816c01d5b30SHugh Dickins 181707ab67c8SLinus Torvalds addr = get_area(file, addr, len, pgoff, flags); 181807ab67c8SLinus Torvalds if (IS_ERR_VALUE(addr)) 181907ab67c8SLinus Torvalds return addr; 182007ab67c8SLinus Torvalds 18211da177e4SLinus Torvalds if (addr > TASK_SIZE - len) 18221da177e4SLinus Torvalds return -ENOMEM; 1823de1741a1SAlexander Kuleshov if (offset_in_page(addr)) 18241da177e4SLinus Torvalds return -EINVAL; 182506abdfb4SBenjamin Herrenschmidt 18269ac4ed4bSAl Viro error = security_mmap_addr(addr); 18279ac4ed4bSAl Viro return error ? error : addr; 18281da177e4SLinus Torvalds } 18291da177e4SLinus Torvalds 18301da177e4SLinus Torvalds EXPORT_SYMBOL(get_unmapped_area); 18311da177e4SLinus Torvalds 1832be8432e7SLiam R. Howlett /** 1833abdba2ddSLiam R. Howlett * find_vma_intersection() - Look up the first VMA which intersects the interval 1834abdba2ddSLiam R. Howlett * @mm: The process address space. 1835abdba2ddSLiam R. Howlett * @start_addr: The inclusive start user address. 1836abdba2ddSLiam R. Howlett * @end_addr: The exclusive end user address. 1837abdba2ddSLiam R. Howlett * 1838abdba2ddSLiam R. Howlett * Returns: The first VMA within the provided range, %NULL otherwise. Assumes 1839abdba2ddSLiam R. Howlett * start_addr < end_addr. 1840abdba2ddSLiam R. Howlett */ 1841abdba2ddSLiam R. Howlett struct vm_area_struct *find_vma_intersection(struct mm_struct *mm, 1842abdba2ddSLiam R. Howlett unsigned long start_addr, 1843abdba2ddSLiam R. Howlett unsigned long end_addr) 1844abdba2ddSLiam R. Howlett { 1845abdba2ddSLiam R. Howlett unsigned long index = start_addr; 1846abdba2ddSLiam R. Howlett 1847abdba2ddSLiam R. Howlett mmap_assert_locked(mm); 18487964cf8cSLiam R. Howlett return mt_find(&mm->mm_mt, &index, end_addr - 1); 1849abdba2ddSLiam R. Howlett } 1850abdba2ddSLiam R. Howlett EXPORT_SYMBOL(find_vma_intersection); 1851abdba2ddSLiam R. Howlett 1852abdba2ddSLiam R. Howlett /** 1853be8432e7SLiam R. Howlett * find_vma() - Find the VMA for a given address, or the next VMA. 1854be8432e7SLiam R. Howlett * @mm: The mm_struct to check 1855be8432e7SLiam R. Howlett * @addr: The address 1856be8432e7SLiam R. Howlett * 1857be8432e7SLiam R. Howlett * Returns: The VMA associated with addr, or the next VMA. 1858be8432e7SLiam R. Howlett * May return %NULL in the case of no VMA at addr or above. 1859be8432e7SLiam R. Howlett */ 18601da177e4SLinus Torvalds struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr) 18611da177e4SLinus Torvalds { 1862be8432e7SLiam R. Howlett unsigned long index = addr; 18631da177e4SLinus Torvalds 18645b78ed24SLuigi Rizzo mmap_assert_locked(mm); 18657964cf8cSLiam R. Howlett return mt_find(&mm->mm_mt, &index, ULONG_MAX); 18661da177e4SLinus Torvalds } 18671da177e4SLinus Torvalds EXPORT_SYMBOL(find_vma); 18681da177e4SLinus Torvalds 18697fdbd37dSLiam R. Howlett /** 18707fdbd37dSLiam R. Howlett * find_vma_prev() - Find the VMA for a given address, or the next vma and 18717fdbd37dSLiam R. Howlett * set %pprev to the previous VMA, if any. 18727fdbd37dSLiam R. Howlett * @mm: The mm_struct to check 18737fdbd37dSLiam R. Howlett * @addr: The address 18747fdbd37dSLiam R. Howlett * @pprev: The pointer to set to the previous VMA 18757fdbd37dSLiam R. Howlett * 18767fdbd37dSLiam R. Howlett * Note that RCU lock is missing here since the external mmap_lock() is used 18777fdbd37dSLiam R. Howlett * instead. 18787fdbd37dSLiam R. Howlett * 18797fdbd37dSLiam R. Howlett * Returns: The VMA associated with @addr, or the next vma. 18807fdbd37dSLiam R. Howlett * May return %NULL in the case of no vma at addr or above. 18816bd4837dSKOSAKI Motohiro */ 18821da177e4SLinus Torvalds struct vm_area_struct * 18831da177e4SLinus Torvalds find_vma_prev(struct mm_struct *mm, unsigned long addr, 18841da177e4SLinus Torvalds struct vm_area_struct **pprev) 18851da177e4SLinus Torvalds { 18866bd4837dSKOSAKI Motohiro struct vm_area_struct *vma; 18877fdbd37dSLiam R. Howlett MA_STATE(mas, &mm->mm_mt, addr, addr); 18881da177e4SLinus Torvalds 18897fdbd37dSLiam R. Howlett vma = mas_walk(&mas); 18907fdbd37dSLiam R. Howlett *pprev = mas_prev(&mas, 0); 18917fdbd37dSLiam R. Howlett if (!vma) 18927fdbd37dSLiam R. Howlett vma = mas_next(&mas, ULONG_MAX); 18936bd4837dSKOSAKI Motohiro return vma; 18941da177e4SLinus Torvalds } 18951da177e4SLinus Torvalds 18961da177e4SLinus Torvalds /* 18971da177e4SLinus Torvalds * Verify that the stack growth is acceptable and 18981da177e4SLinus Torvalds * update accounting. This is shared with both the 18991da177e4SLinus Torvalds * grow-up and grow-down cases. 19001da177e4SLinus Torvalds */ 19011be7107fSHugh Dickins static int acct_stack_growth(struct vm_area_struct *vma, 19021be7107fSHugh Dickins unsigned long size, unsigned long grow) 19031da177e4SLinus Torvalds { 19041da177e4SLinus Torvalds struct mm_struct *mm = vma->vm_mm; 19051be7107fSHugh Dickins unsigned long new_start; 19061da177e4SLinus Torvalds 19071da177e4SLinus Torvalds /* address space limit tests */ 190884638335SKonstantin Khlebnikov if (!may_expand_vm(mm, vma->vm_flags, grow)) 19091da177e4SLinus Torvalds return -ENOMEM; 19101da177e4SLinus Torvalds 19111da177e4SLinus Torvalds /* Stack limit test */ 191224c79d8eSKrzysztof Opasiak if (size > rlimit(RLIMIT_STACK)) 19131da177e4SLinus Torvalds return -ENOMEM; 19141da177e4SLinus Torvalds 19151da177e4SLinus Torvalds /* mlock limit tests */ 1916b0cc5e89SAndrew Morton if (!mlock_future_ok(mm, vma->vm_flags, grow << PAGE_SHIFT)) 19171da177e4SLinus Torvalds return -ENOMEM; 19181da177e4SLinus Torvalds 19190d59a01bSAdam Litke /* Check to ensure the stack will not grow into a hugetlb-only region */ 19200d59a01bSAdam Litke new_start = (vma->vm_flags & VM_GROWSUP) ? vma->vm_start : 19210d59a01bSAdam Litke vma->vm_end - size; 19220d59a01bSAdam Litke if (is_hugepage_only_range(vma->vm_mm, new_start, size)) 19230d59a01bSAdam Litke return -EFAULT; 19240d59a01bSAdam Litke 19251da177e4SLinus Torvalds /* 19261da177e4SLinus Torvalds * Overcommit.. This must be the final test, as it will 19271da177e4SLinus Torvalds * update security statistics. 19281da177e4SLinus Torvalds */ 192905fa199dSHugh Dickins if (security_vm_enough_memory_mm(mm, grow)) 19301da177e4SLinus Torvalds return -ENOMEM; 19311da177e4SLinus Torvalds 19321da177e4SLinus Torvalds return 0; 19331da177e4SLinus Torvalds } 19341da177e4SLinus Torvalds 193546dea3d0SHugh Dickins #if defined(CONFIG_STACK_GROWSUP) || defined(CONFIG_IA64) 19361da177e4SLinus Torvalds /* 193746dea3d0SHugh Dickins * PA-RISC uses this for its stack; IA64 for its Register Backing Store. 193846dea3d0SHugh Dickins * vma is the last one with address > vma->vm_end. Have to extend vma. 19391da177e4SLinus Torvalds */ 19408d7071afSLinus Torvalds static int expand_upwards(struct vm_area_struct *vma, unsigned long address) 19411da177e4SLinus Torvalds { 194209357814SOleg Nesterov struct mm_struct *mm = vma->vm_mm; 19431be7107fSHugh Dickins struct vm_area_struct *next; 19441be7107fSHugh Dickins unsigned long gap_addr; 194512352d3cSKonstantin Khlebnikov int error = 0; 1946d4af56c5SLiam R. Howlett MA_STATE(mas, &mm->mm_mt, 0, 0); 19471da177e4SLinus Torvalds 19481da177e4SLinus Torvalds if (!(vma->vm_flags & VM_GROWSUP)) 19491da177e4SLinus Torvalds return -EFAULT; 19501da177e4SLinus Torvalds 1951bd726c90SHelge Deller /* Guard against exceeding limits of the address space. */ 19521be7107fSHugh Dickins address &= PAGE_MASK; 195337511fb5SHelge Deller if (address >= (TASK_SIZE & PAGE_MASK)) 195412352d3cSKonstantin Khlebnikov return -ENOMEM; 1955bd726c90SHelge Deller address += PAGE_SIZE; 195612352d3cSKonstantin Khlebnikov 19571be7107fSHugh Dickins /* Enforce stack_guard_gap */ 19581be7107fSHugh Dickins gap_addr = address + stack_guard_gap; 1959bd726c90SHelge Deller 1960bd726c90SHelge Deller /* Guard against overflow */ 1961bd726c90SHelge Deller if (gap_addr < address || gap_addr > TASK_SIZE) 1962bd726c90SHelge Deller gap_addr = TASK_SIZE; 1963bd726c90SHelge Deller 1964763ecb03SLiam R. Howlett next = find_vma_intersection(mm, vma->vm_end, gap_addr); 1965763ecb03SLiam R. Howlett if (next && vma_is_accessible(next)) { 19661be7107fSHugh Dickins if (!(next->vm_flags & VM_GROWSUP)) 19671be7107fSHugh Dickins return -ENOMEM; 19681be7107fSHugh Dickins /* Check that both stack segments have the same anon_vma? */ 19691be7107fSHugh Dickins } 19701be7107fSHugh Dickins 1971c5d5546eSVernon Yang if (mas_preallocate(&mas, GFP_KERNEL)) 19721da177e4SLinus Torvalds return -ENOMEM; 19731da177e4SLinus Torvalds 1974d4af56c5SLiam R. Howlett /* We must make sure the anon_vma is allocated. */ 1975d4af56c5SLiam R. Howlett if (unlikely(anon_vma_prepare(vma))) { 1976d4af56c5SLiam R. Howlett mas_destroy(&mas); 1977d4af56c5SLiam R. Howlett return -ENOMEM; 1978d4af56c5SLiam R. Howlett } 1979d4af56c5SLiam R. Howlett 1980c137381fSSuren Baghdasaryan /* Lock the VMA before expanding to prevent concurrent page faults */ 1981c137381fSSuren Baghdasaryan vma_start_write(vma); 19821da177e4SLinus Torvalds /* 19831da177e4SLinus Torvalds * vma->vm_start/vm_end cannot change under us because the caller 1984c1e8d7c6SMichel Lespinasse * is required to hold the mmap_lock in read mode. We need the 19851da177e4SLinus Torvalds * anon_vma lock to serialize against concurrent expand_stacks. 19861da177e4SLinus Torvalds */ 198712352d3cSKonstantin Khlebnikov anon_vma_lock_write(vma->anon_vma); 19881da177e4SLinus Torvalds 19891da177e4SLinus Torvalds /* Somebody else might have raced and expanded it already */ 19901da177e4SLinus Torvalds if (address > vma->vm_end) { 19911da177e4SLinus Torvalds unsigned long size, grow; 19921da177e4SLinus Torvalds 19931da177e4SLinus Torvalds size = address - vma->vm_start; 19941da177e4SLinus Torvalds grow = (address - vma->vm_end) >> PAGE_SHIFT; 19951da177e4SLinus Torvalds 199642c36f63SHugh Dickins error = -ENOMEM; 199742c36f63SHugh Dickins if (vma->vm_pgoff + (size >> PAGE_SHIFT) >= vma->vm_pgoff) { 19981da177e4SLinus Torvalds error = acct_stack_growth(vma, size, grow); 19993af9e859SEric B Munson if (!error) { 20004128997bSMichel Lespinasse /* 2001524e00b3SLiam R. Howlett * We only hold a shared mmap_lock lock here, so 2002524e00b3SLiam R. Howlett * we need to protect against concurrent vma 2003524e00b3SLiam R. Howlett * expansions. anon_vma_lock_write() doesn't 2004524e00b3SLiam R. Howlett * help here, as we don't guarantee that all 2005524e00b3SLiam R. Howlett * growable vmas in a mm share the same root 2006524e00b3SLiam R. Howlett * anon vma. So, we reuse mm->page_table_lock 2007524e00b3SLiam R. Howlett * to guard against concurrent vma expansions. 20084128997bSMichel Lespinasse */ 200909357814SOleg Nesterov spin_lock(&mm->page_table_lock); 201087e8827bSOleg Nesterov if (vma->vm_flags & VM_LOCKED) 201109357814SOleg Nesterov mm->locked_vm += grow; 201284638335SKonstantin Khlebnikov vm_stat_account(mm, vma->vm_flags, grow); 2013bf181b9fSMichel Lespinasse anon_vma_interval_tree_pre_update_vma(vma); 20141da177e4SLinus Torvalds vma->vm_end = address; 2015d4af56c5SLiam R. Howlett /* Overwrite old entry in mtree. */ 2016fbcc3104SLiam R. Howlett mas_set_range(&mas, vma->vm_start, address - 1); 2017fbcc3104SLiam R. Howlett mas_store_prealloc(&mas, vma); 2018bf181b9fSMichel Lespinasse anon_vma_interval_tree_post_update_vma(vma); 201909357814SOleg Nesterov spin_unlock(&mm->page_table_lock); 20204128997bSMichel Lespinasse 20213af9e859SEric B Munson perf_event_mmap(vma); 20223af9e859SEric B Munson } 20231da177e4SLinus Torvalds } 202442c36f63SHugh Dickins } 202512352d3cSKonstantin Khlebnikov anon_vma_unlock_write(vma->anon_vma); 2026c791576cSYang Shi khugepaged_enter_vma(vma, vma->vm_flags); 2027d4af56c5SLiam R. Howlett mas_destroy(&mas); 20281da177e4SLinus Torvalds return error; 20291da177e4SLinus Torvalds } 203046dea3d0SHugh Dickins #endif /* CONFIG_STACK_GROWSUP || CONFIG_IA64 */ 203146dea3d0SHugh Dickins 20321da177e4SLinus Torvalds /* 20331da177e4SLinus Torvalds * vma is the first one with address < vma->vm_start. Have to extend vma. 20348d7071afSLinus Torvalds * mmap_lock held for writing. 20351da177e4SLinus Torvalds */ 2036524e00b3SLiam R. Howlett int expand_downwards(struct vm_area_struct *vma, unsigned long address) 20371da177e4SLinus Torvalds { 203809357814SOleg Nesterov struct mm_struct *mm = vma->vm_mm; 2039763ecb03SLiam R. Howlett MA_STATE(mas, &mm->mm_mt, vma->vm_start, vma->vm_start); 20401be7107fSHugh Dickins struct vm_area_struct *prev; 20410a1d5299SJann Horn int error = 0; 20421da177e4SLinus Torvalds 20438d7071afSLinus Torvalds if (!(vma->vm_flags & VM_GROWSDOWN)) 20448d7071afSLinus Torvalds return -EFAULT; 20458d7071afSLinus Torvalds 20468869477aSEric Paris address &= PAGE_MASK; 20478b35ca3eSBen Hutchings if (address < mmap_min_addr || address < FIRST_USER_ADDRESS) 20480a1d5299SJann Horn return -EPERM; 20498869477aSEric Paris 20501be7107fSHugh Dickins /* Enforce stack_guard_gap */ 2051763ecb03SLiam R. Howlett prev = mas_prev(&mas, 0); 20521be7107fSHugh Dickins /* Check that both stack segments have the same anon_vma? */ 2053f440fa1aSLiam R. Howlett if (prev) { 2054f440fa1aSLiam R. Howlett if (!(prev->vm_flags & VM_GROWSDOWN) && 2055f440fa1aSLiam R. Howlett vma_is_accessible(prev) && 2056f440fa1aSLiam R. Howlett (address - prev->vm_end < stack_guard_gap)) 205732e4e6d5SOleg Nesterov return -ENOMEM; 20581be7107fSHugh Dickins } 20591be7107fSHugh Dickins 2060c5d5546eSVernon Yang if (mas_preallocate(&mas, GFP_KERNEL)) 206112352d3cSKonstantin Khlebnikov return -ENOMEM; 20621da177e4SLinus Torvalds 2063d4af56c5SLiam R. Howlett /* We must make sure the anon_vma is allocated. */ 2064d4af56c5SLiam R. Howlett if (unlikely(anon_vma_prepare(vma))) { 2065d4af56c5SLiam R. Howlett mas_destroy(&mas); 2066d4af56c5SLiam R. Howlett return -ENOMEM; 2067d4af56c5SLiam R. Howlett } 2068d4af56c5SLiam R. Howlett 2069c137381fSSuren Baghdasaryan /* Lock the VMA before expanding to prevent concurrent page faults */ 2070c137381fSSuren Baghdasaryan vma_start_write(vma); 20711da177e4SLinus Torvalds /* 20721da177e4SLinus Torvalds * vma->vm_start/vm_end cannot change under us because the caller 2073c1e8d7c6SMichel Lespinasse * is required to hold the mmap_lock in read mode. We need the 20741da177e4SLinus Torvalds * anon_vma lock to serialize against concurrent expand_stacks. 20751da177e4SLinus Torvalds */ 207612352d3cSKonstantin Khlebnikov anon_vma_lock_write(vma->anon_vma); 20771da177e4SLinus Torvalds 20781da177e4SLinus Torvalds /* Somebody else might have raced and expanded it already */ 20791da177e4SLinus Torvalds if (address < vma->vm_start) { 20801da177e4SLinus Torvalds unsigned long size, grow; 20811da177e4SLinus Torvalds 20821da177e4SLinus Torvalds size = vma->vm_end - address; 20831da177e4SLinus Torvalds grow = (vma->vm_start - address) >> PAGE_SHIFT; 20841da177e4SLinus Torvalds 2085a626ca6aSLinus Torvalds error = -ENOMEM; 2086a626ca6aSLinus Torvalds if (grow <= vma->vm_pgoff) { 20871da177e4SLinus Torvalds error = acct_stack_growth(vma, size, grow); 20881da177e4SLinus Torvalds if (!error) { 20894128997bSMichel Lespinasse /* 2090524e00b3SLiam R. Howlett * We only hold a shared mmap_lock lock here, so 2091524e00b3SLiam R. Howlett * we need to protect against concurrent vma 2092524e00b3SLiam R. Howlett * expansions. anon_vma_lock_write() doesn't 2093524e00b3SLiam R. Howlett * help here, as we don't guarantee that all 2094524e00b3SLiam R. Howlett * growable vmas in a mm share the same root 2095524e00b3SLiam R. Howlett * anon vma. So, we reuse mm->page_table_lock 2096524e00b3SLiam R. Howlett * to guard against concurrent vma expansions. 20974128997bSMichel Lespinasse */ 209809357814SOleg Nesterov spin_lock(&mm->page_table_lock); 209987e8827bSOleg Nesterov if (vma->vm_flags & VM_LOCKED) 210009357814SOleg Nesterov mm->locked_vm += grow; 210184638335SKonstantin Khlebnikov vm_stat_account(mm, vma->vm_flags, grow); 2102bf181b9fSMichel Lespinasse anon_vma_interval_tree_pre_update_vma(vma); 21031da177e4SLinus Torvalds vma->vm_start = address; 21041da177e4SLinus Torvalds vma->vm_pgoff -= grow; 2105d4af56c5SLiam R. Howlett /* Overwrite old entry in mtree. */ 2106fbcc3104SLiam R. Howlett mas_set_range(&mas, address, vma->vm_end - 1); 2107fbcc3104SLiam R. Howlett mas_store_prealloc(&mas, vma); 2108bf181b9fSMichel Lespinasse anon_vma_interval_tree_post_update_vma(vma); 210909357814SOleg Nesterov spin_unlock(&mm->page_table_lock); 21104128997bSMichel Lespinasse 21113af9e859SEric B Munson perf_event_mmap(vma); 21121da177e4SLinus Torvalds } 21131da177e4SLinus Torvalds } 2114a626ca6aSLinus Torvalds } 211512352d3cSKonstantin Khlebnikov anon_vma_unlock_write(vma->anon_vma); 2116c791576cSYang Shi khugepaged_enter_vma(vma, vma->vm_flags); 2117d4af56c5SLiam R. Howlett mas_destroy(&mas); 21181da177e4SLinus Torvalds return error; 21191da177e4SLinus Torvalds } 21201da177e4SLinus Torvalds 21211be7107fSHugh Dickins /* enforced gap between the expanding stack and other mappings. */ 21221be7107fSHugh Dickins unsigned long stack_guard_gap = 256UL<<PAGE_SHIFT; 21231be7107fSHugh Dickins 21241be7107fSHugh Dickins static int __init cmdline_parse_stack_guard_gap(char *p) 21251be7107fSHugh Dickins { 21261be7107fSHugh Dickins unsigned long val; 21271be7107fSHugh Dickins char *endptr; 21281be7107fSHugh Dickins 21291be7107fSHugh Dickins val = simple_strtoul(p, &endptr, 10); 21301be7107fSHugh Dickins if (!*endptr) 21311be7107fSHugh Dickins stack_guard_gap = val << PAGE_SHIFT; 21321be7107fSHugh Dickins 2133e6d09493SRandy Dunlap return 1; 21341be7107fSHugh Dickins } 21351be7107fSHugh Dickins __setup("stack_guard_gap=", cmdline_parse_stack_guard_gap); 21361be7107fSHugh Dickins 2137b6a2fea3SOllie Wild #ifdef CONFIG_STACK_GROWSUP 21388d7071afSLinus Torvalds int expand_stack_locked(struct vm_area_struct *vma, unsigned long address) 2139b6a2fea3SOllie Wild { 2140b6a2fea3SOllie Wild return expand_upwards(vma, address); 2141b6a2fea3SOllie Wild } 2142b6a2fea3SOllie Wild 21438d7071afSLinus Torvalds struct vm_area_struct *find_extend_vma_locked(struct mm_struct *mm, unsigned long addr) 2144b6a2fea3SOllie Wild { 2145b6a2fea3SOllie Wild struct vm_area_struct *vma, *prev; 2146b6a2fea3SOllie Wild 2147b6a2fea3SOllie Wild addr &= PAGE_MASK; 2148b6a2fea3SOllie Wild vma = find_vma_prev(mm, addr, &prev); 2149b6a2fea3SOllie Wild if (vma && (vma->vm_start <= addr)) 2150b6a2fea3SOllie Wild return vma; 2151f440fa1aSLiam R. Howlett if (!prev) 2152f440fa1aSLiam R. Howlett return NULL; 21538d7071afSLinus Torvalds if (expand_stack_locked(prev, addr)) 2154b6a2fea3SOllie Wild return NULL; 2155cea10a19SMichel Lespinasse if (prev->vm_flags & VM_LOCKED) 2156fc05f566SKirill A. Shutemov populate_vma_page_range(prev, addr, prev->vm_end, NULL); 2157b6a2fea3SOllie Wild return prev; 2158b6a2fea3SOllie Wild } 2159b6a2fea3SOllie Wild #else 21608d7071afSLinus Torvalds int expand_stack_locked(struct vm_area_struct *vma, unsigned long address) 2161b6a2fea3SOllie Wild { 2162f440fa1aSLiam R. Howlett if (unlikely(!(vma->vm_flags & VM_GROWSDOWN))) 2163f440fa1aSLiam R. Howlett return -EINVAL; 2164b6a2fea3SOllie Wild return expand_downwards(vma, address); 2165b6a2fea3SOllie Wild } 2166b6a2fea3SOllie Wild 21678d7071afSLinus Torvalds struct vm_area_struct *find_extend_vma_locked(struct mm_struct *mm, unsigned long addr) 21681da177e4SLinus Torvalds { 21691da177e4SLinus Torvalds struct vm_area_struct *vma; 21701da177e4SLinus Torvalds unsigned long start; 21711da177e4SLinus Torvalds 21721da177e4SLinus Torvalds addr &= PAGE_MASK; 21731da177e4SLinus Torvalds vma = find_vma(mm, addr); 21741da177e4SLinus Torvalds if (!vma) 21751da177e4SLinus Torvalds return NULL; 21761da177e4SLinus Torvalds if (vma->vm_start <= addr) 21771da177e4SLinus Torvalds return vma; 21781da177e4SLinus Torvalds start = vma->vm_start; 21798d7071afSLinus Torvalds if (expand_stack_locked(vma, addr)) 21801da177e4SLinus Torvalds return NULL; 2181cea10a19SMichel Lespinasse if (vma->vm_flags & VM_LOCKED) 2182fc05f566SKirill A. Shutemov populate_vma_page_range(vma, addr, start, NULL); 21831da177e4SLinus Torvalds return vma; 21841da177e4SLinus Torvalds } 21851da177e4SLinus Torvalds #endif 21861da177e4SLinus Torvalds 21878d7071afSLinus Torvalds /* 21888d7071afSLinus Torvalds * IA64 has some horrid mapping rules: it can expand both up and down, 21898d7071afSLinus Torvalds * but with various special rules. 21908d7071afSLinus Torvalds * 21918d7071afSLinus Torvalds * We'll get rid of this architecture eventually, so the ugliness is 21928d7071afSLinus Torvalds * temporary. 21938d7071afSLinus Torvalds */ 21948d7071afSLinus Torvalds #ifdef CONFIG_IA64 21958d7071afSLinus Torvalds static inline bool vma_expand_ok(struct vm_area_struct *vma, unsigned long addr) 2196f440fa1aSLiam R. Howlett { 21978d7071afSLinus Torvalds return REGION_NUMBER(addr) == REGION_NUMBER(vma->vm_start) && 21988d7071afSLinus Torvalds REGION_OFFSET(addr) < RGN_MAP_LIMIT; 2199f440fa1aSLiam R. Howlett } 22008d7071afSLinus Torvalds 22018d7071afSLinus Torvalds /* 22028d7071afSLinus Torvalds * IA64 stacks grow down, but there's a special register backing store 22038d7071afSLinus Torvalds * that can grow up. Only sequentially, though, so the new address must 22048d7071afSLinus Torvalds * match vm_end. 22058d7071afSLinus Torvalds */ 22068d7071afSLinus Torvalds static inline int vma_expand_up(struct vm_area_struct *vma, unsigned long addr) 22078d7071afSLinus Torvalds { 22088d7071afSLinus Torvalds if (!vma_expand_ok(vma, addr)) 22098d7071afSLinus Torvalds return -EFAULT; 22108d7071afSLinus Torvalds if (vma->vm_end != (addr & PAGE_MASK)) 22118d7071afSLinus Torvalds return -EFAULT; 22128d7071afSLinus Torvalds return expand_upwards(vma, addr); 22138d7071afSLinus Torvalds } 22148d7071afSLinus Torvalds 22158d7071afSLinus Torvalds static inline bool vma_expand_down(struct vm_area_struct *vma, unsigned long addr) 22168d7071afSLinus Torvalds { 22178d7071afSLinus Torvalds if (!vma_expand_ok(vma, addr)) 22188d7071afSLinus Torvalds return -EFAULT; 22198d7071afSLinus Torvalds return expand_downwards(vma, addr); 22208d7071afSLinus Torvalds } 22218d7071afSLinus Torvalds 22228d7071afSLinus Torvalds #elif defined(CONFIG_STACK_GROWSUP) 22238d7071afSLinus Torvalds 22248d7071afSLinus Torvalds #define vma_expand_up(vma,addr) expand_upwards(vma, addr) 22258d7071afSLinus Torvalds #define vma_expand_down(vma, addr) (-EFAULT) 22268d7071afSLinus Torvalds 22278d7071afSLinus Torvalds #else 22288d7071afSLinus Torvalds 22298d7071afSLinus Torvalds #define vma_expand_up(vma,addr) (-EFAULT) 22308d7071afSLinus Torvalds #define vma_expand_down(vma, addr) expand_downwards(vma, addr) 22318d7071afSLinus Torvalds 22328d7071afSLinus Torvalds #endif 22338d7071afSLinus Torvalds 22348d7071afSLinus Torvalds /* 22358d7071afSLinus Torvalds * expand_stack(): legacy interface for page faulting. Don't use unless 22368d7071afSLinus Torvalds * you have to. 22378d7071afSLinus Torvalds * 22388d7071afSLinus Torvalds * This is called with the mm locked for reading, drops the lock, takes 22398d7071afSLinus Torvalds * the lock for writing, tries to look up a vma again, expands it if 22408d7071afSLinus Torvalds * necessary, and downgrades the lock to reading again. 22418d7071afSLinus Torvalds * 22428d7071afSLinus Torvalds * If no vma is found or it can't be expanded, it returns NULL and has 22438d7071afSLinus Torvalds * dropped the lock. 22448d7071afSLinus Torvalds */ 22458d7071afSLinus Torvalds struct vm_area_struct *expand_stack(struct mm_struct *mm, unsigned long addr) 22468d7071afSLinus Torvalds { 22478d7071afSLinus Torvalds struct vm_area_struct *vma, *prev; 22488d7071afSLinus Torvalds 22498d7071afSLinus Torvalds mmap_read_unlock(mm); 22508d7071afSLinus Torvalds if (mmap_write_lock_killable(mm)) 22518d7071afSLinus Torvalds return NULL; 22528d7071afSLinus Torvalds 22538d7071afSLinus Torvalds vma = find_vma_prev(mm, addr, &prev); 22548d7071afSLinus Torvalds if (vma && vma->vm_start <= addr) 22558d7071afSLinus Torvalds goto success; 22568d7071afSLinus Torvalds 22578d7071afSLinus Torvalds if (prev && !vma_expand_up(prev, addr)) { 22588d7071afSLinus Torvalds vma = prev; 22598d7071afSLinus Torvalds goto success; 22608d7071afSLinus Torvalds } 22618d7071afSLinus Torvalds 22628d7071afSLinus Torvalds if (vma && !vma_expand_down(vma, addr)) 22638d7071afSLinus Torvalds goto success; 22648d7071afSLinus Torvalds 22658d7071afSLinus Torvalds mmap_write_unlock(mm); 22668d7071afSLinus Torvalds return NULL; 22678d7071afSLinus Torvalds 22688d7071afSLinus Torvalds success: 22698d7071afSLinus Torvalds mmap_write_downgrade(mm); 22708d7071afSLinus Torvalds return vma; 22718d7071afSLinus Torvalds } 2272e1d6d01aSJesse Barnes 22732c0b3814SHugh Dickins /* 2274763ecb03SLiam R. Howlett * Ok - we have the memory areas we should free on a maple tree so release them, 2275763ecb03SLiam R. Howlett * and do the vma updates. 22761da177e4SLinus Torvalds * 22772c0b3814SHugh Dickins * Called with the mm semaphore held. 22781da177e4SLinus Torvalds */ 2279763ecb03SLiam R. Howlett static inline void remove_mt(struct mm_struct *mm, struct ma_state *mas) 22801da177e4SLinus Torvalds { 22814f74d2c8SLinus Torvalds unsigned long nr_accounted = 0; 2282763ecb03SLiam R. Howlett struct vm_area_struct *vma; 22834f74d2c8SLinus Torvalds 2284365e9c87SHugh Dickins /* Update high watermark before we lower total_vm */ 2285365e9c87SHugh Dickins update_hiwater_vm(mm); 2286763ecb03SLiam R. Howlett mas_for_each(mas, vma, ULONG_MAX) { 2287ab50b8edSHugh Dickins long nrpages = vma_pages(vma); 22881da177e4SLinus Torvalds 22894f74d2c8SLinus Torvalds if (vma->vm_flags & VM_ACCOUNT) 22904f74d2c8SLinus Torvalds nr_accounted += nrpages; 229184638335SKonstantin Khlebnikov vm_stat_account(mm, vma->vm_flags, -nrpages); 22920d2ebf9cSSuren Baghdasaryan remove_vma(vma, false); 2293763ecb03SLiam R. Howlett } 22944f74d2c8SLinus Torvalds vm_unacct_memory(nr_accounted); 22951da177e4SLinus Torvalds validate_mm(mm); 22961da177e4SLinus Torvalds } 22971da177e4SLinus Torvalds 22981da177e4SLinus Torvalds /* 22991da177e4SLinus Torvalds * Get rid of page table information in the indicated region. 23001da177e4SLinus Torvalds * 2301f10df686SPaolo 'Blaisorblade' Giarrusso * Called with the mm semaphore held. 23021da177e4SLinus Torvalds */ 2303763ecb03SLiam R. Howlett static void unmap_region(struct mm_struct *mm, struct maple_tree *mt, 2304e0da382cSHugh Dickins struct vm_area_struct *vma, struct vm_area_struct *prev, 2305763ecb03SLiam R. Howlett struct vm_area_struct *next, 230668f48381SSuren Baghdasaryan unsigned long start, unsigned long end, bool mm_wr_locked) 23071da177e4SLinus Torvalds { 2308d16dfc55SPeter Zijlstra struct mmu_gather tlb; 23091da177e4SLinus Torvalds 23101da177e4SLinus Torvalds lru_add_drain(); 2311a72afd87SWill Deacon tlb_gather_mmu(&tlb, mm); 2312365e9c87SHugh Dickins update_hiwater_rss(mm); 231368f48381SSuren Baghdasaryan unmap_vmas(&tlb, mt, vma, start, end, mm_wr_locked); 2314763ecb03SLiam R. Howlett free_pgtables(&tlb, mt, vma, prev ? prev->vm_end : FIRST_USER_ADDRESS, 231598e51a22SSuren Baghdasaryan next ? next->vm_start : USER_PGTABLES_CEILING, 231698e51a22SSuren Baghdasaryan mm_wr_locked); 2317ae8eba8bSWill Deacon tlb_finish_mmu(&tlb); 23181da177e4SLinus Torvalds } 23191da177e4SLinus Torvalds 23201da177e4SLinus Torvalds /* 2321def5efe0SDavid Rientjes * __split_vma() bypasses sysctl_max_map_count checking. We use this where it 2322def5efe0SDavid Rientjes * has already been checked or doesn't make sense to fail. 23230fd5a9e2SLiam R. Howlett * VMA Iterator will point to the end VMA. 23241da177e4SLinus Torvalds */ 23259760ebffSLiam R. Howlett int __split_vma(struct vma_iterator *vmi, struct vm_area_struct *vma, 23261da177e4SLinus Torvalds unsigned long addr, int new_below) 23271da177e4SLinus Torvalds { 2328b2b3b886SLiam R. Howlett struct vma_prepare vp; 23291da177e4SLinus Torvalds struct vm_area_struct *new; 2330e3975891SChen Gang int err; 23319760ebffSLiam R. Howlett 2332b50e195fSLiam R. Howlett validate_mm(vma->vm_mm); 23331da177e4SLinus Torvalds 2334b2b3b886SLiam R. Howlett WARN_ON(vma->vm_start >= addr); 2335b2b3b886SLiam R. Howlett WARN_ON(vma->vm_end <= addr); 2336b2b3b886SLiam R. Howlett 2337dd3b614fSDmitry Safonov if (vma->vm_ops && vma->vm_ops->may_split) { 2338dd3b614fSDmitry Safonov err = vma->vm_ops->may_split(vma, addr); 233931383c68SDan Williams if (err) 234031383c68SDan Williams return err; 234131383c68SDan Williams } 23421da177e4SLinus Torvalds 23433928d4f5SLinus Torvalds new = vm_area_dup(vma); 23441da177e4SLinus Torvalds if (!new) 2345e3975891SChen Gang return -ENOMEM; 23461da177e4SLinus Torvalds 2347b2b3b886SLiam R. Howlett err = -ENOMEM; 2348b2b3b886SLiam R. Howlett if (vma_iter_prealloc(vmi)) 2349b2b3b886SLiam R. Howlett goto out_free_vma; 2350b2b3b886SLiam R. Howlett 2351b2b3b886SLiam R. Howlett if (new_below) { 23521da177e4SLinus Torvalds new->vm_end = addr; 2353b2b3b886SLiam R. Howlett } else { 23541da177e4SLinus Torvalds new->vm_start = addr; 23551da177e4SLinus Torvalds new->vm_pgoff += ((addr - vma->vm_start) >> PAGE_SHIFT); 23561da177e4SLinus Torvalds } 23571da177e4SLinus Torvalds 2358ef0855d3SOleg Nesterov err = vma_dup_policy(vma, new); 2359ef0855d3SOleg Nesterov if (err) 2360b2b3b886SLiam R. Howlett goto out_free_vmi; 23611da177e4SLinus Torvalds 2362c4ea95d7SDaniel Forrest err = anon_vma_clone(new, vma); 2363c4ea95d7SDaniel Forrest if (err) 23645beb4930SRik van Riel goto out_free_mpol; 23655beb4930SRik van Riel 2366e9714acfSKonstantin Khlebnikov if (new->vm_file) 23671da177e4SLinus Torvalds get_file(new->vm_file); 23681da177e4SLinus Torvalds 23691da177e4SLinus Torvalds if (new->vm_ops && new->vm_ops->open) 23701da177e4SLinus Torvalds new->vm_ops->open(new); 23711da177e4SLinus Torvalds 2372b2b3b886SLiam R. Howlett init_vma_prep(&vp, vma); 2373b2b3b886SLiam R. Howlett vp.insert = new; 2374b2b3b886SLiam R. Howlett vma_prepare(&vp); 2375ccf1d78dSSuren Baghdasaryan vma_adjust_trans_huge(vma, vma->vm_start, addr, 0); 23761da177e4SLinus Torvalds 2377b2b3b886SLiam R. Howlett if (new_below) { 2378b2b3b886SLiam R. Howlett vma->vm_start = addr; 2379b2b3b886SLiam R. Howlett vma->vm_pgoff += (addr - new->vm_start) >> PAGE_SHIFT; 2380b2b3b886SLiam R. Howlett } else { 2381b2b3b886SLiam R. Howlett vma->vm_end = addr; 23829760ebffSLiam R. Howlett } 23835beb4930SRik van Riel 2384b2b3b886SLiam R. Howlett /* vma_complete stores the new vma */ 2385b2b3b886SLiam R. Howlett vma_complete(&vp, vmi, vma->vm_mm); 2386b2b3b886SLiam R. Howlett 2387b2b3b886SLiam R. Howlett /* Success. */ 2388b2b3b886SLiam R. Howlett if (new_below) 2389b2b3b886SLiam R. Howlett vma_next(vmi); 2390b50e195fSLiam R. Howlett validate_mm(vma->vm_mm); 2391b2b3b886SLiam R. Howlett return 0; 2392b2b3b886SLiam R. Howlett 23935beb4930SRik van Riel out_free_mpol: 2394ef0855d3SOleg Nesterov mpol_put(vma_policy(new)); 2395b2b3b886SLiam R. Howlett out_free_vmi: 2396b2b3b886SLiam R. Howlett vma_iter_free(vmi); 23975beb4930SRik van Riel out_free_vma: 23983928d4f5SLinus Torvalds vm_area_free(new); 2399b50e195fSLiam R. Howlett validate_mm(vma->vm_mm); 24005beb4930SRik van Riel return err; 24011da177e4SLinus Torvalds } 24021da177e4SLinus Torvalds 2403659ace58SKOSAKI Motohiro /* 2404659ace58SKOSAKI Motohiro * Split a vma into two pieces at address 'addr', a new vma is allocated 2405659ace58SKOSAKI Motohiro * either for the first part or the tail. 2406659ace58SKOSAKI Motohiro */ 24079760ebffSLiam R. Howlett int split_vma(struct vma_iterator *vmi, struct vm_area_struct *vma, 2408659ace58SKOSAKI Motohiro unsigned long addr, int new_below) 2409659ace58SKOSAKI Motohiro { 24109760ebffSLiam R. Howlett if (vma->vm_mm->map_count >= sysctl_max_map_count) 2411659ace58SKOSAKI Motohiro return -ENOMEM; 2412659ace58SKOSAKI Motohiro 24139760ebffSLiam R. Howlett return __split_vma(vmi, vma, addr, new_below); 2414f2ebfe43SLiam R. Howlett } 2415f2ebfe43SLiam R. Howlett 241611f9a21aSLiam R. Howlett /* 2417183654ceSLiam R. Howlett * do_vmi_align_munmap() - munmap the aligned region from @start to @end. 2418183654ceSLiam R. Howlett * @vmi: The vma iterator 241911f9a21aSLiam R. Howlett * @vma: The starting vm_area_struct 242011f9a21aSLiam R. Howlett * @mm: The mm_struct 242111f9a21aSLiam R. Howlett * @start: The aligned start address to munmap. 242211f9a21aSLiam R. Howlett * @end: The aligned end address to munmap. 242311f9a21aSLiam R. Howlett * @uf: The userfaultfd list_head 2424408579cdSLiam R. Howlett * @unlock: Set to true to drop the mmap_lock. unlocking only happens on 2425408579cdSLiam R. Howlett * success. 242611f9a21aSLiam R. Howlett * 2427408579cdSLiam R. Howlett * Return: 0 on success and drops the lock if so directed, error and leaves the 2428408579cdSLiam R. Howlett * lock held otherwise. 242911f9a21aSLiam R. Howlett */ 243011f9a21aSLiam R. Howlett static int 2431183654ceSLiam R. Howlett do_vmi_align_munmap(struct vma_iterator *vmi, struct vm_area_struct *vma, 243211f9a21aSLiam R. Howlett struct mm_struct *mm, unsigned long start, 2433408579cdSLiam R. Howlett unsigned long end, struct list_head *uf, bool unlock) 243411f9a21aSLiam R. Howlett { 2435763ecb03SLiam R. Howlett struct vm_area_struct *prev, *next = NULL; 2436763ecb03SLiam R. Howlett struct maple_tree mt_detach; 2437763ecb03SLiam R. Howlett int count = 0; 2438d4af56c5SLiam R. Howlett int error = -ENOMEM; 2439606c812eSLiam R. Howlett unsigned long locked_vm = 0; 2440763ecb03SLiam R. Howlett MA_STATE(mas_detach, &mt_detach, 0, 0); 24413dd44325SLiam R. Howlett mt_init_flags(&mt_detach, vmi->mas.tree->ma_flags & MT_FLAGS_LOCK_MASK); 2442763ecb03SLiam R. Howlett mt_set_external_lock(&mt_detach, &mm->mmap_lock); 2443524e00b3SLiam R. Howlett 24441da177e4SLinus Torvalds /* 24451da177e4SLinus Torvalds * If we need to split any vma, do it now to save pain later. 24461da177e4SLinus Torvalds * 24471da177e4SLinus Torvalds * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially 24481da177e4SLinus Torvalds * unmapped vm_area_struct will remain in use: so lower split_vma 24491da177e4SLinus Torvalds * places tmp vma above, and higher split_vma places tmp vma below. 24501da177e4SLinus Torvalds */ 2451763ecb03SLiam R. Howlett 2452763ecb03SLiam R. Howlett /* Does it split the first one? */ 2453146425a3SHugh Dickins if (start > vma->vm_start) { 2454659ace58SKOSAKI Motohiro 2455659ace58SKOSAKI Motohiro /* 2456659ace58SKOSAKI Motohiro * Make sure that map_count on return from munmap() will 2457659ace58SKOSAKI Motohiro * not exceed its limit; but let map_count go just above 2458659ace58SKOSAKI Motohiro * its limit temporarily, to help free resources as expected. 2459659ace58SKOSAKI Motohiro */ 2460659ace58SKOSAKI Motohiro if (end < vma->vm_end && mm->map_count >= sysctl_max_map_count) 2461d4af56c5SLiam R. Howlett goto map_count_exceeded; 2462659ace58SKOSAKI Motohiro 24639760ebffSLiam R. Howlett error = __split_vma(vmi, vma, start, 0); 24641da177e4SLinus Torvalds if (error) 2465763ecb03SLiam R. Howlett goto start_split_failed; 246611f9a21aSLiam R. Howlett 24670fd5a9e2SLiam R. Howlett vma = vma_iter_load(vmi); 24681da177e4SLinus Torvalds } 24691da177e4SLinus Torvalds 2470183654ceSLiam R. Howlett prev = vma_prev(vmi); 2471763ecb03SLiam R. Howlett if (unlikely((!prev))) 2472183654ceSLiam R. Howlett vma_iter_set(vmi, start); 247311f9a21aSLiam R. Howlett 2474763ecb03SLiam R. Howlett /* 2475763ecb03SLiam R. Howlett * Detach a range of VMAs from the mm. Using next as a temp variable as 2476763ecb03SLiam R. Howlett * it is always overwritten. 2477763ecb03SLiam R. Howlett */ 2478183654ceSLiam R. Howlett for_each_vma_range(*vmi, next, end) { 2479763ecb03SLiam R. Howlett /* Does it split the end? */ 2480763ecb03SLiam R. Howlett if (next->vm_end > end) { 24816b73cff2SLiam R. Howlett error = __split_vma(vmi, next, end, 0); 24821da177e4SLinus Torvalds if (error) 2483763ecb03SLiam R. Howlett goto end_split_failed; 248411f9a21aSLiam R. Howlett } 2485606c812eSLiam R. Howlett vma_start_write(next); 2486606c812eSLiam R. Howlett mas_set_range(&mas_detach, next->vm_start, next->vm_end - 1); 24876c26bd43SDavid Woodhouse error = mas_store_gfp(&mas_detach, next, GFP_KERNEL); 24886c26bd43SDavid Woodhouse if (error) 2489606c812eSLiam R. Howlett goto munmap_gather_failed; 2490606c812eSLiam R. Howlett vma_mark_detached(next, true); 2491606c812eSLiam R. Howlett if (next->vm_flags & VM_LOCKED) 2492606c812eSLiam R. Howlett locked_vm += vma_pages(next); 2493763ecb03SLiam R. Howlett 2494763ecb03SLiam R. Howlett count++; 24952376dd7cSAndrea Arcangeli if (unlikely(uf)) { 24962376dd7cSAndrea Arcangeli /* 24972376dd7cSAndrea Arcangeli * If userfaultfd_unmap_prep returns an error the vmas 2498f0953a1bSIngo Molnar * will remain split, but userland will get a 24992376dd7cSAndrea Arcangeli * highly unexpected error anyway. This is no 25002376dd7cSAndrea Arcangeli * different than the case where the first of the two 25012376dd7cSAndrea Arcangeli * __split_vma fails, but we don't undo the first 25022376dd7cSAndrea Arcangeli * split, despite we could. This is unlikely enough 25032376dd7cSAndrea Arcangeli * failure that it's not worth optimizing it for. 25042376dd7cSAndrea Arcangeli */ 250565ac1320SLiam R. Howlett error = userfaultfd_unmap_prep(next, start, end, uf); 250611f9a21aSLiam R. Howlett 25072376dd7cSAndrea Arcangeli if (error) 2508d4af56c5SLiam R. Howlett goto userfaultfd_error; 25092376dd7cSAndrea Arcangeli } 2510ba470de4SRik van Riel #ifdef CONFIG_DEBUG_VM_MAPLE_TREE 2511ba470de4SRik van Riel BUG_ON(next->vm_start < start); 2512ba470de4SRik van Riel BUG_ON(next->vm_start > end); 25131da177e4SLinus Torvalds #endif 25141da177e4SLinus Torvalds } 2515146425a3SHugh Dickins 251615c0c60bSLiam R. Howlett if (vma_iter_end(vmi) > end) 251715c0c60bSLiam R. Howlett next = vma_iter_load(vmi); 251815c0c60bSLiam R. Howlett 251915c0c60bSLiam R. Howlett if (!next) 25201da177e4SLinus Torvalds next = vma_next(vmi); 25212376dd7cSAndrea Arcangeli 2522763ecb03SLiam R. Howlett #if defined(CONFIG_DEBUG_VM_MAPLE_TREE) 2523763ecb03SLiam R. Howlett /* Make sure no VMAs are about to be lost. */ 2524763ecb03SLiam R. Howlett { 2525763ecb03SLiam R. Howlett MA_STATE(test, &mt_detach, start, end - 1); 2526763ecb03SLiam R. Howlett struct vm_area_struct *vma_mas, *vma_test; 2527763ecb03SLiam R. Howlett int test_count = 0; 2528763ecb03SLiam R. Howlett 2529183654ceSLiam R. Howlett vma_iter_set(vmi, start); 2530763ecb03SLiam R. Howlett rcu_read_lock(); 2531763ecb03SLiam R. Howlett vma_test = mas_find(&test, end - 1); 2532183654ceSLiam R. Howlett for_each_vma_range(*vmi, vma_mas, end) { 2533763ecb03SLiam R. Howlett BUG_ON(vma_mas != vma_test); 2534763ecb03SLiam R. Howlett test_count++; 2535763ecb03SLiam R. Howlett vma_test = mas_next(&test, end - 1); 2536763ecb03SLiam R. Howlett } 2537763ecb03SLiam R. Howlett rcu_read_unlock(); 2538763ecb03SLiam R. Howlett BUG_ON(count != test_count); 2539763ecb03SLiam R. Howlett } 2540763ecb03SLiam R. Howlett #endif 2541183654ceSLiam R. Howlett vma_iter_set(vmi, start); 25426c26bd43SDavid Woodhouse error = vma_iter_clear_gfp(vmi, start, end, GFP_KERNEL); 25436c26bd43SDavid Woodhouse if (error) 2544606c812eSLiam R. Howlett goto clear_tree_failed; 25450378c0a0SLiam R. Howlett 25466c26bd43SDavid Woodhouse /* Point of no return */ 2547606c812eSLiam R. Howlett mm->locked_vm -= locked_vm; 2548763ecb03SLiam R. Howlett mm->map_count -= count; 2549408579cdSLiam R. Howlett if (unlock) 2550d8ed45c5SMichel Lespinasse mmap_write_downgrade(mm); 2551dd2283f2SYang Shi 255268f48381SSuren Baghdasaryan /* 255368f48381SSuren Baghdasaryan * We can free page tables without write-locking mmap_lock because VMAs 255468f48381SSuren Baghdasaryan * were isolated before we downgraded mmap_lock. 255568f48381SSuren Baghdasaryan */ 2556408579cdSLiam R. Howlett unmap_region(mm, &mt_detach, vma, prev, next, start, end, !unlock); 2557763ecb03SLiam R. Howlett /* Statistics and freeing VMAs */ 2558763ecb03SLiam R. Howlett mas_set(&mas_detach, start); 2559763ecb03SLiam R. Howlett remove_mt(mm, &mas_detach); 2560763ecb03SLiam R. Howlett __mt_destroy(&mt_detach); 2561ae80b404SLinus Torvalds validate_mm(mm); 2562408579cdSLiam R. Howlett if (unlock) 2563408579cdSLiam R. Howlett mmap_read_unlock(mm); 25641da177e4SLinus Torvalds 2565408579cdSLiam R. Howlett return 0; 2566d4af56c5SLiam R. Howlett 2567606c812eSLiam R. Howlett clear_tree_failed: 2568d4af56c5SLiam R. Howlett userfaultfd_error: 2569606c812eSLiam R. Howlett munmap_gather_failed: 2570763ecb03SLiam R. Howlett end_split_failed: 2571606c812eSLiam R. Howlett mas_set(&mas_detach, 0); 2572606c812eSLiam R. Howlett mas_for_each(&mas_detach, next, end) 2573606c812eSLiam R. Howlett vma_mark_detached(next, false); 2574606c812eSLiam R. Howlett 2575763ecb03SLiam R. Howlett __mt_destroy(&mt_detach); 2576763ecb03SLiam R. Howlett start_split_failed: 2577763ecb03SLiam R. Howlett map_count_exceeded: 2578b5641a5dSLinus Torvalds validate_mm(mm); 2579d4af56c5SLiam R. Howlett return error; 25801da177e4SLinus Torvalds } 25811da177e4SLinus Torvalds 258211f9a21aSLiam R. Howlett /* 2583183654ceSLiam R. Howlett * do_vmi_munmap() - munmap a given range. 2584183654ceSLiam R. Howlett * @vmi: The vma iterator 258511f9a21aSLiam R. Howlett * @mm: The mm_struct 258611f9a21aSLiam R. Howlett * @start: The start address to munmap 258711f9a21aSLiam R. Howlett * @len: The length of the range to munmap 258811f9a21aSLiam R. Howlett * @uf: The userfaultfd list_head 2589408579cdSLiam R. Howlett * @unlock: set to true if the user wants to drop the mmap_lock on success 259011f9a21aSLiam R. Howlett * 259111f9a21aSLiam R. Howlett * This function takes a @mas that is either pointing to the previous VMA or set 259211f9a21aSLiam R. Howlett * to MA_START and sets it up to remove the mapping(s). The @len will be 259311f9a21aSLiam R. Howlett * aligned and any arch_unmap work will be preformed. 259411f9a21aSLiam R. Howlett * 2595408579cdSLiam R. Howlett * Return: 0 on success and drops the lock if so directed, error and leaves the 2596408579cdSLiam R. Howlett * lock held otherwise. 259711f9a21aSLiam R. Howlett */ 2598183654ceSLiam R. Howlett int do_vmi_munmap(struct vma_iterator *vmi, struct mm_struct *mm, 259911f9a21aSLiam R. Howlett unsigned long start, size_t len, struct list_head *uf, 2600408579cdSLiam R. Howlett bool unlock) 260111f9a21aSLiam R. Howlett { 260211f9a21aSLiam R. Howlett unsigned long end; 260311f9a21aSLiam R. Howlett struct vm_area_struct *vma; 260411f9a21aSLiam R. Howlett 260511f9a21aSLiam R. Howlett if ((offset_in_page(start)) || start > TASK_SIZE || len > TASK_SIZE-start) 260611f9a21aSLiam R. Howlett return -EINVAL; 260711f9a21aSLiam R. Howlett 260811f9a21aSLiam R. Howlett end = start + PAGE_ALIGN(len); 260911f9a21aSLiam R. Howlett if (end == start) 261011f9a21aSLiam R. Howlett return -EINVAL; 261111f9a21aSLiam R. Howlett 261211f9a21aSLiam R. Howlett /* arch_unmap() might do unmaps itself. */ 261311f9a21aSLiam R. Howlett arch_unmap(mm, start, end); 261411f9a21aSLiam R. Howlett 261511f9a21aSLiam R. Howlett /* Find the first overlapping VMA */ 2616183654ceSLiam R. Howlett vma = vma_find(vmi, end); 2617408579cdSLiam R. Howlett if (!vma) { 2618408579cdSLiam R. Howlett if (unlock) 2619408579cdSLiam R. Howlett mmap_write_unlock(mm); 262011f9a21aSLiam R. Howlett return 0; 2621408579cdSLiam R. Howlett } 262211f9a21aSLiam R. Howlett 2623408579cdSLiam R. Howlett return do_vmi_align_munmap(vmi, vma, mm, start, end, uf, unlock); 262411f9a21aSLiam R. Howlett } 262511f9a21aSLiam R. Howlett 262611f9a21aSLiam R. Howlett /* do_munmap() - Wrapper function for non-maple tree aware do_munmap() calls. 262711f9a21aSLiam R. Howlett * @mm: The mm_struct 262811f9a21aSLiam R. Howlett * @start: The start address to munmap 262911f9a21aSLiam R. Howlett * @len: The length to be munmapped. 263011f9a21aSLiam R. Howlett * @uf: The userfaultfd list_head 2631408579cdSLiam R. Howlett * 2632408579cdSLiam R. Howlett * Return: 0 on success, error otherwise. 263311f9a21aSLiam R. Howlett */ 2634dd2283f2SYang Shi int do_munmap(struct mm_struct *mm, unsigned long start, size_t len, 2635dd2283f2SYang Shi struct list_head *uf) 2636dd2283f2SYang Shi { 2637183654ceSLiam R. Howlett VMA_ITERATOR(vmi, mm, start); 263811f9a21aSLiam R. Howlett 2639183654ceSLiam R. Howlett return do_vmi_munmap(&vmi, mm, start, len, uf, false); 2640dd2283f2SYang Shi } 2641dd2283f2SYang Shi 2642e99668a5SLiam R. Howlett unsigned long mmap_region(struct file *file, unsigned long addr, 2643e99668a5SLiam R. Howlett unsigned long len, vm_flags_t vm_flags, unsigned long pgoff, 2644e99668a5SLiam R. Howlett struct list_head *uf) 2645e99668a5SLiam R. Howlett { 2646e99668a5SLiam R. Howlett struct mm_struct *mm = current->mm; 2647e99668a5SLiam R. Howlett struct vm_area_struct *vma = NULL; 2648e99668a5SLiam R. Howlett struct vm_area_struct *next, *prev, *merge; 2649e99668a5SLiam R. Howlett pgoff_t pglen = len >> PAGE_SHIFT; 2650e99668a5SLiam R. Howlett unsigned long charged = 0; 2651e99668a5SLiam R. Howlett unsigned long end = addr + len; 2652e99668a5SLiam R. Howlett unsigned long merge_start = addr, merge_end = end; 2653e99668a5SLiam R. Howlett pgoff_t vm_pgoff; 2654e99668a5SLiam R. Howlett int error; 2655183654ceSLiam R. Howlett VMA_ITERATOR(vmi, mm, addr); 2656e99668a5SLiam R. Howlett 2657e99668a5SLiam R. Howlett /* Check against address space limit. */ 2658e99668a5SLiam R. Howlett if (!may_expand_vm(mm, vm_flags, len >> PAGE_SHIFT)) { 2659e99668a5SLiam R. Howlett unsigned long nr_pages; 2660e99668a5SLiam R. Howlett 2661e99668a5SLiam R. Howlett /* 2662e99668a5SLiam R. Howlett * MAP_FIXED may remove pages of mappings that intersects with 2663e99668a5SLiam R. Howlett * requested mapping. Account for the pages it would unmap. 2664e99668a5SLiam R. Howlett */ 2665e99668a5SLiam R. Howlett nr_pages = count_vma_pages_range(mm, addr, end); 2666e99668a5SLiam R. Howlett 2667e99668a5SLiam R. Howlett if (!may_expand_vm(mm, vm_flags, 2668e99668a5SLiam R. Howlett (len >> PAGE_SHIFT) - nr_pages)) 2669e99668a5SLiam R. Howlett return -ENOMEM; 2670e99668a5SLiam R. Howlett } 2671e99668a5SLiam R. Howlett 2672e99668a5SLiam R. Howlett /* Unmap any existing mapping in the area */ 2673183654ceSLiam R. Howlett if (do_vmi_munmap(&vmi, mm, addr, len, uf, false)) 2674e99668a5SLiam R. Howlett return -ENOMEM; 2675e99668a5SLiam R. Howlett 2676e99668a5SLiam R. Howlett /* 2677e99668a5SLiam R. Howlett * Private writable mapping: check memory availability 2678e99668a5SLiam R. Howlett */ 2679e99668a5SLiam R. Howlett if (accountable_mapping(file, vm_flags)) { 2680e99668a5SLiam R. Howlett charged = len >> PAGE_SHIFT; 2681e99668a5SLiam R. Howlett if (security_vm_enough_memory_mm(mm, charged)) 2682e99668a5SLiam R. Howlett return -ENOMEM; 2683e99668a5SLiam R. Howlett vm_flags |= VM_ACCOUNT; 2684e99668a5SLiam R. Howlett } 2685e99668a5SLiam R. Howlett 2686183654ceSLiam R. Howlett next = vma_next(&vmi); 2687183654ceSLiam R. Howlett prev = vma_prev(&vmi); 2688e99668a5SLiam R. Howlett if (vm_flags & VM_SPECIAL) 2689e99668a5SLiam R. Howlett goto cannot_expand; 2690e99668a5SLiam R. Howlett 2691e99668a5SLiam R. Howlett /* Attempt to expand an old mapping */ 2692e99668a5SLiam R. Howlett /* Check next */ 2693e99668a5SLiam R. Howlett if (next && next->vm_start == end && !vma_policy(next) && 2694e99668a5SLiam R. Howlett can_vma_merge_before(next, vm_flags, NULL, file, pgoff+pglen, 2695e99668a5SLiam R. Howlett NULL_VM_UFFD_CTX, NULL)) { 2696e99668a5SLiam R. Howlett merge_end = next->vm_end; 2697e99668a5SLiam R. Howlett vma = next; 2698e99668a5SLiam R. Howlett vm_pgoff = next->vm_pgoff - pglen; 2699e99668a5SLiam R. Howlett } 2700e99668a5SLiam R. Howlett 2701e99668a5SLiam R. Howlett /* Check prev */ 2702e99668a5SLiam R. Howlett if (prev && prev->vm_end == addr && !vma_policy(prev) && 2703e99668a5SLiam R. Howlett (vma ? can_vma_merge_after(prev, vm_flags, vma->anon_vma, file, 2704e99668a5SLiam R. Howlett pgoff, vma->vm_userfaultfd_ctx, NULL) : 2705e99668a5SLiam R. Howlett can_vma_merge_after(prev, vm_flags, NULL, file, pgoff, 2706e99668a5SLiam R. Howlett NULL_VM_UFFD_CTX, NULL))) { 2707e99668a5SLiam R. Howlett merge_start = prev->vm_start; 2708e99668a5SLiam R. Howlett vma = prev; 2709e99668a5SLiam R. Howlett vm_pgoff = prev->vm_pgoff; 2710e99668a5SLiam R. Howlett } 2711e99668a5SLiam R. Howlett 2712e99668a5SLiam R. Howlett 2713e99668a5SLiam R. Howlett /* Actually expand, if possible */ 2714e99668a5SLiam R. Howlett if (vma && 27153c441ab7SLiam R. Howlett !vma_expand(&vmi, vma, merge_start, merge_end, vm_pgoff, next)) { 2716e99668a5SLiam R. Howlett khugepaged_enter_vma(vma, vm_flags); 2717e99668a5SLiam R. Howlett goto expanded; 2718e99668a5SLiam R. Howlett } 2719e99668a5SLiam R. Howlett 2720e99668a5SLiam R. Howlett cannot_expand: 27215c1c03deSLiam R. Howlett if (prev) 27225c1c03deSLiam R. Howlett vma_iter_next_range(&vmi); 27235c1c03deSLiam R. Howlett 2724e99668a5SLiam R. Howlett /* 2725e99668a5SLiam R. Howlett * Determine the object being mapped and call the appropriate 2726e99668a5SLiam R. Howlett * specific mapper. the address has already been validated, but 2727e99668a5SLiam R. Howlett * not unmapped, but the maps are removed from the list. 2728e99668a5SLiam R. Howlett */ 2729e99668a5SLiam R. Howlett vma = vm_area_alloc(mm); 2730e99668a5SLiam R. Howlett if (!vma) { 2731e99668a5SLiam R. Howlett error = -ENOMEM; 2732e99668a5SLiam R. Howlett goto unacct_error; 2733e99668a5SLiam R. Howlett } 2734e99668a5SLiam R. Howlett 27350fd5a9e2SLiam R. Howlett vma_iter_set(&vmi, addr); 2736e99668a5SLiam R. Howlett vma->vm_start = addr; 2737e99668a5SLiam R. Howlett vma->vm_end = end; 27381c71222eSSuren Baghdasaryan vm_flags_init(vma, vm_flags); 2739e99668a5SLiam R. Howlett vma->vm_page_prot = vm_get_page_prot(vm_flags); 2740e99668a5SLiam R. Howlett vma->vm_pgoff = pgoff; 2741e99668a5SLiam R. Howlett 2742e99668a5SLiam R. Howlett if (file) { 2743e99668a5SLiam R. Howlett if (vm_flags & VM_SHARED) { 2744e99668a5SLiam R. Howlett error = mapping_map_writable(file->f_mapping); 2745e99668a5SLiam R. Howlett if (error) 2746e99668a5SLiam R. Howlett goto free_vma; 2747e99668a5SLiam R. Howlett } 2748e99668a5SLiam R. Howlett 2749e99668a5SLiam R. Howlett vma->vm_file = get_file(file); 2750e99668a5SLiam R. Howlett error = call_mmap(file, vma); 2751e99668a5SLiam R. Howlett if (error) 2752e99668a5SLiam R. Howlett goto unmap_and_free_vma; 2753e99668a5SLiam R. Howlett 2754a57b7051SLiam Howlett /* 2755a57b7051SLiam Howlett * Expansion is handled above, merging is handled below. 2756a57b7051SLiam Howlett * Drivers should not alter the address of the VMA. 2757e99668a5SLiam R. Howlett */ 2758a57b7051SLiam Howlett error = -EINVAL; 2759cc8d1b09SLiam R. Howlett if (WARN_ON((addr != vma->vm_start))) 2760a57b7051SLiam Howlett goto close_and_free_vma; 2761e99668a5SLiam R. Howlett 2762cc8d1b09SLiam R. Howlett vma_iter_set(&vmi, addr); 2763e99668a5SLiam R. Howlett /* 2764e99668a5SLiam R. Howlett * If vm_flags changed after call_mmap(), we should try merge 2765e99668a5SLiam R. Howlett * vma again as we may succeed this time. 2766e99668a5SLiam R. Howlett */ 2767e99668a5SLiam R. Howlett if (unlikely(vm_flags != vma->vm_flags && prev)) { 27689760ebffSLiam R. Howlett merge = vma_merge(&vmi, mm, prev, vma->vm_start, 27699760ebffSLiam R. Howlett vma->vm_end, vma->vm_flags, NULL, 27709760ebffSLiam R. Howlett vma->vm_file, vma->vm_pgoff, NULL, 27719760ebffSLiam R. Howlett NULL_VM_UFFD_CTX, NULL); 2772e99668a5SLiam R. Howlett if (merge) { 2773e99668a5SLiam R. Howlett /* 2774e99668a5SLiam R. Howlett * ->mmap() can change vma->vm_file and fput 2775e99668a5SLiam R. Howlett * the original file. So fput the vma->vm_file 2776e99668a5SLiam R. Howlett * here or we would add an extra fput for file 2777e99668a5SLiam R. Howlett * and cause general protection fault 2778e99668a5SLiam R. Howlett * ultimately. 2779e99668a5SLiam R. Howlett */ 2780e99668a5SLiam R. Howlett fput(vma->vm_file); 2781e99668a5SLiam R. Howlett vm_area_free(vma); 2782e99668a5SLiam R. Howlett vma = merge; 2783e99668a5SLiam R. Howlett /* Update vm_flags to pick up the change. */ 2784e99668a5SLiam R. Howlett vm_flags = vma->vm_flags; 2785e99668a5SLiam R. Howlett goto unmap_writable; 2786e99668a5SLiam R. Howlett } 2787e99668a5SLiam R. Howlett } 2788e99668a5SLiam R. Howlett 2789e99668a5SLiam R. Howlett vm_flags = vma->vm_flags; 2790e99668a5SLiam R. Howlett } else if (vm_flags & VM_SHARED) { 2791e99668a5SLiam R. Howlett error = shmem_zero_setup(vma); 2792e99668a5SLiam R. Howlett if (error) 2793e99668a5SLiam R. Howlett goto free_vma; 2794e99668a5SLiam R. Howlett } else { 2795e99668a5SLiam R. Howlett vma_set_anonymous(vma); 2796e99668a5SLiam R. Howlett } 2797e99668a5SLiam R. Howlett 2798b507808eSJoey Gouly if (map_deny_write_exec(vma, vma->vm_flags)) { 2799b507808eSJoey Gouly error = -EACCES; 2800b507808eSJoey Gouly goto close_and_free_vma; 2801b507808eSJoey Gouly } 2802b507808eSJoey Gouly 2803e99668a5SLiam R. Howlett /* Allow architectures to sanity-check the vm_flags */ 2804e99668a5SLiam R. Howlett error = -EINVAL; 2805cc8d1b09SLiam R. Howlett if (!arch_validate_flags(vma->vm_flags)) 2806deb0f656SCarlos Llamas goto close_and_free_vma; 2807e99668a5SLiam R. Howlett 2808e99668a5SLiam R. Howlett error = -ENOMEM; 2809cc8d1b09SLiam R. Howlett if (vma_iter_prealloc(&vmi)) 28105789151eSMike Kravetz goto close_and_free_vma; 2811e99668a5SLiam R. Howlett 28121c7873e3SHugh Dickins /* Lock the VMA since it is modified after insertion into VMA tree */ 28131c7873e3SHugh Dickins vma_start_write(vma); 2814e99668a5SLiam R. Howlett if (vma->vm_file) 2815e99668a5SLiam R. Howlett i_mmap_lock_write(vma->vm_file->f_mapping); 2816e99668a5SLiam R. Howlett 2817183654ceSLiam R. Howlett vma_iter_store(&vmi, vma); 2818e99668a5SLiam R. Howlett mm->map_count++; 2819e99668a5SLiam R. Howlett if (vma->vm_file) { 2820e99668a5SLiam R. Howlett if (vma->vm_flags & VM_SHARED) 2821e99668a5SLiam R. Howlett mapping_allow_writable(vma->vm_file->f_mapping); 2822e99668a5SLiam R. Howlett 2823e99668a5SLiam R. Howlett flush_dcache_mmap_lock(vma->vm_file->f_mapping); 2824e99668a5SLiam R. Howlett vma_interval_tree_insert(vma, &vma->vm_file->f_mapping->i_mmap); 2825e99668a5SLiam R. Howlett flush_dcache_mmap_unlock(vma->vm_file->f_mapping); 2826e99668a5SLiam R. Howlett i_mmap_unlock_write(vma->vm_file->f_mapping); 2827e99668a5SLiam R. Howlett } 2828e99668a5SLiam R. Howlett 2829e99668a5SLiam R. Howlett /* 2830e99668a5SLiam R. Howlett * vma_merge() calls khugepaged_enter_vma() either, the below 2831e99668a5SLiam R. Howlett * call covers the non-merge case. 2832e99668a5SLiam R. Howlett */ 2833e99668a5SLiam R. Howlett khugepaged_enter_vma(vma, vma->vm_flags); 2834e99668a5SLiam R. Howlett 2835e99668a5SLiam R. Howlett /* Once vma denies write, undo our temporary denial count */ 2836e99668a5SLiam R. Howlett unmap_writable: 2837e99668a5SLiam R. Howlett if (file && vm_flags & VM_SHARED) 2838e99668a5SLiam R. Howlett mapping_unmap_writable(file->f_mapping); 2839e99668a5SLiam R. Howlett file = vma->vm_file; 2840d7597f59SStefan Roesch ksm_add_vma(vma); 2841e99668a5SLiam R. Howlett expanded: 2842e99668a5SLiam R. Howlett perf_event_mmap(vma); 2843e99668a5SLiam R. Howlett 2844e99668a5SLiam R. Howlett vm_stat_account(mm, vm_flags, len >> PAGE_SHIFT); 2845e99668a5SLiam R. Howlett if (vm_flags & VM_LOCKED) { 2846e99668a5SLiam R. Howlett if ((vm_flags & VM_SPECIAL) || vma_is_dax(vma) || 2847e99668a5SLiam R. Howlett is_vm_hugetlb_page(vma) || 2848e99668a5SLiam R. Howlett vma == get_gate_vma(current->mm)) 2849e430a95aSSuren Baghdasaryan vm_flags_clear(vma, VM_LOCKED_MASK); 2850e99668a5SLiam R. Howlett else 2851e99668a5SLiam R. Howlett mm->locked_vm += (len >> PAGE_SHIFT); 2852e99668a5SLiam R. Howlett } 2853e99668a5SLiam R. Howlett 2854e99668a5SLiam R. Howlett if (file) 2855e99668a5SLiam R. Howlett uprobe_mmap(vma); 2856e99668a5SLiam R. Howlett 2857e99668a5SLiam R. Howlett /* 2858e99668a5SLiam R. Howlett * New (or expanded) vma always get soft dirty status. 2859e99668a5SLiam R. Howlett * Otherwise user-space soft-dirty page tracker won't 2860e99668a5SLiam R. Howlett * be able to distinguish situation when vma area unmapped, 2861e99668a5SLiam R. Howlett * then new mapped in-place (which must be aimed as 2862e99668a5SLiam R. Howlett * a completely new data area). 2863e99668a5SLiam R. Howlett */ 28641c71222eSSuren Baghdasaryan vm_flags_set(vma, VM_SOFTDIRTY); 2865e99668a5SLiam R. Howlett 2866e99668a5SLiam R. Howlett vma_set_page_prot(vma); 2867e99668a5SLiam R. Howlett 2868e99668a5SLiam R. Howlett validate_mm(mm); 2869e99668a5SLiam R. Howlett return addr; 2870e99668a5SLiam R. Howlett 2871deb0f656SCarlos Llamas close_and_free_vma: 2872cc8d1b09SLiam R. Howlett if (file && vma->vm_ops && vma->vm_ops->close) 2873deb0f656SCarlos Llamas vma->vm_ops->close(vma); 2874cc8d1b09SLiam R. Howlett 2875cc8d1b09SLiam R. Howlett if (file || vma->vm_file) { 2876e99668a5SLiam R. Howlett unmap_and_free_vma: 2877e99668a5SLiam R. Howlett fput(vma->vm_file); 2878e99668a5SLiam R. Howlett vma->vm_file = NULL; 2879e99668a5SLiam R. Howlett 2880e99668a5SLiam R. Howlett /* Undo any partial mapping done by a device driver. */ 2881cc8d1b09SLiam R. Howlett unmap_region(mm, &mm->mm_mt, vma, prev, next, vma->vm_start, 288268f48381SSuren Baghdasaryan vma->vm_end, true); 2883cc8d1b09SLiam R. Howlett } 2884cc674ab3SLi Zetao if (file && (vm_flags & VM_SHARED)) 2885e99668a5SLiam R. Howlett mapping_unmap_writable(file->f_mapping); 2886e99668a5SLiam R. Howlett free_vma: 2887e99668a5SLiam R. Howlett vm_area_free(vma); 2888e99668a5SLiam R. Howlett unacct_error: 2889e99668a5SLiam R. Howlett if (charged) 2890e99668a5SLiam R. Howlett vm_unacct_memory(charged); 2891e99668a5SLiam R. Howlett validate_mm(mm); 2892e99668a5SLiam R. Howlett return error; 2893e99668a5SLiam R. Howlett } 2894e99668a5SLiam R. Howlett 2895408579cdSLiam R. Howlett static int __vm_munmap(unsigned long start, size_t len, bool unlock) 2896a46ef99dSLinus Torvalds { 2897a46ef99dSLinus Torvalds int ret; 2898bfce281cSAl Viro struct mm_struct *mm = current->mm; 2899897ab3e0SMike Rapoport LIST_HEAD(uf); 2900183654ceSLiam R. Howlett VMA_ITERATOR(vmi, mm, start); 2901a46ef99dSLinus Torvalds 2902d8ed45c5SMichel Lespinasse if (mmap_write_lock_killable(mm)) 2903ae798783SMichal Hocko return -EINTR; 2904ae798783SMichal Hocko 2905408579cdSLiam R. Howlett ret = do_vmi_munmap(&vmi, mm, start, len, &uf, unlock); 2906408579cdSLiam R. Howlett if (ret || !unlock) 2907d8ed45c5SMichel Lespinasse mmap_write_unlock(mm); 2908dd2283f2SYang Shi 2909897ab3e0SMike Rapoport userfaultfd_unmap_complete(mm, &uf); 2910a46ef99dSLinus Torvalds return ret; 2911a46ef99dSLinus Torvalds } 2912dd2283f2SYang Shi 2913dd2283f2SYang Shi int vm_munmap(unsigned long start, size_t len) 2914dd2283f2SYang Shi { 2915dd2283f2SYang Shi return __vm_munmap(start, len, false); 2916dd2283f2SYang Shi } 2917a46ef99dSLinus Torvalds EXPORT_SYMBOL(vm_munmap); 2918a46ef99dSLinus Torvalds 29196a6160a7SHeiko Carstens SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len) 29201da177e4SLinus Torvalds { 2921ce18d171SCatalin Marinas addr = untagged_addr(addr); 2922dd2283f2SYang Shi return __vm_munmap(addr, len, true); 29231da177e4SLinus Torvalds } 29241da177e4SLinus Torvalds 2925c8d78c18SKirill A. Shutemov 2926c8d78c18SKirill A. Shutemov /* 2927c8d78c18SKirill A. Shutemov * Emulation of deprecated remap_file_pages() syscall. 2928c8d78c18SKirill A. Shutemov */ 2929c8d78c18SKirill A. Shutemov SYSCALL_DEFINE5(remap_file_pages, unsigned long, start, unsigned long, size, 2930c8d78c18SKirill A. Shutemov unsigned long, prot, unsigned long, pgoff, unsigned long, flags) 2931c8d78c18SKirill A. Shutemov { 2932c8d78c18SKirill A. Shutemov 2933c8d78c18SKirill A. Shutemov struct mm_struct *mm = current->mm; 2934c8d78c18SKirill A. Shutemov struct vm_area_struct *vma; 2935c8d78c18SKirill A. Shutemov unsigned long populate = 0; 2936c8d78c18SKirill A. Shutemov unsigned long ret = -EINVAL; 2937c8d78c18SKirill A. Shutemov struct file *file; 2938c8d78c18SKirill A. Shutemov 2939ee65728eSMike Rapoport pr_warn_once("%s (%d) uses deprecated remap_file_pages() syscall. See Documentation/mm/remap_file_pages.rst.\n", 2940c8d78c18SKirill A. Shutemov current->comm, current->pid); 2941c8d78c18SKirill A. Shutemov 2942c8d78c18SKirill A. Shutemov if (prot) 2943c8d78c18SKirill A. Shutemov return ret; 2944c8d78c18SKirill A. Shutemov start = start & PAGE_MASK; 2945c8d78c18SKirill A. Shutemov size = size & PAGE_MASK; 2946c8d78c18SKirill A. Shutemov 2947c8d78c18SKirill A. Shutemov if (start + size <= start) 2948c8d78c18SKirill A. Shutemov return ret; 2949c8d78c18SKirill A. Shutemov 2950c8d78c18SKirill A. Shutemov /* Does pgoff wrap? */ 2951c8d78c18SKirill A. Shutemov if (pgoff + (size >> PAGE_SHIFT) < pgoff) 2952c8d78c18SKirill A. Shutemov return ret; 2953c8d78c18SKirill A. Shutemov 2954d8ed45c5SMichel Lespinasse if (mmap_write_lock_killable(mm)) 2955dc0ef0dfSMichal Hocko return -EINTR; 2956dc0ef0dfSMichal Hocko 29579b593cb2SLiam R. Howlett vma = vma_lookup(mm, start); 2958c8d78c18SKirill A. Shutemov 2959c8d78c18SKirill A. Shutemov if (!vma || !(vma->vm_flags & VM_SHARED)) 2960c8d78c18SKirill A. Shutemov goto out; 2961c8d78c18SKirill A. Shutemov 296248f7df32SKirill A. Shutemov if (start + size > vma->vm_end) { 2963763ecb03SLiam R. Howlett VMA_ITERATOR(vmi, mm, vma->vm_end); 2964763ecb03SLiam R. Howlett struct vm_area_struct *next, *prev = vma; 296548f7df32SKirill A. Shutemov 2966763ecb03SLiam R. Howlett for_each_vma_range(vmi, next, start + size) { 296748f7df32SKirill A. Shutemov /* hole between vmas ? */ 2968763ecb03SLiam R. Howlett if (next->vm_start != prev->vm_end) 296948f7df32SKirill A. Shutemov goto out; 297048f7df32SKirill A. Shutemov 297148f7df32SKirill A. Shutemov if (next->vm_file != vma->vm_file) 297248f7df32SKirill A. Shutemov goto out; 297348f7df32SKirill A. Shutemov 297448f7df32SKirill A. Shutemov if (next->vm_flags != vma->vm_flags) 297548f7df32SKirill A. Shutemov goto out; 297648f7df32SKirill A. Shutemov 29771db43d3fSLiam Howlett if (start + size <= next->vm_end) 29781db43d3fSLiam Howlett break; 29791db43d3fSLiam Howlett 2980763ecb03SLiam R. Howlett prev = next; 298148f7df32SKirill A. Shutemov } 298248f7df32SKirill A. Shutemov 298348f7df32SKirill A. Shutemov if (!next) 2984c8d78c18SKirill A. Shutemov goto out; 2985c8d78c18SKirill A. Shutemov } 2986c8d78c18SKirill A. Shutemov 2987c8d78c18SKirill A. Shutemov prot |= vma->vm_flags & VM_READ ? PROT_READ : 0; 2988c8d78c18SKirill A. Shutemov prot |= vma->vm_flags & VM_WRITE ? PROT_WRITE : 0; 2989c8d78c18SKirill A. Shutemov prot |= vma->vm_flags & VM_EXEC ? PROT_EXEC : 0; 2990c8d78c18SKirill A. Shutemov 2991c8d78c18SKirill A. Shutemov flags &= MAP_NONBLOCK; 2992c8d78c18SKirill A. Shutemov flags |= MAP_SHARED | MAP_FIXED | MAP_POPULATE; 2993fce000b1SLiam Howlett if (vma->vm_flags & VM_LOCKED) 2994c8d78c18SKirill A. Shutemov flags |= MAP_LOCKED; 299548f7df32SKirill A. Shutemov 2996c8d78c18SKirill A. Shutemov file = get_file(vma->vm_file); 299745e55300SPeter Collingbourne ret = do_mmap(vma->vm_file, start, size, 2998*592b5fadSYu-cheng Yu prot, flags, 0, pgoff, &populate, NULL); 2999c8d78c18SKirill A. Shutemov fput(file); 3000c8d78c18SKirill A. Shutemov out: 3001d8ed45c5SMichel Lespinasse mmap_write_unlock(mm); 3002c8d78c18SKirill A. Shutemov if (populate) 3003c8d78c18SKirill A. Shutemov mm_populate(ret, populate); 3004c8d78c18SKirill A. Shutemov if (!IS_ERR_VALUE(ret)) 3005c8d78c18SKirill A. Shutemov ret = 0; 3006c8d78c18SKirill A. Shutemov return ret; 3007c8d78c18SKirill A. Shutemov } 3008c8d78c18SKirill A. Shutemov 30091da177e4SLinus Torvalds /* 301027b26701SLiam R. Howlett * do_vma_munmap() - Unmap a full or partial vma. 301127b26701SLiam R. Howlett * @vmi: The vma iterator pointing at the vma 301227b26701SLiam R. Howlett * @vma: The first vma to be munmapped 301327b26701SLiam R. Howlett * @start: the start of the address to unmap 301427b26701SLiam R. Howlett * @end: The end of the address to unmap 30152e7ce7d3SLiam R. Howlett * @uf: The userfaultfd list_head 3016408579cdSLiam R. Howlett * @unlock: Drop the lock on success 30172e7ce7d3SLiam R. Howlett * 301827b26701SLiam R. Howlett * unmaps a VMA mapping when the vma iterator is already in position. 301927b26701SLiam R. Howlett * Does not handle alignment. 3020408579cdSLiam R. Howlett * 3021408579cdSLiam R. Howlett * Return: 0 on success drops the lock of so directed, error on failure and will 3022408579cdSLiam R. Howlett * still hold the lock. 30231da177e4SLinus Torvalds */ 302427b26701SLiam R. Howlett int do_vma_munmap(struct vma_iterator *vmi, struct vm_area_struct *vma, 3025408579cdSLiam R. Howlett unsigned long start, unsigned long end, struct list_head *uf, 3026408579cdSLiam R. Howlett bool unlock) 30272e7ce7d3SLiam R. Howlett { 30282e7ce7d3SLiam R. Howlett struct mm_struct *mm = vma->vm_mm; 30292e7ce7d3SLiam R. Howlett 303027b26701SLiam R. Howlett arch_unmap(mm, start, end); 3031b5641a5dSLinus Torvalds return do_vmi_align_munmap(vmi, vma, mm, start, end, uf, unlock); 30322e7ce7d3SLiam R. Howlett } 30332e7ce7d3SLiam R. Howlett 30342e7ce7d3SLiam R. Howlett /* 30352e7ce7d3SLiam R. Howlett * do_brk_flags() - Increase the brk vma if the flags match. 303692fed820SLiam R. Howlett * @vmi: The vma iterator 30372e7ce7d3SLiam R. Howlett * @addr: The start address 30382e7ce7d3SLiam R. Howlett * @len: The length of the increase 30392e7ce7d3SLiam R. Howlett * @vma: The vma, 30402e7ce7d3SLiam R. Howlett * @flags: The VMA Flags 30412e7ce7d3SLiam R. Howlett * 30422e7ce7d3SLiam R. Howlett * Extend the brk VMA from addr to addr + len. If the VMA is NULL or the flags 30432e7ce7d3SLiam R. Howlett * do not match then create a new anonymous VMA. Eventually we may be able to 30442e7ce7d3SLiam R. Howlett * do some brk-specific accounting here. 30452e7ce7d3SLiam R. Howlett */ 304692fed820SLiam R. Howlett static int do_brk_flags(struct vma_iterator *vmi, struct vm_area_struct *vma, 3047763ecb03SLiam R. Howlett unsigned long addr, unsigned long len, unsigned long flags) 30481da177e4SLinus Torvalds { 30491da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 3050287051b1SLiam R. Howlett struct vma_prepare vp; 30512e7ce7d3SLiam R. Howlett 3052b50e195fSLiam R. Howlett validate_mm(mm); 30532e7ce7d3SLiam R. Howlett /* 30542e7ce7d3SLiam R. Howlett * Check against address space limits by the changed size 30552e7ce7d3SLiam R. Howlett * Note: This happens *after* clearing old mappings in some code paths. 30562e7ce7d3SLiam R. Howlett */ 305716e72e9bSDenys Vlasenko flags |= VM_DATA_DEFAULT_FLAGS | VM_ACCOUNT | mm->def_flags; 305884638335SKonstantin Khlebnikov if (!may_expand_vm(mm, flags, len >> PAGE_SHIFT)) 30591da177e4SLinus Torvalds return -ENOMEM; 30601da177e4SLinus Torvalds 30611da177e4SLinus Torvalds if (mm->map_count > sysctl_max_map_count) 30621da177e4SLinus Torvalds return -ENOMEM; 30631da177e4SLinus Torvalds 3064191c5424SAl Viro if (security_vm_enough_memory_mm(mm, len >> PAGE_SHIFT)) 30651da177e4SLinus Torvalds return -ENOMEM; 30661da177e4SLinus Torvalds 30671da177e4SLinus Torvalds /* 30682e7ce7d3SLiam R. Howlett * Expand the existing vma if possible; Note that singular lists do not 30692e7ce7d3SLiam R. Howlett * occur after forking, so the expand will only happen on new VMAs. 30701da177e4SLinus Torvalds */ 30716c28ca64SLiam Howlett if (vma && vma->vm_end == addr && !vma_policy(vma) && 30726c28ca64SLiam Howlett can_vma_merge_after(vma, flags, NULL, NULL, 30736c28ca64SLiam Howlett addr >> PAGE_SHIFT, NULL_VM_UFFD_CTX, NULL)) { 307492fed820SLiam R. Howlett if (vma_iter_prealloc(vmi)) 3075675eaca1SAlistair Popple goto unacct_fail; 307628c5609fSLiam Howlett 3077287051b1SLiam R. Howlett init_vma_prep(&vp, vma); 3078287051b1SLiam R. Howlett vma_prepare(&vp); 3079ccf1d78dSSuren Baghdasaryan vma_adjust_trans_huge(vma, vma->vm_start, addr + len, 0); 30802e7ce7d3SLiam R. Howlett vma->vm_end = addr + len; 30811c71222eSSuren Baghdasaryan vm_flags_set(vma, VM_SOFTDIRTY); 308292fed820SLiam R. Howlett vma_iter_store(vmi, vma); 30832e7ce7d3SLiam R. Howlett 3084287051b1SLiam R. Howlett vma_complete(&vp, vmi, mm); 30852e7ce7d3SLiam R. Howlett khugepaged_enter_vma(vma, flags); 30862e7ce7d3SLiam R. Howlett goto out; 30872e7ce7d3SLiam R. Howlett } 30882e7ce7d3SLiam R. Howlett 30892e7ce7d3SLiam R. Howlett /* create a vma struct for an anonymous mapping */ 30902e7ce7d3SLiam R. Howlett vma = vm_area_alloc(mm); 30912e7ce7d3SLiam R. Howlett if (!vma) 3092675eaca1SAlistair Popple goto unacct_fail; 30931da177e4SLinus Torvalds 3094bfd40eafSKirill A. Shutemov vma_set_anonymous(vma); 30951da177e4SLinus Torvalds vma->vm_start = addr; 30961da177e4SLinus Torvalds vma->vm_end = addr + len; 30972e7ce7d3SLiam R. Howlett vma->vm_pgoff = addr >> PAGE_SHIFT; 30981c71222eSSuren Baghdasaryan vm_flags_init(vma, flags); 30993ed75eb8SColy Li vma->vm_page_prot = vm_get_page_prot(flags); 310092fed820SLiam R. Howlett if (vma_iter_store_gfp(vmi, vma, GFP_KERNEL)) 31012e7ce7d3SLiam R. Howlett goto mas_store_fail; 3102d4af56c5SLiam R. Howlett 31032e7ce7d3SLiam R. Howlett mm->map_count++; 3104d7597f59SStefan Roesch ksm_add_vma(vma); 31051da177e4SLinus Torvalds out: 31063af9e859SEric B Munson perf_event_mmap(vma); 31071da177e4SLinus Torvalds mm->total_vm += len >> PAGE_SHIFT; 310884638335SKonstantin Khlebnikov mm->data_vm += len >> PAGE_SHIFT; 3109128557ffSMichel Lespinasse if (flags & VM_LOCKED) 3110ba470de4SRik van Riel mm->locked_vm += (len >> PAGE_SHIFT); 31111c71222eSSuren Baghdasaryan vm_flags_set(vma, VM_SOFTDIRTY); 3112763ecb03SLiam R. Howlett validate_mm(mm); 31135d22fc25SLinus Torvalds return 0; 3114d4af56c5SLiam R. Howlett 31152e7ce7d3SLiam R. Howlett mas_store_fail: 3116d4af56c5SLiam R. Howlett vm_area_free(vma); 3117675eaca1SAlistair Popple unacct_fail: 31182e7ce7d3SLiam R. Howlett vm_unacct_memory(len >> PAGE_SHIFT); 31192e7ce7d3SLiam R. Howlett return -ENOMEM; 31201da177e4SLinus Torvalds } 31211da177e4SLinus Torvalds 3122bb177a73SMichal Hocko int vm_brk_flags(unsigned long addr, unsigned long request, unsigned long flags) 3123e4eb1ff6SLinus Torvalds { 3124e4eb1ff6SLinus Torvalds struct mm_struct *mm = current->mm; 31252e7ce7d3SLiam R. Howlett struct vm_area_struct *vma = NULL; 3126bb177a73SMichal Hocko unsigned long len; 31275d22fc25SLinus Torvalds int ret; 3128128557ffSMichel Lespinasse bool populate; 3129897ab3e0SMike Rapoport LIST_HEAD(uf); 313092fed820SLiam R. Howlett VMA_ITERATOR(vmi, mm, addr); 3131e4eb1ff6SLinus Torvalds 3132bb177a73SMichal Hocko len = PAGE_ALIGN(request); 3133bb177a73SMichal Hocko if (len < request) 3134bb177a73SMichal Hocko return -ENOMEM; 3135bb177a73SMichal Hocko if (!len) 3136bb177a73SMichal Hocko return 0; 3137bb177a73SMichal Hocko 3138d8ed45c5SMichel Lespinasse if (mmap_write_lock_killable(mm)) 31392d6c9282SMichal Hocko return -EINTR; 31402d6c9282SMichal Hocko 31412e7ce7d3SLiam R. Howlett /* Until we need other flags, refuse anything except VM_EXEC. */ 31422e7ce7d3SLiam R. Howlett if ((flags & (~VM_EXEC)) != 0) 31432e7ce7d3SLiam R. Howlett return -EINVAL; 31442e7ce7d3SLiam R. Howlett 31452e7ce7d3SLiam R. Howlett ret = check_brk_limits(addr, len); 31462e7ce7d3SLiam R. Howlett if (ret) 31472e7ce7d3SLiam R. Howlett goto limits_failed; 31482e7ce7d3SLiam R. Howlett 3149183654ceSLiam R. Howlett ret = do_vmi_munmap(&vmi, mm, addr, len, &uf, 0); 31502e7ce7d3SLiam R. Howlett if (ret) 31512e7ce7d3SLiam R. Howlett goto munmap_failed; 31522e7ce7d3SLiam R. Howlett 315392fed820SLiam R. Howlett vma = vma_prev(&vmi); 315492fed820SLiam R. Howlett ret = do_brk_flags(&vmi, vma, addr, len, flags); 3155128557ffSMichel Lespinasse populate = ((mm->def_flags & VM_LOCKED) != 0); 3156d8ed45c5SMichel Lespinasse mmap_write_unlock(mm); 3157897ab3e0SMike Rapoport userfaultfd_unmap_complete(mm, &uf); 31585d22fc25SLinus Torvalds if (populate && !ret) 3159128557ffSMichel Lespinasse mm_populate(addr, len); 3160e4eb1ff6SLinus Torvalds return ret; 31612e7ce7d3SLiam R. Howlett 31622e7ce7d3SLiam R. Howlett munmap_failed: 31632e7ce7d3SLiam R. Howlett limits_failed: 31642e7ce7d3SLiam R. Howlett mmap_write_unlock(mm); 31652e7ce7d3SLiam R. Howlett return ret; 3166e4eb1ff6SLinus Torvalds } 316716e72e9bSDenys Vlasenko EXPORT_SYMBOL(vm_brk_flags); 316816e72e9bSDenys Vlasenko 316916e72e9bSDenys Vlasenko int vm_brk(unsigned long addr, unsigned long len) 317016e72e9bSDenys Vlasenko { 317116e72e9bSDenys Vlasenko return vm_brk_flags(addr, len, 0); 317216e72e9bSDenys Vlasenko } 3173e4eb1ff6SLinus Torvalds EXPORT_SYMBOL(vm_brk); 31741da177e4SLinus Torvalds 31751da177e4SLinus Torvalds /* Release all mmaps. */ 31761da177e4SLinus Torvalds void exit_mmap(struct mm_struct *mm) 31771da177e4SLinus Torvalds { 3178d16dfc55SPeter Zijlstra struct mmu_gather tlb; 3179ba470de4SRik van Riel struct vm_area_struct *vma; 31801da177e4SLinus Torvalds unsigned long nr_accounted = 0; 3181763ecb03SLiam R. Howlett MA_STATE(mas, &mm->mm_mt, 0, 0); 3182763ecb03SLiam R. Howlett int count = 0; 31831da177e4SLinus Torvalds 3184d6dd61c8SJeremy Fitzhardinge /* mm's last user has gone, and its about to be pulled down */ 3185cddb8a5cSAndrea Arcangeli mmu_notifier_release(mm); 3186d6dd61c8SJeremy Fitzhardinge 3187bf3980c8SSuren Baghdasaryan mmap_read_lock(mm); 31889480c53eSJeremy Fitzhardinge arch_exit_mmap(mm); 31899480c53eSJeremy Fitzhardinge 3190763ecb03SLiam R. Howlett vma = mas_find(&mas, ULONG_MAX); 319164591e86SSuren Baghdasaryan if (!vma) { 319264591e86SSuren Baghdasaryan /* Can happen if dup_mmap() received an OOM */ 3193bf3980c8SSuren Baghdasaryan mmap_read_unlock(mm); 31949480c53eSJeremy Fitzhardinge return; 319564591e86SSuren Baghdasaryan } 31969480c53eSJeremy Fitzhardinge 31971da177e4SLinus Torvalds lru_add_drain(); 31981da177e4SLinus Torvalds flush_cache_mm(mm); 3199d8b45053SWill Deacon tlb_gather_mmu_fullmm(&tlb, mm); 3200901608d9SOleg Nesterov /* update_hiwater_rss(mm) here? but nobody should be looking */ 3201763ecb03SLiam R. Howlett /* Use ULONG_MAX here to ensure all VMAs in the mm are unmapped */ 320268f48381SSuren Baghdasaryan unmap_vmas(&tlb, &mm->mm_mt, vma, 0, ULONG_MAX, false); 3203bf3980c8SSuren Baghdasaryan mmap_read_unlock(mm); 3204bf3980c8SSuren Baghdasaryan 3205bf3980c8SSuren Baghdasaryan /* 3206bf3980c8SSuren Baghdasaryan * Set MMF_OOM_SKIP to hide this task from the oom killer/reaper 3207b3541d91SSuren Baghdasaryan * because the memory has been already freed. 3208bf3980c8SSuren Baghdasaryan */ 3209bf3980c8SSuren Baghdasaryan set_bit(MMF_OOM_SKIP, &mm->flags); 3210bf3980c8SSuren Baghdasaryan mmap_write_lock(mm); 32113dd44325SLiam R. Howlett mt_clear_in_rcu(&mm->mm_mt); 3212763ecb03SLiam R. Howlett free_pgtables(&tlb, &mm->mm_mt, vma, FIRST_USER_ADDRESS, 321398e51a22SSuren Baghdasaryan USER_PGTABLES_CEILING, true); 3214ae8eba8bSWill Deacon tlb_finish_mmu(&tlb); 32151da177e4SLinus Torvalds 3216763ecb03SLiam R. Howlett /* 3217763ecb03SLiam R. Howlett * Walk the list again, actually closing and freeing it, with preemption 3218763ecb03SLiam R. Howlett * enabled, without holding any MM locks besides the unreachable 3219763ecb03SLiam R. Howlett * mmap_write_lock. 3220763ecb03SLiam R. Howlett */ 3221763ecb03SLiam R. Howlett do { 32224f74d2c8SLinus Torvalds if (vma->vm_flags & VM_ACCOUNT) 32234f74d2c8SLinus Torvalds nr_accounted += vma_pages(vma); 32240d2ebf9cSSuren Baghdasaryan remove_vma(vma, true); 3225763ecb03SLiam R. Howlett count++; 32260a3b3c25SPaul E. McKenney cond_resched(); 3227763ecb03SLiam R. Howlett } while ((vma = mas_find(&mas, ULONG_MAX)) != NULL); 3228763ecb03SLiam R. Howlett 3229763ecb03SLiam R. Howlett BUG_ON(count != mm->map_count); 3230d4af56c5SLiam R. Howlett 3231d4af56c5SLiam R. Howlett trace_exit_mmap(mm); 3232d4af56c5SLiam R. Howlett __mt_destroy(&mm->mm_mt); 323364591e86SSuren Baghdasaryan mmap_write_unlock(mm); 32344f74d2c8SLinus Torvalds vm_unacct_memory(nr_accounted); 32351da177e4SLinus Torvalds } 32361da177e4SLinus Torvalds 32371da177e4SLinus Torvalds /* Insert vm structure into process list sorted by address 32381da177e4SLinus Torvalds * and into the inode's i_mmap tree. If vm_file is non-NULL 3239c8c06efaSDavidlohr Bueso * then i_mmap_rwsem is taken here. 32401da177e4SLinus Torvalds */ 32411da177e4SLinus Torvalds int insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma) 32421da177e4SLinus Torvalds { 3243d4af56c5SLiam R. Howlett unsigned long charged = vma_pages(vma); 32441da177e4SLinus Torvalds 3245d4af56c5SLiam R. Howlett 3246d0601a50SLiam R. Howlett if (find_vma_intersection(mm, vma->vm_start, vma->vm_end)) 3247c9d13f5fSChen Gang return -ENOMEM; 3248d4af56c5SLiam R. Howlett 3249c9d13f5fSChen Gang if ((vma->vm_flags & VM_ACCOUNT) && 3250d4af56c5SLiam R. Howlett security_vm_enough_memory_mm(mm, charged)) 3251c9d13f5fSChen Gang return -ENOMEM; 3252c9d13f5fSChen Gang 32531da177e4SLinus Torvalds /* 32541da177e4SLinus Torvalds * The vm_pgoff of a purely anonymous vma should be irrelevant 32551da177e4SLinus Torvalds * until its first write fault, when page's anon_vma and index 32561da177e4SLinus Torvalds * are set. But now set the vm_pgoff it will almost certainly 32571da177e4SLinus Torvalds * end up with (unless mremap moves it elsewhere before that 32581da177e4SLinus Torvalds * first wfault), so /proc/pid/maps tells a consistent story. 32591da177e4SLinus Torvalds * 32601da177e4SLinus Torvalds * By setting it to reflect the virtual start address of the 32611da177e4SLinus Torvalds * vma, merges and splits can happen in a seamless way, just 32621da177e4SLinus Torvalds * using the existing file pgoff checks and manipulations. 32638332326eSLiao Pingfang * Similarly in do_mmap and in do_brk_flags. 32641da177e4SLinus Torvalds */ 32658a9cc3b5SOleg Nesterov if (vma_is_anonymous(vma)) { 32661da177e4SLinus Torvalds BUG_ON(vma->anon_vma); 32671da177e4SLinus Torvalds vma->vm_pgoff = vma->vm_start >> PAGE_SHIFT; 32681da177e4SLinus Torvalds } 32692b144498SSrikar Dronamraju 3270763ecb03SLiam R. Howlett if (vma_link(mm, vma)) { 3271d4af56c5SLiam R. Howlett vm_unacct_memory(charged); 3272d4af56c5SLiam R. Howlett return -ENOMEM; 3273d4af56c5SLiam R. Howlett } 3274d4af56c5SLiam R. Howlett 32751da177e4SLinus Torvalds return 0; 32761da177e4SLinus Torvalds } 32771da177e4SLinus Torvalds 32781da177e4SLinus Torvalds /* 32791da177e4SLinus Torvalds * Copy the vma structure to a new location in the same mm, 32801da177e4SLinus Torvalds * prior to moving page table entries, to effect an mremap move. 32811da177e4SLinus Torvalds */ 32821da177e4SLinus Torvalds struct vm_area_struct *copy_vma(struct vm_area_struct **vmap, 328338a76013SMichel Lespinasse unsigned long addr, unsigned long len, pgoff_t pgoff, 328438a76013SMichel Lespinasse bool *need_rmap_locks) 32851da177e4SLinus Torvalds { 32861da177e4SLinus Torvalds struct vm_area_struct *vma = *vmap; 32871da177e4SLinus Torvalds unsigned long vma_start = vma->vm_start; 32881da177e4SLinus Torvalds struct mm_struct *mm = vma->vm_mm; 32891da177e4SLinus Torvalds struct vm_area_struct *new_vma, *prev; 3290948f017bSAndrea Arcangeli bool faulted_in_anon_vma = true; 3291076f16bfSLiam R. Howlett VMA_ITERATOR(vmi, mm, addr); 32921da177e4SLinus Torvalds 3293b50e195fSLiam R. Howlett validate_mm(mm); 32941da177e4SLinus Torvalds /* 32951da177e4SLinus Torvalds * If anonymous vma has not yet been faulted, update new pgoff 32961da177e4SLinus Torvalds * to match new location, to increase its chance of merging. 32971da177e4SLinus Torvalds */ 3298ce75799bSOleg Nesterov if (unlikely(vma_is_anonymous(vma) && !vma->anon_vma)) { 32991da177e4SLinus Torvalds pgoff = addr >> PAGE_SHIFT; 3300948f017bSAndrea Arcangeli faulted_in_anon_vma = false; 3301948f017bSAndrea Arcangeli } 33021da177e4SLinus Torvalds 3303763ecb03SLiam R. Howlett new_vma = find_vma_prev(mm, addr, &prev); 3304763ecb03SLiam R. Howlett if (new_vma && new_vma->vm_start < addr + len) 33056597d783SHugh Dickins return NULL; /* should never get here */ 3306524e00b3SLiam R. Howlett 33079760ebffSLiam R. Howlett new_vma = vma_merge(&vmi, mm, prev, addr, addr + len, vma->vm_flags, 330819a809afSAndrea Arcangeli vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma), 33095c26f6acSSuren Baghdasaryan vma->vm_userfaultfd_ctx, anon_vma_name(vma)); 33101da177e4SLinus Torvalds if (new_vma) { 33111da177e4SLinus Torvalds /* 33121da177e4SLinus Torvalds * Source vma may have been merged into new_vma 33131da177e4SLinus Torvalds */ 3314948f017bSAndrea Arcangeli if (unlikely(vma_start >= new_vma->vm_start && 3315948f017bSAndrea Arcangeli vma_start < new_vma->vm_end)) { 3316948f017bSAndrea Arcangeli /* 3317948f017bSAndrea Arcangeli * The only way we can get a vma_merge with 3318948f017bSAndrea Arcangeli * self during an mremap is if the vma hasn't 3319948f017bSAndrea Arcangeli * been faulted in yet and we were allowed to 3320948f017bSAndrea Arcangeli * reset the dst vma->vm_pgoff to the 3321948f017bSAndrea Arcangeli * destination address of the mremap to allow 3322948f017bSAndrea Arcangeli * the merge to happen. mremap must change the 3323948f017bSAndrea Arcangeli * vm_pgoff linearity between src and dst vmas 3324948f017bSAndrea Arcangeli * (in turn preventing a vma_merge) to be 3325948f017bSAndrea Arcangeli * safe. It is only safe to keep the vm_pgoff 3326948f017bSAndrea Arcangeli * linear if there are no pages mapped yet. 3327948f017bSAndrea Arcangeli */ 332881d1b09cSSasha Levin VM_BUG_ON_VMA(faulted_in_anon_vma, new_vma); 332938a76013SMichel Lespinasse *vmap = vma = new_vma; 3330108d6642SMichel Lespinasse } 333138a76013SMichel Lespinasse *need_rmap_locks = (new_vma->vm_pgoff <= vma->vm_pgoff); 33321da177e4SLinus Torvalds } else { 33333928d4f5SLinus Torvalds new_vma = vm_area_dup(vma); 3334e3975891SChen Gang if (!new_vma) 3335e3975891SChen Gang goto out; 33361da177e4SLinus Torvalds new_vma->vm_start = addr; 33371da177e4SLinus Torvalds new_vma->vm_end = addr + len; 33381da177e4SLinus Torvalds new_vma->vm_pgoff = pgoff; 3339ef0855d3SOleg Nesterov if (vma_dup_policy(vma, new_vma)) 3340523d4e20SMichel Lespinasse goto out_free_vma; 3341523d4e20SMichel Lespinasse if (anon_vma_clone(new_vma, vma)) 3342523d4e20SMichel Lespinasse goto out_free_mempol; 3343e9714acfSKonstantin Khlebnikov if (new_vma->vm_file) 33441da177e4SLinus Torvalds get_file(new_vma->vm_file); 33451da177e4SLinus Torvalds if (new_vma->vm_ops && new_vma->vm_ops->open) 33461da177e4SLinus Torvalds new_vma->vm_ops->open(new_vma); 3347d6ac235dSSuren Baghdasaryan vma_start_write(new_vma); 3348763ecb03SLiam R. Howlett if (vma_link(mm, new_vma)) 3349524e00b3SLiam R. Howlett goto out_vma_link; 335038a76013SMichel Lespinasse *need_rmap_locks = false; 33511da177e4SLinus Torvalds } 3352b50e195fSLiam R. Howlett validate_mm(mm); 33531da177e4SLinus Torvalds return new_vma; 33545beb4930SRik van Riel 3355524e00b3SLiam R. Howlett out_vma_link: 3356524e00b3SLiam R. Howlett if (new_vma->vm_ops && new_vma->vm_ops->close) 3357524e00b3SLiam R. Howlett new_vma->vm_ops->close(new_vma); 335892b73996SLiam Howlett 335992b73996SLiam Howlett if (new_vma->vm_file) 336092b73996SLiam Howlett fput(new_vma->vm_file); 336192b73996SLiam Howlett 336292b73996SLiam Howlett unlink_anon_vmas(new_vma); 33635beb4930SRik van Riel out_free_mempol: 3364ef0855d3SOleg Nesterov mpol_put(vma_policy(new_vma)); 33655beb4930SRik van Riel out_free_vma: 33663928d4f5SLinus Torvalds vm_area_free(new_vma); 3367e3975891SChen Gang out: 3368b50e195fSLiam R. Howlett validate_mm(mm); 33695beb4930SRik van Riel return NULL; 33701da177e4SLinus Torvalds } 3371119f657cSakpm@osdl.org 3372119f657cSakpm@osdl.org /* 3373119f657cSakpm@osdl.org * Return true if the calling process may expand its vm space by the passed 3374119f657cSakpm@osdl.org * number of pages 3375119f657cSakpm@osdl.org */ 337684638335SKonstantin Khlebnikov bool may_expand_vm(struct mm_struct *mm, vm_flags_t flags, unsigned long npages) 3377119f657cSakpm@osdl.org { 337884638335SKonstantin Khlebnikov if (mm->total_vm + npages > rlimit(RLIMIT_AS) >> PAGE_SHIFT) 337984638335SKonstantin Khlebnikov return false; 3380119f657cSakpm@osdl.org 3381d977d56cSKonstantin Khlebnikov if (is_data_mapping(flags) && 3382d977d56cSKonstantin Khlebnikov mm->data_vm + npages > rlimit(RLIMIT_DATA) >> PAGE_SHIFT) { 3383f4fcd558SKonstantin Khlebnikov /* Workaround for Valgrind */ 3384f4fcd558SKonstantin Khlebnikov if (rlimit(RLIMIT_DATA) == 0 && 3385f4fcd558SKonstantin Khlebnikov mm->data_vm + npages <= rlimit_max(RLIMIT_DATA) >> PAGE_SHIFT) 3386f4fcd558SKonstantin Khlebnikov return true; 338757a7702bSDavid Woodhouse 338857a7702bSDavid Woodhouse pr_warn_once("%s (%d): VmData %lu exceed data ulimit %lu. Update limits%s.\n", 3389d977d56cSKonstantin Khlebnikov current->comm, current->pid, 3390d977d56cSKonstantin Khlebnikov (mm->data_vm + npages) << PAGE_SHIFT, 339157a7702bSDavid Woodhouse rlimit(RLIMIT_DATA), 339257a7702bSDavid Woodhouse ignore_rlimit_data ? "" : " or use boot option ignore_rlimit_data"); 339357a7702bSDavid Woodhouse 339457a7702bSDavid Woodhouse if (!ignore_rlimit_data) 3395d977d56cSKonstantin Khlebnikov return false; 3396d977d56cSKonstantin Khlebnikov } 3397119f657cSakpm@osdl.org 339884638335SKonstantin Khlebnikov return true; 339984638335SKonstantin Khlebnikov } 340084638335SKonstantin Khlebnikov 340184638335SKonstantin Khlebnikov void vm_stat_account(struct mm_struct *mm, vm_flags_t flags, long npages) 340284638335SKonstantin Khlebnikov { 34037866076bSPeng Liu WRITE_ONCE(mm->total_vm, READ_ONCE(mm->total_vm)+npages); 340484638335SKonstantin Khlebnikov 3405d977d56cSKonstantin Khlebnikov if (is_exec_mapping(flags)) 340684638335SKonstantin Khlebnikov mm->exec_vm += npages; 3407d977d56cSKonstantin Khlebnikov else if (is_stack_mapping(flags)) 340884638335SKonstantin Khlebnikov mm->stack_vm += npages; 3409d977d56cSKonstantin Khlebnikov else if (is_data_mapping(flags)) 341084638335SKonstantin Khlebnikov mm->data_vm += npages; 3411119f657cSakpm@osdl.org } 3412fa5dc22fSRoland McGrath 3413b3ec9f33SSouptick Joarder static vm_fault_t special_mapping_fault(struct vm_fault *vmf); 3414a62c34bdSAndy Lutomirski 3415a62c34bdSAndy Lutomirski /* 3416a62c34bdSAndy Lutomirski * Having a close hook prevents vma merging regardless of flags. 3417a62c34bdSAndy Lutomirski */ 3418a62c34bdSAndy Lutomirski static void special_mapping_close(struct vm_area_struct *vma) 3419a62c34bdSAndy Lutomirski { 3420a62c34bdSAndy Lutomirski } 3421a62c34bdSAndy Lutomirski 3422a62c34bdSAndy Lutomirski static const char *special_mapping_name(struct vm_area_struct *vma) 3423a62c34bdSAndy Lutomirski { 3424a62c34bdSAndy Lutomirski return ((struct vm_special_mapping *)vma->vm_private_data)->name; 3425a62c34bdSAndy Lutomirski } 3426a62c34bdSAndy Lutomirski 342714d07113SBrian Geffon static int special_mapping_mremap(struct vm_area_struct *new_vma) 3428b059a453SDmitry Safonov { 3429b059a453SDmitry Safonov struct vm_special_mapping *sm = new_vma->vm_private_data; 3430b059a453SDmitry Safonov 3431280e87e9SDmitry Safonov if (WARN_ON_ONCE(current->mm != new_vma->vm_mm)) 3432280e87e9SDmitry Safonov return -EFAULT; 3433280e87e9SDmitry Safonov 3434b059a453SDmitry Safonov if (sm->mremap) 3435b059a453SDmitry Safonov return sm->mremap(sm, new_vma); 3436280e87e9SDmitry Safonov 3437b059a453SDmitry Safonov return 0; 3438b059a453SDmitry Safonov } 3439b059a453SDmitry Safonov 3440871402e0SDmitry Safonov static int special_mapping_split(struct vm_area_struct *vma, unsigned long addr) 3441871402e0SDmitry Safonov { 3442871402e0SDmitry Safonov /* 3443871402e0SDmitry Safonov * Forbid splitting special mappings - kernel has expectations over 3444871402e0SDmitry Safonov * the number of pages in mapping. Together with VM_DONTEXPAND 3445871402e0SDmitry Safonov * the size of vma should stay the same over the special mapping's 3446871402e0SDmitry Safonov * lifetime. 3447871402e0SDmitry Safonov */ 3448871402e0SDmitry Safonov return -EINVAL; 3449871402e0SDmitry Safonov } 3450871402e0SDmitry Safonov 3451a62c34bdSAndy Lutomirski static const struct vm_operations_struct special_mapping_vmops = { 3452a62c34bdSAndy Lutomirski .close = special_mapping_close, 3453a62c34bdSAndy Lutomirski .fault = special_mapping_fault, 3454b059a453SDmitry Safonov .mremap = special_mapping_mremap, 3455a62c34bdSAndy Lutomirski .name = special_mapping_name, 3456af34ebebSDmitry Safonov /* vDSO code relies that VVAR can't be accessed remotely */ 3457af34ebebSDmitry Safonov .access = NULL, 3458871402e0SDmitry Safonov .may_split = special_mapping_split, 3459a62c34bdSAndy Lutomirski }; 3460a62c34bdSAndy Lutomirski 3461a62c34bdSAndy Lutomirski static const struct vm_operations_struct legacy_special_mapping_vmops = { 3462a62c34bdSAndy Lutomirski .close = special_mapping_close, 3463a62c34bdSAndy Lutomirski .fault = special_mapping_fault, 3464a62c34bdSAndy Lutomirski }; 3465fa5dc22fSRoland McGrath 3466b3ec9f33SSouptick Joarder static vm_fault_t special_mapping_fault(struct vm_fault *vmf) 3467fa5dc22fSRoland McGrath { 346811bac800SDave Jiang struct vm_area_struct *vma = vmf->vma; 3469b1d0e4f5SNick Piggin pgoff_t pgoff; 3470fa5dc22fSRoland McGrath struct page **pages; 3471fa5dc22fSRoland McGrath 3472f872f540SAndy Lutomirski if (vma->vm_ops == &legacy_special_mapping_vmops) { 3473a62c34bdSAndy Lutomirski pages = vma->vm_private_data; 3474f872f540SAndy Lutomirski } else { 3475f872f540SAndy Lutomirski struct vm_special_mapping *sm = vma->vm_private_data; 3476f872f540SAndy Lutomirski 3477f872f540SAndy Lutomirski if (sm->fault) 347811bac800SDave Jiang return sm->fault(sm, vmf->vma, vmf); 3479f872f540SAndy Lutomirski 3480f872f540SAndy Lutomirski pages = sm->pages; 3481f872f540SAndy Lutomirski } 3482a62c34bdSAndy Lutomirski 34838a9cc3b5SOleg Nesterov for (pgoff = vmf->pgoff; pgoff && *pages; ++pages) 3484b1d0e4f5SNick Piggin pgoff--; 3485fa5dc22fSRoland McGrath 3486fa5dc22fSRoland McGrath if (*pages) { 3487fa5dc22fSRoland McGrath struct page *page = *pages; 3488fa5dc22fSRoland McGrath get_page(page); 3489b1d0e4f5SNick Piggin vmf->page = page; 3490b1d0e4f5SNick Piggin return 0; 3491fa5dc22fSRoland McGrath } 3492fa5dc22fSRoland McGrath 3493b1d0e4f5SNick Piggin return VM_FAULT_SIGBUS; 3494fa5dc22fSRoland McGrath } 3495fa5dc22fSRoland McGrath 3496a62c34bdSAndy Lutomirski static struct vm_area_struct *__install_special_mapping( 3497a62c34bdSAndy Lutomirski struct mm_struct *mm, 3498fa5dc22fSRoland McGrath unsigned long addr, unsigned long len, 349927f28b97SChen Gang unsigned long vm_flags, void *priv, 350027f28b97SChen Gang const struct vm_operations_struct *ops) 3501fa5dc22fSRoland McGrath { 3502462e635eSTavis Ormandy int ret; 3503fa5dc22fSRoland McGrath struct vm_area_struct *vma; 3504fa5dc22fSRoland McGrath 3505b50e195fSLiam R. Howlett validate_mm(mm); 3506490fc053SLinus Torvalds vma = vm_area_alloc(mm); 3507fa5dc22fSRoland McGrath if (unlikely(vma == NULL)) 35083935ed6aSStefani Seibold return ERR_PTR(-ENOMEM); 3509fa5dc22fSRoland McGrath 3510fa5dc22fSRoland McGrath vma->vm_start = addr; 3511fa5dc22fSRoland McGrath vma->vm_end = addr + len; 3512fa5dc22fSRoland McGrath 3513e430a95aSSuren Baghdasaryan vm_flags_init(vma, (vm_flags | mm->def_flags | 3514e430a95aSSuren Baghdasaryan VM_DONTEXPAND | VM_SOFTDIRTY) & ~VM_LOCKED_MASK); 35153ed75eb8SColy Li vma->vm_page_prot = vm_get_page_prot(vma->vm_flags); 3516fa5dc22fSRoland McGrath 3517a62c34bdSAndy Lutomirski vma->vm_ops = ops; 3518a62c34bdSAndy Lutomirski vma->vm_private_data = priv; 3519fa5dc22fSRoland McGrath 3520462e635eSTavis Ormandy ret = insert_vm_struct(mm, vma); 3521462e635eSTavis Ormandy if (ret) 3522462e635eSTavis Ormandy goto out; 3523fa5dc22fSRoland McGrath 352484638335SKonstantin Khlebnikov vm_stat_account(mm, vma->vm_flags, len >> PAGE_SHIFT); 3525fa5dc22fSRoland McGrath 3526cdd6c482SIngo Molnar perf_event_mmap(vma); 3527089dd79dSPeter Zijlstra 3528b50e195fSLiam R. Howlett validate_mm(mm); 35293935ed6aSStefani Seibold return vma; 3530462e635eSTavis Ormandy 3531462e635eSTavis Ormandy out: 35323928d4f5SLinus Torvalds vm_area_free(vma); 3533b50e195fSLiam R. Howlett validate_mm(mm); 35343935ed6aSStefani Seibold return ERR_PTR(ret); 35353935ed6aSStefani Seibold } 35363935ed6aSStefani Seibold 35372eefd878SDmitry Safonov bool vma_is_special_mapping(const struct vm_area_struct *vma, 35382eefd878SDmitry Safonov const struct vm_special_mapping *sm) 35392eefd878SDmitry Safonov { 35402eefd878SDmitry Safonov return vma->vm_private_data == sm && 35412eefd878SDmitry Safonov (vma->vm_ops == &special_mapping_vmops || 35422eefd878SDmitry Safonov vma->vm_ops == &legacy_special_mapping_vmops); 35432eefd878SDmitry Safonov } 35442eefd878SDmitry Safonov 3545a62c34bdSAndy Lutomirski /* 3546c1e8d7c6SMichel Lespinasse * Called with mm->mmap_lock held for writing. 3547a62c34bdSAndy Lutomirski * Insert a new vma covering the given region, with the given flags. 3548a62c34bdSAndy Lutomirski * Its pages are supplied by the given array of struct page *. 3549a62c34bdSAndy Lutomirski * The array can be shorter than len >> PAGE_SHIFT if it's null-terminated. 3550a62c34bdSAndy Lutomirski * The region past the last page supplied will always produce SIGBUS. 3551a62c34bdSAndy Lutomirski * The array pointer and the pages it points to are assumed to stay alive 3552a62c34bdSAndy Lutomirski * for as long as this mapping might exist. 3553a62c34bdSAndy Lutomirski */ 3554a62c34bdSAndy Lutomirski struct vm_area_struct *_install_special_mapping( 3555a62c34bdSAndy Lutomirski struct mm_struct *mm, 3556a62c34bdSAndy Lutomirski unsigned long addr, unsigned long len, 3557a62c34bdSAndy Lutomirski unsigned long vm_flags, const struct vm_special_mapping *spec) 3558a62c34bdSAndy Lutomirski { 355927f28b97SChen Gang return __install_special_mapping(mm, addr, len, vm_flags, (void *)spec, 356027f28b97SChen Gang &special_mapping_vmops); 3561a62c34bdSAndy Lutomirski } 3562a62c34bdSAndy Lutomirski 35633935ed6aSStefani Seibold int install_special_mapping(struct mm_struct *mm, 35643935ed6aSStefani Seibold unsigned long addr, unsigned long len, 35653935ed6aSStefani Seibold unsigned long vm_flags, struct page **pages) 35663935ed6aSStefani Seibold { 3567a62c34bdSAndy Lutomirski struct vm_area_struct *vma = __install_special_mapping( 356827f28b97SChen Gang mm, addr, len, vm_flags, (void *)pages, 356927f28b97SChen Gang &legacy_special_mapping_vmops); 35703935ed6aSStefani Seibold 357114bd5b45SDuan Jiong return PTR_ERR_OR_ZERO(vma); 3572fa5dc22fSRoland McGrath } 35737906d00cSAndrea Arcangeli 35747906d00cSAndrea Arcangeli static DEFINE_MUTEX(mm_all_locks_mutex); 35757906d00cSAndrea Arcangeli 3576454ed842SPeter Zijlstra static void vm_lock_anon_vma(struct mm_struct *mm, struct anon_vma *anon_vma) 35777906d00cSAndrea Arcangeli { 3578f808c13fSDavidlohr Bueso if (!test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_root.rb_node)) { 35797906d00cSAndrea Arcangeli /* 35807906d00cSAndrea Arcangeli * The LSB of head.next can't change from under us 35817906d00cSAndrea Arcangeli * because we hold the mm_all_locks_mutex. 35827906d00cSAndrea Arcangeli */ 3583da1c55f1SMichel Lespinasse down_write_nest_lock(&anon_vma->root->rwsem, &mm->mmap_lock); 35847906d00cSAndrea Arcangeli /* 35857906d00cSAndrea Arcangeli * We can safely modify head.next after taking the 35865a505085SIngo Molnar * anon_vma->root->rwsem. If some other vma in this mm shares 35877906d00cSAndrea Arcangeli * the same anon_vma we won't take it again. 35887906d00cSAndrea Arcangeli * 35897906d00cSAndrea Arcangeli * No need of atomic instructions here, head.next 35907906d00cSAndrea Arcangeli * can't change from under us thanks to the 35915a505085SIngo Molnar * anon_vma->root->rwsem. 35927906d00cSAndrea Arcangeli */ 35937906d00cSAndrea Arcangeli if (__test_and_set_bit(0, (unsigned long *) 3594f808c13fSDavidlohr Bueso &anon_vma->root->rb_root.rb_root.rb_node)) 35957906d00cSAndrea Arcangeli BUG(); 35967906d00cSAndrea Arcangeli } 35977906d00cSAndrea Arcangeli } 35987906d00cSAndrea Arcangeli 3599454ed842SPeter Zijlstra static void vm_lock_mapping(struct mm_struct *mm, struct address_space *mapping) 36007906d00cSAndrea Arcangeli { 36017906d00cSAndrea Arcangeli if (!test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) { 36027906d00cSAndrea Arcangeli /* 36037906d00cSAndrea Arcangeli * AS_MM_ALL_LOCKS can't change from under us because 36047906d00cSAndrea Arcangeli * we hold the mm_all_locks_mutex. 36057906d00cSAndrea Arcangeli * 36067906d00cSAndrea Arcangeli * Operations on ->flags have to be atomic because 36077906d00cSAndrea Arcangeli * even if AS_MM_ALL_LOCKS is stable thanks to the 36087906d00cSAndrea Arcangeli * mm_all_locks_mutex, there may be other cpus 36097906d00cSAndrea Arcangeli * changing other bitflags in parallel to us. 36107906d00cSAndrea Arcangeli */ 36117906d00cSAndrea Arcangeli if (test_and_set_bit(AS_MM_ALL_LOCKS, &mapping->flags)) 36127906d00cSAndrea Arcangeli BUG(); 3613da1c55f1SMichel Lespinasse down_write_nest_lock(&mapping->i_mmap_rwsem, &mm->mmap_lock); 36147906d00cSAndrea Arcangeli } 36157906d00cSAndrea Arcangeli } 36167906d00cSAndrea Arcangeli 36177906d00cSAndrea Arcangeli /* 36187906d00cSAndrea Arcangeli * This operation locks against the VM for all pte/vma/mm related 36197906d00cSAndrea Arcangeli * operations that could ever happen on a certain mm. This includes 36207906d00cSAndrea Arcangeli * vmtruncate, try_to_unmap, and all page faults. 36217906d00cSAndrea Arcangeli * 3622c1e8d7c6SMichel Lespinasse * The caller must take the mmap_lock in write mode before calling 36237906d00cSAndrea Arcangeli * mm_take_all_locks(). The caller isn't allowed to release the 3624c1e8d7c6SMichel Lespinasse * mmap_lock until mm_drop_all_locks() returns. 36257906d00cSAndrea Arcangeli * 3626c1e8d7c6SMichel Lespinasse * mmap_lock in write mode is required in order to block all operations 36277906d00cSAndrea Arcangeli * that could modify pagetables and free pages without need of 362827ba0644SKirill A. Shutemov * altering the vma layout. It's also needed in write mode to avoid new 36297906d00cSAndrea Arcangeli * anon_vmas to be associated with existing vmas. 36307906d00cSAndrea Arcangeli * 36317906d00cSAndrea Arcangeli * A single task can't take more than one mm_take_all_locks() in a row 36327906d00cSAndrea Arcangeli * or it would deadlock. 36337906d00cSAndrea Arcangeli * 3634bf181b9fSMichel Lespinasse * The LSB in anon_vma->rb_root.rb_node and the AS_MM_ALL_LOCKS bitflag in 36357906d00cSAndrea Arcangeli * mapping->flags avoid to take the same lock twice, if more than one 36367906d00cSAndrea Arcangeli * vma in this mm is backed by the same anon_vma or address_space. 36377906d00cSAndrea Arcangeli * 363888f306b6SKirill A. Shutemov * We take locks in following order, accordingly to comment at beginning 363988f306b6SKirill A. Shutemov * of mm/rmap.c: 364088f306b6SKirill A. Shutemov * - all hugetlbfs_i_mmap_rwsem_key locks (aka mapping->i_mmap_rwsem for 364188f306b6SKirill A. Shutemov * hugetlb mapping); 3642eeff9a5dSSuren Baghdasaryan * - all vmas marked locked 364388f306b6SKirill A. Shutemov * - all i_mmap_rwsem locks; 364488f306b6SKirill A. Shutemov * - all anon_vma->rwseml 364588f306b6SKirill A. Shutemov * 364688f306b6SKirill A. Shutemov * We can take all locks within these types randomly because the VM code 364788f306b6SKirill A. Shutemov * doesn't nest them and we protected from parallel mm_take_all_locks() by 364888f306b6SKirill A. Shutemov * mm_all_locks_mutex. 36497906d00cSAndrea Arcangeli * 36507906d00cSAndrea Arcangeli * mm_take_all_locks() and mm_drop_all_locks are expensive operations 36517906d00cSAndrea Arcangeli * that may have to take thousand of locks. 36527906d00cSAndrea Arcangeli * 36537906d00cSAndrea Arcangeli * mm_take_all_locks() can fail if it's interrupted by signals. 36547906d00cSAndrea Arcangeli */ 36557906d00cSAndrea Arcangeli int mm_take_all_locks(struct mm_struct *mm) 36567906d00cSAndrea Arcangeli { 36577906d00cSAndrea Arcangeli struct vm_area_struct *vma; 36585beb4930SRik van Riel struct anon_vma_chain *avc; 3659763ecb03SLiam R. Howlett MA_STATE(mas, &mm->mm_mt, 0, 0); 36607906d00cSAndrea Arcangeli 3661325bca1fSRolf Eike Beer mmap_assert_write_locked(mm); 36627906d00cSAndrea Arcangeli 36637906d00cSAndrea Arcangeli mutex_lock(&mm_all_locks_mutex); 36647906d00cSAndrea Arcangeli 3665763ecb03SLiam R. Howlett mas_for_each(&mas, vma, ULONG_MAX) { 36667906d00cSAndrea Arcangeli if (signal_pending(current)) 36677906d00cSAndrea Arcangeli goto out_unlock; 3668eeff9a5dSSuren Baghdasaryan vma_start_write(vma); 3669eeff9a5dSSuren Baghdasaryan } 3670eeff9a5dSSuren Baghdasaryan 3671eeff9a5dSSuren Baghdasaryan mas_set(&mas, 0); 3672eeff9a5dSSuren Baghdasaryan mas_for_each(&mas, vma, ULONG_MAX) { 3673eeff9a5dSSuren Baghdasaryan if (signal_pending(current)) 3674eeff9a5dSSuren Baghdasaryan goto out_unlock; 367588f306b6SKirill A. Shutemov if (vma->vm_file && vma->vm_file->f_mapping && 367688f306b6SKirill A. Shutemov is_vm_hugetlb_page(vma)) 367788f306b6SKirill A. Shutemov vm_lock_mapping(mm, vma->vm_file->f_mapping); 367888f306b6SKirill A. Shutemov } 367988f306b6SKirill A. Shutemov 3680763ecb03SLiam R. Howlett mas_set(&mas, 0); 3681763ecb03SLiam R. Howlett mas_for_each(&mas, vma, ULONG_MAX) { 368288f306b6SKirill A. Shutemov if (signal_pending(current)) 368388f306b6SKirill A. Shutemov goto out_unlock; 368488f306b6SKirill A. Shutemov if (vma->vm_file && vma->vm_file->f_mapping && 368588f306b6SKirill A. Shutemov !is_vm_hugetlb_page(vma)) 3686454ed842SPeter Zijlstra vm_lock_mapping(mm, vma->vm_file->f_mapping); 36877906d00cSAndrea Arcangeli } 36887cd5a02fSPeter Zijlstra 3689763ecb03SLiam R. Howlett mas_set(&mas, 0); 3690763ecb03SLiam R. Howlett mas_for_each(&mas, vma, ULONG_MAX) { 36917cd5a02fSPeter Zijlstra if (signal_pending(current)) 36927cd5a02fSPeter Zijlstra goto out_unlock; 36937cd5a02fSPeter Zijlstra if (vma->anon_vma) 36945beb4930SRik van Riel list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) 36955beb4930SRik van Riel vm_lock_anon_vma(mm, avc->anon_vma); 36967cd5a02fSPeter Zijlstra } 36977cd5a02fSPeter Zijlstra 3698584cff54SKautuk Consul return 0; 36997906d00cSAndrea Arcangeli 37007906d00cSAndrea Arcangeli out_unlock: 37017906d00cSAndrea Arcangeli mm_drop_all_locks(mm); 3702584cff54SKautuk Consul return -EINTR; 37037906d00cSAndrea Arcangeli } 37047906d00cSAndrea Arcangeli 37057906d00cSAndrea Arcangeli static void vm_unlock_anon_vma(struct anon_vma *anon_vma) 37067906d00cSAndrea Arcangeli { 3707f808c13fSDavidlohr Bueso if (test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_root.rb_node)) { 37087906d00cSAndrea Arcangeli /* 37097906d00cSAndrea Arcangeli * The LSB of head.next can't change to 0 from under 37107906d00cSAndrea Arcangeli * us because we hold the mm_all_locks_mutex. 37117906d00cSAndrea Arcangeli * 37127906d00cSAndrea Arcangeli * We must however clear the bitflag before unlocking 3713bf181b9fSMichel Lespinasse * the vma so the users using the anon_vma->rb_root will 37147906d00cSAndrea Arcangeli * never see our bitflag. 37157906d00cSAndrea Arcangeli * 37167906d00cSAndrea Arcangeli * No need of atomic instructions here, head.next 37177906d00cSAndrea Arcangeli * can't change from under us until we release the 37185a505085SIngo Molnar * anon_vma->root->rwsem. 37197906d00cSAndrea Arcangeli */ 37207906d00cSAndrea Arcangeli if (!__test_and_clear_bit(0, (unsigned long *) 3721f808c13fSDavidlohr Bueso &anon_vma->root->rb_root.rb_root.rb_node)) 37227906d00cSAndrea Arcangeli BUG(); 372308b52706SKonstantin Khlebnikov anon_vma_unlock_write(anon_vma); 37247906d00cSAndrea Arcangeli } 37257906d00cSAndrea Arcangeli } 37267906d00cSAndrea Arcangeli 37277906d00cSAndrea Arcangeli static void vm_unlock_mapping(struct address_space *mapping) 37287906d00cSAndrea Arcangeli { 37297906d00cSAndrea Arcangeli if (test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) { 37307906d00cSAndrea Arcangeli /* 37317906d00cSAndrea Arcangeli * AS_MM_ALL_LOCKS can't change to 0 from under us 37327906d00cSAndrea Arcangeli * because we hold the mm_all_locks_mutex. 37337906d00cSAndrea Arcangeli */ 373483cde9e8SDavidlohr Bueso i_mmap_unlock_write(mapping); 37357906d00cSAndrea Arcangeli if (!test_and_clear_bit(AS_MM_ALL_LOCKS, 37367906d00cSAndrea Arcangeli &mapping->flags)) 37377906d00cSAndrea Arcangeli BUG(); 37387906d00cSAndrea Arcangeli } 37397906d00cSAndrea Arcangeli } 37407906d00cSAndrea Arcangeli 37417906d00cSAndrea Arcangeli /* 3742c1e8d7c6SMichel Lespinasse * The mmap_lock cannot be released by the caller until 37437906d00cSAndrea Arcangeli * mm_drop_all_locks() returns. 37447906d00cSAndrea Arcangeli */ 37457906d00cSAndrea Arcangeli void mm_drop_all_locks(struct mm_struct *mm) 37467906d00cSAndrea Arcangeli { 37477906d00cSAndrea Arcangeli struct vm_area_struct *vma; 37485beb4930SRik van Riel struct anon_vma_chain *avc; 3749763ecb03SLiam R. Howlett MA_STATE(mas, &mm->mm_mt, 0, 0); 37507906d00cSAndrea Arcangeli 3751325bca1fSRolf Eike Beer mmap_assert_write_locked(mm); 37527906d00cSAndrea Arcangeli BUG_ON(!mutex_is_locked(&mm_all_locks_mutex)); 37537906d00cSAndrea Arcangeli 3754763ecb03SLiam R. Howlett mas_for_each(&mas, vma, ULONG_MAX) { 37557906d00cSAndrea Arcangeli if (vma->anon_vma) 37565beb4930SRik van Riel list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) 37575beb4930SRik van Riel vm_unlock_anon_vma(avc->anon_vma); 37587906d00cSAndrea Arcangeli if (vma->vm_file && vma->vm_file->f_mapping) 37597906d00cSAndrea Arcangeli vm_unlock_mapping(vma->vm_file->f_mapping); 37607906d00cSAndrea Arcangeli } 3761eeff9a5dSSuren Baghdasaryan vma_end_write_all(mm); 37627906d00cSAndrea Arcangeli 37637906d00cSAndrea Arcangeli mutex_unlock(&mm_all_locks_mutex); 37647906d00cSAndrea Arcangeli } 37658feae131SDavid Howells 37668feae131SDavid Howells /* 37673edf41d8Sseokhoon.yoon * initialise the percpu counter for VM 37688feae131SDavid Howells */ 37698feae131SDavid Howells void __init mmap_init(void) 37708feae131SDavid Howells { 377100a62ce9SKOSAKI Motohiro int ret; 377200a62ce9SKOSAKI Motohiro 3773908c7f19STejun Heo ret = percpu_counter_init(&vm_committed_as, 0, GFP_KERNEL); 377400a62ce9SKOSAKI Motohiro VM_BUG_ON(ret); 37758feae131SDavid Howells } 3776c9b1d098SAndrew Shewmaker 3777c9b1d098SAndrew Shewmaker /* 3778c9b1d098SAndrew Shewmaker * Initialise sysctl_user_reserve_kbytes. 3779c9b1d098SAndrew Shewmaker * 3780c9b1d098SAndrew Shewmaker * This is intended to prevent a user from starting a single memory hogging 3781c9b1d098SAndrew Shewmaker * process, such that they cannot recover (kill the hog) in OVERCOMMIT_NEVER 3782c9b1d098SAndrew Shewmaker * mode. 3783c9b1d098SAndrew Shewmaker * 3784c9b1d098SAndrew Shewmaker * The default value is min(3% of free memory, 128MB) 3785c9b1d098SAndrew Shewmaker * 128MB is enough to recover with sshd/login, bash, and top/kill. 3786c9b1d098SAndrew Shewmaker */ 37871640879aSAndrew Shewmaker static int init_user_reserve(void) 3788c9b1d098SAndrew Shewmaker { 3789c9b1d098SAndrew Shewmaker unsigned long free_kbytes; 3790c9b1d098SAndrew Shewmaker 3791c41f012aSMichal Hocko free_kbytes = global_zone_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10); 3792c9b1d098SAndrew Shewmaker 3793c9b1d098SAndrew Shewmaker sysctl_user_reserve_kbytes = min(free_kbytes / 32, 1UL << 17); 3794c9b1d098SAndrew Shewmaker return 0; 3795c9b1d098SAndrew Shewmaker } 3796a64fb3cdSPaul Gortmaker subsys_initcall(init_user_reserve); 37974eeab4f5SAndrew Shewmaker 37984eeab4f5SAndrew Shewmaker /* 37994eeab4f5SAndrew Shewmaker * Initialise sysctl_admin_reserve_kbytes. 38004eeab4f5SAndrew Shewmaker * 38014eeab4f5SAndrew Shewmaker * The purpose of sysctl_admin_reserve_kbytes is to allow the sys admin 38024eeab4f5SAndrew Shewmaker * to log in and kill a memory hogging process. 38034eeab4f5SAndrew Shewmaker * 38044eeab4f5SAndrew Shewmaker * Systems with more than 256MB will reserve 8MB, enough to recover 38054eeab4f5SAndrew Shewmaker * with sshd, bash, and top in OVERCOMMIT_GUESS. Smaller systems will 38064eeab4f5SAndrew Shewmaker * only reserve 3% of free pages by default. 38074eeab4f5SAndrew Shewmaker */ 38081640879aSAndrew Shewmaker static int init_admin_reserve(void) 38094eeab4f5SAndrew Shewmaker { 38104eeab4f5SAndrew Shewmaker unsigned long free_kbytes; 38114eeab4f5SAndrew Shewmaker 3812c41f012aSMichal Hocko free_kbytes = global_zone_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10); 38134eeab4f5SAndrew Shewmaker 38144eeab4f5SAndrew Shewmaker sysctl_admin_reserve_kbytes = min(free_kbytes / 32, 1UL << 13); 38154eeab4f5SAndrew Shewmaker return 0; 38164eeab4f5SAndrew Shewmaker } 3817a64fb3cdSPaul Gortmaker subsys_initcall(init_admin_reserve); 38181640879aSAndrew Shewmaker 38191640879aSAndrew Shewmaker /* 38201640879aSAndrew Shewmaker * Reinititalise user and admin reserves if memory is added or removed. 38211640879aSAndrew Shewmaker * 38221640879aSAndrew Shewmaker * The default user reserve max is 128MB, and the default max for the 38231640879aSAndrew Shewmaker * admin reserve is 8MB. These are usually, but not always, enough to 38241640879aSAndrew Shewmaker * enable recovery from a memory hogging process using login/sshd, a shell, 38251640879aSAndrew Shewmaker * and tools like top. It may make sense to increase or even disable the 38261640879aSAndrew Shewmaker * reserve depending on the existence of swap or variations in the recovery 38271640879aSAndrew Shewmaker * tools. So, the admin may have changed them. 38281640879aSAndrew Shewmaker * 38291640879aSAndrew Shewmaker * If memory is added and the reserves have been eliminated or increased above 38301640879aSAndrew Shewmaker * the default max, then we'll trust the admin. 38311640879aSAndrew Shewmaker * 38321640879aSAndrew Shewmaker * If memory is removed and there isn't enough free memory, then we 38331640879aSAndrew Shewmaker * need to reset the reserves. 38341640879aSAndrew Shewmaker * 38351640879aSAndrew Shewmaker * Otherwise keep the reserve set by the admin. 38361640879aSAndrew Shewmaker */ 38371640879aSAndrew Shewmaker static int reserve_mem_notifier(struct notifier_block *nb, 38381640879aSAndrew Shewmaker unsigned long action, void *data) 38391640879aSAndrew Shewmaker { 38401640879aSAndrew Shewmaker unsigned long tmp, free_kbytes; 38411640879aSAndrew Shewmaker 38421640879aSAndrew Shewmaker switch (action) { 38431640879aSAndrew Shewmaker case MEM_ONLINE: 38441640879aSAndrew Shewmaker /* Default max is 128MB. Leave alone if modified by operator. */ 38451640879aSAndrew Shewmaker tmp = sysctl_user_reserve_kbytes; 38461640879aSAndrew Shewmaker if (0 < tmp && tmp < (1UL << 17)) 38471640879aSAndrew Shewmaker init_user_reserve(); 38481640879aSAndrew Shewmaker 38491640879aSAndrew Shewmaker /* Default max is 8MB. Leave alone if modified by operator. */ 38501640879aSAndrew Shewmaker tmp = sysctl_admin_reserve_kbytes; 38511640879aSAndrew Shewmaker if (0 < tmp && tmp < (1UL << 13)) 38521640879aSAndrew Shewmaker init_admin_reserve(); 38531640879aSAndrew Shewmaker 38541640879aSAndrew Shewmaker break; 38551640879aSAndrew Shewmaker case MEM_OFFLINE: 3856c41f012aSMichal Hocko free_kbytes = global_zone_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10); 38571640879aSAndrew Shewmaker 38581640879aSAndrew Shewmaker if (sysctl_user_reserve_kbytes > free_kbytes) { 38591640879aSAndrew Shewmaker init_user_reserve(); 38601640879aSAndrew Shewmaker pr_info("vm.user_reserve_kbytes reset to %lu\n", 38611640879aSAndrew Shewmaker sysctl_user_reserve_kbytes); 38621640879aSAndrew Shewmaker } 38631640879aSAndrew Shewmaker 38641640879aSAndrew Shewmaker if (sysctl_admin_reserve_kbytes > free_kbytes) { 38651640879aSAndrew Shewmaker init_admin_reserve(); 38661640879aSAndrew Shewmaker pr_info("vm.admin_reserve_kbytes reset to %lu\n", 38671640879aSAndrew Shewmaker sysctl_admin_reserve_kbytes); 38681640879aSAndrew Shewmaker } 38691640879aSAndrew Shewmaker break; 38701640879aSAndrew Shewmaker default: 38711640879aSAndrew Shewmaker break; 38721640879aSAndrew Shewmaker } 38731640879aSAndrew Shewmaker return NOTIFY_OK; 38741640879aSAndrew Shewmaker } 38751640879aSAndrew Shewmaker 38761640879aSAndrew Shewmaker static int __meminit init_reserve_notifier(void) 38771640879aSAndrew Shewmaker { 38781eeaa4fdSLiu Shixin if (hotplug_memory_notifier(reserve_mem_notifier, DEFAULT_CALLBACK_PRI)) 3879b1de0d13SMitchel Humpherys pr_err("Failed registering memory add/remove notifier for admin reserve\n"); 38801640879aSAndrew Shewmaker 38811640879aSAndrew Shewmaker return 0; 38821640879aSAndrew Shewmaker } 3883a64fb3cdSPaul Gortmaker subsys_initcall(init_reserve_notifier); 3884