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> 491da177e4SLinus Torvalds 507c0f6ba6SLinus Torvalds #include <linux/uaccess.h> 511da177e4SLinus Torvalds #include <asm/cacheflush.h> 521da177e4SLinus Torvalds #include <asm/tlb.h> 53d6dd61c8SJeremy Fitzhardinge #include <asm/mmu_context.h> 541da177e4SLinus Torvalds 55df529cabSJaewon Kim #define CREATE_TRACE_POINTS 56df529cabSJaewon Kim #include <trace/events/mmap.h> 57df529cabSJaewon Kim 5842b77728SJan Beulich #include "internal.h" 5942b77728SJan Beulich 603a459756SKirill Korotaev #ifndef arch_mmap_check 613a459756SKirill Korotaev #define arch_mmap_check(addr, len, flags) (0) 623a459756SKirill Korotaev #endif 633a459756SKirill Korotaev 64d07e2259SDaniel Cashman #ifdef CONFIG_HAVE_ARCH_MMAP_RND_BITS 65d07e2259SDaniel Cashman const int mmap_rnd_bits_min = CONFIG_ARCH_MMAP_RND_BITS_MIN; 66d07e2259SDaniel Cashman const int mmap_rnd_bits_max = CONFIG_ARCH_MMAP_RND_BITS_MAX; 67d07e2259SDaniel Cashman int mmap_rnd_bits __read_mostly = CONFIG_ARCH_MMAP_RND_BITS; 68d07e2259SDaniel Cashman #endif 69d07e2259SDaniel Cashman #ifdef CONFIG_HAVE_ARCH_MMAP_RND_COMPAT_BITS 70d07e2259SDaniel Cashman const int mmap_rnd_compat_bits_min = CONFIG_ARCH_MMAP_RND_COMPAT_BITS_MIN; 71d07e2259SDaniel Cashman const int mmap_rnd_compat_bits_max = CONFIG_ARCH_MMAP_RND_COMPAT_BITS_MAX; 72d07e2259SDaniel Cashman int mmap_rnd_compat_bits __read_mostly = CONFIG_ARCH_MMAP_RND_COMPAT_BITS; 73d07e2259SDaniel Cashman #endif 74d07e2259SDaniel Cashman 75f4fcd558SKonstantin Khlebnikov static bool ignore_rlimit_data; 76d977d56cSKonstantin Khlebnikov core_param(ignore_rlimit_data, ignore_rlimit_data, bool, 0644); 77d07e2259SDaniel Cashman 78763ecb03SLiam R. Howlett static void unmap_region(struct mm_struct *mm, struct maple_tree *mt, 79e0da382cSHugh Dickins struct vm_area_struct *vma, struct vm_area_struct *prev, 80763ecb03SLiam R. Howlett struct vm_area_struct *next, unsigned long start, 81763ecb03SLiam R. Howlett unsigned long end); 82e0da382cSHugh Dickins 8364e45507SPeter Feiner static pgprot_t vm_pgprot_modify(pgprot_t oldprot, unsigned long vm_flags) 8464e45507SPeter Feiner { 8564e45507SPeter Feiner return pgprot_modify(oldprot, vm_get_page_prot(vm_flags)); 8664e45507SPeter Feiner } 8764e45507SPeter Feiner 8864e45507SPeter Feiner /* Update vma->vm_page_prot to reflect vma->vm_flags. */ 8964e45507SPeter Feiner void vma_set_page_prot(struct vm_area_struct *vma) 9064e45507SPeter Feiner { 9164e45507SPeter Feiner unsigned long vm_flags = vma->vm_flags; 926d2329f8SAndrea Arcangeli pgprot_t vm_page_prot; 9364e45507SPeter Feiner 946d2329f8SAndrea Arcangeli vm_page_prot = vm_pgprot_modify(vma->vm_page_prot, vm_flags); 956d2329f8SAndrea Arcangeli if (vma_wants_writenotify(vma, vm_page_prot)) { 9664e45507SPeter Feiner vm_flags &= ~VM_SHARED; 976d2329f8SAndrea Arcangeli vm_page_prot = vm_pgprot_modify(vm_page_prot, vm_flags); 9864e45507SPeter Feiner } 99c1e8d7c6SMichel Lespinasse /* remove_protection_ptes reads vma->vm_page_prot without mmap_lock */ 1006d2329f8SAndrea Arcangeli WRITE_ONCE(vma->vm_page_prot, vm_page_prot); 10164e45507SPeter Feiner } 10264e45507SPeter Feiner 1031da177e4SLinus Torvalds /* 104c8c06efaSDavidlohr Bueso * Requires inode->i_mapping->i_mmap_rwsem 1051da177e4SLinus Torvalds */ 1061da177e4SLinus Torvalds static void __remove_shared_vm_struct(struct vm_area_struct *vma, 1071da177e4SLinus Torvalds struct file *file, struct address_space *mapping) 1081da177e4SLinus Torvalds { 1091da177e4SLinus Torvalds if (vma->vm_flags & VM_SHARED) 1104bb5f5d9SDavid Herrmann mapping_unmap_writable(mapping); 1111da177e4SLinus Torvalds 1121da177e4SLinus Torvalds flush_dcache_mmap_lock(mapping); 1136b2dbba8SMichel Lespinasse vma_interval_tree_remove(vma, &mapping->i_mmap); 1141da177e4SLinus Torvalds flush_dcache_mmap_unlock(mapping); 1151da177e4SLinus Torvalds } 1161da177e4SLinus Torvalds 1171da177e4SLinus Torvalds /* 1186b2dbba8SMichel Lespinasse * Unlink a file-based vm structure from its interval tree, to hide 119a8fb5618SHugh Dickins * vma from rmap and vmtruncate before freeing its page tables. 1201da177e4SLinus Torvalds */ 121a8fb5618SHugh Dickins void unlink_file_vma(struct vm_area_struct *vma) 1221da177e4SLinus Torvalds { 1231da177e4SLinus Torvalds struct file *file = vma->vm_file; 1241da177e4SLinus Torvalds 1251da177e4SLinus Torvalds if (file) { 1261da177e4SLinus Torvalds struct address_space *mapping = file->f_mapping; 12783cde9e8SDavidlohr Bueso i_mmap_lock_write(mapping); 1281da177e4SLinus Torvalds __remove_shared_vm_struct(vma, file, mapping); 12983cde9e8SDavidlohr Bueso i_mmap_unlock_write(mapping); 1301da177e4SLinus Torvalds } 131a8fb5618SHugh Dickins } 132a8fb5618SHugh Dickins 133a8fb5618SHugh Dickins /* 134763ecb03SLiam R. Howlett * Close a vm structure and free it. 135a8fb5618SHugh Dickins */ 136763ecb03SLiam R. Howlett static void remove_vma(struct vm_area_struct *vma) 137a8fb5618SHugh Dickins { 138a8fb5618SHugh Dickins might_sleep(); 1391da177e4SLinus Torvalds if (vma->vm_ops && vma->vm_ops->close) 1401da177e4SLinus Torvalds vma->vm_ops->close(vma); 141e9714acfSKonstantin Khlebnikov if (vma->vm_file) 142a8fb5618SHugh Dickins fput(vma->vm_file); 143f0be3d32SLee Schermerhorn mpol_put(vma_policy(vma)); 1443928d4f5SLinus Torvalds vm_area_free(vma); 1451da177e4SLinus Torvalds } 1461da177e4SLinus Torvalds 147b62b633eSLiam R. Howlett static inline struct vm_area_struct *vma_prev_limit(struct vma_iterator *vmi, 148b62b633eSLiam R. Howlett unsigned long min) 149b62b633eSLiam R. Howlett { 150b62b633eSLiam R. Howlett return mas_prev(&vmi->mas, min); 151b62b633eSLiam R. Howlett } 152b62b633eSLiam R. Howlett 153b62b633eSLiam R. Howlett static inline int vma_iter_clear_gfp(struct vma_iterator *vmi, 154b62b633eSLiam R. Howlett unsigned long start, unsigned long end, gfp_t gfp) 155b62b633eSLiam R. Howlett { 156b62b633eSLiam R. Howlett vmi->mas.index = start; 157b62b633eSLiam R. Howlett vmi->mas.last = end - 1; 158b62b633eSLiam R. Howlett mas_store_gfp(&vmi->mas, NULL, gfp); 159b62b633eSLiam R. Howlett if (unlikely(mas_is_err(&vmi->mas))) 160b62b633eSLiam R. Howlett return -ENOMEM; 161b62b633eSLiam R. Howlett 162b62b633eSLiam R. Howlett return 0; 163b62b633eSLiam R. Howlett } 164b62b633eSLiam R. Howlett 1652e7ce7d3SLiam R. Howlett /* 1662e7ce7d3SLiam R. Howlett * check_brk_limits() - Use platform specific check of range & verify mlock 1672e7ce7d3SLiam R. Howlett * limits. 1682e7ce7d3SLiam R. Howlett * @addr: The address to check 1692e7ce7d3SLiam R. Howlett * @len: The size of increase. 1702e7ce7d3SLiam R. Howlett * 1712e7ce7d3SLiam R. Howlett * Return: 0 on success. 1722e7ce7d3SLiam R. Howlett */ 1732e7ce7d3SLiam R. Howlett static int check_brk_limits(unsigned long addr, unsigned long len) 1742e7ce7d3SLiam R. Howlett { 1752e7ce7d3SLiam R. Howlett unsigned long mapped_addr; 1762e7ce7d3SLiam R. Howlett 1772e7ce7d3SLiam R. Howlett mapped_addr = get_unmapped_area(NULL, addr, len, 0, MAP_FIXED); 1782e7ce7d3SLiam R. Howlett if (IS_ERR_VALUE(mapped_addr)) 1792e7ce7d3SLiam R. Howlett return mapped_addr; 1802e7ce7d3SLiam R. Howlett 1812e7ce7d3SLiam R. Howlett return mlock_future_check(current->mm, current->mm->def_flags, len); 1822e7ce7d3SLiam R. Howlett } 18392fed820SLiam R. Howlett static int do_brk_flags(struct vma_iterator *vmi, struct vm_area_struct *brkvma, 184763ecb03SLiam R. Howlett unsigned long addr, unsigned long request, unsigned long flags); 1856a6160a7SHeiko Carstens SYSCALL_DEFINE1(brk, unsigned long, brk) 1861da177e4SLinus Torvalds { 1879bc8039eSYang Shi unsigned long newbrk, oldbrk, origbrk; 1881da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 1892e7ce7d3SLiam R. Howlett struct vm_area_struct *brkvma, *next = NULL; 190a5b4592cSJiri Kosina unsigned long min_brk; 191128557ffSMichel Lespinasse bool populate; 1929bc8039eSYang Shi bool downgraded = false; 193897ab3e0SMike Rapoport LIST_HEAD(uf); 19492fed820SLiam R. Howlett struct vma_iterator vmi; 1951da177e4SLinus Torvalds 196d8ed45c5SMichel Lespinasse if (mmap_write_lock_killable(mm)) 197dc0ef0dfSMichal Hocko return -EINTR; 1981da177e4SLinus Torvalds 1999bc8039eSYang Shi origbrk = mm->brk; 2009bc8039eSYang Shi 201a5b4592cSJiri Kosina #ifdef CONFIG_COMPAT_BRK 2025520e894SJiri Kosina /* 2035520e894SJiri Kosina * CONFIG_COMPAT_BRK can still be overridden by setting 2045520e894SJiri Kosina * randomize_va_space to 2, which will still cause mm->start_brk 2055520e894SJiri Kosina * to be arbitrarily shifted 2065520e894SJiri Kosina */ 2074471a675SJiri Kosina if (current->brk_randomized) 2085520e894SJiri Kosina min_brk = mm->start_brk; 2095520e894SJiri Kosina else 2105520e894SJiri Kosina min_brk = mm->end_data; 211a5b4592cSJiri Kosina #else 212a5b4592cSJiri Kosina min_brk = mm->start_brk; 213a5b4592cSJiri Kosina #endif 214a5b4592cSJiri Kosina if (brk < min_brk) 2151da177e4SLinus Torvalds goto out; 2161e624196SRam Gupta 2171e624196SRam Gupta /* 2181e624196SRam Gupta * Check against rlimit here. If this check is done later after the test 2191e624196SRam Gupta * of oldbrk with newbrk then it can escape the test and let the data 2201e624196SRam Gupta * segment grow beyond its set limit the in case where the limit is 2211e624196SRam Gupta * not page aligned -Ram Gupta 2221e624196SRam Gupta */ 2238764b338SCyrill Gorcunov if (check_data_rlimit(rlimit(RLIMIT_DATA), brk, mm->start_brk, 2248764b338SCyrill Gorcunov mm->end_data, mm->start_data)) 2251e624196SRam Gupta goto out; 2261e624196SRam Gupta 2271da177e4SLinus Torvalds newbrk = PAGE_ALIGN(brk); 2281da177e4SLinus Torvalds oldbrk = PAGE_ALIGN(mm->brk); 2299bc8039eSYang Shi if (oldbrk == newbrk) { 2309bc8039eSYang Shi mm->brk = brk; 2319bc8039eSYang Shi goto success; 2329bc8039eSYang Shi } 2331da177e4SLinus Torvalds 2349bc8039eSYang Shi /* 2359bc8039eSYang Shi * Always allow shrinking brk. 23627b26701SLiam R. Howlett * do_vma_munmap() may downgrade mmap_lock to read. 2379bc8039eSYang Shi */ 2381da177e4SLinus Torvalds if (brk <= mm->brk) { 2399bc8039eSYang Shi int ret; 2409bc8039eSYang Shi 2412e7ce7d3SLiam R. Howlett /* Search one past newbrk */ 24292fed820SLiam R. Howlett vma_iter_init(&vmi, mm, newbrk); 24392fed820SLiam R. Howlett brkvma = vma_find(&vmi, oldbrk); 244f5ad5083SJason A. Donenfeld if (!brkvma || brkvma->vm_start >= oldbrk) 2452e7ce7d3SLiam R. Howlett goto out; /* mapping intersects with an existing non-brk vma. */ 2469bc8039eSYang Shi /* 2472e7ce7d3SLiam R. Howlett * mm->brk must be protected by write mmap_lock. 24827b26701SLiam R. Howlett * do_vma_munmap() may downgrade the lock, so update it 24927b26701SLiam R. Howlett * before calling do_vma_munmap(). 2509bc8039eSYang Shi */ 2519bc8039eSYang Shi mm->brk = brk; 25227b26701SLiam R. Howlett ret = do_vma_munmap(&vmi, brkvma, newbrk, oldbrk, &uf, true); 2532e7ce7d3SLiam R. Howlett if (ret == 1) { 2542e7ce7d3SLiam R. Howlett downgraded = true; 2552e7ce7d3SLiam R. Howlett goto success; 2562e7ce7d3SLiam R. Howlett } else if (!ret) 2572e7ce7d3SLiam R. Howlett goto success; 2582e7ce7d3SLiam R. Howlett 2599bc8039eSYang Shi mm->brk = origbrk; 2601da177e4SLinus Torvalds goto out; 2611da177e4SLinus Torvalds } 2621da177e4SLinus Torvalds 2632e7ce7d3SLiam R. Howlett if (check_brk_limits(oldbrk, newbrk - oldbrk)) 2642e7ce7d3SLiam R. Howlett goto out; 2652e7ce7d3SLiam R. Howlett 2662e7ce7d3SLiam R. Howlett /* 2672e7ce7d3SLiam R. Howlett * Only check if the next VMA is within the stack_guard_gap of the 2682e7ce7d3SLiam R. Howlett * expansion area 2692e7ce7d3SLiam R. Howlett */ 27092fed820SLiam R. Howlett vma_iter_init(&vmi, mm, oldbrk); 27192fed820SLiam R. Howlett next = vma_find(&vmi, newbrk + PAGE_SIZE + stack_guard_gap); 2721be7107fSHugh Dickins if (next && newbrk + PAGE_SIZE > vm_start_gap(next)) 2731da177e4SLinus Torvalds goto out; 2741da177e4SLinus Torvalds 27592fed820SLiam R. Howlett brkvma = vma_prev_limit(&vmi, mm->start_brk); 2761da177e4SLinus Torvalds /* Ok, looks good - let it rip. */ 27792fed820SLiam R. Howlett if (do_brk_flags(&vmi, brkvma, oldbrk, newbrk - oldbrk, 0) < 0) 2781da177e4SLinus Torvalds goto out; 2792e7ce7d3SLiam R. Howlett 2801da177e4SLinus Torvalds mm->brk = brk; 2819bc8039eSYang Shi 2829bc8039eSYang Shi success: 283128557ffSMichel Lespinasse populate = newbrk > oldbrk && (mm->def_flags & VM_LOCKED) != 0; 2849bc8039eSYang Shi if (downgraded) 285d8ed45c5SMichel Lespinasse mmap_read_unlock(mm); 2869bc8039eSYang Shi else 287d8ed45c5SMichel Lespinasse mmap_write_unlock(mm); 288897ab3e0SMike Rapoport userfaultfd_unmap_complete(mm, &uf); 289128557ffSMichel Lespinasse if (populate) 290128557ffSMichel Lespinasse mm_populate(oldbrk, newbrk - oldbrk); 291128557ffSMichel Lespinasse return brk; 292128557ffSMichel Lespinasse 2931da177e4SLinus Torvalds out: 294d8ed45c5SMichel Lespinasse mmap_write_unlock(mm); 295b7204006SAdrian Huang return origbrk; 2961da177e4SLinus Torvalds } 2971da177e4SLinus Torvalds 298d4af56c5SLiam R. Howlett #if defined(CONFIG_DEBUG_VM_MAPLE_TREE) 299d4af56c5SLiam R. Howlett extern void mt_validate(struct maple_tree *mt); 300d4af56c5SLiam R. Howlett extern void mt_dump(const struct maple_tree *mt); 3011da177e4SLinus Torvalds 302d4af56c5SLiam R. Howlett /* Validate the maple tree */ 303d4af56c5SLiam R. Howlett static void validate_mm_mt(struct mm_struct *mm) 304d4af56c5SLiam R. Howlett { 305d4af56c5SLiam R. Howlett struct maple_tree *mt = &mm->mm_mt; 306763ecb03SLiam R. Howlett struct vm_area_struct *vma_mt; 307d4af56c5SLiam R. Howlett 308d4af56c5SLiam R. Howlett MA_STATE(mas, mt, 0, 0); 309d4af56c5SLiam R. Howlett 310d4af56c5SLiam R. Howlett mt_validate(&mm->mm_mt); 311d4af56c5SLiam R. Howlett mas_for_each(&mas, vma_mt, ULONG_MAX) { 312763ecb03SLiam R. Howlett if ((vma_mt->vm_start != mas.index) || 313763ecb03SLiam R. Howlett (vma_mt->vm_end - 1 != mas.last)) { 314d4af56c5SLiam R. Howlett pr_emerg("issue in %s\n", current->comm); 315d4af56c5SLiam R. Howlett dump_stack(); 316d4af56c5SLiam R. Howlett dump_vma(vma_mt); 317d4af56c5SLiam R. Howlett pr_emerg("mt piv: %p %lu - %lu\n", vma_mt, 318d4af56c5SLiam R. Howlett mas.index, mas.last); 319d4af56c5SLiam R. Howlett pr_emerg("mt vma: %p %lu - %lu\n", vma_mt, 320d4af56c5SLiam R. Howlett vma_mt->vm_start, vma_mt->vm_end); 321d4af56c5SLiam R. Howlett 322d4af56c5SLiam R. Howlett mt_dump(mas.tree); 323d4af56c5SLiam R. Howlett if (vma_mt->vm_end != mas.last + 1) { 324d4af56c5SLiam R. Howlett pr_err("vma: %p vma_mt %lu-%lu\tmt %lu-%lu\n", 325d4af56c5SLiam R. Howlett mm, vma_mt->vm_start, vma_mt->vm_end, 326d4af56c5SLiam R. Howlett mas.index, mas.last); 327d4af56c5SLiam R. Howlett mt_dump(mas.tree); 328d4af56c5SLiam R. Howlett } 329d4af56c5SLiam R. Howlett VM_BUG_ON_MM(vma_mt->vm_end != mas.last + 1, mm); 330d4af56c5SLiam R. Howlett if (vma_mt->vm_start != mas.index) { 331d4af56c5SLiam R. Howlett pr_err("vma: %p vma_mt %p %lu - %lu doesn't match\n", 332d4af56c5SLiam R. Howlett mm, vma_mt, vma_mt->vm_start, vma_mt->vm_end); 333d4af56c5SLiam R. Howlett mt_dump(mas.tree); 334d4af56c5SLiam R. Howlett } 335d4af56c5SLiam R. Howlett VM_BUG_ON_MM(vma_mt->vm_start != mas.index, mm); 336d4af56c5SLiam R. Howlett } 337d4af56c5SLiam R. Howlett } 338d4af56c5SLiam R. Howlett } 3391da177e4SLinus Torvalds 340eafd4dc4SRashika Kheria static void validate_mm(struct mm_struct *mm) 3411da177e4SLinus Torvalds { 3421da177e4SLinus Torvalds int bug = 0; 3431da177e4SLinus Torvalds int i = 0; 344763ecb03SLiam R. Howlett struct vm_area_struct *vma; 345763ecb03SLiam R. Howlett MA_STATE(mas, &mm->mm_mt, 0, 0); 346ff26f70fSAndrew Morton 347524e00b3SLiam R. Howlett validate_mm_mt(mm); 348524e00b3SLiam R. Howlett 349763ecb03SLiam R. Howlett mas_for_each(&mas, vma, ULONG_MAX) { 350524e00b3SLiam R. Howlett #ifdef CONFIG_DEBUG_VM_RB 35112352d3cSKonstantin Khlebnikov struct anon_vma *anon_vma = vma->anon_vma; 352ed8ea815SMichel Lespinasse struct anon_vma_chain *avc; 353ff26f70fSAndrew Morton 35412352d3cSKonstantin Khlebnikov if (anon_vma) { 35512352d3cSKonstantin Khlebnikov anon_vma_lock_read(anon_vma); 356ed8ea815SMichel Lespinasse list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) 357ed8ea815SMichel Lespinasse anon_vma_interval_tree_verify(avc); 35812352d3cSKonstantin Khlebnikov anon_vma_unlock_read(anon_vma); 35912352d3cSKonstantin Khlebnikov } 360524e00b3SLiam R. Howlett #endif 3611da177e4SLinus Torvalds i++; 3621da177e4SLinus Torvalds } 3635a0768f6SMichel Lespinasse if (i != mm->map_count) { 364763ecb03SLiam R. Howlett pr_emerg("map_count %d mas_for_each %d\n", mm->map_count, i); 3655a0768f6SMichel Lespinasse bug = 1; 3665a0768f6SMichel Lespinasse } 36796dad67fSSasha Levin VM_BUG_ON_MM(bug, mm); 3681da177e4SLinus Torvalds } 369524e00b3SLiam R. Howlett 370524e00b3SLiam R. Howlett #else /* !CONFIG_DEBUG_VM_MAPLE_TREE */ 371d4af56c5SLiam R. Howlett #define validate_mm_mt(root) do { } while (0) 3721da177e4SLinus Torvalds #define validate_mm(mm) do { } while (0) 373524e00b3SLiam R. Howlett #endif /* CONFIG_DEBUG_VM_MAPLE_TREE */ 374d3737187SMichel Lespinasse 375bf181b9fSMichel Lespinasse /* 376bf181b9fSMichel Lespinasse * vma has some anon_vma assigned, and is already inserted on that 377bf181b9fSMichel Lespinasse * anon_vma's interval trees. 378bf181b9fSMichel Lespinasse * 379bf181b9fSMichel Lespinasse * Before updating the vma's vm_start / vm_end / vm_pgoff fields, the 380bf181b9fSMichel Lespinasse * vma must be removed from the anon_vma's interval trees using 381bf181b9fSMichel Lespinasse * anon_vma_interval_tree_pre_update_vma(). 382bf181b9fSMichel Lespinasse * 383bf181b9fSMichel Lespinasse * After the update, the vma will be reinserted using 384bf181b9fSMichel Lespinasse * anon_vma_interval_tree_post_update_vma(). 385bf181b9fSMichel Lespinasse * 386c1e8d7c6SMichel Lespinasse * The entire update must be protected by exclusive mmap_lock and by 387bf181b9fSMichel Lespinasse * the root anon_vma's mutex. 388bf181b9fSMichel Lespinasse */ 389bf181b9fSMichel Lespinasse static inline void 390bf181b9fSMichel Lespinasse anon_vma_interval_tree_pre_update_vma(struct vm_area_struct *vma) 391bf181b9fSMichel Lespinasse { 392bf181b9fSMichel Lespinasse struct anon_vma_chain *avc; 393bf181b9fSMichel Lespinasse 394bf181b9fSMichel Lespinasse list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) 395bf181b9fSMichel Lespinasse anon_vma_interval_tree_remove(avc, &avc->anon_vma->rb_root); 396bf181b9fSMichel Lespinasse } 397bf181b9fSMichel Lespinasse 398bf181b9fSMichel Lespinasse static inline void 399bf181b9fSMichel Lespinasse anon_vma_interval_tree_post_update_vma(struct vm_area_struct *vma) 400bf181b9fSMichel Lespinasse { 401bf181b9fSMichel Lespinasse struct anon_vma_chain *avc; 402bf181b9fSMichel Lespinasse 403bf181b9fSMichel Lespinasse list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) 404bf181b9fSMichel Lespinasse anon_vma_interval_tree_insert(avc, &avc->anon_vma->rb_root); 405bf181b9fSMichel Lespinasse } 406bf181b9fSMichel Lespinasse 407e8420a8eSCyril Hrubis static unsigned long count_vma_pages_range(struct mm_struct *mm, 408e8420a8eSCyril Hrubis unsigned long addr, unsigned long end) 409e8420a8eSCyril Hrubis { 4102e3af1dbSMatthew Wilcox (Oracle) VMA_ITERATOR(vmi, mm, addr); 411e8420a8eSCyril Hrubis struct vm_area_struct *vma; 4122e3af1dbSMatthew Wilcox (Oracle) unsigned long nr_pages = 0; 413e8420a8eSCyril Hrubis 4142e3af1dbSMatthew Wilcox (Oracle) for_each_vma_range(vmi, vma, end) { 4152e3af1dbSMatthew Wilcox (Oracle) unsigned long vm_start = max(addr, vma->vm_start); 4162e3af1dbSMatthew Wilcox (Oracle) unsigned long vm_end = min(end, vma->vm_end); 417e8420a8eSCyril Hrubis 4182e3af1dbSMatthew Wilcox (Oracle) nr_pages += PHYS_PFN(vm_end - vm_start); 419e8420a8eSCyril Hrubis } 420e8420a8eSCyril Hrubis 421e8420a8eSCyril Hrubis return nr_pages; 422e8420a8eSCyril Hrubis } 423e8420a8eSCyril Hrubis 424c154124fSLiam R. Howlett static void __vma_link_file(struct vm_area_struct *vma, 425c154124fSLiam R. Howlett struct address_space *mapping) 4261da177e4SLinus Torvalds { 4271da177e4SLinus Torvalds if (vma->vm_flags & VM_SHARED) 428cf508b58SMiaohe Lin mapping_allow_writable(mapping); 4291da177e4SLinus Torvalds 4301da177e4SLinus Torvalds flush_dcache_mmap_lock(mapping); 4316b2dbba8SMichel Lespinasse vma_interval_tree_insert(vma, &mapping->i_mmap); 4321da177e4SLinus Torvalds flush_dcache_mmap_unlock(mapping); 4331da177e4SLinus Torvalds } 4341da177e4SLinus Torvalds 435763ecb03SLiam R. Howlett static int vma_link(struct mm_struct *mm, struct vm_area_struct *vma) 4361da177e4SLinus Torvalds { 43779e4f2caSLiam R. Howlett VMA_ITERATOR(vmi, mm, 0); 4381da177e4SLinus Torvalds struct address_space *mapping = NULL; 4391da177e4SLinus Torvalds 44079e4f2caSLiam R. Howlett if (vma_iter_prealloc(&vmi)) 441d4af56c5SLiam R. Howlett return -ENOMEM; 442d4af56c5SLiam R. Howlett 44364ac4940SHuang Shijie if (vma->vm_file) { 4441da177e4SLinus Torvalds mapping = vma->vm_file->f_mapping; 44583cde9e8SDavidlohr Bueso i_mmap_lock_write(mapping); 44664ac4940SHuang Shijie } 4471da177e4SLinus Torvalds 44879e4f2caSLiam R. Howlett vma_iter_store(&vmi, vma); 4491da177e4SLinus Torvalds 450c154124fSLiam R. Howlett if (mapping) { 451c154124fSLiam R. Howlett __vma_link_file(vma, mapping); 45283cde9e8SDavidlohr Bueso i_mmap_unlock_write(mapping); 453c154124fSLiam R. Howlett } 4541da177e4SLinus Torvalds 4551da177e4SLinus Torvalds mm->map_count++; 4561da177e4SLinus Torvalds validate_mm(mm); 457d4af56c5SLiam R. Howlett return 0; 4581da177e4SLinus Torvalds } 4591da177e4SLinus Torvalds 4601da177e4SLinus Torvalds /* 461*68cefec5SLiam R. Howlett * init_multi_vma_prep() - Initializer for struct vma_prepare 462*68cefec5SLiam R. Howlett * @vp: The vma_prepare struct 463*68cefec5SLiam R. Howlett * @vma: The vma that will be altered once locked 464*68cefec5SLiam R. Howlett * @next: The next vma if it is to be adjusted 465*68cefec5SLiam R. Howlett * @remove: The first vma to be removed 466*68cefec5SLiam R. Howlett * @remove2: The second vma to be removed 467*68cefec5SLiam R. Howlett */ 468*68cefec5SLiam R. Howlett static inline void init_multi_vma_prep(struct vma_prepare *vp, 469*68cefec5SLiam R. Howlett struct vm_area_struct *vma, struct vm_area_struct *next, 470*68cefec5SLiam R. Howlett struct vm_area_struct *remove, struct vm_area_struct *remove2) 471*68cefec5SLiam R. Howlett { 472*68cefec5SLiam R. Howlett memset(vp, 0, sizeof(struct vma_prepare)); 473*68cefec5SLiam R. Howlett vp->vma = vma; 474*68cefec5SLiam R. Howlett vp->anon_vma = vma->anon_vma; 475*68cefec5SLiam R. Howlett vp->remove = remove; 476*68cefec5SLiam R. Howlett vp->remove2 = remove2; 477*68cefec5SLiam R. Howlett vp->adj_next = next; 478*68cefec5SLiam R. Howlett if (!vp->anon_vma && next) 479*68cefec5SLiam R. Howlett vp->anon_vma = next->anon_vma; 480*68cefec5SLiam R. Howlett 481*68cefec5SLiam R. Howlett vp->file = vma->vm_file; 482*68cefec5SLiam R. Howlett if (vp->file) 483*68cefec5SLiam R. Howlett vp->mapping = vma->vm_file->f_mapping; 484*68cefec5SLiam R. Howlett 485*68cefec5SLiam R. Howlett } 486*68cefec5SLiam R. Howlett 487*68cefec5SLiam R. Howlett /* 488*68cefec5SLiam R. Howlett * init_vma_prep() - Initializer wrapper for vma_prepare struct 489*68cefec5SLiam R. Howlett * @vp: The vma_prepare struct 490*68cefec5SLiam R. Howlett * @vma: The vma that will be altered once locked 491*68cefec5SLiam R. Howlett */ 492*68cefec5SLiam R. Howlett static inline void init_vma_prep(struct vma_prepare *vp, 493*68cefec5SLiam R. Howlett struct vm_area_struct *vma) 494*68cefec5SLiam R. Howlett { 495*68cefec5SLiam R. Howlett init_multi_vma_prep(vp, vma, NULL, NULL, NULL); 496*68cefec5SLiam R. Howlett } 497*68cefec5SLiam R. Howlett 498*68cefec5SLiam R. Howlett 499*68cefec5SLiam R. Howlett /* 500440703e0SLiam R. Howlett * vma_prepare() - Helper function for handling locking VMAs prior to altering 501440703e0SLiam R. Howlett * @vp: The initialized vma_prepare struct 502440703e0SLiam R. Howlett */ 503440703e0SLiam R. Howlett static inline void vma_prepare(struct vma_prepare *vp) 504440703e0SLiam R. Howlett { 505440703e0SLiam R. Howlett if (vp->file) { 506440703e0SLiam R. Howlett uprobe_munmap(vp->vma, vp->vma->vm_start, vp->vma->vm_end); 507440703e0SLiam R. Howlett 508440703e0SLiam R. Howlett if (vp->adj_next) 509440703e0SLiam R. Howlett uprobe_munmap(vp->adj_next, vp->adj_next->vm_start, 510440703e0SLiam R. Howlett vp->adj_next->vm_end); 511440703e0SLiam R. Howlett 512440703e0SLiam R. Howlett i_mmap_lock_write(vp->mapping); 513440703e0SLiam R. Howlett if (vp->insert && vp->insert->vm_file) { 514440703e0SLiam R. Howlett /* 515440703e0SLiam R. Howlett * Put into interval tree now, so instantiated pages 516440703e0SLiam R. Howlett * are visible to arm/parisc __flush_dcache_page 517440703e0SLiam R. Howlett * throughout; but we cannot insert into address 518440703e0SLiam R. Howlett * space until vma start or end is updated. 519440703e0SLiam R. Howlett */ 520440703e0SLiam R. Howlett __vma_link_file(vp->insert, 521440703e0SLiam R. Howlett vp->insert->vm_file->f_mapping); 522440703e0SLiam R. Howlett } 523440703e0SLiam R. Howlett } 524440703e0SLiam R. Howlett 525440703e0SLiam R. Howlett if (vp->anon_vma) { 526440703e0SLiam R. Howlett anon_vma_lock_write(vp->anon_vma); 527440703e0SLiam R. Howlett anon_vma_interval_tree_pre_update_vma(vp->vma); 528440703e0SLiam R. Howlett if (vp->adj_next) 529440703e0SLiam R. Howlett anon_vma_interval_tree_pre_update_vma(vp->adj_next); 530440703e0SLiam R. Howlett } 531440703e0SLiam R. Howlett 532440703e0SLiam R. Howlett if (vp->file) { 533440703e0SLiam R. Howlett flush_dcache_mmap_lock(vp->mapping); 534440703e0SLiam R. Howlett vma_interval_tree_remove(vp->vma, &vp->mapping->i_mmap); 535440703e0SLiam R. Howlett if (vp->adj_next) 536440703e0SLiam R. Howlett vma_interval_tree_remove(vp->adj_next, 537440703e0SLiam R. Howlett &vp->mapping->i_mmap); 538440703e0SLiam R. Howlett } 539440703e0SLiam R. Howlett 540440703e0SLiam R. Howlett } 541440703e0SLiam R. Howlett 542440703e0SLiam R. Howlett /* 543440703e0SLiam R. Howlett * vma_complete- Helper function for handling the unlocking after altering VMAs, 544440703e0SLiam R. Howlett * or for inserting a VMA. 545440703e0SLiam R. Howlett * 546440703e0SLiam R. Howlett * @vp: The vma_prepare struct 547440703e0SLiam R. Howlett * @vmi: The vma iterator 548440703e0SLiam R. Howlett * @mm: The mm_struct 549440703e0SLiam R. Howlett */ 550440703e0SLiam R. Howlett static inline void vma_complete(struct vma_prepare *vp, 551440703e0SLiam R. Howlett struct vma_iterator *vmi, struct mm_struct *mm) 552440703e0SLiam R. Howlett { 553440703e0SLiam R. Howlett if (vp->file) { 554440703e0SLiam R. Howlett if (vp->adj_next) 555440703e0SLiam R. Howlett vma_interval_tree_insert(vp->adj_next, 556440703e0SLiam R. Howlett &vp->mapping->i_mmap); 557440703e0SLiam R. Howlett vma_interval_tree_insert(vp->vma, &vp->mapping->i_mmap); 558440703e0SLiam R. Howlett flush_dcache_mmap_unlock(vp->mapping); 559440703e0SLiam R. Howlett } 560440703e0SLiam R. Howlett 561440703e0SLiam R. Howlett if (vp->remove && vp->file) { 562440703e0SLiam R. Howlett __remove_shared_vm_struct(vp->remove, vp->file, vp->mapping); 563440703e0SLiam R. Howlett if (vp->remove2) 564440703e0SLiam R. Howlett __remove_shared_vm_struct(vp->remove2, vp->file, 565440703e0SLiam R. Howlett vp->mapping); 566440703e0SLiam R. Howlett } else if (vp->insert) { 567440703e0SLiam R. Howlett /* 568440703e0SLiam R. Howlett * split_vma has split insert from vma, and needs 569440703e0SLiam R. Howlett * us to insert it before dropping the locks 570440703e0SLiam R. Howlett * (it may either follow vma or precede it). 571440703e0SLiam R. Howlett */ 572440703e0SLiam R. Howlett vma_iter_store(vmi, vp->insert); 573440703e0SLiam R. Howlett mm->map_count++; 574440703e0SLiam R. Howlett } 575440703e0SLiam R. Howlett 576440703e0SLiam R. Howlett if (vp->anon_vma) { 577440703e0SLiam R. Howlett anon_vma_interval_tree_post_update_vma(vp->vma); 578440703e0SLiam R. Howlett if (vp->adj_next) 579440703e0SLiam R. Howlett anon_vma_interval_tree_post_update_vma(vp->adj_next); 580440703e0SLiam R. Howlett anon_vma_unlock_write(vp->anon_vma); 581440703e0SLiam R. Howlett } 582440703e0SLiam R. Howlett 583440703e0SLiam R. Howlett if (vp->file) { 584440703e0SLiam R. Howlett i_mmap_unlock_write(vp->mapping); 585440703e0SLiam R. Howlett uprobe_mmap(vp->vma); 586440703e0SLiam R. Howlett 587440703e0SLiam R. Howlett if (vp->adj_next) 588440703e0SLiam R. Howlett uprobe_mmap(vp->adj_next); 589440703e0SLiam R. Howlett } 590440703e0SLiam R. Howlett 591440703e0SLiam R. Howlett if (vp->remove) { 592440703e0SLiam R. Howlett again: 593440703e0SLiam R. Howlett if (vp->file) { 594440703e0SLiam R. Howlett uprobe_munmap(vp->remove, vp->remove->vm_start, 595440703e0SLiam R. Howlett vp->remove->vm_end); 596440703e0SLiam R. Howlett fput(vp->file); 597440703e0SLiam R. Howlett } 598440703e0SLiam R. Howlett if (vp->remove->anon_vma) 599440703e0SLiam R. Howlett anon_vma_merge(vp->vma, vp->remove); 600440703e0SLiam R. Howlett mm->map_count--; 601440703e0SLiam R. Howlett mpol_put(vma_policy(vp->remove)); 602440703e0SLiam R. Howlett if (!vp->remove2) 603440703e0SLiam R. Howlett WARN_ON_ONCE(vp->vma->vm_end < vp->remove->vm_end); 604440703e0SLiam R. Howlett vm_area_free(vp->remove); 605440703e0SLiam R. Howlett 606440703e0SLiam R. Howlett /* 607440703e0SLiam R. Howlett * In mprotect's case 6 (see comments on vma_merge), 608*68cefec5SLiam R. Howlett * we must remove the one after next as well. 609440703e0SLiam R. Howlett */ 610440703e0SLiam R. Howlett if (vp->remove2) { 611440703e0SLiam R. Howlett vp->remove = vp->remove2; 612440703e0SLiam R. Howlett vp->remove2 = NULL; 613440703e0SLiam R. Howlett goto again; 614440703e0SLiam R. Howlett } 615440703e0SLiam R. Howlett } 616440703e0SLiam R. Howlett if (vp->insert && vp->file) 617440703e0SLiam R. Howlett uprobe_mmap(vp->insert); 618440703e0SLiam R. Howlett } 619440703e0SLiam R. Howlett 620440703e0SLiam R. Howlett /* 6219303d3e1SLiam R. Howlett * vma_expand - Expand an existing VMA 6229303d3e1SLiam R. Howlett * 6239303d3e1SLiam R. Howlett * @vmi: The vma iterator 6249303d3e1SLiam R. Howlett * @vma: The vma to expand 6259303d3e1SLiam R. Howlett * @start: The start of the vma 6269303d3e1SLiam R. Howlett * @end: The exclusive end of the vma 6279303d3e1SLiam R. Howlett * @pgoff: The page offset of vma 6289303d3e1SLiam R. Howlett * @next: The current of next vma. 6299303d3e1SLiam R. Howlett * 6309303d3e1SLiam R. Howlett * Expand @vma to @start and @end. Can expand off the start and end. Will 6319303d3e1SLiam R. Howlett * expand over @next if it's different from @vma and @end == @next->vm_end. 6329303d3e1SLiam R. Howlett * Checking if the @vma can expand and merge with @next needs to be handled by 6339303d3e1SLiam R. Howlett * the caller. 6349303d3e1SLiam R. Howlett * 6359303d3e1SLiam R. Howlett * Returns: 0 on success 6369303d3e1SLiam R. Howlett */ 6379303d3e1SLiam R. Howlett inline int vma_expand(struct vma_iterator *vmi, struct vm_area_struct *vma, 6389303d3e1SLiam R. Howlett unsigned long start, unsigned long end, pgoff_t pgoff, 6399303d3e1SLiam R. Howlett struct vm_area_struct *next) 6409303d3e1SLiam R. Howlett { 641*68cefec5SLiam R. Howlett bool remove_next = false; 6429303d3e1SLiam R. Howlett struct vma_prepare vp; 6439303d3e1SLiam R. Howlett 6449303d3e1SLiam R. Howlett if (next && (vma != next) && (end == next->vm_end)) { 645*68cefec5SLiam R. Howlett remove_next = true; 6469303d3e1SLiam R. Howlett if (next->anon_vma && !vma->anon_vma) { 6479303d3e1SLiam R. Howlett int error; 6489303d3e1SLiam R. Howlett 6499303d3e1SLiam R. Howlett vma->anon_vma = next->anon_vma; 6509303d3e1SLiam R. Howlett error = anon_vma_clone(vma, next); 6519303d3e1SLiam R. Howlett if (error) 6529303d3e1SLiam R. Howlett return error; 6539303d3e1SLiam R. Howlett } 6549303d3e1SLiam R. Howlett } 6559303d3e1SLiam R. Howlett 656*68cefec5SLiam R. Howlett init_multi_vma_prep(&vp, vma, NULL, remove_next ? next : NULL, NULL); 6579303d3e1SLiam R. Howlett /* Not merging but overwriting any part of next is not handled. */ 6589303d3e1SLiam R. Howlett VM_WARN_ON(next && !vp.remove && 6599303d3e1SLiam R. Howlett next != vma && end > next->vm_start); 6609303d3e1SLiam R. Howlett /* Only handles expanding */ 6619303d3e1SLiam R. Howlett VM_WARN_ON(vma->vm_start < start || vma->vm_end > end); 6629303d3e1SLiam R. Howlett 6639303d3e1SLiam R. Howlett if (vma_iter_prealloc(vmi)) 6649303d3e1SLiam R. Howlett goto nomem; 6659303d3e1SLiam R. Howlett 6669303d3e1SLiam R. Howlett vma_adjust_trans_huge(vma, start, end, 0); 6679303d3e1SLiam R. Howlett /* VMA iterator points to previous, so set to start if necessary */ 6689303d3e1SLiam R. Howlett if (vma_iter_addr(vmi) != start) 6699303d3e1SLiam R. Howlett vma_iter_set(vmi, start); 6709303d3e1SLiam R. Howlett 6719303d3e1SLiam R. Howlett vma_prepare(&vp); 6729303d3e1SLiam R. Howlett vma->vm_start = start; 6739303d3e1SLiam R. Howlett vma->vm_end = end; 6749303d3e1SLiam R. Howlett vma->vm_pgoff = pgoff; 6759303d3e1SLiam R. Howlett /* Note: mas must be pointing to the expanding VMA */ 6769303d3e1SLiam R. Howlett vma_iter_store(vmi, vma); 6779303d3e1SLiam R. Howlett 6789303d3e1SLiam R. Howlett vma_complete(&vp, vmi, vma->vm_mm); 6799303d3e1SLiam R. Howlett validate_mm(vma->vm_mm); 6809303d3e1SLiam R. Howlett return 0; 6819303d3e1SLiam R. Howlett 6829303d3e1SLiam R. Howlett nomem: 6839303d3e1SLiam R. Howlett return -ENOMEM; 6849303d3e1SLiam R. Howlett } 6859303d3e1SLiam R. Howlett /* 6861da177e4SLinus Torvalds * We cannot adjust vm_start, vm_end, vm_pgoff fields of a vma that 6871da177e4SLinus Torvalds * is already present in an i_mmap tree without adjusting the tree. 6881da177e4SLinus Torvalds * The following helper function should be used when such adjustments 6891da177e4SLinus Torvalds * are necessary. The "insert" vma (if any) is to be inserted 6901da177e4SLinus Torvalds * before we drop the necessary locks. 6911da177e4SLinus Torvalds */ 6929e560446SLiam R. Howlett int __vma_adjust(struct vma_iterator *vmi, struct vm_area_struct *vma, 6939e560446SLiam R. Howlett unsigned long start, unsigned long end, pgoff_t pgoff, 6949e560446SLiam R. Howlett struct vm_area_struct *insert, struct vm_area_struct *expand) 6951da177e4SLinus Torvalds { 6961da177e4SLinus Torvalds struct mm_struct *mm = vma->vm_mm; 697*68cefec5SLiam R. Howlett struct vm_area_struct *remove2 = NULL; 698*68cefec5SLiam R. Howlett struct vm_area_struct *remove = NULL; 6991cd916d0SAndrew Morton struct vm_area_struct *next = find_vma(mm, vma->vm_end); 700524e00b3SLiam R. Howlett struct vm_area_struct *orig_vma = vma; 7011da177e4SLinus Torvalds struct file *file = vma->vm_file; 702524e00b3SLiam R. Howlett bool vma_changed = false; 7031da177e4SLinus Torvalds long adjust_next = 0; 704734537c9SKirill A. Shutemov struct vm_area_struct *exporter = NULL, *importer = NULL; 705440703e0SLiam R. Howlett struct vma_prepare vma_prep; 706287d97acSLinus Torvalds 707d4af56c5SLiam R. Howlett if (next && !insert) { 7081da177e4SLinus Torvalds if (end >= next->vm_end) { 7091da177e4SLinus Torvalds /* 7101da177e4SLinus Torvalds * vma expands, overlapping all the next, and 7111da177e4SLinus Torvalds * perhaps the one after too (mprotect case 6). 71286d12e47SAndrea Arcangeli * The only other cases that gets here are 713e86f15eeSAndrea Arcangeli * case 1, case 7 and case 8. 714e86f15eeSAndrea Arcangeli */ 715e86f15eeSAndrea Arcangeli if (next == expand) { 716e86f15eeSAndrea Arcangeli /* 717e86f15eeSAndrea Arcangeli * The only case where we don't expand "vma" 718e86f15eeSAndrea Arcangeli * and we expand "next" instead is case 8. 719e86f15eeSAndrea Arcangeli */ 720e86f15eeSAndrea Arcangeli VM_WARN_ON(end != next->vm_end); 721e86f15eeSAndrea Arcangeli /* 722*68cefec5SLiam R. Howlett * we're removing "vma" and that to do so we 723e86f15eeSAndrea Arcangeli * swapped "vma" and "next". 724e86f15eeSAndrea Arcangeli */ 725e86f15eeSAndrea Arcangeli VM_WARN_ON(file != next->vm_file); 726e86f15eeSAndrea Arcangeli swap(vma, next); 727*68cefec5SLiam R. Howlett remove = next; 728e86f15eeSAndrea Arcangeli } else { 729e86f15eeSAndrea Arcangeli VM_WARN_ON(expand != vma); 730e86f15eeSAndrea Arcangeli /* 731*68cefec5SLiam R. Howlett * case 1, 6, 7, remove next. 732*68cefec5SLiam R. Howlett * case 6 also removes the one beyond next 7331da177e4SLinus Torvalds */ 734*68cefec5SLiam R. Howlett remove = next; 735*68cefec5SLiam R. Howlett if (end > next->vm_end) 736*68cefec5SLiam R. Howlett remove2 = find_vma(mm, next->vm_end); 737d4af56c5SLiam R. Howlett 738*68cefec5SLiam R. Howlett VM_WARN_ON(remove2 != NULL && 739*68cefec5SLiam R. Howlett end != remove2->vm_end); 740e86f15eeSAndrea Arcangeli } 741e86f15eeSAndrea Arcangeli 742287d97acSLinus Torvalds exporter = next; 7431da177e4SLinus Torvalds importer = vma; 744734537c9SKirill A. Shutemov 745734537c9SKirill A. Shutemov /* 746734537c9SKirill A. Shutemov * If next doesn't have anon_vma, import from vma after 747734537c9SKirill A. Shutemov * next, if the vma overlaps with it. 748734537c9SKirill A. Shutemov */ 749*68cefec5SLiam R. Howlett if (remove2 != NULL && !next->anon_vma) 750*68cefec5SLiam R. Howlett exporter = remove2; 751734537c9SKirill A. Shutemov 7521da177e4SLinus Torvalds } else if (end > next->vm_start) { 7531da177e4SLinus Torvalds /* 7541da177e4SLinus Torvalds * vma expands, overlapping part of the next: 7551da177e4SLinus Torvalds * mprotect case 5 shifting the boundary up. 7561da177e4SLinus Torvalds */ 757f9d86a60SWei Yang adjust_next = (end - next->vm_start); 758287d97acSLinus Torvalds exporter = next; 7591da177e4SLinus Torvalds importer = vma; 760e86f15eeSAndrea Arcangeli VM_WARN_ON(expand != importer); 7611da177e4SLinus Torvalds } else if (end < vma->vm_end) { 7621da177e4SLinus Torvalds /* 7631da177e4SLinus Torvalds * vma shrinks, and !insert tells it's not 7641da177e4SLinus Torvalds * split_vma inserting another: so it must be 7651da177e4SLinus Torvalds * mprotect case 4 shifting the boundary down. 7661da177e4SLinus Torvalds */ 767f9d86a60SWei Yang adjust_next = -(vma->vm_end - end); 768287d97acSLinus Torvalds exporter = vma; 7691da177e4SLinus Torvalds importer = next; 770e86f15eeSAndrea Arcangeli VM_WARN_ON(expand != importer); 7711da177e4SLinus Torvalds } 7721da177e4SLinus Torvalds 7735beb4930SRik van Riel /* 7745beb4930SRik van Riel * Easily overlooked: when mprotect shifts the boundary, 7755beb4930SRik van Riel * make sure the expanding vma has anon_vma set if the 7765beb4930SRik van Riel * shrinking vma had, to cover any anon pages imported. 7775beb4930SRik van Riel */ 778287d97acSLinus Torvalds if (exporter && exporter->anon_vma && !importer->anon_vma) { 779c4ea95d7SDaniel Forrest int error; 780c4ea95d7SDaniel Forrest 781287d97acSLinus Torvalds importer->anon_vma = exporter->anon_vma; 782b800c91aSKonstantin Khlebnikov error = anon_vma_clone(importer, exporter); 7833fe89b3eSLeon Yu if (error) 784b800c91aSKonstantin Khlebnikov return error; 785b800c91aSKonstantin Khlebnikov } 7865beb4930SRik van Riel } 78737f9f559SKirill A. Shutemov 7889e560446SLiam R. Howlett if (vma_iter_prealloc(vmi)) 789d4af56c5SLiam R. Howlett return -ENOMEM; 790d4af56c5SLiam R. Howlett 791d4af56c5SLiam R. Howlett vma_adjust_trans_huge(orig_vma, start, end, adjust_next); 792682968e0SSrikar Dronamraju 793*68cefec5SLiam R. Howlett init_multi_vma_prep(&vma_prep, vma, adjust_next ? next : NULL, remove, 794*68cefec5SLiam R. Howlett remove2); 795*68cefec5SLiam R. Howlett VM_WARN_ON(vma_prep.anon_vma && adjust_next && next->anon_vma && 796*68cefec5SLiam R. Howlett vma_prep.anon_vma != next->anon_vma); 7971da177e4SLinus Torvalds 798*68cefec5SLiam R. Howlett vma_prep.insert = insert; 799440703e0SLiam R. Howlett vma_prepare(&vma_prep); 8001da177e4SLinus Torvalds 801d3737187SMichel Lespinasse if (start != vma->vm_start) { 802c465be97SLiam R. Howlett if (vma->vm_start < start) { 803c465be97SLiam R. Howlett if (!insert || (insert->vm_end != start)) { 8049e560446SLiam R. Howlett vma_iter_clear(vmi, vma->vm_start, start); 805c465be97SLiam R. Howlett vma_iter_set(vmi, start); 8063b0e81a1SLiam R. Howlett VM_WARN_ON(insert && insert->vm_start > vma->vm_start); 807c465be97SLiam R. Howlett } 8083b0e81a1SLiam R. Howlett } else { 809524e00b3SLiam R. Howlett vma_changed = true; 8103b0e81a1SLiam R. Howlett } 8111da177e4SLinus Torvalds vma->vm_start = start; 812d3737187SMichel Lespinasse } 813d3737187SMichel Lespinasse if (end != vma->vm_end) { 8143b0e81a1SLiam R. Howlett if (vma->vm_end > end) { 8153b0e81a1SLiam R. Howlett if (!insert || (insert->vm_start != end)) { 8169e560446SLiam R. Howlett vma_iter_clear(vmi, end, vma->vm_end); 8179e560446SLiam R. Howlett vma_iter_set(vmi, vma->vm_end); 8183b0e81a1SLiam R. Howlett VM_WARN_ON(insert && 8193b0e81a1SLiam R. Howlett insert->vm_end < vma->vm_end); 8203b0e81a1SLiam R. Howlett } 8213b0e81a1SLiam R. Howlett } else { 822524e00b3SLiam R. Howlett vma_changed = true; 8233b0e81a1SLiam R. Howlett } 8241da177e4SLinus Torvalds vma->vm_end = end; 825d3737187SMichel Lespinasse } 826d4af56c5SLiam R. Howlett 827524e00b3SLiam R. Howlett if (vma_changed) 8289e560446SLiam R. Howlett vma_iter_store(vmi, vma); 829d4af56c5SLiam R. Howlett 8301da177e4SLinus Torvalds vma->vm_pgoff = pgoff; 8311da177e4SLinus Torvalds if (adjust_next) { 832f9d86a60SWei Yang next->vm_start += adjust_next; 833f9d86a60SWei Yang next->vm_pgoff += adjust_next >> PAGE_SHIFT; 8349e560446SLiam R. Howlett vma_iter_store(vmi, next); 8351da177e4SLinus Torvalds } 8361da177e4SLinus Torvalds 837440703e0SLiam R. Howlett vma_complete(&vma_prep, vmi, mm); 8389e560446SLiam R. Howlett vma_iter_free(vmi); 8391da177e4SLinus Torvalds validate_mm(mm); 840763ecb03SLiam R. Howlett 8415beb4930SRik van Riel return 0; 8421da177e4SLinus Torvalds } 8431da177e4SLinus Torvalds 8441da177e4SLinus Torvalds /* 8451da177e4SLinus Torvalds * If the vma has a ->close operation then the driver probably needs to release 8461da177e4SLinus Torvalds * per-vma resources, so we don't attempt to merge those. 8471da177e4SLinus Torvalds */ 8481da177e4SLinus Torvalds static inline int is_mergeable_vma(struct vm_area_struct *vma, 84919a809afSAndrea Arcangeli struct file *file, unsigned long vm_flags, 8509a10064fSColin Cross struct vm_userfaultfd_ctx vm_userfaultfd_ctx, 8515c26f6acSSuren Baghdasaryan struct anon_vma_name *anon_name) 8521da177e4SLinus Torvalds { 85334228d47SCyrill Gorcunov /* 85434228d47SCyrill Gorcunov * VM_SOFTDIRTY should not prevent from VMA merging, if we 85534228d47SCyrill Gorcunov * match the flags but dirty bit -- the caller should mark 85634228d47SCyrill Gorcunov * merged VMA as dirty. If dirty bit won't be excluded from 8578bb4e7a2SWei Yang * comparison, we increase pressure on the memory system forcing 85834228d47SCyrill Gorcunov * the kernel to generate new VMAs when old one could be 85934228d47SCyrill Gorcunov * extended instead. 86034228d47SCyrill Gorcunov */ 86134228d47SCyrill Gorcunov if ((vma->vm_flags ^ vm_flags) & ~VM_SOFTDIRTY) 8621da177e4SLinus Torvalds return 0; 8631da177e4SLinus Torvalds if (vma->vm_file != file) 8641da177e4SLinus Torvalds return 0; 8651da177e4SLinus Torvalds if (vma->vm_ops && vma->vm_ops->close) 8661da177e4SLinus Torvalds return 0; 86719a809afSAndrea Arcangeli if (!is_mergeable_vm_userfaultfd_ctx(vma, vm_userfaultfd_ctx)) 86819a809afSAndrea Arcangeli return 0; 8695c26f6acSSuren Baghdasaryan if (!anon_vma_name_eq(anon_vma_name(vma), anon_name)) 8709a10064fSColin Cross return 0; 8711da177e4SLinus Torvalds return 1; 8721da177e4SLinus Torvalds } 8731da177e4SLinus Torvalds 8741da177e4SLinus Torvalds static inline int is_mergeable_anon_vma(struct anon_vma *anon_vma1, 875965f55deSShaohua Li struct anon_vma *anon_vma2, 876965f55deSShaohua Li struct vm_area_struct *vma) 8771da177e4SLinus Torvalds { 878965f55deSShaohua Li /* 879965f55deSShaohua Li * The list_is_singular() test is to avoid merging VMA cloned from 880965f55deSShaohua Li * parents. This can improve scalability caused by anon_vma lock. 881965f55deSShaohua Li */ 882965f55deSShaohua Li if ((!anon_vma1 || !anon_vma2) && (!vma || 883965f55deSShaohua Li list_is_singular(&vma->anon_vma_chain))) 884965f55deSShaohua Li return 1; 885965f55deSShaohua Li return anon_vma1 == anon_vma2; 8861da177e4SLinus Torvalds } 8871da177e4SLinus Torvalds 8881da177e4SLinus Torvalds /* 8891da177e4SLinus Torvalds * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff) 8901da177e4SLinus Torvalds * in front of (at a lower virtual address and file offset than) the vma. 8911da177e4SLinus Torvalds * 8921da177e4SLinus Torvalds * We cannot merge two vmas if they have differently assigned (non-NULL) 8931da177e4SLinus Torvalds * anon_vmas, nor if same anon_vma is assigned but offsets incompatible. 8941da177e4SLinus Torvalds * 8951da177e4SLinus Torvalds * We don't check here for the merged mmap wrapping around the end of pagecache 89645e55300SPeter Collingbourne * indices (16TB on ia32) because do_mmap() does not permit mmap's which 8971da177e4SLinus Torvalds * wrap, nor mmaps which cover the final page at index -1UL. 8981da177e4SLinus Torvalds */ 8991da177e4SLinus Torvalds static int 9001da177e4SLinus Torvalds can_vma_merge_before(struct vm_area_struct *vma, unsigned long vm_flags, 90119a809afSAndrea Arcangeli struct anon_vma *anon_vma, struct file *file, 90219a809afSAndrea Arcangeli pgoff_t vm_pgoff, 9039a10064fSColin Cross struct vm_userfaultfd_ctx vm_userfaultfd_ctx, 9045c26f6acSSuren Baghdasaryan struct anon_vma_name *anon_name) 9051da177e4SLinus Torvalds { 9069a10064fSColin Cross if (is_mergeable_vma(vma, file, vm_flags, vm_userfaultfd_ctx, anon_name) && 907965f55deSShaohua Li is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) { 9081da177e4SLinus Torvalds if (vma->vm_pgoff == vm_pgoff) 9091da177e4SLinus Torvalds return 1; 9101da177e4SLinus Torvalds } 9111da177e4SLinus Torvalds return 0; 9121da177e4SLinus Torvalds } 9131da177e4SLinus Torvalds 9141da177e4SLinus Torvalds /* 9151da177e4SLinus Torvalds * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff) 9161da177e4SLinus Torvalds * beyond (at a higher virtual address and file offset than) the vma. 9171da177e4SLinus Torvalds * 9181da177e4SLinus Torvalds * We cannot merge two vmas if they have differently assigned (non-NULL) 9191da177e4SLinus Torvalds * anon_vmas, nor if same anon_vma is assigned but offsets incompatible. 9201da177e4SLinus Torvalds */ 9211da177e4SLinus Torvalds static int 9221da177e4SLinus Torvalds can_vma_merge_after(struct vm_area_struct *vma, unsigned long vm_flags, 92319a809afSAndrea Arcangeli struct anon_vma *anon_vma, struct file *file, 92419a809afSAndrea Arcangeli pgoff_t vm_pgoff, 9259a10064fSColin Cross struct vm_userfaultfd_ctx vm_userfaultfd_ctx, 9265c26f6acSSuren Baghdasaryan struct anon_vma_name *anon_name) 9271da177e4SLinus Torvalds { 9289a10064fSColin Cross if (is_mergeable_vma(vma, file, vm_flags, vm_userfaultfd_ctx, anon_name) && 929965f55deSShaohua Li is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) { 9301da177e4SLinus Torvalds pgoff_t vm_pglen; 931d6e93217SLibin vm_pglen = vma_pages(vma); 9321da177e4SLinus Torvalds if (vma->vm_pgoff + vm_pglen == vm_pgoff) 9331da177e4SLinus Torvalds return 1; 9341da177e4SLinus Torvalds } 9351da177e4SLinus Torvalds return 0; 9361da177e4SLinus Torvalds } 9371da177e4SLinus Torvalds 9381da177e4SLinus Torvalds /* 9399a10064fSColin Cross * Given a mapping request (addr,end,vm_flags,file,pgoff,anon_name), 9409a10064fSColin Cross * figure out whether that can be merged with its predecessor or its 9419a10064fSColin Cross * successor. Or both (it neatly fills a hole). 9421da177e4SLinus Torvalds * 9431da177e4SLinus Torvalds * In most cases - when called for mmap, brk or mremap - [addr,end) is 9441da177e4SLinus Torvalds * certain not to be mapped by the time vma_merge is called; but when 9451da177e4SLinus Torvalds * called for mprotect, it is certain to be already mapped (either at 9461da177e4SLinus Torvalds * an offset within prev, or at the start of next), and the flags of 9471da177e4SLinus Torvalds * this area are about to be changed to vm_flags - and the no-change 9481da177e4SLinus Torvalds * case has already been eliminated. 9491da177e4SLinus Torvalds * 9501da177e4SLinus Torvalds * The following mprotect cases have to be considered, where AAAA is 9511da177e4SLinus Torvalds * the area passed down from mprotect_fixup, never extending beyond one 9521da177e4SLinus Torvalds * vma, PPPPPP is the prev vma specified, and NNNNNN the next vma after: 9531da177e4SLinus Torvalds * 9545d42ab29SWei Yang * AAAA AAAA AAAA 9555d42ab29SWei Yang * PPPPPPNNNNNN PPPPPPNNNNNN PPPPPPNNNNNN 9565d42ab29SWei Yang * cannot merge might become might become 9575d42ab29SWei Yang * PPNNNNNNNNNN PPPPPPPPPPNN 9585d42ab29SWei Yang * mmap, brk or case 4 below case 5 below 9595d42ab29SWei Yang * mremap move: 9605d42ab29SWei Yang * AAAA AAAA 9615d42ab29SWei Yang * PPPP NNNN PPPPNNNNXXXX 9625d42ab29SWei Yang * might become might become 9635d42ab29SWei Yang * PPPPPPPPPPPP 1 or PPPPPPPPPPPP 6 or 9645d42ab29SWei Yang * PPPPPPPPNNNN 2 or PPPPPPPPXXXX 7 or 9655d42ab29SWei Yang * PPPPNNNNNNNN 3 PPPPXXXXXXXX 8 9661da177e4SLinus Torvalds * 9678bb4e7a2SWei Yang * It is important for case 8 that the vma NNNN overlapping the 968e86f15eeSAndrea Arcangeli * region AAAA is never going to extended over XXXX. Instead XXXX must 969e86f15eeSAndrea Arcangeli * be extended in region AAAA and NNNN must be removed. This way in 970e86f15eeSAndrea Arcangeli * all cases where vma_merge succeeds, the moment vma_adjust drops the 971e86f15eeSAndrea Arcangeli * rmap_locks, the properties of the merged vma will be already 972e86f15eeSAndrea Arcangeli * correct for the whole merged range. Some of those properties like 973e86f15eeSAndrea Arcangeli * vm_page_prot/vm_flags may be accessed by rmap_walks and they must 974e86f15eeSAndrea Arcangeli * be correct for the whole merged range immediately after the 975e86f15eeSAndrea Arcangeli * rmap_locks are released. Otherwise if XXXX would be removed and 976e86f15eeSAndrea Arcangeli * NNNN would be extended over the XXXX range, remove_migration_ptes 977e86f15eeSAndrea Arcangeli * or other rmap walkers (if working on addresses beyond the "end" 978e86f15eeSAndrea Arcangeli * parameter) may establish ptes with the wrong permissions of NNNN 979e86f15eeSAndrea Arcangeli * instead of the right permissions of XXXX. 9801da177e4SLinus Torvalds */ 9819760ebffSLiam R. Howlett struct vm_area_struct *vma_merge(struct vma_iterator *vmi, struct mm_struct *mm, 9821da177e4SLinus Torvalds struct vm_area_struct *prev, unsigned long addr, 9831da177e4SLinus Torvalds unsigned long end, unsigned long vm_flags, 9841da177e4SLinus Torvalds struct anon_vma *anon_vma, struct file *file, 98519a809afSAndrea Arcangeli pgoff_t pgoff, struct mempolicy *policy, 9869a10064fSColin Cross struct vm_userfaultfd_ctx vm_userfaultfd_ctx, 9875c26f6acSSuren Baghdasaryan struct anon_vma_name *anon_name) 9881da177e4SLinus Torvalds { 9891da177e4SLinus Torvalds pgoff_t pglen = (end - addr) >> PAGE_SHIFT; 9909760ebffSLiam R. Howlett struct vm_area_struct *mid, *next, *res = NULL; 991eef19944SJakub Matěna int err = -1; 992eef19944SJakub Matěna bool merge_prev = false; 993eef19944SJakub Matěna bool merge_next = false; 9941da177e4SLinus Torvalds 9951da177e4SLinus Torvalds /* 9961da177e4SLinus Torvalds * We later require that vma->vm_flags == vm_flags, 9971da177e4SLinus Torvalds * so this tests vma->vm_flags & VM_SPECIAL, too. 9981da177e4SLinus Torvalds */ 9991da177e4SLinus Torvalds if (vm_flags & VM_SPECIAL) 10001da177e4SLinus Torvalds return NULL; 10011da177e4SLinus Torvalds 1002763ecb03SLiam R. Howlett next = find_vma(mm, prev ? prev->vm_end : 0); 1003eef19944SJakub Matěna mid = next; 1004eef19944SJakub Matěna if (next && next->vm_end == end) /* cases 6, 7, 8 */ 1005763ecb03SLiam R. Howlett next = find_vma(mm, next->vm_end); 10061da177e4SLinus Torvalds 1007e86f15eeSAndrea Arcangeli /* verify some invariant that must be enforced by the caller */ 1008e86f15eeSAndrea Arcangeli VM_WARN_ON(prev && addr <= prev->vm_start); 1009eef19944SJakub Matěna VM_WARN_ON(mid && end > mid->vm_end); 1010e86f15eeSAndrea Arcangeli VM_WARN_ON(addr >= end); 1011e86f15eeSAndrea Arcangeli 1012eef19944SJakub Matěna /* Can we merge the predecessor? */ 10131da177e4SLinus Torvalds if (prev && prev->vm_end == addr && 10141da177e4SLinus Torvalds mpol_equal(vma_policy(prev), policy) && 10151da177e4SLinus Torvalds can_vma_merge_after(prev, vm_flags, 101619a809afSAndrea Arcangeli anon_vma, file, pgoff, 10179a10064fSColin Cross vm_userfaultfd_ctx, anon_name)) { 1018eef19944SJakub Matěna merge_prev = true; 10191da177e4SLinus Torvalds } 1020eef19944SJakub Matěna /* Can we merge the successor? */ 10211da177e4SLinus Torvalds if (next && end == next->vm_start && 10221da177e4SLinus Torvalds mpol_equal(policy, vma_policy(next)) && 10231da177e4SLinus Torvalds can_vma_merge_before(next, vm_flags, 102419a809afSAndrea Arcangeli anon_vma, file, pgoff+pglen, 10259a10064fSColin Cross vm_userfaultfd_ctx, anon_name)) { 1026eef19944SJakub Matěna merge_next = true; 1027eef19944SJakub Matěna } 1028eef19944SJakub Matěna /* Can we merge both the predecessor and the successor? */ 1029eef19944SJakub Matěna if (merge_prev && merge_next && 1030eef19944SJakub Matěna is_mergeable_anon_vma(prev->anon_vma, 1031eef19944SJakub Matěna next->anon_vma, NULL)) { /* cases 1, 6 */ 10329e560446SLiam R. Howlett err = __vma_adjust(vmi, prev, prev->vm_start, 1033eef19944SJakub Matěna next->vm_end, prev->vm_pgoff, NULL, 1034eef19944SJakub Matěna prev); 1035eef19944SJakub Matěna res = prev; 1036eef19944SJakub Matěna } else if (merge_prev) { /* cases 2, 5, 7 */ 10379e560446SLiam R. Howlett err = __vma_adjust(vmi, prev, prev->vm_start, 1038eef19944SJakub Matěna end, prev->vm_pgoff, NULL, prev); 1039eef19944SJakub Matěna res = prev; 1040eef19944SJakub Matěna } else if (merge_next) { 10411da177e4SLinus Torvalds if (prev && addr < prev->vm_end) /* case 4 */ 10429e560446SLiam R. Howlett err = __vma_adjust(vmi, prev, prev->vm_start, 1043e86f15eeSAndrea Arcangeli addr, prev->vm_pgoff, NULL, next); 1044eef19944SJakub Matěna else /* cases 3, 8 */ 10459e560446SLiam R. Howlett err = __vma_adjust(vmi, mid, addr, next->vm_end, 1046e86f15eeSAndrea Arcangeli next->vm_pgoff - pglen, NULL, next); 1047eef19944SJakub Matěna res = next; 10481da177e4SLinus Torvalds } 10491da177e4SLinus Torvalds 1050eef19944SJakub Matěna /* 1051eef19944SJakub Matěna * Cannot merge with predecessor or successor or error in __vma_adjust? 1052eef19944SJakub Matěna */ 1053eef19944SJakub Matěna if (err) 10541da177e4SLinus Torvalds return NULL; 1055eef19944SJakub Matěna khugepaged_enter_vma(res, vm_flags); 10561da177e4SLinus Torvalds 10579760ebffSLiam R. Howlett if (res) 1058f2ebfe43SLiam R. Howlett vma_iter_set(vmi, end); 1059f2ebfe43SLiam R. Howlett 10609760ebffSLiam R. Howlett return res; 1061f2ebfe43SLiam R. Howlett } 1062f2ebfe43SLiam R. Howlett 10631da177e4SLinus Torvalds /* 1064b4f315b4SEthon Paul * Rough compatibility check to quickly see if it's even worth looking 1065d0e9fe17SLinus Torvalds * at sharing an anon_vma. 1066d0e9fe17SLinus Torvalds * 1067d0e9fe17SLinus Torvalds * They need to have the same vm_file, and the flags can only differ 1068d0e9fe17SLinus Torvalds * in things that mprotect may change. 1069d0e9fe17SLinus Torvalds * 1070d0e9fe17SLinus Torvalds * NOTE! The fact that we share an anon_vma doesn't _have_ to mean that 1071d0e9fe17SLinus Torvalds * we can merge the two vma's. For example, we refuse to merge a vma if 1072d0e9fe17SLinus Torvalds * there is a vm_ops->close() function, because that indicates that the 1073d0e9fe17SLinus Torvalds * driver is doing some kind of reference counting. But that doesn't 1074d0e9fe17SLinus Torvalds * really matter for the anon_vma sharing case. 1075d0e9fe17SLinus Torvalds */ 1076d0e9fe17SLinus Torvalds static int anon_vma_compatible(struct vm_area_struct *a, struct vm_area_struct *b) 1077d0e9fe17SLinus Torvalds { 1078d0e9fe17SLinus Torvalds return a->vm_end == b->vm_start && 1079d0e9fe17SLinus Torvalds mpol_equal(vma_policy(a), vma_policy(b)) && 1080d0e9fe17SLinus Torvalds a->vm_file == b->vm_file && 10816cb4d9a2SAnshuman Khandual !((a->vm_flags ^ b->vm_flags) & ~(VM_ACCESS_FLAGS | VM_SOFTDIRTY)) && 1082d0e9fe17SLinus Torvalds b->vm_pgoff == a->vm_pgoff + ((b->vm_start - a->vm_start) >> PAGE_SHIFT); 1083d0e9fe17SLinus Torvalds } 1084d0e9fe17SLinus Torvalds 1085d0e9fe17SLinus Torvalds /* 1086d0e9fe17SLinus Torvalds * Do some basic sanity checking to see if we can re-use the anon_vma 1087d0e9fe17SLinus Torvalds * from 'old'. The 'a'/'b' vma's are in VM order - one of them will be 1088d0e9fe17SLinus Torvalds * the same as 'old', the other will be the new one that is trying 1089d0e9fe17SLinus Torvalds * to share the anon_vma. 1090d0e9fe17SLinus Torvalds * 10915b449489SFlorian Rommel * NOTE! This runs with mmap_lock held for reading, so it is possible that 1092d0e9fe17SLinus Torvalds * the anon_vma of 'old' is concurrently in the process of being set up 1093d0e9fe17SLinus Torvalds * by another page fault trying to merge _that_. But that's ok: if it 1094d0e9fe17SLinus Torvalds * is being set up, that automatically means that it will be a singleton 1095d0e9fe17SLinus Torvalds * acceptable for merging, so we can do all of this optimistically. But 10964db0c3c2SJason Low * we do that READ_ONCE() to make sure that we never re-load the pointer. 1097d0e9fe17SLinus Torvalds * 1098d0e9fe17SLinus Torvalds * IOW: that the "list_is_singular()" test on the anon_vma_chain only 1099d0e9fe17SLinus Torvalds * matters for the 'stable anon_vma' case (ie the thing we want to avoid 1100d0e9fe17SLinus Torvalds * is to return an anon_vma that is "complex" due to having gone through 1101d0e9fe17SLinus Torvalds * a fork). 1102d0e9fe17SLinus Torvalds * 1103d0e9fe17SLinus Torvalds * We also make sure that the two vma's are compatible (adjacent, 1104d0e9fe17SLinus Torvalds * and with the same memory policies). That's all stable, even with just 11055b449489SFlorian Rommel * a read lock on the mmap_lock. 1106d0e9fe17SLinus Torvalds */ 1107d0e9fe17SLinus Torvalds static struct anon_vma *reusable_anon_vma(struct vm_area_struct *old, struct vm_area_struct *a, struct vm_area_struct *b) 1108d0e9fe17SLinus Torvalds { 1109d0e9fe17SLinus Torvalds if (anon_vma_compatible(a, b)) { 11104db0c3c2SJason Low struct anon_vma *anon_vma = READ_ONCE(old->anon_vma); 1111d0e9fe17SLinus Torvalds 1112d0e9fe17SLinus Torvalds if (anon_vma && list_is_singular(&old->anon_vma_chain)) 1113d0e9fe17SLinus Torvalds return anon_vma; 1114d0e9fe17SLinus Torvalds } 1115d0e9fe17SLinus Torvalds return NULL; 1116d0e9fe17SLinus Torvalds } 1117d0e9fe17SLinus Torvalds 1118d0e9fe17SLinus Torvalds /* 11191da177e4SLinus Torvalds * find_mergeable_anon_vma is used by anon_vma_prepare, to check 11201da177e4SLinus Torvalds * neighbouring vmas for a suitable anon_vma, before it goes off 11211da177e4SLinus Torvalds * to allocate a new anon_vma. It checks because a repetitive 11221da177e4SLinus Torvalds * sequence of mprotects and faults may otherwise lead to distinct 11231da177e4SLinus Torvalds * anon_vmas being allocated, preventing vma merge in subsequent 11241da177e4SLinus Torvalds * mprotect. 11251da177e4SLinus Torvalds */ 11261da177e4SLinus Torvalds struct anon_vma *find_mergeable_anon_vma(struct vm_area_struct *vma) 11271da177e4SLinus Torvalds { 1128763ecb03SLiam R. Howlett MA_STATE(mas, &vma->vm_mm->mm_mt, vma->vm_end, vma->vm_end); 1129a67c8caaSMiaohe Lin struct anon_vma *anon_vma = NULL; 1130763ecb03SLiam R. Howlett struct vm_area_struct *prev, *next; 11311da177e4SLinus Torvalds 1132a67c8caaSMiaohe Lin /* Try next first. */ 1133763ecb03SLiam R. Howlett next = mas_walk(&mas); 1134763ecb03SLiam R. Howlett if (next) { 1135763ecb03SLiam R. Howlett anon_vma = reusable_anon_vma(next, vma, next); 1136d0e9fe17SLinus Torvalds if (anon_vma) 1137d0e9fe17SLinus Torvalds return anon_vma; 1138a67c8caaSMiaohe Lin } 11391da177e4SLinus Torvalds 1140763ecb03SLiam R. Howlett prev = mas_prev(&mas, 0); 1141763ecb03SLiam R. Howlett VM_BUG_ON_VMA(prev != vma, vma); 1142763ecb03SLiam R. Howlett prev = mas_prev(&mas, 0); 1143a67c8caaSMiaohe Lin /* Try prev next. */ 1144763ecb03SLiam R. Howlett if (prev) 1145763ecb03SLiam R. Howlett anon_vma = reusable_anon_vma(prev, prev, vma); 1146a67c8caaSMiaohe Lin 11471da177e4SLinus Torvalds /* 1148a67c8caaSMiaohe Lin * We might reach here with anon_vma == NULL if we can't find 1149a67c8caaSMiaohe Lin * any reusable anon_vma. 11501da177e4SLinus Torvalds * There's no absolute need to look only at touching neighbours: 11511da177e4SLinus Torvalds * we could search further afield for "compatible" anon_vmas. 11521da177e4SLinus Torvalds * But it would probably just be a waste of time searching, 11531da177e4SLinus Torvalds * or lead to too many vmas hanging off the same anon_vma. 11541da177e4SLinus Torvalds * We're trying to allow mprotect remerging later on, 11551da177e4SLinus Torvalds * not trying to minimize memory used for anon_vmas. 11561da177e4SLinus Torvalds */ 1157a67c8caaSMiaohe Lin return anon_vma; 11581da177e4SLinus Torvalds } 11591da177e4SLinus Torvalds 11601da177e4SLinus Torvalds /* 116140401530SAl Viro * If a hint addr is less than mmap_min_addr change hint to be as 116240401530SAl Viro * low as possible but still greater than mmap_min_addr 116340401530SAl Viro */ 116440401530SAl Viro static inline unsigned long round_hint_to_min(unsigned long hint) 116540401530SAl Viro { 116640401530SAl Viro hint &= PAGE_MASK; 116740401530SAl Viro if (((void *)hint != NULL) && 116840401530SAl Viro (hint < mmap_min_addr)) 116940401530SAl Viro return PAGE_ALIGN(mmap_min_addr); 117040401530SAl Viro return hint; 117140401530SAl Viro } 117240401530SAl Viro 11736aeb2542SMike Rapoport int mlock_future_check(struct mm_struct *mm, unsigned long flags, 1174363ee17fSDavidlohr Bueso unsigned long len) 1175363ee17fSDavidlohr Bueso { 1176363ee17fSDavidlohr Bueso unsigned long locked, lock_limit; 1177363ee17fSDavidlohr Bueso 1178363ee17fSDavidlohr Bueso /* mlock MCL_FUTURE? */ 1179363ee17fSDavidlohr Bueso if (flags & VM_LOCKED) { 1180363ee17fSDavidlohr Bueso locked = len >> PAGE_SHIFT; 1181363ee17fSDavidlohr Bueso locked += mm->locked_vm; 1182363ee17fSDavidlohr Bueso lock_limit = rlimit(RLIMIT_MEMLOCK); 1183363ee17fSDavidlohr Bueso lock_limit >>= PAGE_SHIFT; 1184363ee17fSDavidlohr Bueso if (locked > lock_limit && !capable(CAP_IPC_LOCK)) 1185363ee17fSDavidlohr Bueso return -EAGAIN; 1186363ee17fSDavidlohr Bueso } 1187363ee17fSDavidlohr Bueso return 0; 1188363ee17fSDavidlohr Bueso } 1189363ee17fSDavidlohr Bueso 1190be83bbf8SLinus Torvalds static inline u64 file_mmap_size_max(struct file *file, struct inode *inode) 1191be83bbf8SLinus Torvalds { 1192be83bbf8SLinus Torvalds if (S_ISREG(inode->i_mode)) 1193423913adSLinus Torvalds return MAX_LFS_FILESIZE; 1194be83bbf8SLinus Torvalds 1195be83bbf8SLinus Torvalds if (S_ISBLK(inode->i_mode)) 1196be83bbf8SLinus Torvalds return MAX_LFS_FILESIZE; 1197be83bbf8SLinus Torvalds 119876f34950SIvan Khoronzhuk if (S_ISSOCK(inode->i_mode)) 119976f34950SIvan Khoronzhuk return MAX_LFS_FILESIZE; 120076f34950SIvan Khoronzhuk 1201be83bbf8SLinus Torvalds /* Special "we do even unsigned file positions" case */ 1202be83bbf8SLinus Torvalds if (file->f_mode & FMODE_UNSIGNED_OFFSET) 1203be83bbf8SLinus Torvalds return 0; 1204be83bbf8SLinus Torvalds 1205be83bbf8SLinus Torvalds /* Yes, random drivers might want more. But I'm tired of buggy drivers */ 1206be83bbf8SLinus Torvalds return ULONG_MAX; 1207be83bbf8SLinus Torvalds } 1208be83bbf8SLinus Torvalds 1209be83bbf8SLinus Torvalds static inline bool file_mmap_ok(struct file *file, struct inode *inode, 1210be83bbf8SLinus Torvalds unsigned long pgoff, unsigned long len) 1211be83bbf8SLinus Torvalds { 1212be83bbf8SLinus Torvalds u64 maxsize = file_mmap_size_max(file, inode); 1213be83bbf8SLinus Torvalds 1214be83bbf8SLinus Torvalds if (maxsize && len > maxsize) 1215be83bbf8SLinus Torvalds return false; 1216be83bbf8SLinus Torvalds maxsize -= len; 1217be83bbf8SLinus Torvalds if (pgoff > maxsize >> PAGE_SHIFT) 1218be83bbf8SLinus Torvalds return false; 1219be83bbf8SLinus Torvalds return true; 1220be83bbf8SLinus Torvalds } 1221be83bbf8SLinus Torvalds 122240401530SAl Viro /* 12233e4e28c5SMichel Lespinasse * The caller must write-lock current->mm->mmap_lock. 12241da177e4SLinus Torvalds */ 12251fcfd8dbSOleg Nesterov unsigned long do_mmap(struct file *file, unsigned long addr, 12261da177e4SLinus Torvalds unsigned long len, unsigned long prot, 122745e55300SPeter Collingbourne unsigned long flags, unsigned long pgoff, 122845e55300SPeter Collingbourne unsigned long *populate, struct list_head *uf) 12291da177e4SLinus Torvalds { 12301da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 123145e55300SPeter Collingbourne vm_flags_t vm_flags; 123262b5f7d0SDave Hansen int pkey = 0; 12331da177e4SLinus Torvalds 1234524e00b3SLiam R. Howlett validate_mm(mm); 123541badc15SMichel Lespinasse *populate = 0; 1236bebeb3d6SMichel Lespinasse 1237e37609bbSPiotr Kwapulinski if (!len) 1238e37609bbSPiotr Kwapulinski return -EINVAL; 1239e37609bbSPiotr Kwapulinski 12401da177e4SLinus Torvalds /* 12411da177e4SLinus Torvalds * Does the application expect PROT_READ to imply PROT_EXEC? 12421da177e4SLinus Torvalds * 12431da177e4SLinus Torvalds * (the exception is when the underlying filesystem is noexec 12441da177e4SLinus Torvalds * mounted, in which case we dont add PROT_EXEC.) 12451da177e4SLinus Torvalds */ 12461da177e4SLinus Torvalds if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC)) 124790f8572bSEric W. Biederman if (!(file && path_noexec(&file->f_path))) 12481da177e4SLinus Torvalds prot |= PROT_EXEC; 12491da177e4SLinus Torvalds 1250a4ff8e86SMichal Hocko /* force arch specific MAP_FIXED handling in get_unmapped_area */ 1251a4ff8e86SMichal Hocko if (flags & MAP_FIXED_NOREPLACE) 1252a4ff8e86SMichal Hocko flags |= MAP_FIXED; 1253a4ff8e86SMichal Hocko 12547cd94146SEric Paris if (!(flags & MAP_FIXED)) 12557cd94146SEric Paris addr = round_hint_to_min(addr); 12567cd94146SEric Paris 12571da177e4SLinus Torvalds /* Careful about overflows.. */ 12581da177e4SLinus Torvalds len = PAGE_ALIGN(len); 12599206de95SAl Viro if (!len) 12601da177e4SLinus Torvalds return -ENOMEM; 12611da177e4SLinus Torvalds 12621da177e4SLinus Torvalds /* offset overflow? */ 12631da177e4SLinus Torvalds if ((pgoff + (len >> PAGE_SHIFT)) < pgoff) 12641da177e4SLinus Torvalds return -EOVERFLOW; 12651da177e4SLinus Torvalds 12661da177e4SLinus Torvalds /* Too many mappings? */ 12671da177e4SLinus Torvalds if (mm->map_count > sysctl_max_map_count) 12681da177e4SLinus Torvalds return -ENOMEM; 12691da177e4SLinus Torvalds 12701da177e4SLinus Torvalds /* Obtain the address to map to. we verify (or select) it and ensure 12711da177e4SLinus Torvalds * that it represents a valid section of the address space. 12721da177e4SLinus Torvalds */ 12731da177e4SLinus Torvalds addr = get_unmapped_area(file, addr, len, pgoff, flags); 1274ff68dac6SGaowei Pu if (IS_ERR_VALUE(addr)) 12751da177e4SLinus Torvalds return addr; 12761da177e4SLinus Torvalds 1277a4ff8e86SMichal Hocko if (flags & MAP_FIXED_NOREPLACE) { 127835e43c5fSLiam Howlett if (find_vma_intersection(mm, addr, addr + len)) 1279a4ff8e86SMichal Hocko return -EEXIST; 1280a4ff8e86SMichal Hocko } 1281a4ff8e86SMichal Hocko 128262b5f7d0SDave Hansen if (prot == PROT_EXEC) { 128362b5f7d0SDave Hansen pkey = execute_only_pkey(mm); 128462b5f7d0SDave Hansen if (pkey < 0) 128562b5f7d0SDave Hansen pkey = 0; 128662b5f7d0SDave Hansen } 128762b5f7d0SDave Hansen 12881da177e4SLinus Torvalds /* Do simple checking here so the lower-level routines won't have 12891da177e4SLinus Torvalds * to. we assume access permissions have been handled by the open 12901da177e4SLinus Torvalds * of the memory object, so we don't do any here. 12911da177e4SLinus Torvalds */ 129245e55300SPeter Collingbourne vm_flags = calc_vm_prot_bits(prot, pkey) | calc_vm_flag_bits(flags) | 12931da177e4SLinus Torvalds mm->def_flags | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC; 12941da177e4SLinus Torvalds 1295cdf7b341SHuang Shijie if (flags & MAP_LOCKED) 12961da177e4SLinus Torvalds if (!can_do_mlock()) 12971da177e4SLinus Torvalds return -EPERM; 1298ba470de4SRik van Riel 1299363ee17fSDavidlohr Bueso if (mlock_future_check(mm, vm_flags, len)) 13001da177e4SLinus Torvalds return -EAGAIN; 13011da177e4SLinus Torvalds 13021da177e4SLinus Torvalds if (file) { 1303077bf22bSOleg Nesterov struct inode *inode = file_inode(file); 13041c972597SDan Williams unsigned long flags_mask; 13051c972597SDan Williams 1306be83bbf8SLinus Torvalds if (!file_mmap_ok(file, inode, pgoff, len)) 1307be83bbf8SLinus Torvalds return -EOVERFLOW; 1308be83bbf8SLinus Torvalds 13091c972597SDan Williams flags_mask = LEGACY_MAP_MASK | file->f_op->mmap_supported_flags; 1310077bf22bSOleg Nesterov 13111da177e4SLinus Torvalds switch (flags & MAP_TYPE) { 13121da177e4SLinus Torvalds case MAP_SHARED: 13131c972597SDan Williams /* 13141c972597SDan Williams * Force use of MAP_SHARED_VALIDATE with non-legacy 13151c972597SDan Williams * flags. E.g. MAP_SYNC is dangerous to use with 13161c972597SDan Williams * MAP_SHARED as you don't know which consistency model 13171c972597SDan Williams * you will get. We silently ignore unsupported flags 13181c972597SDan Williams * with MAP_SHARED to preserve backward compatibility. 13191c972597SDan Williams */ 13201c972597SDan Williams flags &= LEGACY_MAP_MASK; 1321e4a9bc58SJoe Perches fallthrough; 13221c972597SDan Williams case MAP_SHARED_VALIDATE: 13231c972597SDan Williams if (flags & ~flags_mask) 13241c972597SDan Williams return -EOPNOTSUPP; 1325dc617f29SDarrick J. Wong if (prot & PROT_WRITE) { 1326dc617f29SDarrick J. Wong if (!(file->f_mode & FMODE_WRITE)) 13271da177e4SLinus Torvalds return -EACCES; 1328dc617f29SDarrick J. Wong if (IS_SWAPFILE(file->f_mapping->host)) 1329dc617f29SDarrick J. Wong return -ETXTBSY; 1330dc617f29SDarrick J. Wong } 13311da177e4SLinus Torvalds 13321da177e4SLinus Torvalds /* 13331da177e4SLinus Torvalds * Make sure we don't allow writing to an append-only 13341da177e4SLinus Torvalds * file.. 13351da177e4SLinus Torvalds */ 13361da177e4SLinus Torvalds if (IS_APPEND(inode) && (file->f_mode & FMODE_WRITE)) 13371da177e4SLinus Torvalds return -EACCES; 13381da177e4SLinus Torvalds 13391da177e4SLinus Torvalds vm_flags |= VM_SHARED | VM_MAYSHARE; 13401da177e4SLinus Torvalds if (!(file->f_mode & FMODE_WRITE)) 13411da177e4SLinus Torvalds vm_flags &= ~(VM_MAYWRITE | VM_SHARED); 1342e4a9bc58SJoe Perches fallthrough; 13431da177e4SLinus Torvalds case MAP_PRIVATE: 13441da177e4SLinus Torvalds if (!(file->f_mode & FMODE_READ)) 13451da177e4SLinus Torvalds return -EACCES; 134690f8572bSEric W. Biederman if (path_noexec(&file->f_path)) { 134780c5606cSLinus Torvalds if (vm_flags & VM_EXEC) 134880c5606cSLinus Torvalds return -EPERM; 134980c5606cSLinus Torvalds vm_flags &= ~VM_MAYEXEC; 135080c5606cSLinus Torvalds } 135180c5606cSLinus Torvalds 135272c2d531SAl Viro if (!file->f_op->mmap) 135380c5606cSLinus Torvalds return -ENODEV; 1354b2c56e4fSOleg Nesterov if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP)) 1355b2c56e4fSOleg Nesterov return -EINVAL; 13561da177e4SLinus Torvalds break; 13571da177e4SLinus Torvalds 13581da177e4SLinus Torvalds default: 13591da177e4SLinus Torvalds return -EINVAL; 13601da177e4SLinus Torvalds } 13611da177e4SLinus Torvalds } else { 13621da177e4SLinus Torvalds switch (flags & MAP_TYPE) { 13631da177e4SLinus Torvalds case MAP_SHARED: 1364b2c56e4fSOleg Nesterov if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP)) 1365b2c56e4fSOleg Nesterov return -EINVAL; 1366ce363942STejun Heo /* 1367ce363942STejun Heo * Ignore pgoff. 1368ce363942STejun Heo */ 1369ce363942STejun Heo pgoff = 0; 13701da177e4SLinus Torvalds vm_flags |= VM_SHARED | VM_MAYSHARE; 13711da177e4SLinus Torvalds break; 13721da177e4SLinus Torvalds case MAP_PRIVATE: 13731da177e4SLinus Torvalds /* 13741da177e4SLinus Torvalds * Set pgoff according to addr for anon_vma. 13751da177e4SLinus Torvalds */ 13761da177e4SLinus Torvalds pgoff = addr >> PAGE_SHIFT; 13771da177e4SLinus Torvalds break; 13781da177e4SLinus Torvalds default: 13791da177e4SLinus Torvalds return -EINVAL; 13801da177e4SLinus Torvalds } 13811da177e4SLinus Torvalds } 13821da177e4SLinus Torvalds 1383c22c0d63SMichel Lespinasse /* 1384c22c0d63SMichel Lespinasse * Set 'VM_NORESERVE' if we should not account for the 1385c22c0d63SMichel Lespinasse * memory use of this mapping. 1386c22c0d63SMichel Lespinasse */ 1387c22c0d63SMichel Lespinasse if (flags & MAP_NORESERVE) { 1388c22c0d63SMichel Lespinasse /* We honor MAP_NORESERVE if allowed to overcommit */ 1389c22c0d63SMichel Lespinasse if (sysctl_overcommit_memory != OVERCOMMIT_NEVER) 1390c22c0d63SMichel Lespinasse vm_flags |= VM_NORESERVE; 1391c22c0d63SMichel Lespinasse 1392c22c0d63SMichel Lespinasse /* hugetlb applies strict overcommit unless MAP_NORESERVE */ 1393c22c0d63SMichel Lespinasse if (file && is_file_hugepages(file)) 1394c22c0d63SMichel Lespinasse vm_flags |= VM_NORESERVE; 1395c22c0d63SMichel Lespinasse } 1396c22c0d63SMichel Lespinasse 1397897ab3e0SMike Rapoport addr = mmap_region(file, addr, len, vm_flags, pgoff, uf); 139809a9f1d2SMichel Lespinasse if (!IS_ERR_VALUE(addr) && 139909a9f1d2SMichel Lespinasse ((vm_flags & VM_LOCKED) || 140009a9f1d2SMichel Lespinasse (flags & (MAP_POPULATE | MAP_NONBLOCK)) == MAP_POPULATE)) 140141badc15SMichel Lespinasse *populate = len; 1402bebeb3d6SMichel Lespinasse return addr; 14030165ab44SMiklos Szeredi } 14046be5ceb0SLinus Torvalds 1405a90f590aSDominik Brodowski unsigned long ksys_mmap_pgoff(unsigned long addr, unsigned long len, 1406a90f590aSDominik Brodowski unsigned long prot, unsigned long flags, 1407a90f590aSDominik Brodowski unsigned long fd, unsigned long pgoff) 140866f0dc48SHugh Dickins { 140966f0dc48SHugh Dickins struct file *file = NULL; 14101e3ee14bSChen Gang unsigned long retval; 141166f0dc48SHugh Dickins 141266f0dc48SHugh Dickins if (!(flags & MAP_ANONYMOUS)) { 1413120a795dSAl Viro audit_mmap_fd(fd, flags); 141466f0dc48SHugh Dickins file = fget(fd); 141566f0dc48SHugh Dickins if (!file) 14161e3ee14bSChen Gang return -EBADF; 14177bba8f0eSZhen Lei if (is_file_hugepages(file)) { 1418af73e4d9SNaoya Horiguchi len = ALIGN(len, huge_page_size(hstate_file(file))); 14197bba8f0eSZhen Lei } else if (unlikely(flags & MAP_HUGETLB)) { 1420493af578SJörn Engel retval = -EINVAL; 1421493af578SJörn Engel goto out_fput; 14227bba8f0eSZhen Lei } 142366f0dc48SHugh Dickins } else if (flags & MAP_HUGETLB) { 1424c103a4dcSAndrew Morton struct hstate *hs; 1425af73e4d9SNaoya Horiguchi 142620ac2893SAnshuman Khandual hs = hstate_sizelog((flags >> MAP_HUGE_SHIFT) & MAP_HUGE_MASK); 1427091d0d55SLi Zefan if (!hs) 1428091d0d55SLi Zefan return -EINVAL; 1429091d0d55SLi Zefan 1430091d0d55SLi Zefan len = ALIGN(len, huge_page_size(hs)); 143166f0dc48SHugh Dickins /* 143266f0dc48SHugh Dickins * VM_NORESERVE is used because the reservations will be 143366f0dc48SHugh Dickins * taken when vm_ops->mmap() is called 143466f0dc48SHugh Dickins */ 1435af73e4d9SNaoya Horiguchi file = hugetlb_file_setup(HUGETLB_ANON_FILE, len, 143642d7395fSAndi Kleen VM_NORESERVE, 143783c1fd76Szhangyiru HUGETLB_ANONHUGE_INODE, 143842d7395fSAndi Kleen (flags >> MAP_HUGE_SHIFT) & MAP_HUGE_MASK); 143966f0dc48SHugh Dickins if (IS_ERR(file)) 144066f0dc48SHugh Dickins return PTR_ERR(file); 144166f0dc48SHugh Dickins } 144266f0dc48SHugh Dickins 14439fbeb5abSMichal Hocko retval = vm_mmap_pgoff(file, addr, len, prot, flags, pgoff); 1444493af578SJörn Engel out_fput: 144566f0dc48SHugh Dickins if (file) 144666f0dc48SHugh Dickins fput(file); 144766f0dc48SHugh Dickins return retval; 144866f0dc48SHugh Dickins } 144966f0dc48SHugh Dickins 1450a90f590aSDominik Brodowski SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len, 1451a90f590aSDominik Brodowski unsigned long, prot, unsigned long, flags, 1452a90f590aSDominik Brodowski unsigned long, fd, unsigned long, pgoff) 1453a90f590aSDominik Brodowski { 1454a90f590aSDominik Brodowski return ksys_mmap_pgoff(addr, len, prot, flags, fd, pgoff); 1455a90f590aSDominik Brodowski } 1456a90f590aSDominik Brodowski 1457a4679373SChristoph Hellwig #ifdef __ARCH_WANT_SYS_OLD_MMAP 1458a4679373SChristoph Hellwig struct mmap_arg_struct { 1459a4679373SChristoph Hellwig unsigned long addr; 1460a4679373SChristoph Hellwig unsigned long len; 1461a4679373SChristoph Hellwig unsigned long prot; 1462a4679373SChristoph Hellwig unsigned long flags; 1463a4679373SChristoph Hellwig unsigned long fd; 1464a4679373SChristoph Hellwig unsigned long offset; 1465a4679373SChristoph Hellwig }; 1466a4679373SChristoph Hellwig 1467a4679373SChristoph Hellwig SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg) 1468a4679373SChristoph Hellwig { 1469a4679373SChristoph Hellwig struct mmap_arg_struct a; 1470a4679373SChristoph Hellwig 1471a4679373SChristoph Hellwig if (copy_from_user(&a, arg, sizeof(a))) 1472a4679373SChristoph Hellwig return -EFAULT; 1473de1741a1SAlexander Kuleshov if (offset_in_page(a.offset)) 1474a4679373SChristoph Hellwig return -EINVAL; 1475a4679373SChristoph Hellwig 1476a90f590aSDominik Brodowski return ksys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd, 1477a4679373SChristoph Hellwig a.offset >> PAGE_SHIFT); 1478a4679373SChristoph Hellwig } 1479a4679373SChristoph Hellwig #endif /* __ARCH_WANT_SYS_OLD_MMAP */ 1480a4679373SChristoph Hellwig 14814e950f6fSAlexey Dobriyan /* 14828bb4e7a2SWei Yang * Some shared mappings will want the pages marked read-only 14834e950f6fSAlexey Dobriyan * to track write events. If so, we'll downgrade vm_page_prot 14844e950f6fSAlexey Dobriyan * to the private version (using protection_map[] without the 14854e950f6fSAlexey Dobriyan * VM_SHARED bit). 14864e950f6fSAlexey Dobriyan */ 14876d2329f8SAndrea Arcangeli int vma_wants_writenotify(struct vm_area_struct *vma, pgprot_t vm_page_prot) 14884e950f6fSAlexey Dobriyan { 1489ca16d140SKOSAKI Motohiro vm_flags_t vm_flags = vma->vm_flags; 14908a04446aSKirill A. Shutemov const struct vm_operations_struct *vm_ops = vma->vm_ops; 14914e950f6fSAlexey Dobriyan 14924e950f6fSAlexey Dobriyan /* If it was private or non-writable, the write bit is already clear */ 14934e950f6fSAlexey Dobriyan if ((vm_flags & (VM_WRITE|VM_SHARED)) != ((VM_WRITE|VM_SHARED))) 14944e950f6fSAlexey Dobriyan return 0; 14954e950f6fSAlexey Dobriyan 14964e950f6fSAlexey Dobriyan /* The backer wishes to know when pages are first written to? */ 14978a04446aSKirill A. Shutemov if (vm_ops && (vm_ops->page_mkwrite || vm_ops->pfn_mkwrite)) 14984e950f6fSAlexey Dobriyan return 1; 14994e950f6fSAlexey Dobriyan 150064e45507SPeter Feiner /* The open routine did something to the protections that pgprot_modify 150164e45507SPeter Feiner * won't preserve? */ 15026d2329f8SAndrea Arcangeli if (pgprot_val(vm_page_prot) != 15036d2329f8SAndrea Arcangeli pgprot_val(vm_pgprot_modify(vm_page_prot, vm_flags))) 15044e950f6fSAlexey Dobriyan return 0; 15054e950f6fSAlexey Dobriyan 1506f96f7a40SDavid Hildenbrand /* 1507f96f7a40SDavid Hildenbrand * Do we need to track softdirty? hugetlb does not support softdirty 1508f96f7a40SDavid Hildenbrand * tracking yet. 1509f96f7a40SDavid Hildenbrand */ 1510f96f7a40SDavid Hildenbrand if (vma_soft_dirty_enabled(vma) && !is_vm_hugetlb_page(vma)) 151164e45507SPeter Feiner return 1; 151264e45507SPeter Feiner 151351d3d5ebSDavid Hildenbrand /* Do we need write faults for uffd-wp tracking? */ 151451d3d5ebSDavid Hildenbrand if (userfaultfd_wp(vma)) 151551d3d5ebSDavid Hildenbrand return 1; 151651d3d5ebSDavid Hildenbrand 15174e950f6fSAlexey Dobriyan /* Specialty mapping? */ 15184b6e1e37SKonstantin Khlebnikov if (vm_flags & VM_PFNMAP) 15194e950f6fSAlexey Dobriyan return 0; 15204e950f6fSAlexey Dobriyan 15214e950f6fSAlexey Dobriyan /* Can the mapping track the dirty pages? */ 15224e950f6fSAlexey Dobriyan return vma->vm_file && vma->vm_file->f_mapping && 1523f56753acSChristoph Hellwig mapping_can_writeback(vma->vm_file->f_mapping); 15244e950f6fSAlexey Dobriyan } 15254e950f6fSAlexey Dobriyan 1526fc8744adSLinus Torvalds /* 1527fc8744adSLinus Torvalds * We account for memory if it's a private writeable mapping, 15285a6fe125SMel Gorman * not hugepages and VM_NORESERVE wasn't set. 1529fc8744adSLinus Torvalds */ 1530ca16d140SKOSAKI Motohiro static inline int accountable_mapping(struct file *file, vm_flags_t vm_flags) 1531fc8744adSLinus Torvalds { 15325a6fe125SMel Gorman /* 15335a6fe125SMel Gorman * hugetlb has its own accounting separate from the core VM 15345a6fe125SMel Gorman * VM_HUGETLB may not be set yet so we cannot check for that flag. 15355a6fe125SMel Gorman */ 15365a6fe125SMel Gorman if (file && is_file_hugepages(file)) 15375a6fe125SMel Gorman return 0; 15385a6fe125SMel Gorman 1539fc8744adSLinus Torvalds return (vm_flags & (VM_NORESERVE | VM_SHARED | VM_WRITE)) == VM_WRITE; 1540fc8744adSLinus Torvalds } 1541fc8744adSLinus Torvalds 15423499a131SLiam R. Howlett /** 15433499a131SLiam R. Howlett * unmapped_area() - Find an area between the low_limit and the high_limit with 15443499a131SLiam R. Howlett * the correct alignment and offset, all from @info. Note: current->mm is used 15453499a131SLiam R. Howlett * for the search. 15463499a131SLiam R. Howlett * 154782b24936SVernon Yang * @info: The unmapped area information including the range [low_limit - 154882b24936SVernon Yang * high_limit), the alignment offset and mask. 15493499a131SLiam R. Howlett * 15503499a131SLiam R. Howlett * Return: A memory address or -ENOMEM. 15513499a131SLiam R. Howlett */ 1552baceaf1cSJaewon Kim static unsigned long unmapped_area(struct vm_unmapped_area_info *info) 1553db4fbfb9SMichel Lespinasse { 15543499a131SLiam R. Howlett unsigned long length, gap; 1555db4fbfb9SMichel Lespinasse 15563499a131SLiam R. Howlett MA_STATE(mas, ¤t->mm->mm_mt, 0, 0); 1557db4fbfb9SMichel Lespinasse 1558db4fbfb9SMichel Lespinasse /* Adjust search length to account for worst case alignment overhead */ 1559db4fbfb9SMichel Lespinasse length = info->length + info->align_mask; 1560db4fbfb9SMichel Lespinasse if (length < info->length) 1561db4fbfb9SMichel Lespinasse return -ENOMEM; 1562db4fbfb9SMichel Lespinasse 15633499a131SLiam R. Howlett if (mas_empty_area(&mas, info->low_limit, info->high_limit - 1, 15643499a131SLiam R. Howlett length)) 15653499a131SLiam R. Howlett return -ENOMEM; 15663499a131SLiam R. Howlett 1567d4af56c5SLiam R. Howlett gap = mas.index; 1568d4af56c5SLiam R. Howlett gap += (info->align_offset - gap) & info->align_mask; 15693499a131SLiam R. Howlett return gap; 1570db4fbfb9SMichel Lespinasse } 1571db4fbfb9SMichel Lespinasse 15723499a131SLiam R. Howlett /** 15733499a131SLiam R. Howlett * unmapped_area_topdown() - Find an area between the low_limit and the 157482b24936SVernon Yang * high_limit with the correct alignment and offset at the highest available 15753499a131SLiam R. Howlett * address, all from @info. Note: current->mm is used for the search. 15763499a131SLiam R. Howlett * 157782b24936SVernon Yang * @info: The unmapped area information including the range [low_limit - 157882b24936SVernon Yang * high_limit), the alignment offset and mask. 15793499a131SLiam R. Howlett * 15803499a131SLiam R. Howlett * Return: A memory address or -ENOMEM. 15813499a131SLiam R. Howlett */ 1582baceaf1cSJaewon Kim static unsigned long unmapped_area_topdown(struct vm_unmapped_area_info *info) 1583db4fbfb9SMichel Lespinasse { 15843499a131SLiam R. Howlett unsigned long length, gap; 1585d4af56c5SLiam R. Howlett 15863499a131SLiam R. Howlett MA_STATE(mas, ¤t->mm->mm_mt, 0, 0); 1587db4fbfb9SMichel Lespinasse /* Adjust search length to account for worst case alignment overhead */ 1588db4fbfb9SMichel Lespinasse length = info->length + info->align_mask; 1589db4fbfb9SMichel Lespinasse if (length < info->length) 1590db4fbfb9SMichel Lespinasse return -ENOMEM; 1591db4fbfb9SMichel Lespinasse 15923499a131SLiam R. Howlett if (mas_empty_area_rev(&mas, info->low_limit, info->high_limit - 1, 15933499a131SLiam R. Howlett length)) 15943499a131SLiam R. Howlett return -ENOMEM; 15953499a131SLiam R. Howlett 1596d4af56c5SLiam R. Howlett gap = mas.last + 1 - info->length; 1597d4af56c5SLiam R. Howlett gap -= (gap - info->align_offset) & info->align_mask; 15983499a131SLiam R. Howlett return gap; 1599db4fbfb9SMichel Lespinasse } 1600db4fbfb9SMichel Lespinasse 1601baceaf1cSJaewon Kim /* 1602baceaf1cSJaewon Kim * Search for an unmapped address range. 1603baceaf1cSJaewon Kim * 1604baceaf1cSJaewon Kim * We are looking for a range that: 1605baceaf1cSJaewon Kim * - does not intersect with any VMA; 1606baceaf1cSJaewon Kim * - is contained within the [low_limit, high_limit) interval; 1607baceaf1cSJaewon Kim * - is at least the desired size. 1608baceaf1cSJaewon Kim * - satisfies (begin_addr & align_mask) == (align_offset & align_mask) 1609baceaf1cSJaewon Kim */ 1610baceaf1cSJaewon Kim unsigned long vm_unmapped_area(struct vm_unmapped_area_info *info) 1611baceaf1cSJaewon Kim { 1612df529cabSJaewon Kim unsigned long addr; 1613df529cabSJaewon Kim 1614baceaf1cSJaewon Kim if (info->flags & VM_UNMAPPED_AREA_TOPDOWN) 1615df529cabSJaewon Kim addr = unmapped_area_topdown(info); 1616baceaf1cSJaewon Kim else 1617df529cabSJaewon Kim addr = unmapped_area(info); 1618df529cabSJaewon Kim 1619df529cabSJaewon Kim trace_vm_unmapped_area(addr, info); 1620df529cabSJaewon Kim return addr; 1621baceaf1cSJaewon Kim } 1622f6795053SSteve Capper 16231da177e4SLinus Torvalds /* Get an address range which is currently unmapped. 16241da177e4SLinus Torvalds * For shmat() with addr=0. 16251da177e4SLinus Torvalds * 16261da177e4SLinus Torvalds * Ugly calling convention alert: 16271da177e4SLinus Torvalds * Return value with the low bits set means error value, 16281da177e4SLinus Torvalds * ie 16291da177e4SLinus Torvalds * if (ret & ~PAGE_MASK) 16301da177e4SLinus Torvalds * error = ret; 16311da177e4SLinus Torvalds * 16321da177e4SLinus Torvalds * This function "knows" that -ENOMEM has the bits set. 16331da177e4SLinus Torvalds */ 16341da177e4SLinus Torvalds unsigned long 16354b439e25SChristophe Leroy generic_get_unmapped_area(struct file *filp, unsigned long addr, 16364b439e25SChristophe Leroy unsigned long len, unsigned long pgoff, 16374b439e25SChristophe Leroy unsigned long flags) 16381da177e4SLinus Torvalds { 16391da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 16401be7107fSHugh Dickins struct vm_area_struct *vma, *prev; 1641db4fbfb9SMichel Lespinasse struct vm_unmapped_area_info info; 16422cb4de08SChristophe Leroy const unsigned long mmap_end = arch_get_mmap_end(addr, len, flags); 16431da177e4SLinus Torvalds 1644f6795053SSteve Capper if (len > mmap_end - mmap_min_addr) 16451da177e4SLinus Torvalds return -ENOMEM; 16461da177e4SLinus Torvalds 164706abdfb4SBenjamin Herrenschmidt if (flags & MAP_FIXED) 164806abdfb4SBenjamin Herrenschmidt return addr; 164906abdfb4SBenjamin Herrenschmidt 16501da177e4SLinus Torvalds if (addr) { 16511da177e4SLinus Torvalds addr = PAGE_ALIGN(addr); 16521be7107fSHugh Dickins vma = find_vma_prev(mm, addr, &prev); 1653f6795053SSteve Capper if (mmap_end - len >= addr && addr >= mmap_min_addr && 16541be7107fSHugh Dickins (!vma || addr + len <= vm_start_gap(vma)) && 16551be7107fSHugh Dickins (!prev || addr >= vm_end_gap(prev))) 16561da177e4SLinus Torvalds return addr; 16571da177e4SLinus Torvalds } 16581da177e4SLinus Torvalds 1659db4fbfb9SMichel Lespinasse info.flags = 0; 1660db4fbfb9SMichel Lespinasse info.length = len; 16614e99b021SHeiko Carstens info.low_limit = mm->mmap_base; 1662f6795053SSteve Capper info.high_limit = mmap_end; 1663db4fbfb9SMichel Lespinasse info.align_mask = 0; 166409ef5283SJaewon Kim info.align_offset = 0; 1665db4fbfb9SMichel Lespinasse return vm_unmapped_area(&info); 16661da177e4SLinus Torvalds } 16674b439e25SChristophe Leroy 16684b439e25SChristophe Leroy #ifndef HAVE_ARCH_UNMAPPED_AREA 16694b439e25SChristophe Leroy unsigned long 16704b439e25SChristophe Leroy arch_get_unmapped_area(struct file *filp, unsigned long addr, 16714b439e25SChristophe Leroy unsigned long len, unsigned long pgoff, 16724b439e25SChristophe Leroy unsigned long flags) 16734b439e25SChristophe Leroy { 16744b439e25SChristophe Leroy return generic_get_unmapped_area(filp, addr, len, pgoff, flags); 16754b439e25SChristophe Leroy } 16761da177e4SLinus Torvalds #endif 16771da177e4SLinus Torvalds 16781da177e4SLinus Torvalds /* 16791da177e4SLinus Torvalds * This mmap-allocator allocates new areas top-down from below the 16801da177e4SLinus Torvalds * stack's low limit (the base): 16811da177e4SLinus Torvalds */ 16821da177e4SLinus Torvalds unsigned long 16834b439e25SChristophe Leroy generic_get_unmapped_area_topdown(struct file *filp, unsigned long addr, 168443cca0b1SYang Fan unsigned long len, unsigned long pgoff, 168543cca0b1SYang Fan unsigned long flags) 16861da177e4SLinus Torvalds { 16871be7107fSHugh Dickins struct vm_area_struct *vma, *prev; 16881da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 1689db4fbfb9SMichel Lespinasse struct vm_unmapped_area_info info; 16902cb4de08SChristophe Leroy const unsigned long mmap_end = arch_get_mmap_end(addr, len, flags); 16911da177e4SLinus Torvalds 16921da177e4SLinus Torvalds /* requested length too big for entire address space */ 1693f6795053SSteve Capper if (len > mmap_end - mmap_min_addr) 16941da177e4SLinus Torvalds return -ENOMEM; 16951da177e4SLinus Torvalds 169606abdfb4SBenjamin Herrenschmidt if (flags & MAP_FIXED) 169706abdfb4SBenjamin Herrenschmidt return addr; 169806abdfb4SBenjamin Herrenschmidt 16991da177e4SLinus Torvalds /* requesting a specific address */ 17001da177e4SLinus Torvalds if (addr) { 17011da177e4SLinus Torvalds addr = PAGE_ALIGN(addr); 17021be7107fSHugh Dickins vma = find_vma_prev(mm, addr, &prev); 1703f6795053SSteve Capper if (mmap_end - len >= addr && addr >= mmap_min_addr && 17041be7107fSHugh Dickins (!vma || addr + len <= vm_start_gap(vma)) && 17051be7107fSHugh Dickins (!prev || addr >= vm_end_gap(prev))) 17061da177e4SLinus Torvalds return addr; 17071da177e4SLinus Torvalds } 17081da177e4SLinus Torvalds 1709db4fbfb9SMichel Lespinasse info.flags = VM_UNMAPPED_AREA_TOPDOWN; 1710db4fbfb9SMichel Lespinasse info.length = len; 17112afc745fSAkira Takeuchi info.low_limit = max(PAGE_SIZE, mmap_min_addr); 1712f6795053SSteve Capper info.high_limit = arch_get_mmap_base(addr, mm->mmap_base); 1713db4fbfb9SMichel Lespinasse info.align_mask = 0; 171409ef5283SJaewon Kim info.align_offset = 0; 1715db4fbfb9SMichel Lespinasse addr = vm_unmapped_area(&info); 1716b716ad95SXiao Guangrong 17171da177e4SLinus Torvalds /* 17181da177e4SLinus Torvalds * A failed mmap() very likely causes application failure, 17191da177e4SLinus Torvalds * so fall back to the bottom-up function here. This scenario 17201da177e4SLinus Torvalds * can happen with large stack limits and large mmap() 17211da177e4SLinus Torvalds * allocations. 17221da177e4SLinus Torvalds */ 1723de1741a1SAlexander Kuleshov if (offset_in_page(addr)) { 1724db4fbfb9SMichel Lespinasse VM_BUG_ON(addr != -ENOMEM); 1725db4fbfb9SMichel Lespinasse info.flags = 0; 1726db4fbfb9SMichel Lespinasse info.low_limit = TASK_UNMAPPED_BASE; 1727f6795053SSteve Capper info.high_limit = mmap_end; 1728db4fbfb9SMichel Lespinasse addr = vm_unmapped_area(&info); 1729db4fbfb9SMichel Lespinasse } 17301da177e4SLinus Torvalds 17311da177e4SLinus Torvalds return addr; 17321da177e4SLinus Torvalds } 17334b439e25SChristophe Leroy 17344b439e25SChristophe Leroy #ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN 17354b439e25SChristophe Leroy unsigned long 17364b439e25SChristophe Leroy arch_get_unmapped_area_topdown(struct file *filp, unsigned long addr, 17374b439e25SChristophe Leroy unsigned long len, unsigned long pgoff, 17384b439e25SChristophe Leroy unsigned long flags) 17394b439e25SChristophe Leroy { 17404b439e25SChristophe Leroy return generic_get_unmapped_area_topdown(filp, addr, len, pgoff, flags); 17414b439e25SChristophe Leroy } 17421da177e4SLinus Torvalds #endif 17431da177e4SLinus Torvalds 17441da177e4SLinus Torvalds unsigned long 17451da177e4SLinus Torvalds get_unmapped_area(struct file *file, unsigned long addr, unsigned long len, 17461da177e4SLinus Torvalds unsigned long pgoff, unsigned long flags) 17471da177e4SLinus Torvalds { 174806abdfb4SBenjamin Herrenschmidt unsigned long (*get_area)(struct file *, unsigned long, 174906abdfb4SBenjamin Herrenschmidt unsigned long, unsigned long, unsigned long); 175007ab67c8SLinus Torvalds 17519206de95SAl Viro unsigned long error = arch_mmap_check(addr, len, flags); 17529206de95SAl Viro if (error) 17539206de95SAl Viro return error; 17549206de95SAl Viro 17559206de95SAl Viro /* Careful about overflows.. */ 17569206de95SAl Viro if (len > TASK_SIZE) 17579206de95SAl Viro return -ENOMEM; 17589206de95SAl Viro 175907ab67c8SLinus Torvalds get_area = current->mm->get_unmapped_area; 1760c01d5b30SHugh Dickins if (file) { 1761c01d5b30SHugh Dickins if (file->f_op->get_unmapped_area) 176207ab67c8SLinus Torvalds get_area = file->f_op->get_unmapped_area; 1763c01d5b30SHugh Dickins } else if (flags & MAP_SHARED) { 1764c01d5b30SHugh Dickins /* 1765c01d5b30SHugh Dickins * mmap_region() will call shmem_zero_setup() to create a file, 1766c01d5b30SHugh Dickins * so use shmem's get_unmapped_area in case it can be huge. 176745e55300SPeter Collingbourne * do_mmap() will clear pgoff, so match alignment. 1768c01d5b30SHugh Dickins */ 1769c01d5b30SHugh Dickins pgoff = 0; 1770c01d5b30SHugh Dickins get_area = shmem_get_unmapped_area; 1771c01d5b30SHugh Dickins } 1772c01d5b30SHugh Dickins 177307ab67c8SLinus Torvalds addr = get_area(file, addr, len, pgoff, flags); 177407ab67c8SLinus Torvalds if (IS_ERR_VALUE(addr)) 177507ab67c8SLinus Torvalds return addr; 177607ab67c8SLinus Torvalds 17771da177e4SLinus Torvalds if (addr > TASK_SIZE - len) 17781da177e4SLinus Torvalds return -ENOMEM; 1779de1741a1SAlexander Kuleshov if (offset_in_page(addr)) 17801da177e4SLinus Torvalds return -EINVAL; 178106abdfb4SBenjamin Herrenschmidt 17829ac4ed4bSAl Viro error = security_mmap_addr(addr); 17839ac4ed4bSAl Viro return error ? error : addr; 17841da177e4SLinus Torvalds } 17851da177e4SLinus Torvalds 17861da177e4SLinus Torvalds EXPORT_SYMBOL(get_unmapped_area); 17871da177e4SLinus Torvalds 1788be8432e7SLiam R. Howlett /** 1789abdba2ddSLiam R. Howlett * find_vma_intersection() - Look up the first VMA which intersects the interval 1790abdba2ddSLiam R. Howlett * @mm: The process address space. 1791abdba2ddSLiam R. Howlett * @start_addr: The inclusive start user address. 1792abdba2ddSLiam R. Howlett * @end_addr: The exclusive end user address. 1793abdba2ddSLiam R. Howlett * 1794abdba2ddSLiam R. Howlett * Returns: The first VMA within the provided range, %NULL otherwise. Assumes 1795abdba2ddSLiam R. Howlett * start_addr < end_addr. 1796abdba2ddSLiam R. Howlett */ 1797abdba2ddSLiam R. Howlett struct vm_area_struct *find_vma_intersection(struct mm_struct *mm, 1798abdba2ddSLiam R. Howlett unsigned long start_addr, 1799abdba2ddSLiam R. Howlett unsigned long end_addr) 1800abdba2ddSLiam R. Howlett { 1801abdba2ddSLiam R. Howlett unsigned long index = start_addr; 1802abdba2ddSLiam R. Howlett 1803abdba2ddSLiam R. Howlett mmap_assert_locked(mm); 18047964cf8cSLiam R. Howlett return mt_find(&mm->mm_mt, &index, end_addr - 1); 1805abdba2ddSLiam R. Howlett } 1806abdba2ddSLiam R. Howlett EXPORT_SYMBOL(find_vma_intersection); 1807abdba2ddSLiam R. Howlett 1808abdba2ddSLiam R. Howlett /** 1809be8432e7SLiam R. Howlett * find_vma() - Find the VMA for a given address, or the next VMA. 1810be8432e7SLiam R. Howlett * @mm: The mm_struct to check 1811be8432e7SLiam R. Howlett * @addr: The address 1812be8432e7SLiam R. Howlett * 1813be8432e7SLiam R. Howlett * Returns: The VMA associated with addr, or the next VMA. 1814be8432e7SLiam R. Howlett * May return %NULL in the case of no VMA at addr or above. 1815be8432e7SLiam R. Howlett */ 18161da177e4SLinus Torvalds struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr) 18171da177e4SLinus Torvalds { 1818be8432e7SLiam R. Howlett unsigned long index = addr; 18191da177e4SLinus Torvalds 18205b78ed24SLuigi Rizzo mmap_assert_locked(mm); 18217964cf8cSLiam R. Howlett return mt_find(&mm->mm_mt, &index, ULONG_MAX); 18221da177e4SLinus Torvalds } 18231da177e4SLinus Torvalds EXPORT_SYMBOL(find_vma); 18241da177e4SLinus Torvalds 18257fdbd37dSLiam R. Howlett /** 18267fdbd37dSLiam R. Howlett * find_vma_prev() - Find the VMA for a given address, or the next vma and 18277fdbd37dSLiam R. Howlett * set %pprev to the previous VMA, if any. 18287fdbd37dSLiam R. Howlett * @mm: The mm_struct to check 18297fdbd37dSLiam R. Howlett * @addr: The address 18307fdbd37dSLiam R. Howlett * @pprev: The pointer to set to the previous VMA 18317fdbd37dSLiam R. Howlett * 18327fdbd37dSLiam R. Howlett * Note that RCU lock is missing here since the external mmap_lock() is used 18337fdbd37dSLiam R. Howlett * instead. 18347fdbd37dSLiam R. Howlett * 18357fdbd37dSLiam R. Howlett * Returns: The VMA associated with @addr, or the next vma. 18367fdbd37dSLiam R. Howlett * May return %NULL in the case of no vma at addr or above. 18376bd4837dSKOSAKI Motohiro */ 18381da177e4SLinus Torvalds struct vm_area_struct * 18391da177e4SLinus Torvalds find_vma_prev(struct mm_struct *mm, unsigned long addr, 18401da177e4SLinus Torvalds struct vm_area_struct **pprev) 18411da177e4SLinus Torvalds { 18426bd4837dSKOSAKI Motohiro struct vm_area_struct *vma; 18437fdbd37dSLiam R. Howlett MA_STATE(mas, &mm->mm_mt, addr, addr); 18441da177e4SLinus Torvalds 18457fdbd37dSLiam R. Howlett vma = mas_walk(&mas); 18467fdbd37dSLiam R. Howlett *pprev = mas_prev(&mas, 0); 18477fdbd37dSLiam R. Howlett if (!vma) 18487fdbd37dSLiam R. Howlett vma = mas_next(&mas, ULONG_MAX); 18496bd4837dSKOSAKI Motohiro return vma; 18501da177e4SLinus Torvalds } 18511da177e4SLinus Torvalds 18521da177e4SLinus Torvalds /* 18531da177e4SLinus Torvalds * Verify that the stack growth is acceptable and 18541da177e4SLinus Torvalds * update accounting. This is shared with both the 18551da177e4SLinus Torvalds * grow-up and grow-down cases. 18561da177e4SLinus Torvalds */ 18571be7107fSHugh Dickins static int acct_stack_growth(struct vm_area_struct *vma, 18581be7107fSHugh Dickins unsigned long size, unsigned long grow) 18591da177e4SLinus Torvalds { 18601da177e4SLinus Torvalds struct mm_struct *mm = vma->vm_mm; 18611be7107fSHugh Dickins unsigned long new_start; 18621da177e4SLinus Torvalds 18631da177e4SLinus Torvalds /* address space limit tests */ 186484638335SKonstantin Khlebnikov if (!may_expand_vm(mm, vma->vm_flags, grow)) 18651da177e4SLinus Torvalds return -ENOMEM; 18661da177e4SLinus Torvalds 18671da177e4SLinus Torvalds /* Stack limit test */ 186824c79d8eSKrzysztof Opasiak if (size > rlimit(RLIMIT_STACK)) 18691da177e4SLinus Torvalds return -ENOMEM; 18701da177e4SLinus Torvalds 18711da177e4SLinus Torvalds /* mlock limit tests */ 1872c5d8a364SMiaohe Lin if (mlock_future_check(mm, vma->vm_flags, grow << PAGE_SHIFT)) 18731da177e4SLinus Torvalds return -ENOMEM; 18741da177e4SLinus Torvalds 18750d59a01bSAdam Litke /* Check to ensure the stack will not grow into a hugetlb-only region */ 18760d59a01bSAdam Litke new_start = (vma->vm_flags & VM_GROWSUP) ? vma->vm_start : 18770d59a01bSAdam Litke vma->vm_end - size; 18780d59a01bSAdam Litke if (is_hugepage_only_range(vma->vm_mm, new_start, size)) 18790d59a01bSAdam Litke return -EFAULT; 18800d59a01bSAdam Litke 18811da177e4SLinus Torvalds /* 18821da177e4SLinus Torvalds * Overcommit.. This must be the final test, as it will 18831da177e4SLinus Torvalds * update security statistics. 18841da177e4SLinus Torvalds */ 188505fa199dSHugh Dickins if (security_vm_enough_memory_mm(mm, grow)) 18861da177e4SLinus Torvalds return -ENOMEM; 18871da177e4SLinus Torvalds 18881da177e4SLinus Torvalds return 0; 18891da177e4SLinus Torvalds } 18901da177e4SLinus Torvalds 189146dea3d0SHugh Dickins #if defined(CONFIG_STACK_GROWSUP) || defined(CONFIG_IA64) 18921da177e4SLinus Torvalds /* 189346dea3d0SHugh Dickins * PA-RISC uses this for its stack; IA64 for its Register Backing Store. 189446dea3d0SHugh Dickins * vma is the last one with address > vma->vm_end. Have to extend vma. 18951da177e4SLinus Torvalds */ 189646dea3d0SHugh Dickins int expand_upwards(struct vm_area_struct *vma, unsigned long address) 18971da177e4SLinus Torvalds { 189809357814SOleg Nesterov struct mm_struct *mm = vma->vm_mm; 18991be7107fSHugh Dickins struct vm_area_struct *next; 19001be7107fSHugh Dickins unsigned long gap_addr; 190112352d3cSKonstantin Khlebnikov int error = 0; 1902d4af56c5SLiam R. Howlett MA_STATE(mas, &mm->mm_mt, 0, 0); 19031da177e4SLinus Torvalds 19041da177e4SLinus Torvalds if (!(vma->vm_flags & VM_GROWSUP)) 19051da177e4SLinus Torvalds return -EFAULT; 19061da177e4SLinus Torvalds 1907bd726c90SHelge Deller /* Guard against exceeding limits of the address space. */ 19081be7107fSHugh Dickins address &= PAGE_MASK; 190937511fb5SHelge Deller if (address >= (TASK_SIZE & PAGE_MASK)) 191012352d3cSKonstantin Khlebnikov return -ENOMEM; 1911bd726c90SHelge Deller address += PAGE_SIZE; 191212352d3cSKonstantin Khlebnikov 19131be7107fSHugh Dickins /* Enforce stack_guard_gap */ 19141be7107fSHugh Dickins gap_addr = address + stack_guard_gap; 1915bd726c90SHelge Deller 1916bd726c90SHelge Deller /* Guard against overflow */ 1917bd726c90SHelge Deller if (gap_addr < address || gap_addr > TASK_SIZE) 1918bd726c90SHelge Deller gap_addr = TASK_SIZE; 1919bd726c90SHelge Deller 1920763ecb03SLiam R. Howlett next = find_vma_intersection(mm, vma->vm_end, gap_addr); 1921763ecb03SLiam R. Howlett if (next && vma_is_accessible(next)) { 19221be7107fSHugh Dickins if (!(next->vm_flags & VM_GROWSUP)) 19231be7107fSHugh Dickins return -ENOMEM; 19241be7107fSHugh Dickins /* Check that both stack segments have the same anon_vma? */ 19251be7107fSHugh Dickins } 19261be7107fSHugh Dickins 1927c5d5546eSVernon Yang if (mas_preallocate(&mas, GFP_KERNEL)) 19281da177e4SLinus Torvalds return -ENOMEM; 19291da177e4SLinus Torvalds 1930d4af56c5SLiam R. Howlett /* We must make sure the anon_vma is allocated. */ 1931d4af56c5SLiam R. Howlett if (unlikely(anon_vma_prepare(vma))) { 1932d4af56c5SLiam R. Howlett mas_destroy(&mas); 1933d4af56c5SLiam R. Howlett return -ENOMEM; 1934d4af56c5SLiam R. Howlett } 1935d4af56c5SLiam R. Howlett 19361da177e4SLinus Torvalds /* 19371da177e4SLinus Torvalds * vma->vm_start/vm_end cannot change under us because the caller 1938c1e8d7c6SMichel Lespinasse * is required to hold the mmap_lock in read mode. We need the 19391da177e4SLinus Torvalds * anon_vma lock to serialize against concurrent expand_stacks. 19401da177e4SLinus Torvalds */ 194112352d3cSKonstantin Khlebnikov anon_vma_lock_write(vma->anon_vma); 19421da177e4SLinus Torvalds 19431da177e4SLinus Torvalds /* Somebody else might have raced and expanded it already */ 19441da177e4SLinus Torvalds if (address > vma->vm_end) { 19451da177e4SLinus Torvalds unsigned long size, grow; 19461da177e4SLinus Torvalds 19471da177e4SLinus Torvalds size = address - vma->vm_start; 19481da177e4SLinus Torvalds grow = (address - vma->vm_end) >> PAGE_SHIFT; 19491da177e4SLinus Torvalds 195042c36f63SHugh Dickins error = -ENOMEM; 195142c36f63SHugh Dickins if (vma->vm_pgoff + (size >> PAGE_SHIFT) >= vma->vm_pgoff) { 19521da177e4SLinus Torvalds error = acct_stack_growth(vma, size, grow); 19533af9e859SEric B Munson if (!error) { 19544128997bSMichel Lespinasse /* 1955524e00b3SLiam R. Howlett * We only hold a shared mmap_lock lock here, so 1956524e00b3SLiam R. Howlett * we need to protect against concurrent vma 1957524e00b3SLiam R. Howlett * expansions. anon_vma_lock_write() doesn't 1958524e00b3SLiam R. Howlett * help here, as we don't guarantee that all 1959524e00b3SLiam R. Howlett * growable vmas in a mm share the same root 1960524e00b3SLiam R. Howlett * anon vma. So, we reuse mm->page_table_lock 1961524e00b3SLiam R. Howlett * to guard against concurrent vma expansions. 19624128997bSMichel Lespinasse */ 196309357814SOleg Nesterov spin_lock(&mm->page_table_lock); 196487e8827bSOleg Nesterov if (vma->vm_flags & VM_LOCKED) 196509357814SOleg Nesterov mm->locked_vm += grow; 196684638335SKonstantin Khlebnikov vm_stat_account(mm, vma->vm_flags, grow); 1967bf181b9fSMichel Lespinasse anon_vma_interval_tree_pre_update_vma(vma); 19681da177e4SLinus Torvalds vma->vm_end = address; 1969d4af56c5SLiam R. Howlett /* Overwrite old entry in mtree. */ 1970fbcc3104SLiam R. Howlett mas_set_range(&mas, vma->vm_start, address - 1); 1971fbcc3104SLiam R. Howlett mas_store_prealloc(&mas, vma); 1972bf181b9fSMichel Lespinasse anon_vma_interval_tree_post_update_vma(vma); 197309357814SOleg Nesterov spin_unlock(&mm->page_table_lock); 19744128997bSMichel Lespinasse 19753af9e859SEric B Munson perf_event_mmap(vma); 19763af9e859SEric B Munson } 19771da177e4SLinus Torvalds } 197842c36f63SHugh Dickins } 197912352d3cSKonstantin Khlebnikov anon_vma_unlock_write(vma->anon_vma); 1980c791576cSYang Shi khugepaged_enter_vma(vma, vma->vm_flags); 1981d4af56c5SLiam R. Howlett mas_destroy(&mas); 19821da177e4SLinus Torvalds return error; 19831da177e4SLinus Torvalds } 198446dea3d0SHugh Dickins #endif /* CONFIG_STACK_GROWSUP || CONFIG_IA64 */ 198546dea3d0SHugh Dickins 19861da177e4SLinus Torvalds /* 19871da177e4SLinus Torvalds * vma is the first one with address < vma->vm_start. Have to extend vma. 19881da177e4SLinus Torvalds */ 1989524e00b3SLiam R. Howlett int expand_downwards(struct vm_area_struct *vma, unsigned long address) 19901da177e4SLinus Torvalds { 199109357814SOleg Nesterov struct mm_struct *mm = vma->vm_mm; 1992763ecb03SLiam R. Howlett MA_STATE(mas, &mm->mm_mt, vma->vm_start, vma->vm_start); 19931be7107fSHugh Dickins struct vm_area_struct *prev; 19940a1d5299SJann Horn int error = 0; 19951da177e4SLinus Torvalds 19968869477aSEric Paris address &= PAGE_MASK; 19970a1d5299SJann Horn if (address < mmap_min_addr) 19980a1d5299SJann Horn return -EPERM; 19998869477aSEric Paris 20001be7107fSHugh Dickins /* Enforce stack_guard_gap */ 2001763ecb03SLiam R. Howlett prev = mas_prev(&mas, 0); 20021be7107fSHugh Dickins /* Check that both stack segments have the same anon_vma? */ 200332e4e6d5SOleg Nesterov if (prev && !(prev->vm_flags & VM_GROWSDOWN) && 20043122e80eSAnshuman Khandual vma_is_accessible(prev)) { 200532e4e6d5SOleg Nesterov if (address - prev->vm_end < stack_guard_gap) 200632e4e6d5SOleg Nesterov return -ENOMEM; 20071be7107fSHugh Dickins } 20081be7107fSHugh Dickins 2009c5d5546eSVernon Yang if (mas_preallocate(&mas, GFP_KERNEL)) 201012352d3cSKonstantin Khlebnikov return -ENOMEM; 20111da177e4SLinus Torvalds 2012d4af56c5SLiam R. Howlett /* We must make sure the anon_vma is allocated. */ 2013d4af56c5SLiam R. Howlett if (unlikely(anon_vma_prepare(vma))) { 2014d4af56c5SLiam R. Howlett mas_destroy(&mas); 2015d4af56c5SLiam R. Howlett return -ENOMEM; 2016d4af56c5SLiam R. Howlett } 2017d4af56c5SLiam R. Howlett 20181da177e4SLinus Torvalds /* 20191da177e4SLinus Torvalds * vma->vm_start/vm_end cannot change under us because the caller 2020c1e8d7c6SMichel Lespinasse * is required to hold the mmap_lock in read mode. We need the 20211da177e4SLinus Torvalds * anon_vma lock to serialize against concurrent expand_stacks. 20221da177e4SLinus Torvalds */ 202312352d3cSKonstantin Khlebnikov anon_vma_lock_write(vma->anon_vma); 20241da177e4SLinus Torvalds 20251da177e4SLinus Torvalds /* Somebody else might have raced and expanded it already */ 20261da177e4SLinus Torvalds if (address < vma->vm_start) { 20271da177e4SLinus Torvalds unsigned long size, grow; 20281da177e4SLinus Torvalds 20291da177e4SLinus Torvalds size = vma->vm_end - address; 20301da177e4SLinus Torvalds grow = (vma->vm_start - address) >> PAGE_SHIFT; 20311da177e4SLinus Torvalds 2032a626ca6aSLinus Torvalds error = -ENOMEM; 2033a626ca6aSLinus Torvalds if (grow <= vma->vm_pgoff) { 20341da177e4SLinus Torvalds error = acct_stack_growth(vma, size, grow); 20351da177e4SLinus Torvalds if (!error) { 20364128997bSMichel Lespinasse /* 2037524e00b3SLiam R. Howlett * We only hold a shared mmap_lock lock here, so 2038524e00b3SLiam R. Howlett * we need to protect against concurrent vma 2039524e00b3SLiam R. Howlett * expansions. anon_vma_lock_write() doesn't 2040524e00b3SLiam R. Howlett * help here, as we don't guarantee that all 2041524e00b3SLiam R. Howlett * growable vmas in a mm share the same root 2042524e00b3SLiam R. Howlett * anon vma. So, we reuse mm->page_table_lock 2043524e00b3SLiam R. Howlett * to guard against concurrent vma expansions. 20444128997bSMichel Lespinasse */ 204509357814SOleg Nesterov spin_lock(&mm->page_table_lock); 204687e8827bSOleg Nesterov if (vma->vm_flags & VM_LOCKED) 204709357814SOleg Nesterov mm->locked_vm += grow; 204884638335SKonstantin Khlebnikov vm_stat_account(mm, vma->vm_flags, grow); 2049bf181b9fSMichel Lespinasse anon_vma_interval_tree_pre_update_vma(vma); 20501da177e4SLinus Torvalds vma->vm_start = address; 20511da177e4SLinus Torvalds vma->vm_pgoff -= grow; 2052d4af56c5SLiam R. Howlett /* Overwrite old entry in mtree. */ 2053fbcc3104SLiam R. Howlett mas_set_range(&mas, address, vma->vm_end - 1); 2054fbcc3104SLiam R. Howlett mas_store_prealloc(&mas, vma); 2055bf181b9fSMichel Lespinasse anon_vma_interval_tree_post_update_vma(vma); 205609357814SOleg Nesterov spin_unlock(&mm->page_table_lock); 20574128997bSMichel Lespinasse 20583af9e859SEric B Munson perf_event_mmap(vma); 20591da177e4SLinus Torvalds } 20601da177e4SLinus Torvalds } 2061a626ca6aSLinus Torvalds } 206212352d3cSKonstantin Khlebnikov anon_vma_unlock_write(vma->anon_vma); 2063c791576cSYang Shi khugepaged_enter_vma(vma, vma->vm_flags); 2064d4af56c5SLiam R. Howlett mas_destroy(&mas); 20651da177e4SLinus Torvalds return error; 20661da177e4SLinus Torvalds } 20671da177e4SLinus Torvalds 20681be7107fSHugh Dickins /* enforced gap between the expanding stack and other mappings. */ 20691be7107fSHugh Dickins unsigned long stack_guard_gap = 256UL<<PAGE_SHIFT; 20701be7107fSHugh Dickins 20711be7107fSHugh Dickins static int __init cmdline_parse_stack_guard_gap(char *p) 20721be7107fSHugh Dickins { 20731be7107fSHugh Dickins unsigned long val; 20741be7107fSHugh Dickins char *endptr; 20751be7107fSHugh Dickins 20761be7107fSHugh Dickins val = simple_strtoul(p, &endptr, 10); 20771be7107fSHugh Dickins if (!*endptr) 20781be7107fSHugh Dickins stack_guard_gap = val << PAGE_SHIFT; 20791be7107fSHugh Dickins 2080e6d09493SRandy Dunlap return 1; 20811be7107fSHugh Dickins } 20821be7107fSHugh Dickins __setup("stack_guard_gap=", cmdline_parse_stack_guard_gap); 20831be7107fSHugh Dickins 2084b6a2fea3SOllie Wild #ifdef CONFIG_STACK_GROWSUP 2085b6a2fea3SOllie Wild int expand_stack(struct vm_area_struct *vma, unsigned long address) 2086b6a2fea3SOllie Wild { 2087b6a2fea3SOllie Wild return expand_upwards(vma, address); 2088b6a2fea3SOllie Wild } 2089b6a2fea3SOllie Wild 2090b6a2fea3SOllie Wild struct vm_area_struct * 2091b6a2fea3SOllie Wild find_extend_vma(struct mm_struct *mm, unsigned long addr) 2092b6a2fea3SOllie Wild { 2093b6a2fea3SOllie Wild struct vm_area_struct *vma, *prev; 2094b6a2fea3SOllie Wild 2095b6a2fea3SOllie Wild addr &= PAGE_MASK; 2096b6a2fea3SOllie Wild vma = find_vma_prev(mm, addr, &prev); 2097b6a2fea3SOllie Wild if (vma && (vma->vm_start <= addr)) 2098b6a2fea3SOllie Wild return vma; 20994d45e75aSJann Horn if (!prev || expand_stack(prev, addr)) 2100b6a2fea3SOllie Wild return NULL; 2101cea10a19SMichel Lespinasse if (prev->vm_flags & VM_LOCKED) 2102fc05f566SKirill A. Shutemov populate_vma_page_range(prev, addr, prev->vm_end, NULL); 2103b6a2fea3SOllie Wild return prev; 2104b6a2fea3SOllie Wild } 2105b6a2fea3SOllie Wild #else 2106b6a2fea3SOllie Wild int expand_stack(struct vm_area_struct *vma, unsigned long address) 2107b6a2fea3SOllie Wild { 2108b6a2fea3SOllie Wild return expand_downwards(vma, address); 2109b6a2fea3SOllie Wild } 2110b6a2fea3SOllie Wild 21111da177e4SLinus Torvalds struct vm_area_struct * 21121da177e4SLinus Torvalds find_extend_vma(struct mm_struct *mm, unsigned long addr) 21131da177e4SLinus Torvalds { 21141da177e4SLinus Torvalds struct vm_area_struct *vma; 21151da177e4SLinus Torvalds unsigned long start; 21161da177e4SLinus Torvalds 21171da177e4SLinus Torvalds addr &= PAGE_MASK; 21181da177e4SLinus Torvalds vma = find_vma(mm, addr); 21191da177e4SLinus Torvalds if (!vma) 21201da177e4SLinus Torvalds return NULL; 21211da177e4SLinus Torvalds if (vma->vm_start <= addr) 21221da177e4SLinus Torvalds return vma; 21231da177e4SLinus Torvalds if (!(vma->vm_flags & VM_GROWSDOWN)) 21241da177e4SLinus Torvalds return NULL; 21251da177e4SLinus Torvalds start = vma->vm_start; 21261da177e4SLinus Torvalds if (expand_stack(vma, addr)) 21271da177e4SLinus Torvalds return NULL; 2128cea10a19SMichel Lespinasse if (vma->vm_flags & VM_LOCKED) 2129fc05f566SKirill A. Shutemov populate_vma_page_range(vma, addr, start, NULL); 21301da177e4SLinus Torvalds return vma; 21311da177e4SLinus Torvalds } 21321da177e4SLinus Torvalds #endif 21331da177e4SLinus Torvalds 2134e1d6d01aSJesse Barnes EXPORT_SYMBOL_GPL(find_extend_vma); 2135e1d6d01aSJesse Barnes 21362c0b3814SHugh Dickins /* 2137763ecb03SLiam R. Howlett * Ok - we have the memory areas we should free on a maple tree so release them, 2138763ecb03SLiam R. Howlett * and do the vma updates. 21391da177e4SLinus Torvalds * 21402c0b3814SHugh Dickins * Called with the mm semaphore held. 21411da177e4SLinus Torvalds */ 2142763ecb03SLiam R. Howlett static inline void remove_mt(struct mm_struct *mm, struct ma_state *mas) 21431da177e4SLinus Torvalds { 21444f74d2c8SLinus Torvalds unsigned long nr_accounted = 0; 2145763ecb03SLiam R. Howlett struct vm_area_struct *vma; 21464f74d2c8SLinus Torvalds 2147365e9c87SHugh Dickins /* Update high watermark before we lower total_vm */ 2148365e9c87SHugh Dickins update_hiwater_vm(mm); 2149763ecb03SLiam R. Howlett mas_for_each(mas, vma, ULONG_MAX) { 2150ab50b8edSHugh Dickins long nrpages = vma_pages(vma); 21511da177e4SLinus Torvalds 21524f74d2c8SLinus Torvalds if (vma->vm_flags & VM_ACCOUNT) 21534f74d2c8SLinus Torvalds nr_accounted += nrpages; 215484638335SKonstantin Khlebnikov vm_stat_account(mm, vma->vm_flags, -nrpages); 2155763ecb03SLiam R. Howlett remove_vma(vma); 2156763ecb03SLiam R. Howlett } 21574f74d2c8SLinus Torvalds vm_unacct_memory(nr_accounted); 21581da177e4SLinus Torvalds validate_mm(mm); 21591da177e4SLinus Torvalds } 21601da177e4SLinus Torvalds 21611da177e4SLinus Torvalds /* 21621da177e4SLinus Torvalds * Get rid of page table information in the indicated region. 21631da177e4SLinus Torvalds * 2164f10df686SPaolo 'Blaisorblade' Giarrusso * Called with the mm semaphore held. 21651da177e4SLinus Torvalds */ 2166763ecb03SLiam R. Howlett static void unmap_region(struct mm_struct *mm, struct maple_tree *mt, 2167e0da382cSHugh Dickins struct vm_area_struct *vma, struct vm_area_struct *prev, 2168763ecb03SLiam R. Howlett struct vm_area_struct *next, 2169e0da382cSHugh Dickins unsigned long start, unsigned long end) 21701da177e4SLinus Torvalds { 2171d16dfc55SPeter Zijlstra struct mmu_gather tlb; 21721da177e4SLinus Torvalds 21731da177e4SLinus Torvalds lru_add_drain(); 2174a72afd87SWill Deacon tlb_gather_mmu(&tlb, mm); 2175365e9c87SHugh Dickins update_hiwater_rss(mm); 2176763ecb03SLiam R. Howlett unmap_vmas(&tlb, mt, vma, start, end); 2177763ecb03SLiam R. Howlett free_pgtables(&tlb, mt, vma, prev ? prev->vm_end : FIRST_USER_ADDRESS, 21786ee8630eSHugh Dickins next ? next->vm_start : USER_PGTABLES_CEILING); 2179ae8eba8bSWill Deacon tlb_finish_mmu(&tlb); 21801da177e4SLinus Torvalds } 21811da177e4SLinus Torvalds 21821da177e4SLinus Torvalds /* 2183def5efe0SDavid Rientjes * __split_vma() bypasses sysctl_max_map_count checking. We use this where it 2184def5efe0SDavid Rientjes * has already been checked or doesn't make sense to fail. 21850fd5a9e2SLiam R. Howlett * VMA Iterator will point to the end VMA. 21861da177e4SLinus Torvalds */ 21879760ebffSLiam R. Howlett int __split_vma(struct vma_iterator *vmi, struct vm_area_struct *vma, 21881da177e4SLinus Torvalds unsigned long addr, int new_below) 21891da177e4SLinus Torvalds { 21901da177e4SLinus Torvalds struct vm_area_struct *new; 2191e3975891SChen Gang int err; 21929760ebffSLiam R. Howlett 21939760ebffSLiam R. Howlett validate_mm_mt(vma->vm_mm); 21941da177e4SLinus Torvalds 2195dd3b614fSDmitry Safonov if (vma->vm_ops && vma->vm_ops->may_split) { 2196dd3b614fSDmitry Safonov err = vma->vm_ops->may_split(vma, addr); 219731383c68SDan Williams if (err) 219831383c68SDan Williams return err; 219931383c68SDan Williams } 22001da177e4SLinus Torvalds 22013928d4f5SLinus Torvalds new = vm_area_dup(vma); 22021da177e4SLinus Torvalds if (!new) 2203e3975891SChen Gang return -ENOMEM; 22041da177e4SLinus Torvalds 22051da177e4SLinus Torvalds if (new_below) 22061da177e4SLinus Torvalds new->vm_end = addr; 22071da177e4SLinus Torvalds else { 22081da177e4SLinus Torvalds new->vm_start = addr; 22091da177e4SLinus Torvalds new->vm_pgoff += ((addr - vma->vm_start) >> PAGE_SHIFT); 22101da177e4SLinus Torvalds } 22111da177e4SLinus Torvalds 2212ef0855d3SOleg Nesterov err = vma_dup_policy(vma, new); 2213ef0855d3SOleg Nesterov if (err) 22145beb4930SRik van Riel goto out_free_vma; 22151da177e4SLinus Torvalds 2216c4ea95d7SDaniel Forrest err = anon_vma_clone(new, vma); 2217c4ea95d7SDaniel Forrest if (err) 22185beb4930SRik van Riel goto out_free_mpol; 22195beb4930SRik van Riel 2220e9714acfSKonstantin Khlebnikov if (new->vm_file) 22211da177e4SLinus Torvalds get_file(new->vm_file); 22221da177e4SLinus Torvalds 22231da177e4SLinus Torvalds if (new->vm_ops && new->vm_ops->open) 22241da177e4SLinus Torvalds new->vm_ops->open(new); 22251da177e4SLinus Torvalds 22261da177e4SLinus Torvalds if (new_below) 2227b373037fSLiam R. Howlett err = vma_adjust(vmi, vma, addr, vma->vm_end, 22280fd5a9e2SLiam R. Howlett vma->vm_pgoff + ((addr - new->vm_start) >> PAGE_SHIFT), 2229b373037fSLiam R. Howlett new); 22301da177e4SLinus Torvalds else 2231b373037fSLiam R. Howlett err = vma_adjust(vmi, vma, vma->vm_start, addr, vma->vm_pgoff, 2232b373037fSLiam R. Howlett new); 22331da177e4SLinus Torvalds 22345beb4930SRik van Riel /* Success. */ 22359760ebffSLiam R. Howlett if (!err) { 22360fd5a9e2SLiam R. Howlett if (new_below) 22370fd5a9e2SLiam R. Howlett vma_next(vmi); 22381da177e4SLinus Torvalds return 0; 22399760ebffSLiam R. Howlett } 22405beb4930SRik van Riel 2241d4af56c5SLiam R. Howlett /* Avoid vm accounting in close() operation */ 2242d4af56c5SLiam R. Howlett new->vm_start = new->vm_end; 2243d4af56c5SLiam R. Howlett new->vm_pgoff = 0; 22445beb4930SRik van Riel /* Clean everything up if vma_adjust failed. */ 224558927533SRik van Riel if (new->vm_ops && new->vm_ops->close) 22465beb4930SRik van Riel new->vm_ops->close(new); 2247e9714acfSKonstantin Khlebnikov if (new->vm_file) 22485beb4930SRik van Riel fput(new->vm_file); 22492aeadc30SAndrea Arcangeli unlink_anon_vmas(new); 22505beb4930SRik van Riel out_free_mpol: 2251ef0855d3SOleg Nesterov mpol_put(vma_policy(new)); 22525beb4930SRik van Riel out_free_vma: 22533928d4f5SLinus Torvalds vm_area_free(new); 22549760ebffSLiam R. Howlett validate_mm_mt(vma->vm_mm); 22555beb4930SRik van Riel return err; 22561da177e4SLinus Torvalds } 22571da177e4SLinus Torvalds 2258659ace58SKOSAKI Motohiro /* 2259659ace58SKOSAKI Motohiro * Split a vma into two pieces at address 'addr', a new vma is allocated 2260659ace58SKOSAKI Motohiro * either for the first part or the tail. 2261659ace58SKOSAKI Motohiro */ 22629760ebffSLiam R. Howlett int split_vma(struct vma_iterator *vmi, struct vm_area_struct *vma, 2263659ace58SKOSAKI Motohiro unsigned long addr, int new_below) 2264659ace58SKOSAKI Motohiro { 22659760ebffSLiam R. Howlett if (vma->vm_mm->map_count >= sysctl_max_map_count) 2266659ace58SKOSAKI Motohiro return -ENOMEM; 2267659ace58SKOSAKI Motohiro 22689760ebffSLiam R. Howlett return __split_vma(vmi, vma, addr, new_below); 2269f2ebfe43SLiam R. Howlett } 2270f2ebfe43SLiam R. Howlett 2271763ecb03SLiam R. Howlett static inline int munmap_sidetree(struct vm_area_struct *vma, 2272763ecb03SLiam R. Howlett struct ma_state *mas_detach) 22731da177e4SLinus Torvalds { 2274763ecb03SLiam R. Howlett mas_set_range(mas_detach, vma->vm_start, vma->vm_end - 1); 2275763ecb03SLiam R. Howlett if (mas_store_gfp(mas_detach, vma, GFP_KERNEL)) 2276763ecb03SLiam R. Howlett return -ENOMEM; 227711f9a21aSLiam R. Howlett 2278763ecb03SLiam R. Howlett if (vma->vm_flags & VM_LOCKED) 2279763ecb03SLiam R. Howlett vma->vm_mm->locked_vm -= vma_pages(vma); 228011f9a21aSLiam R. Howlett 2281763ecb03SLiam R. Howlett return 0; 228211f9a21aSLiam R. Howlett } 228311f9a21aSLiam R. Howlett 228411f9a21aSLiam R. Howlett /* 2285183654ceSLiam R. Howlett * do_vmi_align_munmap() - munmap the aligned region from @start to @end. 2286183654ceSLiam R. Howlett * @vmi: The vma iterator 228711f9a21aSLiam R. Howlett * @vma: The starting vm_area_struct 228811f9a21aSLiam R. Howlett * @mm: The mm_struct 228911f9a21aSLiam R. Howlett * @start: The aligned start address to munmap. 229011f9a21aSLiam R. Howlett * @end: The aligned end address to munmap. 229111f9a21aSLiam R. Howlett * @uf: The userfaultfd list_head 22928651a137SLorenzo Stoakes * @downgrade: Set to true to attempt a write downgrade of the mmap_lock 229311f9a21aSLiam R. Howlett * 229411f9a21aSLiam R. Howlett * If @downgrade is true, check return code for potential release of the lock. 229511f9a21aSLiam R. Howlett */ 229611f9a21aSLiam R. Howlett static int 2297183654ceSLiam R. Howlett do_vmi_align_munmap(struct vma_iterator *vmi, struct vm_area_struct *vma, 229811f9a21aSLiam R. Howlett struct mm_struct *mm, unsigned long start, 229911f9a21aSLiam R. Howlett unsigned long end, struct list_head *uf, bool downgrade) 230011f9a21aSLiam R. Howlett { 2301763ecb03SLiam R. Howlett struct vm_area_struct *prev, *next = NULL; 2302763ecb03SLiam R. Howlett struct maple_tree mt_detach; 2303763ecb03SLiam R. Howlett int count = 0; 2304d4af56c5SLiam R. Howlett int error = -ENOMEM; 2305763ecb03SLiam R. Howlett MA_STATE(mas_detach, &mt_detach, 0, 0); 2306763ecb03SLiam R. Howlett mt_init_flags(&mt_detach, MT_FLAGS_LOCK_EXTERN); 2307763ecb03SLiam R. Howlett mt_set_external_lock(&mt_detach, &mm->mmap_lock); 2308524e00b3SLiam R. Howlett 23091da177e4SLinus Torvalds /* 23101da177e4SLinus Torvalds * If we need to split any vma, do it now to save pain later. 23111da177e4SLinus Torvalds * 23121da177e4SLinus Torvalds * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially 23131da177e4SLinus Torvalds * unmapped vm_area_struct will remain in use: so lower split_vma 23141da177e4SLinus Torvalds * places tmp vma above, and higher split_vma places tmp vma below. 23151da177e4SLinus Torvalds */ 2316763ecb03SLiam R. Howlett 2317763ecb03SLiam R. Howlett /* Does it split the first one? */ 2318146425a3SHugh Dickins if (start > vma->vm_start) { 2319659ace58SKOSAKI Motohiro 2320659ace58SKOSAKI Motohiro /* 2321659ace58SKOSAKI Motohiro * Make sure that map_count on return from munmap() will 2322659ace58SKOSAKI Motohiro * not exceed its limit; but let map_count go just above 2323659ace58SKOSAKI Motohiro * its limit temporarily, to help free resources as expected. 2324659ace58SKOSAKI Motohiro */ 2325659ace58SKOSAKI Motohiro if (end < vma->vm_end && mm->map_count >= sysctl_max_map_count) 2326d4af56c5SLiam R. Howlett goto map_count_exceeded; 2327659ace58SKOSAKI Motohiro 23289760ebffSLiam R. Howlett error = __split_vma(vmi, vma, start, 0); 23291da177e4SLinus Torvalds if (error) 2330763ecb03SLiam R. Howlett goto start_split_failed; 233111f9a21aSLiam R. Howlett 23320fd5a9e2SLiam R. Howlett vma = vma_iter_load(vmi); 23331da177e4SLinus Torvalds } 23341da177e4SLinus Torvalds 2335183654ceSLiam R. Howlett prev = vma_prev(vmi); 2336763ecb03SLiam R. Howlett if (unlikely((!prev))) 2337183654ceSLiam R. Howlett vma_iter_set(vmi, start); 233811f9a21aSLiam R. Howlett 2339763ecb03SLiam R. Howlett /* 2340763ecb03SLiam R. Howlett * Detach a range of VMAs from the mm. Using next as a temp variable as 2341763ecb03SLiam R. Howlett * it is always overwritten. 2342763ecb03SLiam R. Howlett */ 2343183654ceSLiam R. Howlett for_each_vma_range(*vmi, next, end) { 2344763ecb03SLiam R. Howlett /* Does it split the end? */ 2345763ecb03SLiam R. Howlett if (next->vm_end > end) { 23466b73cff2SLiam R. Howlett error = __split_vma(vmi, next, end, 0); 23471da177e4SLinus Torvalds if (error) 2348763ecb03SLiam R. Howlett goto end_split_failed; 234911f9a21aSLiam R. Howlett } 2350763ecb03SLiam R. Howlett error = munmap_sidetree(next, &mas_detach); 2351763ecb03SLiam R. Howlett if (error) 2352763ecb03SLiam R. Howlett goto munmap_sidetree_failed; 2353763ecb03SLiam R. Howlett 2354763ecb03SLiam R. Howlett count++; 2355763ecb03SLiam R. Howlett #ifdef CONFIG_DEBUG_VM_MAPLE_TREE 2356763ecb03SLiam R. Howlett BUG_ON(next->vm_start < start); 2357763ecb03SLiam R. Howlett BUG_ON(next->vm_start > end); 2358763ecb03SLiam R. Howlett #endif 2359763ecb03SLiam R. Howlett } 2360763ecb03SLiam R. Howlett 2361183654ceSLiam R. Howlett next = vma_next(vmi); 23622376dd7cSAndrea Arcangeli if (unlikely(uf)) { 23632376dd7cSAndrea Arcangeli /* 23642376dd7cSAndrea Arcangeli * If userfaultfd_unmap_prep returns an error the vmas 2365f0953a1bSIngo Molnar * will remain split, but userland will get a 23662376dd7cSAndrea Arcangeli * highly unexpected error anyway. This is no 23672376dd7cSAndrea Arcangeli * different than the case where the first of the two 23682376dd7cSAndrea Arcangeli * __split_vma fails, but we don't undo the first 23692376dd7cSAndrea Arcangeli * split, despite we could. This is unlikely enough 23702376dd7cSAndrea Arcangeli * failure that it's not worth optimizing it for. 23712376dd7cSAndrea Arcangeli */ 237269dbe6daSLiam R. Howlett error = userfaultfd_unmap_prep(mm, start, end, uf); 237311f9a21aSLiam R. Howlett 23742376dd7cSAndrea Arcangeli if (error) 2375d4af56c5SLiam R. Howlett goto userfaultfd_error; 23762376dd7cSAndrea Arcangeli } 23772376dd7cSAndrea Arcangeli 2378763ecb03SLiam R. Howlett #if defined(CONFIG_DEBUG_VM_MAPLE_TREE) 2379763ecb03SLiam R. Howlett /* Make sure no VMAs are about to be lost. */ 2380763ecb03SLiam R. Howlett { 2381763ecb03SLiam R. Howlett MA_STATE(test, &mt_detach, start, end - 1); 2382763ecb03SLiam R. Howlett struct vm_area_struct *vma_mas, *vma_test; 2383763ecb03SLiam R. Howlett int test_count = 0; 2384763ecb03SLiam R. Howlett 2385183654ceSLiam R. Howlett vma_iter_set(vmi, start); 2386763ecb03SLiam R. Howlett rcu_read_lock(); 2387763ecb03SLiam R. Howlett vma_test = mas_find(&test, end - 1); 2388183654ceSLiam R. Howlett for_each_vma_range(*vmi, vma_mas, end) { 2389763ecb03SLiam R. Howlett BUG_ON(vma_mas != vma_test); 2390763ecb03SLiam R. Howlett test_count++; 2391763ecb03SLiam R. Howlett vma_test = mas_next(&test, end - 1); 2392763ecb03SLiam R. Howlett } 2393763ecb03SLiam R. Howlett rcu_read_unlock(); 2394763ecb03SLiam R. Howlett BUG_ON(count != test_count); 2395763ecb03SLiam R. Howlett } 2396763ecb03SLiam R. Howlett #endif 23970378c0a0SLiam R. Howlett /* Point of no return */ 2398183654ceSLiam R. Howlett vma_iter_set(vmi, start); 2399183654ceSLiam R. Howlett if (vma_iter_clear_gfp(vmi, start, end, GFP_KERNEL)) 24000378c0a0SLiam R. Howlett return -ENOMEM; 24010378c0a0SLiam R. Howlett 2402763ecb03SLiam R. Howlett mm->map_count -= count; 240311f9a21aSLiam R. Howlett /* 240411f9a21aSLiam R. Howlett * Do not downgrade mmap_lock if we are next to VM_GROWSDOWN or 240511f9a21aSLiam R. Howlett * VM_GROWSUP VMA. Such VMAs can change their size under 240611f9a21aSLiam R. Howlett * down_read(mmap_lock) and collide with the VMA we are about to unmap. 240711f9a21aSLiam R. Howlett */ 240811f9a21aSLiam R. Howlett if (downgrade) { 2409763ecb03SLiam R. Howlett if (next && (next->vm_flags & VM_GROWSDOWN)) 241011f9a21aSLiam R. Howlett downgrade = false; 241111f9a21aSLiam R. Howlett else if (prev && (prev->vm_flags & VM_GROWSUP)) 241211f9a21aSLiam R. Howlett downgrade = false; 241311f9a21aSLiam R. Howlett else 2414d8ed45c5SMichel Lespinasse mmap_write_downgrade(mm); 241511f9a21aSLiam R. Howlett } 2416dd2283f2SYang Shi 2417763ecb03SLiam R. Howlett unmap_region(mm, &mt_detach, vma, prev, next, start, end); 2418763ecb03SLiam R. Howlett /* Statistics and freeing VMAs */ 2419763ecb03SLiam R. Howlett mas_set(&mas_detach, start); 2420763ecb03SLiam R. Howlett remove_mt(mm, &mas_detach); 2421763ecb03SLiam R. Howlett __mt_destroy(&mt_detach); 24221da177e4SLinus Torvalds 2423524e00b3SLiam R. Howlett 2424524e00b3SLiam R. Howlett validate_mm(mm); 2425dd2283f2SYang Shi return downgrade ? 1 : 0; 2426d4af56c5SLiam R. Howlett 2427d4af56c5SLiam R. Howlett userfaultfd_error: 2428763ecb03SLiam R. Howlett munmap_sidetree_failed: 2429763ecb03SLiam R. Howlett end_split_failed: 2430763ecb03SLiam R. Howlett __mt_destroy(&mt_detach); 2431763ecb03SLiam R. Howlett start_split_failed: 2432763ecb03SLiam R. Howlett map_count_exceeded: 2433d4af56c5SLiam R. Howlett return error; 24341da177e4SLinus Torvalds } 24351da177e4SLinus Torvalds 243611f9a21aSLiam R. Howlett /* 2437183654ceSLiam R. Howlett * do_vmi_munmap() - munmap a given range. 2438183654ceSLiam R. Howlett * @vmi: The vma iterator 243911f9a21aSLiam R. Howlett * @mm: The mm_struct 244011f9a21aSLiam R. Howlett * @start: The start address to munmap 244111f9a21aSLiam R. Howlett * @len: The length of the range to munmap 244211f9a21aSLiam R. Howlett * @uf: The userfaultfd list_head 244311f9a21aSLiam R. Howlett * @downgrade: set to true if the user wants to attempt to write_downgrade the 24448651a137SLorenzo Stoakes * mmap_lock 244511f9a21aSLiam R. Howlett * 244611f9a21aSLiam R. Howlett * This function takes a @mas that is either pointing to the previous VMA or set 244711f9a21aSLiam R. Howlett * to MA_START and sets it up to remove the mapping(s). The @len will be 244811f9a21aSLiam R. Howlett * aligned and any arch_unmap work will be preformed. 244911f9a21aSLiam R. Howlett * 245011f9a21aSLiam R. Howlett * Returns: -EINVAL on failure, 1 on success and unlock, 0 otherwise. 245111f9a21aSLiam R. Howlett */ 2452183654ceSLiam R. Howlett int do_vmi_munmap(struct vma_iterator *vmi, struct mm_struct *mm, 245311f9a21aSLiam R. Howlett unsigned long start, size_t len, struct list_head *uf, 245411f9a21aSLiam R. Howlett bool downgrade) 245511f9a21aSLiam R. Howlett { 245611f9a21aSLiam R. Howlett unsigned long end; 245711f9a21aSLiam R. Howlett struct vm_area_struct *vma; 245811f9a21aSLiam R. Howlett 245911f9a21aSLiam R. Howlett if ((offset_in_page(start)) || start > TASK_SIZE || len > TASK_SIZE-start) 246011f9a21aSLiam R. Howlett return -EINVAL; 246111f9a21aSLiam R. Howlett 246211f9a21aSLiam R. Howlett end = start + PAGE_ALIGN(len); 246311f9a21aSLiam R. Howlett if (end == start) 246411f9a21aSLiam R. Howlett return -EINVAL; 246511f9a21aSLiam R. Howlett 246611f9a21aSLiam R. Howlett /* arch_unmap() might do unmaps itself. */ 246711f9a21aSLiam R. Howlett arch_unmap(mm, start, end); 246811f9a21aSLiam R. Howlett 246911f9a21aSLiam R. Howlett /* Find the first overlapping VMA */ 2470183654ceSLiam R. Howlett vma = vma_find(vmi, end); 247111f9a21aSLiam R. Howlett if (!vma) 247211f9a21aSLiam R. Howlett return 0; 247311f9a21aSLiam R. Howlett 2474183654ceSLiam R. Howlett return do_vmi_align_munmap(vmi, vma, mm, start, end, uf, downgrade); 247511f9a21aSLiam R. Howlett } 247611f9a21aSLiam R. Howlett 247711f9a21aSLiam R. Howlett /* do_munmap() - Wrapper function for non-maple tree aware do_munmap() calls. 247811f9a21aSLiam R. Howlett * @mm: The mm_struct 247911f9a21aSLiam R. Howlett * @start: The start address to munmap 248011f9a21aSLiam R. Howlett * @len: The length to be munmapped. 248111f9a21aSLiam R. Howlett * @uf: The userfaultfd list_head 248211f9a21aSLiam R. Howlett */ 2483dd2283f2SYang Shi int do_munmap(struct mm_struct *mm, unsigned long start, size_t len, 2484dd2283f2SYang Shi struct list_head *uf) 2485dd2283f2SYang Shi { 2486183654ceSLiam R. Howlett VMA_ITERATOR(vmi, mm, start); 248711f9a21aSLiam R. Howlett 2488183654ceSLiam R. Howlett return do_vmi_munmap(&vmi, mm, start, len, uf, false); 2489dd2283f2SYang Shi } 2490dd2283f2SYang Shi 2491e99668a5SLiam R. Howlett unsigned long mmap_region(struct file *file, unsigned long addr, 2492e99668a5SLiam R. Howlett unsigned long len, vm_flags_t vm_flags, unsigned long pgoff, 2493e99668a5SLiam R. Howlett struct list_head *uf) 2494e99668a5SLiam R. Howlett { 2495e99668a5SLiam R. Howlett struct mm_struct *mm = current->mm; 2496e99668a5SLiam R. Howlett struct vm_area_struct *vma = NULL; 2497e99668a5SLiam R. Howlett struct vm_area_struct *next, *prev, *merge; 2498e99668a5SLiam R. Howlett pgoff_t pglen = len >> PAGE_SHIFT; 2499e99668a5SLiam R. Howlett unsigned long charged = 0; 2500e99668a5SLiam R. Howlett unsigned long end = addr + len; 2501e99668a5SLiam R. Howlett unsigned long merge_start = addr, merge_end = end; 2502e99668a5SLiam R. Howlett pgoff_t vm_pgoff; 2503e99668a5SLiam R. Howlett int error; 2504183654ceSLiam R. Howlett VMA_ITERATOR(vmi, mm, addr); 2505e99668a5SLiam R. Howlett 2506e99668a5SLiam R. Howlett /* Check against address space limit. */ 2507e99668a5SLiam R. Howlett if (!may_expand_vm(mm, vm_flags, len >> PAGE_SHIFT)) { 2508e99668a5SLiam R. Howlett unsigned long nr_pages; 2509e99668a5SLiam R. Howlett 2510e99668a5SLiam R. Howlett /* 2511e99668a5SLiam R. Howlett * MAP_FIXED may remove pages of mappings that intersects with 2512e99668a5SLiam R. Howlett * requested mapping. Account for the pages it would unmap. 2513e99668a5SLiam R. Howlett */ 2514e99668a5SLiam R. Howlett nr_pages = count_vma_pages_range(mm, addr, end); 2515e99668a5SLiam R. Howlett 2516e99668a5SLiam R. Howlett if (!may_expand_vm(mm, vm_flags, 2517e99668a5SLiam R. Howlett (len >> PAGE_SHIFT) - nr_pages)) 2518e99668a5SLiam R. Howlett return -ENOMEM; 2519e99668a5SLiam R. Howlett } 2520e99668a5SLiam R. Howlett 2521e99668a5SLiam R. Howlett /* Unmap any existing mapping in the area */ 2522183654ceSLiam R. Howlett if (do_vmi_munmap(&vmi, mm, addr, len, uf, false)) 2523e99668a5SLiam R. Howlett return -ENOMEM; 2524e99668a5SLiam R. Howlett 2525e99668a5SLiam R. Howlett /* 2526e99668a5SLiam R. Howlett * Private writable mapping: check memory availability 2527e99668a5SLiam R. Howlett */ 2528e99668a5SLiam R. Howlett if (accountable_mapping(file, vm_flags)) { 2529e99668a5SLiam R. Howlett charged = len >> PAGE_SHIFT; 2530e99668a5SLiam R. Howlett if (security_vm_enough_memory_mm(mm, charged)) 2531e99668a5SLiam R. Howlett return -ENOMEM; 2532e99668a5SLiam R. Howlett vm_flags |= VM_ACCOUNT; 2533e99668a5SLiam R. Howlett } 2534e99668a5SLiam R. Howlett 2535183654ceSLiam R. Howlett next = vma_next(&vmi); 2536183654ceSLiam R. Howlett prev = vma_prev(&vmi); 2537e99668a5SLiam R. Howlett if (vm_flags & VM_SPECIAL) 2538e99668a5SLiam R. Howlett goto cannot_expand; 2539e99668a5SLiam R. Howlett 2540e99668a5SLiam R. Howlett /* Attempt to expand an old mapping */ 2541e99668a5SLiam R. Howlett /* Check next */ 2542e99668a5SLiam R. Howlett if (next && next->vm_start == end && !vma_policy(next) && 2543e99668a5SLiam R. Howlett can_vma_merge_before(next, vm_flags, NULL, file, pgoff+pglen, 2544e99668a5SLiam R. Howlett NULL_VM_UFFD_CTX, NULL)) { 2545e99668a5SLiam R. Howlett merge_end = next->vm_end; 2546e99668a5SLiam R. Howlett vma = next; 2547e99668a5SLiam R. Howlett vm_pgoff = next->vm_pgoff - pglen; 2548e99668a5SLiam R. Howlett } 2549e99668a5SLiam R. Howlett 2550e99668a5SLiam R. Howlett /* Check prev */ 2551e99668a5SLiam R. Howlett if (prev && prev->vm_end == addr && !vma_policy(prev) && 2552e99668a5SLiam R. Howlett (vma ? can_vma_merge_after(prev, vm_flags, vma->anon_vma, file, 2553e99668a5SLiam R. Howlett pgoff, vma->vm_userfaultfd_ctx, NULL) : 2554e99668a5SLiam R. Howlett can_vma_merge_after(prev, vm_flags, NULL, file, pgoff, 2555e99668a5SLiam R. Howlett NULL_VM_UFFD_CTX, NULL))) { 2556e99668a5SLiam R. Howlett merge_start = prev->vm_start; 2557e99668a5SLiam R. Howlett vma = prev; 2558e99668a5SLiam R. Howlett vm_pgoff = prev->vm_pgoff; 2559e99668a5SLiam R. Howlett } 2560e99668a5SLiam R. Howlett 2561e99668a5SLiam R. Howlett 2562e99668a5SLiam R. Howlett /* Actually expand, if possible */ 2563e99668a5SLiam R. Howlett if (vma && 25643c441ab7SLiam R. Howlett !vma_expand(&vmi, vma, merge_start, merge_end, vm_pgoff, next)) { 2565e99668a5SLiam R. Howlett khugepaged_enter_vma(vma, vm_flags); 2566e99668a5SLiam R. Howlett goto expanded; 2567e99668a5SLiam R. Howlett } 2568e99668a5SLiam R. Howlett 2569e99668a5SLiam R. Howlett cannot_expand: 2570e99668a5SLiam R. Howlett /* 2571e99668a5SLiam R. Howlett * Determine the object being mapped and call the appropriate 2572e99668a5SLiam R. Howlett * specific mapper. the address has already been validated, but 2573e99668a5SLiam R. Howlett * not unmapped, but the maps are removed from the list. 2574e99668a5SLiam R. Howlett */ 2575e99668a5SLiam R. Howlett vma = vm_area_alloc(mm); 2576e99668a5SLiam R. Howlett if (!vma) { 2577e99668a5SLiam R. Howlett error = -ENOMEM; 2578e99668a5SLiam R. Howlett goto unacct_error; 2579e99668a5SLiam R. Howlett } 2580e99668a5SLiam R. Howlett 25810fd5a9e2SLiam R. Howlett vma_iter_set(&vmi, addr); 2582e99668a5SLiam R. Howlett vma->vm_start = addr; 2583e99668a5SLiam R. Howlett vma->vm_end = end; 2584e99668a5SLiam R. Howlett vma->vm_flags = vm_flags; 2585e99668a5SLiam R. Howlett vma->vm_page_prot = vm_get_page_prot(vm_flags); 2586e99668a5SLiam R. Howlett vma->vm_pgoff = pgoff; 2587e99668a5SLiam R. Howlett 2588e99668a5SLiam R. Howlett if (file) { 2589e99668a5SLiam R. Howlett if (vm_flags & VM_SHARED) { 2590e99668a5SLiam R. Howlett error = mapping_map_writable(file->f_mapping); 2591e99668a5SLiam R. Howlett if (error) 2592e99668a5SLiam R. Howlett goto free_vma; 2593e99668a5SLiam R. Howlett } 2594e99668a5SLiam R. Howlett 2595e99668a5SLiam R. Howlett vma->vm_file = get_file(file); 2596e99668a5SLiam R. Howlett error = call_mmap(file, vma); 2597e99668a5SLiam R. Howlett if (error) 2598e99668a5SLiam R. Howlett goto unmap_and_free_vma; 2599e99668a5SLiam R. Howlett 2600a57b7051SLiam Howlett /* 2601a57b7051SLiam Howlett * Expansion is handled above, merging is handled below. 2602a57b7051SLiam Howlett * Drivers should not alter the address of the VMA. 2603e99668a5SLiam R. Howlett */ 2604a57b7051SLiam Howlett error = -EINVAL; 2605cc8d1b09SLiam R. Howlett if (WARN_ON((addr != vma->vm_start))) 2606a57b7051SLiam Howlett goto close_and_free_vma; 2607e99668a5SLiam R. Howlett 2608cc8d1b09SLiam R. Howlett vma_iter_set(&vmi, addr); 2609e99668a5SLiam R. Howlett /* 2610e99668a5SLiam R. Howlett * If vm_flags changed after call_mmap(), we should try merge 2611e99668a5SLiam R. Howlett * vma again as we may succeed this time. 2612e99668a5SLiam R. Howlett */ 2613e99668a5SLiam R. Howlett if (unlikely(vm_flags != vma->vm_flags && prev)) { 26149760ebffSLiam R. Howlett merge = vma_merge(&vmi, mm, prev, vma->vm_start, 26159760ebffSLiam R. Howlett vma->vm_end, vma->vm_flags, NULL, 26169760ebffSLiam R. Howlett vma->vm_file, vma->vm_pgoff, NULL, 26179760ebffSLiam R. Howlett NULL_VM_UFFD_CTX, NULL); 2618e99668a5SLiam R. Howlett if (merge) { 2619e99668a5SLiam R. Howlett /* 2620e99668a5SLiam R. Howlett * ->mmap() can change vma->vm_file and fput 2621e99668a5SLiam R. Howlett * the original file. So fput the vma->vm_file 2622e99668a5SLiam R. Howlett * here or we would add an extra fput for file 2623e99668a5SLiam R. Howlett * and cause general protection fault 2624e99668a5SLiam R. Howlett * ultimately. 2625e99668a5SLiam R. Howlett */ 2626e99668a5SLiam R. Howlett fput(vma->vm_file); 2627e99668a5SLiam R. Howlett vm_area_free(vma); 2628e99668a5SLiam R. Howlett vma = merge; 2629e99668a5SLiam R. Howlett /* Update vm_flags to pick up the change. */ 2630e99668a5SLiam R. Howlett vm_flags = vma->vm_flags; 2631e99668a5SLiam R. Howlett goto unmap_writable; 2632e99668a5SLiam R. Howlett } 2633e99668a5SLiam R. Howlett } 2634e99668a5SLiam R. Howlett 2635e99668a5SLiam R. Howlett vm_flags = vma->vm_flags; 2636e99668a5SLiam R. Howlett } else if (vm_flags & VM_SHARED) { 2637e99668a5SLiam R. Howlett error = shmem_zero_setup(vma); 2638e99668a5SLiam R. Howlett if (error) 2639e99668a5SLiam R. Howlett goto free_vma; 2640e99668a5SLiam R. Howlett } else { 2641e99668a5SLiam R. Howlett vma_set_anonymous(vma); 2642e99668a5SLiam R. Howlett } 2643e99668a5SLiam R. Howlett 2644b507808eSJoey Gouly if (map_deny_write_exec(vma, vma->vm_flags)) { 2645b507808eSJoey Gouly error = -EACCES; 2646b507808eSJoey Gouly if (file) 2647b507808eSJoey Gouly goto close_and_free_vma; 2648b507808eSJoey Gouly else if (vma->vm_file) 2649b507808eSJoey Gouly goto unmap_and_free_vma; 2650b507808eSJoey Gouly else 2651b507808eSJoey Gouly goto free_vma; 2652b507808eSJoey Gouly } 2653b507808eSJoey Gouly 2654e99668a5SLiam R. Howlett /* Allow architectures to sanity-check the vm_flags */ 2655e99668a5SLiam R. Howlett error = -EINVAL; 2656cc8d1b09SLiam R. Howlett if (!arch_validate_flags(vma->vm_flags)) 2657deb0f656SCarlos Llamas goto close_and_free_vma; 2658e99668a5SLiam R. Howlett 2659e99668a5SLiam R. Howlett error = -ENOMEM; 2660cc8d1b09SLiam R. Howlett if (vma_iter_prealloc(&vmi)) 26615789151eSMike Kravetz goto close_and_free_vma; 2662e99668a5SLiam R. Howlett 2663e99668a5SLiam R. Howlett if (vma->vm_file) 2664e99668a5SLiam R. Howlett i_mmap_lock_write(vma->vm_file->f_mapping); 2665e99668a5SLiam R. Howlett 2666183654ceSLiam R. Howlett vma_iter_store(&vmi, vma); 2667e99668a5SLiam R. Howlett mm->map_count++; 2668e99668a5SLiam R. Howlett if (vma->vm_file) { 2669e99668a5SLiam R. Howlett if (vma->vm_flags & VM_SHARED) 2670e99668a5SLiam R. Howlett mapping_allow_writable(vma->vm_file->f_mapping); 2671e99668a5SLiam R. Howlett 2672e99668a5SLiam R. Howlett flush_dcache_mmap_lock(vma->vm_file->f_mapping); 2673e99668a5SLiam R. Howlett vma_interval_tree_insert(vma, &vma->vm_file->f_mapping->i_mmap); 2674e99668a5SLiam R. Howlett flush_dcache_mmap_unlock(vma->vm_file->f_mapping); 2675e99668a5SLiam R. Howlett i_mmap_unlock_write(vma->vm_file->f_mapping); 2676e99668a5SLiam R. Howlett } 2677e99668a5SLiam R. Howlett 2678e99668a5SLiam R. Howlett /* 2679e99668a5SLiam R. Howlett * vma_merge() calls khugepaged_enter_vma() either, the below 2680e99668a5SLiam R. Howlett * call covers the non-merge case. 2681e99668a5SLiam R. Howlett */ 2682e99668a5SLiam R. Howlett khugepaged_enter_vma(vma, vma->vm_flags); 2683e99668a5SLiam R. Howlett 2684e99668a5SLiam R. Howlett /* Once vma denies write, undo our temporary denial count */ 2685e99668a5SLiam R. Howlett unmap_writable: 2686e99668a5SLiam R. Howlett if (file && vm_flags & VM_SHARED) 2687e99668a5SLiam R. Howlett mapping_unmap_writable(file->f_mapping); 2688e99668a5SLiam R. Howlett file = vma->vm_file; 2689e99668a5SLiam R. Howlett expanded: 2690e99668a5SLiam R. Howlett perf_event_mmap(vma); 2691e99668a5SLiam R. Howlett 2692e99668a5SLiam R. Howlett vm_stat_account(mm, vm_flags, len >> PAGE_SHIFT); 2693e99668a5SLiam R. Howlett if (vm_flags & VM_LOCKED) { 2694e99668a5SLiam R. Howlett if ((vm_flags & VM_SPECIAL) || vma_is_dax(vma) || 2695e99668a5SLiam R. Howlett is_vm_hugetlb_page(vma) || 2696e99668a5SLiam R. Howlett vma == get_gate_vma(current->mm)) 2697e99668a5SLiam R. Howlett vma->vm_flags &= VM_LOCKED_CLEAR_MASK; 2698e99668a5SLiam R. Howlett else 2699e99668a5SLiam R. Howlett mm->locked_vm += (len >> PAGE_SHIFT); 2700e99668a5SLiam R. Howlett } 2701e99668a5SLiam R. Howlett 2702e99668a5SLiam R. Howlett if (file) 2703e99668a5SLiam R. Howlett uprobe_mmap(vma); 2704e99668a5SLiam R. Howlett 2705e99668a5SLiam R. Howlett /* 2706e99668a5SLiam R. Howlett * New (or expanded) vma always get soft dirty status. 2707e99668a5SLiam R. Howlett * Otherwise user-space soft-dirty page tracker won't 2708e99668a5SLiam R. Howlett * be able to distinguish situation when vma area unmapped, 2709e99668a5SLiam R. Howlett * then new mapped in-place (which must be aimed as 2710e99668a5SLiam R. Howlett * a completely new data area). 2711e99668a5SLiam R. Howlett */ 2712e99668a5SLiam R. Howlett vma->vm_flags |= VM_SOFTDIRTY; 2713e99668a5SLiam R. Howlett 2714e99668a5SLiam R. Howlett vma_set_page_prot(vma); 2715e99668a5SLiam R. Howlett 2716e99668a5SLiam R. Howlett validate_mm(mm); 2717e99668a5SLiam R. Howlett return addr; 2718e99668a5SLiam R. Howlett 2719deb0f656SCarlos Llamas close_and_free_vma: 2720cc8d1b09SLiam R. Howlett if (file && vma->vm_ops && vma->vm_ops->close) 2721deb0f656SCarlos Llamas vma->vm_ops->close(vma); 2722cc8d1b09SLiam R. Howlett 2723cc8d1b09SLiam R. Howlett if (file || vma->vm_file) { 2724e99668a5SLiam R. Howlett unmap_and_free_vma: 2725e99668a5SLiam R. Howlett fput(vma->vm_file); 2726e99668a5SLiam R. Howlett vma->vm_file = NULL; 2727e99668a5SLiam R. Howlett 2728e99668a5SLiam R. Howlett /* Undo any partial mapping done by a device driver. */ 2729cc8d1b09SLiam R. Howlett unmap_region(mm, &mm->mm_mt, vma, prev, next, vma->vm_start, 2730cc8d1b09SLiam R. Howlett vma->vm_end); 2731cc8d1b09SLiam R. Howlett } 2732cc674ab3SLi Zetao if (file && (vm_flags & VM_SHARED)) 2733e99668a5SLiam R. Howlett mapping_unmap_writable(file->f_mapping); 2734e99668a5SLiam R. Howlett free_vma: 2735e99668a5SLiam R. Howlett vm_area_free(vma); 2736e99668a5SLiam R. Howlett unacct_error: 2737e99668a5SLiam R. Howlett if (charged) 2738e99668a5SLiam R. Howlett vm_unacct_memory(charged); 2739e99668a5SLiam R. Howlett validate_mm(mm); 2740e99668a5SLiam R. Howlett return error; 2741e99668a5SLiam R. Howlett } 2742e99668a5SLiam R. Howlett 2743dd2283f2SYang Shi static int __vm_munmap(unsigned long start, size_t len, bool downgrade) 2744a46ef99dSLinus Torvalds { 2745a46ef99dSLinus Torvalds int ret; 2746bfce281cSAl Viro struct mm_struct *mm = current->mm; 2747897ab3e0SMike Rapoport LIST_HEAD(uf); 2748183654ceSLiam R. Howlett VMA_ITERATOR(vmi, mm, start); 2749a46ef99dSLinus Torvalds 2750d8ed45c5SMichel Lespinasse if (mmap_write_lock_killable(mm)) 2751ae798783SMichal Hocko return -EINTR; 2752ae798783SMichal Hocko 2753183654ceSLiam R. Howlett ret = do_vmi_munmap(&vmi, mm, start, len, &uf, downgrade); 2754dd2283f2SYang Shi /* 2755c1e8d7c6SMichel Lespinasse * Returning 1 indicates mmap_lock is downgraded. 2756dd2283f2SYang Shi * But 1 is not legal return value of vm_munmap() and munmap(), reset 2757dd2283f2SYang Shi * it to 0 before return. 2758dd2283f2SYang Shi */ 2759dd2283f2SYang Shi if (ret == 1) { 2760d8ed45c5SMichel Lespinasse mmap_read_unlock(mm); 2761dd2283f2SYang Shi ret = 0; 2762dd2283f2SYang Shi } else 2763d8ed45c5SMichel Lespinasse mmap_write_unlock(mm); 2764dd2283f2SYang Shi 2765897ab3e0SMike Rapoport userfaultfd_unmap_complete(mm, &uf); 2766a46ef99dSLinus Torvalds return ret; 2767a46ef99dSLinus Torvalds } 2768dd2283f2SYang Shi 2769dd2283f2SYang Shi int vm_munmap(unsigned long start, size_t len) 2770dd2283f2SYang Shi { 2771dd2283f2SYang Shi return __vm_munmap(start, len, false); 2772dd2283f2SYang Shi } 2773a46ef99dSLinus Torvalds EXPORT_SYMBOL(vm_munmap); 2774a46ef99dSLinus Torvalds 27756a6160a7SHeiko Carstens SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len) 27761da177e4SLinus Torvalds { 2777ce18d171SCatalin Marinas addr = untagged_addr(addr); 2778dd2283f2SYang Shi return __vm_munmap(addr, len, true); 27791da177e4SLinus Torvalds } 27801da177e4SLinus Torvalds 2781c8d78c18SKirill A. Shutemov 2782c8d78c18SKirill A. Shutemov /* 2783c8d78c18SKirill A. Shutemov * Emulation of deprecated remap_file_pages() syscall. 2784c8d78c18SKirill A. Shutemov */ 2785c8d78c18SKirill A. Shutemov SYSCALL_DEFINE5(remap_file_pages, unsigned long, start, unsigned long, size, 2786c8d78c18SKirill A. Shutemov unsigned long, prot, unsigned long, pgoff, unsigned long, flags) 2787c8d78c18SKirill A. Shutemov { 2788c8d78c18SKirill A. Shutemov 2789c8d78c18SKirill A. Shutemov struct mm_struct *mm = current->mm; 2790c8d78c18SKirill A. Shutemov struct vm_area_struct *vma; 2791c8d78c18SKirill A. Shutemov unsigned long populate = 0; 2792c8d78c18SKirill A. Shutemov unsigned long ret = -EINVAL; 2793c8d78c18SKirill A. Shutemov struct file *file; 2794c8d78c18SKirill A. Shutemov 2795ee65728eSMike Rapoport pr_warn_once("%s (%d) uses deprecated remap_file_pages() syscall. See Documentation/mm/remap_file_pages.rst.\n", 2796c8d78c18SKirill A. Shutemov current->comm, current->pid); 2797c8d78c18SKirill A. Shutemov 2798c8d78c18SKirill A. Shutemov if (prot) 2799c8d78c18SKirill A. Shutemov return ret; 2800c8d78c18SKirill A. Shutemov start = start & PAGE_MASK; 2801c8d78c18SKirill A. Shutemov size = size & PAGE_MASK; 2802c8d78c18SKirill A. Shutemov 2803c8d78c18SKirill A. Shutemov if (start + size <= start) 2804c8d78c18SKirill A. Shutemov return ret; 2805c8d78c18SKirill A. Shutemov 2806c8d78c18SKirill A. Shutemov /* Does pgoff wrap? */ 2807c8d78c18SKirill A. Shutemov if (pgoff + (size >> PAGE_SHIFT) < pgoff) 2808c8d78c18SKirill A. Shutemov return ret; 2809c8d78c18SKirill A. Shutemov 2810d8ed45c5SMichel Lespinasse if (mmap_write_lock_killable(mm)) 2811dc0ef0dfSMichal Hocko return -EINTR; 2812dc0ef0dfSMichal Hocko 28139b593cb2SLiam R. Howlett vma = vma_lookup(mm, start); 2814c8d78c18SKirill A. Shutemov 2815c8d78c18SKirill A. Shutemov if (!vma || !(vma->vm_flags & VM_SHARED)) 2816c8d78c18SKirill A. Shutemov goto out; 2817c8d78c18SKirill A. Shutemov 281848f7df32SKirill A. Shutemov if (start + size > vma->vm_end) { 2819763ecb03SLiam R. Howlett VMA_ITERATOR(vmi, mm, vma->vm_end); 2820763ecb03SLiam R. Howlett struct vm_area_struct *next, *prev = vma; 282148f7df32SKirill A. Shutemov 2822763ecb03SLiam R. Howlett for_each_vma_range(vmi, next, start + size) { 282348f7df32SKirill A. Shutemov /* hole between vmas ? */ 2824763ecb03SLiam R. Howlett if (next->vm_start != prev->vm_end) 282548f7df32SKirill A. Shutemov goto out; 282648f7df32SKirill A. Shutemov 282748f7df32SKirill A. Shutemov if (next->vm_file != vma->vm_file) 282848f7df32SKirill A. Shutemov goto out; 282948f7df32SKirill A. Shutemov 283048f7df32SKirill A. Shutemov if (next->vm_flags != vma->vm_flags) 283148f7df32SKirill A. Shutemov goto out; 283248f7df32SKirill A. Shutemov 28331db43d3fSLiam Howlett if (start + size <= next->vm_end) 28341db43d3fSLiam Howlett break; 28351db43d3fSLiam Howlett 2836763ecb03SLiam R. Howlett prev = next; 283748f7df32SKirill A. Shutemov } 283848f7df32SKirill A. Shutemov 283948f7df32SKirill A. Shutemov if (!next) 2840c8d78c18SKirill A. Shutemov goto out; 2841c8d78c18SKirill A. Shutemov } 2842c8d78c18SKirill A. Shutemov 2843c8d78c18SKirill A. Shutemov prot |= vma->vm_flags & VM_READ ? PROT_READ : 0; 2844c8d78c18SKirill A. Shutemov prot |= vma->vm_flags & VM_WRITE ? PROT_WRITE : 0; 2845c8d78c18SKirill A. Shutemov prot |= vma->vm_flags & VM_EXEC ? PROT_EXEC : 0; 2846c8d78c18SKirill A. Shutemov 2847c8d78c18SKirill A. Shutemov flags &= MAP_NONBLOCK; 2848c8d78c18SKirill A. Shutemov flags |= MAP_SHARED | MAP_FIXED | MAP_POPULATE; 2849fce000b1SLiam Howlett if (vma->vm_flags & VM_LOCKED) 2850c8d78c18SKirill A. Shutemov flags |= MAP_LOCKED; 285148f7df32SKirill A. Shutemov 2852c8d78c18SKirill A. Shutemov file = get_file(vma->vm_file); 285345e55300SPeter Collingbourne ret = do_mmap(vma->vm_file, start, size, 2854897ab3e0SMike Rapoport prot, flags, pgoff, &populate, NULL); 2855c8d78c18SKirill A. Shutemov fput(file); 2856c8d78c18SKirill A. Shutemov out: 2857d8ed45c5SMichel Lespinasse mmap_write_unlock(mm); 2858c8d78c18SKirill A. Shutemov if (populate) 2859c8d78c18SKirill A. Shutemov mm_populate(ret, populate); 2860c8d78c18SKirill A. Shutemov if (!IS_ERR_VALUE(ret)) 2861c8d78c18SKirill A. Shutemov ret = 0; 2862c8d78c18SKirill A. Shutemov return ret; 2863c8d78c18SKirill A. Shutemov } 2864c8d78c18SKirill A. Shutemov 28651da177e4SLinus Torvalds /* 286627b26701SLiam R. Howlett * do_vma_munmap() - Unmap a full or partial vma. 286727b26701SLiam R. Howlett * @vmi: The vma iterator pointing at the vma 286827b26701SLiam R. Howlett * @vma: The first vma to be munmapped 286927b26701SLiam R. Howlett * @start: the start of the address to unmap 287027b26701SLiam R. Howlett * @end: The end of the address to unmap 28712e7ce7d3SLiam R. Howlett * @uf: The userfaultfd list_head 287227b26701SLiam R. Howlett * @downgrade: Attempt to downgrade or not 28732e7ce7d3SLiam R. Howlett * 287427b26701SLiam R. Howlett * Returns: 0 on success and not downgraded, 1 on success and downgraded. 287527b26701SLiam R. Howlett * unmaps a VMA mapping when the vma iterator is already in position. 287627b26701SLiam R. Howlett * Does not handle alignment. 28771da177e4SLinus Torvalds */ 287827b26701SLiam R. Howlett int do_vma_munmap(struct vma_iterator *vmi, struct vm_area_struct *vma, 287927b26701SLiam R. Howlett unsigned long start, unsigned long end, 288027b26701SLiam R. Howlett struct list_head *uf, bool downgrade) 28812e7ce7d3SLiam R. Howlett { 28822e7ce7d3SLiam R. Howlett struct mm_struct *mm = vma->vm_mm; 28832e7ce7d3SLiam R. Howlett int ret; 28842e7ce7d3SLiam R. Howlett 288527b26701SLiam R. Howlett arch_unmap(mm, start, end); 288627b26701SLiam R. Howlett ret = do_vmi_align_munmap(vmi, vma, mm, start, end, uf, downgrade); 28872e7ce7d3SLiam R. Howlett validate_mm_mt(mm); 28882e7ce7d3SLiam R. Howlett return ret; 28892e7ce7d3SLiam R. Howlett } 28902e7ce7d3SLiam R. Howlett 28912e7ce7d3SLiam R. Howlett /* 28922e7ce7d3SLiam R. Howlett * do_brk_flags() - Increase the brk vma if the flags match. 289392fed820SLiam R. Howlett * @vmi: The vma iterator 28942e7ce7d3SLiam R. Howlett * @addr: The start address 28952e7ce7d3SLiam R. Howlett * @len: The length of the increase 28962e7ce7d3SLiam R. Howlett * @vma: The vma, 28972e7ce7d3SLiam R. Howlett * @flags: The VMA Flags 28982e7ce7d3SLiam R. Howlett * 28992e7ce7d3SLiam R. Howlett * Extend the brk VMA from addr to addr + len. If the VMA is NULL or the flags 29002e7ce7d3SLiam R. Howlett * do not match then create a new anonymous VMA. Eventually we may be able to 29012e7ce7d3SLiam R. Howlett * do some brk-specific accounting here. 29022e7ce7d3SLiam R. Howlett */ 290392fed820SLiam R. Howlett static int do_brk_flags(struct vma_iterator *vmi, struct vm_area_struct *vma, 2904763ecb03SLiam R. Howlett unsigned long addr, unsigned long len, unsigned long flags) 29051da177e4SLinus Torvalds { 29061da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 29072e7ce7d3SLiam R. Howlett 2908d4af56c5SLiam R. Howlett validate_mm_mt(mm); 29092e7ce7d3SLiam R. Howlett /* 29102e7ce7d3SLiam R. Howlett * Check against address space limits by the changed size 29112e7ce7d3SLiam R. Howlett * Note: This happens *after* clearing old mappings in some code paths. 29122e7ce7d3SLiam R. Howlett */ 291316e72e9bSDenys Vlasenko flags |= VM_DATA_DEFAULT_FLAGS | VM_ACCOUNT | mm->def_flags; 291484638335SKonstantin Khlebnikov if (!may_expand_vm(mm, flags, len >> PAGE_SHIFT)) 29151da177e4SLinus Torvalds return -ENOMEM; 29161da177e4SLinus Torvalds 29171da177e4SLinus Torvalds if (mm->map_count > sysctl_max_map_count) 29181da177e4SLinus Torvalds return -ENOMEM; 29191da177e4SLinus Torvalds 2920191c5424SAl Viro if (security_vm_enough_memory_mm(mm, len >> PAGE_SHIFT)) 29211da177e4SLinus Torvalds return -ENOMEM; 29221da177e4SLinus Torvalds 29231da177e4SLinus Torvalds /* 29242e7ce7d3SLiam R. Howlett * Expand the existing vma if possible; Note that singular lists do not 29252e7ce7d3SLiam R. Howlett * occur after forking, so the expand will only happen on new VMAs. 29261da177e4SLinus Torvalds */ 29276c28ca64SLiam Howlett if (vma && vma->vm_end == addr && !vma_policy(vma) && 29286c28ca64SLiam Howlett can_vma_merge_after(vma, flags, NULL, NULL, 29296c28ca64SLiam Howlett addr >> PAGE_SHIFT, NULL_VM_UFFD_CTX, NULL)) { 293092fed820SLiam R. Howlett if (vma_iter_prealloc(vmi)) 2931675eaca1SAlistair Popple goto unacct_fail; 293228c5609fSLiam Howlett 293328c5609fSLiam Howlett vma_adjust_trans_huge(vma, vma->vm_start, addr + len, 0); 29342e7ce7d3SLiam R. Howlett if (vma->anon_vma) { 29352e7ce7d3SLiam R. Howlett anon_vma_lock_write(vma->anon_vma); 29362e7ce7d3SLiam R. Howlett anon_vma_interval_tree_pre_update_vma(vma); 29371da177e4SLinus Torvalds } 29382e7ce7d3SLiam R. Howlett vma->vm_end = addr + len; 29392e7ce7d3SLiam R. Howlett vma->vm_flags |= VM_SOFTDIRTY; 294092fed820SLiam R. Howlett vma_iter_store(vmi, vma); 29412e7ce7d3SLiam R. Howlett 29422e7ce7d3SLiam R. Howlett if (vma->anon_vma) { 29432e7ce7d3SLiam R. Howlett anon_vma_interval_tree_post_update_vma(vma); 29442e7ce7d3SLiam R. Howlett anon_vma_unlock_write(vma->anon_vma); 29452e7ce7d3SLiam R. Howlett } 29462e7ce7d3SLiam R. Howlett khugepaged_enter_vma(vma, flags); 29472e7ce7d3SLiam R. Howlett goto out; 29482e7ce7d3SLiam R. Howlett } 29492e7ce7d3SLiam R. Howlett 29502e7ce7d3SLiam R. Howlett /* create a vma struct for an anonymous mapping */ 29512e7ce7d3SLiam R. Howlett vma = vm_area_alloc(mm); 29522e7ce7d3SLiam R. Howlett if (!vma) 2953675eaca1SAlistair Popple goto unacct_fail; 29541da177e4SLinus Torvalds 2955bfd40eafSKirill A. Shutemov vma_set_anonymous(vma); 29561da177e4SLinus Torvalds vma->vm_start = addr; 29571da177e4SLinus Torvalds vma->vm_end = addr + len; 29582e7ce7d3SLiam R. Howlett vma->vm_pgoff = addr >> PAGE_SHIFT; 29591da177e4SLinus Torvalds vma->vm_flags = flags; 29603ed75eb8SColy Li vma->vm_page_prot = vm_get_page_prot(flags); 296192fed820SLiam R. Howlett if (vma_iter_store_gfp(vmi, vma, GFP_KERNEL)) 29622e7ce7d3SLiam R. Howlett goto mas_store_fail; 2963d4af56c5SLiam R. Howlett 29642e7ce7d3SLiam R. Howlett mm->map_count++; 29651da177e4SLinus Torvalds out: 29663af9e859SEric B Munson perf_event_mmap(vma); 29671da177e4SLinus Torvalds mm->total_vm += len >> PAGE_SHIFT; 296884638335SKonstantin Khlebnikov mm->data_vm += len >> PAGE_SHIFT; 2969128557ffSMichel Lespinasse if (flags & VM_LOCKED) 2970ba470de4SRik van Riel mm->locked_vm += (len >> PAGE_SHIFT); 2971d9104d1cSCyrill Gorcunov vma->vm_flags |= VM_SOFTDIRTY; 2972763ecb03SLiam R. Howlett validate_mm(mm); 29735d22fc25SLinus Torvalds return 0; 2974d4af56c5SLiam R. Howlett 29752e7ce7d3SLiam R. Howlett mas_store_fail: 2976d4af56c5SLiam R. Howlett vm_area_free(vma); 2977675eaca1SAlistair Popple unacct_fail: 29782e7ce7d3SLiam R. Howlett vm_unacct_memory(len >> PAGE_SHIFT); 29792e7ce7d3SLiam R. Howlett return -ENOMEM; 29801da177e4SLinus Torvalds } 29811da177e4SLinus Torvalds 2982bb177a73SMichal Hocko int vm_brk_flags(unsigned long addr, unsigned long request, unsigned long flags) 2983e4eb1ff6SLinus Torvalds { 2984e4eb1ff6SLinus Torvalds struct mm_struct *mm = current->mm; 29852e7ce7d3SLiam R. Howlett struct vm_area_struct *vma = NULL; 2986bb177a73SMichal Hocko unsigned long len; 29875d22fc25SLinus Torvalds int ret; 2988128557ffSMichel Lespinasse bool populate; 2989897ab3e0SMike Rapoport LIST_HEAD(uf); 299092fed820SLiam R. Howlett VMA_ITERATOR(vmi, mm, addr); 2991e4eb1ff6SLinus Torvalds 2992bb177a73SMichal Hocko len = PAGE_ALIGN(request); 2993bb177a73SMichal Hocko if (len < request) 2994bb177a73SMichal Hocko return -ENOMEM; 2995bb177a73SMichal Hocko if (!len) 2996bb177a73SMichal Hocko return 0; 2997bb177a73SMichal Hocko 2998d8ed45c5SMichel Lespinasse if (mmap_write_lock_killable(mm)) 29992d6c9282SMichal Hocko return -EINTR; 30002d6c9282SMichal Hocko 30012e7ce7d3SLiam R. Howlett /* Until we need other flags, refuse anything except VM_EXEC. */ 30022e7ce7d3SLiam R. Howlett if ((flags & (~VM_EXEC)) != 0) 30032e7ce7d3SLiam R. Howlett return -EINVAL; 30042e7ce7d3SLiam R. Howlett 30052e7ce7d3SLiam R. Howlett ret = check_brk_limits(addr, len); 30062e7ce7d3SLiam R. Howlett if (ret) 30072e7ce7d3SLiam R. Howlett goto limits_failed; 30082e7ce7d3SLiam R. Howlett 3009183654ceSLiam R. Howlett ret = do_vmi_munmap(&vmi, mm, addr, len, &uf, 0); 30102e7ce7d3SLiam R. Howlett if (ret) 30112e7ce7d3SLiam R. Howlett goto munmap_failed; 30122e7ce7d3SLiam R. Howlett 301392fed820SLiam R. Howlett vma = vma_prev(&vmi); 301492fed820SLiam R. Howlett ret = do_brk_flags(&vmi, vma, addr, len, flags); 3015128557ffSMichel Lespinasse populate = ((mm->def_flags & VM_LOCKED) != 0); 3016d8ed45c5SMichel Lespinasse mmap_write_unlock(mm); 3017897ab3e0SMike Rapoport userfaultfd_unmap_complete(mm, &uf); 30185d22fc25SLinus Torvalds if (populate && !ret) 3019128557ffSMichel Lespinasse mm_populate(addr, len); 3020e4eb1ff6SLinus Torvalds return ret; 30212e7ce7d3SLiam R. Howlett 30222e7ce7d3SLiam R. Howlett munmap_failed: 30232e7ce7d3SLiam R. Howlett limits_failed: 30242e7ce7d3SLiam R. Howlett mmap_write_unlock(mm); 30252e7ce7d3SLiam R. Howlett return ret; 3026e4eb1ff6SLinus Torvalds } 302716e72e9bSDenys Vlasenko EXPORT_SYMBOL(vm_brk_flags); 302816e72e9bSDenys Vlasenko 302916e72e9bSDenys Vlasenko int vm_brk(unsigned long addr, unsigned long len) 303016e72e9bSDenys Vlasenko { 303116e72e9bSDenys Vlasenko return vm_brk_flags(addr, len, 0); 303216e72e9bSDenys Vlasenko } 3033e4eb1ff6SLinus Torvalds EXPORT_SYMBOL(vm_brk); 30341da177e4SLinus Torvalds 30351da177e4SLinus Torvalds /* Release all mmaps. */ 30361da177e4SLinus Torvalds void exit_mmap(struct mm_struct *mm) 30371da177e4SLinus Torvalds { 3038d16dfc55SPeter Zijlstra struct mmu_gather tlb; 3039ba470de4SRik van Riel struct vm_area_struct *vma; 30401da177e4SLinus Torvalds unsigned long nr_accounted = 0; 3041763ecb03SLiam R. Howlett MA_STATE(mas, &mm->mm_mt, 0, 0); 3042763ecb03SLiam R. Howlett int count = 0; 30431da177e4SLinus Torvalds 3044d6dd61c8SJeremy Fitzhardinge /* mm's last user has gone, and its about to be pulled down */ 3045cddb8a5cSAndrea Arcangeli mmu_notifier_release(mm); 3046d6dd61c8SJeremy Fitzhardinge 3047bf3980c8SSuren Baghdasaryan mmap_read_lock(mm); 30489480c53eSJeremy Fitzhardinge arch_exit_mmap(mm); 30499480c53eSJeremy Fitzhardinge 3050763ecb03SLiam R. Howlett vma = mas_find(&mas, ULONG_MAX); 305164591e86SSuren Baghdasaryan if (!vma) { 305264591e86SSuren Baghdasaryan /* Can happen if dup_mmap() received an OOM */ 3053bf3980c8SSuren Baghdasaryan mmap_read_unlock(mm); 30549480c53eSJeremy Fitzhardinge return; 305564591e86SSuren Baghdasaryan } 30569480c53eSJeremy Fitzhardinge 30571da177e4SLinus Torvalds lru_add_drain(); 30581da177e4SLinus Torvalds flush_cache_mm(mm); 3059d8b45053SWill Deacon tlb_gather_mmu_fullmm(&tlb, mm); 3060901608d9SOleg Nesterov /* update_hiwater_rss(mm) here? but nobody should be looking */ 3061763ecb03SLiam R. Howlett /* Use ULONG_MAX here to ensure all VMAs in the mm are unmapped */ 3062763ecb03SLiam R. Howlett unmap_vmas(&tlb, &mm->mm_mt, vma, 0, ULONG_MAX); 3063bf3980c8SSuren Baghdasaryan mmap_read_unlock(mm); 3064bf3980c8SSuren Baghdasaryan 3065bf3980c8SSuren Baghdasaryan /* 3066bf3980c8SSuren Baghdasaryan * Set MMF_OOM_SKIP to hide this task from the oom killer/reaper 3067b3541d91SSuren Baghdasaryan * because the memory has been already freed. 3068bf3980c8SSuren Baghdasaryan */ 3069bf3980c8SSuren Baghdasaryan set_bit(MMF_OOM_SKIP, &mm->flags); 3070bf3980c8SSuren Baghdasaryan mmap_write_lock(mm); 3071763ecb03SLiam R. Howlett free_pgtables(&tlb, &mm->mm_mt, vma, FIRST_USER_ADDRESS, 3072763ecb03SLiam R. Howlett USER_PGTABLES_CEILING); 3073ae8eba8bSWill Deacon tlb_finish_mmu(&tlb); 30741da177e4SLinus Torvalds 3075763ecb03SLiam R. Howlett /* 3076763ecb03SLiam R. Howlett * Walk the list again, actually closing and freeing it, with preemption 3077763ecb03SLiam R. Howlett * enabled, without holding any MM locks besides the unreachable 3078763ecb03SLiam R. Howlett * mmap_write_lock. 3079763ecb03SLiam R. Howlett */ 3080763ecb03SLiam R. Howlett do { 30814f74d2c8SLinus Torvalds if (vma->vm_flags & VM_ACCOUNT) 30824f74d2c8SLinus Torvalds nr_accounted += vma_pages(vma); 3083763ecb03SLiam R. Howlett remove_vma(vma); 3084763ecb03SLiam R. Howlett count++; 30850a3b3c25SPaul E. McKenney cond_resched(); 3086763ecb03SLiam R. Howlett } while ((vma = mas_find(&mas, ULONG_MAX)) != NULL); 3087763ecb03SLiam R. Howlett 3088763ecb03SLiam R. Howlett BUG_ON(count != mm->map_count); 3089d4af56c5SLiam R. Howlett 3090d4af56c5SLiam R. Howlett trace_exit_mmap(mm); 3091d4af56c5SLiam R. Howlett __mt_destroy(&mm->mm_mt); 309264591e86SSuren Baghdasaryan mmap_write_unlock(mm); 30934f74d2c8SLinus Torvalds vm_unacct_memory(nr_accounted); 30941da177e4SLinus Torvalds } 30951da177e4SLinus Torvalds 30961da177e4SLinus Torvalds /* Insert vm structure into process list sorted by address 30971da177e4SLinus Torvalds * and into the inode's i_mmap tree. If vm_file is non-NULL 3098c8c06efaSDavidlohr Bueso * then i_mmap_rwsem is taken here. 30991da177e4SLinus Torvalds */ 31001da177e4SLinus Torvalds int insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma) 31011da177e4SLinus Torvalds { 3102d4af56c5SLiam R. Howlett unsigned long charged = vma_pages(vma); 31031da177e4SLinus Torvalds 3104d4af56c5SLiam R. Howlett 3105d0601a50SLiam R. Howlett if (find_vma_intersection(mm, vma->vm_start, vma->vm_end)) 3106c9d13f5fSChen Gang return -ENOMEM; 3107d4af56c5SLiam R. Howlett 3108c9d13f5fSChen Gang if ((vma->vm_flags & VM_ACCOUNT) && 3109d4af56c5SLiam R. Howlett security_vm_enough_memory_mm(mm, charged)) 3110c9d13f5fSChen Gang return -ENOMEM; 3111c9d13f5fSChen Gang 31121da177e4SLinus Torvalds /* 31131da177e4SLinus Torvalds * The vm_pgoff of a purely anonymous vma should be irrelevant 31141da177e4SLinus Torvalds * until its first write fault, when page's anon_vma and index 31151da177e4SLinus Torvalds * are set. But now set the vm_pgoff it will almost certainly 31161da177e4SLinus Torvalds * end up with (unless mremap moves it elsewhere before that 31171da177e4SLinus Torvalds * first wfault), so /proc/pid/maps tells a consistent story. 31181da177e4SLinus Torvalds * 31191da177e4SLinus Torvalds * By setting it to reflect the virtual start address of the 31201da177e4SLinus Torvalds * vma, merges and splits can happen in a seamless way, just 31211da177e4SLinus Torvalds * using the existing file pgoff checks and manipulations. 31228332326eSLiao Pingfang * Similarly in do_mmap and in do_brk_flags. 31231da177e4SLinus Torvalds */ 31248a9cc3b5SOleg Nesterov if (vma_is_anonymous(vma)) { 31251da177e4SLinus Torvalds BUG_ON(vma->anon_vma); 31261da177e4SLinus Torvalds vma->vm_pgoff = vma->vm_start >> PAGE_SHIFT; 31271da177e4SLinus Torvalds } 31282b144498SSrikar Dronamraju 3129763ecb03SLiam R. Howlett if (vma_link(mm, vma)) { 3130d4af56c5SLiam R. Howlett vm_unacct_memory(charged); 3131d4af56c5SLiam R. Howlett return -ENOMEM; 3132d4af56c5SLiam R. Howlett } 3133d4af56c5SLiam R. Howlett 31341da177e4SLinus Torvalds return 0; 31351da177e4SLinus Torvalds } 31361da177e4SLinus Torvalds 31371da177e4SLinus Torvalds /* 31381da177e4SLinus Torvalds * Copy the vma structure to a new location in the same mm, 31391da177e4SLinus Torvalds * prior to moving page table entries, to effect an mremap move. 31401da177e4SLinus Torvalds */ 31411da177e4SLinus Torvalds struct vm_area_struct *copy_vma(struct vm_area_struct **vmap, 314238a76013SMichel Lespinasse unsigned long addr, unsigned long len, pgoff_t pgoff, 314338a76013SMichel Lespinasse bool *need_rmap_locks) 31441da177e4SLinus Torvalds { 31451da177e4SLinus Torvalds struct vm_area_struct *vma = *vmap; 31461da177e4SLinus Torvalds unsigned long vma_start = vma->vm_start; 31471da177e4SLinus Torvalds struct mm_struct *mm = vma->vm_mm; 31481da177e4SLinus Torvalds struct vm_area_struct *new_vma, *prev; 3149948f017bSAndrea Arcangeli bool faulted_in_anon_vma = true; 3150076f16bfSLiam R. Howlett VMA_ITERATOR(vmi, mm, addr); 31511da177e4SLinus Torvalds 3152d4af56c5SLiam R. Howlett validate_mm_mt(mm); 31531da177e4SLinus Torvalds /* 31541da177e4SLinus Torvalds * If anonymous vma has not yet been faulted, update new pgoff 31551da177e4SLinus Torvalds * to match new location, to increase its chance of merging. 31561da177e4SLinus Torvalds */ 3157ce75799bSOleg Nesterov if (unlikely(vma_is_anonymous(vma) && !vma->anon_vma)) { 31581da177e4SLinus Torvalds pgoff = addr >> PAGE_SHIFT; 3159948f017bSAndrea Arcangeli faulted_in_anon_vma = false; 3160948f017bSAndrea Arcangeli } 31611da177e4SLinus Torvalds 3162763ecb03SLiam R. Howlett new_vma = find_vma_prev(mm, addr, &prev); 3163763ecb03SLiam R. Howlett if (new_vma && new_vma->vm_start < addr + len) 31646597d783SHugh Dickins return NULL; /* should never get here */ 3165524e00b3SLiam R. Howlett 31669760ebffSLiam R. Howlett new_vma = vma_merge(&vmi, mm, prev, addr, addr + len, vma->vm_flags, 316719a809afSAndrea Arcangeli vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma), 31685c26f6acSSuren Baghdasaryan vma->vm_userfaultfd_ctx, anon_vma_name(vma)); 31691da177e4SLinus Torvalds if (new_vma) { 31701da177e4SLinus Torvalds /* 31711da177e4SLinus Torvalds * Source vma may have been merged into new_vma 31721da177e4SLinus Torvalds */ 3173948f017bSAndrea Arcangeli if (unlikely(vma_start >= new_vma->vm_start && 3174948f017bSAndrea Arcangeli vma_start < new_vma->vm_end)) { 3175948f017bSAndrea Arcangeli /* 3176948f017bSAndrea Arcangeli * The only way we can get a vma_merge with 3177948f017bSAndrea Arcangeli * self during an mremap is if the vma hasn't 3178948f017bSAndrea Arcangeli * been faulted in yet and we were allowed to 3179948f017bSAndrea Arcangeli * reset the dst vma->vm_pgoff to the 3180948f017bSAndrea Arcangeli * destination address of the mremap to allow 3181948f017bSAndrea Arcangeli * the merge to happen. mremap must change the 3182948f017bSAndrea Arcangeli * vm_pgoff linearity between src and dst vmas 3183948f017bSAndrea Arcangeli * (in turn preventing a vma_merge) to be 3184948f017bSAndrea Arcangeli * safe. It is only safe to keep the vm_pgoff 3185948f017bSAndrea Arcangeli * linear if there are no pages mapped yet. 3186948f017bSAndrea Arcangeli */ 318781d1b09cSSasha Levin VM_BUG_ON_VMA(faulted_in_anon_vma, new_vma); 318838a76013SMichel Lespinasse *vmap = vma = new_vma; 3189108d6642SMichel Lespinasse } 319038a76013SMichel Lespinasse *need_rmap_locks = (new_vma->vm_pgoff <= vma->vm_pgoff); 31911da177e4SLinus Torvalds } else { 31923928d4f5SLinus Torvalds new_vma = vm_area_dup(vma); 3193e3975891SChen Gang if (!new_vma) 3194e3975891SChen Gang goto out; 31951da177e4SLinus Torvalds new_vma->vm_start = addr; 31961da177e4SLinus Torvalds new_vma->vm_end = addr + len; 31971da177e4SLinus Torvalds new_vma->vm_pgoff = pgoff; 3198ef0855d3SOleg Nesterov if (vma_dup_policy(vma, new_vma)) 3199523d4e20SMichel Lespinasse goto out_free_vma; 3200523d4e20SMichel Lespinasse if (anon_vma_clone(new_vma, vma)) 3201523d4e20SMichel Lespinasse goto out_free_mempol; 3202e9714acfSKonstantin Khlebnikov if (new_vma->vm_file) 32031da177e4SLinus Torvalds get_file(new_vma->vm_file); 32041da177e4SLinus Torvalds if (new_vma->vm_ops && new_vma->vm_ops->open) 32051da177e4SLinus Torvalds new_vma->vm_ops->open(new_vma); 3206763ecb03SLiam R. Howlett if (vma_link(mm, new_vma)) 3207524e00b3SLiam R. Howlett goto out_vma_link; 320838a76013SMichel Lespinasse *need_rmap_locks = false; 32091da177e4SLinus Torvalds } 3210d4af56c5SLiam R. Howlett validate_mm_mt(mm); 32111da177e4SLinus Torvalds return new_vma; 32125beb4930SRik van Riel 3213524e00b3SLiam R. Howlett out_vma_link: 3214524e00b3SLiam R. Howlett if (new_vma->vm_ops && new_vma->vm_ops->close) 3215524e00b3SLiam R. Howlett new_vma->vm_ops->close(new_vma); 321692b73996SLiam Howlett 321792b73996SLiam Howlett if (new_vma->vm_file) 321892b73996SLiam Howlett fput(new_vma->vm_file); 321992b73996SLiam Howlett 322092b73996SLiam Howlett unlink_anon_vmas(new_vma); 32215beb4930SRik van Riel out_free_mempol: 3222ef0855d3SOleg Nesterov mpol_put(vma_policy(new_vma)); 32235beb4930SRik van Riel out_free_vma: 32243928d4f5SLinus Torvalds vm_area_free(new_vma); 3225e3975891SChen Gang out: 3226d4af56c5SLiam R. Howlett validate_mm_mt(mm); 32275beb4930SRik van Riel return NULL; 32281da177e4SLinus Torvalds } 3229119f657cSakpm@osdl.org 3230119f657cSakpm@osdl.org /* 3231119f657cSakpm@osdl.org * Return true if the calling process may expand its vm space by the passed 3232119f657cSakpm@osdl.org * number of pages 3233119f657cSakpm@osdl.org */ 323484638335SKonstantin Khlebnikov bool may_expand_vm(struct mm_struct *mm, vm_flags_t flags, unsigned long npages) 3235119f657cSakpm@osdl.org { 323684638335SKonstantin Khlebnikov if (mm->total_vm + npages > rlimit(RLIMIT_AS) >> PAGE_SHIFT) 323784638335SKonstantin Khlebnikov return false; 3238119f657cSakpm@osdl.org 3239d977d56cSKonstantin Khlebnikov if (is_data_mapping(flags) && 3240d977d56cSKonstantin Khlebnikov mm->data_vm + npages > rlimit(RLIMIT_DATA) >> PAGE_SHIFT) { 3241f4fcd558SKonstantin Khlebnikov /* Workaround for Valgrind */ 3242f4fcd558SKonstantin Khlebnikov if (rlimit(RLIMIT_DATA) == 0 && 3243f4fcd558SKonstantin Khlebnikov mm->data_vm + npages <= rlimit_max(RLIMIT_DATA) >> PAGE_SHIFT) 3244f4fcd558SKonstantin Khlebnikov return true; 324557a7702bSDavid Woodhouse 324657a7702bSDavid Woodhouse pr_warn_once("%s (%d): VmData %lu exceed data ulimit %lu. Update limits%s.\n", 3247d977d56cSKonstantin Khlebnikov current->comm, current->pid, 3248d977d56cSKonstantin Khlebnikov (mm->data_vm + npages) << PAGE_SHIFT, 324957a7702bSDavid Woodhouse rlimit(RLIMIT_DATA), 325057a7702bSDavid Woodhouse ignore_rlimit_data ? "" : " or use boot option ignore_rlimit_data"); 325157a7702bSDavid Woodhouse 325257a7702bSDavid Woodhouse if (!ignore_rlimit_data) 3253d977d56cSKonstantin Khlebnikov return false; 3254d977d56cSKonstantin Khlebnikov } 3255119f657cSakpm@osdl.org 325684638335SKonstantin Khlebnikov return true; 325784638335SKonstantin Khlebnikov } 325884638335SKonstantin Khlebnikov 325984638335SKonstantin Khlebnikov void vm_stat_account(struct mm_struct *mm, vm_flags_t flags, long npages) 326084638335SKonstantin Khlebnikov { 32617866076bSPeng Liu WRITE_ONCE(mm->total_vm, READ_ONCE(mm->total_vm)+npages); 326284638335SKonstantin Khlebnikov 3263d977d56cSKonstantin Khlebnikov if (is_exec_mapping(flags)) 326484638335SKonstantin Khlebnikov mm->exec_vm += npages; 3265d977d56cSKonstantin Khlebnikov else if (is_stack_mapping(flags)) 326684638335SKonstantin Khlebnikov mm->stack_vm += npages; 3267d977d56cSKonstantin Khlebnikov else if (is_data_mapping(flags)) 326884638335SKonstantin Khlebnikov mm->data_vm += npages; 3269119f657cSakpm@osdl.org } 3270fa5dc22fSRoland McGrath 3271b3ec9f33SSouptick Joarder static vm_fault_t special_mapping_fault(struct vm_fault *vmf); 3272a62c34bdSAndy Lutomirski 3273a62c34bdSAndy Lutomirski /* 3274a62c34bdSAndy Lutomirski * Having a close hook prevents vma merging regardless of flags. 3275a62c34bdSAndy Lutomirski */ 3276a62c34bdSAndy Lutomirski static void special_mapping_close(struct vm_area_struct *vma) 3277a62c34bdSAndy Lutomirski { 3278a62c34bdSAndy Lutomirski } 3279a62c34bdSAndy Lutomirski 3280a62c34bdSAndy Lutomirski static const char *special_mapping_name(struct vm_area_struct *vma) 3281a62c34bdSAndy Lutomirski { 3282a62c34bdSAndy Lutomirski return ((struct vm_special_mapping *)vma->vm_private_data)->name; 3283a62c34bdSAndy Lutomirski } 3284a62c34bdSAndy Lutomirski 328514d07113SBrian Geffon static int special_mapping_mremap(struct vm_area_struct *new_vma) 3286b059a453SDmitry Safonov { 3287b059a453SDmitry Safonov struct vm_special_mapping *sm = new_vma->vm_private_data; 3288b059a453SDmitry Safonov 3289280e87e9SDmitry Safonov if (WARN_ON_ONCE(current->mm != new_vma->vm_mm)) 3290280e87e9SDmitry Safonov return -EFAULT; 3291280e87e9SDmitry Safonov 3292b059a453SDmitry Safonov if (sm->mremap) 3293b059a453SDmitry Safonov return sm->mremap(sm, new_vma); 3294280e87e9SDmitry Safonov 3295b059a453SDmitry Safonov return 0; 3296b059a453SDmitry Safonov } 3297b059a453SDmitry Safonov 3298871402e0SDmitry Safonov static int special_mapping_split(struct vm_area_struct *vma, unsigned long addr) 3299871402e0SDmitry Safonov { 3300871402e0SDmitry Safonov /* 3301871402e0SDmitry Safonov * Forbid splitting special mappings - kernel has expectations over 3302871402e0SDmitry Safonov * the number of pages in mapping. Together with VM_DONTEXPAND 3303871402e0SDmitry Safonov * the size of vma should stay the same over the special mapping's 3304871402e0SDmitry Safonov * lifetime. 3305871402e0SDmitry Safonov */ 3306871402e0SDmitry Safonov return -EINVAL; 3307871402e0SDmitry Safonov } 3308871402e0SDmitry Safonov 3309a62c34bdSAndy Lutomirski static const struct vm_operations_struct special_mapping_vmops = { 3310a62c34bdSAndy Lutomirski .close = special_mapping_close, 3311a62c34bdSAndy Lutomirski .fault = special_mapping_fault, 3312b059a453SDmitry Safonov .mremap = special_mapping_mremap, 3313a62c34bdSAndy Lutomirski .name = special_mapping_name, 3314af34ebebSDmitry Safonov /* vDSO code relies that VVAR can't be accessed remotely */ 3315af34ebebSDmitry Safonov .access = NULL, 3316871402e0SDmitry Safonov .may_split = special_mapping_split, 3317a62c34bdSAndy Lutomirski }; 3318a62c34bdSAndy Lutomirski 3319a62c34bdSAndy Lutomirski static const struct vm_operations_struct legacy_special_mapping_vmops = { 3320a62c34bdSAndy Lutomirski .close = special_mapping_close, 3321a62c34bdSAndy Lutomirski .fault = special_mapping_fault, 3322a62c34bdSAndy Lutomirski }; 3323fa5dc22fSRoland McGrath 3324b3ec9f33SSouptick Joarder static vm_fault_t special_mapping_fault(struct vm_fault *vmf) 3325fa5dc22fSRoland McGrath { 332611bac800SDave Jiang struct vm_area_struct *vma = vmf->vma; 3327b1d0e4f5SNick Piggin pgoff_t pgoff; 3328fa5dc22fSRoland McGrath struct page **pages; 3329fa5dc22fSRoland McGrath 3330f872f540SAndy Lutomirski if (vma->vm_ops == &legacy_special_mapping_vmops) { 3331a62c34bdSAndy Lutomirski pages = vma->vm_private_data; 3332f872f540SAndy Lutomirski } else { 3333f872f540SAndy Lutomirski struct vm_special_mapping *sm = vma->vm_private_data; 3334f872f540SAndy Lutomirski 3335f872f540SAndy Lutomirski if (sm->fault) 333611bac800SDave Jiang return sm->fault(sm, vmf->vma, vmf); 3337f872f540SAndy Lutomirski 3338f872f540SAndy Lutomirski pages = sm->pages; 3339f872f540SAndy Lutomirski } 3340a62c34bdSAndy Lutomirski 33418a9cc3b5SOleg Nesterov for (pgoff = vmf->pgoff; pgoff && *pages; ++pages) 3342b1d0e4f5SNick Piggin pgoff--; 3343fa5dc22fSRoland McGrath 3344fa5dc22fSRoland McGrath if (*pages) { 3345fa5dc22fSRoland McGrath struct page *page = *pages; 3346fa5dc22fSRoland McGrath get_page(page); 3347b1d0e4f5SNick Piggin vmf->page = page; 3348b1d0e4f5SNick Piggin return 0; 3349fa5dc22fSRoland McGrath } 3350fa5dc22fSRoland McGrath 3351b1d0e4f5SNick Piggin return VM_FAULT_SIGBUS; 3352fa5dc22fSRoland McGrath } 3353fa5dc22fSRoland McGrath 3354a62c34bdSAndy Lutomirski static struct vm_area_struct *__install_special_mapping( 3355a62c34bdSAndy Lutomirski struct mm_struct *mm, 3356fa5dc22fSRoland McGrath unsigned long addr, unsigned long len, 335727f28b97SChen Gang unsigned long vm_flags, void *priv, 335827f28b97SChen Gang const struct vm_operations_struct *ops) 3359fa5dc22fSRoland McGrath { 3360462e635eSTavis Ormandy int ret; 3361fa5dc22fSRoland McGrath struct vm_area_struct *vma; 3362fa5dc22fSRoland McGrath 3363d4af56c5SLiam R. Howlett validate_mm_mt(mm); 3364490fc053SLinus Torvalds vma = vm_area_alloc(mm); 3365fa5dc22fSRoland McGrath if (unlikely(vma == NULL)) 33663935ed6aSStefani Seibold return ERR_PTR(-ENOMEM); 3367fa5dc22fSRoland McGrath 3368fa5dc22fSRoland McGrath vma->vm_start = addr; 3369fa5dc22fSRoland McGrath vma->vm_end = addr + len; 3370fa5dc22fSRoland McGrath 3371d9104d1cSCyrill Gorcunov vma->vm_flags = vm_flags | mm->def_flags | VM_DONTEXPAND | VM_SOFTDIRTY; 33721fc09228SHugh Dickins vma->vm_flags &= VM_LOCKED_CLEAR_MASK; 33733ed75eb8SColy Li vma->vm_page_prot = vm_get_page_prot(vma->vm_flags); 3374fa5dc22fSRoland McGrath 3375a62c34bdSAndy Lutomirski vma->vm_ops = ops; 3376a62c34bdSAndy Lutomirski vma->vm_private_data = priv; 3377fa5dc22fSRoland McGrath 3378462e635eSTavis Ormandy ret = insert_vm_struct(mm, vma); 3379462e635eSTavis Ormandy if (ret) 3380462e635eSTavis Ormandy goto out; 3381fa5dc22fSRoland McGrath 338284638335SKonstantin Khlebnikov vm_stat_account(mm, vma->vm_flags, len >> PAGE_SHIFT); 3383fa5dc22fSRoland McGrath 3384cdd6c482SIngo Molnar perf_event_mmap(vma); 3385089dd79dSPeter Zijlstra 3386d4af56c5SLiam R. Howlett validate_mm_mt(mm); 33873935ed6aSStefani Seibold return vma; 3388462e635eSTavis Ormandy 3389462e635eSTavis Ormandy out: 33903928d4f5SLinus Torvalds vm_area_free(vma); 3391d4af56c5SLiam R. Howlett validate_mm_mt(mm); 33923935ed6aSStefani Seibold return ERR_PTR(ret); 33933935ed6aSStefani Seibold } 33943935ed6aSStefani Seibold 33952eefd878SDmitry Safonov bool vma_is_special_mapping(const struct vm_area_struct *vma, 33962eefd878SDmitry Safonov const struct vm_special_mapping *sm) 33972eefd878SDmitry Safonov { 33982eefd878SDmitry Safonov return vma->vm_private_data == sm && 33992eefd878SDmitry Safonov (vma->vm_ops == &special_mapping_vmops || 34002eefd878SDmitry Safonov vma->vm_ops == &legacy_special_mapping_vmops); 34012eefd878SDmitry Safonov } 34022eefd878SDmitry Safonov 3403a62c34bdSAndy Lutomirski /* 3404c1e8d7c6SMichel Lespinasse * Called with mm->mmap_lock held for writing. 3405a62c34bdSAndy Lutomirski * Insert a new vma covering the given region, with the given flags. 3406a62c34bdSAndy Lutomirski * Its pages are supplied by the given array of struct page *. 3407a62c34bdSAndy Lutomirski * The array can be shorter than len >> PAGE_SHIFT if it's null-terminated. 3408a62c34bdSAndy Lutomirski * The region past the last page supplied will always produce SIGBUS. 3409a62c34bdSAndy Lutomirski * The array pointer and the pages it points to are assumed to stay alive 3410a62c34bdSAndy Lutomirski * for as long as this mapping might exist. 3411a62c34bdSAndy Lutomirski */ 3412a62c34bdSAndy Lutomirski struct vm_area_struct *_install_special_mapping( 3413a62c34bdSAndy Lutomirski struct mm_struct *mm, 3414a62c34bdSAndy Lutomirski unsigned long addr, unsigned long len, 3415a62c34bdSAndy Lutomirski unsigned long vm_flags, const struct vm_special_mapping *spec) 3416a62c34bdSAndy Lutomirski { 341727f28b97SChen Gang return __install_special_mapping(mm, addr, len, vm_flags, (void *)spec, 341827f28b97SChen Gang &special_mapping_vmops); 3419a62c34bdSAndy Lutomirski } 3420a62c34bdSAndy Lutomirski 34213935ed6aSStefani Seibold int install_special_mapping(struct mm_struct *mm, 34223935ed6aSStefani Seibold unsigned long addr, unsigned long len, 34233935ed6aSStefani Seibold unsigned long vm_flags, struct page **pages) 34243935ed6aSStefani Seibold { 3425a62c34bdSAndy Lutomirski struct vm_area_struct *vma = __install_special_mapping( 342627f28b97SChen Gang mm, addr, len, vm_flags, (void *)pages, 342727f28b97SChen Gang &legacy_special_mapping_vmops); 34283935ed6aSStefani Seibold 342914bd5b45SDuan Jiong return PTR_ERR_OR_ZERO(vma); 3430fa5dc22fSRoland McGrath } 34317906d00cSAndrea Arcangeli 34327906d00cSAndrea Arcangeli static DEFINE_MUTEX(mm_all_locks_mutex); 34337906d00cSAndrea Arcangeli 3434454ed842SPeter Zijlstra static void vm_lock_anon_vma(struct mm_struct *mm, struct anon_vma *anon_vma) 34357906d00cSAndrea Arcangeli { 3436f808c13fSDavidlohr Bueso if (!test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_root.rb_node)) { 34377906d00cSAndrea Arcangeli /* 34387906d00cSAndrea Arcangeli * The LSB of head.next can't change from under us 34397906d00cSAndrea Arcangeli * because we hold the mm_all_locks_mutex. 34407906d00cSAndrea Arcangeli */ 3441da1c55f1SMichel Lespinasse down_write_nest_lock(&anon_vma->root->rwsem, &mm->mmap_lock); 34427906d00cSAndrea Arcangeli /* 34437906d00cSAndrea Arcangeli * We can safely modify head.next after taking the 34445a505085SIngo Molnar * anon_vma->root->rwsem. If some other vma in this mm shares 34457906d00cSAndrea Arcangeli * the same anon_vma we won't take it again. 34467906d00cSAndrea Arcangeli * 34477906d00cSAndrea Arcangeli * No need of atomic instructions here, head.next 34487906d00cSAndrea Arcangeli * can't change from under us thanks to the 34495a505085SIngo Molnar * anon_vma->root->rwsem. 34507906d00cSAndrea Arcangeli */ 34517906d00cSAndrea Arcangeli if (__test_and_set_bit(0, (unsigned long *) 3452f808c13fSDavidlohr Bueso &anon_vma->root->rb_root.rb_root.rb_node)) 34537906d00cSAndrea Arcangeli BUG(); 34547906d00cSAndrea Arcangeli } 34557906d00cSAndrea Arcangeli } 34567906d00cSAndrea Arcangeli 3457454ed842SPeter Zijlstra static void vm_lock_mapping(struct mm_struct *mm, struct address_space *mapping) 34587906d00cSAndrea Arcangeli { 34597906d00cSAndrea Arcangeli if (!test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) { 34607906d00cSAndrea Arcangeli /* 34617906d00cSAndrea Arcangeli * AS_MM_ALL_LOCKS can't change from under us because 34627906d00cSAndrea Arcangeli * we hold the mm_all_locks_mutex. 34637906d00cSAndrea Arcangeli * 34647906d00cSAndrea Arcangeli * Operations on ->flags have to be atomic because 34657906d00cSAndrea Arcangeli * even if AS_MM_ALL_LOCKS is stable thanks to the 34667906d00cSAndrea Arcangeli * mm_all_locks_mutex, there may be other cpus 34677906d00cSAndrea Arcangeli * changing other bitflags in parallel to us. 34687906d00cSAndrea Arcangeli */ 34697906d00cSAndrea Arcangeli if (test_and_set_bit(AS_MM_ALL_LOCKS, &mapping->flags)) 34707906d00cSAndrea Arcangeli BUG(); 3471da1c55f1SMichel Lespinasse down_write_nest_lock(&mapping->i_mmap_rwsem, &mm->mmap_lock); 34727906d00cSAndrea Arcangeli } 34737906d00cSAndrea Arcangeli } 34747906d00cSAndrea Arcangeli 34757906d00cSAndrea Arcangeli /* 34767906d00cSAndrea Arcangeli * This operation locks against the VM for all pte/vma/mm related 34777906d00cSAndrea Arcangeli * operations that could ever happen on a certain mm. This includes 34787906d00cSAndrea Arcangeli * vmtruncate, try_to_unmap, and all page faults. 34797906d00cSAndrea Arcangeli * 3480c1e8d7c6SMichel Lespinasse * The caller must take the mmap_lock in write mode before calling 34817906d00cSAndrea Arcangeli * mm_take_all_locks(). The caller isn't allowed to release the 3482c1e8d7c6SMichel Lespinasse * mmap_lock until mm_drop_all_locks() returns. 34837906d00cSAndrea Arcangeli * 3484c1e8d7c6SMichel Lespinasse * mmap_lock in write mode is required in order to block all operations 34857906d00cSAndrea Arcangeli * that could modify pagetables and free pages without need of 348627ba0644SKirill A. Shutemov * altering the vma layout. It's also needed in write mode to avoid new 34877906d00cSAndrea Arcangeli * anon_vmas to be associated with existing vmas. 34887906d00cSAndrea Arcangeli * 34897906d00cSAndrea Arcangeli * A single task can't take more than one mm_take_all_locks() in a row 34907906d00cSAndrea Arcangeli * or it would deadlock. 34917906d00cSAndrea Arcangeli * 3492bf181b9fSMichel Lespinasse * The LSB in anon_vma->rb_root.rb_node and the AS_MM_ALL_LOCKS bitflag in 34937906d00cSAndrea Arcangeli * mapping->flags avoid to take the same lock twice, if more than one 34947906d00cSAndrea Arcangeli * vma in this mm is backed by the same anon_vma or address_space. 34957906d00cSAndrea Arcangeli * 349688f306b6SKirill A. Shutemov * We take locks in following order, accordingly to comment at beginning 349788f306b6SKirill A. Shutemov * of mm/rmap.c: 349888f306b6SKirill A. Shutemov * - all hugetlbfs_i_mmap_rwsem_key locks (aka mapping->i_mmap_rwsem for 349988f306b6SKirill A. Shutemov * hugetlb mapping); 350088f306b6SKirill A. Shutemov * - all i_mmap_rwsem locks; 350188f306b6SKirill A. Shutemov * - all anon_vma->rwseml 350288f306b6SKirill A. Shutemov * 350388f306b6SKirill A. Shutemov * We can take all locks within these types randomly because the VM code 350488f306b6SKirill A. Shutemov * doesn't nest them and we protected from parallel mm_take_all_locks() by 350588f306b6SKirill A. Shutemov * mm_all_locks_mutex. 35067906d00cSAndrea Arcangeli * 35077906d00cSAndrea Arcangeli * mm_take_all_locks() and mm_drop_all_locks are expensive operations 35087906d00cSAndrea Arcangeli * that may have to take thousand of locks. 35097906d00cSAndrea Arcangeli * 35107906d00cSAndrea Arcangeli * mm_take_all_locks() can fail if it's interrupted by signals. 35117906d00cSAndrea Arcangeli */ 35127906d00cSAndrea Arcangeli int mm_take_all_locks(struct mm_struct *mm) 35137906d00cSAndrea Arcangeli { 35147906d00cSAndrea Arcangeli struct vm_area_struct *vma; 35155beb4930SRik van Riel struct anon_vma_chain *avc; 3516763ecb03SLiam R. Howlett MA_STATE(mas, &mm->mm_mt, 0, 0); 35177906d00cSAndrea Arcangeli 3518325bca1fSRolf Eike Beer mmap_assert_write_locked(mm); 35197906d00cSAndrea Arcangeli 35207906d00cSAndrea Arcangeli mutex_lock(&mm_all_locks_mutex); 35217906d00cSAndrea Arcangeli 3522763ecb03SLiam R. Howlett mas_for_each(&mas, vma, ULONG_MAX) { 35237906d00cSAndrea Arcangeli if (signal_pending(current)) 35247906d00cSAndrea Arcangeli goto out_unlock; 352588f306b6SKirill A. Shutemov if (vma->vm_file && vma->vm_file->f_mapping && 352688f306b6SKirill A. Shutemov is_vm_hugetlb_page(vma)) 352788f306b6SKirill A. Shutemov vm_lock_mapping(mm, vma->vm_file->f_mapping); 352888f306b6SKirill A. Shutemov } 352988f306b6SKirill A. Shutemov 3530763ecb03SLiam R. Howlett mas_set(&mas, 0); 3531763ecb03SLiam R. Howlett mas_for_each(&mas, vma, ULONG_MAX) { 353288f306b6SKirill A. Shutemov if (signal_pending(current)) 353388f306b6SKirill A. Shutemov goto out_unlock; 353488f306b6SKirill A. Shutemov if (vma->vm_file && vma->vm_file->f_mapping && 353588f306b6SKirill A. Shutemov !is_vm_hugetlb_page(vma)) 3536454ed842SPeter Zijlstra vm_lock_mapping(mm, vma->vm_file->f_mapping); 35377906d00cSAndrea Arcangeli } 35387cd5a02fSPeter Zijlstra 3539763ecb03SLiam R. Howlett mas_set(&mas, 0); 3540763ecb03SLiam R. Howlett mas_for_each(&mas, vma, ULONG_MAX) { 35417cd5a02fSPeter Zijlstra if (signal_pending(current)) 35427cd5a02fSPeter Zijlstra goto out_unlock; 35437cd5a02fSPeter Zijlstra if (vma->anon_vma) 35445beb4930SRik van Riel list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) 35455beb4930SRik van Riel vm_lock_anon_vma(mm, avc->anon_vma); 35467cd5a02fSPeter Zijlstra } 35477cd5a02fSPeter Zijlstra 3548584cff54SKautuk Consul return 0; 35497906d00cSAndrea Arcangeli 35507906d00cSAndrea Arcangeli out_unlock: 35517906d00cSAndrea Arcangeli mm_drop_all_locks(mm); 3552584cff54SKautuk Consul return -EINTR; 35537906d00cSAndrea Arcangeli } 35547906d00cSAndrea Arcangeli 35557906d00cSAndrea Arcangeli static void vm_unlock_anon_vma(struct anon_vma *anon_vma) 35567906d00cSAndrea Arcangeli { 3557f808c13fSDavidlohr Bueso if (test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_root.rb_node)) { 35587906d00cSAndrea Arcangeli /* 35597906d00cSAndrea Arcangeli * The LSB of head.next can't change to 0 from under 35607906d00cSAndrea Arcangeli * us because we hold the mm_all_locks_mutex. 35617906d00cSAndrea Arcangeli * 35627906d00cSAndrea Arcangeli * We must however clear the bitflag before unlocking 3563bf181b9fSMichel Lespinasse * the vma so the users using the anon_vma->rb_root will 35647906d00cSAndrea Arcangeli * never see our bitflag. 35657906d00cSAndrea Arcangeli * 35667906d00cSAndrea Arcangeli * No need of atomic instructions here, head.next 35677906d00cSAndrea Arcangeli * can't change from under us until we release the 35685a505085SIngo Molnar * anon_vma->root->rwsem. 35697906d00cSAndrea Arcangeli */ 35707906d00cSAndrea Arcangeli if (!__test_and_clear_bit(0, (unsigned long *) 3571f808c13fSDavidlohr Bueso &anon_vma->root->rb_root.rb_root.rb_node)) 35727906d00cSAndrea Arcangeli BUG(); 357308b52706SKonstantin Khlebnikov anon_vma_unlock_write(anon_vma); 35747906d00cSAndrea Arcangeli } 35757906d00cSAndrea Arcangeli } 35767906d00cSAndrea Arcangeli 35777906d00cSAndrea Arcangeli static void vm_unlock_mapping(struct address_space *mapping) 35787906d00cSAndrea Arcangeli { 35797906d00cSAndrea Arcangeli if (test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) { 35807906d00cSAndrea Arcangeli /* 35817906d00cSAndrea Arcangeli * AS_MM_ALL_LOCKS can't change to 0 from under us 35827906d00cSAndrea Arcangeli * because we hold the mm_all_locks_mutex. 35837906d00cSAndrea Arcangeli */ 358483cde9e8SDavidlohr Bueso i_mmap_unlock_write(mapping); 35857906d00cSAndrea Arcangeli if (!test_and_clear_bit(AS_MM_ALL_LOCKS, 35867906d00cSAndrea Arcangeli &mapping->flags)) 35877906d00cSAndrea Arcangeli BUG(); 35887906d00cSAndrea Arcangeli } 35897906d00cSAndrea Arcangeli } 35907906d00cSAndrea Arcangeli 35917906d00cSAndrea Arcangeli /* 3592c1e8d7c6SMichel Lespinasse * The mmap_lock cannot be released by the caller until 35937906d00cSAndrea Arcangeli * mm_drop_all_locks() returns. 35947906d00cSAndrea Arcangeli */ 35957906d00cSAndrea Arcangeli void mm_drop_all_locks(struct mm_struct *mm) 35967906d00cSAndrea Arcangeli { 35977906d00cSAndrea Arcangeli struct vm_area_struct *vma; 35985beb4930SRik van Riel struct anon_vma_chain *avc; 3599763ecb03SLiam R. Howlett MA_STATE(mas, &mm->mm_mt, 0, 0); 36007906d00cSAndrea Arcangeli 3601325bca1fSRolf Eike Beer mmap_assert_write_locked(mm); 36027906d00cSAndrea Arcangeli BUG_ON(!mutex_is_locked(&mm_all_locks_mutex)); 36037906d00cSAndrea Arcangeli 3604763ecb03SLiam R. Howlett mas_for_each(&mas, vma, ULONG_MAX) { 36057906d00cSAndrea Arcangeli if (vma->anon_vma) 36065beb4930SRik van Riel list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) 36075beb4930SRik van Riel vm_unlock_anon_vma(avc->anon_vma); 36087906d00cSAndrea Arcangeli if (vma->vm_file && vma->vm_file->f_mapping) 36097906d00cSAndrea Arcangeli vm_unlock_mapping(vma->vm_file->f_mapping); 36107906d00cSAndrea Arcangeli } 36117906d00cSAndrea Arcangeli 36127906d00cSAndrea Arcangeli mutex_unlock(&mm_all_locks_mutex); 36137906d00cSAndrea Arcangeli } 36148feae131SDavid Howells 36158feae131SDavid Howells /* 36163edf41d8Sseokhoon.yoon * initialise the percpu counter for VM 36178feae131SDavid Howells */ 36188feae131SDavid Howells void __init mmap_init(void) 36198feae131SDavid Howells { 362000a62ce9SKOSAKI Motohiro int ret; 362100a62ce9SKOSAKI Motohiro 3622908c7f19STejun Heo ret = percpu_counter_init(&vm_committed_as, 0, GFP_KERNEL); 362300a62ce9SKOSAKI Motohiro VM_BUG_ON(ret); 36248feae131SDavid Howells } 3625c9b1d098SAndrew Shewmaker 3626c9b1d098SAndrew Shewmaker /* 3627c9b1d098SAndrew Shewmaker * Initialise sysctl_user_reserve_kbytes. 3628c9b1d098SAndrew Shewmaker * 3629c9b1d098SAndrew Shewmaker * This is intended to prevent a user from starting a single memory hogging 3630c9b1d098SAndrew Shewmaker * process, such that they cannot recover (kill the hog) in OVERCOMMIT_NEVER 3631c9b1d098SAndrew Shewmaker * mode. 3632c9b1d098SAndrew Shewmaker * 3633c9b1d098SAndrew Shewmaker * The default value is min(3% of free memory, 128MB) 3634c9b1d098SAndrew Shewmaker * 128MB is enough to recover with sshd/login, bash, and top/kill. 3635c9b1d098SAndrew Shewmaker */ 36361640879aSAndrew Shewmaker static int init_user_reserve(void) 3637c9b1d098SAndrew Shewmaker { 3638c9b1d098SAndrew Shewmaker unsigned long free_kbytes; 3639c9b1d098SAndrew Shewmaker 3640c41f012aSMichal Hocko free_kbytes = global_zone_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10); 3641c9b1d098SAndrew Shewmaker 3642c9b1d098SAndrew Shewmaker sysctl_user_reserve_kbytes = min(free_kbytes / 32, 1UL << 17); 3643c9b1d098SAndrew Shewmaker return 0; 3644c9b1d098SAndrew Shewmaker } 3645a64fb3cdSPaul Gortmaker subsys_initcall(init_user_reserve); 36464eeab4f5SAndrew Shewmaker 36474eeab4f5SAndrew Shewmaker /* 36484eeab4f5SAndrew Shewmaker * Initialise sysctl_admin_reserve_kbytes. 36494eeab4f5SAndrew Shewmaker * 36504eeab4f5SAndrew Shewmaker * The purpose of sysctl_admin_reserve_kbytes is to allow the sys admin 36514eeab4f5SAndrew Shewmaker * to log in and kill a memory hogging process. 36524eeab4f5SAndrew Shewmaker * 36534eeab4f5SAndrew Shewmaker * Systems with more than 256MB will reserve 8MB, enough to recover 36544eeab4f5SAndrew Shewmaker * with sshd, bash, and top in OVERCOMMIT_GUESS. Smaller systems will 36554eeab4f5SAndrew Shewmaker * only reserve 3% of free pages by default. 36564eeab4f5SAndrew Shewmaker */ 36571640879aSAndrew Shewmaker static int init_admin_reserve(void) 36584eeab4f5SAndrew Shewmaker { 36594eeab4f5SAndrew Shewmaker unsigned long free_kbytes; 36604eeab4f5SAndrew Shewmaker 3661c41f012aSMichal Hocko free_kbytes = global_zone_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10); 36624eeab4f5SAndrew Shewmaker 36634eeab4f5SAndrew Shewmaker sysctl_admin_reserve_kbytes = min(free_kbytes / 32, 1UL << 13); 36644eeab4f5SAndrew Shewmaker return 0; 36654eeab4f5SAndrew Shewmaker } 3666a64fb3cdSPaul Gortmaker subsys_initcall(init_admin_reserve); 36671640879aSAndrew Shewmaker 36681640879aSAndrew Shewmaker /* 36691640879aSAndrew Shewmaker * Reinititalise user and admin reserves if memory is added or removed. 36701640879aSAndrew Shewmaker * 36711640879aSAndrew Shewmaker * The default user reserve max is 128MB, and the default max for the 36721640879aSAndrew Shewmaker * admin reserve is 8MB. These are usually, but not always, enough to 36731640879aSAndrew Shewmaker * enable recovery from a memory hogging process using login/sshd, a shell, 36741640879aSAndrew Shewmaker * and tools like top. It may make sense to increase or even disable the 36751640879aSAndrew Shewmaker * reserve depending on the existence of swap or variations in the recovery 36761640879aSAndrew Shewmaker * tools. So, the admin may have changed them. 36771640879aSAndrew Shewmaker * 36781640879aSAndrew Shewmaker * If memory is added and the reserves have been eliminated or increased above 36791640879aSAndrew Shewmaker * the default max, then we'll trust the admin. 36801640879aSAndrew Shewmaker * 36811640879aSAndrew Shewmaker * If memory is removed and there isn't enough free memory, then we 36821640879aSAndrew Shewmaker * need to reset the reserves. 36831640879aSAndrew Shewmaker * 36841640879aSAndrew Shewmaker * Otherwise keep the reserve set by the admin. 36851640879aSAndrew Shewmaker */ 36861640879aSAndrew Shewmaker static int reserve_mem_notifier(struct notifier_block *nb, 36871640879aSAndrew Shewmaker unsigned long action, void *data) 36881640879aSAndrew Shewmaker { 36891640879aSAndrew Shewmaker unsigned long tmp, free_kbytes; 36901640879aSAndrew Shewmaker 36911640879aSAndrew Shewmaker switch (action) { 36921640879aSAndrew Shewmaker case MEM_ONLINE: 36931640879aSAndrew Shewmaker /* Default max is 128MB. Leave alone if modified by operator. */ 36941640879aSAndrew Shewmaker tmp = sysctl_user_reserve_kbytes; 36951640879aSAndrew Shewmaker if (0 < tmp && tmp < (1UL << 17)) 36961640879aSAndrew Shewmaker init_user_reserve(); 36971640879aSAndrew Shewmaker 36981640879aSAndrew Shewmaker /* Default max is 8MB. Leave alone if modified by operator. */ 36991640879aSAndrew Shewmaker tmp = sysctl_admin_reserve_kbytes; 37001640879aSAndrew Shewmaker if (0 < tmp && tmp < (1UL << 13)) 37011640879aSAndrew Shewmaker init_admin_reserve(); 37021640879aSAndrew Shewmaker 37031640879aSAndrew Shewmaker break; 37041640879aSAndrew Shewmaker case MEM_OFFLINE: 3705c41f012aSMichal Hocko free_kbytes = global_zone_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10); 37061640879aSAndrew Shewmaker 37071640879aSAndrew Shewmaker if (sysctl_user_reserve_kbytes > free_kbytes) { 37081640879aSAndrew Shewmaker init_user_reserve(); 37091640879aSAndrew Shewmaker pr_info("vm.user_reserve_kbytes reset to %lu\n", 37101640879aSAndrew Shewmaker sysctl_user_reserve_kbytes); 37111640879aSAndrew Shewmaker } 37121640879aSAndrew Shewmaker 37131640879aSAndrew Shewmaker if (sysctl_admin_reserve_kbytes > free_kbytes) { 37141640879aSAndrew Shewmaker init_admin_reserve(); 37151640879aSAndrew Shewmaker pr_info("vm.admin_reserve_kbytes reset to %lu\n", 37161640879aSAndrew Shewmaker sysctl_admin_reserve_kbytes); 37171640879aSAndrew Shewmaker } 37181640879aSAndrew Shewmaker break; 37191640879aSAndrew Shewmaker default: 37201640879aSAndrew Shewmaker break; 37211640879aSAndrew Shewmaker } 37221640879aSAndrew Shewmaker return NOTIFY_OK; 37231640879aSAndrew Shewmaker } 37241640879aSAndrew Shewmaker 37251640879aSAndrew Shewmaker static int __meminit init_reserve_notifier(void) 37261640879aSAndrew Shewmaker { 37271eeaa4fdSLiu Shixin if (hotplug_memory_notifier(reserve_mem_notifier, DEFAULT_CALLBACK_PRI)) 3728b1de0d13SMitchel Humpherys pr_err("Failed registering memory add/remove notifier for admin reserve\n"); 37291640879aSAndrew Shewmaker 37301640879aSAndrew Shewmaker return 0; 37311640879aSAndrew Shewmaker } 3732a64fb3cdSPaul Gortmaker subsys_initcall(init_reserve_notifier); 3733