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 /* 46168cefec5SLiam R. Howlett * init_multi_vma_prep() - Initializer for struct vma_prepare 46268cefec5SLiam R. Howlett * @vp: The vma_prepare struct 46368cefec5SLiam R. Howlett * @vma: The vma that will be altered once locked 46468cefec5SLiam R. Howlett * @next: The next vma if it is to be adjusted 46568cefec5SLiam R. Howlett * @remove: The first vma to be removed 46668cefec5SLiam R. Howlett * @remove2: The second vma to be removed 46768cefec5SLiam R. Howlett */ 46868cefec5SLiam R. Howlett static inline void init_multi_vma_prep(struct vma_prepare *vp, 46968cefec5SLiam R. Howlett struct vm_area_struct *vma, struct vm_area_struct *next, 47068cefec5SLiam R. Howlett struct vm_area_struct *remove, struct vm_area_struct *remove2) 47168cefec5SLiam R. Howlett { 47268cefec5SLiam R. Howlett memset(vp, 0, sizeof(struct vma_prepare)); 47368cefec5SLiam R. Howlett vp->vma = vma; 47468cefec5SLiam R. Howlett vp->anon_vma = vma->anon_vma; 47568cefec5SLiam R. Howlett vp->remove = remove; 47668cefec5SLiam R. Howlett vp->remove2 = remove2; 47768cefec5SLiam R. Howlett vp->adj_next = next; 47868cefec5SLiam R. Howlett if (!vp->anon_vma && next) 47968cefec5SLiam R. Howlett vp->anon_vma = next->anon_vma; 48068cefec5SLiam R. Howlett 48168cefec5SLiam R. Howlett vp->file = vma->vm_file; 48268cefec5SLiam R. Howlett if (vp->file) 48368cefec5SLiam R. Howlett vp->mapping = vma->vm_file->f_mapping; 48468cefec5SLiam R. Howlett 48568cefec5SLiam R. Howlett } 48668cefec5SLiam R. Howlett 48768cefec5SLiam R. Howlett /* 48868cefec5SLiam R. Howlett * init_vma_prep() - Initializer wrapper for vma_prepare struct 48968cefec5SLiam R. Howlett * @vp: The vma_prepare struct 49068cefec5SLiam R. Howlett * @vma: The vma that will be altered once locked 49168cefec5SLiam R. Howlett */ 49268cefec5SLiam R. Howlett static inline void init_vma_prep(struct vma_prepare *vp, 49368cefec5SLiam R. Howlett struct vm_area_struct *vma) 49468cefec5SLiam R. Howlett { 49568cefec5SLiam R. Howlett init_multi_vma_prep(vp, vma, NULL, NULL, NULL); 49668cefec5SLiam R. Howlett } 49768cefec5SLiam R. Howlett 49868cefec5SLiam R. Howlett 49968cefec5SLiam 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), 60868cefec5SLiam 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 /* 62104241ffeSLiam R. Howlett * dup_anon_vma() - Helper function to duplicate anon_vma 62204241ffeSLiam R. Howlett * @dst: The destination VMA 62304241ffeSLiam R. Howlett * @src: The source VMA 62404241ffeSLiam R. Howlett * 62504241ffeSLiam R. Howlett * Returns: 0 on success. 62604241ffeSLiam R. Howlett */ 62704241ffeSLiam R. Howlett static inline int dup_anon_vma(struct vm_area_struct *dst, 62804241ffeSLiam R. Howlett struct vm_area_struct *src) 62904241ffeSLiam R. Howlett { 63004241ffeSLiam R. Howlett /* 63104241ffeSLiam R. Howlett * Easily overlooked: when mprotect shifts the boundary, make sure the 63204241ffeSLiam R. Howlett * expanding vma has anon_vma set if the shrinking vma had, to cover any 63304241ffeSLiam R. Howlett * anon pages imported. 63404241ffeSLiam R. Howlett */ 63504241ffeSLiam R. Howlett if (src->anon_vma && !dst->anon_vma) { 63604241ffeSLiam R. Howlett dst->anon_vma = src->anon_vma; 63704241ffeSLiam R. Howlett return anon_vma_clone(dst, src); 63804241ffeSLiam R. Howlett } 63904241ffeSLiam R. Howlett 64004241ffeSLiam R. Howlett return 0; 64104241ffeSLiam R. Howlett } 64204241ffeSLiam R. Howlett 64304241ffeSLiam R. Howlett /* 6449303d3e1SLiam R. Howlett * vma_expand - Expand an existing VMA 6459303d3e1SLiam R. Howlett * 6469303d3e1SLiam R. Howlett * @vmi: The vma iterator 6479303d3e1SLiam R. Howlett * @vma: The vma to expand 6489303d3e1SLiam R. Howlett * @start: The start of the vma 6499303d3e1SLiam R. Howlett * @end: The exclusive end of the vma 6509303d3e1SLiam R. Howlett * @pgoff: The page offset of vma 6519303d3e1SLiam R. Howlett * @next: The current of next vma. 6529303d3e1SLiam R. Howlett * 6539303d3e1SLiam R. Howlett * Expand @vma to @start and @end. Can expand off the start and end. Will 6549303d3e1SLiam R. Howlett * expand over @next if it's different from @vma and @end == @next->vm_end. 6559303d3e1SLiam R. Howlett * Checking if the @vma can expand and merge with @next needs to be handled by 6569303d3e1SLiam R. Howlett * the caller. 6579303d3e1SLiam R. Howlett * 6589303d3e1SLiam R. Howlett * Returns: 0 on success 6599303d3e1SLiam R. Howlett */ 6607c9813e8SLiam R. Howlett int vma_expand(struct vma_iterator *vmi, struct vm_area_struct *vma, 6619303d3e1SLiam R. Howlett unsigned long start, unsigned long end, pgoff_t pgoff, 6629303d3e1SLiam R. Howlett struct vm_area_struct *next) 6639303d3e1SLiam R. Howlett { 66468cefec5SLiam R. Howlett bool remove_next = false; 6659303d3e1SLiam R. Howlett struct vma_prepare vp; 6669303d3e1SLiam R. Howlett 6679303d3e1SLiam R. Howlett if (next && (vma != next) && (end == next->vm_end)) { 66804241ffeSLiam R. Howlett int ret; 6699303d3e1SLiam R. Howlett 67004241ffeSLiam R. Howlett remove_next = true; 67104241ffeSLiam R. Howlett ret = dup_anon_vma(vma, next); 67204241ffeSLiam R. Howlett if (ret) 67304241ffeSLiam R. Howlett return ret; 6749303d3e1SLiam R. Howlett } 6759303d3e1SLiam R. Howlett 67668cefec5SLiam R. Howlett init_multi_vma_prep(&vp, vma, NULL, remove_next ? next : NULL, NULL); 6779303d3e1SLiam R. Howlett /* Not merging but overwriting any part of next is not handled. */ 6789303d3e1SLiam R. Howlett VM_WARN_ON(next && !vp.remove && 6799303d3e1SLiam R. Howlett next != vma && end > next->vm_start); 6809303d3e1SLiam R. Howlett /* Only handles expanding */ 6819303d3e1SLiam R. Howlett VM_WARN_ON(vma->vm_start < start || vma->vm_end > end); 6829303d3e1SLiam R. Howlett 6839303d3e1SLiam R. Howlett if (vma_iter_prealloc(vmi)) 6849303d3e1SLiam R. Howlett goto nomem; 6859303d3e1SLiam R. Howlett 6869303d3e1SLiam R. Howlett vma_adjust_trans_huge(vma, start, end, 0); 6879303d3e1SLiam R. Howlett /* VMA iterator points to previous, so set to start if necessary */ 6889303d3e1SLiam R. Howlett if (vma_iter_addr(vmi) != start) 6899303d3e1SLiam R. Howlett vma_iter_set(vmi, start); 6909303d3e1SLiam R. Howlett 6919303d3e1SLiam R. Howlett vma_prepare(&vp); 6929303d3e1SLiam R. Howlett vma->vm_start = start; 6939303d3e1SLiam R. Howlett vma->vm_end = end; 6949303d3e1SLiam R. Howlett vma->vm_pgoff = pgoff; 6959303d3e1SLiam R. Howlett /* Note: mas must be pointing to the expanding VMA */ 6969303d3e1SLiam R. Howlett vma_iter_store(vmi, vma); 6979303d3e1SLiam R. Howlett 6989303d3e1SLiam R. Howlett vma_complete(&vp, vmi, vma->vm_mm); 6999303d3e1SLiam R. Howlett validate_mm(vma->vm_mm); 7009303d3e1SLiam R. Howlett return 0; 7019303d3e1SLiam R. Howlett 7029303d3e1SLiam R. Howlett nomem: 7039303d3e1SLiam R. Howlett return -ENOMEM; 7049303d3e1SLiam R. Howlett } 705cf51e86dSLiam R. Howlett 706cf51e86dSLiam R. Howlett /* 707cf51e86dSLiam R. Howlett * vma_shrink() - Reduce an existing VMAs memory area 708cf51e86dSLiam R. Howlett * @vmi: The vma iterator 709cf51e86dSLiam R. Howlett * @vma: The VMA to modify 710cf51e86dSLiam R. Howlett * @start: The new start 711cf51e86dSLiam R. Howlett * @end: The new end 712cf51e86dSLiam R. Howlett * 713cf51e86dSLiam R. Howlett * Returns: 0 on success, -ENOMEM otherwise 714cf51e86dSLiam R. Howlett */ 715cf51e86dSLiam R. Howlett int vma_shrink(struct vma_iterator *vmi, struct vm_area_struct *vma, 716cf51e86dSLiam R. Howlett unsigned long start, unsigned long end, pgoff_t pgoff) 717cf51e86dSLiam R. Howlett { 718cf51e86dSLiam R. Howlett struct vma_prepare vp; 719cf51e86dSLiam R. Howlett 720cf51e86dSLiam R. Howlett WARN_ON((vma->vm_start != start) && (vma->vm_end != end)); 721cf51e86dSLiam R. Howlett 722cf51e86dSLiam R. Howlett if (vma_iter_prealloc(vmi)) 723cf51e86dSLiam R. Howlett return -ENOMEM; 724cf51e86dSLiam R. Howlett 725cf51e86dSLiam R. Howlett init_vma_prep(&vp, vma); 726cf51e86dSLiam R. Howlett vma_adjust_trans_huge(vma, start, end, 0); 727cf51e86dSLiam R. Howlett vma_prepare(&vp); 728cf51e86dSLiam R. Howlett 729cf51e86dSLiam R. Howlett if (vma->vm_start < start) 730cf51e86dSLiam R. Howlett vma_iter_clear(vmi, vma->vm_start, start); 731cf51e86dSLiam R. Howlett 732cf51e86dSLiam R. Howlett if (vma->vm_end > end) 733cf51e86dSLiam R. Howlett vma_iter_clear(vmi, end, vma->vm_end); 734cf51e86dSLiam R. Howlett 735cf51e86dSLiam R. Howlett vma->vm_start = start; 736cf51e86dSLiam R. Howlett vma->vm_end = end; 737cf51e86dSLiam R. Howlett vma->vm_pgoff = pgoff; 738cf51e86dSLiam R. Howlett vma_complete(&vp, vmi, vma->vm_mm); 739cf51e86dSLiam R. Howlett validate_mm(vma->vm_mm); 740cf51e86dSLiam R. Howlett return 0; 741cf51e86dSLiam R. Howlett } 742cf51e86dSLiam R. Howlett 7439303d3e1SLiam R. Howlett /* 7441da177e4SLinus Torvalds * If the vma has a ->close operation then the driver probably needs to release 7451da177e4SLinus Torvalds * per-vma resources, so we don't attempt to merge those. 7461da177e4SLinus Torvalds */ 7471da177e4SLinus Torvalds static inline int is_mergeable_vma(struct vm_area_struct *vma, 74819a809afSAndrea Arcangeli struct file *file, unsigned long vm_flags, 7499a10064fSColin Cross struct vm_userfaultfd_ctx vm_userfaultfd_ctx, 7505c26f6acSSuren Baghdasaryan struct anon_vma_name *anon_name) 7511da177e4SLinus Torvalds { 75234228d47SCyrill Gorcunov /* 75334228d47SCyrill Gorcunov * VM_SOFTDIRTY should not prevent from VMA merging, if we 75434228d47SCyrill Gorcunov * match the flags but dirty bit -- the caller should mark 75534228d47SCyrill Gorcunov * merged VMA as dirty. If dirty bit won't be excluded from 7568bb4e7a2SWei Yang * comparison, we increase pressure on the memory system forcing 75734228d47SCyrill Gorcunov * the kernel to generate new VMAs when old one could be 75834228d47SCyrill Gorcunov * extended instead. 75934228d47SCyrill Gorcunov */ 76034228d47SCyrill Gorcunov if ((vma->vm_flags ^ vm_flags) & ~VM_SOFTDIRTY) 7611da177e4SLinus Torvalds return 0; 7621da177e4SLinus Torvalds if (vma->vm_file != file) 7631da177e4SLinus Torvalds return 0; 7641da177e4SLinus Torvalds if (vma->vm_ops && vma->vm_ops->close) 7651da177e4SLinus Torvalds return 0; 76619a809afSAndrea Arcangeli if (!is_mergeable_vm_userfaultfd_ctx(vma, vm_userfaultfd_ctx)) 76719a809afSAndrea Arcangeli return 0; 7685c26f6acSSuren Baghdasaryan if (!anon_vma_name_eq(anon_vma_name(vma), anon_name)) 7699a10064fSColin Cross return 0; 7701da177e4SLinus Torvalds return 1; 7711da177e4SLinus Torvalds } 7721da177e4SLinus Torvalds 7731da177e4SLinus Torvalds static inline int is_mergeable_anon_vma(struct anon_vma *anon_vma1, 774965f55deSShaohua Li struct anon_vma *anon_vma2, 775965f55deSShaohua Li struct vm_area_struct *vma) 7761da177e4SLinus Torvalds { 777965f55deSShaohua Li /* 778965f55deSShaohua Li * The list_is_singular() test is to avoid merging VMA cloned from 779965f55deSShaohua Li * parents. This can improve scalability caused by anon_vma lock. 780965f55deSShaohua Li */ 781965f55deSShaohua Li if ((!anon_vma1 || !anon_vma2) && (!vma || 782965f55deSShaohua Li list_is_singular(&vma->anon_vma_chain))) 783965f55deSShaohua Li return 1; 784965f55deSShaohua Li return anon_vma1 == anon_vma2; 7851da177e4SLinus Torvalds } 7861da177e4SLinus Torvalds 7871da177e4SLinus Torvalds /* 7881da177e4SLinus Torvalds * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff) 7891da177e4SLinus Torvalds * in front of (at a lower virtual address and file offset than) the vma. 7901da177e4SLinus Torvalds * 7911da177e4SLinus Torvalds * We cannot merge two vmas if they have differently assigned (non-NULL) 7921da177e4SLinus Torvalds * anon_vmas, nor if same anon_vma is assigned but offsets incompatible. 7931da177e4SLinus Torvalds * 7941da177e4SLinus Torvalds * We don't check here for the merged mmap wrapping around the end of pagecache 79545e55300SPeter Collingbourne * indices (16TB on ia32) because do_mmap() does not permit mmap's which 7961da177e4SLinus Torvalds * wrap, nor mmaps which cover the final page at index -1UL. 7971da177e4SLinus Torvalds */ 7981da177e4SLinus Torvalds static int 7991da177e4SLinus Torvalds can_vma_merge_before(struct vm_area_struct *vma, unsigned long vm_flags, 80019a809afSAndrea Arcangeli struct anon_vma *anon_vma, struct file *file, 80119a809afSAndrea Arcangeli pgoff_t vm_pgoff, 8029a10064fSColin Cross struct vm_userfaultfd_ctx vm_userfaultfd_ctx, 8035c26f6acSSuren Baghdasaryan struct anon_vma_name *anon_name) 8041da177e4SLinus Torvalds { 8059a10064fSColin Cross if (is_mergeable_vma(vma, file, vm_flags, vm_userfaultfd_ctx, anon_name) && 806965f55deSShaohua Li is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) { 8071da177e4SLinus Torvalds if (vma->vm_pgoff == vm_pgoff) 8081da177e4SLinus Torvalds return 1; 8091da177e4SLinus Torvalds } 8101da177e4SLinus Torvalds return 0; 8111da177e4SLinus Torvalds } 8121da177e4SLinus Torvalds 8131da177e4SLinus Torvalds /* 8141da177e4SLinus Torvalds * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff) 8151da177e4SLinus Torvalds * beyond (at a higher virtual address and file offset than) the vma. 8161da177e4SLinus Torvalds * 8171da177e4SLinus Torvalds * We cannot merge two vmas if they have differently assigned (non-NULL) 8181da177e4SLinus Torvalds * anon_vmas, nor if same anon_vma is assigned but offsets incompatible. 8191da177e4SLinus Torvalds */ 8201da177e4SLinus Torvalds static int 8211da177e4SLinus Torvalds can_vma_merge_after(struct vm_area_struct *vma, unsigned long vm_flags, 82219a809afSAndrea Arcangeli struct anon_vma *anon_vma, struct file *file, 82319a809afSAndrea Arcangeli pgoff_t vm_pgoff, 8249a10064fSColin Cross struct vm_userfaultfd_ctx vm_userfaultfd_ctx, 8255c26f6acSSuren Baghdasaryan struct anon_vma_name *anon_name) 8261da177e4SLinus Torvalds { 8279a10064fSColin Cross if (is_mergeable_vma(vma, file, vm_flags, vm_userfaultfd_ctx, anon_name) && 828965f55deSShaohua Li is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) { 8291da177e4SLinus Torvalds pgoff_t vm_pglen; 830d6e93217SLibin vm_pglen = vma_pages(vma); 8311da177e4SLinus Torvalds if (vma->vm_pgoff + vm_pglen == vm_pgoff) 8321da177e4SLinus Torvalds return 1; 8331da177e4SLinus Torvalds } 8341da177e4SLinus Torvalds return 0; 8351da177e4SLinus Torvalds } 8361da177e4SLinus Torvalds 8371da177e4SLinus Torvalds /* 8389a10064fSColin Cross * Given a mapping request (addr,end,vm_flags,file,pgoff,anon_name), 8399a10064fSColin Cross * figure out whether that can be merged with its predecessor or its 8409a10064fSColin Cross * successor. Or both (it neatly fills a hole). 8411da177e4SLinus Torvalds * 8421da177e4SLinus Torvalds * In most cases - when called for mmap, brk or mremap - [addr,end) is 8431da177e4SLinus Torvalds * certain not to be mapped by the time vma_merge is called; but when 8441da177e4SLinus Torvalds * called for mprotect, it is certain to be already mapped (either at 8451da177e4SLinus Torvalds * an offset within prev, or at the start of next), and the flags of 8461da177e4SLinus Torvalds * this area are about to be changed to vm_flags - and the no-change 8471da177e4SLinus Torvalds * case has already been eliminated. 8481da177e4SLinus Torvalds * 8491da177e4SLinus Torvalds * The following mprotect cases have to be considered, where AAAA is 8501da177e4SLinus Torvalds * the area passed down from mprotect_fixup, never extending beyond one 8511da177e4SLinus Torvalds * vma, PPPPPP is the prev vma specified, and NNNNNN the next vma after: 8521da177e4SLinus Torvalds * 8535d42ab29SWei Yang * AAAA AAAA AAAA 8545d42ab29SWei Yang * PPPPPPNNNNNN PPPPPPNNNNNN PPPPPPNNNNNN 8555d42ab29SWei Yang * cannot merge might become might become 8565d42ab29SWei Yang * PPNNNNNNNNNN PPPPPPPPPPNN 8575d42ab29SWei Yang * mmap, brk or case 4 below case 5 below 8585d42ab29SWei Yang * mremap move: 8595d42ab29SWei Yang * AAAA AAAA 8605d42ab29SWei Yang * PPPP NNNN PPPPNNNNXXXX 8615d42ab29SWei Yang * might become might become 8625d42ab29SWei Yang * PPPPPPPPPPPP 1 or PPPPPPPPPPPP 6 or 8635d42ab29SWei Yang * PPPPPPPPNNNN 2 or PPPPPPPPXXXX 7 or 8645d42ab29SWei Yang * PPPPNNNNNNNN 3 PPPPXXXXXXXX 8 8651da177e4SLinus Torvalds * 8668bb4e7a2SWei Yang * It is important for case 8 that the vma NNNN overlapping the 867e86f15eeSAndrea Arcangeli * region AAAA is never going to extended over XXXX. Instead XXXX must 868e86f15eeSAndrea Arcangeli * be extended in region AAAA and NNNN must be removed. This way in 8690503ea8fSLiam R. Howlett * all cases where vma_merge succeeds, the moment vma_merge drops the 870e86f15eeSAndrea Arcangeli * rmap_locks, the properties of the merged vma will be already 871e86f15eeSAndrea Arcangeli * correct for the whole merged range. Some of those properties like 872e86f15eeSAndrea Arcangeli * vm_page_prot/vm_flags may be accessed by rmap_walks and they must 873e86f15eeSAndrea Arcangeli * be correct for the whole merged range immediately after the 874e86f15eeSAndrea Arcangeli * rmap_locks are released. Otherwise if XXXX would be removed and 875e86f15eeSAndrea Arcangeli * NNNN would be extended over the XXXX range, remove_migration_ptes 876e86f15eeSAndrea Arcangeli * or other rmap walkers (if working on addresses beyond the "end" 877e86f15eeSAndrea Arcangeli * parameter) may establish ptes with the wrong permissions of NNNN 878e86f15eeSAndrea Arcangeli * instead of the right permissions of XXXX. 8790503ea8fSLiam R. Howlett * 8800503ea8fSLiam R. Howlett * In the code below: 8810503ea8fSLiam R. Howlett * PPPP is represented by *prev 8820503ea8fSLiam R. Howlett * NNNN is represented by *mid (and possibly equal to *next) 8830503ea8fSLiam R. Howlett * XXXX is represented by *next or not represented at all. 8840503ea8fSLiam R. Howlett * AAAA is not represented - it will be merged or the function will return NULL 8851da177e4SLinus Torvalds */ 8869760ebffSLiam R. Howlett struct vm_area_struct *vma_merge(struct vma_iterator *vmi, struct mm_struct *mm, 8871da177e4SLinus Torvalds struct vm_area_struct *prev, unsigned long addr, 8881da177e4SLinus Torvalds unsigned long end, unsigned long vm_flags, 8891da177e4SLinus Torvalds struct anon_vma *anon_vma, struct file *file, 89019a809afSAndrea Arcangeli pgoff_t pgoff, struct mempolicy *policy, 8919a10064fSColin Cross struct vm_userfaultfd_ctx vm_userfaultfd_ctx, 8925c26f6acSSuren Baghdasaryan struct anon_vma_name *anon_name) 8931da177e4SLinus Torvalds { 8941da177e4SLinus Torvalds pgoff_t pglen = (end - addr) >> PAGE_SHIFT; 8950503ea8fSLiam R. Howlett pgoff_t vma_pgoff; 8969760ebffSLiam R. Howlett struct vm_area_struct *mid, *next, *res = NULL; 8970503ea8fSLiam R. Howlett struct vm_area_struct *vma, *adjust, *remove, *remove2; 898eef19944SJakub Matěna int err = -1; 899eef19944SJakub Matěna bool merge_prev = false; 900eef19944SJakub Matěna bool merge_next = false; 9010503ea8fSLiam R. Howlett bool vma_expanded = false; 9020503ea8fSLiam R. Howlett struct vma_prepare vp; 9030503ea8fSLiam R. Howlett unsigned long vma_end = end; 9040503ea8fSLiam R. Howlett long adj_next = 0; 9050503ea8fSLiam R. Howlett unsigned long vma_start = addr; 9061da177e4SLinus Torvalds 9070503ea8fSLiam R. Howlett validate_mm(mm); 9081da177e4SLinus Torvalds /* 9091da177e4SLinus Torvalds * We later require that vma->vm_flags == vm_flags, 9101da177e4SLinus Torvalds * so this tests vma->vm_flags & VM_SPECIAL, too. 9111da177e4SLinus Torvalds */ 9121da177e4SLinus Torvalds if (vm_flags & VM_SPECIAL) 9131da177e4SLinus Torvalds return NULL; 9141da177e4SLinus Torvalds 915763ecb03SLiam R. Howlett next = find_vma(mm, prev ? prev->vm_end : 0); 916eef19944SJakub Matěna mid = next; 917eef19944SJakub Matěna if (next && next->vm_end == end) /* cases 6, 7, 8 */ 918763ecb03SLiam R. Howlett next = find_vma(mm, next->vm_end); 9191da177e4SLinus Torvalds 920e86f15eeSAndrea Arcangeli /* verify some invariant that must be enforced by the caller */ 921e86f15eeSAndrea Arcangeli VM_WARN_ON(prev && addr <= prev->vm_start); 922eef19944SJakub Matěna VM_WARN_ON(mid && end > mid->vm_end); 923e86f15eeSAndrea Arcangeli VM_WARN_ON(addr >= end); 924e86f15eeSAndrea Arcangeli 9250503ea8fSLiam R. Howlett if (prev) { 9260503ea8fSLiam R. Howlett res = prev; 9270503ea8fSLiam R. Howlett vma = prev; 9280503ea8fSLiam R. Howlett vma_start = prev->vm_start; 9290503ea8fSLiam R. Howlett vma_pgoff = prev->vm_pgoff; 930eef19944SJakub Matěna /* Can we merge the predecessor? */ 9310503ea8fSLiam R. Howlett if (prev->vm_end == addr && mpol_equal(vma_policy(prev), policy) 9320503ea8fSLiam R. Howlett && can_vma_merge_after(prev, vm_flags, anon_vma, file, 9330503ea8fSLiam R. Howlett pgoff, vm_userfaultfd_ctx, anon_name)) { 934eef19944SJakub Matěna merge_prev = true; 93518b098afSLiam R. Howlett vma_prev(vmi); 9361da177e4SLinus Torvalds } 9370503ea8fSLiam R. Howlett } 938eef19944SJakub Matěna /* Can we merge the successor? */ 9391da177e4SLinus Torvalds if (next && end == next->vm_start && 9401da177e4SLinus Torvalds mpol_equal(policy, vma_policy(next)) && 9411da177e4SLinus Torvalds can_vma_merge_before(next, vm_flags, 94219a809afSAndrea Arcangeli anon_vma, file, pgoff+pglen, 9439a10064fSColin Cross vm_userfaultfd_ctx, anon_name)) { 944eef19944SJakub Matěna merge_next = true; 945eef19944SJakub Matěna } 9460503ea8fSLiam R. Howlett 9470503ea8fSLiam R. Howlett remove = remove2 = adjust = NULL; 948eef19944SJakub Matěna /* Can we merge both the predecessor and the successor? */ 949eef19944SJakub Matěna if (merge_prev && merge_next && 9500503ea8fSLiam R. Howlett is_mergeable_anon_vma(prev->anon_vma, next->anon_vma, NULL)) { 9510503ea8fSLiam R. Howlett remove = mid; /* case 1 */ 9520503ea8fSLiam R. Howlett vma_end = next->vm_end; 9530503ea8fSLiam R. Howlett err = dup_anon_vma(res, remove); 9540503ea8fSLiam R. Howlett if (mid != next) { /* case 6 */ 9550503ea8fSLiam R. Howlett remove2 = next; 9560503ea8fSLiam R. Howlett if (!remove->anon_vma) 9570503ea8fSLiam R. Howlett err = dup_anon_vma(res, remove2); 9580503ea8fSLiam R. Howlett } 9590503ea8fSLiam R. Howlett } else if (merge_prev) { 9600503ea8fSLiam R. Howlett err = 0; /* case 2 */ 9610503ea8fSLiam R. Howlett if (mid && end > mid->vm_start) { 9620503ea8fSLiam R. Howlett err = dup_anon_vma(res, mid); 9630503ea8fSLiam R. Howlett if (end == mid->vm_end) { /* case 7 */ 9640503ea8fSLiam R. Howlett remove = mid; 9650503ea8fSLiam R. Howlett } else { /* case 5 */ 9660503ea8fSLiam R. Howlett adjust = mid; 9670503ea8fSLiam R. Howlett adj_next = (end - mid->vm_start); 9680503ea8fSLiam R. Howlett } 9690503ea8fSLiam R. Howlett } 970eef19944SJakub Matěna } else if (merge_next) { 971eef19944SJakub Matěna res = next; 9720503ea8fSLiam R. Howlett if (prev && addr < prev->vm_end) { /* case 4 */ 9730503ea8fSLiam R. Howlett vma_end = addr; 9740503ea8fSLiam R. Howlett adjust = mid; 9750503ea8fSLiam R. Howlett adj_next = -(vma->vm_end - addr); 9760503ea8fSLiam R. Howlett err = dup_anon_vma(res, adjust); 9770503ea8fSLiam R. Howlett } else { 9780503ea8fSLiam R. Howlett vma = next; /* case 3 */ 9790503ea8fSLiam R. Howlett vma_start = addr; 9800503ea8fSLiam R. Howlett vma_end = next->vm_end; 9810503ea8fSLiam R. Howlett vma_pgoff = mid->vm_pgoff; 9820503ea8fSLiam R. Howlett err = 0; 9830503ea8fSLiam R. Howlett if (mid != next) { /* case 8 */ 9840503ea8fSLiam R. Howlett remove = mid; 9850503ea8fSLiam R. Howlett err = dup_anon_vma(res, remove); 9860503ea8fSLiam R. Howlett } 9870503ea8fSLiam R. Howlett } 9881da177e4SLinus Torvalds } 9891da177e4SLinus Torvalds 9900503ea8fSLiam R. Howlett /* Cannot merge or error in anon_vma clone */ 991eef19944SJakub Matěna if (err) 9921da177e4SLinus Torvalds return NULL; 9930503ea8fSLiam R. Howlett 9940503ea8fSLiam R. Howlett if (vma_iter_prealloc(vmi)) 9950503ea8fSLiam R. Howlett return NULL; 9960503ea8fSLiam R. Howlett 9970503ea8fSLiam R. Howlett vma_adjust_trans_huge(vma, vma_start, vma_end, adj_next); 9980503ea8fSLiam R. Howlett init_multi_vma_prep(&vp, vma, adjust, remove, remove2); 9990503ea8fSLiam R. Howlett VM_WARN_ON(vp.anon_vma && adjust && adjust->anon_vma && 10000503ea8fSLiam R. Howlett vp.anon_vma != adjust->anon_vma); 10010503ea8fSLiam R. Howlett 10020503ea8fSLiam R. Howlett vma_prepare(&vp); 10030503ea8fSLiam R. Howlett if (vma_start < vma->vm_start || vma_end > vma->vm_end) 10040503ea8fSLiam R. Howlett vma_expanded = true; 10050503ea8fSLiam R. Howlett 10060503ea8fSLiam R. Howlett vma->vm_start = vma_start; 10070503ea8fSLiam R. Howlett vma->vm_end = vma_end; 10080503ea8fSLiam R. Howlett vma->vm_pgoff = vma_pgoff; 10090503ea8fSLiam R. Howlett 10100503ea8fSLiam R. Howlett if (vma_expanded) 10110503ea8fSLiam R. Howlett vma_iter_store(vmi, vma); 10120503ea8fSLiam R. Howlett 10130503ea8fSLiam R. Howlett if (adj_next) { 10140503ea8fSLiam R. Howlett adjust->vm_start += adj_next; 10150503ea8fSLiam R. Howlett adjust->vm_pgoff += adj_next >> PAGE_SHIFT; 10160503ea8fSLiam R. Howlett if (adj_next < 0) { 10170503ea8fSLiam R. Howlett WARN_ON(vma_expanded); 10180503ea8fSLiam R. Howlett vma_iter_store(vmi, next); 10190503ea8fSLiam R. Howlett } 10200503ea8fSLiam R. Howlett } 10210503ea8fSLiam R. Howlett 10220503ea8fSLiam R. Howlett vma_complete(&vp, vmi, mm); 10230503ea8fSLiam R. Howlett vma_iter_free(vmi); 10240503ea8fSLiam R. Howlett validate_mm(mm); 1025eef19944SJakub Matěna khugepaged_enter_vma(res, vm_flags); 10261da177e4SLinus Torvalds 10279760ebffSLiam R. Howlett return res; 1028f2ebfe43SLiam R. Howlett } 1029f2ebfe43SLiam R. Howlett 10301da177e4SLinus Torvalds /* 1031b4f315b4SEthon Paul * Rough compatibility check to quickly see if it's even worth looking 1032d0e9fe17SLinus Torvalds * at sharing an anon_vma. 1033d0e9fe17SLinus Torvalds * 1034d0e9fe17SLinus Torvalds * They need to have the same vm_file, and the flags can only differ 1035d0e9fe17SLinus Torvalds * in things that mprotect may change. 1036d0e9fe17SLinus Torvalds * 1037d0e9fe17SLinus Torvalds * NOTE! The fact that we share an anon_vma doesn't _have_ to mean that 1038d0e9fe17SLinus Torvalds * we can merge the two vma's. For example, we refuse to merge a vma if 1039d0e9fe17SLinus Torvalds * there is a vm_ops->close() function, because that indicates that the 1040d0e9fe17SLinus Torvalds * driver is doing some kind of reference counting. But that doesn't 1041d0e9fe17SLinus Torvalds * really matter for the anon_vma sharing case. 1042d0e9fe17SLinus Torvalds */ 1043d0e9fe17SLinus Torvalds static int anon_vma_compatible(struct vm_area_struct *a, struct vm_area_struct *b) 1044d0e9fe17SLinus Torvalds { 1045d0e9fe17SLinus Torvalds return a->vm_end == b->vm_start && 1046d0e9fe17SLinus Torvalds mpol_equal(vma_policy(a), vma_policy(b)) && 1047d0e9fe17SLinus Torvalds a->vm_file == b->vm_file && 10486cb4d9a2SAnshuman Khandual !((a->vm_flags ^ b->vm_flags) & ~(VM_ACCESS_FLAGS | VM_SOFTDIRTY)) && 1049d0e9fe17SLinus Torvalds b->vm_pgoff == a->vm_pgoff + ((b->vm_start - a->vm_start) >> PAGE_SHIFT); 1050d0e9fe17SLinus Torvalds } 1051d0e9fe17SLinus Torvalds 1052d0e9fe17SLinus Torvalds /* 1053d0e9fe17SLinus Torvalds * Do some basic sanity checking to see if we can re-use the anon_vma 1054d0e9fe17SLinus Torvalds * from 'old'. The 'a'/'b' vma's are in VM order - one of them will be 1055d0e9fe17SLinus Torvalds * the same as 'old', the other will be the new one that is trying 1056d0e9fe17SLinus Torvalds * to share the anon_vma. 1057d0e9fe17SLinus Torvalds * 10585b449489SFlorian Rommel * NOTE! This runs with mmap_lock held for reading, so it is possible that 1059d0e9fe17SLinus Torvalds * the anon_vma of 'old' is concurrently in the process of being set up 1060d0e9fe17SLinus Torvalds * by another page fault trying to merge _that_. But that's ok: if it 1061d0e9fe17SLinus Torvalds * is being set up, that automatically means that it will be a singleton 1062d0e9fe17SLinus Torvalds * acceptable for merging, so we can do all of this optimistically. But 10634db0c3c2SJason Low * we do that READ_ONCE() to make sure that we never re-load the pointer. 1064d0e9fe17SLinus Torvalds * 1065d0e9fe17SLinus Torvalds * IOW: that the "list_is_singular()" test on the anon_vma_chain only 1066d0e9fe17SLinus Torvalds * matters for the 'stable anon_vma' case (ie the thing we want to avoid 1067d0e9fe17SLinus Torvalds * is to return an anon_vma that is "complex" due to having gone through 1068d0e9fe17SLinus Torvalds * a fork). 1069d0e9fe17SLinus Torvalds * 1070d0e9fe17SLinus Torvalds * We also make sure that the two vma's are compatible (adjacent, 1071d0e9fe17SLinus Torvalds * and with the same memory policies). That's all stable, even with just 10725b449489SFlorian Rommel * a read lock on the mmap_lock. 1073d0e9fe17SLinus Torvalds */ 1074d0e9fe17SLinus Torvalds static struct anon_vma *reusable_anon_vma(struct vm_area_struct *old, struct vm_area_struct *a, struct vm_area_struct *b) 1075d0e9fe17SLinus Torvalds { 1076d0e9fe17SLinus Torvalds if (anon_vma_compatible(a, b)) { 10774db0c3c2SJason Low struct anon_vma *anon_vma = READ_ONCE(old->anon_vma); 1078d0e9fe17SLinus Torvalds 1079d0e9fe17SLinus Torvalds if (anon_vma && list_is_singular(&old->anon_vma_chain)) 1080d0e9fe17SLinus Torvalds return anon_vma; 1081d0e9fe17SLinus Torvalds } 1082d0e9fe17SLinus Torvalds return NULL; 1083d0e9fe17SLinus Torvalds } 1084d0e9fe17SLinus Torvalds 1085d0e9fe17SLinus Torvalds /* 10861da177e4SLinus Torvalds * find_mergeable_anon_vma is used by anon_vma_prepare, to check 10871da177e4SLinus Torvalds * neighbouring vmas for a suitable anon_vma, before it goes off 10881da177e4SLinus Torvalds * to allocate a new anon_vma. It checks because a repetitive 10891da177e4SLinus Torvalds * sequence of mprotects and faults may otherwise lead to distinct 10901da177e4SLinus Torvalds * anon_vmas being allocated, preventing vma merge in subsequent 10911da177e4SLinus Torvalds * mprotect. 10921da177e4SLinus Torvalds */ 10931da177e4SLinus Torvalds struct anon_vma *find_mergeable_anon_vma(struct vm_area_struct *vma) 10941da177e4SLinus Torvalds { 1095763ecb03SLiam R. Howlett MA_STATE(mas, &vma->vm_mm->mm_mt, vma->vm_end, vma->vm_end); 1096a67c8caaSMiaohe Lin struct anon_vma *anon_vma = NULL; 1097763ecb03SLiam R. Howlett struct vm_area_struct *prev, *next; 10981da177e4SLinus Torvalds 1099a67c8caaSMiaohe Lin /* Try next first. */ 1100763ecb03SLiam R. Howlett next = mas_walk(&mas); 1101763ecb03SLiam R. Howlett if (next) { 1102763ecb03SLiam R. Howlett anon_vma = reusable_anon_vma(next, vma, next); 1103d0e9fe17SLinus Torvalds if (anon_vma) 1104d0e9fe17SLinus Torvalds return anon_vma; 1105a67c8caaSMiaohe Lin } 11061da177e4SLinus Torvalds 1107763ecb03SLiam R. Howlett prev = mas_prev(&mas, 0); 1108763ecb03SLiam R. Howlett VM_BUG_ON_VMA(prev != vma, vma); 1109763ecb03SLiam R. Howlett prev = mas_prev(&mas, 0); 1110a67c8caaSMiaohe Lin /* Try prev next. */ 1111763ecb03SLiam R. Howlett if (prev) 1112763ecb03SLiam R. Howlett anon_vma = reusable_anon_vma(prev, prev, vma); 1113a67c8caaSMiaohe Lin 11141da177e4SLinus Torvalds /* 1115a67c8caaSMiaohe Lin * We might reach here with anon_vma == NULL if we can't find 1116a67c8caaSMiaohe Lin * any reusable anon_vma. 11171da177e4SLinus Torvalds * There's no absolute need to look only at touching neighbours: 11181da177e4SLinus Torvalds * we could search further afield for "compatible" anon_vmas. 11191da177e4SLinus Torvalds * But it would probably just be a waste of time searching, 11201da177e4SLinus Torvalds * or lead to too many vmas hanging off the same anon_vma. 11211da177e4SLinus Torvalds * We're trying to allow mprotect remerging later on, 11221da177e4SLinus Torvalds * not trying to minimize memory used for anon_vmas. 11231da177e4SLinus Torvalds */ 1124a67c8caaSMiaohe Lin return anon_vma; 11251da177e4SLinus Torvalds } 11261da177e4SLinus Torvalds 11271da177e4SLinus Torvalds /* 112840401530SAl Viro * If a hint addr is less than mmap_min_addr change hint to be as 112940401530SAl Viro * low as possible but still greater than mmap_min_addr 113040401530SAl Viro */ 113140401530SAl Viro static inline unsigned long round_hint_to_min(unsigned long hint) 113240401530SAl Viro { 113340401530SAl Viro hint &= PAGE_MASK; 113440401530SAl Viro if (((void *)hint != NULL) && 113540401530SAl Viro (hint < mmap_min_addr)) 113640401530SAl Viro return PAGE_ALIGN(mmap_min_addr); 113740401530SAl Viro return hint; 113840401530SAl Viro } 113940401530SAl Viro 11406aeb2542SMike Rapoport int mlock_future_check(struct mm_struct *mm, unsigned long flags, 1141363ee17fSDavidlohr Bueso unsigned long len) 1142363ee17fSDavidlohr Bueso { 1143363ee17fSDavidlohr Bueso unsigned long locked, lock_limit; 1144363ee17fSDavidlohr Bueso 1145363ee17fSDavidlohr Bueso /* mlock MCL_FUTURE? */ 1146363ee17fSDavidlohr Bueso if (flags & VM_LOCKED) { 1147363ee17fSDavidlohr Bueso locked = len >> PAGE_SHIFT; 1148363ee17fSDavidlohr Bueso locked += mm->locked_vm; 1149363ee17fSDavidlohr Bueso lock_limit = rlimit(RLIMIT_MEMLOCK); 1150363ee17fSDavidlohr Bueso lock_limit >>= PAGE_SHIFT; 1151363ee17fSDavidlohr Bueso if (locked > lock_limit && !capable(CAP_IPC_LOCK)) 1152363ee17fSDavidlohr Bueso return -EAGAIN; 1153363ee17fSDavidlohr Bueso } 1154363ee17fSDavidlohr Bueso return 0; 1155363ee17fSDavidlohr Bueso } 1156363ee17fSDavidlohr Bueso 1157be83bbf8SLinus Torvalds static inline u64 file_mmap_size_max(struct file *file, struct inode *inode) 1158be83bbf8SLinus Torvalds { 1159be83bbf8SLinus Torvalds if (S_ISREG(inode->i_mode)) 1160423913adSLinus Torvalds return MAX_LFS_FILESIZE; 1161be83bbf8SLinus Torvalds 1162be83bbf8SLinus Torvalds if (S_ISBLK(inode->i_mode)) 1163be83bbf8SLinus Torvalds return MAX_LFS_FILESIZE; 1164be83bbf8SLinus Torvalds 116576f34950SIvan Khoronzhuk if (S_ISSOCK(inode->i_mode)) 116676f34950SIvan Khoronzhuk return MAX_LFS_FILESIZE; 116776f34950SIvan Khoronzhuk 1168be83bbf8SLinus Torvalds /* Special "we do even unsigned file positions" case */ 1169be83bbf8SLinus Torvalds if (file->f_mode & FMODE_UNSIGNED_OFFSET) 1170be83bbf8SLinus Torvalds return 0; 1171be83bbf8SLinus Torvalds 1172be83bbf8SLinus Torvalds /* Yes, random drivers might want more. But I'm tired of buggy drivers */ 1173be83bbf8SLinus Torvalds return ULONG_MAX; 1174be83bbf8SLinus Torvalds } 1175be83bbf8SLinus Torvalds 1176be83bbf8SLinus Torvalds static inline bool file_mmap_ok(struct file *file, struct inode *inode, 1177be83bbf8SLinus Torvalds unsigned long pgoff, unsigned long len) 1178be83bbf8SLinus Torvalds { 1179be83bbf8SLinus Torvalds u64 maxsize = file_mmap_size_max(file, inode); 1180be83bbf8SLinus Torvalds 1181be83bbf8SLinus Torvalds if (maxsize && len > maxsize) 1182be83bbf8SLinus Torvalds return false; 1183be83bbf8SLinus Torvalds maxsize -= len; 1184be83bbf8SLinus Torvalds if (pgoff > maxsize >> PAGE_SHIFT) 1185be83bbf8SLinus Torvalds return false; 1186be83bbf8SLinus Torvalds return true; 1187be83bbf8SLinus Torvalds } 1188be83bbf8SLinus Torvalds 118940401530SAl Viro /* 11903e4e28c5SMichel Lespinasse * The caller must write-lock current->mm->mmap_lock. 11911da177e4SLinus Torvalds */ 11921fcfd8dbSOleg Nesterov unsigned long do_mmap(struct file *file, unsigned long addr, 11931da177e4SLinus Torvalds unsigned long len, unsigned long prot, 119445e55300SPeter Collingbourne unsigned long flags, unsigned long pgoff, 119545e55300SPeter Collingbourne unsigned long *populate, struct list_head *uf) 11961da177e4SLinus Torvalds { 11971da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 119845e55300SPeter Collingbourne vm_flags_t vm_flags; 119962b5f7d0SDave Hansen int pkey = 0; 12001da177e4SLinus Torvalds 1201524e00b3SLiam R. Howlett validate_mm(mm); 120241badc15SMichel Lespinasse *populate = 0; 1203bebeb3d6SMichel Lespinasse 1204e37609bbSPiotr Kwapulinski if (!len) 1205e37609bbSPiotr Kwapulinski return -EINVAL; 1206e37609bbSPiotr Kwapulinski 12071da177e4SLinus Torvalds /* 12081da177e4SLinus Torvalds * Does the application expect PROT_READ to imply PROT_EXEC? 12091da177e4SLinus Torvalds * 12101da177e4SLinus Torvalds * (the exception is when the underlying filesystem is noexec 12111da177e4SLinus Torvalds * mounted, in which case we dont add PROT_EXEC.) 12121da177e4SLinus Torvalds */ 12131da177e4SLinus Torvalds if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC)) 121490f8572bSEric W. Biederman if (!(file && path_noexec(&file->f_path))) 12151da177e4SLinus Torvalds prot |= PROT_EXEC; 12161da177e4SLinus Torvalds 1217a4ff8e86SMichal Hocko /* force arch specific MAP_FIXED handling in get_unmapped_area */ 1218a4ff8e86SMichal Hocko if (flags & MAP_FIXED_NOREPLACE) 1219a4ff8e86SMichal Hocko flags |= MAP_FIXED; 1220a4ff8e86SMichal Hocko 12217cd94146SEric Paris if (!(flags & MAP_FIXED)) 12227cd94146SEric Paris addr = round_hint_to_min(addr); 12237cd94146SEric Paris 12241da177e4SLinus Torvalds /* Careful about overflows.. */ 12251da177e4SLinus Torvalds len = PAGE_ALIGN(len); 12269206de95SAl Viro if (!len) 12271da177e4SLinus Torvalds return -ENOMEM; 12281da177e4SLinus Torvalds 12291da177e4SLinus Torvalds /* offset overflow? */ 12301da177e4SLinus Torvalds if ((pgoff + (len >> PAGE_SHIFT)) < pgoff) 12311da177e4SLinus Torvalds return -EOVERFLOW; 12321da177e4SLinus Torvalds 12331da177e4SLinus Torvalds /* Too many mappings? */ 12341da177e4SLinus Torvalds if (mm->map_count > sysctl_max_map_count) 12351da177e4SLinus Torvalds return -ENOMEM; 12361da177e4SLinus Torvalds 12371da177e4SLinus Torvalds /* Obtain the address to map to. we verify (or select) it and ensure 12381da177e4SLinus Torvalds * that it represents a valid section of the address space. 12391da177e4SLinus Torvalds */ 12401da177e4SLinus Torvalds addr = get_unmapped_area(file, addr, len, pgoff, flags); 1241ff68dac6SGaowei Pu if (IS_ERR_VALUE(addr)) 12421da177e4SLinus Torvalds return addr; 12431da177e4SLinus Torvalds 1244a4ff8e86SMichal Hocko if (flags & MAP_FIXED_NOREPLACE) { 124535e43c5fSLiam Howlett if (find_vma_intersection(mm, addr, addr + len)) 1246a4ff8e86SMichal Hocko return -EEXIST; 1247a4ff8e86SMichal Hocko } 1248a4ff8e86SMichal Hocko 124962b5f7d0SDave Hansen if (prot == PROT_EXEC) { 125062b5f7d0SDave Hansen pkey = execute_only_pkey(mm); 125162b5f7d0SDave Hansen if (pkey < 0) 125262b5f7d0SDave Hansen pkey = 0; 125362b5f7d0SDave Hansen } 125462b5f7d0SDave Hansen 12551da177e4SLinus Torvalds /* Do simple checking here so the lower-level routines won't have 12561da177e4SLinus Torvalds * to. we assume access permissions have been handled by the open 12571da177e4SLinus Torvalds * of the memory object, so we don't do any here. 12581da177e4SLinus Torvalds */ 125945e55300SPeter Collingbourne vm_flags = calc_vm_prot_bits(prot, pkey) | calc_vm_flag_bits(flags) | 12601da177e4SLinus Torvalds mm->def_flags | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC; 12611da177e4SLinus Torvalds 1262cdf7b341SHuang Shijie if (flags & MAP_LOCKED) 12631da177e4SLinus Torvalds if (!can_do_mlock()) 12641da177e4SLinus Torvalds return -EPERM; 1265ba470de4SRik van Riel 1266363ee17fSDavidlohr Bueso if (mlock_future_check(mm, vm_flags, len)) 12671da177e4SLinus Torvalds return -EAGAIN; 12681da177e4SLinus Torvalds 12691da177e4SLinus Torvalds if (file) { 1270077bf22bSOleg Nesterov struct inode *inode = file_inode(file); 12711c972597SDan Williams unsigned long flags_mask; 12721c972597SDan Williams 1273be83bbf8SLinus Torvalds if (!file_mmap_ok(file, inode, pgoff, len)) 1274be83bbf8SLinus Torvalds return -EOVERFLOW; 1275be83bbf8SLinus Torvalds 12761c972597SDan Williams flags_mask = LEGACY_MAP_MASK | file->f_op->mmap_supported_flags; 1277077bf22bSOleg Nesterov 12781da177e4SLinus Torvalds switch (flags & MAP_TYPE) { 12791da177e4SLinus Torvalds case MAP_SHARED: 12801c972597SDan Williams /* 12811c972597SDan Williams * Force use of MAP_SHARED_VALIDATE with non-legacy 12821c972597SDan Williams * flags. E.g. MAP_SYNC is dangerous to use with 12831c972597SDan Williams * MAP_SHARED as you don't know which consistency model 12841c972597SDan Williams * you will get. We silently ignore unsupported flags 12851c972597SDan Williams * with MAP_SHARED to preserve backward compatibility. 12861c972597SDan Williams */ 12871c972597SDan Williams flags &= LEGACY_MAP_MASK; 1288e4a9bc58SJoe Perches fallthrough; 12891c972597SDan Williams case MAP_SHARED_VALIDATE: 12901c972597SDan Williams if (flags & ~flags_mask) 12911c972597SDan Williams return -EOPNOTSUPP; 1292dc617f29SDarrick J. Wong if (prot & PROT_WRITE) { 1293dc617f29SDarrick J. Wong if (!(file->f_mode & FMODE_WRITE)) 12941da177e4SLinus Torvalds return -EACCES; 1295dc617f29SDarrick J. Wong if (IS_SWAPFILE(file->f_mapping->host)) 1296dc617f29SDarrick J. Wong return -ETXTBSY; 1297dc617f29SDarrick J. Wong } 12981da177e4SLinus Torvalds 12991da177e4SLinus Torvalds /* 13001da177e4SLinus Torvalds * Make sure we don't allow writing to an append-only 13011da177e4SLinus Torvalds * file.. 13021da177e4SLinus Torvalds */ 13031da177e4SLinus Torvalds if (IS_APPEND(inode) && (file->f_mode & FMODE_WRITE)) 13041da177e4SLinus Torvalds return -EACCES; 13051da177e4SLinus Torvalds 13061da177e4SLinus Torvalds vm_flags |= VM_SHARED | VM_MAYSHARE; 13071da177e4SLinus Torvalds if (!(file->f_mode & FMODE_WRITE)) 13081da177e4SLinus Torvalds vm_flags &= ~(VM_MAYWRITE | VM_SHARED); 1309e4a9bc58SJoe Perches fallthrough; 13101da177e4SLinus Torvalds case MAP_PRIVATE: 13111da177e4SLinus Torvalds if (!(file->f_mode & FMODE_READ)) 13121da177e4SLinus Torvalds return -EACCES; 131390f8572bSEric W. Biederman if (path_noexec(&file->f_path)) { 131480c5606cSLinus Torvalds if (vm_flags & VM_EXEC) 131580c5606cSLinus Torvalds return -EPERM; 131680c5606cSLinus Torvalds vm_flags &= ~VM_MAYEXEC; 131780c5606cSLinus Torvalds } 131880c5606cSLinus Torvalds 131972c2d531SAl Viro if (!file->f_op->mmap) 132080c5606cSLinus Torvalds return -ENODEV; 1321b2c56e4fSOleg Nesterov if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP)) 1322b2c56e4fSOleg Nesterov return -EINVAL; 13231da177e4SLinus Torvalds break; 13241da177e4SLinus Torvalds 13251da177e4SLinus Torvalds default: 13261da177e4SLinus Torvalds return -EINVAL; 13271da177e4SLinus Torvalds } 13281da177e4SLinus Torvalds } else { 13291da177e4SLinus Torvalds switch (flags & MAP_TYPE) { 13301da177e4SLinus Torvalds case MAP_SHARED: 1331b2c56e4fSOleg Nesterov if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP)) 1332b2c56e4fSOleg Nesterov return -EINVAL; 1333ce363942STejun Heo /* 1334ce363942STejun Heo * Ignore pgoff. 1335ce363942STejun Heo */ 1336ce363942STejun Heo pgoff = 0; 13371da177e4SLinus Torvalds vm_flags |= VM_SHARED | VM_MAYSHARE; 13381da177e4SLinus Torvalds break; 13391da177e4SLinus Torvalds case MAP_PRIVATE: 13401da177e4SLinus Torvalds /* 13411da177e4SLinus Torvalds * Set pgoff according to addr for anon_vma. 13421da177e4SLinus Torvalds */ 13431da177e4SLinus Torvalds pgoff = addr >> PAGE_SHIFT; 13441da177e4SLinus Torvalds break; 13451da177e4SLinus Torvalds default: 13461da177e4SLinus Torvalds return -EINVAL; 13471da177e4SLinus Torvalds } 13481da177e4SLinus Torvalds } 13491da177e4SLinus Torvalds 1350c22c0d63SMichel Lespinasse /* 1351c22c0d63SMichel Lespinasse * Set 'VM_NORESERVE' if we should not account for the 1352c22c0d63SMichel Lespinasse * memory use of this mapping. 1353c22c0d63SMichel Lespinasse */ 1354c22c0d63SMichel Lespinasse if (flags & MAP_NORESERVE) { 1355c22c0d63SMichel Lespinasse /* We honor MAP_NORESERVE if allowed to overcommit */ 1356c22c0d63SMichel Lespinasse if (sysctl_overcommit_memory != OVERCOMMIT_NEVER) 1357c22c0d63SMichel Lespinasse vm_flags |= VM_NORESERVE; 1358c22c0d63SMichel Lespinasse 1359c22c0d63SMichel Lespinasse /* hugetlb applies strict overcommit unless MAP_NORESERVE */ 1360c22c0d63SMichel Lespinasse if (file && is_file_hugepages(file)) 1361c22c0d63SMichel Lespinasse vm_flags |= VM_NORESERVE; 1362c22c0d63SMichel Lespinasse } 1363c22c0d63SMichel Lespinasse 1364897ab3e0SMike Rapoport addr = mmap_region(file, addr, len, vm_flags, pgoff, uf); 136509a9f1d2SMichel Lespinasse if (!IS_ERR_VALUE(addr) && 136609a9f1d2SMichel Lespinasse ((vm_flags & VM_LOCKED) || 136709a9f1d2SMichel Lespinasse (flags & (MAP_POPULATE | MAP_NONBLOCK)) == MAP_POPULATE)) 136841badc15SMichel Lespinasse *populate = len; 1369bebeb3d6SMichel Lespinasse return addr; 13700165ab44SMiklos Szeredi } 13716be5ceb0SLinus Torvalds 1372a90f590aSDominik Brodowski unsigned long ksys_mmap_pgoff(unsigned long addr, unsigned long len, 1373a90f590aSDominik Brodowski unsigned long prot, unsigned long flags, 1374a90f590aSDominik Brodowski unsigned long fd, unsigned long pgoff) 137566f0dc48SHugh Dickins { 137666f0dc48SHugh Dickins struct file *file = NULL; 13771e3ee14bSChen Gang unsigned long retval; 137866f0dc48SHugh Dickins 137966f0dc48SHugh Dickins if (!(flags & MAP_ANONYMOUS)) { 1380120a795dSAl Viro audit_mmap_fd(fd, flags); 138166f0dc48SHugh Dickins file = fget(fd); 138266f0dc48SHugh Dickins if (!file) 13831e3ee14bSChen Gang return -EBADF; 13847bba8f0eSZhen Lei if (is_file_hugepages(file)) { 1385af73e4d9SNaoya Horiguchi len = ALIGN(len, huge_page_size(hstate_file(file))); 13867bba8f0eSZhen Lei } else if (unlikely(flags & MAP_HUGETLB)) { 1387493af578SJörn Engel retval = -EINVAL; 1388493af578SJörn Engel goto out_fput; 13897bba8f0eSZhen Lei } 139066f0dc48SHugh Dickins } else if (flags & MAP_HUGETLB) { 1391c103a4dcSAndrew Morton struct hstate *hs; 1392af73e4d9SNaoya Horiguchi 139320ac2893SAnshuman Khandual hs = hstate_sizelog((flags >> MAP_HUGE_SHIFT) & MAP_HUGE_MASK); 1394091d0d55SLi Zefan if (!hs) 1395091d0d55SLi Zefan return -EINVAL; 1396091d0d55SLi Zefan 1397091d0d55SLi Zefan len = ALIGN(len, huge_page_size(hs)); 139866f0dc48SHugh Dickins /* 139966f0dc48SHugh Dickins * VM_NORESERVE is used because the reservations will be 140066f0dc48SHugh Dickins * taken when vm_ops->mmap() is called 140166f0dc48SHugh Dickins */ 1402af73e4d9SNaoya Horiguchi file = hugetlb_file_setup(HUGETLB_ANON_FILE, len, 140342d7395fSAndi Kleen VM_NORESERVE, 140483c1fd76Szhangyiru HUGETLB_ANONHUGE_INODE, 140542d7395fSAndi Kleen (flags >> MAP_HUGE_SHIFT) & MAP_HUGE_MASK); 140666f0dc48SHugh Dickins if (IS_ERR(file)) 140766f0dc48SHugh Dickins return PTR_ERR(file); 140866f0dc48SHugh Dickins } 140966f0dc48SHugh Dickins 14109fbeb5abSMichal Hocko retval = vm_mmap_pgoff(file, addr, len, prot, flags, pgoff); 1411493af578SJörn Engel out_fput: 141266f0dc48SHugh Dickins if (file) 141366f0dc48SHugh Dickins fput(file); 141466f0dc48SHugh Dickins return retval; 141566f0dc48SHugh Dickins } 141666f0dc48SHugh Dickins 1417a90f590aSDominik Brodowski SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len, 1418a90f590aSDominik Brodowski unsigned long, prot, unsigned long, flags, 1419a90f590aSDominik Brodowski unsigned long, fd, unsigned long, pgoff) 1420a90f590aSDominik Brodowski { 1421a90f590aSDominik Brodowski return ksys_mmap_pgoff(addr, len, prot, flags, fd, pgoff); 1422a90f590aSDominik Brodowski } 1423a90f590aSDominik Brodowski 1424a4679373SChristoph Hellwig #ifdef __ARCH_WANT_SYS_OLD_MMAP 1425a4679373SChristoph Hellwig struct mmap_arg_struct { 1426a4679373SChristoph Hellwig unsigned long addr; 1427a4679373SChristoph Hellwig unsigned long len; 1428a4679373SChristoph Hellwig unsigned long prot; 1429a4679373SChristoph Hellwig unsigned long flags; 1430a4679373SChristoph Hellwig unsigned long fd; 1431a4679373SChristoph Hellwig unsigned long offset; 1432a4679373SChristoph Hellwig }; 1433a4679373SChristoph Hellwig 1434a4679373SChristoph Hellwig SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg) 1435a4679373SChristoph Hellwig { 1436a4679373SChristoph Hellwig struct mmap_arg_struct a; 1437a4679373SChristoph Hellwig 1438a4679373SChristoph Hellwig if (copy_from_user(&a, arg, sizeof(a))) 1439a4679373SChristoph Hellwig return -EFAULT; 1440de1741a1SAlexander Kuleshov if (offset_in_page(a.offset)) 1441a4679373SChristoph Hellwig return -EINVAL; 1442a4679373SChristoph Hellwig 1443a90f590aSDominik Brodowski return ksys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd, 1444a4679373SChristoph Hellwig a.offset >> PAGE_SHIFT); 1445a4679373SChristoph Hellwig } 1446a4679373SChristoph Hellwig #endif /* __ARCH_WANT_SYS_OLD_MMAP */ 1447a4679373SChristoph Hellwig 14484e950f6fSAlexey Dobriyan /* 14498bb4e7a2SWei Yang * Some shared mappings will want the pages marked read-only 14504e950f6fSAlexey Dobriyan * to track write events. If so, we'll downgrade vm_page_prot 14514e950f6fSAlexey Dobriyan * to the private version (using protection_map[] without the 14524e950f6fSAlexey Dobriyan * VM_SHARED bit). 14534e950f6fSAlexey Dobriyan */ 14546d2329f8SAndrea Arcangeli int vma_wants_writenotify(struct vm_area_struct *vma, pgprot_t vm_page_prot) 14554e950f6fSAlexey Dobriyan { 1456ca16d140SKOSAKI Motohiro vm_flags_t vm_flags = vma->vm_flags; 14578a04446aSKirill A. Shutemov const struct vm_operations_struct *vm_ops = vma->vm_ops; 14584e950f6fSAlexey Dobriyan 14594e950f6fSAlexey Dobriyan /* If it was private or non-writable, the write bit is already clear */ 14604e950f6fSAlexey Dobriyan if ((vm_flags & (VM_WRITE|VM_SHARED)) != ((VM_WRITE|VM_SHARED))) 14614e950f6fSAlexey Dobriyan return 0; 14624e950f6fSAlexey Dobriyan 14634e950f6fSAlexey Dobriyan /* The backer wishes to know when pages are first written to? */ 14648a04446aSKirill A. Shutemov if (vm_ops && (vm_ops->page_mkwrite || vm_ops->pfn_mkwrite)) 14654e950f6fSAlexey Dobriyan return 1; 14664e950f6fSAlexey Dobriyan 146764e45507SPeter Feiner /* The open routine did something to the protections that pgprot_modify 146864e45507SPeter Feiner * won't preserve? */ 14696d2329f8SAndrea Arcangeli if (pgprot_val(vm_page_prot) != 14706d2329f8SAndrea Arcangeli pgprot_val(vm_pgprot_modify(vm_page_prot, vm_flags))) 14714e950f6fSAlexey Dobriyan return 0; 14724e950f6fSAlexey Dobriyan 1473f96f7a40SDavid Hildenbrand /* 1474f96f7a40SDavid Hildenbrand * Do we need to track softdirty? hugetlb does not support softdirty 1475f96f7a40SDavid Hildenbrand * tracking yet. 1476f96f7a40SDavid Hildenbrand */ 1477f96f7a40SDavid Hildenbrand if (vma_soft_dirty_enabled(vma) && !is_vm_hugetlb_page(vma)) 147864e45507SPeter Feiner return 1; 147964e45507SPeter Feiner 148051d3d5ebSDavid Hildenbrand /* Do we need write faults for uffd-wp tracking? */ 148151d3d5ebSDavid Hildenbrand if (userfaultfd_wp(vma)) 148251d3d5ebSDavid Hildenbrand return 1; 148351d3d5ebSDavid Hildenbrand 14844e950f6fSAlexey Dobriyan /* Specialty mapping? */ 14854b6e1e37SKonstantin Khlebnikov if (vm_flags & VM_PFNMAP) 14864e950f6fSAlexey Dobriyan return 0; 14874e950f6fSAlexey Dobriyan 14884e950f6fSAlexey Dobriyan /* Can the mapping track the dirty pages? */ 14894e950f6fSAlexey Dobriyan return vma->vm_file && vma->vm_file->f_mapping && 1490f56753acSChristoph Hellwig mapping_can_writeback(vma->vm_file->f_mapping); 14914e950f6fSAlexey Dobriyan } 14924e950f6fSAlexey Dobriyan 1493fc8744adSLinus Torvalds /* 1494fc8744adSLinus Torvalds * We account for memory if it's a private writeable mapping, 14955a6fe125SMel Gorman * not hugepages and VM_NORESERVE wasn't set. 1496fc8744adSLinus Torvalds */ 1497ca16d140SKOSAKI Motohiro static inline int accountable_mapping(struct file *file, vm_flags_t vm_flags) 1498fc8744adSLinus Torvalds { 14995a6fe125SMel Gorman /* 15005a6fe125SMel Gorman * hugetlb has its own accounting separate from the core VM 15015a6fe125SMel Gorman * VM_HUGETLB may not be set yet so we cannot check for that flag. 15025a6fe125SMel Gorman */ 15035a6fe125SMel Gorman if (file && is_file_hugepages(file)) 15045a6fe125SMel Gorman return 0; 15055a6fe125SMel Gorman 1506fc8744adSLinus Torvalds return (vm_flags & (VM_NORESERVE | VM_SHARED | VM_WRITE)) == VM_WRITE; 1507fc8744adSLinus Torvalds } 1508fc8744adSLinus Torvalds 15093499a131SLiam R. Howlett /** 15103499a131SLiam R. Howlett * unmapped_area() - Find an area between the low_limit and the high_limit with 15113499a131SLiam R. Howlett * the correct alignment and offset, all from @info. Note: current->mm is used 15123499a131SLiam R. Howlett * for the search. 15133499a131SLiam R. Howlett * 151482b24936SVernon Yang * @info: The unmapped area information including the range [low_limit - 151582b24936SVernon Yang * high_limit), the alignment offset and mask. 15163499a131SLiam R. Howlett * 15173499a131SLiam R. Howlett * Return: A memory address or -ENOMEM. 15183499a131SLiam R. Howlett */ 1519baceaf1cSJaewon Kim static unsigned long unmapped_area(struct vm_unmapped_area_info *info) 1520db4fbfb9SMichel Lespinasse { 15213499a131SLiam R. Howlett unsigned long length, gap; 1522db4fbfb9SMichel Lespinasse 15233499a131SLiam R. Howlett MA_STATE(mas, ¤t->mm->mm_mt, 0, 0); 1524db4fbfb9SMichel Lespinasse 1525db4fbfb9SMichel Lespinasse /* Adjust search length to account for worst case alignment overhead */ 1526db4fbfb9SMichel Lespinasse length = info->length + info->align_mask; 1527db4fbfb9SMichel Lespinasse if (length < info->length) 1528db4fbfb9SMichel Lespinasse return -ENOMEM; 1529db4fbfb9SMichel Lespinasse 15303499a131SLiam R. Howlett if (mas_empty_area(&mas, info->low_limit, info->high_limit - 1, 15313499a131SLiam R. Howlett length)) 15323499a131SLiam R. Howlett return -ENOMEM; 15333499a131SLiam R. Howlett 1534d4af56c5SLiam R. Howlett gap = mas.index; 1535d4af56c5SLiam R. Howlett gap += (info->align_offset - gap) & info->align_mask; 15363499a131SLiam R. Howlett return gap; 1537db4fbfb9SMichel Lespinasse } 1538db4fbfb9SMichel Lespinasse 15393499a131SLiam R. Howlett /** 15403499a131SLiam R. Howlett * unmapped_area_topdown() - Find an area between the low_limit and the 154182b24936SVernon Yang * high_limit with the correct alignment and offset at the highest available 15423499a131SLiam R. Howlett * address, all from @info. Note: current->mm is used for the search. 15433499a131SLiam R. Howlett * 154482b24936SVernon Yang * @info: The unmapped area information including the range [low_limit - 154582b24936SVernon Yang * high_limit), the alignment offset and mask. 15463499a131SLiam R. Howlett * 15473499a131SLiam R. Howlett * Return: A memory address or -ENOMEM. 15483499a131SLiam R. Howlett */ 1549baceaf1cSJaewon Kim static unsigned long unmapped_area_topdown(struct vm_unmapped_area_info *info) 1550db4fbfb9SMichel Lespinasse { 15513499a131SLiam R. Howlett unsigned long length, gap; 1552d4af56c5SLiam R. Howlett 15533499a131SLiam R. Howlett MA_STATE(mas, ¤t->mm->mm_mt, 0, 0); 1554db4fbfb9SMichel Lespinasse /* Adjust search length to account for worst case alignment overhead */ 1555db4fbfb9SMichel Lespinasse length = info->length + info->align_mask; 1556db4fbfb9SMichel Lespinasse if (length < info->length) 1557db4fbfb9SMichel Lespinasse return -ENOMEM; 1558db4fbfb9SMichel Lespinasse 15593499a131SLiam R. Howlett if (mas_empty_area_rev(&mas, info->low_limit, info->high_limit - 1, 15603499a131SLiam R. Howlett length)) 15613499a131SLiam R. Howlett return -ENOMEM; 15623499a131SLiam R. Howlett 1563d4af56c5SLiam R. Howlett gap = mas.last + 1 - info->length; 1564d4af56c5SLiam R. Howlett gap -= (gap - info->align_offset) & info->align_mask; 15653499a131SLiam R. Howlett return gap; 1566db4fbfb9SMichel Lespinasse } 1567db4fbfb9SMichel Lespinasse 1568baceaf1cSJaewon Kim /* 1569baceaf1cSJaewon Kim * Search for an unmapped address range. 1570baceaf1cSJaewon Kim * 1571baceaf1cSJaewon Kim * We are looking for a range that: 1572baceaf1cSJaewon Kim * - does not intersect with any VMA; 1573baceaf1cSJaewon Kim * - is contained within the [low_limit, high_limit) interval; 1574baceaf1cSJaewon Kim * - is at least the desired size. 1575baceaf1cSJaewon Kim * - satisfies (begin_addr & align_mask) == (align_offset & align_mask) 1576baceaf1cSJaewon Kim */ 1577baceaf1cSJaewon Kim unsigned long vm_unmapped_area(struct vm_unmapped_area_info *info) 1578baceaf1cSJaewon Kim { 1579df529cabSJaewon Kim unsigned long addr; 1580df529cabSJaewon Kim 1581baceaf1cSJaewon Kim if (info->flags & VM_UNMAPPED_AREA_TOPDOWN) 1582df529cabSJaewon Kim addr = unmapped_area_topdown(info); 1583baceaf1cSJaewon Kim else 1584df529cabSJaewon Kim addr = unmapped_area(info); 1585df529cabSJaewon Kim 1586df529cabSJaewon Kim trace_vm_unmapped_area(addr, info); 1587df529cabSJaewon Kim return addr; 1588baceaf1cSJaewon Kim } 1589f6795053SSteve Capper 15901da177e4SLinus Torvalds /* Get an address range which is currently unmapped. 15911da177e4SLinus Torvalds * For shmat() with addr=0. 15921da177e4SLinus Torvalds * 15931da177e4SLinus Torvalds * Ugly calling convention alert: 15941da177e4SLinus Torvalds * Return value with the low bits set means error value, 15951da177e4SLinus Torvalds * ie 15961da177e4SLinus Torvalds * if (ret & ~PAGE_MASK) 15971da177e4SLinus Torvalds * error = ret; 15981da177e4SLinus Torvalds * 15991da177e4SLinus Torvalds * This function "knows" that -ENOMEM has the bits set. 16001da177e4SLinus Torvalds */ 16011da177e4SLinus Torvalds unsigned long 16024b439e25SChristophe Leroy generic_get_unmapped_area(struct file *filp, unsigned long addr, 16034b439e25SChristophe Leroy unsigned long len, unsigned long pgoff, 16044b439e25SChristophe Leroy unsigned long flags) 16051da177e4SLinus Torvalds { 16061da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 16071be7107fSHugh Dickins struct vm_area_struct *vma, *prev; 1608db4fbfb9SMichel Lespinasse struct vm_unmapped_area_info info; 16092cb4de08SChristophe Leroy const unsigned long mmap_end = arch_get_mmap_end(addr, len, flags); 16101da177e4SLinus Torvalds 1611f6795053SSteve Capper if (len > mmap_end - mmap_min_addr) 16121da177e4SLinus Torvalds return -ENOMEM; 16131da177e4SLinus Torvalds 161406abdfb4SBenjamin Herrenschmidt if (flags & MAP_FIXED) 161506abdfb4SBenjamin Herrenschmidt return addr; 161606abdfb4SBenjamin Herrenschmidt 16171da177e4SLinus Torvalds if (addr) { 16181da177e4SLinus Torvalds addr = PAGE_ALIGN(addr); 16191be7107fSHugh Dickins vma = find_vma_prev(mm, addr, &prev); 1620f6795053SSteve Capper if (mmap_end - len >= addr && addr >= mmap_min_addr && 16211be7107fSHugh Dickins (!vma || addr + len <= vm_start_gap(vma)) && 16221be7107fSHugh Dickins (!prev || addr >= vm_end_gap(prev))) 16231da177e4SLinus Torvalds return addr; 16241da177e4SLinus Torvalds } 16251da177e4SLinus Torvalds 1626db4fbfb9SMichel Lespinasse info.flags = 0; 1627db4fbfb9SMichel Lespinasse info.length = len; 16284e99b021SHeiko Carstens info.low_limit = mm->mmap_base; 1629f6795053SSteve Capper info.high_limit = mmap_end; 1630db4fbfb9SMichel Lespinasse info.align_mask = 0; 163109ef5283SJaewon Kim info.align_offset = 0; 1632db4fbfb9SMichel Lespinasse return vm_unmapped_area(&info); 16331da177e4SLinus Torvalds } 16344b439e25SChristophe Leroy 16354b439e25SChristophe Leroy #ifndef HAVE_ARCH_UNMAPPED_AREA 16364b439e25SChristophe Leroy unsigned long 16374b439e25SChristophe Leroy arch_get_unmapped_area(struct file *filp, unsigned long addr, 16384b439e25SChristophe Leroy unsigned long len, unsigned long pgoff, 16394b439e25SChristophe Leroy unsigned long flags) 16404b439e25SChristophe Leroy { 16414b439e25SChristophe Leroy return generic_get_unmapped_area(filp, addr, len, pgoff, flags); 16424b439e25SChristophe Leroy } 16431da177e4SLinus Torvalds #endif 16441da177e4SLinus Torvalds 16451da177e4SLinus Torvalds /* 16461da177e4SLinus Torvalds * This mmap-allocator allocates new areas top-down from below the 16471da177e4SLinus Torvalds * stack's low limit (the base): 16481da177e4SLinus Torvalds */ 16491da177e4SLinus Torvalds unsigned long 16504b439e25SChristophe Leroy generic_get_unmapped_area_topdown(struct file *filp, unsigned long addr, 165143cca0b1SYang Fan unsigned long len, unsigned long pgoff, 165243cca0b1SYang Fan unsigned long flags) 16531da177e4SLinus Torvalds { 16541be7107fSHugh Dickins struct vm_area_struct *vma, *prev; 16551da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 1656db4fbfb9SMichel Lespinasse struct vm_unmapped_area_info info; 16572cb4de08SChristophe Leroy const unsigned long mmap_end = arch_get_mmap_end(addr, len, flags); 16581da177e4SLinus Torvalds 16591da177e4SLinus Torvalds /* requested length too big for entire address space */ 1660f6795053SSteve Capper if (len > mmap_end - mmap_min_addr) 16611da177e4SLinus Torvalds return -ENOMEM; 16621da177e4SLinus Torvalds 166306abdfb4SBenjamin Herrenschmidt if (flags & MAP_FIXED) 166406abdfb4SBenjamin Herrenschmidt return addr; 166506abdfb4SBenjamin Herrenschmidt 16661da177e4SLinus Torvalds /* requesting a specific address */ 16671da177e4SLinus Torvalds if (addr) { 16681da177e4SLinus Torvalds addr = PAGE_ALIGN(addr); 16691be7107fSHugh Dickins vma = find_vma_prev(mm, addr, &prev); 1670f6795053SSteve Capper if (mmap_end - len >= addr && addr >= mmap_min_addr && 16711be7107fSHugh Dickins (!vma || addr + len <= vm_start_gap(vma)) && 16721be7107fSHugh Dickins (!prev || addr >= vm_end_gap(prev))) 16731da177e4SLinus Torvalds return addr; 16741da177e4SLinus Torvalds } 16751da177e4SLinus Torvalds 1676db4fbfb9SMichel Lespinasse info.flags = VM_UNMAPPED_AREA_TOPDOWN; 1677db4fbfb9SMichel Lespinasse info.length = len; 16782afc745fSAkira Takeuchi info.low_limit = max(PAGE_SIZE, mmap_min_addr); 1679f6795053SSteve Capper info.high_limit = arch_get_mmap_base(addr, mm->mmap_base); 1680db4fbfb9SMichel Lespinasse info.align_mask = 0; 168109ef5283SJaewon Kim info.align_offset = 0; 1682db4fbfb9SMichel Lespinasse addr = vm_unmapped_area(&info); 1683b716ad95SXiao Guangrong 16841da177e4SLinus Torvalds /* 16851da177e4SLinus Torvalds * A failed mmap() very likely causes application failure, 16861da177e4SLinus Torvalds * so fall back to the bottom-up function here. This scenario 16871da177e4SLinus Torvalds * can happen with large stack limits and large mmap() 16881da177e4SLinus Torvalds * allocations. 16891da177e4SLinus Torvalds */ 1690de1741a1SAlexander Kuleshov if (offset_in_page(addr)) { 1691db4fbfb9SMichel Lespinasse VM_BUG_ON(addr != -ENOMEM); 1692db4fbfb9SMichel Lespinasse info.flags = 0; 1693db4fbfb9SMichel Lespinasse info.low_limit = TASK_UNMAPPED_BASE; 1694f6795053SSteve Capper info.high_limit = mmap_end; 1695db4fbfb9SMichel Lespinasse addr = vm_unmapped_area(&info); 1696db4fbfb9SMichel Lespinasse } 16971da177e4SLinus Torvalds 16981da177e4SLinus Torvalds return addr; 16991da177e4SLinus Torvalds } 17004b439e25SChristophe Leroy 17014b439e25SChristophe Leroy #ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN 17024b439e25SChristophe Leroy unsigned long 17034b439e25SChristophe Leroy arch_get_unmapped_area_topdown(struct file *filp, unsigned long addr, 17044b439e25SChristophe Leroy unsigned long len, unsigned long pgoff, 17054b439e25SChristophe Leroy unsigned long flags) 17064b439e25SChristophe Leroy { 17074b439e25SChristophe Leroy return generic_get_unmapped_area_topdown(filp, addr, len, pgoff, flags); 17084b439e25SChristophe Leroy } 17091da177e4SLinus Torvalds #endif 17101da177e4SLinus Torvalds 17111da177e4SLinus Torvalds unsigned long 17121da177e4SLinus Torvalds get_unmapped_area(struct file *file, unsigned long addr, unsigned long len, 17131da177e4SLinus Torvalds unsigned long pgoff, unsigned long flags) 17141da177e4SLinus Torvalds { 171506abdfb4SBenjamin Herrenschmidt unsigned long (*get_area)(struct file *, unsigned long, 171606abdfb4SBenjamin Herrenschmidt unsigned long, unsigned long, unsigned long); 171707ab67c8SLinus Torvalds 17189206de95SAl Viro unsigned long error = arch_mmap_check(addr, len, flags); 17199206de95SAl Viro if (error) 17209206de95SAl Viro return error; 17219206de95SAl Viro 17229206de95SAl Viro /* Careful about overflows.. */ 17239206de95SAl Viro if (len > TASK_SIZE) 17249206de95SAl Viro return -ENOMEM; 17259206de95SAl Viro 172607ab67c8SLinus Torvalds get_area = current->mm->get_unmapped_area; 1727c01d5b30SHugh Dickins if (file) { 1728c01d5b30SHugh Dickins if (file->f_op->get_unmapped_area) 172907ab67c8SLinus Torvalds get_area = file->f_op->get_unmapped_area; 1730c01d5b30SHugh Dickins } else if (flags & MAP_SHARED) { 1731c01d5b30SHugh Dickins /* 1732c01d5b30SHugh Dickins * mmap_region() will call shmem_zero_setup() to create a file, 1733c01d5b30SHugh Dickins * so use shmem's get_unmapped_area in case it can be huge. 173445e55300SPeter Collingbourne * do_mmap() will clear pgoff, so match alignment. 1735c01d5b30SHugh Dickins */ 1736c01d5b30SHugh Dickins pgoff = 0; 1737c01d5b30SHugh Dickins get_area = shmem_get_unmapped_area; 1738c01d5b30SHugh Dickins } 1739c01d5b30SHugh Dickins 174007ab67c8SLinus Torvalds addr = get_area(file, addr, len, pgoff, flags); 174107ab67c8SLinus Torvalds if (IS_ERR_VALUE(addr)) 174207ab67c8SLinus Torvalds return addr; 174307ab67c8SLinus Torvalds 17441da177e4SLinus Torvalds if (addr > TASK_SIZE - len) 17451da177e4SLinus Torvalds return -ENOMEM; 1746de1741a1SAlexander Kuleshov if (offset_in_page(addr)) 17471da177e4SLinus Torvalds return -EINVAL; 174806abdfb4SBenjamin Herrenschmidt 17499ac4ed4bSAl Viro error = security_mmap_addr(addr); 17509ac4ed4bSAl Viro return error ? error : addr; 17511da177e4SLinus Torvalds } 17521da177e4SLinus Torvalds 17531da177e4SLinus Torvalds EXPORT_SYMBOL(get_unmapped_area); 17541da177e4SLinus Torvalds 1755be8432e7SLiam R. Howlett /** 1756abdba2ddSLiam R. Howlett * find_vma_intersection() - Look up the first VMA which intersects the interval 1757abdba2ddSLiam R. Howlett * @mm: The process address space. 1758abdba2ddSLiam R. Howlett * @start_addr: The inclusive start user address. 1759abdba2ddSLiam R. Howlett * @end_addr: The exclusive end user address. 1760abdba2ddSLiam R. Howlett * 1761abdba2ddSLiam R. Howlett * Returns: The first VMA within the provided range, %NULL otherwise. Assumes 1762abdba2ddSLiam R. Howlett * start_addr < end_addr. 1763abdba2ddSLiam R. Howlett */ 1764abdba2ddSLiam R. Howlett struct vm_area_struct *find_vma_intersection(struct mm_struct *mm, 1765abdba2ddSLiam R. Howlett unsigned long start_addr, 1766abdba2ddSLiam R. Howlett unsigned long end_addr) 1767abdba2ddSLiam R. Howlett { 1768abdba2ddSLiam R. Howlett unsigned long index = start_addr; 1769abdba2ddSLiam R. Howlett 1770abdba2ddSLiam R. Howlett mmap_assert_locked(mm); 17717964cf8cSLiam R. Howlett return mt_find(&mm->mm_mt, &index, end_addr - 1); 1772abdba2ddSLiam R. Howlett } 1773abdba2ddSLiam R. Howlett EXPORT_SYMBOL(find_vma_intersection); 1774abdba2ddSLiam R. Howlett 1775abdba2ddSLiam R. Howlett /** 1776be8432e7SLiam R. Howlett * find_vma() - Find the VMA for a given address, or the next VMA. 1777be8432e7SLiam R. Howlett * @mm: The mm_struct to check 1778be8432e7SLiam R. Howlett * @addr: The address 1779be8432e7SLiam R. Howlett * 1780be8432e7SLiam R. Howlett * Returns: The VMA associated with addr, or the next VMA. 1781be8432e7SLiam R. Howlett * May return %NULL in the case of no VMA at addr or above. 1782be8432e7SLiam R. Howlett */ 17831da177e4SLinus Torvalds struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr) 17841da177e4SLinus Torvalds { 1785be8432e7SLiam R. Howlett unsigned long index = addr; 17861da177e4SLinus Torvalds 17875b78ed24SLuigi Rizzo mmap_assert_locked(mm); 17887964cf8cSLiam R. Howlett return mt_find(&mm->mm_mt, &index, ULONG_MAX); 17891da177e4SLinus Torvalds } 17901da177e4SLinus Torvalds EXPORT_SYMBOL(find_vma); 17911da177e4SLinus Torvalds 17927fdbd37dSLiam R. Howlett /** 17937fdbd37dSLiam R. Howlett * find_vma_prev() - Find the VMA for a given address, or the next vma and 17947fdbd37dSLiam R. Howlett * set %pprev to the previous VMA, if any. 17957fdbd37dSLiam R. Howlett * @mm: The mm_struct to check 17967fdbd37dSLiam R. Howlett * @addr: The address 17977fdbd37dSLiam R. Howlett * @pprev: The pointer to set to the previous VMA 17987fdbd37dSLiam R. Howlett * 17997fdbd37dSLiam R. Howlett * Note that RCU lock is missing here since the external mmap_lock() is used 18007fdbd37dSLiam R. Howlett * instead. 18017fdbd37dSLiam R. Howlett * 18027fdbd37dSLiam R. Howlett * Returns: The VMA associated with @addr, or the next vma. 18037fdbd37dSLiam R. Howlett * May return %NULL in the case of no vma at addr or above. 18046bd4837dSKOSAKI Motohiro */ 18051da177e4SLinus Torvalds struct vm_area_struct * 18061da177e4SLinus Torvalds find_vma_prev(struct mm_struct *mm, unsigned long addr, 18071da177e4SLinus Torvalds struct vm_area_struct **pprev) 18081da177e4SLinus Torvalds { 18096bd4837dSKOSAKI Motohiro struct vm_area_struct *vma; 18107fdbd37dSLiam R. Howlett MA_STATE(mas, &mm->mm_mt, addr, addr); 18111da177e4SLinus Torvalds 18127fdbd37dSLiam R. Howlett vma = mas_walk(&mas); 18137fdbd37dSLiam R. Howlett *pprev = mas_prev(&mas, 0); 18147fdbd37dSLiam R. Howlett if (!vma) 18157fdbd37dSLiam R. Howlett vma = mas_next(&mas, ULONG_MAX); 18166bd4837dSKOSAKI Motohiro return vma; 18171da177e4SLinus Torvalds } 18181da177e4SLinus Torvalds 18191da177e4SLinus Torvalds /* 18201da177e4SLinus Torvalds * Verify that the stack growth is acceptable and 18211da177e4SLinus Torvalds * update accounting. This is shared with both the 18221da177e4SLinus Torvalds * grow-up and grow-down cases. 18231da177e4SLinus Torvalds */ 18241be7107fSHugh Dickins static int acct_stack_growth(struct vm_area_struct *vma, 18251be7107fSHugh Dickins unsigned long size, unsigned long grow) 18261da177e4SLinus Torvalds { 18271da177e4SLinus Torvalds struct mm_struct *mm = vma->vm_mm; 18281be7107fSHugh Dickins unsigned long new_start; 18291da177e4SLinus Torvalds 18301da177e4SLinus Torvalds /* address space limit tests */ 183184638335SKonstantin Khlebnikov if (!may_expand_vm(mm, vma->vm_flags, grow)) 18321da177e4SLinus Torvalds return -ENOMEM; 18331da177e4SLinus Torvalds 18341da177e4SLinus Torvalds /* Stack limit test */ 183524c79d8eSKrzysztof Opasiak if (size > rlimit(RLIMIT_STACK)) 18361da177e4SLinus Torvalds return -ENOMEM; 18371da177e4SLinus Torvalds 18381da177e4SLinus Torvalds /* mlock limit tests */ 1839c5d8a364SMiaohe Lin if (mlock_future_check(mm, vma->vm_flags, grow << PAGE_SHIFT)) 18401da177e4SLinus Torvalds return -ENOMEM; 18411da177e4SLinus Torvalds 18420d59a01bSAdam Litke /* Check to ensure the stack will not grow into a hugetlb-only region */ 18430d59a01bSAdam Litke new_start = (vma->vm_flags & VM_GROWSUP) ? vma->vm_start : 18440d59a01bSAdam Litke vma->vm_end - size; 18450d59a01bSAdam Litke if (is_hugepage_only_range(vma->vm_mm, new_start, size)) 18460d59a01bSAdam Litke return -EFAULT; 18470d59a01bSAdam Litke 18481da177e4SLinus Torvalds /* 18491da177e4SLinus Torvalds * Overcommit.. This must be the final test, as it will 18501da177e4SLinus Torvalds * update security statistics. 18511da177e4SLinus Torvalds */ 185205fa199dSHugh Dickins if (security_vm_enough_memory_mm(mm, grow)) 18531da177e4SLinus Torvalds return -ENOMEM; 18541da177e4SLinus Torvalds 18551da177e4SLinus Torvalds return 0; 18561da177e4SLinus Torvalds } 18571da177e4SLinus Torvalds 185846dea3d0SHugh Dickins #if defined(CONFIG_STACK_GROWSUP) || defined(CONFIG_IA64) 18591da177e4SLinus Torvalds /* 186046dea3d0SHugh Dickins * PA-RISC uses this for its stack; IA64 for its Register Backing Store. 186146dea3d0SHugh Dickins * vma is the last one with address > vma->vm_end. Have to extend vma. 18621da177e4SLinus Torvalds */ 186346dea3d0SHugh Dickins int expand_upwards(struct vm_area_struct *vma, unsigned long address) 18641da177e4SLinus Torvalds { 186509357814SOleg Nesterov struct mm_struct *mm = vma->vm_mm; 18661be7107fSHugh Dickins struct vm_area_struct *next; 18671be7107fSHugh Dickins unsigned long gap_addr; 186812352d3cSKonstantin Khlebnikov int error = 0; 1869d4af56c5SLiam R. Howlett MA_STATE(mas, &mm->mm_mt, 0, 0); 18701da177e4SLinus Torvalds 18711da177e4SLinus Torvalds if (!(vma->vm_flags & VM_GROWSUP)) 18721da177e4SLinus Torvalds return -EFAULT; 18731da177e4SLinus Torvalds 1874bd726c90SHelge Deller /* Guard against exceeding limits of the address space. */ 18751be7107fSHugh Dickins address &= PAGE_MASK; 187637511fb5SHelge Deller if (address >= (TASK_SIZE & PAGE_MASK)) 187712352d3cSKonstantin Khlebnikov return -ENOMEM; 1878bd726c90SHelge Deller address += PAGE_SIZE; 187912352d3cSKonstantin Khlebnikov 18801be7107fSHugh Dickins /* Enforce stack_guard_gap */ 18811be7107fSHugh Dickins gap_addr = address + stack_guard_gap; 1882bd726c90SHelge Deller 1883bd726c90SHelge Deller /* Guard against overflow */ 1884bd726c90SHelge Deller if (gap_addr < address || gap_addr > TASK_SIZE) 1885bd726c90SHelge Deller gap_addr = TASK_SIZE; 1886bd726c90SHelge Deller 1887763ecb03SLiam R. Howlett next = find_vma_intersection(mm, vma->vm_end, gap_addr); 1888763ecb03SLiam R. Howlett if (next && vma_is_accessible(next)) { 18891be7107fSHugh Dickins if (!(next->vm_flags & VM_GROWSUP)) 18901be7107fSHugh Dickins return -ENOMEM; 18911be7107fSHugh Dickins /* Check that both stack segments have the same anon_vma? */ 18921be7107fSHugh Dickins } 18931be7107fSHugh Dickins 1894c5d5546eSVernon Yang if (mas_preallocate(&mas, GFP_KERNEL)) 18951da177e4SLinus Torvalds return -ENOMEM; 18961da177e4SLinus Torvalds 1897d4af56c5SLiam R. Howlett /* We must make sure the anon_vma is allocated. */ 1898d4af56c5SLiam R. Howlett if (unlikely(anon_vma_prepare(vma))) { 1899d4af56c5SLiam R. Howlett mas_destroy(&mas); 1900d4af56c5SLiam R. Howlett return -ENOMEM; 1901d4af56c5SLiam R. Howlett } 1902d4af56c5SLiam R. Howlett 19031da177e4SLinus Torvalds /* 19041da177e4SLinus Torvalds * vma->vm_start/vm_end cannot change under us because the caller 1905c1e8d7c6SMichel Lespinasse * is required to hold the mmap_lock in read mode. We need the 19061da177e4SLinus Torvalds * anon_vma lock to serialize against concurrent expand_stacks. 19071da177e4SLinus Torvalds */ 190812352d3cSKonstantin Khlebnikov anon_vma_lock_write(vma->anon_vma); 19091da177e4SLinus Torvalds 19101da177e4SLinus Torvalds /* Somebody else might have raced and expanded it already */ 19111da177e4SLinus Torvalds if (address > vma->vm_end) { 19121da177e4SLinus Torvalds unsigned long size, grow; 19131da177e4SLinus Torvalds 19141da177e4SLinus Torvalds size = address - vma->vm_start; 19151da177e4SLinus Torvalds grow = (address - vma->vm_end) >> PAGE_SHIFT; 19161da177e4SLinus Torvalds 191742c36f63SHugh Dickins error = -ENOMEM; 191842c36f63SHugh Dickins if (vma->vm_pgoff + (size >> PAGE_SHIFT) >= vma->vm_pgoff) { 19191da177e4SLinus Torvalds error = acct_stack_growth(vma, size, grow); 19203af9e859SEric B Munson if (!error) { 19214128997bSMichel Lespinasse /* 1922524e00b3SLiam R. Howlett * We only hold a shared mmap_lock lock here, so 1923524e00b3SLiam R. Howlett * we need to protect against concurrent vma 1924524e00b3SLiam R. Howlett * expansions. anon_vma_lock_write() doesn't 1925524e00b3SLiam R. Howlett * help here, as we don't guarantee that all 1926524e00b3SLiam R. Howlett * growable vmas in a mm share the same root 1927524e00b3SLiam R. Howlett * anon vma. So, we reuse mm->page_table_lock 1928524e00b3SLiam R. Howlett * to guard against concurrent vma expansions. 19294128997bSMichel Lespinasse */ 193009357814SOleg Nesterov spin_lock(&mm->page_table_lock); 193187e8827bSOleg Nesterov if (vma->vm_flags & VM_LOCKED) 193209357814SOleg Nesterov mm->locked_vm += grow; 193384638335SKonstantin Khlebnikov vm_stat_account(mm, vma->vm_flags, grow); 1934bf181b9fSMichel Lespinasse anon_vma_interval_tree_pre_update_vma(vma); 19351da177e4SLinus Torvalds vma->vm_end = address; 1936d4af56c5SLiam R. Howlett /* Overwrite old entry in mtree. */ 1937fbcc3104SLiam R. Howlett mas_set_range(&mas, vma->vm_start, address - 1); 1938fbcc3104SLiam R. Howlett mas_store_prealloc(&mas, vma); 1939bf181b9fSMichel Lespinasse anon_vma_interval_tree_post_update_vma(vma); 194009357814SOleg Nesterov spin_unlock(&mm->page_table_lock); 19414128997bSMichel Lespinasse 19423af9e859SEric B Munson perf_event_mmap(vma); 19433af9e859SEric B Munson } 19441da177e4SLinus Torvalds } 194542c36f63SHugh Dickins } 194612352d3cSKonstantin Khlebnikov anon_vma_unlock_write(vma->anon_vma); 1947c791576cSYang Shi khugepaged_enter_vma(vma, vma->vm_flags); 1948d4af56c5SLiam R. Howlett mas_destroy(&mas); 19491da177e4SLinus Torvalds return error; 19501da177e4SLinus Torvalds } 195146dea3d0SHugh Dickins #endif /* CONFIG_STACK_GROWSUP || CONFIG_IA64 */ 195246dea3d0SHugh Dickins 19531da177e4SLinus Torvalds /* 19541da177e4SLinus Torvalds * vma is the first one with address < vma->vm_start. Have to extend vma. 19551da177e4SLinus Torvalds */ 1956524e00b3SLiam R. Howlett int expand_downwards(struct vm_area_struct *vma, unsigned long address) 19571da177e4SLinus Torvalds { 195809357814SOleg Nesterov struct mm_struct *mm = vma->vm_mm; 1959763ecb03SLiam R. Howlett MA_STATE(mas, &mm->mm_mt, vma->vm_start, vma->vm_start); 19601be7107fSHugh Dickins struct vm_area_struct *prev; 19610a1d5299SJann Horn int error = 0; 19621da177e4SLinus Torvalds 19638869477aSEric Paris address &= PAGE_MASK; 19640a1d5299SJann Horn if (address < mmap_min_addr) 19650a1d5299SJann Horn return -EPERM; 19668869477aSEric Paris 19671be7107fSHugh Dickins /* Enforce stack_guard_gap */ 1968763ecb03SLiam R. Howlett prev = mas_prev(&mas, 0); 19691be7107fSHugh Dickins /* Check that both stack segments have the same anon_vma? */ 197032e4e6d5SOleg Nesterov if (prev && !(prev->vm_flags & VM_GROWSDOWN) && 19713122e80eSAnshuman Khandual vma_is_accessible(prev)) { 197232e4e6d5SOleg Nesterov if (address - prev->vm_end < stack_guard_gap) 197332e4e6d5SOleg Nesterov return -ENOMEM; 19741be7107fSHugh Dickins } 19751be7107fSHugh Dickins 1976c5d5546eSVernon Yang if (mas_preallocate(&mas, GFP_KERNEL)) 197712352d3cSKonstantin Khlebnikov return -ENOMEM; 19781da177e4SLinus Torvalds 1979d4af56c5SLiam R. Howlett /* We must make sure the anon_vma is allocated. */ 1980d4af56c5SLiam R. Howlett if (unlikely(anon_vma_prepare(vma))) { 1981d4af56c5SLiam R. Howlett mas_destroy(&mas); 1982d4af56c5SLiam R. Howlett return -ENOMEM; 1983d4af56c5SLiam R. Howlett } 1984d4af56c5SLiam R. Howlett 19851da177e4SLinus Torvalds /* 19861da177e4SLinus Torvalds * vma->vm_start/vm_end cannot change under us because the caller 1987c1e8d7c6SMichel Lespinasse * is required to hold the mmap_lock in read mode. We need the 19881da177e4SLinus Torvalds * anon_vma lock to serialize against concurrent expand_stacks. 19891da177e4SLinus Torvalds */ 199012352d3cSKonstantin Khlebnikov anon_vma_lock_write(vma->anon_vma); 19911da177e4SLinus Torvalds 19921da177e4SLinus Torvalds /* Somebody else might have raced and expanded it already */ 19931da177e4SLinus Torvalds if (address < vma->vm_start) { 19941da177e4SLinus Torvalds unsigned long size, grow; 19951da177e4SLinus Torvalds 19961da177e4SLinus Torvalds size = vma->vm_end - address; 19971da177e4SLinus Torvalds grow = (vma->vm_start - address) >> PAGE_SHIFT; 19981da177e4SLinus Torvalds 1999a626ca6aSLinus Torvalds error = -ENOMEM; 2000a626ca6aSLinus Torvalds if (grow <= vma->vm_pgoff) { 20011da177e4SLinus Torvalds error = acct_stack_growth(vma, size, grow); 20021da177e4SLinus Torvalds if (!error) { 20034128997bSMichel Lespinasse /* 2004524e00b3SLiam R. Howlett * We only hold a shared mmap_lock lock here, so 2005524e00b3SLiam R. Howlett * we need to protect against concurrent vma 2006524e00b3SLiam R. Howlett * expansions. anon_vma_lock_write() doesn't 2007524e00b3SLiam R. Howlett * help here, as we don't guarantee that all 2008524e00b3SLiam R. Howlett * growable vmas in a mm share the same root 2009524e00b3SLiam R. Howlett * anon vma. So, we reuse mm->page_table_lock 2010524e00b3SLiam R. Howlett * to guard against concurrent vma expansions. 20114128997bSMichel Lespinasse */ 201209357814SOleg Nesterov spin_lock(&mm->page_table_lock); 201387e8827bSOleg Nesterov if (vma->vm_flags & VM_LOCKED) 201409357814SOleg Nesterov mm->locked_vm += grow; 201584638335SKonstantin Khlebnikov vm_stat_account(mm, vma->vm_flags, grow); 2016bf181b9fSMichel Lespinasse anon_vma_interval_tree_pre_update_vma(vma); 20171da177e4SLinus Torvalds vma->vm_start = address; 20181da177e4SLinus Torvalds vma->vm_pgoff -= grow; 2019d4af56c5SLiam R. Howlett /* Overwrite old entry in mtree. */ 2020fbcc3104SLiam R. Howlett mas_set_range(&mas, address, vma->vm_end - 1); 2021fbcc3104SLiam R. Howlett mas_store_prealloc(&mas, vma); 2022bf181b9fSMichel Lespinasse anon_vma_interval_tree_post_update_vma(vma); 202309357814SOleg Nesterov spin_unlock(&mm->page_table_lock); 20244128997bSMichel Lespinasse 20253af9e859SEric B Munson perf_event_mmap(vma); 20261da177e4SLinus Torvalds } 20271da177e4SLinus Torvalds } 2028a626ca6aSLinus Torvalds } 202912352d3cSKonstantin Khlebnikov anon_vma_unlock_write(vma->anon_vma); 2030c791576cSYang Shi khugepaged_enter_vma(vma, vma->vm_flags); 2031d4af56c5SLiam R. Howlett mas_destroy(&mas); 20321da177e4SLinus Torvalds return error; 20331da177e4SLinus Torvalds } 20341da177e4SLinus Torvalds 20351be7107fSHugh Dickins /* enforced gap between the expanding stack and other mappings. */ 20361be7107fSHugh Dickins unsigned long stack_guard_gap = 256UL<<PAGE_SHIFT; 20371be7107fSHugh Dickins 20381be7107fSHugh Dickins static int __init cmdline_parse_stack_guard_gap(char *p) 20391be7107fSHugh Dickins { 20401be7107fSHugh Dickins unsigned long val; 20411be7107fSHugh Dickins char *endptr; 20421be7107fSHugh Dickins 20431be7107fSHugh Dickins val = simple_strtoul(p, &endptr, 10); 20441be7107fSHugh Dickins if (!*endptr) 20451be7107fSHugh Dickins stack_guard_gap = val << PAGE_SHIFT; 20461be7107fSHugh Dickins 2047e6d09493SRandy Dunlap return 1; 20481be7107fSHugh Dickins } 20491be7107fSHugh Dickins __setup("stack_guard_gap=", cmdline_parse_stack_guard_gap); 20501be7107fSHugh Dickins 2051b6a2fea3SOllie Wild #ifdef CONFIG_STACK_GROWSUP 2052b6a2fea3SOllie Wild int expand_stack(struct vm_area_struct *vma, unsigned long address) 2053b6a2fea3SOllie Wild { 2054b6a2fea3SOllie Wild return expand_upwards(vma, address); 2055b6a2fea3SOllie Wild } 2056b6a2fea3SOllie Wild 2057b6a2fea3SOllie Wild struct vm_area_struct * 2058b6a2fea3SOllie Wild find_extend_vma(struct mm_struct *mm, unsigned long addr) 2059b6a2fea3SOllie Wild { 2060b6a2fea3SOllie Wild struct vm_area_struct *vma, *prev; 2061b6a2fea3SOllie Wild 2062b6a2fea3SOllie Wild addr &= PAGE_MASK; 2063b6a2fea3SOllie Wild vma = find_vma_prev(mm, addr, &prev); 2064b6a2fea3SOllie Wild if (vma && (vma->vm_start <= addr)) 2065b6a2fea3SOllie Wild return vma; 20664d45e75aSJann Horn if (!prev || expand_stack(prev, addr)) 2067b6a2fea3SOllie Wild return NULL; 2068cea10a19SMichel Lespinasse if (prev->vm_flags & VM_LOCKED) 2069fc05f566SKirill A. Shutemov populate_vma_page_range(prev, addr, prev->vm_end, NULL); 2070b6a2fea3SOllie Wild return prev; 2071b6a2fea3SOllie Wild } 2072b6a2fea3SOllie Wild #else 2073b6a2fea3SOllie Wild int expand_stack(struct vm_area_struct *vma, unsigned long address) 2074b6a2fea3SOllie Wild { 2075b6a2fea3SOllie Wild return expand_downwards(vma, address); 2076b6a2fea3SOllie Wild } 2077b6a2fea3SOllie Wild 20781da177e4SLinus Torvalds struct vm_area_struct * 20791da177e4SLinus Torvalds find_extend_vma(struct mm_struct *mm, unsigned long addr) 20801da177e4SLinus Torvalds { 20811da177e4SLinus Torvalds struct vm_area_struct *vma; 20821da177e4SLinus Torvalds unsigned long start; 20831da177e4SLinus Torvalds 20841da177e4SLinus Torvalds addr &= PAGE_MASK; 20851da177e4SLinus Torvalds vma = find_vma(mm, addr); 20861da177e4SLinus Torvalds if (!vma) 20871da177e4SLinus Torvalds return NULL; 20881da177e4SLinus Torvalds if (vma->vm_start <= addr) 20891da177e4SLinus Torvalds return vma; 20901da177e4SLinus Torvalds if (!(vma->vm_flags & VM_GROWSDOWN)) 20911da177e4SLinus Torvalds return NULL; 20921da177e4SLinus Torvalds start = vma->vm_start; 20931da177e4SLinus Torvalds if (expand_stack(vma, addr)) 20941da177e4SLinus Torvalds return NULL; 2095cea10a19SMichel Lespinasse if (vma->vm_flags & VM_LOCKED) 2096fc05f566SKirill A. Shutemov populate_vma_page_range(vma, addr, start, NULL); 20971da177e4SLinus Torvalds return vma; 20981da177e4SLinus Torvalds } 20991da177e4SLinus Torvalds #endif 21001da177e4SLinus Torvalds 2101e1d6d01aSJesse Barnes EXPORT_SYMBOL_GPL(find_extend_vma); 2102e1d6d01aSJesse Barnes 21032c0b3814SHugh Dickins /* 2104763ecb03SLiam R. Howlett * Ok - we have the memory areas we should free on a maple tree so release them, 2105763ecb03SLiam R. Howlett * and do the vma updates. 21061da177e4SLinus Torvalds * 21072c0b3814SHugh Dickins * Called with the mm semaphore held. 21081da177e4SLinus Torvalds */ 2109763ecb03SLiam R. Howlett static inline void remove_mt(struct mm_struct *mm, struct ma_state *mas) 21101da177e4SLinus Torvalds { 21114f74d2c8SLinus Torvalds unsigned long nr_accounted = 0; 2112763ecb03SLiam R. Howlett struct vm_area_struct *vma; 21134f74d2c8SLinus Torvalds 2114365e9c87SHugh Dickins /* Update high watermark before we lower total_vm */ 2115365e9c87SHugh Dickins update_hiwater_vm(mm); 2116763ecb03SLiam R. Howlett mas_for_each(mas, vma, ULONG_MAX) { 2117ab50b8edSHugh Dickins long nrpages = vma_pages(vma); 21181da177e4SLinus Torvalds 21194f74d2c8SLinus Torvalds if (vma->vm_flags & VM_ACCOUNT) 21204f74d2c8SLinus Torvalds nr_accounted += nrpages; 212184638335SKonstantin Khlebnikov vm_stat_account(mm, vma->vm_flags, -nrpages); 2122763ecb03SLiam R. Howlett remove_vma(vma); 2123763ecb03SLiam R. Howlett } 21244f74d2c8SLinus Torvalds vm_unacct_memory(nr_accounted); 21251da177e4SLinus Torvalds validate_mm(mm); 21261da177e4SLinus Torvalds } 21271da177e4SLinus Torvalds 21281da177e4SLinus Torvalds /* 21291da177e4SLinus Torvalds * Get rid of page table information in the indicated region. 21301da177e4SLinus Torvalds * 2131f10df686SPaolo 'Blaisorblade' Giarrusso * Called with the mm semaphore held. 21321da177e4SLinus Torvalds */ 2133763ecb03SLiam R. Howlett static void unmap_region(struct mm_struct *mm, struct maple_tree *mt, 2134e0da382cSHugh Dickins struct vm_area_struct *vma, struct vm_area_struct *prev, 2135763ecb03SLiam R. Howlett struct vm_area_struct *next, 2136e0da382cSHugh Dickins unsigned long start, unsigned long end) 21371da177e4SLinus Torvalds { 2138d16dfc55SPeter Zijlstra struct mmu_gather tlb; 21391da177e4SLinus Torvalds 21401da177e4SLinus Torvalds lru_add_drain(); 2141a72afd87SWill Deacon tlb_gather_mmu(&tlb, mm); 2142365e9c87SHugh Dickins update_hiwater_rss(mm); 2143763ecb03SLiam R. Howlett unmap_vmas(&tlb, mt, vma, start, end); 2144763ecb03SLiam R. Howlett free_pgtables(&tlb, mt, vma, prev ? prev->vm_end : FIRST_USER_ADDRESS, 21456ee8630eSHugh Dickins next ? next->vm_start : USER_PGTABLES_CEILING); 2146ae8eba8bSWill Deacon tlb_finish_mmu(&tlb); 21471da177e4SLinus Torvalds } 21481da177e4SLinus Torvalds 21491da177e4SLinus Torvalds /* 2150def5efe0SDavid Rientjes * __split_vma() bypasses sysctl_max_map_count checking. We use this where it 2151def5efe0SDavid Rientjes * has already been checked or doesn't make sense to fail. 21520fd5a9e2SLiam R. Howlett * VMA Iterator will point to the end VMA. 21531da177e4SLinus Torvalds */ 21549760ebffSLiam R. Howlett int __split_vma(struct vma_iterator *vmi, struct vm_area_struct *vma, 21551da177e4SLinus Torvalds unsigned long addr, int new_below) 21561da177e4SLinus Torvalds { 2157b2b3b886SLiam R. Howlett struct vma_prepare vp; 21581da177e4SLinus Torvalds struct vm_area_struct *new; 2159e3975891SChen Gang int err; 21609760ebffSLiam R. Howlett 21619760ebffSLiam R. Howlett validate_mm_mt(vma->vm_mm); 21621da177e4SLinus Torvalds 2163b2b3b886SLiam R. Howlett WARN_ON(vma->vm_start >= addr); 2164b2b3b886SLiam R. Howlett WARN_ON(vma->vm_end <= addr); 2165b2b3b886SLiam R. Howlett 2166dd3b614fSDmitry Safonov if (vma->vm_ops && vma->vm_ops->may_split) { 2167dd3b614fSDmitry Safonov err = vma->vm_ops->may_split(vma, addr); 216831383c68SDan Williams if (err) 216931383c68SDan Williams return err; 217031383c68SDan Williams } 21711da177e4SLinus Torvalds 21723928d4f5SLinus Torvalds new = vm_area_dup(vma); 21731da177e4SLinus Torvalds if (!new) 2174e3975891SChen Gang return -ENOMEM; 21751da177e4SLinus Torvalds 2176b2b3b886SLiam R. Howlett err = -ENOMEM; 2177b2b3b886SLiam R. Howlett if (vma_iter_prealloc(vmi)) 2178b2b3b886SLiam R. Howlett goto out_free_vma; 2179b2b3b886SLiam R. Howlett 2180b2b3b886SLiam R. Howlett if (new_below) { 21811da177e4SLinus Torvalds new->vm_end = addr; 2182b2b3b886SLiam R. Howlett } else { 21831da177e4SLinus Torvalds new->vm_start = addr; 21841da177e4SLinus Torvalds new->vm_pgoff += ((addr - vma->vm_start) >> PAGE_SHIFT); 21851da177e4SLinus Torvalds } 21861da177e4SLinus Torvalds 2187ef0855d3SOleg Nesterov err = vma_dup_policy(vma, new); 2188ef0855d3SOleg Nesterov if (err) 2189b2b3b886SLiam R. Howlett goto out_free_vmi; 21901da177e4SLinus Torvalds 2191c4ea95d7SDaniel Forrest err = anon_vma_clone(new, vma); 2192c4ea95d7SDaniel Forrest if (err) 21935beb4930SRik van Riel goto out_free_mpol; 21945beb4930SRik van Riel 2195e9714acfSKonstantin Khlebnikov if (new->vm_file) 21961da177e4SLinus Torvalds get_file(new->vm_file); 21971da177e4SLinus Torvalds 21981da177e4SLinus Torvalds if (new->vm_ops && new->vm_ops->open) 21991da177e4SLinus Torvalds new->vm_ops->open(new); 22001da177e4SLinus Torvalds 2201b2b3b886SLiam R. Howlett vma_adjust_trans_huge(vma, vma->vm_start, addr, 0); 2202b2b3b886SLiam R. Howlett init_vma_prep(&vp, vma); 2203b2b3b886SLiam R. Howlett vp.insert = new; 2204b2b3b886SLiam R. Howlett vma_prepare(&vp); 22051da177e4SLinus Torvalds 2206b2b3b886SLiam R. Howlett if (new_below) { 2207b2b3b886SLiam R. Howlett vma->vm_start = addr; 2208b2b3b886SLiam R. Howlett vma->vm_pgoff += (addr - new->vm_start) >> PAGE_SHIFT; 2209b2b3b886SLiam R. Howlett } else { 2210b2b3b886SLiam R. Howlett vma->vm_end = addr; 22119760ebffSLiam R. Howlett } 22125beb4930SRik van Riel 2213b2b3b886SLiam R. Howlett /* vma_complete stores the new vma */ 2214b2b3b886SLiam R. Howlett vma_complete(&vp, vmi, vma->vm_mm); 2215b2b3b886SLiam R. Howlett 2216b2b3b886SLiam R. Howlett /* Success. */ 2217b2b3b886SLiam R. Howlett if (new_below) 2218b2b3b886SLiam R. Howlett vma_next(vmi); 2219b2b3b886SLiam R. Howlett validate_mm_mt(vma->vm_mm); 2220b2b3b886SLiam R. Howlett return 0; 2221b2b3b886SLiam R. Howlett 22225beb4930SRik van Riel out_free_mpol: 2223ef0855d3SOleg Nesterov mpol_put(vma_policy(new)); 2224b2b3b886SLiam R. Howlett out_free_vmi: 2225b2b3b886SLiam R. Howlett vma_iter_free(vmi); 22265beb4930SRik van Riel out_free_vma: 22273928d4f5SLinus Torvalds vm_area_free(new); 22289760ebffSLiam R. Howlett validate_mm_mt(vma->vm_mm); 22295beb4930SRik van Riel return err; 22301da177e4SLinus Torvalds } 22311da177e4SLinus Torvalds 2232659ace58SKOSAKI Motohiro /* 2233659ace58SKOSAKI Motohiro * Split a vma into two pieces at address 'addr', a new vma is allocated 2234659ace58SKOSAKI Motohiro * either for the first part or the tail. 2235659ace58SKOSAKI Motohiro */ 22369760ebffSLiam R. Howlett int split_vma(struct vma_iterator *vmi, struct vm_area_struct *vma, 2237659ace58SKOSAKI Motohiro unsigned long addr, int new_below) 2238659ace58SKOSAKI Motohiro { 22399760ebffSLiam R. Howlett if (vma->vm_mm->map_count >= sysctl_max_map_count) 2240659ace58SKOSAKI Motohiro return -ENOMEM; 2241659ace58SKOSAKI Motohiro 22429760ebffSLiam R. Howlett return __split_vma(vmi, vma, addr, new_below); 2243f2ebfe43SLiam R. Howlett } 2244f2ebfe43SLiam R. Howlett 2245763ecb03SLiam R. Howlett static inline int munmap_sidetree(struct vm_area_struct *vma, 2246763ecb03SLiam R. Howlett struct ma_state *mas_detach) 22471da177e4SLinus Torvalds { 2248763ecb03SLiam R. Howlett mas_set_range(mas_detach, vma->vm_start, vma->vm_end - 1); 2249763ecb03SLiam R. Howlett if (mas_store_gfp(mas_detach, vma, GFP_KERNEL)) 2250763ecb03SLiam R. Howlett return -ENOMEM; 225111f9a21aSLiam R. Howlett 2252763ecb03SLiam R. Howlett if (vma->vm_flags & VM_LOCKED) 2253763ecb03SLiam R. Howlett vma->vm_mm->locked_vm -= vma_pages(vma); 225411f9a21aSLiam R. Howlett 2255763ecb03SLiam R. Howlett return 0; 225611f9a21aSLiam R. Howlett } 225711f9a21aSLiam R. Howlett 225811f9a21aSLiam R. Howlett /* 2259183654ceSLiam R. Howlett * do_vmi_align_munmap() - munmap the aligned region from @start to @end. 2260183654ceSLiam R. Howlett * @vmi: The vma iterator 226111f9a21aSLiam R. Howlett * @vma: The starting vm_area_struct 226211f9a21aSLiam R. Howlett * @mm: The mm_struct 226311f9a21aSLiam R. Howlett * @start: The aligned start address to munmap. 226411f9a21aSLiam R. Howlett * @end: The aligned end address to munmap. 226511f9a21aSLiam R. Howlett * @uf: The userfaultfd list_head 22668651a137SLorenzo Stoakes * @downgrade: Set to true to attempt a write downgrade of the mmap_lock 226711f9a21aSLiam R. Howlett * 226811f9a21aSLiam R. Howlett * If @downgrade is true, check return code for potential release of the lock. 226911f9a21aSLiam R. Howlett */ 227011f9a21aSLiam R. Howlett static int 2271183654ceSLiam R. Howlett do_vmi_align_munmap(struct vma_iterator *vmi, struct vm_area_struct *vma, 227211f9a21aSLiam R. Howlett struct mm_struct *mm, unsigned long start, 227311f9a21aSLiam R. Howlett unsigned long end, struct list_head *uf, bool downgrade) 227411f9a21aSLiam R. Howlett { 2275763ecb03SLiam R. Howlett struct vm_area_struct *prev, *next = NULL; 2276763ecb03SLiam R. Howlett struct maple_tree mt_detach; 2277763ecb03SLiam R. Howlett int count = 0; 2278d4af56c5SLiam R. Howlett int error = -ENOMEM; 2279763ecb03SLiam R. Howlett MA_STATE(mas_detach, &mt_detach, 0, 0); 2280763ecb03SLiam R. Howlett mt_init_flags(&mt_detach, MT_FLAGS_LOCK_EXTERN); 2281763ecb03SLiam R. Howlett mt_set_external_lock(&mt_detach, &mm->mmap_lock); 2282524e00b3SLiam R. Howlett 22831da177e4SLinus Torvalds /* 22841da177e4SLinus Torvalds * If we need to split any vma, do it now to save pain later. 22851da177e4SLinus Torvalds * 22861da177e4SLinus Torvalds * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially 22871da177e4SLinus Torvalds * unmapped vm_area_struct will remain in use: so lower split_vma 22881da177e4SLinus Torvalds * places tmp vma above, and higher split_vma places tmp vma below. 22891da177e4SLinus Torvalds */ 2290763ecb03SLiam R. Howlett 2291763ecb03SLiam R. Howlett /* Does it split the first one? */ 2292146425a3SHugh Dickins if (start > vma->vm_start) { 2293659ace58SKOSAKI Motohiro 2294659ace58SKOSAKI Motohiro /* 2295659ace58SKOSAKI Motohiro * Make sure that map_count on return from munmap() will 2296659ace58SKOSAKI Motohiro * not exceed its limit; but let map_count go just above 2297659ace58SKOSAKI Motohiro * its limit temporarily, to help free resources as expected. 2298659ace58SKOSAKI Motohiro */ 2299659ace58SKOSAKI Motohiro if (end < vma->vm_end && mm->map_count >= sysctl_max_map_count) 2300d4af56c5SLiam R. Howlett goto map_count_exceeded; 2301659ace58SKOSAKI Motohiro 23029760ebffSLiam R. Howlett error = __split_vma(vmi, vma, start, 0); 23031da177e4SLinus Torvalds if (error) 2304763ecb03SLiam R. Howlett goto start_split_failed; 230511f9a21aSLiam R. Howlett 23060fd5a9e2SLiam R. Howlett vma = vma_iter_load(vmi); 23071da177e4SLinus Torvalds } 23081da177e4SLinus Torvalds 2309183654ceSLiam R. Howlett prev = vma_prev(vmi); 2310763ecb03SLiam R. Howlett if (unlikely((!prev))) 2311183654ceSLiam R. Howlett vma_iter_set(vmi, start); 231211f9a21aSLiam R. Howlett 2313763ecb03SLiam R. Howlett /* 2314763ecb03SLiam R. Howlett * Detach a range of VMAs from the mm. Using next as a temp variable as 2315763ecb03SLiam R. Howlett * it is always overwritten. 2316763ecb03SLiam R. Howlett */ 2317183654ceSLiam R. Howlett for_each_vma_range(*vmi, next, end) { 2318763ecb03SLiam R. Howlett /* Does it split the end? */ 2319763ecb03SLiam R. Howlett if (next->vm_end > end) { 23206b73cff2SLiam R. Howlett error = __split_vma(vmi, next, end, 0); 23211da177e4SLinus Torvalds if (error) 2322763ecb03SLiam R. Howlett goto end_split_failed; 232311f9a21aSLiam R. Howlett } 2324763ecb03SLiam R. Howlett error = munmap_sidetree(next, &mas_detach); 2325763ecb03SLiam R. Howlett if (error) 2326763ecb03SLiam R. Howlett goto munmap_sidetree_failed; 2327763ecb03SLiam R. Howlett 2328763ecb03SLiam R. Howlett count++; 2329763ecb03SLiam R. Howlett #ifdef CONFIG_DEBUG_VM_MAPLE_TREE 2330763ecb03SLiam R. Howlett BUG_ON(next->vm_start < start); 2331763ecb03SLiam R. Howlett BUG_ON(next->vm_start > end); 2332763ecb03SLiam R. Howlett #endif 2333763ecb03SLiam R. Howlett } 2334763ecb03SLiam R. Howlett 2335183654ceSLiam R. Howlett next = vma_next(vmi); 23362376dd7cSAndrea Arcangeli if (unlikely(uf)) { 23372376dd7cSAndrea Arcangeli /* 23382376dd7cSAndrea Arcangeli * If userfaultfd_unmap_prep returns an error the vmas 2339f0953a1bSIngo Molnar * will remain split, but userland will get a 23402376dd7cSAndrea Arcangeli * highly unexpected error anyway. This is no 23412376dd7cSAndrea Arcangeli * different than the case where the first of the two 23422376dd7cSAndrea Arcangeli * __split_vma fails, but we don't undo the first 23432376dd7cSAndrea Arcangeli * split, despite we could. This is unlikely enough 23442376dd7cSAndrea Arcangeli * failure that it's not worth optimizing it for. 23452376dd7cSAndrea Arcangeli */ 234669dbe6daSLiam R. Howlett error = userfaultfd_unmap_prep(mm, start, end, uf); 234711f9a21aSLiam R. Howlett 23482376dd7cSAndrea Arcangeli if (error) 2349d4af56c5SLiam R. Howlett goto userfaultfd_error; 23502376dd7cSAndrea Arcangeli } 23512376dd7cSAndrea Arcangeli 2352763ecb03SLiam R. Howlett #if defined(CONFIG_DEBUG_VM_MAPLE_TREE) 2353763ecb03SLiam R. Howlett /* Make sure no VMAs are about to be lost. */ 2354763ecb03SLiam R. Howlett { 2355763ecb03SLiam R. Howlett MA_STATE(test, &mt_detach, start, end - 1); 2356763ecb03SLiam R. Howlett struct vm_area_struct *vma_mas, *vma_test; 2357763ecb03SLiam R. Howlett int test_count = 0; 2358763ecb03SLiam R. Howlett 2359183654ceSLiam R. Howlett vma_iter_set(vmi, start); 2360763ecb03SLiam R. Howlett rcu_read_lock(); 2361763ecb03SLiam R. Howlett vma_test = mas_find(&test, end - 1); 2362183654ceSLiam R. Howlett for_each_vma_range(*vmi, vma_mas, end) { 2363763ecb03SLiam R. Howlett BUG_ON(vma_mas != vma_test); 2364763ecb03SLiam R. Howlett test_count++; 2365763ecb03SLiam R. Howlett vma_test = mas_next(&test, end - 1); 2366763ecb03SLiam R. Howlett } 2367763ecb03SLiam R. Howlett rcu_read_unlock(); 2368763ecb03SLiam R. Howlett BUG_ON(count != test_count); 2369763ecb03SLiam R. Howlett } 2370763ecb03SLiam R. Howlett #endif 23710378c0a0SLiam R. Howlett /* Point of no return */ 2372183654ceSLiam R. Howlett vma_iter_set(vmi, start); 2373183654ceSLiam R. Howlett if (vma_iter_clear_gfp(vmi, start, end, GFP_KERNEL)) 23740378c0a0SLiam R. Howlett return -ENOMEM; 23750378c0a0SLiam R. Howlett 2376763ecb03SLiam R. Howlett mm->map_count -= count; 237711f9a21aSLiam R. Howlett /* 237811f9a21aSLiam R. Howlett * Do not downgrade mmap_lock if we are next to VM_GROWSDOWN or 237911f9a21aSLiam R. Howlett * VM_GROWSUP VMA. Such VMAs can change their size under 238011f9a21aSLiam R. Howlett * down_read(mmap_lock) and collide with the VMA we are about to unmap. 238111f9a21aSLiam R. Howlett */ 238211f9a21aSLiam R. Howlett if (downgrade) { 2383763ecb03SLiam R. Howlett if (next && (next->vm_flags & VM_GROWSDOWN)) 238411f9a21aSLiam R. Howlett downgrade = false; 238511f9a21aSLiam R. Howlett else if (prev && (prev->vm_flags & VM_GROWSUP)) 238611f9a21aSLiam R. Howlett downgrade = false; 238711f9a21aSLiam R. Howlett else 2388d8ed45c5SMichel Lespinasse mmap_write_downgrade(mm); 238911f9a21aSLiam R. Howlett } 2390dd2283f2SYang Shi 2391763ecb03SLiam R. Howlett unmap_region(mm, &mt_detach, vma, prev, next, start, end); 2392763ecb03SLiam R. Howlett /* Statistics and freeing VMAs */ 2393763ecb03SLiam R. Howlett mas_set(&mas_detach, start); 2394763ecb03SLiam R. Howlett remove_mt(mm, &mas_detach); 2395763ecb03SLiam R. Howlett __mt_destroy(&mt_detach); 23961da177e4SLinus Torvalds 2397524e00b3SLiam R. Howlett 2398524e00b3SLiam R. Howlett validate_mm(mm); 2399dd2283f2SYang Shi return downgrade ? 1 : 0; 2400d4af56c5SLiam R. Howlett 2401d4af56c5SLiam R. Howlett userfaultfd_error: 2402763ecb03SLiam R. Howlett munmap_sidetree_failed: 2403763ecb03SLiam R. Howlett end_split_failed: 2404763ecb03SLiam R. Howlett __mt_destroy(&mt_detach); 2405763ecb03SLiam R. Howlett start_split_failed: 2406763ecb03SLiam R. Howlett map_count_exceeded: 2407d4af56c5SLiam R. Howlett return error; 24081da177e4SLinus Torvalds } 24091da177e4SLinus Torvalds 241011f9a21aSLiam R. Howlett /* 2411183654ceSLiam R. Howlett * do_vmi_munmap() - munmap a given range. 2412183654ceSLiam R. Howlett * @vmi: The vma iterator 241311f9a21aSLiam R. Howlett * @mm: The mm_struct 241411f9a21aSLiam R. Howlett * @start: The start address to munmap 241511f9a21aSLiam R. Howlett * @len: The length of the range to munmap 241611f9a21aSLiam R. Howlett * @uf: The userfaultfd list_head 241711f9a21aSLiam R. Howlett * @downgrade: set to true if the user wants to attempt to write_downgrade the 24188651a137SLorenzo Stoakes * mmap_lock 241911f9a21aSLiam R. Howlett * 242011f9a21aSLiam R. Howlett * This function takes a @mas that is either pointing to the previous VMA or set 242111f9a21aSLiam R. Howlett * to MA_START and sets it up to remove the mapping(s). The @len will be 242211f9a21aSLiam R. Howlett * aligned and any arch_unmap work will be preformed. 242311f9a21aSLiam R. Howlett * 242411f9a21aSLiam R. Howlett * Returns: -EINVAL on failure, 1 on success and unlock, 0 otherwise. 242511f9a21aSLiam R. Howlett */ 2426183654ceSLiam R. Howlett int do_vmi_munmap(struct vma_iterator *vmi, struct mm_struct *mm, 242711f9a21aSLiam R. Howlett unsigned long start, size_t len, struct list_head *uf, 242811f9a21aSLiam R. Howlett bool downgrade) 242911f9a21aSLiam R. Howlett { 243011f9a21aSLiam R. Howlett unsigned long end; 243111f9a21aSLiam R. Howlett struct vm_area_struct *vma; 243211f9a21aSLiam R. Howlett 243311f9a21aSLiam R. Howlett if ((offset_in_page(start)) || start > TASK_SIZE || len > TASK_SIZE-start) 243411f9a21aSLiam R. Howlett return -EINVAL; 243511f9a21aSLiam R. Howlett 243611f9a21aSLiam R. Howlett end = start + PAGE_ALIGN(len); 243711f9a21aSLiam R. Howlett if (end == start) 243811f9a21aSLiam R. Howlett return -EINVAL; 243911f9a21aSLiam R. Howlett 244011f9a21aSLiam R. Howlett /* arch_unmap() might do unmaps itself. */ 244111f9a21aSLiam R. Howlett arch_unmap(mm, start, end); 244211f9a21aSLiam R. Howlett 244311f9a21aSLiam R. Howlett /* Find the first overlapping VMA */ 2444183654ceSLiam R. Howlett vma = vma_find(vmi, end); 244511f9a21aSLiam R. Howlett if (!vma) 244611f9a21aSLiam R. Howlett return 0; 244711f9a21aSLiam R. Howlett 2448183654ceSLiam R. Howlett return do_vmi_align_munmap(vmi, vma, mm, start, end, uf, downgrade); 244911f9a21aSLiam R. Howlett } 245011f9a21aSLiam R. Howlett 245111f9a21aSLiam R. Howlett /* do_munmap() - Wrapper function for non-maple tree aware do_munmap() calls. 245211f9a21aSLiam R. Howlett * @mm: The mm_struct 245311f9a21aSLiam R. Howlett * @start: The start address to munmap 245411f9a21aSLiam R. Howlett * @len: The length to be munmapped. 245511f9a21aSLiam R. Howlett * @uf: The userfaultfd list_head 245611f9a21aSLiam R. Howlett */ 2457dd2283f2SYang Shi int do_munmap(struct mm_struct *mm, unsigned long start, size_t len, 2458dd2283f2SYang Shi struct list_head *uf) 2459dd2283f2SYang Shi { 2460183654ceSLiam R. Howlett VMA_ITERATOR(vmi, mm, start); 246111f9a21aSLiam R. Howlett 2462183654ceSLiam R. Howlett return do_vmi_munmap(&vmi, mm, start, len, uf, false); 2463dd2283f2SYang Shi } 2464dd2283f2SYang Shi 2465e99668a5SLiam R. Howlett unsigned long mmap_region(struct file *file, unsigned long addr, 2466e99668a5SLiam R. Howlett unsigned long len, vm_flags_t vm_flags, unsigned long pgoff, 2467e99668a5SLiam R. Howlett struct list_head *uf) 2468e99668a5SLiam R. Howlett { 2469e99668a5SLiam R. Howlett struct mm_struct *mm = current->mm; 2470e99668a5SLiam R. Howlett struct vm_area_struct *vma = NULL; 2471e99668a5SLiam R. Howlett struct vm_area_struct *next, *prev, *merge; 2472e99668a5SLiam R. Howlett pgoff_t pglen = len >> PAGE_SHIFT; 2473e99668a5SLiam R. Howlett unsigned long charged = 0; 2474e99668a5SLiam R. Howlett unsigned long end = addr + len; 2475e99668a5SLiam R. Howlett unsigned long merge_start = addr, merge_end = end; 2476e99668a5SLiam R. Howlett pgoff_t vm_pgoff; 2477e99668a5SLiam R. Howlett int error; 2478183654ceSLiam R. Howlett VMA_ITERATOR(vmi, mm, addr); 2479e99668a5SLiam R. Howlett 2480e99668a5SLiam R. Howlett /* Check against address space limit. */ 2481e99668a5SLiam R. Howlett if (!may_expand_vm(mm, vm_flags, len >> PAGE_SHIFT)) { 2482e99668a5SLiam R. Howlett unsigned long nr_pages; 2483e99668a5SLiam R. Howlett 2484e99668a5SLiam R. Howlett /* 2485e99668a5SLiam R. Howlett * MAP_FIXED may remove pages of mappings that intersects with 2486e99668a5SLiam R. Howlett * requested mapping. Account for the pages it would unmap. 2487e99668a5SLiam R. Howlett */ 2488e99668a5SLiam R. Howlett nr_pages = count_vma_pages_range(mm, addr, end); 2489e99668a5SLiam R. Howlett 2490e99668a5SLiam R. Howlett if (!may_expand_vm(mm, vm_flags, 2491e99668a5SLiam R. Howlett (len >> PAGE_SHIFT) - nr_pages)) 2492e99668a5SLiam R. Howlett return -ENOMEM; 2493e99668a5SLiam R. Howlett } 2494e99668a5SLiam R. Howlett 2495e99668a5SLiam R. Howlett /* Unmap any existing mapping in the area */ 2496183654ceSLiam R. Howlett if (do_vmi_munmap(&vmi, mm, addr, len, uf, false)) 2497e99668a5SLiam R. Howlett return -ENOMEM; 2498e99668a5SLiam R. Howlett 2499e99668a5SLiam R. Howlett /* 2500e99668a5SLiam R. Howlett * Private writable mapping: check memory availability 2501e99668a5SLiam R. Howlett */ 2502e99668a5SLiam R. Howlett if (accountable_mapping(file, vm_flags)) { 2503e99668a5SLiam R. Howlett charged = len >> PAGE_SHIFT; 2504e99668a5SLiam R. Howlett if (security_vm_enough_memory_mm(mm, charged)) 2505e99668a5SLiam R. Howlett return -ENOMEM; 2506e99668a5SLiam R. Howlett vm_flags |= VM_ACCOUNT; 2507e99668a5SLiam R. Howlett } 2508e99668a5SLiam R. Howlett 2509183654ceSLiam R. Howlett next = vma_next(&vmi); 2510183654ceSLiam R. Howlett prev = vma_prev(&vmi); 2511e99668a5SLiam R. Howlett if (vm_flags & VM_SPECIAL) 2512e99668a5SLiam R. Howlett goto cannot_expand; 2513e99668a5SLiam R. Howlett 2514e99668a5SLiam R. Howlett /* Attempt to expand an old mapping */ 2515e99668a5SLiam R. Howlett /* Check next */ 2516e99668a5SLiam R. Howlett if (next && next->vm_start == end && !vma_policy(next) && 2517e99668a5SLiam R. Howlett can_vma_merge_before(next, vm_flags, NULL, file, pgoff+pglen, 2518e99668a5SLiam R. Howlett NULL_VM_UFFD_CTX, NULL)) { 2519e99668a5SLiam R. Howlett merge_end = next->vm_end; 2520e99668a5SLiam R. Howlett vma = next; 2521e99668a5SLiam R. Howlett vm_pgoff = next->vm_pgoff - pglen; 2522e99668a5SLiam R. Howlett } 2523e99668a5SLiam R. Howlett 2524e99668a5SLiam R. Howlett /* Check prev */ 2525e99668a5SLiam R. Howlett if (prev && prev->vm_end == addr && !vma_policy(prev) && 2526e99668a5SLiam R. Howlett (vma ? can_vma_merge_after(prev, vm_flags, vma->anon_vma, file, 2527e99668a5SLiam R. Howlett pgoff, vma->vm_userfaultfd_ctx, NULL) : 2528e99668a5SLiam R. Howlett can_vma_merge_after(prev, vm_flags, NULL, file, pgoff, 2529e99668a5SLiam R. Howlett NULL_VM_UFFD_CTX, NULL))) { 2530e99668a5SLiam R. Howlett merge_start = prev->vm_start; 2531e99668a5SLiam R. Howlett vma = prev; 2532e99668a5SLiam R. Howlett vm_pgoff = prev->vm_pgoff; 2533e99668a5SLiam R. Howlett } 2534e99668a5SLiam R. Howlett 2535e99668a5SLiam R. Howlett 2536e99668a5SLiam R. Howlett /* Actually expand, if possible */ 2537e99668a5SLiam R. Howlett if (vma && 25383c441ab7SLiam R. Howlett !vma_expand(&vmi, vma, merge_start, merge_end, vm_pgoff, next)) { 2539e99668a5SLiam R. Howlett khugepaged_enter_vma(vma, vm_flags); 2540e99668a5SLiam R. Howlett goto expanded; 2541e99668a5SLiam R. Howlett } 2542e99668a5SLiam R. Howlett 2543e99668a5SLiam R. Howlett cannot_expand: 2544e99668a5SLiam R. Howlett /* 2545e99668a5SLiam R. Howlett * Determine the object being mapped and call the appropriate 2546e99668a5SLiam R. Howlett * specific mapper. the address has already been validated, but 2547e99668a5SLiam R. Howlett * not unmapped, but the maps are removed from the list. 2548e99668a5SLiam R. Howlett */ 2549e99668a5SLiam R. Howlett vma = vm_area_alloc(mm); 2550e99668a5SLiam R. Howlett if (!vma) { 2551e99668a5SLiam R. Howlett error = -ENOMEM; 2552e99668a5SLiam R. Howlett goto unacct_error; 2553e99668a5SLiam R. Howlett } 2554e99668a5SLiam R. Howlett 25550fd5a9e2SLiam R. Howlett vma_iter_set(&vmi, addr); 2556e99668a5SLiam R. Howlett vma->vm_start = addr; 2557e99668a5SLiam R. Howlett vma->vm_end = end; 2558*1c71222eSSuren Baghdasaryan vm_flags_init(vma, vm_flags); 2559e99668a5SLiam R. Howlett vma->vm_page_prot = vm_get_page_prot(vm_flags); 2560e99668a5SLiam R. Howlett vma->vm_pgoff = pgoff; 2561e99668a5SLiam R. Howlett 2562e99668a5SLiam R. Howlett if (file) { 2563e99668a5SLiam R. Howlett if (vm_flags & VM_SHARED) { 2564e99668a5SLiam R. Howlett error = mapping_map_writable(file->f_mapping); 2565e99668a5SLiam R. Howlett if (error) 2566e99668a5SLiam R. Howlett goto free_vma; 2567e99668a5SLiam R. Howlett } 2568e99668a5SLiam R. Howlett 2569e99668a5SLiam R. Howlett vma->vm_file = get_file(file); 2570e99668a5SLiam R. Howlett error = call_mmap(file, vma); 2571e99668a5SLiam R. Howlett if (error) 2572e99668a5SLiam R. Howlett goto unmap_and_free_vma; 2573e99668a5SLiam R. Howlett 2574a57b7051SLiam Howlett /* 2575a57b7051SLiam Howlett * Expansion is handled above, merging is handled below. 2576a57b7051SLiam Howlett * Drivers should not alter the address of the VMA. 2577e99668a5SLiam R. Howlett */ 2578a57b7051SLiam Howlett error = -EINVAL; 2579cc8d1b09SLiam R. Howlett if (WARN_ON((addr != vma->vm_start))) 2580a57b7051SLiam Howlett goto close_and_free_vma; 2581e99668a5SLiam R. Howlett 2582cc8d1b09SLiam R. Howlett vma_iter_set(&vmi, addr); 2583e99668a5SLiam R. Howlett /* 2584e99668a5SLiam R. Howlett * If vm_flags changed after call_mmap(), we should try merge 2585e99668a5SLiam R. Howlett * vma again as we may succeed this time. 2586e99668a5SLiam R. Howlett */ 2587e99668a5SLiam R. Howlett if (unlikely(vm_flags != vma->vm_flags && prev)) { 25889760ebffSLiam R. Howlett merge = vma_merge(&vmi, mm, prev, vma->vm_start, 25899760ebffSLiam R. Howlett vma->vm_end, vma->vm_flags, NULL, 25909760ebffSLiam R. Howlett vma->vm_file, vma->vm_pgoff, NULL, 25919760ebffSLiam R. Howlett NULL_VM_UFFD_CTX, NULL); 2592e99668a5SLiam R. Howlett if (merge) { 2593e99668a5SLiam R. Howlett /* 2594e99668a5SLiam R. Howlett * ->mmap() can change vma->vm_file and fput 2595e99668a5SLiam R. Howlett * the original file. So fput the vma->vm_file 2596e99668a5SLiam R. Howlett * here or we would add an extra fput for file 2597e99668a5SLiam R. Howlett * and cause general protection fault 2598e99668a5SLiam R. Howlett * ultimately. 2599e99668a5SLiam R. Howlett */ 2600e99668a5SLiam R. Howlett fput(vma->vm_file); 2601e99668a5SLiam R. Howlett vm_area_free(vma); 2602e99668a5SLiam R. Howlett vma = merge; 2603e99668a5SLiam R. Howlett /* Update vm_flags to pick up the change. */ 2604e99668a5SLiam R. Howlett vm_flags = vma->vm_flags; 2605e99668a5SLiam R. Howlett goto unmap_writable; 2606e99668a5SLiam R. Howlett } 2607e99668a5SLiam R. Howlett } 2608e99668a5SLiam R. Howlett 2609e99668a5SLiam R. Howlett vm_flags = vma->vm_flags; 2610e99668a5SLiam R. Howlett } else if (vm_flags & VM_SHARED) { 2611e99668a5SLiam R. Howlett error = shmem_zero_setup(vma); 2612e99668a5SLiam R. Howlett if (error) 2613e99668a5SLiam R. Howlett goto free_vma; 2614e99668a5SLiam R. Howlett } else { 2615e99668a5SLiam R. Howlett vma_set_anonymous(vma); 2616e99668a5SLiam R. Howlett } 2617e99668a5SLiam R. Howlett 2618b507808eSJoey Gouly if (map_deny_write_exec(vma, vma->vm_flags)) { 2619b507808eSJoey Gouly error = -EACCES; 2620b507808eSJoey Gouly if (file) 2621b507808eSJoey Gouly goto close_and_free_vma; 2622b507808eSJoey Gouly else if (vma->vm_file) 2623b507808eSJoey Gouly goto unmap_and_free_vma; 2624b507808eSJoey Gouly else 2625b507808eSJoey Gouly goto free_vma; 2626b507808eSJoey Gouly } 2627b507808eSJoey Gouly 2628e99668a5SLiam R. Howlett /* Allow architectures to sanity-check the vm_flags */ 2629e99668a5SLiam R. Howlett error = -EINVAL; 2630cc8d1b09SLiam R. Howlett if (!arch_validate_flags(vma->vm_flags)) 2631deb0f656SCarlos Llamas goto close_and_free_vma; 2632e99668a5SLiam R. Howlett 2633e99668a5SLiam R. Howlett error = -ENOMEM; 2634cc8d1b09SLiam R. Howlett if (vma_iter_prealloc(&vmi)) 26355789151eSMike Kravetz goto close_and_free_vma; 2636e99668a5SLiam R. Howlett 2637e99668a5SLiam R. Howlett if (vma->vm_file) 2638e99668a5SLiam R. Howlett i_mmap_lock_write(vma->vm_file->f_mapping); 2639e99668a5SLiam R. Howlett 2640183654ceSLiam R. Howlett vma_iter_store(&vmi, vma); 2641e99668a5SLiam R. Howlett mm->map_count++; 2642e99668a5SLiam R. Howlett if (vma->vm_file) { 2643e99668a5SLiam R. Howlett if (vma->vm_flags & VM_SHARED) 2644e99668a5SLiam R. Howlett mapping_allow_writable(vma->vm_file->f_mapping); 2645e99668a5SLiam R. Howlett 2646e99668a5SLiam R. Howlett flush_dcache_mmap_lock(vma->vm_file->f_mapping); 2647e99668a5SLiam R. Howlett vma_interval_tree_insert(vma, &vma->vm_file->f_mapping->i_mmap); 2648e99668a5SLiam R. Howlett flush_dcache_mmap_unlock(vma->vm_file->f_mapping); 2649e99668a5SLiam R. Howlett i_mmap_unlock_write(vma->vm_file->f_mapping); 2650e99668a5SLiam R. Howlett } 2651e99668a5SLiam R. Howlett 2652e99668a5SLiam R. Howlett /* 2653e99668a5SLiam R. Howlett * vma_merge() calls khugepaged_enter_vma() either, the below 2654e99668a5SLiam R. Howlett * call covers the non-merge case. 2655e99668a5SLiam R. Howlett */ 2656e99668a5SLiam R. Howlett khugepaged_enter_vma(vma, vma->vm_flags); 2657e99668a5SLiam R. Howlett 2658e99668a5SLiam R. Howlett /* Once vma denies write, undo our temporary denial count */ 2659e99668a5SLiam R. Howlett unmap_writable: 2660e99668a5SLiam R. Howlett if (file && vm_flags & VM_SHARED) 2661e99668a5SLiam R. Howlett mapping_unmap_writable(file->f_mapping); 2662e99668a5SLiam R. Howlett file = vma->vm_file; 2663e99668a5SLiam R. Howlett expanded: 2664e99668a5SLiam R. Howlett perf_event_mmap(vma); 2665e99668a5SLiam R. Howlett 2666e99668a5SLiam R. Howlett vm_stat_account(mm, vm_flags, len >> PAGE_SHIFT); 2667e99668a5SLiam R. Howlett if (vm_flags & VM_LOCKED) { 2668e99668a5SLiam R. Howlett if ((vm_flags & VM_SPECIAL) || vma_is_dax(vma) || 2669e99668a5SLiam R. Howlett is_vm_hugetlb_page(vma) || 2670e99668a5SLiam R. Howlett vma == get_gate_vma(current->mm)) 2671e430a95aSSuren Baghdasaryan vm_flags_clear(vma, VM_LOCKED_MASK); 2672e99668a5SLiam R. Howlett else 2673e99668a5SLiam R. Howlett mm->locked_vm += (len >> PAGE_SHIFT); 2674e99668a5SLiam R. Howlett } 2675e99668a5SLiam R. Howlett 2676e99668a5SLiam R. Howlett if (file) 2677e99668a5SLiam R. Howlett uprobe_mmap(vma); 2678e99668a5SLiam R. Howlett 2679e99668a5SLiam R. Howlett /* 2680e99668a5SLiam R. Howlett * New (or expanded) vma always get soft dirty status. 2681e99668a5SLiam R. Howlett * Otherwise user-space soft-dirty page tracker won't 2682e99668a5SLiam R. Howlett * be able to distinguish situation when vma area unmapped, 2683e99668a5SLiam R. Howlett * then new mapped in-place (which must be aimed as 2684e99668a5SLiam R. Howlett * a completely new data area). 2685e99668a5SLiam R. Howlett */ 2686*1c71222eSSuren Baghdasaryan vm_flags_set(vma, VM_SOFTDIRTY); 2687e99668a5SLiam R. Howlett 2688e99668a5SLiam R. Howlett vma_set_page_prot(vma); 2689e99668a5SLiam R. Howlett 2690e99668a5SLiam R. Howlett validate_mm(mm); 2691e99668a5SLiam R. Howlett return addr; 2692e99668a5SLiam R. Howlett 2693deb0f656SCarlos Llamas close_and_free_vma: 2694cc8d1b09SLiam R. Howlett if (file && vma->vm_ops && vma->vm_ops->close) 2695deb0f656SCarlos Llamas vma->vm_ops->close(vma); 2696cc8d1b09SLiam R. Howlett 2697cc8d1b09SLiam R. Howlett if (file || vma->vm_file) { 2698e99668a5SLiam R. Howlett unmap_and_free_vma: 2699e99668a5SLiam R. Howlett fput(vma->vm_file); 2700e99668a5SLiam R. Howlett vma->vm_file = NULL; 2701e99668a5SLiam R. Howlett 2702e99668a5SLiam R. Howlett /* Undo any partial mapping done by a device driver. */ 2703cc8d1b09SLiam R. Howlett unmap_region(mm, &mm->mm_mt, vma, prev, next, vma->vm_start, 2704cc8d1b09SLiam R. Howlett vma->vm_end); 2705cc8d1b09SLiam R. Howlett } 2706cc674ab3SLi Zetao if (file && (vm_flags & VM_SHARED)) 2707e99668a5SLiam R. Howlett mapping_unmap_writable(file->f_mapping); 2708e99668a5SLiam R. Howlett free_vma: 2709e99668a5SLiam R. Howlett vm_area_free(vma); 2710e99668a5SLiam R. Howlett unacct_error: 2711e99668a5SLiam R. Howlett if (charged) 2712e99668a5SLiam R. Howlett vm_unacct_memory(charged); 2713e99668a5SLiam R. Howlett validate_mm(mm); 2714e99668a5SLiam R. Howlett return error; 2715e99668a5SLiam R. Howlett } 2716e99668a5SLiam R. Howlett 2717dd2283f2SYang Shi static int __vm_munmap(unsigned long start, size_t len, bool downgrade) 2718a46ef99dSLinus Torvalds { 2719a46ef99dSLinus Torvalds int ret; 2720bfce281cSAl Viro struct mm_struct *mm = current->mm; 2721897ab3e0SMike Rapoport LIST_HEAD(uf); 2722183654ceSLiam R. Howlett VMA_ITERATOR(vmi, mm, start); 2723a46ef99dSLinus Torvalds 2724d8ed45c5SMichel Lespinasse if (mmap_write_lock_killable(mm)) 2725ae798783SMichal Hocko return -EINTR; 2726ae798783SMichal Hocko 2727183654ceSLiam R. Howlett ret = do_vmi_munmap(&vmi, mm, start, len, &uf, downgrade); 2728dd2283f2SYang Shi /* 2729c1e8d7c6SMichel Lespinasse * Returning 1 indicates mmap_lock is downgraded. 2730dd2283f2SYang Shi * But 1 is not legal return value of vm_munmap() and munmap(), reset 2731dd2283f2SYang Shi * it to 0 before return. 2732dd2283f2SYang Shi */ 2733dd2283f2SYang Shi if (ret == 1) { 2734d8ed45c5SMichel Lespinasse mmap_read_unlock(mm); 2735dd2283f2SYang Shi ret = 0; 2736dd2283f2SYang Shi } else 2737d8ed45c5SMichel Lespinasse mmap_write_unlock(mm); 2738dd2283f2SYang Shi 2739897ab3e0SMike Rapoport userfaultfd_unmap_complete(mm, &uf); 2740a46ef99dSLinus Torvalds return ret; 2741a46ef99dSLinus Torvalds } 2742dd2283f2SYang Shi 2743dd2283f2SYang Shi int vm_munmap(unsigned long start, size_t len) 2744dd2283f2SYang Shi { 2745dd2283f2SYang Shi return __vm_munmap(start, len, false); 2746dd2283f2SYang Shi } 2747a46ef99dSLinus Torvalds EXPORT_SYMBOL(vm_munmap); 2748a46ef99dSLinus Torvalds 27496a6160a7SHeiko Carstens SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len) 27501da177e4SLinus Torvalds { 2751ce18d171SCatalin Marinas addr = untagged_addr(addr); 2752dd2283f2SYang Shi return __vm_munmap(addr, len, true); 27531da177e4SLinus Torvalds } 27541da177e4SLinus Torvalds 2755c8d78c18SKirill A. Shutemov 2756c8d78c18SKirill A. Shutemov /* 2757c8d78c18SKirill A. Shutemov * Emulation of deprecated remap_file_pages() syscall. 2758c8d78c18SKirill A. Shutemov */ 2759c8d78c18SKirill A. Shutemov SYSCALL_DEFINE5(remap_file_pages, unsigned long, start, unsigned long, size, 2760c8d78c18SKirill A. Shutemov unsigned long, prot, unsigned long, pgoff, unsigned long, flags) 2761c8d78c18SKirill A. Shutemov { 2762c8d78c18SKirill A. Shutemov 2763c8d78c18SKirill A. Shutemov struct mm_struct *mm = current->mm; 2764c8d78c18SKirill A. Shutemov struct vm_area_struct *vma; 2765c8d78c18SKirill A. Shutemov unsigned long populate = 0; 2766c8d78c18SKirill A. Shutemov unsigned long ret = -EINVAL; 2767c8d78c18SKirill A. Shutemov struct file *file; 2768c8d78c18SKirill A. Shutemov 2769ee65728eSMike Rapoport pr_warn_once("%s (%d) uses deprecated remap_file_pages() syscall. See Documentation/mm/remap_file_pages.rst.\n", 2770c8d78c18SKirill A. Shutemov current->comm, current->pid); 2771c8d78c18SKirill A. Shutemov 2772c8d78c18SKirill A. Shutemov if (prot) 2773c8d78c18SKirill A. Shutemov return ret; 2774c8d78c18SKirill A. Shutemov start = start & PAGE_MASK; 2775c8d78c18SKirill A. Shutemov size = size & PAGE_MASK; 2776c8d78c18SKirill A. Shutemov 2777c8d78c18SKirill A. Shutemov if (start + size <= start) 2778c8d78c18SKirill A. Shutemov return ret; 2779c8d78c18SKirill A. Shutemov 2780c8d78c18SKirill A. Shutemov /* Does pgoff wrap? */ 2781c8d78c18SKirill A. Shutemov if (pgoff + (size >> PAGE_SHIFT) < pgoff) 2782c8d78c18SKirill A. Shutemov return ret; 2783c8d78c18SKirill A. Shutemov 2784d8ed45c5SMichel Lespinasse if (mmap_write_lock_killable(mm)) 2785dc0ef0dfSMichal Hocko return -EINTR; 2786dc0ef0dfSMichal Hocko 27879b593cb2SLiam R. Howlett vma = vma_lookup(mm, start); 2788c8d78c18SKirill A. Shutemov 2789c8d78c18SKirill A. Shutemov if (!vma || !(vma->vm_flags & VM_SHARED)) 2790c8d78c18SKirill A. Shutemov goto out; 2791c8d78c18SKirill A. Shutemov 279248f7df32SKirill A. Shutemov if (start + size > vma->vm_end) { 2793763ecb03SLiam R. Howlett VMA_ITERATOR(vmi, mm, vma->vm_end); 2794763ecb03SLiam R. Howlett struct vm_area_struct *next, *prev = vma; 279548f7df32SKirill A. Shutemov 2796763ecb03SLiam R. Howlett for_each_vma_range(vmi, next, start + size) { 279748f7df32SKirill A. Shutemov /* hole between vmas ? */ 2798763ecb03SLiam R. Howlett if (next->vm_start != prev->vm_end) 279948f7df32SKirill A. Shutemov goto out; 280048f7df32SKirill A. Shutemov 280148f7df32SKirill A. Shutemov if (next->vm_file != vma->vm_file) 280248f7df32SKirill A. Shutemov goto out; 280348f7df32SKirill A. Shutemov 280448f7df32SKirill A. Shutemov if (next->vm_flags != vma->vm_flags) 280548f7df32SKirill A. Shutemov goto out; 280648f7df32SKirill A. Shutemov 28071db43d3fSLiam Howlett if (start + size <= next->vm_end) 28081db43d3fSLiam Howlett break; 28091db43d3fSLiam Howlett 2810763ecb03SLiam R. Howlett prev = next; 281148f7df32SKirill A. Shutemov } 281248f7df32SKirill A. Shutemov 281348f7df32SKirill A. Shutemov if (!next) 2814c8d78c18SKirill A. Shutemov goto out; 2815c8d78c18SKirill A. Shutemov } 2816c8d78c18SKirill A. Shutemov 2817c8d78c18SKirill A. Shutemov prot |= vma->vm_flags & VM_READ ? PROT_READ : 0; 2818c8d78c18SKirill A. Shutemov prot |= vma->vm_flags & VM_WRITE ? PROT_WRITE : 0; 2819c8d78c18SKirill A. Shutemov prot |= vma->vm_flags & VM_EXEC ? PROT_EXEC : 0; 2820c8d78c18SKirill A. Shutemov 2821c8d78c18SKirill A. Shutemov flags &= MAP_NONBLOCK; 2822c8d78c18SKirill A. Shutemov flags |= MAP_SHARED | MAP_FIXED | MAP_POPULATE; 2823fce000b1SLiam Howlett if (vma->vm_flags & VM_LOCKED) 2824c8d78c18SKirill A. Shutemov flags |= MAP_LOCKED; 282548f7df32SKirill A. Shutemov 2826c8d78c18SKirill A. Shutemov file = get_file(vma->vm_file); 282745e55300SPeter Collingbourne ret = do_mmap(vma->vm_file, start, size, 2828897ab3e0SMike Rapoport prot, flags, pgoff, &populate, NULL); 2829c8d78c18SKirill A. Shutemov fput(file); 2830c8d78c18SKirill A. Shutemov out: 2831d8ed45c5SMichel Lespinasse mmap_write_unlock(mm); 2832c8d78c18SKirill A. Shutemov if (populate) 2833c8d78c18SKirill A. Shutemov mm_populate(ret, populate); 2834c8d78c18SKirill A. Shutemov if (!IS_ERR_VALUE(ret)) 2835c8d78c18SKirill A. Shutemov ret = 0; 2836c8d78c18SKirill A. Shutemov return ret; 2837c8d78c18SKirill A. Shutemov } 2838c8d78c18SKirill A. Shutemov 28391da177e4SLinus Torvalds /* 284027b26701SLiam R. Howlett * do_vma_munmap() - Unmap a full or partial vma. 284127b26701SLiam R. Howlett * @vmi: The vma iterator pointing at the vma 284227b26701SLiam R. Howlett * @vma: The first vma to be munmapped 284327b26701SLiam R. Howlett * @start: the start of the address to unmap 284427b26701SLiam R. Howlett * @end: The end of the address to unmap 28452e7ce7d3SLiam R. Howlett * @uf: The userfaultfd list_head 284627b26701SLiam R. Howlett * @downgrade: Attempt to downgrade or not 28472e7ce7d3SLiam R. Howlett * 284827b26701SLiam R. Howlett * Returns: 0 on success and not downgraded, 1 on success and downgraded. 284927b26701SLiam R. Howlett * unmaps a VMA mapping when the vma iterator is already in position. 285027b26701SLiam R. Howlett * Does not handle alignment. 28511da177e4SLinus Torvalds */ 285227b26701SLiam R. Howlett int do_vma_munmap(struct vma_iterator *vmi, struct vm_area_struct *vma, 285327b26701SLiam R. Howlett unsigned long start, unsigned long end, 285427b26701SLiam R. Howlett struct list_head *uf, bool downgrade) 28552e7ce7d3SLiam R. Howlett { 28562e7ce7d3SLiam R. Howlett struct mm_struct *mm = vma->vm_mm; 28572e7ce7d3SLiam R. Howlett int ret; 28582e7ce7d3SLiam R. Howlett 285927b26701SLiam R. Howlett arch_unmap(mm, start, end); 286027b26701SLiam R. Howlett ret = do_vmi_align_munmap(vmi, vma, mm, start, end, uf, downgrade); 28612e7ce7d3SLiam R. Howlett validate_mm_mt(mm); 28622e7ce7d3SLiam R. Howlett return ret; 28632e7ce7d3SLiam R. Howlett } 28642e7ce7d3SLiam R. Howlett 28652e7ce7d3SLiam R. Howlett /* 28662e7ce7d3SLiam R. Howlett * do_brk_flags() - Increase the brk vma if the flags match. 286792fed820SLiam R. Howlett * @vmi: The vma iterator 28682e7ce7d3SLiam R. Howlett * @addr: The start address 28692e7ce7d3SLiam R. Howlett * @len: The length of the increase 28702e7ce7d3SLiam R. Howlett * @vma: The vma, 28712e7ce7d3SLiam R. Howlett * @flags: The VMA Flags 28722e7ce7d3SLiam R. Howlett * 28732e7ce7d3SLiam R. Howlett * Extend the brk VMA from addr to addr + len. If the VMA is NULL or the flags 28742e7ce7d3SLiam R. Howlett * do not match then create a new anonymous VMA. Eventually we may be able to 28752e7ce7d3SLiam R. Howlett * do some brk-specific accounting here. 28762e7ce7d3SLiam R. Howlett */ 287792fed820SLiam R. Howlett static int do_brk_flags(struct vma_iterator *vmi, struct vm_area_struct *vma, 2878763ecb03SLiam R. Howlett unsigned long addr, unsigned long len, unsigned long flags) 28791da177e4SLinus Torvalds { 28801da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 2881287051b1SLiam R. Howlett struct vma_prepare vp; 28822e7ce7d3SLiam R. Howlett 2883d4af56c5SLiam R. Howlett validate_mm_mt(mm); 28842e7ce7d3SLiam R. Howlett /* 28852e7ce7d3SLiam R. Howlett * Check against address space limits by the changed size 28862e7ce7d3SLiam R. Howlett * Note: This happens *after* clearing old mappings in some code paths. 28872e7ce7d3SLiam R. Howlett */ 288816e72e9bSDenys Vlasenko flags |= VM_DATA_DEFAULT_FLAGS | VM_ACCOUNT | mm->def_flags; 288984638335SKonstantin Khlebnikov if (!may_expand_vm(mm, flags, len >> PAGE_SHIFT)) 28901da177e4SLinus Torvalds return -ENOMEM; 28911da177e4SLinus Torvalds 28921da177e4SLinus Torvalds if (mm->map_count > sysctl_max_map_count) 28931da177e4SLinus Torvalds return -ENOMEM; 28941da177e4SLinus Torvalds 2895191c5424SAl Viro if (security_vm_enough_memory_mm(mm, len >> PAGE_SHIFT)) 28961da177e4SLinus Torvalds return -ENOMEM; 28971da177e4SLinus Torvalds 28981da177e4SLinus Torvalds /* 28992e7ce7d3SLiam R. Howlett * Expand the existing vma if possible; Note that singular lists do not 29002e7ce7d3SLiam R. Howlett * occur after forking, so the expand will only happen on new VMAs. 29011da177e4SLinus Torvalds */ 29026c28ca64SLiam Howlett if (vma && vma->vm_end == addr && !vma_policy(vma) && 29036c28ca64SLiam Howlett can_vma_merge_after(vma, flags, NULL, NULL, 29046c28ca64SLiam Howlett addr >> PAGE_SHIFT, NULL_VM_UFFD_CTX, NULL)) { 290592fed820SLiam R. Howlett if (vma_iter_prealloc(vmi)) 2906675eaca1SAlistair Popple goto unacct_fail; 290728c5609fSLiam Howlett 290828c5609fSLiam Howlett vma_adjust_trans_huge(vma, vma->vm_start, addr + len, 0); 2909287051b1SLiam R. Howlett init_vma_prep(&vp, vma); 2910287051b1SLiam R. Howlett vma_prepare(&vp); 29112e7ce7d3SLiam R. Howlett vma->vm_end = addr + len; 2912*1c71222eSSuren Baghdasaryan vm_flags_set(vma, VM_SOFTDIRTY); 291392fed820SLiam R. Howlett vma_iter_store(vmi, vma); 29142e7ce7d3SLiam R. Howlett 2915287051b1SLiam R. Howlett vma_complete(&vp, vmi, mm); 29162e7ce7d3SLiam R. Howlett khugepaged_enter_vma(vma, flags); 29172e7ce7d3SLiam R. Howlett goto out; 29182e7ce7d3SLiam R. Howlett } 29192e7ce7d3SLiam R. Howlett 29202e7ce7d3SLiam R. Howlett /* create a vma struct for an anonymous mapping */ 29212e7ce7d3SLiam R. Howlett vma = vm_area_alloc(mm); 29222e7ce7d3SLiam R. Howlett if (!vma) 2923675eaca1SAlistair Popple goto unacct_fail; 29241da177e4SLinus Torvalds 2925bfd40eafSKirill A. Shutemov vma_set_anonymous(vma); 29261da177e4SLinus Torvalds vma->vm_start = addr; 29271da177e4SLinus Torvalds vma->vm_end = addr + len; 29282e7ce7d3SLiam R. Howlett vma->vm_pgoff = addr >> PAGE_SHIFT; 2929*1c71222eSSuren Baghdasaryan vm_flags_init(vma, flags); 29303ed75eb8SColy Li vma->vm_page_prot = vm_get_page_prot(flags); 293192fed820SLiam R. Howlett if (vma_iter_store_gfp(vmi, vma, GFP_KERNEL)) 29322e7ce7d3SLiam R. Howlett goto mas_store_fail; 2933d4af56c5SLiam R. Howlett 29342e7ce7d3SLiam R. Howlett mm->map_count++; 29351da177e4SLinus Torvalds out: 29363af9e859SEric B Munson perf_event_mmap(vma); 29371da177e4SLinus Torvalds mm->total_vm += len >> PAGE_SHIFT; 293884638335SKonstantin Khlebnikov mm->data_vm += len >> PAGE_SHIFT; 2939128557ffSMichel Lespinasse if (flags & VM_LOCKED) 2940ba470de4SRik van Riel mm->locked_vm += (len >> PAGE_SHIFT); 2941*1c71222eSSuren Baghdasaryan vm_flags_set(vma, VM_SOFTDIRTY); 2942763ecb03SLiam R. Howlett validate_mm(mm); 29435d22fc25SLinus Torvalds return 0; 2944d4af56c5SLiam R. Howlett 29452e7ce7d3SLiam R. Howlett mas_store_fail: 2946d4af56c5SLiam R. Howlett vm_area_free(vma); 2947675eaca1SAlistair Popple unacct_fail: 29482e7ce7d3SLiam R. Howlett vm_unacct_memory(len >> PAGE_SHIFT); 29492e7ce7d3SLiam R. Howlett return -ENOMEM; 29501da177e4SLinus Torvalds } 29511da177e4SLinus Torvalds 2952bb177a73SMichal Hocko int vm_brk_flags(unsigned long addr, unsigned long request, unsigned long flags) 2953e4eb1ff6SLinus Torvalds { 2954e4eb1ff6SLinus Torvalds struct mm_struct *mm = current->mm; 29552e7ce7d3SLiam R. Howlett struct vm_area_struct *vma = NULL; 2956bb177a73SMichal Hocko unsigned long len; 29575d22fc25SLinus Torvalds int ret; 2958128557ffSMichel Lespinasse bool populate; 2959897ab3e0SMike Rapoport LIST_HEAD(uf); 296092fed820SLiam R. Howlett VMA_ITERATOR(vmi, mm, addr); 2961e4eb1ff6SLinus Torvalds 2962bb177a73SMichal Hocko len = PAGE_ALIGN(request); 2963bb177a73SMichal Hocko if (len < request) 2964bb177a73SMichal Hocko return -ENOMEM; 2965bb177a73SMichal Hocko if (!len) 2966bb177a73SMichal Hocko return 0; 2967bb177a73SMichal Hocko 2968d8ed45c5SMichel Lespinasse if (mmap_write_lock_killable(mm)) 29692d6c9282SMichal Hocko return -EINTR; 29702d6c9282SMichal Hocko 29712e7ce7d3SLiam R. Howlett /* Until we need other flags, refuse anything except VM_EXEC. */ 29722e7ce7d3SLiam R. Howlett if ((flags & (~VM_EXEC)) != 0) 29732e7ce7d3SLiam R. Howlett return -EINVAL; 29742e7ce7d3SLiam R. Howlett 29752e7ce7d3SLiam R. Howlett ret = check_brk_limits(addr, len); 29762e7ce7d3SLiam R. Howlett if (ret) 29772e7ce7d3SLiam R. Howlett goto limits_failed; 29782e7ce7d3SLiam R. Howlett 2979183654ceSLiam R. Howlett ret = do_vmi_munmap(&vmi, mm, addr, len, &uf, 0); 29802e7ce7d3SLiam R. Howlett if (ret) 29812e7ce7d3SLiam R. Howlett goto munmap_failed; 29822e7ce7d3SLiam R. Howlett 298392fed820SLiam R. Howlett vma = vma_prev(&vmi); 298492fed820SLiam R. Howlett ret = do_brk_flags(&vmi, vma, addr, len, flags); 2985128557ffSMichel Lespinasse populate = ((mm->def_flags & VM_LOCKED) != 0); 2986d8ed45c5SMichel Lespinasse mmap_write_unlock(mm); 2987897ab3e0SMike Rapoport userfaultfd_unmap_complete(mm, &uf); 29885d22fc25SLinus Torvalds if (populate && !ret) 2989128557ffSMichel Lespinasse mm_populate(addr, len); 2990e4eb1ff6SLinus Torvalds return ret; 29912e7ce7d3SLiam R. Howlett 29922e7ce7d3SLiam R. Howlett munmap_failed: 29932e7ce7d3SLiam R. Howlett limits_failed: 29942e7ce7d3SLiam R. Howlett mmap_write_unlock(mm); 29952e7ce7d3SLiam R. Howlett return ret; 2996e4eb1ff6SLinus Torvalds } 299716e72e9bSDenys Vlasenko EXPORT_SYMBOL(vm_brk_flags); 299816e72e9bSDenys Vlasenko 299916e72e9bSDenys Vlasenko int vm_brk(unsigned long addr, unsigned long len) 300016e72e9bSDenys Vlasenko { 300116e72e9bSDenys Vlasenko return vm_brk_flags(addr, len, 0); 300216e72e9bSDenys Vlasenko } 3003e4eb1ff6SLinus Torvalds EXPORT_SYMBOL(vm_brk); 30041da177e4SLinus Torvalds 30051da177e4SLinus Torvalds /* Release all mmaps. */ 30061da177e4SLinus Torvalds void exit_mmap(struct mm_struct *mm) 30071da177e4SLinus Torvalds { 3008d16dfc55SPeter Zijlstra struct mmu_gather tlb; 3009ba470de4SRik van Riel struct vm_area_struct *vma; 30101da177e4SLinus Torvalds unsigned long nr_accounted = 0; 3011763ecb03SLiam R. Howlett MA_STATE(mas, &mm->mm_mt, 0, 0); 3012763ecb03SLiam R. Howlett int count = 0; 30131da177e4SLinus Torvalds 3014d6dd61c8SJeremy Fitzhardinge /* mm's last user has gone, and its about to be pulled down */ 3015cddb8a5cSAndrea Arcangeli mmu_notifier_release(mm); 3016d6dd61c8SJeremy Fitzhardinge 3017bf3980c8SSuren Baghdasaryan mmap_read_lock(mm); 30189480c53eSJeremy Fitzhardinge arch_exit_mmap(mm); 30199480c53eSJeremy Fitzhardinge 3020763ecb03SLiam R. Howlett vma = mas_find(&mas, ULONG_MAX); 302164591e86SSuren Baghdasaryan if (!vma) { 302264591e86SSuren Baghdasaryan /* Can happen if dup_mmap() received an OOM */ 3023bf3980c8SSuren Baghdasaryan mmap_read_unlock(mm); 30249480c53eSJeremy Fitzhardinge return; 302564591e86SSuren Baghdasaryan } 30269480c53eSJeremy Fitzhardinge 30271da177e4SLinus Torvalds lru_add_drain(); 30281da177e4SLinus Torvalds flush_cache_mm(mm); 3029d8b45053SWill Deacon tlb_gather_mmu_fullmm(&tlb, mm); 3030901608d9SOleg Nesterov /* update_hiwater_rss(mm) here? but nobody should be looking */ 3031763ecb03SLiam R. Howlett /* Use ULONG_MAX here to ensure all VMAs in the mm are unmapped */ 3032763ecb03SLiam R. Howlett unmap_vmas(&tlb, &mm->mm_mt, vma, 0, ULONG_MAX); 3033bf3980c8SSuren Baghdasaryan mmap_read_unlock(mm); 3034bf3980c8SSuren Baghdasaryan 3035bf3980c8SSuren Baghdasaryan /* 3036bf3980c8SSuren Baghdasaryan * Set MMF_OOM_SKIP to hide this task from the oom killer/reaper 3037b3541d91SSuren Baghdasaryan * because the memory has been already freed. 3038bf3980c8SSuren Baghdasaryan */ 3039bf3980c8SSuren Baghdasaryan set_bit(MMF_OOM_SKIP, &mm->flags); 3040bf3980c8SSuren Baghdasaryan mmap_write_lock(mm); 3041763ecb03SLiam R. Howlett free_pgtables(&tlb, &mm->mm_mt, vma, FIRST_USER_ADDRESS, 3042763ecb03SLiam R. Howlett USER_PGTABLES_CEILING); 3043ae8eba8bSWill Deacon tlb_finish_mmu(&tlb); 30441da177e4SLinus Torvalds 3045763ecb03SLiam R. Howlett /* 3046763ecb03SLiam R. Howlett * Walk the list again, actually closing and freeing it, with preemption 3047763ecb03SLiam R. Howlett * enabled, without holding any MM locks besides the unreachable 3048763ecb03SLiam R. Howlett * mmap_write_lock. 3049763ecb03SLiam R. Howlett */ 3050763ecb03SLiam R. Howlett do { 30514f74d2c8SLinus Torvalds if (vma->vm_flags & VM_ACCOUNT) 30524f74d2c8SLinus Torvalds nr_accounted += vma_pages(vma); 3053763ecb03SLiam R. Howlett remove_vma(vma); 3054763ecb03SLiam R. Howlett count++; 30550a3b3c25SPaul E. McKenney cond_resched(); 3056763ecb03SLiam R. Howlett } while ((vma = mas_find(&mas, ULONG_MAX)) != NULL); 3057763ecb03SLiam R. Howlett 3058763ecb03SLiam R. Howlett BUG_ON(count != mm->map_count); 3059d4af56c5SLiam R. Howlett 3060d4af56c5SLiam R. Howlett trace_exit_mmap(mm); 3061d4af56c5SLiam R. Howlett __mt_destroy(&mm->mm_mt); 306264591e86SSuren Baghdasaryan mmap_write_unlock(mm); 30634f74d2c8SLinus Torvalds vm_unacct_memory(nr_accounted); 30641da177e4SLinus Torvalds } 30651da177e4SLinus Torvalds 30661da177e4SLinus Torvalds /* Insert vm structure into process list sorted by address 30671da177e4SLinus Torvalds * and into the inode's i_mmap tree. If vm_file is non-NULL 3068c8c06efaSDavidlohr Bueso * then i_mmap_rwsem is taken here. 30691da177e4SLinus Torvalds */ 30701da177e4SLinus Torvalds int insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma) 30711da177e4SLinus Torvalds { 3072d4af56c5SLiam R. Howlett unsigned long charged = vma_pages(vma); 30731da177e4SLinus Torvalds 3074d4af56c5SLiam R. Howlett 3075d0601a50SLiam R. Howlett if (find_vma_intersection(mm, vma->vm_start, vma->vm_end)) 3076c9d13f5fSChen Gang return -ENOMEM; 3077d4af56c5SLiam R. Howlett 3078c9d13f5fSChen Gang if ((vma->vm_flags & VM_ACCOUNT) && 3079d4af56c5SLiam R. Howlett security_vm_enough_memory_mm(mm, charged)) 3080c9d13f5fSChen Gang return -ENOMEM; 3081c9d13f5fSChen Gang 30821da177e4SLinus Torvalds /* 30831da177e4SLinus Torvalds * The vm_pgoff of a purely anonymous vma should be irrelevant 30841da177e4SLinus Torvalds * until its first write fault, when page's anon_vma and index 30851da177e4SLinus Torvalds * are set. But now set the vm_pgoff it will almost certainly 30861da177e4SLinus Torvalds * end up with (unless mremap moves it elsewhere before that 30871da177e4SLinus Torvalds * first wfault), so /proc/pid/maps tells a consistent story. 30881da177e4SLinus Torvalds * 30891da177e4SLinus Torvalds * By setting it to reflect the virtual start address of the 30901da177e4SLinus Torvalds * vma, merges and splits can happen in a seamless way, just 30911da177e4SLinus Torvalds * using the existing file pgoff checks and manipulations. 30928332326eSLiao Pingfang * Similarly in do_mmap and in do_brk_flags. 30931da177e4SLinus Torvalds */ 30948a9cc3b5SOleg Nesterov if (vma_is_anonymous(vma)) { 30951da177e4SLinus Torvalds BUG_ON(vma->anon_vma); 30961da177e4SLinus Torvalds vma->vm_pgoff = vma->vm_start >> PAGE_SHIFT; 30971da177e4SLinus Torvalds } 30982b144498SSrikar Dronamraju 3099763ecb03SLiam R. Howlett if (vma_link(mm, vma)) { 3100d4af56c5SLiam R. Howlett vm_unacct_memory(charged); 3101d4af56c5SLiam R. Howlett return -ENOMEM; 3102d4af56c5SLiam R. Howlett } 3103d4af56c5SLiam R. Howlett 31041da177e4SLinus Torvalds return 0; 31051da177e4SLinus Torvalds } 31061da177e4SLinus Torvalds 31071da177e4SLinus Torvalds /* 31081da177e4SLinus Torvalds * Copy the vma structure to a new location in the same mm, 31091da177e4SLinus Torvalds * prior to moving page table entries, to effect an mremap move. 31101da177e4SLinus Torvalds */ 31111da177e4SLinus Torvalds struct vm_area_struct *copy_vma(struct vm_area_struct **vmap, 311238a76013SMichel Lespinasse unsigned long addr, unsigned long len, pgoff_t pgoff, 311338a76013SMichel Lespinasse bool *need_rmap_locks) 31141da177e4SLinus Torvalds { 31151da177e4SLinus Torvalds struct vm_area_struct *vma = *vmap; 31161da177e4SLinus Torvalds unsigned long vma_start = vma->vm_start; 31171da177e4SLinus Torvalds struct mm_struct *mm = vma->vm_mm; 31181da177e4SLinus Torvalds struct vm_area_struct *new_vma, *prev; 3119948f017bSAndrea Arcangeli bool faulted_in_anon_vma = true; 3120076f16bfSLiam R. Howlett VMA_ITERATOR(vmi, mm, addr); 31211da177e4SLinus Torvalds 3122d4af56c5SLiam R. Howlett validate_mm_mt(mm); 31231da177e4SLinus Torvalds /* 31241da177e4SLinus Torvalds * If anonymous vma has not yet been faulted, update new pgoff 31251da177e4SLinus Torvalds * to match new location, to increase its chance of merging. 31261da177e4SLinus Torvalds */ 3127ce75799bSOleg Nesterov if (unlikely(vma_is_anonymous(vma) && !vma->anon_vma)) { 31281da177e4SLinus Torvalds pgoff = addr >> PAGE_SHIFT; 3129948f017bSAndrea Arcangeli faulted_in_anon_vma = false; 3130948f017bSAndrea Arcangeli } 31311da177e4SLinus Torvalds 3132763ecb03SLiam R. Howlett new_vma = find_vma_prev(mm, addr, &prev); 3133763ecb03SLiam R. Howlett if (new_vma && new_vma->vm_start < addr + len) 31346597d783SHugh Dickins return NULL; /* should never get here */ 3135524e00b3SLiam R. Howlett 31369760ebffSLiam R. Howlett new_vma = vma_merge(&vmi, mm, prev, addr, addr + len, vma->vm_flags, 313719a809afSAndrea Arcangeli vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma), 31385c26f6acSSuren Baghdasaryan vma->vm_userfaultfd_ctx, anon_vma_name(vma)); 31391da177e4SLinus Torvalds if (new_vma) { 31401da177e4SLinus Torvalds /* 31411da177e4SLinus Torvalds * Source vma may have been merged into new_vma 31421da177e4SLinus Torvalds */ 3143948f017bSAndrea Arcangeli if (unlikely(vma_start >= new_vma->vm_start && 3144948f017bSAndrea Arcangeli vma_start < new_vma->vm_end)) { 3145948f017bSAndrea Arcangeli /* 3146948f017bSAndrea Arcangeli * The only way we can get a vma_merge with 3147948f017bSAndrea Arcangeli * self during an mremap is if the vma hasn't 3148948f017bSAndrea Arcangeli * been faulted in yet and we were allowed to 3149948f017bSAndrea Arcangeli * reset the dst vma->vm_pgoff to the 3150948f017bSAndrea Arcangeli * destination address of the mremap to allow 3151948f017bSAndrea Arcangeli * the merge to happen. mremap must change the 3152948f017bSAndrea Arcangeli * vm_pgoff linearity between src and dst vmas 3153948f017bSAndrea Arcangeli * (in turn preventing a vma_merge) to be 3154948f017bSAndrea Arcangeli * safe. It is only safe to keep the vm_pgoff 3155948f017bSAndrea Arcangeli * linear if there are no pages mapped yet. 3156948f017bSAndrea Arcangeli */ 315781d1b09cSSasha Levin VM_BUG_ON_VMA(faulted_in_anon_vma, new_vma); 315838a76013SMichel Lespinasse *vmap = vma = new_vma; 3159108d6642SMichel Lespinasse } 316038a76013SMichel Lespinasse *need_rmap_locks = (new_vma->vm_pgoff <= vma->vm_pgoff); 31611da177e4SLinus Torvalds } else { 31623928d4f5SLinus Torvalds new_vma = vm_area_dup(vma); 3163e3975891SChen Gang if (!new_vma) 3164e3975891SChen Gang goto out; 31651da177e4SLinus Torvalds new_vma->vm_start = addr; 31661da177e4SLinus Torvalds new_vma->vm_end = addr + len; 31671da177e4SLinus Torvalds new_vma->vm_pgoff = pgoff; 3168ef0855d3SOleg Nesterov if (vma_dup_policy(vma, new_vma)) 3169523d4e20SMichel Lespinasse goto out_free_vma; 3170523d4e20SMichel Lespinasse if (anon_vma_clone(new_vma, vma)) 3171523d4e20SMichel Lespinasse goto out_free_mempol; 3172e9714acfSKonstantin Khlebnikov if (new_vma->vm_file) 31731da177e4SLinus Torvalds get_file(new_vma->vm_file); 31741da177e4SLinus Torvalds if (new_vma->vm_ops && new_vma->vm_ops->open) 31751da177e4SLinus Torvalds new_vma->vm_ops->open(new_vma); 3176763ecb03SLiam R. Howlett if (vma_link(mm, new_vma)) 3177524e00b3SLiam R. Howlett goto out_vma_link; 317838a76013SMichel Lespinasse *need_rmap_locks = false; 31791da177e4SLinus Torvalds } 3180d4af56c5SLiam R. Howlett validate_mm_mt(mm); 31811da177e4SLinus Torvalds return new_vma; 31825beb4930SRik van Riel 3183524e00b3SLiam R. Howlett out_vma_link: 3184524e00b3SLiam R. Howlett if (new_vma->vm_ops && new_vma->vm_ops->close) 3185524e00b3SLiam R. Howlett new_vma->vm_ops->close(new_vma); 318692b73996SLiam Howlett 318792b73996SLiam Howlett if (new_vma->vm_file) 318892b73996SLiam Howlett fput(new_vma->vm_file); 318992b73996SLiam Howlett 319092b73996SLiam Howlett unlink_anon_vmas(new_vma); 31915beb4930SRik van Riel out_free_mempol: 3192ef0855d3SOleg Nesterov mpol_put(vma_policy(new_vma)); 31935beb4930SRik van Riel out_free_vma: 31943928d4f5SLinus Torvalds vm_area_free(new_vma); 3195e3975891SChen Gang out: 3196d4af56c5SLiam R. Howlett validate_mm_mt(mm); 31975beb4930SRik van Riel return NULL; 31981da177e4SLinus Torvalds } 3199119f657cSakpm@osdl.org 3200119f657cSakpm@osdl.org /* 3201119f657cSakpm@osdl.org * Return true if the calling process may expand its vm space by the passed 3202119f657cSakpm@osdl.org * number of pages 3203119f657cSakpm@osdl.org */ 320484638335SKonstantin Khlebnikov bool may_expand_vm(struct mm_struct *mm, vm_flags_t flags, unsigned long npages) 3205119f657cSakpm@osdl.org { 320684638335SKonstantin Khlebnikov if (mm->total_vm + npages > rlimit(RLIMIT_AS) >> PAGE_SHIFT) 320784638335SKonstantin Khlebnikov return false; 3208119f657cSakpm@osdl.org 3209d977d56cSKonstantin Khlebnikov if (is_data_mapping(flags) && 3210d977d56cSKonstantin Khlebnikov mm->data_vm + npages > rlimit(RLIMIT_DATA) >> PAGE_SHIFT) { 3211f4fcd558SKonstantin Khlebnikov /* Workaround for Valgrind */ 3212f4fcd558SKonstantin Khlebnikov if (rlimit(RLIMIT_DATA) == 0 && 3213f4fcd558SKonstantin Khlebnikov mm->data_vm + npages <= rlimit_max(RLIMIT_DATA) >> PAGE_SHIFT) 3214f4fcd558SKonstantin Khlebnikov return true; 321557a7702bSDavid Woodhouse 321657a7702bSDavid Woodhouse pr_warn_once("%s (%d): VmData %lu exceed data ulimit %lu. Update limits%s.\n", 3217d977d56cSKonstantin Khlebnikov current->comm, current->pid, 3218d977d56cSKonstantin Khlebnikov (mm->data_vm + npages) << PAGE_SHIFT, 321957a7702bSDavid Woodhouse rlimit(RLIMIT_DATA), 322057a7702bSDavid Woodhouse ignore_rlimit_data ? "" : " or use boot option ignore_rlimit_data"); 322157a7702bSDavid Woodhouse 322257a7702bSDavid Woodhouse if (!ignore_rlimit_data) 3223d977d56cSKonstantin Khlebnikov return false; 3224d977d56cSKonstantin Khlebnikov } 3225119f657cSakpm@osdl.org 322684638335SKonstantin Khlebnikov return true; 322784638335SKonstantin Khlebnikov } 322884638335SKonstantin Khlebnikov 322984638335SKonstantin Khlebnikov void vm_stat_account(struct mm_struct *mm, vm_flags_t flags, long npages) 323084638335SKonstantin Khlebnikov { 32317866076bSPeng Liu WRITE_ONCE(mm->total_vm, READ_ONCE(mm->total_vm)+npages); 323284638335SKonstantin Khlebnikov 3233d977d56cSKonstantin Khlebnikov if (is_exec_mapping(flags)) 323484638335SKonstantin Khlebnikov mm->exec_vm += npages; 3235d977d56cSKonstantin Khlebnikov else if (is_stack_mapping(flags)) 323684638335SKonstantin Khlebnikov mm->stack_vm += npages; 3237d977d56cSKonstantin Khlebnikov else if (is_data_mapping(flags)) 323884638335SKonstantin Khlebnikov mm->data_vm += npages; 3239119f657cSakpm@osdl.org } 3240fa5dc22fSRoland McGrath 3241b3ec9f33SSouptick Joarder static vm_fault_t special_mapping_fault(struct vm_fault *vmf); 3242a62c34bdSAndy Lutomirski 3243a62c34bdSAndy Lutomirski /* 3244a62c34bdSAndy Lutomirski * Having a close hook prevents vma merging regardless of flags. 3245a62c34bdSAndy Lutomirski */ 3246a62c34bdSAndy Lutomirski static void special_mapping_close(struct vm_area_struct *vma) 3247a62c34bdSAndy Lutomirski { 3248a62c34bdSAndy Lutomirski } 3249a62c34bdSAndy Lutomirski 3250a62c34bdSAndy Lutomirski static const char *special_mapping_name(struct vm_area_struct *vma) 3251a62c34bdSAndy Lutomirski { 3252a62c34bdSAndy Lutomirski return ((struct vm_special_mapping *)vma->vm_private_data)->name; 3253a62c34bdSAndy Lutomirski } 3254a62c34bdSAndy Lutomirski 325514d07113SBrian Geffon static int special_mapping_mremap(struct vm_area_struct *new_vma) 3256b059a453SDmitry Safonov { 3257b059a453SDmitry Safonov struct vm_special_mapping *sm = new_vma->vm_private_data; 3258b059a453SDmitry Safonov 3259280e87e9SDmitry Safonov if (WARN_ON_ONCE(current->mm != new_vma->vm_mm)) 3260280e87e9SDmitry Safonov return -EFAULT; 3261280e87e9SDmitry Safonov 3262b059a453SDmitry Safonov if (sm->mremap) 3263b059a453SDmitry Safonov return sm->mremap(sm, new_vma); 3264280e87e9SDmitry Safonov 3265b059a453SDmitry Safonov return 0; 3266b059a453SDmitry Safonov } 3267b059a453SDmitry Safonov 3268871402e0SDmitry Safonov static int special_mapping_split(struct vm_area_struct *vma, unsigned long addr) 3269871402e0SDmitry Safonov { 3270871402e0SDmitry Safonov /* 3271871402e0SDmitry Safonov * Forbid splitting special mappings - kernel has expectations over 3272871402e0SDmitry Safonov * the number of pages in mapping. Together with VM_DONTEXPAND 3273871402e0SDmitry Safonov * the size of vma should stay the same over the special mapping's 3274871402e0SDmitry Safonov * lifetime. 3275871402e0SDmitry Safonov */ 3276871402e0SDmitry Safonov return -EINVAL; 3277871402e0SDmitry Safonov } 3278871402e0SDmitry Safonov 3279a62c34bdSAndy Lutomirski static const struct vm_operations_struct special_mapping_vmops = { 3280a62c34bdSAndy Lutomirski .close = special_mapping_close, 3281a62c34bdSAndy Lutomirski .fault = special_mapping_fault, 3282b059a453SDmitry Safonov .mremap = special_mapping_mremap, 3283a62c34bdSAndy Lutomirski .name = special_mapping_name, 3284af34ebebSDmitry Safonov /* vDSO code relies that VVAR can't be accessed remotely */ 3285af34ebebSDmitry Safonov .access = NULL, 3286871402e0SDmitry Safonov .may_split = special_mapping_split, 3287a62c34bdSAndy Lutomirski }; 3288a62c34bdSAndy Lutomirski 3289a62c34bdSAndy Lutomirski static const struct vm_operations_struct legacy_special_mapping_vmops = { 3290a62c34bdSAndy Lutomirski .close = special_mapping_close, 3291a62c34bdSAndy Lutomirski .fault = special_mapping_fault, 3292a62c34bdSAndy Lutomirski }; 3293fa5dc22fSRoland McGrath 3294b3ec9f33SSouptick Joarder static vm_fault_t special_mapping_fault(struct vm_fault *vmf) 3295fa5dc22fSRoland McGrath { 329611bac800SDave Jiang struct vm_area_struct *vma = vmf->vma; 3297b1d0e4f5SNick Piggin pgoff_t pgoff; 3298fa5dc22fSRoland McGrath struct page **pages; 3299fa5dc22fSRoland McGrath 3300f872f540SAndy Lutomirski if (vma->vm_ops == &legacy_special_mapping_vmops) { 3301a62c34bdSAndy Lutomirski pages = vma->vm_private_data; 3302f872f540SAndy Lutomirski } else { 3303f872f540SAndy Lutomirski struct vm_special_mapping *sm = vma->vm_private_data; 3304f872f540SAndy Lutomirski 3305f872f540SAndy Lutomirski if (sm->fault) 330611bac800SDave Jiang return sm->fault(sm, vmf->vma, vmf); 3307f872f540SAndy Lutomirski 3308f872f540SAndy Lutomirski pages = sm->pages; 3309f872f540SAndy Lutomirski } 3310a62c34bdSAndy Lutomirski 33118a9cc3b5SOleg Nesterov for (pgoff = vmf->pgoff; pgoff && *pages; ++pages) 3312b1d0e4f5SNick Piggin pgoff--; 3313fa5dc22fSRoland McGrath 3314fa5dc22fSRoland McGrath if (*pages) { 3315fa5dc22fSRoland McGrath struct page *page = *pages; 3316fa5dc22fSRoland McGrath get_page(page); 3317b1d0e4f5SNick Piggin vmf->page = page; 3318b1d0e4f5SNick Piggin return 0; 3319fa5dc22fSRoland McGrath } 3320fa5dc22fSRoland McGrath 3321b1d0e4f5SNick Piggin return VM_FAULT_SIGBUS; 3322fa5dc22fSRoland McGrath } 3323fa5dc22fSRoland McGrath 3324a62c34bdSAndy Lutomirski static struct vm_area_struct *__install_special_mapping( 3325a62c34bdSAndy Lutomirski struct mm_struct *mm, 3326fa5dc22fSRoland McGrath unsigned long addr, unsigned long len, 332727f28b97SChen Gang unsigned long vm_flags, void *priv, 332827f28b97SChen Gang const struct vm_operations_struct *ops) 3329fa5dc22fSRoland McGrath { 3330462e635eSTavis Ormandy int ret; 3331fa5dc22fSRoland McGrath struct vm_area_struct *vma; 3332fa5dc22fSRoland McGrath 3333d4af56c5SLiam R. Howlett validate_mm_mt(mm); 3334490fc053SLinus Torvalds vma = vm_area_alloc(mm); 3335fa5dc22fSRoland McGrath if (unlikely(vma == NULL)) 33363935ed6aSStefani Seibold return ERR_PTR(-ENOMEM); 3337fa5dc22fSRoland McGrath 3338fa5dc22fSRoland McGrath vma->vm_start = addr; 3339fa5dc22fSRoland McGrath vma->vm_end = addr + len; 3340fa5dc22fSRoland McGrath 3341e430a95aSSuren Baghdasaryan vm_flags_init(vma, (vm_flags | mm->def_flags | 3342e430a95aSSuren Baghdasaryan VM_DONTEXPAND | VM_SOFTDIRTY) & ~VM_LOCKED_MASK); 33433ed75eb8SColy Li vma->vm_page_prot = vm_get_page_prot(vma->vm_flags); 3344fa5dc22fSRoland McGrath 3345a62c34bdSAndy Lutomirski vma->vm_ops = ops; 3346a62c34bdSAndy Lutomirski vma->vm_private_data = priv; 3347fa5dc22fSRoland McGrath 3348462e635eSTavis Ormandy ret = insert_vm_struct(mm, vma); 3349462e635eSTavis Ormandy if (ret) 3350462e635eSTavis Ormandy goto out; 3351fa5dc22fSRoland McGrath 335284638335SKonstantin Khlebnikov vm_stat_account(mm, vma->vm_flags, len >> PAGE_SHIFT); 3353fa5dc22fSRoland McGrath 3354cdd6c482SIngo Molnar perf_event_mmap(vma); 3355089dd79dSPeter Zijlstra 3356d4af56c5SLiam R. Howlett validate_mm_mt(mm); 33573935ed6aSStefani Seibold return vma; 3358462e635eSTavis Ormandy 3359462e635eSTavis Ormandy out: 33603928d4f5SLinus Torvalds vm_area_free(vma); 3361d4af56c5SLiam R. Howlett validate_mm_mt(mm); 33623935ed6aSStefani Seibold return ERR_PTR(ret); 33633935ed6aSStefani Seibold } 33643935ed6aSStefani Seibold 33652eefd878SDmitry Safonov bool vma_is_special_mapping(const struct vm_area_struct *vma, 33662eefd878SDmitry Safonov const struct vm_special_mapping *sm) 33672eefd878SDmitry Safonov { 33682eefd878SDmitry Safonov return vma->vm_private_data == sm && 33692eefd878SDmitry Safonov (vma->vm_ops == &special_mapping_vmops || 33702eefd878SDmitry Safonov vma->vm_ops == &legacy_special_mapping_vmops); 33712eefd878SDmitry Safonov } 33722eefd878SDmitry Safonov 3373a62c34bdSAndy Lutomirski /* 3374c1e8d7c6SMichel Lespinasse * Called with mm->mmap_lock held for writing. 3375a62c34bdSAndy Lutomirski * Insert a new vma covering the given region, with the given flags. 3376a62c34bdSAndy Lutomirski * Its pages are supplied by the given array of struct page *. 3377a62c34bdSAndy Lutomirski * The array can be shorter than len >> PAGE_SHIFT if it's null-terminated. 3378a62c34bdSAndy Lutomirski * The region past the last page supplied will always produce SIGBUS. 3379a62c34bdSAndy Lutomirski * The array pointer and the pages it points to are assumed to stay alive 3380a62c34bdSAndy Lutomirski * for as long as this mapping might exist. 3381a62c34bdSAndy Lutomirski */ 3382a62c34bdSAndy Lutomirski struct vm_area_struct *_install_special_mapping( 3383a62c34bdSAndy Lutomirski struct mm_struct *mm, 3384a62c34bdSAndy Lutomirski unsigned long addr, unsigned long len, 3385a62c34bdSAndy Lutomirski unsigned long vm_flags, const struct vm_special_mapping *spec) 3386a62c34bdSAndy Lutomirski { 338727f28b97SChen Gang return __install_special_mapping(mm, addr, len, vm_flags, (void *)spec, 338827f28b97SChen Gang &special_mapping_vmops); 3389a62c34bdSAndy Lutomirski } 3390a62c34bdSAndy Lutomirski 33913935ed6aSStefani Seibold int install_special_mapping(struct mm_struct *mm, 33923935ed6aSStefani Seibold unsigned long addr, unsigned long len, 33933935ed6aSStefani Seibold unsigned long vm_flags, struct page **pages) 33943935ed6aSStefani Seibold { 3395a62c34bdSAndy Lutomirski struct vm_area_struct *vma = __install_special_mapping( 339627f28b97SChen Gang mm, addr, len, vm_flags, (void *)pages, 339727f28b97SChen Gang &legacy_special_mapping_vmops); 33983935ed6aSStefani Seibold 339914bd5b45SDuan Jiong return PTR_ERR_OR_ZERO(vma); 3400fa5dc22fSRoland McGrath } 34017906d00cSAndrea Arcangeli 34027906d00cSAndrea Arcangeli static DEFINE_MUTEX(mm_all_locks_mutex); 34037906d00cSAndrea Arcangeli 3404454ed842SPeter Zijlstra static void vm_lock_anon_vma(struct mm_struct *mm, struct anon_vma *anon_vma) 34057906d00cSAndrea Arcangeli { 3406f808c13fSDavidlohr Bueso if (!test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_root.rb_node)) { 34077906d00cSAndrea Arcangeli /* 34087906d00cSAndrea Arcangeli * The LSB of head.next can't change from under us 34097906d00cSAndrea Arcangeli * because we hold the mm_all_locks_mutex. 34107906d00cSAndrea Arcangeli */ 3411da1c55f1SMichel Lespinasse down_write_nest_lock(&anon_vma->root->rwsem, &mm->mmap_lock); 34127906d00cSAndrea Arcangeli /* 34137906d00cSAndrea Arcangeli * We can safely modify head.next after taking the 34145a505085SIngo Molnar * anon_vma->root->rwsem. If some other vma in this mm shares 34157906d00cSAndrea Arcangeli * the same anon_vma we won't take it again. 34167906d00cSAndrea Arcangeli * 34177906d00cSAndrea Arcangeli * No need of atomic instructions here, head.next 34187906d00cSAndrea Arcangeli * can't change from under us thanks to the 34195a505085SIngo Molnar * anon_vma->root->rwsem. 34207906d00cSAndrea Arcangeli */ 34217906d00cSAndrea Arcangeli if (__test_and_set_bit(0, (unsigned long *) 3422f808c13fSDavidlohr Bueso &anon_vma->root->rb_root.rb_root.rb_node)) 34237906d00cSAndrea Arcangeli BUG(); 34247906d00cSAndrea Arcangeli } 34257906d00cSAndrea Arcangeli } 34267906d00cSAndrea Arcangeli 3427454ed842SPeter Zijlstra static void vm_lock_mapping(struct mm_struct *mm, struct address_space *mapping) 34287906d00cSAndrea Arcangeli { 34297906d00cSAndrea Arcangeli if (!test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) { 34307906d00cSAndrea Arcangeli /* 34317906d00cSAndrea Arcangeli * AS_MM_ALL_LOCKS can't change from under us because 34327906d00cSAndrea Arcangeli * we hold the mm_all_locks_mutex. 34337906d00cSAndrea Arcangeli * 34347906d00cSAndrea Arcangeli * Operations on ->flags have to be atomic because 34357906d00cSAndrea Arcangeli * even if AS_MM_ALL_LOCKS is stable thanks to the 34367906d00cSAndrea Arcangeli * mm_all_locks_mutex, there may be other cpus 34377906d00cSAndrea Arcangeli * changing other bitflags in parallel to us. 34387906d00cSAndrea Arcangeli */ 34397906d00cSAndrea Arcangeli if (test_and_set_bit(AS_MM_ALL_LOCKS, &mapping->flags)) 34407906d00cSAndrea Arcangeli BUG(); 3441da1c55f1SMichel Lespinasse down_write_nest_lock(&mapping->i_mmap_rwsem, &mm->mmap_lock); 34427906d00cSAndrea Arcangeli } 34437906d00cSAndrea Arcangeli } 34447906d00cSAndrea Arcangeli 34457906d00cSAndrea Arcangeli /* 34467906d00cSAndrea Arcangeli * This operation locks against the VM for all pte/vma/mm related 34477906d00cSAndrea Arcangeli * operations that could ever happen on a certain mm. This includes 34487906d00cSAndrea Arcangeli * vmtruncate, try_to_unmap, and all page faults. 34497906d00cSAndrea Arcangeli * 3450c1e8d7c6SMichel Lespinasse * The caller must take the mmap_lock in write mode before calling 34517906d00cSAndrea Arcangeli * mm_take_all_locks(). The caller isn't allowed to release the 3452c1e8d7c6SMichel Lespinasse * mmap_lock until mm_drop_all_locks() returns. 34537906d00cSAndrea Arcangeli * 3454c1e8d7c6SMichel Lespinasse * mmap_lock in write mode is required in order to block all operations 34557906d00cSAndrea Arcangeli * that could modify pagetables and free pages without need of 345627ba0644SKirill A. Shutemov * altering the vma layout. It's also needed in write mode to avoid new 34577906d00cSAndrea Arcangeli * anon_vmas to be associated with existing vmas. 34587906d00cSAndrea Arcangeli * 34597906d00cSAndrea Arcangeli * A single task can't take more than one mm_take_all_locks() in a row 34607906d00cSAndrea Arcangeli * or it would deadlock. 34617906d00cSAndrea Arcangeli * 3462bf181b9fSMichel Lespinasse * The LSB in anon_vma->rb_root.rb_node and the AS_MM_ALL_LOCKS bitflag in 34637906d00cSAndrea Arcangeli * mapping->flags avoid to take the same lock twice, if more than one 34647906d00cSAndrea Arcangeli * vma in this mm is backed by the same anon_vma or address_space. 34657906d00cSAndrea Arcangeli * 346688f306b6SKirill A. Shutemov * We take locks in following order, accordingly to comment at beginning 346788f306b6SKirill A. Shutemov * of mm/rmap.c: 346888f306b6SKirill A. Shutemov * - all hugetlbfs_i_mmap_rwsem_key locks (aka mapping->i_mmap_rwsem for 346988f306b6SKirill A. Shutemov * hugetlb mapping); 347088f306b6SKirill A. Shutemov * - all i_mmap_rwsem locks; 347188f306b6SKirill A. Shutemov * - all anon_vma->rwseml 347288f306b6SKirill A. Shutemov * 347388f306b6SKirill A. Shutemov * We can take all locks within these types randomly because the VM code 347488f306b6SKirill A. Shutemov * doesn't nest them and we protected from parallel mm_take_all_locks() by 347588f306b6SKirill A. Shutemov * mm_all_locks_mutex. 34767906d00cSAndrea Arcangeli * 34777906d00cSAndrea Arcangeli * mm_take_all_locks() and mm_drop_all_locks are expensive operations 34787906d00cSAndrea Arcangeli * that may have to take thousand of locks. 34797906d00cSAndrea Arcangeli * 34807906d00cSAndrea Arcangeli * mm_take_all_locks() can fail if it's interrupted by signals. 34817906d00cSAndrea Arcangeli */ 34827906d00cSAndrea Arcangeli int mm_take_all_locks(struct mm_struct *mm) 34837906d00cSAndrea Arcangeli { 34847906d00cSAndrea Arcangeli struct vm_area_struct *vma; 34855beb4930SRik van Riel struct anon_vma_chain *avc; 3486763ecb03SLiam R. Howlett MA_STATE(mas, &mm->mm_mt, 0, 0); 34877906d00cSAndrea Arcangeli 3488325bca1fSRolf Eike Beer mmap_assert_write_locked(mm); 34897906d00cSAndrea Arcangeli 34907906d00cSAndrea Arcangeli mutex_lock(&mm_all_locks_mutex); 34917906d00cSAndrea Arcangeli 3492763ecb03SLiam R. Howlett mas_for_each(&mas, vma, ULONG_MAX) { 34937906d00cSAndrea Arcangeli if (signal_pending(current)) 34947906d00cSAndrea Arcangeli goto out_unlock; 349588f306b6SKirill A. Shutemov if (vma->vm_file && vma->vm_file->f_mapping && 349688f306b6SKirill A. Shutemov is_vm_hugetlb_page(vma)) 349788f306b6SKirill A. Shutemov vm_lock_mapping(mm, vma->vm_file->f_mapping); 349888f306b6SKirill A. Shutemov } 349988f306b6SKirill A. Shutemov 3500763ecb03SLiam R. Howlett mas_set(&mas, 0); 3501763ecb03SLiam R. Howlett mas_for_each(&mas, vma, ULONG_MAX) { 350288f306b6SKirill A. Shutemov if (signal_pending(current)) 350388f306b6SKirill A. Shutemov goto out_unlock; 350488f306b6SKirill A. Shutemov if (vma->vm_file && vma->vm_file->f_mapping && 350588f306b6SKirill A. Shutemov !is_vm_hugetlb_page(vma)) 3506454ed842SPeter Zijlstra vm_lock_mapping(mm, vma->vm_file->f_mapping); 35077906d00cSAndrea Arcangeli } 35087cd5a02fSPeter Zijlstra 3509763ecb03SLiam R. Howlett mas_set(&mas, 0); 3510763ecb03SLiam R. Howlett mas_for_each(&mas, vma, ULONG_MAX) { 35117cd5a02fSPeter Zijlstra if (signal_pending(current)) 35127cd5a02fSPeter Zijlstra goto out_unlock; 35137cd5a02fSPeter Zijlstra if (vma->anon_vma) 35145beb4930SRik van Riel list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) 35155beb4930SRik van Riel vm_lock_anon_vma(mm, avc->anon_vma); 35167cd5a02fSPeter Zijlstra } 35177cd5a02fSPeter Zijlstra 3518584cff54SKautuk Consul return 0; 35197906d00cSAndrea Arcangeli 35207906d00cSAndrea Arcangeli out_unlock: 35217906d00cSAndrea Arcangeli mm_drop_all_locks(mm); 3522584cff54SKautuk Consul return -EINTR; 35237906d00cSAndrea Arcangeli } 35247906d00cSAndrea Arcangeli 35257906d00cSAndrea Arcangeli static void vm_unlock_anon_vma(struct anon_vma *anon_vma) 35267906d00cSAndrea Arcangeli { 3527f808c13fSDavidlohr Bueso if (test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_root.rb_node)) { 35287906d00cSAndrea Arcangeli /* 35297906d00cSAndrea Arcangeli * The LSB of head.next can't change to 0 from under 35307906d00cSAndrea Arcangeli * us because we hold the mm_all_locks_mutex. 35317906d00cSAndrea Arcangeli * 35327906d00cSAndrea Arcangeli * We must however clear the bitflag before unlocking 3533bf181b9fSMichel Lespinasse * the vma so the users using the anon_vma->rb_root will 35347906d00cSAndrea Arcangeli * never see our bitflag. 35357906d00cSAndrea Arcangeli * 35367906d00cSAndrea Arcangeli * No need of atomic instructions here, head.next 35377906d00cSAndrea Arcangeli * can't change from under us until we release the 35385a505085SIngo Molnar * anon_vma->root->rwsem. 35397906d00cSAndrea Arcangeli */ 35407906d00cSAndrea Arcangeli if (!__test_and_clear_bit(0, (unsigned long *) 3541f808c13fSDavidlohr Bueso &anon_vma->root->rb_root.rb_root.rb_node)) 35427906d00cSAndrea Arcangeli BUG(); 354308b52706SKonstantin Khlebnikov anon_vma_unlock_write(anon_vma); 35447906d00cSAndrea Arcangeli } 35457906d00cSAndrea Arcangeli } 35467906d00cSAndrea Arcangeli 35477906d00cSAndrea Arcangeli static void vm_unlock_mapping(struct address_space *mapping) 35487906d00cSAndrea Arcangeli { 35497906d00cSAndrea Arcangeli if (test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) { 35507906d00cSAndrea Arcangeli /* 35517906d00cSAndrea Arcangeli * AS_MM_ALL_LOCKS can't change to 0 from under us 35527906d00cSAndrea Arcangeli * because we hold the mm_all_locks_mutex. 35537906d00cSAndrea Arcangeli */ 355483cde9e8SDavidlohr Bueso i_mmap_unlock_write(mapping); 35557906d00cSAndrea Arcangeli if (!test_and_clear_bit(AS_MM_ALL_LOCKS, 35567906d00cSAndrea Arcangeli &mapping->flags)) 35577906d00cSAndrea Arcangeli BUG(); 35587906d00cSAndrea Arcangeli } 35597906d00cSAndrea Arcangeli } 35607906d00cSAndrea Arcangeli 35617906d00cSAndrea Arcangeli /* 3562c1e8d7c6SMichel Lespinasse * The mmap_lock cannot be released by the caller until 35637906d00cSAndrea Arcangeli * mm_drop_all_locks() returns. 35647906d00cSAndrea Arcangeli */ 35657906d00cSAndrea Arcangeli void mm_drop_all_locks(struct mm_struct *mm) 35667906d00cSAndrea Arcangeli { 35677906d00cSAndrea Arcangeli struct vm_area_struct *vma; 35685beb4930SRik van Riel struct anon_vma_chain *avc; 3569763ecb03SLiam R. Howlett MA_STATE(mas, &mm->mm_mt, 0, 0); 35707906d00cSAndrea Arcangeli 3571325bca1fSRolf Eike Beer mmap_assert_write_locked(mm); 35727906d00cSAndrea Arcangeli BUG_ON(!mutex_is_locked(&mm_all_locks_mutex)); 35737906d00cSAndrea Arcangeli 3574763ecb03SLiam R. Howlett mas_for_each(&mas, vma, ULONG_MAX) { 35757906d00cSAndrea Arcangeli if (vma->anon_vma) 35765beb4930SRik van Riel list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) 35775beb4930SRik van Riel vm_unlock_anon_vma(avc->anon_vma); 35787906d00cSAndrea Arcangeli if (vma->vm_file && vma->vm_file->f_mapping) 35797906d00cSAndrea Arcangeli vm_unlock_mapping(vma->vm_file->f_mapping); 35807906d00cSAndrea Arcangeli } 35817906d00cSAndrea Arcangeli 35827906d00cSAndrea Arcangeli mutex_unlock(&mm_all_locks_mutex); 35837906d00cSAndrea Arcangeli } 35848feae131SDavid Howells 35858feae131SDavid Howells /* 35863edf41d8Sseokhoon.yoon * initialise the percpu counter for VM 35878feae131SDavid Howells */ 35888feae131SDavid Howells void __init mmap_init(void) 35898feae131SDavid Howells { 359000a62ce9SKOSAKI Motohiro int ret; 359100a62ce9SKOSAKI Motohiro 3592908c7f19STejun Heo ret = percpu_counter_init(&vm_committed_as, 0, GFP_KERNEL); 359300a62ce9SKOSAKI Motohiro VM_BUG_ON(ret); 35948feae131SDavid Howells } 3595c9b1d098SAndrew Shewmaker 3596c9b1d098SAndrew Shewmaker /* 3597c9b1d098SAndrew Shewmaker * Initialise sysctl_user_reserve_kbytes. 3598c9b1d098SAndrew Shewmaker * 3599c9b1d098SAndrew Shewmaker * This is intended to prevent a user from starting a single memory hogging 3600c9b1d098SAndrew Shewmaker * process, such that they cannot recover (kill the hog) in OVERCOMMIT_NEVER 3601c9b1d098SAndrew Shewmaker * mode. 3602c9b1d098SAndrew Shewmaker * 3603c9b1d098SAndrew Shewmaker * The default value is min(3% of free memory, 128MB) 3604c9b1d098SAndrew Shewmaker * 128MB is enough to recover with sshd/login, bash, and top/kill. 3605c9b1d098SAndrew Shewmaker */ 36061640879aSAndrew Shewmaker static int init_user_reserve(void) 3607c9b1d098SAndrew Shewmaker { 3608c9b1d098SAndrew Shewmaker unsigned long free_kbytes; 3609c9b1d098SAndrew Shewmaker 3610c41f012aSMichal Hocko free_kbytes = global_zone_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10); 3611c9b1d098SAndrew Shewmaker 3612c9b1d098SAndrew Shewmaker sysctl_user_reserve_kbytes = min(free_kbytes / 32, 1UL << 17); 3613c9b1d098SAndrew Shewmaker return 0; 3614c9b1d098SAndrew Shewmaker } 3615a64fb3cdSPaul Gortmaker subsys_initcall(init_user_reserve); 36164eeab4f5SAndrew Shewmaker 36174eeab4f5SAndrew Shewmaker /* 36184eeab4f5SAndrew Shewmaker * Initialise sysctl_admin_reserve_kbytes. 36194eeab4f5SAndrew Shewmaker * 36204eeab4f5SAndrew Shewmaker * The purpose of sysctl_admin_reserve_kbytes is to allow the sys admin 36214eeab4f5SAndrew Shewmaker * to log in and kill a memory hogging process. 36224eeab4f5SAndrew Shewmaker * 36234eeab4f5SAndrew Shewmaker * Systems with more than 256MB will reserve 8MB, enough to recover 36244eeab4f5SAndrew Shewmaker * with sshd, bash, and top in OVERCOMMIT_GUESS. Smaller systems will 36254eeab4f5SAndrew Shewmaker * only reserve 3% of free pages by default. 36264eeab4f5SAndrew Shewmaker */ 36271640879aSAndrew Shewmaker static int init_admin_reserve(void) 36284eeab4f5SAndrew Shewmaker { 36294eeab4f5SAndrew Shewmaker unsigned long free_kbytes; 36304eeab4f5SAndrew Shewmaker 3631c41f012aSMichal Hocko free_kbytes = global_zone_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10); 36324eeab4f5SAndrew Shewmaker 36334eeab4f5SAndrew Shewmaker sysctl_admin_reserve_kbytes = min(free_kbytes / 32, 1UL << 13); 36344eeab4f5SAndrew Shewmaker return 0; 36354eeab4f5SAndrew Shewmaker } 3636a64fb3cdSPaul Gortmaker subsys_initcall(init_admin_reserve); 36371640879aSAndrew Shewmaker 36381640879aSAndrew Shewmaker /* 36391640879aSAndrew Shewmaker * Reinititalise user and admin reserves if memory is added or removed. 36401640879aSAndrew Shewmaker * 36411640879aSAndrew Shewmaker * The default user reserve max is 128MB, and the default max for the 36421640879aSAndrew Shewmaker * admin reserve is 8MB. These are usually, but not always, enough to 36431640879aSAndrew Shewmaker * enable recovery from a memory hogging process using login/sshd, a shell, 36441640879aSAndrew Shewmaker * and tools like top. It may make sense to increase or even disable the 36451640879aSAndrew Shewmaker * reserve depending on the existence of swap or variations in the recovery 36461640879aSAndrew Shewmaker * tools. So, the admin may have changed them. 36471640879aSAndrew Shewmaker * 36481640879aSAndrew Shewmaker * If memory is added and the reserves have been eliminated or increased above 36491640879aSAndrew Shewmaker * the default max, then we'll trust the admin. 36501640879aSAndrew Shewmaker * 36511640879aSAndrew Shewmaker * If memory is removed and there isn't enough free memory, then we 36521640879aSAndrew Shewmaker * need to reset the reserves. 36531640879aSAndrew Shewmaker * 36541640879aSAndrew Shewmaker * Otherwise keep the reserve set by the admin. 36551640879aSAndrew Shewmaker */ 36561640879aSAndrew Shewmaker static int reserve_mem_notifier(struct notifier_block *nb, 36571640879aSAndrew Shewmaker unsigned long action, void *data) 36581640879aSAndrew Shewmaker { 36591640879aSAndrew Shewmaker unsigned long tmp, free_kbytes; 36601640879aSAndrew Shewmaker 36611640879aSAndrew Shewmaker switch (action) { 36621640879aSAndrew Shewmaker case MEM_ONLINE: 36631640879aSAndrew Shewmaker /* Default max is 128MB. Leave alone if modified by operator. */ 36641640879aSAndrew Shewmaker tmp = sysctl_user_reserve_kbytes; 36651640879aSAndrew Shewmaker if (0 < tmp && tmp < (1UL << 17)) 36661640879aSAndrew Shewmaker init_user_reserve(); 36671640879aSAndrew Shewmaker 36681640879aSAndrew Shewmaker /* Default max is 8MB. Leave alone if modified by operator. */ 36691640879aSAndrew Shewmaker tmp = sysctl_admin_reserve_kbytes; 36701640879aSAndrew Shewmaker if (0 < tmp && tmp < (1UL << 13)) 36711640879aSAndrew Shewmaker init_admin_reserve(); 36721640879aSAndrew Shewmaker 36731640879aSAndrew Shewmaker break; 36741640879aSAndrew Shewmaker case MEM_OFFLINE: 3675c41f012aSMichal Hocko free_kbytes = global_zone_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10); 36761640879aSAndrew Shewmaker 36771640879aSAndrew Shewmaker if (sysctl_user_reserve_kbytes > free_kbytes) { 36781640879aSAndrew Shewmaker init_user_reserve(); 36791640879aSAndrew Shewmaker pr_info("vm.user_reserve_kbytes reset to %lu\n", 36801640879aSAndrew Shewmaker sysctl_user_reserve_kbytes); 36811640879aSAndrew Shewmaker } 36821640879aSAndrew Shewmaker 36831640879aSAndrew Shewmaker if (sysctl_admin_reserve_kbytes > free_kbytes) { 36841640879aSAndrew Shewmaker init_admin_reserve(); 36851640879aSAndrew Shewmaker pr_info("vm.admin_reserve_kbytes reset to %lu\n", 36861640879aSAndrew Shewmaker sysctl_admin_reserve_kbytes); 36871640879aSAndrew Shewmaker } 36881640879aSAndrew Shewmaker break; 36891640879aSAndrew Shewmaker default: 36901640879aSAndrew Shewmaker break; 36911640879aSAndrew Shewmaker } 36921640879aSAndrew Shewmaker return NOTIFY_OK; 36931640879aSAndrew Shewmaker } 36941640879aSAndrew Shewmaker 36951640879aSAndrew Shewmaker static int __meminit init_reserve_notifier(void) 36961640879aSAndrew Shewmaker { 36971eeaa4fdSLiu Shixin if (hotplug_memory_notifier(reserve_mem_notifier, DEFAULT_CALLBACK_PRI)) 3698b1de0d13SMitchel Humpherys pr_err("Failed registering memory add/remove notifier for admin reserve\n"); 36991640879aSAndrew Shewmaker 37001640879aSAndrew Shewmaker return 0; 37011640879aSAndrew Shewmaker } 3702a64fb3cdSPaul Gortmaker subsys_initcall(init_reserve_notifier); 3703