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, 8168f48381SSuren Baghdasaryan unsigned long end, bool mm_wr_locked); 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), 6085ff783f1SVlastimil Babka * we are removing both mid and next vmas 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 */ 747*2dbf4010SVlastimil Babka static inline bool 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) 761*2dbf4010SVlastimil Babka return false; 7621da177e4SLinus Torvalds if (vma->vm_file != file) 763*2dbf4010SVlastimil Babka return false; 7641da177e4SLinus Torvalds if (vma->vm_ops && vma->vm_ops->close) 765*2dbf4010SVlastimil Babka return false; 76619a809afSAndrea Arcangeli if (!is_mergeable_vm_userfaultfd_ctx(vma, vm_userfaultfd_ctx)) 767*2dbf4010SVlastimil Babka return false; 7685c26f6acSSuren Baghdasaryan if (!anon_vma_name_eq(anon_vma_name(vma), anon_name)) 769*2dbf4010SVlastimil Babka return false; 770*2dbf4010SVlastimil Babka return true; 7711da177e4SLinus Torvalds } 7721da177e4SLinus Torvalds 773*2dbf4010SVlastimil Babka static inline bool is_mergeable_anon_vma(struct anon_vma *anon_vma1, 774*2dbf4010SVlastimil Babka struct anon_vma *anon_vma2, struct vm_area_struct *vma) 7751da177e4SLinus Torvalds { 776965f55deSShaohua Li /* 777965f55deSShaohua Li * The list_is_singular() test is to avoid merging VMA cloned from 778965f55deSShaohua Li * parents. This can improve scalability caused by anon_vma lock. 779965f55deSShaohua Li */ 780965f55deSShaohua Li if ((!anon_vma1 || !anon_vma2) && (!vma || 781965f55deSShaohua Li list_is_singular(&vma->anon_vma_chain))) 782*2dbf4010SVlastimil Babka return true; 783965f55deSShaohua Li return anon_vma1 == anon_vma2; 7841da177e4SLinus Torvalds } 7851da177e4SLinus Torvalds 7861da177e4SLinus Torvalds /* 7871da177e4SLinus Torvalds * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff) 7881da177e4SLinus Torvalds * in front of (at a lower virtual address and file offset than) the vma. 7891da177e4SLinus Torvalds * 7901da177e4SLinus Torvalds * We cannot merge two vmas if they have differently assigned (non-NULL) 7911da177e4SLinus Torvalds * anon_vmas, nor if same anon_vma is assigned but offsets incompatible. 7921da177e4SLinus Torvalds * 7931da177e4SLinus Torvalds * We don't check here for the merged mmap wrapping around the end of pagecache 79445e55300SPeter Collingbourne * indices (16TB on ia32) because do_mmap() does not permit mmap's which 7951da177e4SLinus Torvalds * wrap, nor mmaps which cover the final page at index -1UL. 7961da177e4SLinus Torvalds */ 797*2dbf4010SVlastimil Babka static bool 7981da177e4SLinus Torvalds can_vma_merge_before(struct vm_area_struct *vma, unsigned long vm_flags, 79919a809afSAndrea Arcangeli struct anon_vma *anon_vma, struct file *file, 800*2dbf4010SVlastimil Babka pgoff_t vm_pgoff, struct vm_userfaultfd_ctx vm_userfaultfd_ctx, 8015c26f6acSSuren Baghdasaryan struct anon_vma_name *anon_name) 8021da177e4SLinus Torvalds { 8039a10064fSColin Cross if (is_mergeable_vma(vma, file, vm_flags, vm_userfaultfd_ctx, anon_name) && 804965f55deSShaohua Li is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) { 8051da177e4SLinus Torvalds if (vma->vm_pgoff == vm_pgoff) 806*2dbf4010SVlastimil Babka return true; 8071da177e4SLinus Torvalds } 808*2dbf4010SVlastimil Babka return false; 8091da177e4SLinus Torvalds } 8101da177e4SLinus Torvalds 8111da177e4SLinus Torvalds /* 8121da177e4SLinus Torvalds * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff) 8131da177e4SLinus Torvalds * beyond (at a higher virtual address and file offset than) the vma. 8141da177e4SLinus Torvalds * 8151da177e4SLinus Torvalds * We cannot merge two vmas if they have differently assigned (non-NULL) 8161da177e4SLinus Torvalds * anon_vmas, nor if same anon_vma is assigned but offsets incompatible. 8171da177e4SLinus Torvalds */ 818*2dbf4010SVlastimil Babka static bool 8191da177e4SLinus Torvalds can_vma_merge_after(struct vm_area_struct *vma, unsigned long vm_flags, 82019a809afSAndrea Arcangeli struct anon_vma *anon_vma, struct file *file, 821*2dbf4010SVlastimil Babka pgoff_t vm_pgoff, struct vm_userfaultfd_ctx vm_userfaultfd_ctx, 8225c26f6acSSuren Baghdasaryan struct anon_vma_name *anon_name) 8231da177e4SLinus Torvalds { 8249a10064fSColin Cross if (is_mergeable_vma(vma, file, vm_flags, vm_userfaultfd_ctx, anon_name) && 825965f55deSShaohua Li is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) { 8261da177e4SLinus Torvalds pgoff_t vm_pglen; 827d6e93217SLibin vm_pglen = vma_pages(vma); 8281da177e4SLinus Torvalds if (vma->vm_pgoff + vm_pglen == vm_pgoff) 829*2dbf4010SVlastimil Babka return true; 8301da177e4SLinus Torvalds } 831*2dbf4010SVlastimil Babka return false; 8321da177e4SLinus Torvalds } 8331da177e4SLinus Torvalds 8341da177e4SLinus Torvalds /* 8359a10064fSColin Cross * Given a mapping request (addr,end,vm_flags,file,pgoff,anon_name), 8369a10064fSColin Cross * figure out whether that can be merged with its predecessor or its 8379a10064fSColin Cross * successor. Or both (it neatly fills a hole). 8381da177e4SLinus Torvalds * 8391da177e4SLinus Torvalds * In most cases - when called for mmap, brk or mremap - [addr,end) is 8401da177e4SLinus Torvalds * certain not to be mapped by the time vma_merge is called; but when 8411da177e4SLinus Torvalds * called for mprotect, it is certain to be already mapped (either at 8421da177e4SLinus Torvalds * an offset within prev, or at the start of next), and the flags of 8431da177e4SLinus Torvalds * this area are about to be changed to vm_flags - and the no-change 8441da177e4SLinus Torvalds * case has already been eliminated. 8451da177e4SLinus Torvalds * 8461da177e4SLinus Torvalds * The following mprotect cases have to be considered, where AAAA is 8471da177e4SLinus Torvalds * the area passed down from mprotect_fixup, never extending beyond one 8489e8a39d2SVlastimil Babka * vma, PPPP is the previous vma, NNNN is a vma that starts at the same 8499e8a39d2SVlastimil Babka * address as AAAA and is of the same or larger span, and XXXX the next 8509e8a39d2SVlastimil Babka * vma after AAAA: 8511da177e4SLinus Torvalds * 8525d42ab29SWei Yang * AAAA AAAA AAAA 8539e8a39d2SVlastimil Babka * PPPPPPXXXXXX PPPPPPXXXXXX PPPPPPNNNNNN 8545d42ab29SWei Yang * cannot merge might become might become 855183b7a60SVlastimil Babka * PPXXXXXXXXXX PPPPPPPPPPNN 8565d42ab29SWei Yang * mmap, brk or case 4 below case 5 below 8575d42ab29SWei Yang * mremap move: 8585d42ab29SWei Yang * AAAA AAAA 859097d70c6SVlastimil Babka * PPPP XXXX PPPPNNNNXXXX 8605d42ab29SWei Yang * might become might become 8615d42ab29SWei Yang * PPPPPPPPPPPP 1 or PPPPPPPPPPPP 6 or 862097d70c6SVlastimil Babka * PPPPPPPPXXXX 2 or PPPPPPPPXXXX 7 or 863097d70c6SVlastimil Babka * PPPPXXXXXXXX 3 PPPPXXXXXXXX 8 8641da177e4SLinus Torvalds * 8658bb4e7a2SWei Yang * It is important for case 8 that the vma NNNN overlapping the 866e86f15eeSAndrea Arcangeli * region AAAA is never going to extended over XXXX. Instead XXXX must 867e86f15eeSAndrea Arcangeli * be extended in region AAAA and NNNN must be removed. This way in 8680503ea8fSLiam R. Howlett * all cases where vma_merge succeeds, the moment vma_merge drops the 869e86f15eeSAndrea Arcangeli * rmap_locks, the properties of the merged vma will be already 870e86f15eeSAndrea Arcangeli * correct for the whole merged range. Some of those properties like 871e86f15eeSAndrea Arcangeli * vm_page_prot/vm_flags may be accessed by rmap_walks and they must 872e86f15eeSAndrea Arcangeli * be correct for the whole merged range immediately after the 873e86f15eeSAndrea Arcangeli * rmap_locks are released. Otherwise if XXXX would be removed and 874e86f15eeSAndrea Arcangeli * NNNN would be extended over the XXXX range, remove_migration_ptes 875e86f15eeSAndrea Arcangeli * or other rmap walkers (if working on addresses beyond the "end" 876e86f15eeSAndrea Arcangeli * parameter) may establish ptes with the wrong permissions of NNNN 877e86f15eeSAndrea Arcangeli * instead of the right permissions of XXXX. 8780503ea8fSLiam R. Howlett * 8790503ea8fSLiam R. Howlett * In the code below: 8800503ea8fSLiam R. Howlett * PPPP is represented by *prev 8819e8a39d2SVlastimil Babka * NNNN is represented by *mid or not represented at all (NULL) 8829e8a39d2SVlastimil Babka * XXXX is represented by *next or not represented at all (NULL) 8839e8a39d2SVlastimil Babka * AAAA is not represented - it will be merged and the vma containing the 8849e8a39d2SVlastimil Babka * area is returned, 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; 9041e76454fSVlastimil Babka long adj_start = 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 9155cd70b96SVlastimil Babka mid = find_vma(mm, prev ? prev->vm_end : 0); 9165cd70b96SVlastimil Babka if (mid && mid->vm_end == end) /* cases 6, 7, 8 */ 9175cd70b96SVlastimil Babka next = find_vma(mm, mid->vm_end); 9185cd70b96SVlastimil Babka else 9195cd70b96SVlastimil Babka next = mid; 9201da177e4SLinus Torvalds 9219e8a39d2SVlastimil Babka /* In cases 1 - 4 there's no NNNN vma */ 9229e8a39d2SVlastimil Babka if (mid && end <= mid->vm_start) 9239e8a39d2SVlastimil Babka mid = NULL; 9249e8a39d2SVlastimil Babka 925e86f15eeSAndrea Arcangeli /* verify some invariant that must be enforced by the caller */ 926e86f15eeSAndrea Arcangeli VM_WARN_ON(prev && addr <= prev->vm_start); 927eef19944SJakub Matěna VM_WARN_ON(mid && end > mid->vm_end); 928e86f15eeSAndrea Arcangeli VM_WARN_ON(addr >= end); 929e86f15eeSAndrea Arcangeli 9300503ea8fSLiam R. Howlett if (prev) { 9310503ea8fSLiam R. Howlett res = prev; 9320503ea8fSLiam R. Howlett vma = prev; 9330503ea8fSLiam R. Howlett vma_start = prev->vm_start; 9340503ea8fSLiam R. Howlett vma_pgoff = prev->vm_pgoff; 935eef19944SJakub Matěna /* Can we merge the predecessor? */ 9360503ea8fSLiam R. Howlett if (prev->vm_end == addr && mpol_equal(vma_policy(prev), policy) 9370503ea8fSLiam R. Howlett && can_vma_merge_after(prev, vm_flags, anon_vma, file, 9380503ea8fSLiam R. Howlett pgoff, vm_userfaultfd_ctx, anon_name)) { 939eef19944SJakub Matěna merge_prev = true; 94018b098afSLiam R. Howlett vma_prev(vmi); 9411da177e4SLinus Torvalds } 9420503ea8fSLiam R. Howlett } 943eef19944SJakub Matěna /* Can we merge the successor? */ 9441da177e4SLinus Torvalds if (next && end == next->vm_start && 9451da177e4SLinus Torvalds mpol_equal(policy, vma_policy(next)) && 9461da177e4SLinus Torvalds can_vma_merge_before(next, vm_flags, 94719a809afSAndrea Arcangeli anon_vma, file, pgoff+pglen, 9489a10064fSColin Cross vm_userfaultfd_ctx, anon_name)) { 949eef19944SJakub Matěna merge_next = true; 950eef19944SJakub Matěna } 9510503ea8fSLiam R. Howlett 9520503ea8fSLiam R. Howlett remove = remove2 = adjust = NULL; 953eef19944SJakub Matěna /* Can we merge both the predecessor and the successor? */ 954eef19944SJakub Matěna if (merge_prev && merge_next && 9550503ea8fSLiam R. Howlett is_mergeable_anon_vma(prev->anon_vma, next->anon_vma, NULL)) { 9565ff783f1SVlastimil Babka remove = next; /* case 1 */ 9570503ea8fSLiam R. Howlett vma_end = next->vm_end; 95850dac011SVlastimil Babka err = dup_anon_vma(prev, next); 9599e8a39d2SVlastimil Babka if (mid) { /* case 6 */ 9605ff783f1SVlastimil Babka remove = mid; 9615ff783f1SVlastimil Babka remove2 = next; 9625ff783f1SVlastimil Babka if (!next->anon_vma) 9635ff783f1SVlastimil Babka err = dup_anon_vma(prev, mid); 9640503ea8fSLiam R. Howlett } 9650503ea8fSLiam R. Howlett } else if (merge_prev) { 9660503ea8fSLiam R. Howlett err = 0; /* case 2 */ 9679e8a39d2SVlastimil Babka if (mid) { 96850dac011SVlastimil Babka err = dup_anon_vma(prev, mid); 9690503ea8fSLiam R. Howlett if (end == mid->vm_end) { /* case 7 */ 9700503ea8fSLiam R. Howlett remove = mid; 9710503ea8fSLiam R. Howlett } else { /* case 5 */ 9720503ea8fSLiam R. Howlett adjust = mid; 9731e76454fSVlastimil Babka adj_start = (end - mid->vm_start); 9740503ea8fSLiam R. Howlett } 9750503ea8fSLiam R. Howlett } 976eef19944SJakub Matěna } else if (merge_next) { 977eef19944SJakub Matěna res = next; 9780503ea8fSLiam R. Howlett if (prev && addr < prev->vm_end) { /* case 4 */ 9790503ea8fSLiam R. Howlett vma_end = addr; 980183b7a60SVlastimil Babka adjust = next; 9811e76454fSVlastimil Babka adj_start = -(prev->vm_end - addr); 982183b7a60SVlastimil Babka err = dup_anon_vma(next, prev); 9830503ea8fSLiam R. Howlett } else { 9840503ea8fSLiam R. Howlett vma = next; /* case 3 */ 9850503ea8fSLiam R. Howlett vma_start = addr; 9860503ea8fSLiam R. Howlett vma_end = next->vm_end; 987097d70c6SVlastimil Babka vma_pgoff = next->vm_pgoff; 9880503ea8fSLiam R. Howlett err = 0; 9899e8a39d2SVlastimil Babka if (mid) { /* case 8 */ 990097d70c6SVlastimil Babka vma_pgoff = mid->vm_pgoff; 9910503ea8fSLiam R. Howlett remove = mid; 99250dac011SVlastimil Babka err = dup_anon_vma(next, mid); 9930503ea8fSLiam R. Howlett } 9940503ea8fSLiam R. Howlett } 9951da177e4SLinus Torvalds } 9961da177e4SLinus Torvalds 9970503ea8fSLiam R. Howlett /* Cannot merge or error in anon_vma clone */ 998eef19944SJakub Matěna if (err) 9991da177e4SLinus Torvalds return NULL; 10000503ea8fSLiam R. Howlett 10010503ea8fSLiam R. Howlett if (vma_iter_prealloc(vmi)) 10020503ea8fSLiam R. Howlett return NULL; 10030503ea8fSLiam R. Howlett 10041e76454fSVlastimil Babka vma_adjust_trans_huge(vma, vma_start, vma_end, adj_start); 10050503ea8fSLiam R. Howlett init_multi_vma_prep(&vp, vma, adjust, remove, remove2); 10060503ea8fSLiam R. Howlett VM_WARN_ON(vp.anon_vma && adjust && adjust->anon_vma && 10070503ea8fSLiam R. Howlett vp.anon_vma != adjust->anon_vma); 10080503ea8fSLiam R. Howlett 10090503ea8fSLiam R. Howlett vma_prepare(&vp); 10100503ea8fSLiam R. Howlett if (vma_start < vma->vm_start || vma_end > vma->vm_end) 10110503ea8fSLiam R. Howlett vma_expanded = true; 10120503ea8fSLiam R. Howlett 10130503ea8fSLiam R. Howlett vma->vm_start = vma_start; 10140503ea8fSLiam R. Howlett vma->vm_end = vma_end; 10150503ea8fSLiam R. Howlett vma->vm_pgoff = vma_pgoff; 10160503ea8fSLiam R. Howlett 10170503ea8fSLiam R. Howlett if (vma_expanded) 10180503ea8fSLiam R. Howlett vma_iter_store(vmi, vma); 10190503ea8fSLiam R. Howlett 10201e76454fSVlastimil Babka if (adj_start) { 10211e76454fSVlastimil Babka adjust->vm_start += adj_start; 10221e76454fSVlastimil Babka adjust->vm_pgoff += adj_start >> PAGE_SHIFT; 10231e76454fSVlastimil Babka if (adj_start < 0) { 10240503ea8fSLiam R. Howlett WARN_ON(vma_expanded); 10250503ea8fSLiam R. Howlett vma_iter_store(vmi, next); 10260503ea8fSLiam R. Howlett } 10270503ea8fSLiam R. Howlett } 10280503ea8fSLiam R. Howlett 10290503ea8fSLiam R. Howlett vma_complete(&vp, vmi, mm); 10300503ea8fSLiam R. Howlett vma_iter_free(vmi); 10310503ea8fSLiam R. Howlett validate_mm(mm); 1032eef19944SJakub Matěna khugepaged_enter_vma(res, vm_flags); 10331da177e4SLinus Torvalds 10349760ebffSLiam R. Howlett return res; 1035f2ebfe43SLiam R. Howlett } 1036f2ebfe43SLiam R. Howlett 10371da177e4SLinus Torvalds /* 1038b4f315b4SEthon Paul * Rough compatibility check to quickly see if it's even worth looking 1039d0e9fe17SLinus Torvalds * at sharing an anon_vma. 1040d0e9fe17SLinus Torvalds * 1041d0e9fe17SLinus Torvalds * They need to have the same vm_file, and the flags can only differ 1042d0e9fe17SLinus Torvalds * in things that mprotect may change. 1043d0e9fe17SLinus Torvalds * 1044d0e9fe17SLinus Torvalds * NOTE! The fact that we share an anon_vma doesn't _have_ to mean that 1045d0e9fe17SLinus Torvalds * we can merge the two vma's. For example, we refuse to merge a vma if 1046d0e9fe17SLinus Torvalds * there is a vm_ops->close() function, because that indicates that the 1047d0e9fe17SLinus Torvalds * driver is doing some kind of reference counting. But that doesn't 1048d0e9fe17SLinus Torvalds * really matter for the anon_vma sharing case. 1049d0e9fe17SLinus Torvalds */ 1050d0e9fe17SLinus Torvalds static int anon_vma_compatible(struct vm_area_struct *a, struct vm_area_struct *b) 1051d0e9fe17SLinus Torvalds { 1052d0e9fe17SLinus Torvalds return a->vm_end == b->vm_start && 1053d0e9fe17SLinus Torvalds mpol_equal(vma_policy(a), vma_policy(b)) && 1054d0e9fe17SLinus Torvalds a->vm_file == b->vm_file && 10556cb4d9a2SAnshuman Khandual !((a->vm_flags ^ b->vm_flags) & ~(VM_ACCESS_FLAGS | VM_SOFTDIRTY)) && 1056d0e9fe17SLinus Torvalds b->vm_pgoff == a->vm_pgoff + ((b->vm_start - a->vm_start) >> PAGE_SHIFT); 1057d0e9fe17SLinus Torvalds } 1058d0e9fe17SLinus Torvalds 1059d0e9fe17SLinus Torvalds /* 1060d0e9fe17SLinus Torvalds * Do some basic sanity checking to see if we can re-use the anon_vma 1061d0e9fe17SLinus Torvalds * from 'old'. The 'a'/'b' vma's are in VM order - one of them will be 1062d0e9fe17SLinus Torvalds * the same as 'old', the other will be the new one that is trying 1063d0e9fe17SLinus Torvalds * to share the anon_vma. 1064d0e9fe17SLinus Torvalds * 10655b449489SFlorian Rommel * NOTE! This runs with mmap_lock held for reading, so it is possible that 1066d0e9fe17SLinus Torvalds * the anon_vma of 'old' is concurrently in the process of being set up 1067d0e9fe17SLinus Torvalds * by another page fault trying to merge _that_. But that's ok: if it 1068d0e9fe17SLinus Torvalds * is being set up, that automatically means that it will be a singleton 1069d0e9fe17SLinus Torvalds * acceptable for merging, so we can do all of this optimistically. But 10704db0c3c2SJason Low * we do that READ_ONCE() to make sure that we never re-load the pointer. 1071d0e9fe17SLinus Torvalds * 1072d0e9fe17SLinus Torvalds * IOW: that the "list_is_singular()" test on the anon_vma_chain only 1073d0e9fe17SLinus Torvalds * matters for the 'stable anon_vma' case (ie the thing we want to avoid 1074d0e9fe17SLinus Torvalds * is to return an anon_vma that is "complex" due to having gone through 1075d0e9fe17SLinus Torvalds * a fork). 1076d0e9fe17SLinus Torvalds * 1077d0e9fe17SLinus Torvalds * We also make sure that the two vma's are compatible (adjacent, 1078d0e9fe17SLinus Torvalds * and with the same memory policies). That's all stable, even with just 10795b449489SFlorian Rommel * a read lock on the mmap_lock. 1080d0e9fe17SLinus Torvalds */ 1081d0e9fe17SLinus Torvalds static struct anon_vma *reusable_anon_vma(struct vm_area_struct *old, struct vm_area_struct *a, struct vm_area_struct *b) 1082d0e9fe17SLinus Torvalds { 1083d0e9fe17SLinus Torvalds if (anon_vma_compatible(a, b)) { 10844db0c3c2SJason Low struct anon_vma *anon_vma = READ_ONCE(old->anon_vma); 1085d0e9fe17SLinus Torvalds 1086d0e9fe17SLinus Torvalds if (anon_vma && list_is_singular(&old->anon_vma_chain)) 1087d0e9fe17SLinus Torvalds return anon_vma; 1088d0e9fe17SLinus Torvalds } 1089d0e9fe17SLinus Torvalds return NULL; 1090d0e9fe17SLinus Torvalds } 1091d0e9fe17SLinus Torvalds 1092d0e9fe17SLinus Torvalds /* 10931da177e4SLinus Torvalds * find_mergeable_anon_vma is used by anon_vma_prepare, to check 10941da177e4SLinus Torvalds * neighbouring vmas for a suitable anon_vma, before it goes off 10951da177e4SLinus Torvalds * to allocate a new anon_vma. It checks because a repetitive 10961da177e4SLinus Torvalds * sequence of mprotects and faults may otherwise lead to distinct 10971da177e4SLinus Torvalds * anon_vmas being allocated, preventing vma merge in subsequent 10981da177e4SLinus Torvalds * mprotect. 10991da177e4SLinus Torvalds */ 11001da177e4SLinus Torvalds struct anon_vma *find_mergeable_anon_vma(struct vm_area_struct *vma) 11011da177e4SLinus Torvalds { 1102763ecb03SLiam R. Howlett MA_STATE(mas, &vma->vm_mm->mm_mt, vma->vm_end, vma->vm_end); 1103a67c8caaSMiaohe Lin struct anon_vma *anon_vma = NULL; 1104763ecb03SLiam R. Howlett struct vm_area_struct *prev, *next; 11051da177e4SLinus Torvalds 1106a67c8caaSMiaohe Lin /* Try next first. */ 1107763ecb03SLiam R. Howlett next = mas_walk(&mas); 1108763ecb03SLiam R. Howlett if (next) { 1109763ecb03SLiam R. Howlett anon_vma = reusable_anon_vma(next, vma, next); 1110d0e9fe17SLinus Torvalds if (anon_vma) 1111d0e9fe17SLinus Torvalds return anon_vma; 1112a67c8caaSMiaohe Lin } 11131da177e4SLinus Torvalds 1114763ecb03SLiam R. Howlett prev = mas_prev(&mas, 0); 1115763ecb03SLiam R. Howlett VM_BUG_ON_VMA(prev != vma, vma); 1116763ecb03SLiam R. Howlett prev = mas_prev(&mas, 0); 1117a67c8caaSMiaohe Lin /* Try prev next. */ 1118763ecb03SLiam R. Howlett if (prev) 1119763ecb03SLiam R. Howlett anon_vma = reusable_anon_vma(prev, prev, vma); 1120a67c8caaSMiaohe Lin 11211da177e4SLinus Torvalds /* 1122a67c8caaSMiaohe Lin * We might reach here with anon_vma == NULL if we can't find 1123a67c8caaSMiaohe Lin * any reusable anon_vma. 11241da177e4SLinus Torvalds * There's no absolute need to look only at touching neighbours: 11251da177e4SLinus Torvalds * we could search further afield for "compatible" anon_vmas. 11261da177e4SLinus Torvalds * But it would probably just be a waste of time searching, 11271da177e4SLinus Torvalds * or lead to too many vmas hanging off the same anon_vma. 11281da177e4SLinus Torvalds * We're trying to allow mprotect remerging later on, 11291da177e4SLinus Torvalds * not trying to minimize memory used for anon_vmas. 11301da177e4SLinus Torvalds */ 1131a67c8caaSMiaohe Lin return anon_vma; 11321da177e4SLinus Torvalds } 11331da177e4SLinus Torvalds 11341da177e4SLinus Torvalds /* 113540401530SAl Viro * If a hint addr is less than mmap_min_addr change hint to be as 113640401530SAl Viro * low as possible but still greater than mmap_min_addr 113740401530SAl Viro */ 113840401530SAl Viro static inline unsigned long round_hint_to_min(unsigned long hint) 113940401530SAl Viro { 114040401530SAl Viro hint &= PAGE_MASK; 114140401530SAl Viro if (((void *)hint != NULL) && 114240401530SAl Viro (hint < mmap_min_addr)) 114340401530SAl Viro return PAGE_ALIGN(mmap_min_addr); 114440401530SAl Viro return hint; 114540401530SAl Viro } 114640401530SAl Viro 11476aeb2542SMike Rapoport int mlock_future_check(struct mm_struct *mm, unsigned long flags, 1148363ee17fSDavidlohr Bueso unsigned long len) 1149363ee17fSDavidlohr Bueso { 1150363ee17fSDavidlohr Bueso unsigned long locked, lock_limit; 1151363ee17fSDavidlohr Bueso 1152363ee17fSDavidlohr Bueso /* mlock MCL_FUTURE? */ 1153363ee17fSDavidlohr Bueso if (flags & VM_LOCKED) { 1154363ee17fSDavidlohr Bueso locked = len >> PAGE_SHIFT; 1155363ee17fSDavidlohr Bueso locked += mm->locked_vm; 1156363ee17fSDavidlohr Bueso lock_limit = rlimit(RLIMIT_MEMLOCK); 1157363ee17fSDavidlohr Bueso lock_limit >>= PAGE_SHIFT; 1158363ee17fSDavidlohr Bueso if (locked > lock_limit && !capable(CAP_IPC_LOCK)) 1159363ee17fSDavidlohr Bueso return -EAGAIN; 1160363ee17fSDavidlohr Bueso } 1161363ee17fSDavidlohr Bueso return 0; 1162363ee17fSDavidlohr Bueso } 1163363ee17fSDavidlohr Bueso 1164be83bbf8SLinus Torvalds static inline u64 file_mmap_size_max(struct file *file, struct inode *inode) 1165be83bbf8SLinus Torvalds { 1166be83bbf8SLinus Torvalds if (S_ISREG(inode->i_mode)) 1167423913adSLinus Torvalds return MAX_LFS_FILESIZE; 1168be83bbf8SLinus Torvalds 1169be83bbf8SLinus Torvalds if (S_ISBLK(inode->i_mode)) 1170be83bbf8SLinus Torvalds return MAX_LFS_FILESIZE; 1171be83bbf8SLinus Torvalds 117276f34950SIvan Khoronzhuk if (S_ISSOCK(inode->i_mode)) 117376f34950SIvan Khoronzhuk return MAX_LFS_FILESIZE; 117476f34950SIvan Khoronzhuk 1175be83bbf8SLinus Torvalds /* Special "we do even unsigned file positions" case */ 1176be83bbf8SLinus Torvalds if (file->f_mode & FMODE_UNSIGNED_OFFSET) 1177be83bbf8SLinus Torvalds return 0; 1178be83bbf8SLinus Torvalds 1179be83bbf8SLinus Torvalds /* Yes, random drivers might want more. But I'm tired of buggy drivers */ 1180be83bbf8SLinus Torvalds return ULONG_MAX; 1181be83bbf8SLinus Torvalds } 1182be83bbf8SLinus Torvalds 1183be83bbf8SLinus Torvalds static inline bool file_mmap_ok(struct file *file, struct inode *inode, 1184be83bbf8SLinus Torvalds unsigned long pgoff, unsigned long len) 1185be83bbf8SLinus Torvalds { 1186be83bbf8SLinus Torvalds u64 maxsize = file_mmap_size_max(file, inode); 1187be83bbf8SLinus Torvalds 1188be83bbf8SLinus Torvalds if (maxsize && len > maxsize) 1189be83bbf8SLinus Torvalds return false; 1190be83bbf8SLinus Torvalds maxsize -= len; 1191be83bbf8SLinus Torvalds if (pgoff > maxsize >> PAGE_SHIFT) 1192be83bbf8SLinus Torvalds return false; 1193be83bbf8SLinus Torvalds return true; 1194be83bbf8SLinus Torvalds } 1195be83bbf8SLinus Torvalds 119640401530SAl Viro /* 11973e4e28c5SMichel Lespinasse * The caller must write-lock current->mm->mmap_lock. 11981da177e4SLinus Torvalds */ 11991fcfd8dbSOleg Nesterov unsigned long do_mmap(struct file *file, unsigned long addr, 12001da177e4SLinus Torvalds unsigned long len, unsigned long prot, 120145e55300SPeter Collingbourne unsigned long flags, unsigned long pgoff, 120245e55300SPeter Collingbourne unsigned long *populate, struct list_head *uf) 12031da177e4SLinus Torvalds { 12041da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 120545e55300SPeter Collingbourne vm_flags_t vm_flags; 120662b5f7d0SDave Hansen int pkey = 0; 12071da177e4SLinus Torvalds 1208524e00b3SLiam R. Howlett validate_mm(mm); 120941badc15SMichel Lespinasse *populate = 0; 1210bebeb3d6SMichel Lespinasse 1211e37609bbSPiotr Kwapulinski if (!len) 1212e37609bbSPiotr Kwapulinski return -EINVAL; 1213e37609bbSPiotr Kwapulinski 12141da177e4SLinus Torvalds /* 12151da177e4SLinus Torvalds * Does the application expect PROT_READ to imply PROT_EXEC? 12161da177e4SLinus Torvalds * 12171da177e4SLinus Torvalds * (the exception is when the underlying filesystem is noexec 12181da177e4SLinus Torvalds * mounted, in which case we dont add PROT_EXEC.) 12191da177e4SLinus Torvalds */ 12201da177e4SLinus Torvalds if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC)) 122190f8572bSEric W. Biederman if (!(file && path_noexec(&file->f_path))) 12221da177e4SLinus Torvalds prot |= PROT_EXEC; 12231da177e4SLinus Torvalds 1224a4ff8e86SMichal Hocko /* force arch specific MAP_FIXED handling in get_unmapped_area */ 1225a4ff8e86SMichal Hocko if (flags & MAP_FIXED_NOREPLACE) 1226a4ff8e86SMichal Hocko flags |= MAP_FIXED; 1227a4ff8e86SMichal Hocko 12287cd94146SEric Paris if (!(flags & MAP_FIXED)) 12297cd94146SEric Paris addr = round_hint_to_min(addr); 12307cd94146SEric Paris 12311da177e4SLinus Torvalds /* Careful about overflows.. */ 12321da177e4SLinus Torvalds len = PAGE_ALIGN(len); 12339206de95SAl Viro if (!len) 12341da177e4SLinus Torvalds return -ENOMEM; 12351da177e4SLinus Torvalds 12361da177e4SLinus Torvalds /* offset overflow? */ 12371da177e4SLinus Torvalds if ((pgoff + (len >> PAGE_SHIFT)) < pgoff) 12381da177e4SLinus Torvalds return -EOVERFLOW; 12391da177e4SLinus Torvalds 12401da177e4SLinus Torvalds /* Too many mappings? */ 12411da177e4SLinus Torvalds if (mm->map_count > sysctl_max_map_count) 12421da177e4SLinus Torvalds return -ENOMEM; 12431da177e4SLinus Torvalds 12441da177e4SLinus Torvalds /* Obtain the address to map to. we verify (or select) it and ensure 12451da177e4SLinus Torvalds * that it represents a valid section of the address space. 12461da177e4SLinus Torvalds */ 12471da177e4SLinus Torvalds addr = get_unmapped_area(file, addr, len, pgoff, flags); 1248ff68dac6SGaowei Pu if (IS_ERR_VALUE(addr)) 12491da177e4SLinus Torvalds return addr; 12501da177e4SLinus Torvalds 1251a4ff8e86SMichal Hocko if (flags & MAP_FIXED_NOREPLACE) { 125235e43c5fSLiam Howlett if (find_vma_intersection(mm, addr, addr + len)) 1253a4ff8e86SMichal Hocko return -EEXIST; 1254a4ff8e86SMichal Hocko } 1255a4ff8e86SMichal Hocko 125662b5f7d0SDave Hansen if (prot == PROT_EXEC) { 125762b5f7d0SDave Hansen pkey = execute_only_pkey(mm); 125862b5f7d0SDave Hansen if (pkey < 0) 125962b5f7d0SDave Hansen pkey = 0; 126062b5f7d0SDave Hansen } 126162b5f7d0SDave Hansen 12621da177e4SLinus Torvalds /* Do simple checking here so the lower-level routines won't have 12631da177e4SLinus Torvalds * to. we assume access permissions have been handled by the open 12641da177e4SLinus Torvalds * of the memory object, so we don't do any here. 12651da177e4SLinus Torvalds */ 126645e55300SPeter Collingbourne vm_flags = calc_vm_prot_bits(prot, pkey) | calc_vm_flag_bits(flags) | 12671da177e4SLinus Torvalds mm->def_flags | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC; 12681da177e4SLinus Torvalds 1269cdf7b341SHuang Shijie if (flags & MAP_LOCKED) 12701da177e4SLinus Torvalds if (!can_do_mlock()) 12711da177e4SLinus Torvalds return -EPERM; 1272ba470de4SRik van Riel 1273363ee17fSDavidlohr Bueso if (mlock_future_check(mm, vm_flags, len)) 12741da177e4SLinus Torvalds return -EAGAIN; 12751da177e4SLinus Torvalds 12761da177e4SLinus Torvalds if (file) { 1277077bf22bSOleg Nesterov struct inode *inode = file_inode(file); 12781c972597SDan Williams unsigned long flags_mask; 12791c972597SDan Williams 1280be83bbf8SLinus Torvalds if (!file_mmap_ok(file, inode, pgoff, len)) 1281be83bbf8SLinus Torvalds return -EOVERFLOW; 1282be83bbf8SLinus Torvalds 12831c972597SDan Williams flags_mask = LEGACY_MAP_MASK | file->f_op->mmap_supported_flags; 1284077bf22bSOleg Nesterov 12851da177e4SLinus Torvalds switch (flags & MAP_TYPE) { 12861da177e4SLinus Torvalds case MAP_SHARED: 12871c972597SDan Williams /* 12881c972597SDan Williams * Force use of MAP_SHARED_VALIDATE with non-legacy 12891c972597SDan Williams * flags. E.g. MAP_SYNC is dangerous to use with 12901c972597SDan Williams * MAP_SHARED as you don't know which consistency model 12911c972597SDan Williams * you will get. We silently ignore unsupported flags 12921c972597SDan Williams * with MAP_SHARED to preserve backward compatibility. 12931c972597SDan Williams */ 12941c972597SDan Williams flags &= LEGACY_MAP_MASK; 1295e4a9bc58SJoe Perches fallthrough; 12961c972597SDan Williams case MAP_SHARED_VALIDATE: 12971c972597SDan Williams if (flags & ~flags_mask) 12981c972597SDan Williams return -EOPNOTSUPP; 1299dc617f29SDarrick J. Wong if (prot & PROT_WRITE) { 1300dc617f29SDarrick J. Wong if (!(file->f_mode & FMODE_WRITE)) 13011da177e4SLinus Torvalds return -EACCES; 1302dc617f29SDarrick J. Wong if (IS_SWAPFILE(file->f_mapping->host)) 1303dc617f29SDarrick J. Wong return -ETXTBSY; 1304dc617f29SDarrick J. Wong } 13051da177e4SLinus Torvalds 13061da177e4SLinus Torvalds /* 13071da177e4SLinus Torvalds * Make sure we don't allow writing to an append-only 13081da177e4SLinus Torvalds * file.. 13091da177e4SLinus Torvalds */ 13101da177e4SLinus Torvalds if (IS_APPEND(inode) && (file->f_mode & FMODE_WRITE)) 13111da177e4SLinus Torvalds return -EACCES; 13121da177e4SLinus Torvalds 13131da177e4SLinus Torvalds vm_flags |= VM_SHARED | VM_MAYSHARE; 13141da177e4SLinus Torvalds if (!(file->f_mode & FMODE_WRITE)) 13151da177e4SLinus Torvalds vm_flags &= ~(VM_MAYWRITE | VM_SHARED); 1316e4a9bc58SJoe Perches fallthrough; 13171da177e4SLinus Torvalds case MAP_PRIVATE: 13181da177e4SLinus Torvalds if (!(file->f_mode & FMODE_READ)) 13191da177e4SLinus Torvalds return -EACCES; 132090f8572bSEric W. Biederman if (path_noexec(&file->f_path)) { 132180c5606cSLinus Torvalds if (vm_flags & VM_EXEC) 132280c5606cSLinus Torvalds return -EPERM; 132380c5606cSLinus Torvalds vm_flags &= ~VM_MAYEXEC; 132480c5606cSLinus Torvalds } 132580c5606cSLinus Torvalds 132672c2d531SAl Viro if (!file->f_op->mmap) 132780c5606cSLinus Torvalds return -ENODEV; 1328b2c56e4fSOleg Nesterov if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP)) 1329b2c56e4fSOleg Nesterov return -EINVAL; 13301da177e4SLinus Torvalds break; 13311da177e4SLinus Torvalds 13321da177e4SLinus Torvalds default: 13331da177e4SLinus Torvalds return -EINVAL; 13341da177e4SLinus Torvalds } 13351da177e4SLinus Torvalds } else { 13361da177e4SLinus Torvalds switch (flags & MAP_TYPE) { 13371da177e4SLinus Torvalds case MAP_SHARED: 1338b2c56e4fSOleg Nesterov if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP)) 1339b2c56e4fSOleg Nesterov return -EINVAL; 1340ce363942STejun Heo /* 1341ce363942STejun Heo * Ignore pgoff. 1342ce363942STejun Heo */ 1343ce363942STejun Heo pgoff = 0; 13441da177e4SLinus Torvalds vm_flags |= VM_SHARED | VM_MAYSHARE; 13451da177e4SLinus Torvalds break; 13461da177e4SLinus Torvalds case MAP_PRIVATE: 13471da177e4SLinus Torvalds /* 13481da177e4SLinus Torvalds * Set pgoff according to addr for anon_vma. 13491da177e4SLinus Torvalds */ 13501da177e4SLinus Torvalds pgoff = addr >> PAGE_SHIFT; 13511da177e4SLinus Torvalds break; 13521da177e4SLinus Torvalds default: 13531da177e4SLinus Torvalds return -EINVAL; 13541da177e4SLinus Torvalds } 13551da177e4SLinus Torvalds } 13561da177e4SLinus Torvalds 1357c22c0d63SMichel Lespinasse /* 1358c22c0d63SMichel Lespinasse * Set 'VM_NORESERVE' if we should not account for the 1359c22c0d63SMichel Lespinasse * memory use of this mapping. 1360c22c0d63SMichel Lespinasse */ 1361c22c0d63SMichel Lespinasse if (flags & MAP_NORESERVE) { 1362c22c0d63SMichel Lespinasse /* We honor MAP_NORESERVE if allowed to overcommit */ 1363c22c0d63SMichel Lespinasse if (sysctl_overcommit_memory != OVERCOMMIT_NEVER) 1364c22c0d63SMichel Lespinasse vm_flags |= VM_NORESERVE; 1365c22c0d63SMichel Lespinasse 1366c22c0d63SMichel Lespinasse /* hugetlb applies strict overcommit unless MAP_NORESERVE */ 1367c22c0d63SMichel Lespinasse if (file && is_file_hugepages(file)) 1368c22c0d63SMichel Lespinasse vm_flags |= VM_NORESERVE; 1369c22c0d63SMichel Lespinasse } 1370c22c0d63SMichel Lespinasse 1371897ab3e0SMike Rapoport addr = mmap_region(file, addr, len, vm_flags, pgoff, uf); 137209a9f1d2SMichel Lespinasse if (!IS_ERR_VALUE(addr) && 137309a9f1d2SMichel Lespinasse ((vm_flags & VM_LOCKED) || 137409a9f1d2SMichel Lespinasse (flags & (MAP_POPULATE | MAP_NONBLOCK)) == MAP_POPULATE)) 137541badc15SMichel Lespinasse *populate = len; 1376bebeb3d6SMichel Lespinasse return addr; 13770165ab44SMiklos Szeredi } 13786be5ceb0SLinus Torvalds 1379a90f590aSDominik Brodowski unsigned long ksys_mmap_pgoff(unsigned long addr, unsigned long len, 1380a90f590aSDominik Brodowski unsigned long prot, unsigned long flags, 1381a90f590aSDominik Brodowski unsigned long fd, unsigned long pgoff) 138266f0dc48SHugh Dickins { 138366f0dc48SHugh Dickins struct file *file = NULL; 13841e3ee14bSChen Gang unsigned long retval; 138566f0dc48SHugh Dickins 138666f0dc48SHugh Dickins if (!(flags & MAP_ANONYMOUS)) { 1387120a795dSAl Viro audit_mmap_fd(fd, flags); 138866f0dc48SHugh Dickins file = fget(fd); 138966f0dc48SHugh Dickins if (!file) 13901e3ee14bSChen Gang return -EBADF; 13917bba8f0eSZhen Lei if (is_file_hugepages(file)) { 1392af73e4d9SNaoya Horiguchi len = ALIGN(len, huge_page_size(hstate_file(file))); 13937bba8f0eSZhen Lei } else if (unlikely(flags & MAP_HUGETLB)) { 1394493af578SJörn Engel retval = -EINVAL; 1395493af578SJörn Engel goto out_fput; 13967bba8f0eSZhen Lei } 139766f0dc48SHugh Dickins } else if (flags & MAP_HUGETLB) { 1398c103a4dcSAndrew Morton struct hstate *hs; 1399af73e4d9SNaoya Horiguchi 140020ac2893SAnshuman Khandual hs = hstate_sizelog((flags >> MAP_HUGE_SHIFT) & MAP_HUGE_MASK); 1401091d0d55SLi Zefan if (!hs) 1402091d0d55SLi Zefan return -EINVAL; 1403091d0d55SLi Zefan 1404091d0d55SLi Zefan len = ALIGN(len, huge_page_size(hs)); 140566f0dc48SHugh Dickins /* 140666f0dc48SHugh Dickins * VM_NORESERVE is used because the reservations will be 140766f0dc48SHugh Dickins * taken when vm_ops->mmap() is called 140866f0dc48SHugh Dickins */ 1409af73e4d9SNaoya Horiguchi file = hugetlb_file_setup(HUGETLB_ANON_FILE, len, 141042d7395fSAndi Kleen VM_NORESERVE, 141183c1fd76Szhangyiru HUGETLB_ANONHUGE_INODE, 141242d7395fSAndi Kleen (flags >> MAP_HUGE_SHIFT) & MAP_HUGE_MASK); 141366f0dc48SHugh Dickins if (IS_ERR(file)) 141466f0dc48SHugh Dickins return PTR_ERR(file); 141566f0dc48SHugh Dickins } 141666f0dc48SHugh Dickins 14179fbeb5abSMichal Hocko retval = vm_mmap_pgoff(file, addr, len, prot, flags, pgoff); 1418493af578SJörn Engel out_fput: 141966f0dc48SHugh Dickins if (file) 142066f0dc48SHugh Dickins fput(file); 142166f0dc48SHugh Dickins return retval; 142266f0dc48SHugh Dickins } 142366f0dc48SHugh Dickins 1424a90f590aSDominik Brodowski SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len, 1425a90f590aSDominik Brodowski unsigned long, prot, unsigned long, flags, 1426a90f590aSDominik Brodowski unsigned long, fd, unsigned long, pgoff) 1427a90f590aSDominik Brodowski { 1428a90f590aSDominik Brodowski return ksys_mmap_pgoff(addr, len, prot, flags, fd, pgoff); 1429a90f590aSDominik Brodowski } 1430a90f590aSDominik Brodowski 1431a4679373SChristoph Hellwig #ifdef __ARCH_WANT_SYS_OLD_MMAP 1432a4679373SChristoph Hellwig struct mmap_arg_struct { 1433a4679373SChristoph Hellwig unsigned long addr; 1434a4679373SChristoph Hellwig unsigned long len; 1435a4679373SChristoph Hellwig unsigned long prot; 1436a4679373SChristoph Hellwig unsigned long flags; 1437a4679373SChristoph Hellwig unsigned long fd; 1438a4679373SChristoph Hellwig unsigned long offset; 1439a4679373SChristoph Hellwig }; 1440a4679373SChristoph Hellwig 1441a4679373SChristoph Hellwig SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg) 1442a4679373SChristoph Hellwig { 1443a4679373SChristoph Hellwig struct mmap_arg_struct a; 1444a4679373SChristoph Hellwig 1445a4679373SChristoph Hellwig if (copy_from_user(&a, arg, sizeof(a))) 1446a4679373SChristoph Hellwig return -EFAULT; 1447de1741a1SAlexander Kuleshov if (offset_in_page(a.offset)) 1448a4679373SChristoph Hellwig return -EINVAL; 1449a4679373SChristoph Hellwig 1450a90f590aSDominik Brodowski return ksys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd, 1451a4679373SChristoph Hellwig a.offset >> PAGE_SHIFT); 1452a4679373SChristoph Hellwig } 1453a4679373SChristoph Hellwig #endif /* __ARCH_WANT_SYS_OLD_MMAP */ 1454a4679373SChristoph Hellwig 14554e950f6fSAlexey Dobriyan /* 14568bb4e7a2SWei Yang * Some shared mappings will want the pages marked read-only 14574e950f6fSAlexey Dobriyan * to track write events. If so, we'll downgrade vm_page_prot 14584e950f6fSAlexey Dobriyan * to the private version (using protection_map[] without the 14594e950f6fSAlexey Dobriyan * VM_SHARED bit). 14604e950f6fSAlexey Dobriyan */ 14616d2329f8SAndrea Arcangeli int vma_wants_writenotify(struct vm_area_struct *vma, pgprot_t vm_page_prot) 14624e950f6fSAlexey Dobriyan { 1463ca16d140SKOSAKI Motohiro vm_flags_t vm_flags = vma->vm_flags; 14648a04446aSKirill A. Shutemov const struct vm_operations_struct *vm_ops = vma->vm_ops; 14654e950f6fSAlexey Dobriyan 14664e950f6fSAlexey Dobriyan /* If it was private or non-writable, the write bit is already clear */ 14674e950f6fSAlexey Dobriyan if ((vm_flags & (VM_WRITE|VM_SHARED)) != ((VM_WRITE|VM_SHARED))) 14684e950f6fSAlexey Dobriyan return 0; 14694e950f6fSAlexey Dobriyan 14704e950f6fSAlexey Dobriyan /* The backer wishes to know when pages are first written to? */ 14718a04446aSKirill A. Shutemov if (vm_ops && (vm_ops->page_mkwrite || vm_ops->pfn_mkwrite)) 14724e950f6fSAlexey Dobriyan return 1; 14734e950f6fSAlexey Dobriyan 147464e45507SPeter Feiner /* The open routine did something to the protections that pgprot_modify 147564e45507SPeter Feiner * won't preserve? */ 14766d2329f8SAndrea Arcangeli if (pgprot_val(vm_page_prot) != 14776d2329f8SAndrea Arcangeli pgprot_val(vm_pgprot_modify(vm_page_prot, vm_flags))) 14784e950f6fSAlexey Dobriyan return 0; 14794e950f6fSAlexey Dobriyan 1480f96f7a40SDavid Hildenbrand /* 1481f96f7a40SDavid Hildenbrand * Do we need to track softdirty? hugetlb does not support softdirty 1482f96f7a40SDavid Hildenbrand * tracking yet. 1483f96f7a40SDavid Hildenbrand */ 1484f96f7a40SDavid Hildenbrand if (vma_soft_dirty_enabled(vma) && !is_vm_hugetlb_page(vma)) 148564e45507SPeter Feiner return 1; 148664e45507SPeter Feiner 148751d3d5ebSDavid Hildenbrand /* Do we need write faults for uffd-wp tracking? */ 148851d3d5ebSDavid Hildenbrand if (userfaultfd_wp(vma)) 148951d3d5ebSDavid Hildenbrand return 1; 149051d3d5ebSDavid Hildenbrand 14914e950f6fSAlexey Dobriyan /* Specialty mapping? */ 14924b6e1e37SKonstantin Khlebnikov if (vm_flags & VM_PFNMAP) 14934e950f6fSAlexey Dobriyan return 0; 14944e950f6fSAlexey Dobriyan 14954e950f6fSAlexey Dobriyan /* Can the mapping track the dirty pages? */ 14964e950f6fSAlexey Dobriyan return vma->vm_file && vma->vm_file->f_mapping && 1497f56753acSChristoph Hellwig mapping_can_writeback(vma->vm_file->f_mapping); 14984e950f6fSAlexey Dobriyan } 14994e950f6fSAlexey Dobriyan 1500fc8744adSLinus Torvalds /* 1501fc8744adSLinus Torvalds * We account for memory if it's a private writeable mapping, 15025a6fe125SMel Gorman * not hugepages and VM_NORESERVE wasn't set. 1503fc8744adSLinus Torvalds */ 1504ca16d140SKOSAKI Motohiro static inline int accountable_mapping(struct file *file, vm_flags_t vm_flags) 1505fc8744adSLinus Torvalds { 15065a6fe125SMel Gorman /* 15075a6fe125SMel Gorman * hugetlb has its own accounting separate from the core VM 15085a6fe125SMel Gorman * VM_HUGETLB may not be set yet so we cannot check for that flag. 15095a6fe125SMel Gorman */ 15105a6fe125SMel Gorman if (file && is_file_hugepages(file)) 15115a6fe125SMel Gorman return 0; 15125a6fe125SMel Gorman 1513fc8744adSLinus Torvalds return (vm_flags & (VM_NORESERVE | VM_SHARED | VM_WRITE)) == VM_WRITE; 1514fc8744adSLinus Torvalds } 1515fc8744adSLinus Torvalds 15163499a131SLiam R. Howlett /** 15173499a131SLiam R. Howlett * unmapped_area() - Find an area between the low_limit and the high_limit with 15183499a131SLiam R. Howlett * the correct alignment and offset, all from @info. Note: current->mm is used 15193499a131SLiam R. Howlett * for the search. 15203499a131SLiam R. Howlett * 152182b24936SVernon Yang * @info: The unmapped area information including the range [low_limit - 152282b24936SVernon Yang * high_limit), the alignment offset and mask. 15233499a131SLiam R. Howlett * 15243499a131SLiam R. Howlett * Return: A memory address or -ENOMEM. 15253499a131SLiam R. Howlett */ 1526baceaf1cSJaewon Kim static unsigned long unmapped_area(struct vm_unmapped_area_info *info) 1527db4fbfb9SMichel Lespinasse { 15283499a131SLiam R. Howlett unsigned long length, gap; 1529db4fbfb9SMichel Lespinasse 15303499a131SLiam R. Howlett MA_STATE(mas, ¤t->mm->mm_mt, 0, 0); 1531db4fbfb9SMichel Lespinasse 1532db4fbfb9SMichel Lespinasse /* Adjust search length to account for worst case alignment overhead */ 1533db4fbfb9SMichel Lespinasse length = info->length + info->align_mask; 1534db4fbfb9SMichel Lespinasse if (length < info->length) 1535db4fbfb9SMichel Lespinasse return -ENOMEM; 1536db4fbfb9SMichel Lespinasse 15373499a131SLiam R. Howlett if (mas_empty_area(&mas, info->low_limit, info->high_limit - 1, 15383499a131SLiam R. Howlett length)) 15393499a131SLiam R. Howlett return -ENOMEM; 15403499a131SLiam R. Howlett 1541d4af56c5SLiam R. Howlett gap = mas.index; 1542d4af56c5SLiam R. Howlett gap += (info->align_offset - gap) & info->align_mask; 15433499a131SLiam R. Howlett return gap; 1544db4fbfb9SMichel Lespinasse } 1545db4fbfb9SMichel Lespinasse 15463499a131SLiam R. Howlett /** 15473499a131SLiam R. Howlett * unmapped_area_topdown() - Find an area between the low_limit and the 154882b24936SVernon Yang * high_limit with the correct alignment and offset at the highest available 15493499a131SLiam R. Howlett * address, all from @info. Note: current->mm is used for the search. 15503499a131SLiam R. Howlett * 155182b24936SVernon Yang * @info: The unmapped area information including the range [low_limit - 155282b24936SVernon Yang * high_limit), the alignment offset and mask. 15533499a131SLiam R. Howlett * 15543499a131SLiam R. Howlett * Return: A memory address or -ENOMEM. 15553499a131SLiam R. Howlett */ 1556baceaf1cSJaewon Kim static unsigned long unmapped_area_topdown(struct vm_unmapped_area_info *info) 1557db4fbfb9SMichel Lespinasse { 15583499a131SLiam R. Howlett unsigned long length, gap; 1559d4af56c5SLiam R. Howlett 15603499a131SLiam R. Howlett MA_STATE(mas, ¤t->mm->mm_mt, 0, 0); 1561db4fbfb9SMichel Lespinasse /* Adjust search length to account for worst case alignment overhead */ 1562db4fbfb9SMichel Lespinasse length = info->length + info->align_mask; 1563db4fbfb9SMichel Lespinasse if (length < info->length) 1564db4fbfb9SMichel Lespinasse return -ENOMEM; 1565db4fbfb9SMichel Lespinasse 15663499a131SLiam R. Howlett if (mas_empty_area_rev(&mas, info->low_limit, info->high_limit - 1, 15673499a131SLiam R. Howlett length)) 15683499a131SLiam R. Howlett return -ENOMEM; 15693499a131SLiam R. Howlett 1570d4af56c5SLiam R. Howlett gap = mas.last + 1 - info->length; 1571d4af56c5SLiam R. Howlett gap -= (gap - info->align_offset) & info->align_mask; 15723499a131SLiam R. Howlett return gap; 1573db4fbfb9SMichel Lespinasse } 1574db4fbfb9SMichel Lespinasse 1575baceaf1cSJaewon Kim /* 1576baceaf1cSJaewon Kim * Search for an unmapped address range. 1577baceaf1cSJaewon Kim * 1578baceaf1cSJaewon Kim * We are looking for a range that: 1579baceaf1cSJaewon Kim * - does not intersect with any VMA; 1580baceaf1cSJaewon Kim * - is contained within the [low_limit, high_limit) interval; 1581baceaf1cSJaewon Kim * - is at least the desired size. 1582baceaf1cSJaewon Kim * - satisfies (begin_addr & align_mask) == (align_offset & align_mask) 1583baceaf1cSJaewon Kim */ 1584baceaf1cSJaewon Kim unsigned long vm_unmapped_area(struct vm_unmapped_area_info *info) 1585baceaf1cSJaewon Kim { 1586df529cabSJaewon Kim unsigned long addr; 1587df529cabSJaewon Kim 1588baceaf1cSJaewon Kim if (info->flags & VM_UNMAPPED_AREA_TOPDOWN) 1589df529cabSJaewon Kim addr = unmapped_area_topdown(info); 1590baceaf1cSJaewon Kim else 1591df529cabSJaewon Kim addr = unmapped_area(info); 1592df529cabSJaewon Kim 1593df529cabSJaewon Kim trace_vm_unmapped_area(addr, info); 1594df529cabSJaewon Kim return addr; 1595baceaf1cSJaewon Kim } 1596f6795053SSteve Capper 15971da177e4SLinus Torvalds /* Get an address range which is currently unmapped. 15981da177e4SLinus Torvalds * For shmat() with addr=0. 15991da177e4SLinus Torvalds * 16001da177e4SLinus Torvalds * Ugly calling convention alert: 16011da177e4SLinus Torvalds * Return value with the low bits set means error value, 16021da177e4SLinus Torvalds * ie 16031da177e4SLinus Torvalds * if (ret & ~PAGE_MASK) 16041da177e4SLinus Torvalds * error = ret; 16051da177e4SLinus Torvalds * 16061da177e4SLinus Torvalds * This function "knows" that -ENOMEM has the bits set. 16071da177e4SLinus Torvalds */ 16081da177e4SLinus Torvalds unsigned long 16094b439e25SChristophe Leroy generic_get_unmapped_area(struct file *filp, unsigned long addr, 16104b439e25SChristophe Leroy unsigned long len, unsigned long pgoff, 16114b439e25SChristophe Leroy unsigned long flags) 16121da177e4SLinus Torvalds { 16131da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 16141be7107fSHugh Dickins struct vm_area_struct *vma, *prev; 1615db4fbfb9SMichel Lespinasse struct vm_unmapped_area_info info; 16162cb4de08SChristophe Leroy const unsigned long mmap_end = arch_get_mmap_end(addr, len, flags); 16171da177e4SLinus Torvalds 1618f6795053SSteve Capper if (len > mmap_end - mmap_min_addr) 16191da177e4SLinus Torvalds return -ENOMEM; 16201da177e4SLinus Torvalds 162106abdfb4SBenjamin Herrenschmidt if (flags & MAP_FIXED) 162206abdfb4SBenjamin Herrenschmidt return addr; 162306abdfb4SBenjamin Herrenschmidt 16241da177e4SLinus Torvalds if (addr) { 16251da177e4SLinus Torvalds addr = PAGE_ALIGN(addr); 16261be7107fSHugh Dickins vma = find_vma_prev(mm, addr, &prev); 1627f6795053SSteve Capper if (mmap_end - len >= addr && addr >= mmap_min_addr && 16281be7107fSHugh Dickins (!vma || addr + len <= vm_start_gap(vma)) && 16291be7107fSHugh Dickins (!prev || addr >= vm_end_gap(prev))) 16301da177e4SLinus Torvalds return addr; 16311da177e4SLinus Torvalds } 16321da177e4SLinus Torvalds 1633db4fbfb9SMichel Lespinasse info.flags = 0; 1634db4fbfb9SMichel Lespinasse info.length = len; 16354e99b021SHeiko Carstens info.low_limit = mm->mmap_base; 1636f6795053SSteve Capper info.high_limit = mmap_end; 1637db4fbfb9SMichel Lespinasse info.align_mask = 0; 163809ef5283SJaewon Kim info.align_offset = 0; 1639db4fbfb9SMichel Lespinasse return vm_unmapped_area(&info); 16401da177e4SLinus Torvalds } 16414b439e25SChristophe Leroy 16424b439e25SChristophe Leroy #ifndef HAVE_ARCH_UNMAPPED_AREA 16434b439e25SChristophe Leroy unsigned long 16444b439e25SChristophe Leroy arch_get_unmapped_area(struct file *filp, unsigned long addr, 16454b439e25SChristophe Leroy unsigned long len, unsigned long pgoff, 16464b439e25SChristophe Leroy unsigned long flags) 16474b439e25SChristophe Leroy { 16484b439e25SChristophe Leroy return generic_get_unmapped_area(filp, addr, len, pgoff, flags); 16494b439e25SChristophe Leroy } 16501da177e4SLinus Torvalds #endif 16511da177e4SLinus Torvalds 16521da177e4SLinus Torvalds /* 16531da177e4SLinus Torvalds * This mmap-allocator allocates new areas top-down from below the 16541da177e4SLinus Torvalds * stack's low limit (the base): 16551da177e4SLinus Torvalds */ 16561da177e4SLinus Torvalds unsigned long 16574b439e25SChristophe Leroy generic_get_unmapped_area_topdown(struct file *filp, unsigned long addr, 165843cca0b1SYang Fan unsigned long len, unsigned long pgoff, 165943cca0b1SYang Fan unsigned long flags) 16601da177e4SLinus Torvalds { 16611be7107fSHugh Dickins struct vm_area_struct *vma, *prev; 16621da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 1663db4fbfb9SMichel Lespinasse struct vm_unmapped_area_info info; 16642cb4de08SChristophe Leroy const unsigned long mmap_end = arch_get_mmap_end(addr, len, flags); 16651da177e4SLinus Torvalds 16661da177e4SLinus Torvalds /* requested length too big for entire address space */ 1667f6795053SSteve Capper if (len > mmap_end - mmap_min_addr) 16681da177e4SLinus Torvalds return -ENOMEM; 16691da177e4SLinus Torvalds 167006abdfb4SBenjamin Herrenschmidt if (flags & MAP_FIXED) 167106abdfb4SBenjamin Herrenschmidt return addr; 167206abdfb4SBenjamin Herrenschmidt 16731da177e4SLinus Torvalds /* requesting a specific address */ 16741da177e4SLinus Torvalds if (addr) { 16751da177e4SLinus Torvalds addr = PAGE_ALIGN(addr); 16761be7107fSHugh Dickins vma = find_vma_prev(mm, addr, &prev); 1677f6795053SSteve Capper if (mmap_end - len >= addr && addr >= mmap_min_addr && 16781be7107fSHugh Dickins (!vma || addr + len <= vm_start_gap(vma)) && 16791be7107fSHugh Dickins (!prev || addr >= vm_end_gap(prev))) 16801da177e4SLinus Torvalds return addr; 16811da177e4SLinus Torvalds } 16821da177e4SLinus Torvalds 1683db4fbfb9SMichel Lespinasse info.flags = VM_UNMAPPED_AREA_TOPDOWN; 1684db4fbfb9SMichel Lespinasse info.length = len; 16852afc745fSAkira Takeuchi info.low_limit = max(PAGE_SIZE, mmap_min_addr); 1686f6795053SSteve Capper info.high_limit = arch_get_mmap_base(addr, mm->mmap_base); 1687db4fbfb9SMichel Lespinasse info.align_mask = 0; 168809ef5283SJaewon Kim info.align_offset = 0; 1689db4fbfb9SMichel Lespinasse addr = vm_unmapped_area(&info); 1690b716ad95SXiao Guangrong 16911da177e4SLinus Torvalds /* 16921da177e4SLinus Torvalds * A failed mmap() very likely causes application failure, 16931da177e4SLinus Torvalds * so fall back to the bottom-up function here. This scenario 16941da177e4SLinus Torvalds * can happen with large stack limits and large mmap() 16951da177e4SLinus Torvalds * allocations. 16961da177e4SLinus Torvalds */ 1697de1741a1SAlexander Kuleshov if (offset_in_page(addr)) { 1698db4fbfb9SMichel Lespinasse VM_BUG_ON(addr != -ENOMEM); 1699db4fbfb9SMichel Lespinasse info.flags = 0; 1700db4fbfb9SMichel Lespinasse info.low_limit = TASK_UNMAPPED_BASE; 1701f6795053SSteve Capper info.high_limit = mmap_end; 1702db4fbfb9SMichel Lespinasse addr = vm_unmapped_area(&info); 1703db4fbfb9SMichel Lespinasse } 17041da177e4SLinus Torvalds 17051da177e4SLinus Torvalds return addr; 17061da177e4SLinus Torvalds } 17074b439e25SChristophe Leroy 17084b439e25SChristophe Leroy #ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN 17094b439e25SChristophe Leroy unsigned long 17104b439e25SChristophe Leroy arch_get_unmapped_area_topdown(struct file *filp, unsigned long addr, 17114b439e25SChristophe Leroy unsigned long len, unsigned long pgoff, 17124b439e25SChristophe Leroy unsigned long flags) 17134b439e25SChristophe Leroy { 17144b439e25SChristophe Leroy return generic_get_unmapped_area_topdown(filp, addr, len, pgoff, flags); 17154b439e25SChristophe Leroy } 17161da177e4SLinus Torvalds #endif 17171da177e4SLinus Torvalds 17181da177e4SLinus Torvalds unsigned long 17191da177e4SLinus Torvalds get_unmapped_area(struct file *file, unsigned long addr, unsigned long len, 17201da177e4SLinus Torvalds unsigned long pgoff, unsigned long flags) 17211da177e4SLinus Torvalds { 172206abdfb4SBenjamin Herrenschmidt unsigned long (*get_area)(struct file *, unsigned long, 172306abdfb4SBenjamin Herrenschmidt unsigned long, unsigned long, unsigned long); 172407ab67c8SLinus Torvalds 17259206de95SAl Viro unsigned long error = arch_mmap_check(addr, len, flags); 17269206de95SAl Viro if (error) 17279206de95SAl Viro return error; 17289206de95SAl Viro 17299206de95SAl Viro /* Careful about overflows.. */ 17309206de95SAl Viro if (len > TASK_SIZE) 17319206de95SAl Viro return -ENOMEM; 17329206de95SAl Viro 173307ab67c8SLinus Torvalds get_area = current->mm->get_unmapped_area; 1734c01d5b30SHugh Dickins if (file) { 1735c01d5b30SHugh Dickins if (file->f_op->get_unmapped_area) 173607ab67c8SLinus Torvalds get_area = file->f_op->get_unmapped_area; 1737c01d5b30SHugh Dickins } else if (flags & MAP_SHARED) { 1738c01d5b30SHugh Dickins /* 1739c01d5b30SHugh Dickins * mmap_region() will call shmem_zero_setup() to create a file, 1740c01d5b30SHugh Dickins * so use shmem's get_unmapped_area in case it can be huge. 174145e55300SPeter Collingbourne * do_mmap() will clear pgoff, so match alignment. 1742c01d5b30SHugh Dickins */ 1743c01d5b30SHugh Dickins pgoff = 0; 1744c01d5b30SHugh Dickins get_area = shmem_get_unmapped_area; 1745c01d5b30SHugh Dickins } 1746c01d5b30SHugh Dickins 174707ab67c8SLinus Torvalds addr = get_area(file, addr, len, pgoff, flags); 174807ab67c8SLinus Torvalds if (IS_ERR_VALUE(addr)) 174907ab67c8SLinus Torvalds return addr; 175007ab67c8SLinus Torvalds 17511da177e4SLinus Torvalds if (addr > TASK_SIZE - len) 17521da177e4SLinus Torvalds return -ENOMEM; 1753de1741a1SAlexander Kuleshov if (offset_in_page(addr)) 17541da177e4SLinus Torvalds return -EINVAL; 175506abdfb4SBenjamin Herrenschmidt 17569ac4ed4bSAl Viro error = security_mmap_addr(addr); 17579ac4ed4bSAl Viro return error ? error : addr; 17581da177e4SLinus Torvalds } 17591da177e4SLinus Torvalds 17601da177e4SLinus Torvalds EXPORT_SYMBOL(get_unmapped_area); 17611da177e4SLinus Torvalds 1762be8432e7SLiam R. Howlett /** 1763abdba2ddSLiam R. Howlett * find_vma_intersection() - Look up the first VMA which intersects the interval 1764abdba2ddSLiam R. Howlett * @mm: The process address space. 1765abdba2ddSLiam R. Howlett * @start_addr: The inclusive start user address. 1766abdba2ddSLiam R. Howlett * @end_addr: The exclusive end user address. 1767abdba2ddSLiam R. Howlett * 1768abdba2ddSLiam R. Howlett * Returns: The first VMA within the provided range, %NULL otherwise. Assumes 1769abdba2ddSLiam R. Howlett * start_addr < end_addr. 1770abdba2ddSLiam R. Howlett */ 1771abdba2ddSLiam R. Howlett struct vm_area_struct *find_vma_intersection(struct mm_struct *mm, 1772abdba2ddSLiam R. Howlett unsigned long start_addr, 1773abdba2ddSLiam R. Howlett unsigned long end_addr) 1774abdba2ddSLiam R. Howlett { 1775abdba2ddSLiam R. Howlett unsigned long index = start_addr; 1776abdba2ddSLiam R. Howlett 1777abdba2ddSLiam R. Howlett mmap_assert_locked(mm); 17787964cf8cSLiam R. Howlett return mt_find(&mm->mm_mt, &index, end_addr - 1); 1779abdba2ddSLiam R. Howlett } 1780abdba2ddSLiam R. Howlett EXPORT_SYMBOL(find_vma_intersection); 1781abdba2ddSLiam R. Howlett 1782abdba2ddSLiam R. Howlett /** 1783be8432e7SLiam R. Howlett * find_vma() - Find the VMA for a given address, or the next VMA. 1784be8432e7SLiam R. Howlett * @mm: The mm_struct to check 1785be8432e7SLiam R. Howlett * @addr: The address 1786be8432e7SLiam R. Howlett * 1787be8432e7SLiam R. Howlett * Returns: The VMA associated with addr, or the next VMA. 1788be8432e7SLiam R. Howlett * May return %NULL in the case of no VMA at addr or above. 1789be8432e7SLiam R. Howlett */ 17901da177e4SLinus Torvalds struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr) 17911da177e4SLinus Torvalds { 1792be8432e7SLiam R. Howlett unsigned long index = addr; 17931da177e4SLinus Torvalds 17945b78ed24SLuigi Rizzo mmap_assert_locked(mm); 17957964cf8cSLiam R. Howlett return mt_find(&mm->mm_mt, &index, ULONG_MAX); 17961da177e4SLinus Torvalds } 17971da177e4SLinus Torvalds EXPORT_SYMBOL(find_vma); 17981da177e4SLinus Torvalds 17997fdbd37dSLiam R. Howlett /** 18007fdbd37dSLiam R. Howlett * find_vma_prev() - Find the VMA for a given address, or the next vma and 18017fdbd37dSLiam R. Howlett * set %pprev to the previous VMA, if any. 18027fdbd37dSLiam R. Howlett * @mm: The mm_struct to check 18037fdbd37dSLiam R. Howlett * @addr: The address 18047fdbd37dSLiam R. Howlett * @pprev: The pointer to set to the previous VMA 18057fdbd37dSLiam R. Howlett * 18067fdbd37dSLiam R. Howlett * Note that RCU lock is missing here since the external mmap_lock() is used 18077fdbd37dSLiam R. Howlett * instead. 18087fdbd37dSLiam R. Howlett * 18097fdbd37dSLiam R. Howlett * Returns: The VMA associated with @addr, or the next vma. 18107fdbd37dSLiam R. Howlett * May return %NULL in the case of no vma at addr or above. 18116bd4837dSKOSAKI Motohiro */ 18121da177e4SLinus Torvalds struct vm_area_struct * 18131da177e4SLinus Torvalds find_vma_prev(struct mm_struct *mm, unsigned long addr, 18141da177e4SLinus Torvalds struct vm_area_struct **pprev) 18151da177e4SLinus Torvalds { 18166bd4837dSKOSAKI Motohiro struct vm_area_struct *vma; 18177fdbd37dSLiam R. Howlett MA_STATE(mas, &mm->mm_mt, addr, addr); 18181da177e4SLinus Torvalds 18197fdbd37dSLiam R. Howlett vma = mas_walk(&mas); 18207fdbd37dSLiam R. Howlett *pprev = mas_prev(&mas, 0); 18217fdbd37dSLiam R. Howlett if (!vma) 18227fdbd37dSLiam R. Howlett vma = mas_next(&mas, ULONG_MAX); 18236bd4837dSKOSAKI Motohiro return vma; 18241da177e4SLinus Torvalds } 18251da177e4SLinus Torvalds 18261da177e4SLinus Torvalds /* 18271da177e4SLinus Torvalds * Verify that the stack growth is acceptable and 18281da177e4SLinus Torvalds * update accounting. This is shared with both the 18291da177e4SLinus Torvalds * grow-up and grow-down cases. 18301da177e4SLinus Torvalds */ 18311be7107fSHugh Dickins static int acct_stack_growth(struct vm_area_struct *vma, 18321be7107fSHugh Dickins unsigned long size, unsigned long grow) 18331da177e4SLinus Torvalds { 18341da177e4SLinus Torvalds struct mm_struct *mm = vma->vm_mm; 18351be7107fSHugh Dickins unsigned long new_start; 18361da177e4SLinus Torvalds 18371da177e4SLinus Torvalds /* address space limit tests */ 183884638335SKonstantin Khlebnikov if (!may_expand_vm(mm, vma->vm_flags, grow)) 18391da177e4SLinus Torvalds return -ENOMEM; 18401da177e4SLinus Torvalds 18411da177e4SLinus Torvalds /* Stack limit test */ 184224c79d8eSKrzysztof Opasiak if (size > rlimit(RLIMIT_STACK)) 18431da177e4SLinus Torvalds return -ENOMEM; 18441da177e4SLinus Torvalds 18451da177e4SLinus Torvalds /* mlock limit tests */ 1846c5d8a364SMiaohe Lin if (mlock_future_check(mm, vma->vm_flags, grow << PAGE_SHIFT)) 18471da177e4SLinus Torvalds return -ENOMEM; 18481da177e4SLinus Torvalds 18490d59a01bSAdam Litke /* Check to ensure the stack will not grow into a hugetlb-only region */ 18500d59a01bSAdam Litke new_start = (vma->vm_flags & VM_GROWSUP) ? vma->vm_start : 18510d59a01bSAdam Litke vma->vm_end - size; 18520d59a01bSAdam Litke if (is_hugepage_only_range(vma->vm_mm, new_start, size)) 18530d59a01bSAdam Litke return -EFAULT; 18540d59a01bSAdam Litke 18551da177e4SLinus Torvalds /* 18561da177e4SLinus Torvalds * Overcommit.. This must be the final test, as it will 18571da177e4SLinus Torvalds * update security statistics. 18581da177e4SLinus Torvalds */ 185905fa199dSHugh Dickins if (security_vm_enough_memory_mm(mm, grow)) 18601da177e4SLinus Torvalds return -ENOMEM; 18611da177e4SLinus Torvalds 18621da177e4SLinus Torvalds return 0; 18631da177e4SLinus Torvalds } 18641da177e4SLinus Torvalds 186546dea3d0SHugh Dickins #if defined(CONFIG_STACK_GROWSUP) || defined(CONFIG_IA64) 18661da177e4SLinus Torvalds /* 186746dea3d0SHugh Dickins * PA-RISC uses this for its stack; IA64 for its Register Backing Store. 186846dea3d0SHugh Dickins * vma is the last one with address > vma->vm_end. Have to extend vma. 18691da177e4SLinus Torvalds */ 187046dea3d0SHugh Dickins int expand_upwards(struct vm_area_struct *vma, unsigned long address) 18711da177e4SLinus Torvalds { 187209357814SOleg Nesterov struct mm_struct *mm = vma->vm_mm; 18731be7107fSHugh Dickins struct vm_area_struct *next; 18741be7107fSHugh Dickins unsigned long gap_addr; 187512352d3cSKonstantin Khlebnikov int error = 0; 1876d4af56c5SLiam R. Howlett MA_STATE(mas, &mm->mm_mt, 0, 0); 18771da177e4SLinus Torvalds 18781da177e4SLinus Torvalds if (!(vma->vm_flags & VM_GROWSUP)) 18791da177e4SLinus Torvalds return -EFAULT; 18801da177e4SLinus Torvalds 1881bd726c90SHelge Deller /* Guard against exceeding limits of the address space. */ 18821be7107fSHugh Dickins address &= PAGE_MASK; 188337511fb5SHelge Deller if (address >= (TASK_SIZE & PAGE_MASK)) 188412352d3cSKonstantin Khlebnikov return -ENOMEM; 1885bd726c90SHelge Deller address += PAGE_SIZE; 188612352d3cSKonstantin Khlebnikov 18871be7107fSHugh Dickins /* Enforce stack_guard_gap */ 18881be7107fSHugh Dickins gap_addr = address + stack_guard_gap; 1889bd726c90SHelge Deller 1890bd726c90SHelge Deller /* Guard against overflow */ 1891bd726c90SHelge Deller if (gap_addr < address || gap_addr > TASK_SIZE) 1892bd726c90SHelge Deller gap_addr = TASK_SIZE; 1893bd726c90SHelge Deller 1894763ecb03SLiam R. Howlett next = find_vma_intersection(mm, vma->vm_end, gap_addr); 1895763ecb03SLiam R. Howlett if (next && vma_is_accessible(next)) { 18961be7107fSHugh Dickins if (!(next->vm_flags & VM_GROWSUP)) 18971be7107fSHugh Dickins return -ENOMEM; 18981be7107fSHugh Dickins /* Check that both stack segments have the same anon_vma? */ 18991be7107fSHugh Dickins } 19001be7107fSHugh Dickins 1901c5d5546eSVernon Yang if (mas_preallocate(&mas, GFP_KERNEL)) 19021da177e4SLinus Torvalds return -ENOMEM; 19031da177e4SLinus Torvalds 1904d4af56c5SLiam R. Howlett /* We must make sure the anon_vma is allocated. */ 1905d4af56c5SLiam R. Howlett if (unlikely(anon_vma_prepare(vma))) { 1906d4af56c5SLiam R. Howlett mas_destroy(&mas); 1907d4af56c5SLiam R. Howlett return -ENOMEM; 1908d4af56c5SLiam R. Howlett } 1909d4af56c5SLiam R. Howlett 19101da177e4SLinus Torvalds /* 19111da177e4SLinus Torvalds * vma->vm_start/vm_end cannot change under us because the caller 1912c1e8d7c6SMichel Lespinasse * is required to hold the mmap_lock in read mode. We need the 19131da177e4SLinus Torvalds * anon_vma lock to serialize against concurrent expand_stacks. 19141da177e4SLinus Torvalds */ 191512352d3cSKonstantin Khlebnikov anon_vma_lock_write(vma->anon_vma); 19161da177e4SLinus Torvalds 19171da177e4SLinus Torvalds /* Somebody else might have raced and expanded it already */ 19181da177e4SLinus Torvalds if (address > vma->vm_end) { 19191da177e4SLinus Torvalds unsigned long size, grow; 19201da177e4SLinus Torvalds 19211da177e4SLinus Torvalds size = address - vma->vm_start; 19221da177e4SLinus Torvalds grow = (address - vma->vm_end) >> PAGE_SHIFT; 19231da177e4SLinus Torvalds 192442c36f63SHugh Dickins error = -ENOMEM; 192542c36f63SHugh Dickins if (vma->vm_pgoff + (size >> PAGE_SHIFT) >= vma->vm_pgoff) { 19261da177e4SLinus Torvalds error = acct_stack_growth(vma, size, grow); 19273af9e859SEric B Munson if (!error) { 19284128997bSMichel Lespinasse /* 1929524e00b3SLiam R. Howlett * We only hold a shared mmap_lock lock here, so 1930524e00b3SLiam R. Howlett * we need to protect against concurrent vma 1931524e00b3SLiam R. Howlett * expansions. anon_vma_lock_write() doesn't 1932524e00b3SLiam R. Howlett * help here, as we don't guarantee that all 1933524e00b3SLiam R. Howlett * growable vmas in a mm share the same root 1934524e00b3SLiam R. Howlett * anon vma. So, we reuse mm->page_table_lock 1935524e00b3SLiam R. Howlett * to guard against concurrent vma expansions. 19364128997bSMichel Lespinasse */ 193709357814SOleg Nesterov spin_lock(&mm->page_table_lock); 193887e8827bSOleg Nesterov if (vma->vm_flags & VM_LOCKED) 193909357814SOleg Nesterov mm->locked_vm += grow; 194084638335SKonstantin Khlebnikov vm_stat_account(mm, vma->vm_flags, grow); 1941bf181b9fSMichel Lespinasse anon_vma_interval_tree_pre_update_vma(vma); 19421da177e4SLinus Torvalds vma->vm_end = address; 1943d4af56c5SLiam R. Howlett /* Overwrite old entry in mtree. */ 1944fbcc3104SLiam R. Howlett mas_set_range(&mas, vma->vm_start, address - 1); 1945fbcc3104SLiam R. Howlett mas_store_prealloc(&mas, vma); 1946bf181b9fSMichel Lespinasse anon_vma_interval_tree_post_update_vma(vma); 194709357814SOleg Nesterov spin_unlock(&mm->page_table_lock); 19484128997bSMichel Lespinasse 19493af9e859SEric B Munson perf_event_mmap(vma); 19503af9e859SEric B Munson } 19511da177e4SLinus Torvalds } 195242c36f63SHugh Dickins } 195312352d3cSKonstantin Khlebnikov anon_vma_unlock_write(vma->anon_vma); 1954c791576cSYang Shi khugepaged_enter_vma(vma, vma->vm_flags); 1955d4af56c5SLiam R. Howlett mas_destroy(&mas); 19561da177e4SLinus Torvalds return error; 19571da177e4SLinus Torvalds } 195846dea3d0SHugh Dickins #endif /* CONFIG_STACK_GROWSUP || CONFIG_IA64 */ 195946dea3d0SHugh Dickins 19601da177e4SLinus Torvalds /* 19611da177e4SLinus Torvalds * vma is the first one with address < vma->vm_start. Have to extend vma. 19621da177e4SLinus Torvalds */ 1963524e00b3SLiam R. Howlett int expand_downwards(struct vm_area_struct *vma, unsigned long address) 19641da177e4SLinus Torvalds { 196509357814SOleg Nesterov struct mm_struct *mm = vma->vm_mm; 1966763ecb03SLiam R. Howlett MA_STATE(mas, &mm->mm_mt, vma->vm_start, vma->vm_start); 19671be7107fSHugh Dickins struct vm_area_struct *prev; 19680a1d5299SJann Horn int error = 0; 19691da177e4SLinus Torvalds 19708869477aSEric Paris address &= PAGE_MASK; 19710a1d5299SJann Horn if (address < mmap_min_addr) 19720a1d5299SJann Horn return -EPERM; 19738869477aSEric Paris 19741be7107fSHugh Dickins /* Enforce stack_guard_gap */ 1975763ecb03SLiam R. Howlett prev = mas_prev(&mas, 0); 19761be7107fSHugh Dickins /* Check that both stack segments have the same anon_vma? */ 197732e4e6d5SOleg Nesterov if (prev && !(prev->vm_flags & VM_GROWSDOWN) && 19783122e80eSAnshuman Khandual vma_is_accessible(prev)) { 197932e4e6d5SOleg Nesterov if (address - prev->vm_end < stack_guard_gap) 198032e4e6d5SOleg Nesterov return -ENOMEM; 19811be7107fSHugh Dickins } 19821be7107fSHugh Dickins 1983c5d5546eSVernon Yang if (mas_preallocate(&mas, GFP_KERNEL)) 198412352d3cSKonstantin Khlebnikov return -ENOMEM; 19851da177e4SLinus Torvalds 1986d4af56c5SLiam R. Howlett /* We must make sure the anon_vma is allocated. */ 1987d4af56c5SLiam R. Howlett if (unlikely(anon_vma_prepare(vma))) { 1988d4af56c5SLiam R. Howlett mas_destroy(&mas); 1989d4af56c5SLiam R. Howlett return -ENOMEM; 1990d4af56c5SLiam R. Howlett } 1991d4af56c5SLiam R. Howlett 19921da177e4SLinus Torvalds /* 19931da177e4SLinus Torvalds * vma->vm_start/vm_end cannot change under us because the caller 1994c1e8d7c6SMichel Lespinasse * is required to hold the mmap_lock in read mode. We need the 19951da177e4SLinus Torvalds * anon_vma lock to serialize against concurrent expand_stacks. 19961da177e4SLinus Torvalds */ 199712352d3cSKonstantin Khlebnikov anon_vma_lock_write(vma->anon_vma); 19981da177e4SLinus Torvalds 19991da177e4SLinus Torvalds /* Somebody else might have raced and expanded it already */ 20001da177e4SLinus Torvalds if (address < vma->vm_start) { 20011da177e4SLinus Torvalds unsigned long size, grow; 20021da177e4SLinus Torvalds 20031da177e4SLinus Torvalds size = vma->vm_end - address; 20041da177e4SLinus Torvalds grow = (vma->vm_start - address) >> PAGE_SHIFT; 20051da177e4SLinus Torvalds 2006a626ca6aSLinus Torvalds error = -ENOMEM; 2007a626ca6aSLinus Torvalds if (grow <= vma->vm_pgoff) { 20081da177e4SLinus Torvalds error = acct_stack_growth(vma, size, grow); 20091da177e4SLinus Torvalds if (!error) { 20104128997bSMichel Lespinasse /* 2011524e00b3SLiam R. Howlett * We only hold a shared mmap_lock lock here, so 2012524e00b3SLiam R. Howlett * we need to protect against concurrent vma 2013524e00b3SLiam R. Howlett * expansions. anon_vma_lock_write() doesn't 2014524e00b3SLiam R. Howlett * help here, as we don't guarantee that all 2015524e00b3SLiam R. Howlett * growable vmas in a mm share the same root 2016524e00b3SLiam R. Howlett * anon vma. So, we reuse mm->page_table_lock 2017524e00b3SLiam R. Howlett * to guard against concurrent vma expansions. 20184128997bSMichel Lespinasse */ 201909357814SOleg Nesterov spin_lock(&mm->page_table_lock); 202087e8827bSOleg Nesterov if (vma->vm_flags & VM_LOCKED) 202109357814SOleg Nesterov mm->locked_vm += grow; 202284638335SKonstantin Khlebnikov vm_stat_account(mm, vma->vm_flags, grow); 2023bf181b9fSMichel Lespinasse anon_vma_interval_tree_pre_update_vma(vma); 20241da177e4SLinus Torvalds vma->vm_start = address; 20251da177e4SLinus Torvalds vma->vm_pgoff -= grow; 2026d4af56c5SLiam R. Howlett /* Overwrite old entry in mtree. */ 2027fbcc3104SLiam R. Howlett mas_set_range(&mas, address, vma->vm_end - 1); 2028fbcc3104SLiam R. Howlett mas_store_prealloc(&mas, vma); 2029bf181b9fSMichel Lespinasse anon_vma_interval_tree_post_update_vma(vma); 203009357814SOleg Nesterov spin_unlock(&mm->page_table_lock); 20314128997bSMichel Lespinasse 20323af9e859SEric B Munson perf_event_mmap(vma); 20331da177e4SLinus Torvalds } 20341da177e4SLinus Torvalds } 2035a626ca6aSLinus Torvalds } 203612352d3cSKonstantin Khlebnikov anon_vma_unlock_write(vma->anon_vma); 2037c791576cSYang Shi khugepaged_enter_vma(vma, vma->vm_flags); 2038d4af56c5SLiam R. Howlett mas_destroy(&mas); 20391da177e4SLinus Torvalds return error; 20401da177e4SLinus Torvalds } 20411da177e4SLinus Torvalds 20421be7107fSHugh Dickins /* enforced gap between the expanding stack and other mappings. */ 20431be7107fSHugh Dickins unsigned long stack_guard_gap = 256UL<<PAGE_SHIFT; 20441be7107fSHugh Dickins 20451be7107fSHugh Dickins static int __init cmdline_parse_stack_guard_gap(char *p) 20461be7107fSHugh Dickins { 20471be7107fSHugh Dickins unsigned long val; 20481be7107fSHugh Dickins char *endptr; 20491be7107fSHugh Dickins 20501be7107fSHugh Dickins val = simple_strtoul(p, &endptr, 10); 20511be7107fSHugh Dickins if (!*endptr) 20521be7107fSHugh Dickins stack_guard_gap = val << PAGE_SHIFT; 20531be7107fSHugh Dickins 2054e6d09493SRandy Dunlap return 1; 20551be7107fSHugh Dickins } 20561be7107fSHugh Dickins __setup("stack_guard_gap=", cmdline_parse_stack_guard_gap); 20571be7107fSHugh Dickins 2058b6a2fea3SOllie Wild #ifdef CONFIG_STACK_GROWSUP 2059b6a2fea3SOllie Wild int expand_stack(struct vm_area_struct *vma, unsigned long address) 2060b6a2fea3SOllie Wild { 2061b6a2fea3SOllie Wild return expand_upwards(vma, address); 2062b6a2fea3SOllie Wild } 2063b6a2fea3SOllie Wild 2064b6a2fea3SOllie Wild struct vm_area_struct * 2065b6a2fea3SOllie Wild find_extend_vma(struct mm_struct *mm, unsigned long addr) 2066b6a2fea3SOllie Wild { 2067b6a2fea3SOllie Wild struct vm_area_struct *vma, *prev; 2068b6a2fea3SOllie Wild 2069b6a2fea3SOllie Wild addr &= PAGE_MASK; 2070b6a2fea3SOllie Wild vma = find_vma_prev(mm, addr, &prev); 2071b6a2fea3SOllie Wild if (vma && (vma->vm_start <= addr)) 2072b6a2fea3SOllie Wild return vma; 20734d45e75aSJann Horn if (!prev || expand_stack(prev, addr)) 2074b6a2fea3SOllie Wild return NULL; 2075cea10a19SMichel Lespinasse if (prev->vm_flags & VM_LOCKED) 2076fc05f566SKirill A. Shutemov populate_vma_page_range(prev, addr, prev->vm_end, NULL); 2077b6a2fea3SOllie Wild return prev; 2078b6a2fea3SOllie Wild } 2079b6a2fea3SOllie Wild #else 2080b6a2fea3SOllie Wild int expand_stack(struct vm_area_struct *vma, unsigned long address) 2081b6a2fea3SOllie Wild { 2082b6a2fea3SOllie Wild return expand_downwards(vma, address); 2083b6a2fea3SOllie Wild } 2084b6a2fea3SOllie Wild 20851da177e4SLinus Torvalds struct vm_area_struct * 20861da177e4SLinus Torvalds find_extend_vma(struct mm_struct *mm, unsigned long addr) 20871da177e4SLinus Torvalds { 20881da177e4SLinus Torvalds struct vm_area_struct *vma; 20891da177e4SLinus Torvalds unsigned long start; 20901da177e4SLinus Torvalds 20911da177e4SLinus Torvalds addr &= PAGE_MASK; 20921da177e4SLinus Torvalds vma = find_vma(mm, addr); 20931da177e4SLinus Torvalds if (!vma) 20941da177e4SLinus Torvalds return NULL; 20951da177e4SLinus Torvalds if (vma->vm_start <= addr) 20961da177e4SLinus Torvalds return vma; 20971da177e4SLinus Torvalds if (!(vma->vm_flags & VM_GROWSDOWN)) 20981da177e4SLinus Torvalds return NULL; 20991da177e4SLinus Torvalds start = vma->vm_start; 21001da177e4SLinus Torvalds if (expand_stack(vma, addr)) 21011da177e4SLinus Torvalds return NULL; 2102cea10a19SMichel Lespinasse if (vma->vm_flags & VM_LOCKED) 2103fc05f566SKirill A. Shutemov populate_vma_page_range(vma, addr, start, NULL); 21041da177e4SLinus Torvalds return vma; 21051da177e4SLinus Torvalds } 21061da177e4SLinus Torvalds #endif 21071da177e4SLinus Torvalds 2108e1d6d01aSJesse Barnes EXPORT_SYMBOL_GPL(find_extend_vma); 2109e1d6d01aSJesse Barnes 21102c0b3814SHugh Dickins /* 2111763ecb03SLiam R. Howlett * Ok - we have the memory areas we should free on a maple tree so release them, 2112763ecb03SLiam R. Howlett * and do the vma updates. 21131da177e4SLinus Torvalds * 21142c0b3814SHugh Dickins * Called with the mm semaphore held. 21151da177e4SLinus Torvalds */ 2116763ecb03SLiam R. Howlett static inline void remove_mt(struct mm_struct *mm, struct ma_state *mas) 21171da177e4SLinus Torvalds { 21184f74d2c8SLinus Torvalds unsigned long nr_accounted = 0; 2119763ecb03SLiam R. Howlett struct vm_area_struct *vma; 21204f74d2c8SLinus Torvalds 2121365e9c87SHugh Dickins /* Update high watermark before we lower total_vm */ 2122365e9c87SHugh Dickins update_hiwater_vm(mm); 2123763ecb03SLiam R. Howlett mas_for_each(mas, vma, ULONG_MAX) { 2124ab50b8edSHugh Dickins long nrpages = vma_pages(vma); 21251da177e4SLinus Torvalds 21264f74d2c8SLinus Torvalds if (vma->vm_flags & VM_ACCOUNT) 21274f74d2c8SLinus Torvalds nr_accounted += nrpages; 212884638335SKonstantin Khlebnikov vm_stat_account(mm, vma->vm_flags, -nrpages); 2129763ecb03SLiam R. Howlett remove_vma(vma); 2130763ecb03SLiam R. Howlett } 21314f74d2c8SLinus Torvalds vm_unacct_memory(nr_accounted); 21321da177e4SLinus Torvalds validate_mm(mm); 21331da177e4SLinus Torvalds } 21341da177e4SLinus Torvalds 21351da177e4SLinus Torvalds /* 21361da177e4SLinus Torvalds * Get rid of page table information in the indicated region. 21371da177e4SLinus Torvalds * 2138f10df686SPaolo 'Blaisorblade' Giarrusso * Called with the mm semaphore held. 21391da177e4SLinus Torvalds */ 2140763ecb03SLiam R. Howlett static void unmap_region(struct mm_struct *mm, struct maple_tree *mt, 2141e0da382cSHugh Dickins struct vm_area_struct *vma, struct vm_area_struct *prev, 2142763ecb03SLiam R. Howlett struct vm_area_struct *next, 214368f48381SSuren Baghdasaryan unsigned long start, unsigned long end, bool mm_wr_locked) 21441da177e4SLinus Torvalds { 2145d16dfc55SPeter Zijlstra struct mmu_gather tlb; 21461da177e4SLinus Torvalds 21471da177e4SLinus Torvalds lru_add_drain(); 2148a72afd87SWill Deacon tlb_gather_mmu(&tlb, mm); 2149365e9c87SHugh Dickins update_hiwater_rss(mm); 215068f48381SSuren Baghdasaryan unmap_vmas(&tlb, mt, vma, start, end, mm_wr_locked); 2151763ecb03SLiam R. Howlett free_pgtables(&tlb, mt, vma, prev ? prev->vm_end : FIRST_USER_ADDRESS, 21526ee8630eSHugh Dickins next ? next->vm_start : USER_PGTABLES_CEILING); 2153ae8eba8bSWill Deacon tlb_finish_mmu(&tlb); 21541da177e4SLinus Torvalds } 21551da177e4SLinus Torvalds 21561da177e4SLinus Torvalds /* 2157def5efe0SDavid Rientjes * __split_vma() bypasses sysctl_max_map_count checking. We use this where it 2158def5efe0SDavid Rientjes * has already been checked or doesn't make sense to fail. 21590fd5a9e2SLiam R. Howlett * VMA Iterator will point to the end VMA. 21601da177e4SLinus Torvalds */ 21619760ebffSLiam R. Howlett int __split_vma(struct vma_iterator *vmi, struct vm_area_struct *vma, 21621da177e4SLinus Torvalds unsigned long addr, int new_below) 21631da177e4SLinus Torvalds { 2164b2b3b886SLiam R. Howlett struct vma_prepare vp; 21651da177e4SLinus Torvalds struct vm_area_struct *new; 2166e3975891SChen Gang int err; 21679760ebffSLiam R. Howlett 21689760ebffSLiam R. Howlett validate_mm_mt(vma->vm_mm); 21691da177e4SLinus Torvalds 2170b2b3b886SLiam R. Howlett WARN_ON(vma->vm_start >= addr); 2171b2b3b886SLiam R. Howlett WARN_ON(vma->vm_end <= addr); 2172b2b3b886SLiam R. Howlett 2173dd3b614fSDmitry Safonov if (vma->vm_ops && vma->vm_ops->may_split) { 2174dd3b614fSDmitry Safonov err = vma->vm_ops->may_split(vma, addr); 217531383c68SDan Williams if (err) 217631383c68SDan Williams return err; 217731383c68SDan Williams } 21781da177e4SLinus Torvalds 21793928d4f5SLinus Torvalds new = vm_area_dup(vma); 21801da177e4SLinus Torvalds if (!new) 2181e3975891SChen Gang return -ENOMEM; 21821da177e4SLinus Torvalds 2183b2b3b886SLiam R. Howlett err = -ENOMEM; 2184b2b3b886SLiam R. Howlett if (vma_iter_prealloc(vmi)) 2185b2b3b886SLiam R. Howlett goto out_free_vma; 2186b2b3b886SLiam R. Howlett 2187b2b3b886SLiam R. Howlett if (new_below) { 21881da177e4SLinus Torvalds new->vm_end = addr; 2189b2b3b886SLiam R. Howlett } else { 21901da177e4SLinus Torvalds new->vm_start = addr; 21911da177e4SLinus Torvalds new->vm_pgoff += ((addr - vma->vm_start) >> PAGE_SHIFT); 21921da177e4SLinus Torvalds } 21931da177e4SLinus Torvalds 2194ef0855d3SOleg Nesterov err = vma_dup_policy(vma, new); 2195ef0855d3SOleg Nesterov if (err) 2196b2b3b886SLiam R. Howlett goto out_free_vmi; 21971da177e4SLinus Torvalds 2198c4ea95d7SDaniel Forrest err = anon_vma_clone(new, vma); 2199c4ea95d7SDaniel Forrest if (err) 22005beb4930SRik van Riel goto out_free_mpol; 22015beb4930SRik van Riel 2202e9714acfSKonstantin Khlebnikov if (new->vm_file) 22031da177e4SLinus Torvalds get_file(new->vm_file); 22041da177e4SLinus Torvalds 22051da177e4SLinus Torvalds if (new->vm_ops && new->vm_ops->open) 22061da177e4SLinus Torvalds new->vm_ops->open(new); 22071da177e4SLinus Torvalds 2208b2b3b886SLiam R. Howlett vma_adjust_trans_huge(vma, vma->vm_start, addr, 0); 2209b2b3b886SLiam R. Howlett init_vma_prep(&vp, vma); 2210b2b3b886SLiam R. Howlett vp.insert = new; 2211b2b3b886SLiam R. Howlett vma_prepare(&vp); 22121da177e4SLinus Torvalds 2213b2b3b886SLiam R. Howlett if (new_below) { 2214b2b3b886SLiam R. Howlett vma->vm_start = addr; 2215b2b3b886SLiam R. Howlett vma->vm_pgoff += (addr - new->vm_start) >> PAGE_SHIFT; 2216b2b3b886SLiam R. Howlett } else { 2217b2b3b886SLiam R. Howlett vma->vm_end = addr; 22189760ebffSLiam R. Howlett } 22195beb4930SRik van Riel 2220b2b3b886SLiam R. Howlett /* vma_complete stores the new vma */ 2221b2b3b886SLiam R. Howlett vma_complete(&vp, vmi, vma->vm_mm); 2222b2b3b886SLiam R. Howlett 2223b2b3b886SLiam R. Howlett /* Success. */ 2224b2b3b886SLiam R. Howlett if (new_below) 2225b2b3b886SLiam R. Howlett vma_next(vmi); 2226b2b3b886SLiam R. Howlett validate_mm_mt(vma->vm_mm); 2227b2b3b886SLiam R. Howlett return 0; 2228b2b3b886SLiam R. Howlett 22295beb4930SRik van Riel out_free_mpol: 2230ef0855d3SOleg Nesterov mpol_put(vma_policy(new)); 2231b2b3b886SLiam R. Howlett out_free_vmi: 2232b2b3b886SLiam R. Howlett vma_iter_free(vmi); 22335beb4930SRik van Riel out_free_vma: 22343928d4f5SLinus Torvalds vm_area_free(new); 22359760ebffSLiam R. Howlett validate_mm_mt(vma->vm_mm); 22365beb4930SRik van Riel return err; 22371da177e4SLinus Torvalds } 22381da177e4SLinus Torvalds 2239659ace58SKOSAKI Motohiro /* 2240659ace58SKOSAKI Motohiro * Split a vma into two pieces at address 'addr', a new vma is allocated 2241659ace58SKOSAKI Motohiro * either for the first part or the tail. 2242659ace58SKOSAKI Motohiro */ 22439760ebffSLiam R. Howlett int split_vma(struct vma_iterator *vmi, struct vm_area_struct *vma, 2244659ace58SKOSAKI Motohiro unsigned long addr, int new_below) 2245659ace58SKOSAKI Motohiro { 22469760ebffSLiam R. Howlett if (vma->vm_mm->map_count >= sysctl_max_map_count) 2247659ace58SKOSAKI Motohiro return -ENOMEM; 2248659ace58SKOSAKI Motohiro 22499760ebffSLiam R. Howlett return __split_vma(vmi, vma, addr, new_below); 2250f2ebfe43SLiam R. Howlett } 2251f2ebfe43SLiam R. Howlett 2252763ecb03SLiam R. Howlett static inline int munmap_sidetree(struct vm_area_struct *vma, 2253763ecb03SLiam R. Howlett struct ma_state *mas_detach) 22541da177e4SLinus Torvalds { 2255763ecb03SLiam R. Howlett mas_set_range(mas_detach, vma->vm_start, vma->vm_end - 1); 2256763ecb03SLiam R. Howlett if (mas_store_gfp(mas_detach, vma, GFP_KERNEL)) 2257763ecb03SLiam R. Howlett return -ENOMEM; 225811f9a21aSLiam R. Howlett 2259763ecb03SLiam R. Howlett if (vma->vm_flags & VM_LOCKED) 2260763ecb03SLiam R. Howlett vma->vm_mm->locked_vm -= vma_pages(vma); 226111f9a21aSLiam R. Howlett 2262763ecb03SLiam R. Howlett return 0; 226311f9a21aSLiam R. Howlett } 226411f9a21aSLiam R. Howlett 226511f9a21aSLiam R. Howlett /* 2266183654ceSLiam R. Howlett * do_vmi_align_munmap() - munmap the aligned region from @start to @end. 2267183654ceSLiam R. Howlett * @vmi: The vma iterator 226811f9a21aSLiam R. Howlett * @vma: The starting vm_area_struct 226911f9a21aSLiam R. Howlett * @mm: The mm_struct 227011f9a21aSLiam R. Howlett * @start: The aligned start address to munmap. 227111f9a21aSLiam R. Howlett * @end: The aligned end address to munmap. 227211f9a21aSLiam R. Howlett * @uf: The userfaultfd list_head 22738651a137SLorenzo Stoakes * @downgrade: Set to true to attempt a write downgrade of the mmap_lock 227411f9a21aSLiam R. Howlett * 227511f9a21aSLiam R. Howlett * If @downgrade is true, check return code for potential release of the lock. 227611f9a21aSLiam R. Howlett */ 227711f9a21aSLiam R. Howlett static int 2278183654ceSLiam R. Howlett do_vmi_align_munmap(struct vma_iterator *vmi, struct vm_area_struct *vma, 227911f9a21aSLiam R. Howlett struct mm_struct *mm, unsigned long start, 228011f9a21aSLiam R. Howlett unsigned long end, struct list_head *uf, bool downgrade) 228111f9a21aSLiam R. Howlett { 2282763ecb03SLiam R. Howlett struct vm_area_struct *prev, *next = NULL; 2283763ecb03SLiam R. Howlett struct maple_tree mt_detach; 2284763ecb03SLiam R. Howlett int count = 0; 2285d4af56c5SLiam R. Howlett int error = -ENOMEM; 2286763ecb03SLiam R. Howlett MA_STATE(mas_detach, &mt_detach, 0, 0); 2287763ecb03SLiam R. Howlett mt_init_flags(&mt_detach, MT_FLAGS_LOCK_EXTERN); 2288763ecb03SLiam R. Howlett mt_set_external_lock(&mt_detach, &mm->mmap_lock); 2289524e00b3SLiam R. Howlett 22901da177e4SLinus Torvalds /* 22911da177e4SLinus Torvalds * If we need to split any vma, do it now to save pain later. 22921da177e4SLinus Torvalds * 22931da177e4SLinus Torvalds * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially 22941da177e4SLinus Torvalds * unmapped vm_area_struct will remain in use: so lower split_vma 22951da177e4SLinus Torvalds * places tmp vma above, and higher split_vma places tmp vma below. 22961da177e4SLinus Torvalds */ 2297763ecb03SLiam R. Howlett 2298763ecb03SLiam R. Howlett /* Does it split the first one? */ 2299146425a3SHugh Dickins if (start > vma->vm_start) { 2300659ace58SKOSAKI Motohiro 2301659ace58SKOSAKI Motohiro /* 2302659ace58SKOSAKI Motohiro * Make sure that map_count on return from munmap() will 2303659ace58SKOSAKI Motohiro * not exceed its limit; but let map_count go just above 2304659ace58SKOSAKI Motohiro * its limit temporarily, to help free resources as expected. 2305659ace58SKOSAKI Motohiro */ 2306659ace58SKOSAKI Motohiro if (end < vma->vm_end && mm->map_count >= sysctl_max_map_count) 2307d4af56c5SLiam R. Howlett goto map_count_exceeded; 2308659ace58SKOSAKI Motohiro 23099760ebffSLiam R. Howlett error = __split_vma(vmi, vma, start, 0); 23101da177e4SLinus Torvalds if (error) 2311763ecb03SLiam R. Howlett goto start_split_failed; 231211f9a21aSLiam R. Howlett 23130fd5a9e2SLiam R. Howlett vma = vma_iter_load(vmi); 23141da177e4SLinus Torvalds } 23151da177e4SLinus Torvalds 2316183654ceSLiam R. Howlett prev = vma_prev(vmi); 2317763ecb03SLiam R. Howlett if (unlikely((!prev))) 2318183654ceSLiam R. Howlett vma_iter_set(vmi, start); 231911f9a21aSLiam R. Howlett 2320763ecb03SLiam R. Howlett /* 2321763ecb03SLiam R. Howlett * Detach a range of VMAs from the mm. Using next as a temp variable as 2322763ecb03SLiam R. Howlett * it is always overwritten. 2323763ecb03SLiam R. Howlett */ 2324183654ceSLiam R. Howlett for_each_vma_range(*vmi, next, end) { 2325763ecb03SLiam R. Howlett /* Does it split the end? */ 2326763ecb03SLiam R. Howlett if (next->vm_end > end) { 23276b73cff2SLiam R. Howlett error = __split_vma(vmi, next, end, 0); 23281da177e4SLinus Torvalds if (error) 2329763ecb03SLiam R. Howlett goto end_split_failed; 233011f9a21aSLiam R. Howlett } 2331763ecb03SLiam R. Howlett error = munmap_sidetree(next, &mas_detach); 2332763ecb03SLiam R. Howlett if (error) 2333763ecb03SLiam R. Howlett goto munmap_sidetree_failed; 2334763ecb03SLiam R. Howlett 2335763ecb03SLiam R. Howlett count++; 2336763ecb03SLiam R. Howlett #ifdef CONFIG_DEBUG_VM_MAPLE_TREE 2337763ecb03SLiam R. Howlett BUG_ON(next->vm_start < start); 2338763ecb03SLiam R. Howlett BUG_ON(next->vm_start > end); 2339763ecb03SLiam R. Howlett #endif 2340763ecb03SLiam R. Howlett } 2341763ecb03SLiam R. Howlett 2342183654ceSLiam R. Howlett next = vma_next(vmi); 23432376dd7cSAndrea Arcangeli if (unlikely(uf)) { 23442376dd7cSAndrea Arcangeli /* 23452376dd7cSAndrea Arcangeli * If userfaultfd_unmap_prep returns an error the vmas 2346f0953a1bSIngo Molnar * will remain split, but userland will get a 23472376dd7cSAndrea Arcangeli * highly unexpected error anyway. This is no 23482376dd7cSAndrea Arcangeli * different than the case where the first of the two 23492376dd7cSAndrea Arcangeli * __split_vma fails, but we don't undo the first 23502376dd7cSAndrea Arcangeli * split, despite we could. This is unlikely enough 23512376dd7cSAndrea Arcangeli * failure that it's not worth optimizing it for. 23522376dd7cSAndrea Arcangeli */ 235369dbe6daSLiam R. Howlett error = userfaultfd_unmap_prep(mm, start, end, uf); 235411f9a21aSLiam R. Howlett 23552376dd7cSAndrea Arcangeli if (error) 2356d4af56c5SLiam R. Howlett goto userfaultfd_error; 23572376dd7cSAndrea Arcangeli } 23582376dd7cSAndrea Arcangeli 2359763ecb03SLiam R. Howlett #if defined(CONFIG_DEBUG_VM_MAPLE_TREE) 2360763ecb03SLiam R. Howlett /* Make sure no VMAs are about to be lost. */ 2361763ecb03SLiam R. Howlett { 2362763ecb03SLiam R. Howlett MA_STATE(test, &mt_detach, start, end - 1); 2363763ecb03SLiam R. Howlett struct vm_area_struct *vma_mas, *vma_test; 2364763ecb03SLiam R. Howlett int test_count = 0; 2365763ecb03SLiam R. Howlett 2366183654ceSLiam R. Howlett vma_iter_set(vmi, start); 2367763ecb03SLiam R. Howlett rcu_read_lock(); 2368763ecb03SLiam R. Howlett vma_test = mas_find(&test, end - 1); 2369183654ceSLiam R. Howlett for_each_vma_range(*vmi, vma_mas, end) { 2370763ecb03SLiam R. Howlett BUG_ON(vma_mas != vma_test); 2371763ecb03SLiam R. Howlett test_count++; 2372763ecb03SLiam R. Howlett vma_test = mas_next(&test, end - 1); 2373763ecb03SLiam R. Howlett } 2374763ecb03SLiam R. Howlett rcu_read_unlock(); 2375763ecb03SLiam R. Howlett BUG_ON(count != test_count); 2376763ecb03SLiam R. Howlett } 2377763ecb03SLiam R. Howlett #endif 23780378c0a0SLiam R. Howlett /* Point of no return */ 2379183654ceSLiam R. Howlett vma_iter_set(vmi, start); 2380183654ceSLiam R. Howlett if (vma_iter_clear_gfp(vmi, start, end, GFP_KERNEL)) 23810378c0a0SLiam R. Howlett return -ENOMEM; 23820378c0a0SLiam R. Howlett 2383763ecb03SLiam R. Howlett mm->map_count -= count; 238411f9a21aSLiam R. Howlett /* 238511f9a21aSLiam R. Howlett * Do not downgrade mmap_lock if we are next to VM_GROWSDOWN or 238611f9a21aSLiam R. Howlett * VM_GROWSUP VMA. Such VMAs can change their size under 238711f9a21aSLiam R. Howlett * down_read(mmap_lock) and collide with the VMA we are about to unmap. 238811f9a21aSLiam R. Howlett */ 238911f9a21aSLiam R. Howlett if (downgrade) { 2390763ecb03SLiam R. Howlett if (next && (next->vm_flags & VM_GROWSDOWN)) 239111f9a21aSLiam R. Howlett downgrade = false; 239211f9a21aSLiam R. Howlett else if (prev && (prev->vm_flags & VM_GROWSUP)) 239311f9a21aSLiam R. Howlett downgrade = false; 239411f9a21aSLiam R. Howlett else 2395d8ed45c5SMichel Lespinasse mmap_write_downgrade(mm); 239611f9a21aSLiam R. Howlett } 2397dd2283f2SYang Shi 239868f48381SSuren Baghdasaryan /* 239968f48381SSuren Baghdasaryan * We can free page tables without write-locking mmap_lock because VMAs 240068f48381SSuren Baghdasaryan * were isolated before we downgraded mmap_lock. 240168f48381SSuren Baghdasaryan */ 240268f48381SSuren Baghdasaryan unmap_region(mm, &mt_detach, vma, prev, next, start, end, !downgrade); 2403763ecb03SLiam R. Howlett /* Statistics and freeing VMAs */ 2404763ecb03SLiam R. Howlett mas_set(&mas_detach, start); 2405763ecb03SLiam R. Howlett remove_mt(mm, &mas_detach); 2406763ecb03SLiam R. Howlett __mt_destroy(&mt_detach); 24071da177e4SLinus Torvalds 2408524e00b3SLiam R. Howlett 2409524e00b3SLiam R. Howlett validate_mm(mm); 2410dd2283f2SYang Shi return downgrade ? 1 : 0; 2411d4af56c5SLiam R. Howlett 2412d4af56c5SLiam R. Howlett userfaultfd_error: 2413763ecb03SLiam R. Howlett munmap_sidetree_failed: 2414763ecb03SLiam R. Howlett end_split_failed: 2415763ecb03SLiam R. Howlett __mt_destroy(&mt_detach); 2416763ecb03SLiam R. Howlett start_split_failed: 2417763ecb03SLiam R. Howlett map_count_exceeded: 2418d4af56c5SLiam R. Howlett return error; 24191da177e4SLinus Torvalds } 24201da177e4SLinus Torvalds 242111f9a21aSLiam R. Howlett /* 2422183654ceSLiam R. Howlett * do_vmi_munmap() - munmap a given range. 2423183654ceSLiam R. Howlett * @vmi: The vma iterator 242411f9a21aSLiam R. Howlett * @mm: The mm_struct 242511f9a21aSLiam R. Howlett * @start: The start address to munmap 242611f9a21aSLiam R. Howlett * @len: The length of the range to munmap 242711f9a21aSLiam R. Howlett * @uf: The userfaultfd list_head 242811f9a21aSLiam R. Howlett * @downgrade: set to true if the user wants to attempt to write_downgrade the 24298651a137SLorenzo Stoakes * mmap_lock 243011f9a21aSLiam R. Howlett * 243111f9a21aSLiam R. Howlett * This function takes a @mas that is either pointing to the previous VMA or set 243211f9a21aSLiam R. Howlett * to MA_START and sets it up to remove the mapping(s). The @len will be 243311f9a21aSLiam R. Howlett * aligned and any arch_unmap work will be preformed. 243411f9a21aSLiam R. Howlett * 243511f9a21aSLiam R. Howlett * Returns: -EINVAL on failure, 1 on success and unlock, 0 otherwise. 243611f9a21aSLiam R. Howlett */ 2437183654ceSLiam R. Howlett int do_vmi_munmap(struct vma_iterator *vmi, struct mm_struct *mm, 243811f9a21aSLiam R. Howlett unsigned long start, size_t len, struct list_head *uf, 243911f9a21aSLiam R. Howlett bool downgrade) 244011f9a21aSLiam R. Howlett { 244111f9a21aSLiam R. Howlett unsigned long end; 244211f9a21aSLiam R. Howlett struct vm_area_struct *vma; 244311f9a21aSLiam R. Howlett 244411f9a21aSLiam R. Howlett if ((offset_in_page(start)) || start > TASK_SIZE || len > TASK_SIZE-start) 244511f9a21aSLiam R. Howlett return -EINVAL; 244611f9a21aSLiam R. Howlett 244711f9a21aSLiam R. Howlett end = start + PAGE_ALIGN(len); 244811f9a21aSLiam R. Howlett if (end == start) 244911f9a21aSLiam R. Howlett return -EINVAL; 245011f9a21aSLiam R. Howlett 245111f9a21aSLiam R. Howlett /* arch_unmap() might do unmaps itself. */ 245211f9a21aSLiam R. Howlett arch_unmap(mm, start, end); 245311f9a21aSLiam R. Howlett 245411f9a21aSLiam R. Howlett /* Find the first overlapping VMA */ 2455183654ceSLiam R. Howlett vma = vma_find(vmi, end); 245611f9a21aSLiam R. Howlett if (!vma) 245711f9a21aSLiam R. Howlett return 0; 245811f9a21aSLiam R. Howlett 2459183654ceSLiam R. Howlett return do_vmi_align_munmap(vmi, vma, mm, start, end, uf, downgrade); 246011f9a21aSLiam R. Howlett } 246111f9a21aSLiam R. Howlett 246211f9a21aSLiam R. Howlett /* do_munmap() - Wrapper function for non-maple tree aware do_munmap() calls. 246311f9a21aSLiam R. Howlett * @mm: The mm_struct 246411f9a21aSLiam R. Howlett * @start: The start address to munmap 246511f9a21aSLiam R. Howlett * @len: The length to be munmapped. 246611f9a21aSLiam R. Howlett * @uf: The userfaultfd list_head 246711f9a21aSLiam R. Howlett */ 2468dd2283f2SYang Shi int do_munmap(struct mm_struct *mm, unsigned long start, size_t len, 2469dd2283f2SYang Shi struct list_head *uf) 2470dd2283f2SYang Shi { 2471183654ceSLiam R. Howlett VMA_ITERATOR(vmi, mm, start); 247211f9a21aSLiam R. Howlett 2473183654ceSLiam R. Howlett return do_vmi_munmap(&vmi, mm, start, len, uf, false); 2474dd2283f2SYang Shi } 2475dd2283f2SYang Shi 2476e99668a5SLiam R. Howlett unsigned long mmap_region(struct file *file, unsigned long addr, 2477e99668a5SLiam R. Howlett unsigned long len, vm_flags_t vm_flags, unsigned long pgoff, 2478e99668a5SLiam R. Howlett struct list_head *uf) 2479e99668a5SLiam R. Howlett { 2480e99668a5SLiam R. Howlett struct mm_struct *mm = current->mm; 2481e99668a5SLiam R. Howlett struct vm_area_struct *vma = NULL; 2482e99668a5SLiam R. Howlett struct vm_area_struct *next, *prev, *merge; 2483e99668a5SLiam R. Howlett pgoff_t pglen = len >> PAGE_SHIFT; 2484e99668a5SLiam R. Howlett unsigned long charged = 0; 2485e99668a5SLiam R. Howlett unsigned long end = addr + len; 2486e99668a5SLiam R. Howlett unsigned long merge_start = addr, merge_end = end; 2487e99668a5SLiam R. Howlett pgoff_t vm_pgoff; 2488e99668a5SLiam R. Howlett int error; 2489183654ceSLiam R. Howlett VMA_ITERATOR(vmi, mm, addr); 2490e99668a5SLiam R. Howlett 2491e99668a5SLiam R. Howlett /* Check against address space limit. */ 2492e99668a5SLiam R. Howlett if (!may_expand_vm(mm, vm_flags, len >> PAGE_SHIFT)) { 2493e99668a5SLiam R. Howlett unsigned long nr_pages; 2494e99668a5SLiam R. Howlett 2495e99668a5SLiam R. Howlett /* 2496e99668a5SLiam R. Howlett * MAP_FIXED may remove pages of mappings that intersects with 2497e99668a5SLiam R. Howlett * requested mapping. Account for the pages it would unmap. 2498e99668a5SLiam R. Howlett */ 2499e99668a5SLiam R. Howlett nr_pages = count_vma_pages_range(mm, addr, end); 2500e99668a5SLiam R. Howlett 2501e99668a5SLiam R. Howlett if (!may_expand_vm(mm, vm_flags, 2502e99668a5SLiam R. Howlett (len >> PAGE_SHIFT) - nr_pages)) 2503e99668a5SLiam R. Howlett return -ENOMEM; 2504e99668a5SLiam R. Howlett } 2505e99668a5SLiam R. Howlett 2506e99668a5SLiam R. Howlett /* Unmap any existing mapping in the area */ 2507183654ceSLiam R. Howlett if (do_vmi_munmap(&vmi, mm, addr, len, uf, false)) 2508e99668a5SLiam R. Howlett return -ENOMEM; 2509e99668a5SLiam R. Howlett 2510e99668a5SLiam R. Howlett /* 2511e99668a5SLiam R. Howlett * Private writable mapping: check memory availability 2512e99668a5SLiam R. Howlett */ 2513e99668a5SLiam R. Howlett if (accountable_mapping(file, vm_flags)) { 2514e99668a5SLiam R. Howlett charged = len >> PAGE_SHIFT; 2515e99668a5SLiam R. Howlett if (security_vm_enough_memory_mm(mm, charged)) 2516e99668a5SLiam R. Howlett return -ENOMEM; 2517e99668a5SLiam R. Howlett vm_flags |= VM_ACCOUNT; 2518e99668a5SLiam R. Howlett } 2519e99668a5SLiam R. Howlett 2520183654ceSLiam R. Howlett next = vma_next(&vmi); 2521183654ceSLiam R. Howlett prev = vma_prev(&vmi); 2522e99668a5SLiam R. Howlett if (vm_flags & VM_SPECIAL) 2523e99668a5SLiam R. Howlett goto cannot_expand; 2524e99668a5SLiam R. Howlett 2525e99668a5SLiam R. Howlett /* Attempt to expand an old mapping */ 2526e99668a5SLiam R. Howlett /* Check next */ 2527e99668a5SLiam R. Howlett if (next && next->vm_start == end && !vma_policy(next) && 2528e99668a5SLiam R. Howlett can_vma_merge_before(next, vm_flags, NULL, file, pgoff+pglen, 2529e99668a5SLiam R. Howlett NULL_VM_UFFD_CTX, NULL)) { 2530e99668a5SLiam R. Howlett merge_end = next->vm_end; 2531e99668a5SLiam R. Howlett vma = next; 2532e99668a5SLiam R. Howlett vm_pgoff = next->vm_pgoff - pglen; 2533e99668a5SLiam R. Howlett } 2534e99668a5SLiam R. Howlett 2535e99668a5SLiam R. Howlett /* Check prev */ 2536e99668a5SLiam R. Howlett if (prev && prev->vm_end == addr && !vma_policy(prev) && 2537e99668a5SLiam R. Howlett (vma ? can_vma_merge_after(prev, vm_flags, vma->anon_vma, file, 2538e99668a5SLiam R. Howlett pgoff, vma->vm_userfaultfd_ctx, NULL) : 2539e99668a5SLiam R. Howlett can_vma_merge_after(prev, vm_flags, NULL, file, pgoff, 2540e99668a5SLiam R. Howlett NULL_VM_UFFD_CTX, NULL))) { 2541e99668a5SLiam R. Howlett merge_start = prev->vm_start; 2542e99668a5SLiam R. Howlett vma = prev; 2543e99668a5SLiam R. Howlett vm_pgoff = prev->vm_pgoff; 2544e99668a5SLiam R. Howlett } 2545e99668a5SLiam R. Howlett 2546e99668a5SLiam R. Howlett 2547e99668a5SLiam R. Howlett /* Actually expand, if possible */ 2548e99668a5SLiam R. Howlett if (vma && 25493c441ab7SLiam R. Howlett !vma_expand(&vmi, vma, merge_start, merge_end, vm_pgoff, next)) { 2550e99668a5SLiam R. Howlett khugepaged_enter_vma(vma, vm_flags); 2551e99668a5SLiam R. Howlett goto expanded; 2552e99668a5SLiam R. Howlett } 2553e99668a5SLiam R. Howlett 2554e99668a5SLiam R. Howlett cannot_expand: 2555e99668a5SLiam R. Howlett /* 2556e99668a5SLiam R. Howlett * Determine the object being mapped and call the appropriate 2557e99668a5SLiam R. Howlett * specific mapper. the address has already been validated, but 2558e99668a5SLiam R. Howlett * not unmapped, but the maps are removed from the list. 2559e99668a5SLiam R. Howlett */ 2560e99668a5SLiam R. Howlett vma = vm_area_alloc(mm); 2561e99668a5SLiam R. Howlett if (!vma) { 2562e99668a5SLiam R. Howlett error = -ENOMEM; 2563e99668a5SLiam R. Howlett goto unacct_error; 2564e99668a5SLiam R. Howlett } 2565e99668a5SLiam R. Howlett 25660fd5a9e2SLiam R. Howlett vma_iter_set(&vmi, addr); 2567e99668a5SLiam R. Howlett vma->vm_start = addr; 2568e99668a5SLiam R. Howlett vma->vm_end = end; 25691c71222eSSuren Baghdasaryan vm_flags_init(vma, vm_flags); 2570e99668a5SLiam R. Howlett vma->vm_page_prot = vm_get_page_prot(vm_flags); 2571e99668a5SLiam R. Howlett vma->vm_pgoff = pgoff; 2572e99668a5SLiam R. Howlett 2573e99668a5SLiam R. Howlett if (file) { 2574e99668a5SLiam R. Howlett if (vm_flags & VM_SHARED) { 2575e99668a5SLiam R. Howlett error = mapping_map_writable(file->f_mapping); 2576e99668a5SLiam R. Howlett if (error) 2577e99668a5SLiam R. Howlett goto free_vma; 2578e99668a5SLiam R. Howlett } 2579e99668a5SLiam R. Howlett 2580e99668a5SLiam R. Howlett vma->vm_file = get_file(file); 2581e99668a5SLiam R. Howlett error = call_mmap(file, vma); 2582e99668a5SLiam R. Howlett if (error) 2583e99668a5SLiam R. Howlett goto unmap_and_free_vma; 2584e99668a5SLiam R. Howlett 2585a57b7051SLiam Howlett /* 2586a57b7051SLiam Howlett * Expansion is handled above, merging is handled below. 2587a57b7051SLiam Howlett * Drivers should not alter the address of the VMA. 2588e99668a5SLiam R. Howlett */ 2589a57b7051SLiam Howlett error = -EINVAL; 2590cc8d1b09SLiam R. Howlett if (WARN_ON((addr != vma->vm_start))) 2591a57b7051SLiam Howlett goto close_and_free_vma; 2592e99668a5SLiam R. Howlett 2593cc8d1b09SLiam R. Howlett vma_iter_set(&vmi, addr); 2594e99668a5SLiam R. Howlett /* 2595e99668a5SLiam R. Howlett * If vm_flags changed after call_mmap(), we should try merge 2596e99668a5SLiam R. Howlett * vma again as we may succeed this time. 2597e99668a5SLiam R. Howlett */ 2598e99668a5SLiam R. Howlett if (unlikely(vm_flags != vma->vm_flags && prev)) { 25999760ebffSLiam R. Howlett merge = vma_merge(&vmi, mm, prev, vma->vm_start, 26009760ebffSLiam R. Howlett vma->vm_end, vma->vm_flags, NULL, 26019760ebffSLiam R. Howlett vma->vm_file, vma->vm_pgoff, NULL, 26029760ebffSLiam R. Howlett NULL_VM_UFFD_CTX, NULL); 2603e99668a5SLiam R. Howlett if (merge) { 2604e99668a5SLiam R. Howlett /* 2605e99668a5SLiam R. Howlett * ->mmap() can change vma->vm_file and fput 2606e99668a5SLiam R. Howlett * the original file. So fput the vma->vm_file 2607e99668a5SLiam R. Howlett * here or we would add an extra fput for file 2608e99668a5SLiam R. Howlett * and cause general protection fault 2609e99668a5SLiam R. Howlett * ultimately. 2610e99668a5SLiam R. Howlett */ 2611e99668a5SLiam R. Howlett fput(vma->vm_file); 2612e99668a5SLiam R. Howlett vm_area_free(vma); 2613e99668a5SLiam R. Howlett vma = merge; 2614e99668a5SLiam R. Howlett /* Update vm_flags to pick up the change. */ 2615e99668a5SLiam R. Howlett vm_flags = vma->vm_flags; 2616e99668a5SLiam R. Howlett goto unmap_writable; 2617e99668a5SLiam R. Howlett } 2618e99668a5SLiam R. Howlett } 2619e99668a5SLiam R. Howlett 2620e99668a5SLiam R. Howlett vm_flags = vma->vm_flags; 2621e99668a5SLiam R. Howlett } else if (vm_flags & VM_SHARED) { 2622e99668a5SLiam R. Howlett error = shmem_zero_setup(vma); 2623e99668a5SLiam R. Howlett if (error) 2624e99668a5SLiam R. Howlett goto free_vma; 2625e99668a5SLiam R. Howlett } else { 2626e99668a5SLiam R. Howlett vma_set_anonymous(vma); 2627e99668a5SLiam R. Howlett } 2628e99668a5SLiam R. Howlett 2629b507808eSJoey Gouly if (map_deny_write_exec(vma, vma->vm_flags)) { 2630b507808eSJoey Gouly error = -EACCES; 2631b507808eSJoey Gouly goto close_and_free_vma; 2632b507808eSJoey Gouly } 2633b507808eSJoey Gouly 2634e99668a5SLiam R. Howlett /* Allow architectures to sanity-check the vm_flags */ 2635e99668a5SLiam R. Howlett error = -EINVAL; 2636cc8d1b09SLiam R. Howlett if (!arch_validate_flags(vma->vm_flags)) 2637deb0f656SCarlos Llamas goto close_and_free_vma; 2638e99668a5SLiam R. Howlett 2639e99668a5SLiam R. Howlett error = -ENOMEM; 2640cc8d1b09SLiam R. Howlett if (vma_iter_prealloc(&vmi)) 26415789151eSMike Kravetz goto close_and_free_vma; 2642e99668a5SLiam R. Howlett 2643e99668a5SLiam R. Howlett if (vma->vm_file) 2644e99668a5SLiam R. Howlett i_mmap_lock_write(vma->vm_file->f_mapping); 2645e99668a5SLiam R. Howlett 2646183654ceSLiam R. Howlett vma_iter_store(&vmi, vma); 2647e99668a5SLiam R. Howlett mm->map_count++; 2648e99668a5SLiam R. Howlett if (vma->vm_file) { 2649e99668a5SLiam R. Howlett if (vma->vm_flags & VM_SHARED) 2650e99668a5SLiam R. Howlett mapping_allow_writable(vma->vm_file->f_mapping); 2651e99668a5SLiam R. Howlett 2652e99668a5SLiam R. Howlett flush_dcache_mmap_lock(vma->vm_file->f_mapping); 2653e99668a5SLiam R. Howlett vma_interval_tree_insert(vma, &vma->vm_file->f_mapping->i_mmap); 2654e99668a5SLiam R. Howlett flush_dcache_mmap_unlock(vma->vm_file->f_mapping); 2655e99668a5SLiam R. Howlett i_mmap_unlock_write(vma->vm_file->f_mapping); 2656e99668a5SLiam R. Howlett } 2657e99668a5SLiam R. Howlett 2658e99668a5SLiam R. Howlett /* 2659e99668a5SLiam R. Howlett * vma_merge() calls khugepaged_enter_vma() either, the below 2660e99668a5SLiam R. Howlett * call covers the non-merge case. 2661e99668a5SLiam R. Howlett */ 2662e99668a5SLiam R. Howlett khugepaged_enter_vma(vma, vma->vm_flags); 2663e99668a5SLiam R. Howlett 2664e99668a5SLiam R. Howlett /* Once vma denies write, undo our temporary denial count */ 2665e99668a5SLiam R. Howlett unmap_writable: 2666e99668a5SLiam R. Howlett if (file && vm_flags & VM_SHARED) 2667e99668a5SLiam R. Howlett mapping_unmap_writable(file->f_mapping); 2668e99668a5SLiam R. Howlett file = vma->vm_file; 2669e99668a5SLiam R. Howlett expanded: 2670e99668a5SLiam R. Howlett perf_event_mmap(vma); 2671e99668a5SLiam R. Howlett 2672e99668a5SLiam R. Howlett vm_stat_account(mm, vm_flags, len >> PAGE_SHIFT); 2673e99668a5SLiam R. Howlett if (vm_flags & VM_LOCKED) { 2674e99668a5SLiam R. Howlett if ((vm_flags & VM_SPECIAL) || vma_is_dax(vma) || 2675e99668a5SLiam R. Howlett is_vm_hugetlb_page(vma) || 2676e99668a5SLiam R. Howlett vma == get_gate_vma(current->mm)) 2677e430a95aSSuren Baghdasaryan vm_flags_clear(vma, VM_LOCKED_MASK); 2678e99668a5SLiam R. Howlett else 2679e99668a5SLiam R. Howlett mm->locked_vm += (len >> PAGE_SHIFT); 2680e99668a5SLiam R. Howlett } 2681e99668a5SLiam R. Howlett 2682e99668a5SLiam R. Howlett if (file) 2683e99668a5SLiam R. Howlett uprobe_mmap(vma); 2684e99668a5SLiam R. Howlett 2685e99668a5SLiam R. Howlett /* 2686e99668a5SLiam R. Howlett * New (or expanded) vma always get soft dirty status. 2687e99668a5SLiam R. Howlett * Otherwise user-space soft-dirty page tracker won't 2688e99668a5SLiam R. Howlett * be able to distinguish situation when vma area unmapped, 2689e99668a5SLiam R. Howlett * then new mapped in-place (which must be aimed as 2690e99668a5SLiam R. Howlett * a completely new data area). 2691e99668a5SLiam R. Howlett */ 26921c71222eSSuren Baghdasaryan vm_flags_set(vma, VM_SOFTDIRTY); 2693e99668a5SLiam R. Howlett 2694e99668a5SLiam R. Howlett vma_set_page_prot(vma); 2695e99668a5SLiam R. Howlett 2696e99668a5SLiam R. Howlett validate_mm(mm); 2697e99668a5SLiam R. Howlett return addr; 2698e99668a5SLiam R. Howlett 2699deb0f656SCarlos Llamas close_and_free_vma: 2700cc8d1b09SLiam R. Howlett if (file && vma->vm_ops && vma->vm_ops->close) 2701deb0f656SCarlos Llamas vma->vm_ops->close(vma); 2702cc8d1b09SLiam R. Howlett 2703cc8d1b09SLiam R. Howlett if (file || vma->vm_file) { 2704e99668a5SLiam R. Howlett unmap_and_free_vma: 2705e99668a5SLiam R. Howlett fput(vma->vm_file); 2706e99668a5SLiam R. Howlett vma->vm_file = NULL; 2707e99668a5SLiam R. Howlett 2708e99668a5SLiam R. Howlett /* Undo any partial mapping done by a device driver. */ 2709cc8d1b09SLiam R. Howlett unmap_region(mm, &mm->mm_mt, vma, prev, next, vma->vm_start, 271068f48381SSuren Baghdasaryan vma->vm_end, true); 2711cc8d1b09SLiam R. Howlett } 2712cc674ab3SLi Zetao if (file && (vm_flags & VM_SHARED)) 2713e99668a5SLiam R. Howlett mapping_unmap_writable(file->f_mapping); 2714e99668a5SLiam R. Howlett free_vma: 2715e99668a5SLiam R. Howlett vm_area_free(vma); 2716e99668a5SLiam R. Howlett unacct_error: 2717e99668a5SLiam R. Howlett if (charged) 2718e99668a5SLiam R. Howlett vm_unacct_memory(charged); 2719e99668a5SLiam R. Howlett validate_mm(mm); 2720e99668a5SLiam R. Howlett return error; 2721e99668a5SLiam R. Howlett } 2722e99668a5SLiam R. Howlett 2723dd2283f2SYang Shi static int __vm_munmap(unsigned long start, size_t len, bool downgrade) 2724a46ef99dSLinus Torvalds { 2725a46ef99dSLinus Torvalds int ret; 2726bfce281cSAl Viro struct mm_struct *mm = current->mm; 2727897ab3e0SMike Rapoport LIST_HEAD(uf); 2728183654ceSLiam R. Howlett VMA_ITERATOR(vmi, mm, start); 2729a46ef99dSLinus Torvalds 2730d8ed45c5SMichel Lespinasse if (mmap_write_lock_killable(mm)) 2731ae798783SMichal Hocko return -EINTR; 2732ae798783SMichal Hocko 2733183654ceSLiam R. Howlett ret = do_vmi_munmap(&vmi, mm, start, len, &uf, downgrade); 2734dd2283f2SYang Shi /* 2735c1e8d7c6SMichel Lespinasse * Returning 1 indicates mmap_lock is downgraded. 2736dd2283f2SYang Shi * But 1 is not legal return value of vm_munmap() and munmap(), reset 2737dd2283f2SYang Shi * it to 0 before return. 2738dd2283f2SYang Shi */ 2739dd2283f2SYang Shi if (ret == 1) { 2740d8ed45c5SMichel Lespinasse mmap_read_unlock(mm); 2741dd2283f2SYang Shi ret = 0; 2742dd2283f2SYang Shi } else 2743d8ed45c5SMichel Lespinasse mmap_write_unlock(mm); 2744dd2283f2SYang Shi 2745897ab3e0SMike Rapoport userfaultfd_unmap_complete(mm, &uf); 2746a46ef99dSLinus Torvalds return ret; 2747a46ef99dSLinus Torvalds } 2748dd2283f2SYang Shi 2749dd2283f2SYang Shi int vm_munmap(unsigned long start, size_t len) 2750dd2283f2SYang Shi { 2751dd2283f2SYang Shi return __vm_munmap(start, len, false); 2752dd2283f2SYang Shi } 2753a46ef99dSLinus Torvalds EXPORT_SYMBOL(vm_munmap); 2754a46ef99dSLinus Torvalds 27556a6160a7SHeiko Carstens SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len) 27561da177e4SLinus Torvalds { 2757ce18d171SCatalin Marinas addr = untagged_addr(addr); 2758dd2283f2SYang Shi return __vm_munmap(addr, len, true); 27591da177e4SLinus Torvalds } 27601da177e4SLinus Torvalds 2761c8d78c18SKirill A. Shutemov 2762c8d78c18SKirill A. Shutemov /* 2763c8d78c18SKirill A. Shutemov * Emulation of deprecated remap_file_pages() syscall. 2764c8d78c18SKirill A. Shutemov */ 2765c8d78c18SKirill A. Shutemov SYSCALL_DEFINE5(remap_file_pages, unsigned long, start, unsigned long, size, 2766c8d78c18SKirill A. Shutemov unsigned long, prot, unsigned long, pgoff, unsigned long, flags) 2767c8d78c18SKirill A. Shutemov { 2768c8d78c18SKirill A. Shutemov 2769c8d78c18SKirill A. Shutemov struct mm_struct *mm = current->mm; 2770c8d78c18SKirill A. Shutemov struct vm_area_struct *vma; 2771c8d78c18SKirill A. Shutemov unsigned long populate = 0; 2772c8d78c18SKirill A. Shutemov unsigned long ret = -EINVAL; 2773c8d78c18SKirill A. Shutemov struct file *file; 2774c8d78c18SKirill A. Shutemov 2775ee65728eSMike Rapoport pr_warn_once("%s (%d) uses deprecated remap_file_pages() syscall. See Documentation/mm/remap_file_pages.rst.\n", 2776c8d78c18SKirill A. Shutemov current->comm, current->pid); 2777c8d78c18SKirill A. Shutemov 2778c8d78c18SKirill A. Shutemov if (prot) 2779c8d78c18SKirill A. Shutemov return ret; 2780c8d78c18SKirill A. Shutemov start = start & PAGE_MASK; 2781c8d78c18SKirill A. Shutemov size = size & PAGE_MASK; 2782c8d78c18SKirill A. Shutemov 2783c8d78c18SKirill A. Shutemov if (start + size <= start) 2784c8d78c18SKirill A. Shutemov return ret; 2785c8d78c18SKirill A. Shutemov 2786c8d78c18SKirill A. Shutemov /* Does pgoff wrap? */ 2787c8d78c18SKirill A. Shutemov if (pgoff + (size >> PAGE_SHIFT) < pgoff) 2788c8d78c18SKirill A. Shutemov return ret; 2789c8d78c18SKirill A. Shutemov 2790d8ed45c5SMichel Lespinasse if (mmap_write_lock_killable(mm)) 2791dc0ef0dfSMichal Hocko return -EINTR; 2792dc0ef0dfSMichal Hocko 27939b593cb2SLiam R. Howlett vma = vma_lookup(mm, start); 2794c8d78c18SKirill A. Shutemov 2795c8d78c18SKirill A. Shutemov if (!vma || !(vma->vm_flags & VM_SHARED)) 2796c8d78c18SKirill A. Shutemov goto out; 2797c8d78c18SKirill A. Shutemov 279848f7df32SKirill A. Shutemov if (start + size > vma->vm_end) { 2799763ecb03SLiam R. Howlett VMA_ITERATOR(vmi, mm, vma->vm_end); 2800763ecb03SLiam R. Howlett struct vm_area_struct *next, *prev = vma; 280148f7df32SKirill A. Shutemov 2802763ecb03SLiam R. Howlett for_each_vma_range(vmi, next, start + size) { 280348f7df32SKirill A. Shutemov /* hole between vmas ? */ 2804763ecb03SLiam R. Howlett if (next->vm_start != prev->vm_end) 280548f7df32SKirill A. Shutemov goto out; 280648f7df32SKirill A. Shutemov 280748f7df32SKirill A. Shutemov if (next->vm_file != vma->vm_file) 280848f7df32SKirill A. Shutemov goto out; 280948f7df32SKirill A. Shutemov 281048f7df32SKirill A. Shutemov if (next->vm_flags != vma->vm_flags) 281148f7df32SKirill A. Shutemov goto out; 281248f7df32SKirill A. Shutemov 28131db43d3fSLiam Howlett if (start + size <= next->vm_end) 28141db43d3fSLiam Howlett break; 28151db43d3fSLiam Howlett 2816763ecb03SLiam R. Howlett prev = next; 281748f7df32SKirill A. Shutemov } 281848f7df32SKirill A. Shutemov 281948f7df32SKirill A. Shutemov if (!next) 2820c8d78c18SKirill A. Shutemov goto out; 2821c8d78c18SKirill A. Shutemov } 2822c8d78c18SKirill A. Shutemov 2823c8d78c18SKirill A. Shutemov prot |= vma->vm_flags & VM_READ ? PROT_READ : 0; 2824c8d78c18SKirill A. Shutemov prot |= vma->vm_flags & VM_WRITE ? PROT_WRITE : 0; 2825c8d78c18SKirill A. Shutemov prot |= vma->vm_flags & VM_EXEC ? PROT_EXEC : 0; 2826c8d78c18SKirill A. Shutemov 2827c8d78c18SKirill A. Shutemov flags &= MAP_NONBLOCK; 2828c8d78c18SKirill A. Shutemov flags |= MAP_SHARED | MAP_FIXED | MAP_POPULATE; 2829fce000b1SLiam Howlett if (vma->vm_flags & VM_LOCKED) 2830c8d78c18SKirill A. Shutemov flags |= MAP_LOCKED; 283148f7df32SKirill A. Shutemov 2832c8d78c18SKirill A. Shutemov file = get_file(vma->vm_file); 283345e55300SPeter Collingbourne ret = do_mmap(vma->vm_file, start, size, 2834897ab3e0SMike Rapoport prot, flags, pgoff, &populate, NULL); 2835c8d78c18SKirill A. Shutemov fput(file); 2836c8d78c18SKirill A. Shutemov out: 2837d8ed45c5SMichel Lespinasse mmap_write_unlock(mm); 2838c8d78c18SKirill A. Shutemov if (populate) 2839c8d78c18SKirill A. Shutemov mm_populate(ret, populate); 2840c8d78c18SKirill A. Shutemov if (!IS_ERR_VALUE(ret)) 2841c8d78c18SKirill A. Shutemov ret = 0; 2842c8d78c18SKirill A. Shutemov return ret; 2843c8d78c18SKirill A. Shutemov } 2844c8d78c18SKirill A. Shutemov 28451da177e4SLinus Torvalds /* 284627b26701SLiam R. Howlett * do_vma_munmap() - Unmap a full or partial vma. 284727b26701SLiam R. Howlett * @vmi: The vma iterator pointing at the vma 284827b26701SLiam R. Howlett * @vma: The first vma to be munmapped 284927b26701SLiam R. Howlett * @start: the start of the address to unmap 285027b26701SLiam R. Howlett * @end: The end of the address to unmap 28512e7ce7d3SLiam R. Howlett * @uf: The userfaultfd list_head 285227b26701SLiam R. Howlett * @downgrade: Attempt to downgrade or not 28532e7ce7d3SLiam R. Howlett * 285427b26701SLiam R. Howlett * Returns: 0 on success and not downgraded, 1 on success and downgraded. 285527b26701SLiam R. Howlett * unmaps a VMA mapping when the vma iterator is already in position. 285627b26701SLiam R. Howlett * Does not handle alignment. 28571da177e4SLinus Torvalds */ 285827b26701SLiam R. Howlett int do_vma_munmap(struct vma_iterator *vmi, struct vm_area_struct *vma, 285927b26701SLiam R. Howlett unsigned long start, unsigned long end, 286027b26701SLiam R. Howlett struct list_head *uf, bool downgrade) 28612e7ce7d3SLiam R. Howlett { 28622e7ce7d3SLiam R. Howlett struct mm_struct *mm = vma->vm_mm; 28632e7ce7d3SLiam R. Howlett int ret; 28642e7ce7d3SLiam R. Howlett 286527b26701SLiam R. Howlett arch_unmap(mm, start, end); 286627b26701SLiam R. Howlett ret = do_vmi_align_munmap(vmi, vma, mm, start, end, uf, downgrade); 28672e7ce7d3SLiam R. Howlett validate_mm_mt(mm); 28682e7ce7d3SLiam R. Howlett return ret; 28692e7ce7d3SLiam R. Howlett } 28702e7ce7d3SLiam R. Howlett 28712e7ce7d3SLiam R. Howlett /* 28722e7ce7d3SLiam R. Howlett * do_brk_flags() - Increase the brk vma if the flags match. 287392fed820SLiam R. Howlett * @vmi: The vma iterator 28742e7ce7d3SLiam R. Howlett * @addr: The start address 28752e7ce7d3SLiam R. Howlett * @len: The length of the increase 28762e7ce7d3SLiam R. Howlett * @vma: The vma, 28772e7ce7d3SLiam R. Howlett * @flags: The VMA Flags 28782e7ce7d3SLiam R. Howlett * 28792e7ce7d3SLiam R. Howlett * Extend the brk VMA from addr to addr + len. If the VMA is NULL or the flags 28802e7ce7d3SLiam R. Howlett * do not match then create a new anonymous VMA. Eventually we may be able to 28812e7ce7d3SLiam R. Howlett * do some brk-specific accounting here. 28822e7ce7d3SLiam R. Howlett */ 288392fed820SLiam R. Howlett static int do_brk_flags(struct vma_iterator *vmi, struct vm_area_struct *vma, 2884763ecb03SLiam R. Howlett unsigned long addr, unsigned long len, unsigned long flags) 28851da177e4SLinus Torvalds { 28861da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 2887287051b1SLiam R. Howlett struct vma_prepare vp; 28882e7ce7d3SLiam R. Howlett 2889d4af56c5SLiam R. Howlett validate_mm_mt(mm); 28902e7ce7d3SLiam R. Howlett /* 28912e7ce7d3SLiam R. Howlett * Check against address space limits by the changed size 28922e7ce7d3SLiam R. Howlett * Note: This happens *after* clearing old mappings in some code paths. 28932e7ce7d3SLiam R. Howlett */ 289416e72e9bSDenys Vlasenko flags |= VM_DATA_DEFAULT_FLAGS | VM_ACCOUNT | mm->def_flags; 289584638335SKonstantin Khlebnikov if (!may_expand_vm(mm, flags, len >> PAGE_SHIFT)) 28961da177e4SLinus Torvalds return -ENOMEM; 28971da177e4SLinus Torvalds 28981da177e4SLinus Torvalds if (mm->map_count > sysctl_max_map_count) 28991da177e4SLinus Torvalds return -ENOMEM; 29001da177e4SLinus Torvalds 2901191c5424SAl Viro if (security_vm_enough_memory_mm(mm, len >> PAGE_SHIFT)) 29021da177e4SLinus Torvalds return -ENOMEM; 29031da177e4SLinus Torvalds 29041da177e4SLinus Torvalds /* 29052e7ce7d3SLiam R. Howlett * Expand the existing vma if possible; Note that singular lists do not 29062e7ce7d3SLiam R. Howlett * occur after forking, so the expand will only happen on new VMAs. 29071da177e4SLinus Torvalds */ 29086c28ca64SLiam Howlett if (vma && vma->vm_end == addr && !vma_policy(vma) && 29096c28ca64SLiam Howlett can_vma_merge_after(vma, flags, NULL, NULL, 29106c28ca64SLiam Howlett addr >> PAGE_SHIFT, NULL_VM_UFFD_CTX, NULL)) { 291192fed820SLiam R. Howlett if (vma_iter_prealloc(vmi)) 2912675eaca1SAlistair Popple goto unacct_fail; 291328c5609fSLiam Howlett 291428c5609fSLiam Howlett vma_adjust_trans_huge(vma, vma->vm_start, addr + len, 0); 2915287051b1SLiam R. Howlett init_vma_prep(&vp, vma); 2916287051b1SLiam R. Howlett vma_prepare(&vp); 29172e7ce7d3SLiam R. Howlett vma->vm_end = addr + len; 29181c71222eSSuren Baghdasaryan vm_flags_set(vma, VM_SOFTDIRTY); 291992fed820SLiam R. Howlett vma_iter_store(vmi, vma); 29202e7ce7d3SLiam R. Howlett 2921287051b1SLiam R. Howlett vma_complete(&vp, vmi, mm); 29222e7ce7d3SLiam R. Howlett khugepaged_enter_vma(vma, flags); 29232e7ce7d3SLiam R. Howlett goto out; 29242e7ce7d3SLiam R. Howlett } 29252e7ce7d3SLiam R. Howlett 29262e7ce7d3SLiam R. Howlett /* create a vma struct for an anonymous mapping */ 29272e7ce7d3SLiam R. Howlett vma = vm_area_alloc(mm); 29282e7ce7d3SLiam R. Howlett if (!vma) 2929675eaca1SAlistair Popple goto unacct_fail; 29301da177e4SLinus Torvalds 2931bfd40eafSKirill A. Shutemov vma_set_anonymous(vma); 29321da177e4SLinus Torvalds vma->vm_start = addr; 29331da177e4SLinus Torvalds vma->vm_end = addr + len; 29342e7ce7d3SLiam R. Howlett vma->vm_pgoff = addr >> PAGE_SHIFT; 29351c71222eSSuren Baghdasaryan vm_flags_init(vma, flags); 29363ed75eb8SColy Li vma->vm_page_prot = vm_get_page_prot(flags); 293792fed820SLiam R. Howlett if (vma_iter_store_gfp(vmi, vma, GFP_KERNEL)) 29382e7ce7d3SLiam R. Howlett goto mas_store_fail; 2939d4af56c5SLiam R. Howlett 29402e7ce7d3SLiam R. Howlett mm->map_count++; 29411da177e4SLinus Torvalds out: 29423af9e859SEric B Munson perf_event_mmap(vma); 29431da177e4SLinus Torvalds mm->total_vm += len >> PAGE_SHIFT; 294484638335SKonstantin Khlebnikov mm->data_vm += len >> PAGE_SHIFT; 2945128557ffSMichel Lespinasse if (flags & VM_LOCKED) 2946ba470de4SRik van Riel mm->locked_vm += (len >> PAGE_SHIFT); 29471c71222eSSuren Baghdasaryan vm_flags_set(vma, VM_SOFTDIRTY); 2948763ecb03SLiam R. Howlett validate_mm(mm); 29495d22fc25SLinus Torvalds return 0; 2950d4af56c5SLiam R. Howlett 29512e7ce7d3SLiam R. Howlett mas_store_fail: 2952d4af56c5SLiam R. Howlett vm_area_free(vma); 2953675eaca1SAlistair Popple unacct_fail: 29542e7ce7d3SLiam R. Howlett vm_unacct_memory(len >> PAGE_SHIFT); 29552e7ce7d3SLiam R. Howlett return -ENOMEM; 29561da177e4SLinus Torvalds } 29571da177e4SLinus Torvalds 2958bb177a73SMichal Hocko int vm_brk_flags(unsigned long addr, unsigned long request, unsigned long flags) 2959e4eb1ff6SLinus Torvalds { 2960e4eb1ff6SLinus Torvalds struct mm_struct *mm = current->mm; 29612e7ce7d3SLiam R. Howlett struct vm_area_struct *vma = NULL; 2962bb177a73SMichal Hocko unsigned long len; 29635d22fc25SLinus Torvalds int ret; 2964128557ffSMichel Lespinasse bool populate; 2965897ab3e0SMike Rapoport LIST_HEAD(uf); 296692fed820SLiam R. Howlett VMA_ITERATOR(vmi, mm, addr); 2967e4eb1ff6SLinus Torvalds 2968bb177a73SMichal Hocko len = PAGE_ALIGN(request); 2969bb177a73SMichal Hocko if (len < request) 2970bb177a73SMichal Hocko return -ENOMEM; 2971bb177a73SMichal Hocko if (!len) 2972bb177a73SMichal Hocko return 0; 2973bb177a73SMichal Hocko 2974d8ed45c5SMichel Lespinasse if (mmap_write_lock_killable(mm)) 29752d6c9282SMichal Hocko return -EINTR; 29762d6c9282SMichal Hocko 29772e7ce7d3SLiam R. Howlett /* Until we need other flags, refuse anything except VM_EXEC. */ 29782e7ce7d3SLiam R. Howlett if ((flags & (~VM_EXEC)) != 0) 29792e7ce7d3SLiam R. Howlett return -EINVAL; 29802e7ce7d3SLiam R. Howlett 29812e7ce7d3SLiam R. Howlett ret = check_brk_limits(addr, len); 29822e7ce7d3SLiam R. Howlett if (ret) 29832e7ce7d3SLiam R. Howlett goto limits_failed; 29842e7ce7d3SLiam R. Howlett 2985183654ceSLiam R. Howlett ret = do_vmi_munmap(&vmi, mm, addr, len, &uf, 0); 29862e7ce7d3SLiam R. Howlett if (ret) 29872e7ce7d3SLiam R. Howlett goto munmap_failed; 29882e7ce7d3SLiam R. Howlett 298992fed820SLiam R. Howlett vma = vma_prev(&vmi); 299092fed820SLiam R. Howlett ret = do_brk_flags(&vmi, vma, addr, len, flags); 2991128557ffSMichel Lespinasse populate = ((mm->def_flags & VM_LOCKED) != 0); 2992d8ed45c5SMichel Lespinasse mmap_write_unlock(mm); 2993897ab3e0SMike Rapoport userfaultfd_unmap_complete(mm, &uf); 29945d22fc25SLinus Torvalds if (populate && !ret) 2995128557ffSMichel Lespinasse mm_populate(addr, len); 2996e4eb1ff6SLinus Torvalds return ret; 29972e7ce7d3SLiam R. Howlett 29982e7ce7d3SLiam R. Howlett munmap_failed: 29992e7ce7d3SLiam R. Howlett limits_failed: 30002e7ce7d3SLiam R. Howlett mmap_write_unlock(mm); 30012e7ce7d3SLiam R. Howlett return ret; 3002e4eb1ff6SLinus Torvalds } 300316e72e9bSDenys Vlasenko EXPORT_SYMBOL(vm_brk_flags); 300416e72e9bSDenys Vlasenko 300516e72e9bSDenys Vlasenko int vm_brk(unsigned long addr, unsigned long len) 300616e72e9bSDenys Vlasenko { 300716e72e9bSDenys Vlasenko return vm_brk_flags(addr, len, 0); 300816e72e9bSDenys Vlasenko } 3009e4eb1ff6SLinus Torvalds EXPORT_SYMBOL(vm_brk); 30101da177e4SLinus Torvalds 30111da177e4SLinus Torvalds /* Release all mmaps. */ 30121da177e4SLinus Torvalds void exit_mmap(struct mm_struct *mm) 30131da177e4SLinus Torvalds { 3014d16dfc55SPeter Zijlstra struct mmu_gather tlb; 3015ba470de4SRik van Riel struct vm_area_struct *vma; 30161da177e4SLinus Torvalds unsigned long nr_accounted = 0; 3017763ecb03SLiam R. Howlett MA_STATE(mas, &mm->mm_mt, 0, 0); 3018763ecb03SLiam R. Howlett int count = 0; 30191da177e4SLinus Torvalds 3020d6dd61c8SJeremy Fitzhardinge /* mm's last user has gone, and its about to be pulled down */ 3021cddb8a5cSAndrea Arcangeli mmu_notifier_release(mm); 3022d6dd61c8SJeremy Fitzhardinge 3023bf3980c8SSuren Baghdasaryan mmap_read_lock(mm); 30249480c53eSJeremy Fitzhardinge arch_exit_mmap(mm); 30259480c53eSJeremy Fitzhardinge 3026763ecb03SLiam R. Howlett vma = mas_find(&mas, ULONG_MAX); 302764591e86SSuren Baghdasaryan if (!vma) { 302864591e86SSuren Baghdasaryan /* Can happen if dup_mmap() received an OOM */ 3029bf3980c8SSuren Baghdasaryan mmap_read_unlock(mm); 30309480c53eSJeremy Fitzhardinge return; 303164591e86SSuren Baghdasaryan } 30329480c53eSJeremy Fitzhardinge 30331da177e4SLinus Torvalds lru_add_drain(); 30341da177e4SLinus Torvalds flush_cache_mm(mm); 3035d8b45053SWill Deacon tlb_gather_mmu_fullmm(&tlb, mm); 3036901608d9SOleg Nesterov /* update_hiwater_rss(mm) here? but nobody should be looking */ 3037763ecb03SLiam R. Howlett /* Use ULONG_MAX here to ensure all VMAs in the mm are unmapped */ 303868f48381SSuren Baghdasaryan unmap_vmas(&tlb, &mm->mm_mt, vma, 0, ULONG_MAX, false); 3039bf3980c8SSuren Baghdasaryan mmap_read_unlock(mm); 3040bf3980c8SSuren Baghdasaryan 3041bf3980c8SSuren Baghdasaryan /* 3042bf3980c8SSuren Baghdasaryan * Set MMF_OOM_SKIP to hide this task from the oom killer/reaper 3043b3541d91SSuren Baghdasaryan * because the memory has been already freed. 3044bf3980c8SSuren Baghdasaryan */ 3045bf3980c8SSuren Baghdasaryan set_bit(MMF_OOM_SKIP, &mm->flags); 3046bf3980c8SSuren Baghdasaryan mmap_write_lock(mm); 3047763ecb03SLiam R. Howlett free_pgtables(&tlb, &mm->mm_mt, vma, FIRST_USER_ADDRESS, 3048763ecb03SLiam R. Howlett USER_PGTABLES_CEILING); 3049ae8eba8bSWill Deacon tlb_finish_mmu(&tlb); 30501da177e4SLinus Torvalds 3051763ecb03SLiam R. Howlett /* 3052763ecb03SLiam R. Howlett * Walk the list again, actually closing and freeing it, with preemption 3053763ecb03SLiam R. Howlett * enabled, without holding any MM locks besides the unreachable 3054763ecb03SLiam R. Howlett * mmap_write_lock. 3055763ecb03SLiam R. Howlett */ 3056763ecb03SLiam R. Howlett do { 30574f74d2c8SLinus Torvalds if (vma->vm_flags & VM_ACCOUNT) 30584f74d2c8SLinus Torvalds nr_accounted += vma_pages(vma); 3059763ecb03SLiam R. Howlett remove_vma(vma); 3060763ecb03SLiam R. Howlett count++; 30610a3b3c25SPaul E. McKenney cond_resched(); 3062763ecb03SLiam R. Howlett } while ((vma = mas_find(&mas, ULONG_MAX)) != NULL); 3063763ecb03SLiam R. Howlett 3064763ecb03SLiam R. Howlett BUG_ON(count != mm->map_count); 3065d4af56c5SLiam R. Howlett 3066d4af56c5SLiam R. Howlett trace_exit_mmap(mm); 3067d4af56c5SLiam R. Howlett __mt_destroy(&mm->mm_mt); 306864591e86SSuren Baghdasaryan mmap_write_unlock(mm); 30694f74d2c8SLinus Torvalds vm_unacct_memory(nr_accounted); 30701da177e4SLinus Torvalds } 30711da177e4SLinus Torvalds 30721da177e4SLinus Torvalds /* Insert vm structure into process list sorted by address 30731da177e4SLinus Torvalds * and into the inode's i_mmap tree. If vm_file is non-NULL 3074c8c06efaSDavidlohr Bueso * then i_mmap_rwsem is taken here. 30751da177e4SLinus Torvalds */ 30761da177e4SLinus Torvalds int insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma) 30771da177e4SLinus Torvalds { 3078d4af56c5SLiam R. Howlett unsigned long charged = vma_pages(vma); 30791da177e4SLinus Torvalds 3080d4af56c5SLiam R. Howlett 3081d0601a50SLiam R. Howlett if (find_vma_intersection(mm, vma->vm_start, vma->vm_end)) 3082c9d13f5fSChen Gang return -ENOMEM; 3083d4af56c5SLiam R. Howlett 3084c9d13f5fSChen Gang if ((vma->vm_flags & VM_ACCOUNT) && 3085d4af56c5SLiam R. Howlett security_vm_enough_memory_mm(mm, charged)) 3086c9d13f5fSChen Gang return -ENOMEM; 3087c9d13f5fSChen Gang 30881da177e4SLinus Torvalds /* 30891da177e4SLinus Torvalds * The vm_pgoff of a purely anonymous vma should be irrelevant 30901da177e4SLinus Torvalds * until its first write fault, when page's anon_vma and index 30911da177e4SLinus Torvalds * are set. But now set the vm_pgoff it will almost certainly 30921da177e4SLinus Torvalds * end up with (unless mremap moves it elsewhere before that 30931da177e4SLinus Torvalds * first wfault), so /proc/pid/maps tells a consistent story. 30941da177e4SLinus Torvalds * 30951da177e4SLinus Torvalds * By setting it to reflect the virtual start address of the 30961da177e4SLinus Torvalds * vma, merges and splits can happen in a seamless way, just 30971da177e4SLinus Torvalds * using the existing file pgoff checks and manipulations. 30988332326eSLiao Pingfang * Similarly in do_mmap and in do_brk_flags. 30991da177e4SLinus Torvalds */ 31008a9cc3b5SOleg Nesterov if (vma_is_anonymous(vma)) { 31011da177e4SLinus Torvalds BUG_ON(vma->anon_vma); 31021da177e4SLinus Torvalds vma->vm_pgoff = vma->vm_start >> PAGE_SHIFT; 31031da177e4SLinus Torvalds } 31042b144498SSrikar Dronamraju 3105763ecb03SLiam R. Howlett if (vma_link(mm, vma)) { 3106d4af56c5SLiam R. Howlett vm_unacct_memory(charged); 3107d4af56c5SLiam R. Howlett return -ENOMEM; 3108d4af56c5SLiam R. Howlett } 3109d4af56c5SLiam R. Howlett 31101da177e4SLinus Torvalds return 0; 31111da177e4SLinus Torvalds } 31121da177e4SLinus Torvalds 31131da177e4SLinus Torvalds /* 31141da177e4SLinus Torvalds * Copy the vma structure to a new location in the same mm, 31151da177e4SLinus Torvalds * prior to moving page table entries, to effect an mremap move. 31161da177e4SLinus Torvalds */ 31171da177e4SLinus Torvalds struct vm_area_struct *copy_vma(struct vm_area_struct **vmap, 311838a76013SMichel Lespinasse unsigned long addr, unsigned long len, pgoff_t pgoff, 311938a76013SMichel Lespinasse bool *need_rmap_locks) 31201da177e4SLinus Torvalds { 31211da177e4SLinus Torvalds struct vm_area_struct *vma = *vmap; 31221da177e4SLinus Torvalds unsigned long vma_start = vma->vm_start; 31231da177e4SLinus Torvalds struct mm_struct *mm = vma->vm_mm; 31241da177e4SLinus Torvalds struct vm_area_struct *new_vma, *prev; 3125948f017bSAndrea Arcangeli bool faulted_in_anon_vma = true; 3126076f16bfSLiam R. Howlett VMA_ITERATOR(vmi, mm, addr); 31271da177e4SLinus Torvalds 3128d4af56c5SLiam R. Howlett validate_mm_mt(mm); 31291da177e4SLinus Torvalds /* 31301da177e4SLinus Torvalds * If anonymous vma has not yet been faulted, update new pgoff 31311da177e4SLinus Torvalds * to match new location, to increase its chance of merging. 31321da177e4SLinus Torvalds */ 3133ce75799bSOleg Nesterov if (unlikely(vma_is_anonymous(vma) && !vma->anon_vma)) { 31341da177e4SLinus Torvalds pgoff = addr >> PAGE_SHIFT; 3135948f017bSAndrea Arcangeli faulted_in_anon_vma = false; 3136948f017bSAndrea Arcangeli } 31371da177e4SLinus Torvalds 3138763ecb03SLiam R. Howlett new_vma = find_vma_prev(mm, addr, &prev); 3139763ecb03SLiam R. Howlett if (new_vma && new_vma->vm_start < addr + len) 31406597d783SHugh Dickins return NULL; /* should never get here */ 3141524e00b3SLiam R. Howlett 31429760ebffSLiam R. Howlett new_vma = vma_merge(&vmi, mm, prev, addr, addr + len, vma->vm_flags, 314319a809afSAndrea Arcangeli vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma), 31445c26f6acSSuren Baghdasaryan vma->vm_userfaultfd_ctx, anon_vma_name(vma)); 31451da177e4SLinus Torvalds if (new_vma) { 31461da177e4SLinus Torvalds /* 31471da177e4SLinus Torvalds * Source vma may have been merged into new_vma 31481da177e4SLinus Torvalds */ 3149948f017bSAndrea Arcangeli if (unlikely(vma_start >= new_vma->vm_start && 3150948f017bSAndrea Arcangeli vma_start < new_vma->vm_end)) { 3151948f017bSAndrea Arcangeli /* 3152948f017bSAndrea Arcangeli * The only way we can get a vma_merge with 3153948f017bSAndrea Arcangeli * self during an mremap is if the vma hasn't 3154948f017bSAndrea Arcangeli * been faulted in yet and we were allowed to 3155948f017bSAndrea Arcangeli * reset the dst vma->vm_pgoff to the 3156948f017bSAndrea Arcangeli * destination address of the mremap to allow 3157948f017bSAndrea Arcangeli * the merge to happen. mremap must change the 3158948f017bSAndrea Arcangeli * vm_pgoff linearity between src and dst vmas 3159948f017bSAndrea Arcangeli * (in turn preventing a vma_merge) to be 3160948f017bSAndrea Arcangeli * safe. It is only safe to keep the vm_pgoff 3161948f017bSAndrea Arcangeli * linear if there are no pages mapped yet. 3162948f017bSAndrea Arcangeli */ 316381d1b09cSSasha Levin VM_BUG_ON_VMA(faulted_in_anon_vma, new_vma); 316438a76013SMichel Lespinasse *vmap = vma = new_vma; 3165108d6642SMichel Lespinasse } 316638a76013SMichel Lespinasse *need_rmap_locks = (new_vma->vm_pgoff <= vma->vm_pgoff); 31671da177e4SLinus Torvalds } else { 31683928d4f5SLinus Torvalds new_vma = vm_area_dup(vma); 3169e3975891SChen Gang if (!new_vma) 3170e3975891SChen Gang goto out; 31711da177e4SLinus Torvalds new_vma->vm_start = addr; 31721da177e4SLinus Torvalds new_vma->vm_end = addr + len; 31731da177e4SLinus Torvalds new_vma->vm_pgoff = pgoff; 3174ef0855d3SOleg Nesterov if (vma_dup_policy(vma, new_vma)) 3175523d4e20SMichel Lespinasse goto out_free_vma; 3176523d4e20SMichel Lespinasse if (anon_vma_clone(new_vma, vma)) 3177523d4e20SMichel Lespinasse goto out_free_mempol; 3178e9714acfSKonstantin Khlebnikov if (new_vma->vm_file) 31791da177e4SLinus Torvalds get_file(new_vma->vm_file); 31801da177e4SLinus Torvalds if (new_vma->vm_ops && new_vma->vm_ops->open) 31811da177e4SLinus Torvalds new_vma->vm_ops->open(new_vma); 3182763ecb03SLiam R. Howlett if (vma_link(mm, new_vma)) 3183524e00b3SLiam R. Howlett goto out_vma_link; 318438a76013SMichel Lespinasse *need_rmap_locks = false; 31851da177e4SLinus Torvalds } 3186d4af56c5SLiam R. Howlett validate_mm_mt(mm); 31871da177e4SLinus Torvalds return new_vma; 31885beb4930SRik van Riel 3189524e00b3SLiam R. Howlett out_vma_link: 3190524e00b3SLiam R. Howlett if (new_vma->vm_ops && new_vma->vm_ops->close) 3191524e00b3SLiam R. Howlett new_vma->vm_ops->close(new_vma); 319292b73996SLiam Howlett 319392b73996SLiam Howlett if (new_vma->vm_file) 319492b73996SLiam Howlett fput(new_vma->vm_file); 319592b73996SLiam Howlett 319692b73996SLiam Howlett unlink_anon_vmas(new_vma); 31975beb4930SRik van Riel out_free_mempol: 3198ef0855d3SOleg Nesterov mpol_put(vma_policy(new_vma)); 31995beb4930SRik van Riel out_free_vma: 32003928d4f5SLinus Torvalds vm_area_free(new_vma); 3201e3975891SChen Gang out: 3202d4af56c5SLiam R. Howlett validate_mm_mt(mm); 32035beb4930SRik van Riel return NULL; 32041da177e4SLinus Torvalds } 3205119f657cSakpm@osdl.org 3206119f657cSakpm@osdl.org /* 3207119f657cSakpm@osdl.org * Return true if the calling process may expand its vm space by the passed 3208119f657cSakpm@osdl.org * number of pages 3209119f657cSakpm@osdl.org */ 321084638335SKonstantin Khlebnikov bool may_expand_vm(struct mm_struct *mm, vm_flags_t flags, unsigned long npages) 3211119f657cSakpm@osdl.org { 321284638335SKonstantin Khlebnikov if (mm->total_vm + npages > rlimit(RLIMIT_AS) >> PAGE_SHIFT) 321384638335SKonstantin Khlebnikov return false; 3214119f657cSakpm@osdl.org 3215d977d56cSKonstantin Khlebnikov if (is_data_mapping(flags) && 3216d977d56cSKonstantin Khlebnikov mm->data_vm + npages > rlimit(RLIMIT_DATA) >> PAGE_SHIFT) { 3217f4fcd558SKonstantin Khlebnikov /* Workaround for Valgrind */ 3218f4fcd558SKonstantin Khlebnikov if (rlimit(RLIMIT_DATA) == 0 && 3219f4fcd558SKonstantin Khlebnikov mm->data_vm + npages <= rlimit_max(RLIMIT_DATA) >> PAGE_SHIFT) 3220f4fcd558SKonstantin Khlebnikov return true; 322157a7702bSDavid Woodhouse 322257a7702bSDavid Woodhouse pr_warn_once("%s (%d): VmData %lu exceed data ulimit %lu. Update limits%s.\n", 3223d977d56cSKonstantin Khlebnikov current->comm, current->pid, 3224d977d56cSKonstantin Khlebnikov (mm->data_vm + npages) << PAGE_SHIFT, 322557a7702bSDavid Woodhouse rlimit(RLIMIT_DATA), 322657a7702bSDavid Woodhouse ignore_rlimit_data ? "" : " or use boot option ignore_rlimit_data"); 322757a7702bSDavid Woodhouse 322857a7702bSDavid Woodhouse if (!ignore_rlimit_data) 3229d977d56cSKonstantin Khlebnikov return false; 3230d977d56cSKonstantin Khlebnikov } 3231119f657cSakpm@osdl.org 323284638335SKonstantin Khlebnikov return true; 323384638335SKonstantin Khlebnikov } 323484638335SKonstantin Khlebnikov 323584638335SKonstantin Khlebnikov void vm_stat_account(struct mm_struct *mm, vm_flags_t flags, long npages) 323684638335SKonstantin Khlebnikov { 32377866076bSPeng Liu WRITE_ONCE(mm->total_vm, READ_ONCE(mm->total_vm)+npages); 323884638335SKonstantin Khlebnikov 3239d977d56cSKonstantin Khlebnikov if (is_exec_mapping(flags)) 324084638335SKonstantin Khlebnikov mm->exec_vm += npages; 3241d977d56cSKonstantin Khlebnikov else if (is_stack_mapping(flags)) 324284638335SKonstantin Khlebnikov mm->stack_vm += npages; 3243d977d56cSKonstantin Khlebnikov else if (is_data_mapping(flags)) 324484638335SKonstantin Khlebnikov mm->data_vm += npages; 3245119f657cSakpm@osdl.org } 3246fa5dc22fSRoland McGrath 3247b3ec9f33SSouptick Joarder static vm_fault_t special_mapping_fault(struct vm_fault *vmf); 3248a62c34bdSAndy Lutomirski 3249a62c34bdSAndy Lutomirski /* 3250a62c34bdSAndy Lutomirski * Having a close hook prevents vma merging regardless of flags. 3251a62c34bdSAndy Lutomirski */ 3252a62c34bdSAndy Lutomirski static void special_mapping_close(struct vm_area_struct *vma) 3253a62c34bdSAndy Lutomirski { 3254a62c34bdSAndy Lutomirski } 3255a62c34bdSAndy Lutomirski 3256a62c34bdSAndy Lutomirski static const char *special_mapping_name(struct vm_area_struct *vma) 3257a62c34bdSAndy Lutomirski { 3258a62c34bdSAndy Lutomirski return ((struct vm_special_mapping *)vma->vm_private_data)->name; 3259a62c34bdSAndy Lutomirski } 3260a62c34bdSAndy Lutomirski 326114d07113SBrian Geffon static int special_mapping_mremap(struct vm_area_struct *new_vma) 3262b059a453SDmitry Safonov { 3263b059a453SDmitry Safonov struct vm_special_mapping *sm = new_vma->vm_private_data; 3264b059a453SDmitry Safonov 3265280e87e9SDmitry Safonov if (WARN_ON_ONCE(current->mm != new_vma->vm_mm)) 3266280e87e9SDmitry Safonov return -EFAULT; 3267280e87e9SDmitry Safonov 3268b059a453SDmitry Safonov if (sm->mremap) 3269b059a453SDmitry Safonov return sm->mremap(sm, new_vma); 3270280e87e9SDmitry Safonov 3271b059a453SDmitry Safonov return 0; 3272b059a453SDmitry Safonov } 3273b059a453SDmitry Safonov 3274871402e0SDmitry Safonov static int special_mapping_split(struct vm_area_struct *vma, unsigned long addr) 3275871402e0SDmitry Safonov { 3276871402e0SDmitry Safonov /* 3277871402e0SDmitry Safonov * Forbid splitting special mappings - kernel has expectations over 3278871402e0SDmitry Safonov * the number of pages in mapping. Together with VM_DONTEXPAND 3279871402e0SDmitry Safonov * the size of vma should stay the same over the special mapping's 3280871402e0SDmitry Safonov * lifetime. 3281871402e0SDmitry Safonov */ 3282871402e0SDmitry Safonov return -EINVAL; 3283871402e0SDmitry Safonov } 3284871402e0SDmitry Safonov 3285a62c34bdSAndy Lutomirski static const struct vm_operations_struct special_mapping_vmops = { 3286a62c34bdSAndy Lutomirski .close = special_mapping_close, 3287a62c34bdSAndy Lutomirski .fault = special_mapping_fault, 3288b059a453SDmitry Safonov .mremap = special_mapping_mremap, 3289a62c34bdSAndy Lutomirski .name = special_mapping_name, 3290af34ebebSDmitry Safonov /* vDSO code relies that VVAR can't be accessed remotely */ 3291af34ebebSDmitry Safonov .access = NULL, 3292871402e0SDmitry Safonov .may_split = special_mapping_split, 3293a62c34bdSAndy Lutomirski }; 3294a62c34bdSAndy Lutomirski 3295a62c34bdSAndy Lutomirski static const struct vm_operations_struct legacy_special_mapping_vmops = { 3296a62c34bdSAndy Lutomirski .close = special_mapping_close, 3297a62c34bdSAndy Lutomirski .fault = special_mapping_fault, 3298a62c34bdSAndy Lutomirski }; 3299fa5dc22fSRoland McGrath 3300b3ec9f33SSouptick Joarder static vm_fault_t special_mapping_fault(struct vm_fault *vmf) 3301fa5dc22fSRoland McGrath { 330211bac800SDave Jiang struct vm_area_struct *vma = vmf->vma; 3303b1d0e4f5SNick Piggin pgoff_t pgoff; 3304fa5dc22fSRoland McGrath struct page **pages; 3305fa5dc22fSRoland McGrath 3306f872f540SAndy Lutomirski if (vma->vm_ops == &legacy_special_mapping_vmops) { 3307a62c34bdSAndy Lutomirski pages = vma->vm_private_data; 3308f872f540SAndy Lutomirski } else { 3309f872f540SAndy Lutomirski struct vm_special_mapping *sm = vma->vm_private_data; 3310f872f540SAndy Lutomirski 3311f872f540SAndy Lutomirski if (sm->fault) 331211bac800SDave Jiang return sm->fault(sm, vmf->vma, vmf); 3313f872f540SAndy Lutomirski 3314f872f540SAndy Lutomirski pages = sm->pages; 3315f872f540SAndy Lutomirski } 3316a62c34bdSAndy Lutomirski 33178a9cc3b5SOleg Nesterov for (pgoff = vmf->pgoff; pgoff && *pages; ++pages) 3318b1d0e4f5SNick Piggin pgoff--; 3319fa5dc22fSRoland McGrath 3320fa5dc22fSRoland McGrath if (*pages) { 3321fa5dc22fSRoland McGrath struct page *page = *pages; 3322fa5dc22fSRoland McGrath get_page(page); 3323b1d0e4f5SNick Piggin vmf->page = page; 3324b1d0e4f5SNick Piggin return 0; 3325fa5dc22fSRoland McGrath } 3326fa5dc22fSRoland McGrath 3327b1d0e4f5SNick Piggin return VM_FAULT_SIGBUS; 3328fa5dc22fSRoland McGrath } 3329fa5dc22fSRoland McGrath 3330a62c34bdSAndy Lutomirski static struct vm_area_struct *__install_special_mapping( 3331a62c34bdSAndy Lutomirski struct mm_struct *mm, 3332fa5dc22fSRoland McGrath unsigned long addr, unsigned long len, 333327f28b97SChen Gang unsigned long vm_flags, void *priv, 333427f28b97SChen Gang const struct vm_operations_struct *ops) 3335fa5dc22fSRoland McGrath { 3336462e635eSTavis Ormandy int ret; 3337fa5dc22fSRoland McGrath struct vm_area_struct *vma; 3338fa5dc22fSRoland McGrath 3339d4af56c5SLiam R. Howlett validate_mm_mt(mm); 3340490fc053SLinus Torvalds vma = vm_area_alloc(mm); 3341fa5dc22fSRoland McGrath if (unlikely(vma == NULL)) 33423935ed6aSStefani Seibold return ERR_PTR(-ENOMEM); 3343fa5dc22fSRoland McGrath 3344fa5dc22fSRoland McGrath vma->vm_start = addr; 3345fa5dc22fSRoland McGrath vma->vm_end = addr + len; 3346fa5dc22fSRoland McGrath 3347e430a95aSSuren Baghdasaryan vm_flags_init(vma, (vm_flags | mm->def_flags | 3348e430a95aSSuren Baghdasaryan VM_DONTEXPAND | VM_SOFTDIRTY) & ~VM_LOCKED_MASK); 33493ed75eb8SColy Li vma->vm_page_prot = vm_get_page_prot(vma->vm_flags); 3350fa5dc22fSRoland McGrath 3351a62c34bdSAndy Lutomirski vma->vm_ops = ops; 3352a62c34bdSAndy Lutomirski vma->vm_private_data = priv; 3353fa5dc22fSRoland McGrath 3354462e635eSTavis Ormandy ret = insert_vm_struct(mm, vma); 3355462e635eSTavis Ormandy if (ret) 3356462e635eSTavis Ormandy goto out; 3357fa5dc22fSRoland McGrath 335884638335SKonstantin Khlebnikov vm_stat_account(mm, vma->vm_flags, len >> PAGE_SHIFT); 3359fa5dc22fSRoland McGrath 3360cdd6c482SIngo Molnar perf_event_mmap(vma); 3361089dd79dSPeter Zijlstra 3362d4af56c5SLiam R. Howlett validate_mm_mt(mm); 33633935ed6aSStefani Seibold return vma; 3364462e635eSTavis Ormandy 3365462e635eSTavis Ormandy out: 33663928d4f5SLinus Torvalds vm_area_free(vma); 3367d4af56c5SLiam R. Howlett validate_mm_mt(mm); 33683935ed6aSStefani Seibold return ERR_PTR(ret); 33693935ed6aSStefani Seibold } 33703935ed6aSStefani Seibold 33712eefd878SDmitry Safonov bool vma_is_special_mapping(const struct vm_area_struct *vma, 33722eefd878SDmitry Safonov const struct vm_special_mapping *sm) 33732eefd878SDmitry Safonov { 33742eefd878SDmitry Safonov return vma->vm_private_data == sm && 33752eefd878SDmitry Safonov (vma->vm_ops == &special_mapping_vmops || 33762eefd878SDmitry Safonov vma->vm_ops == &legacy_special_mapping_vmops); 33772eefd878SDmitry Safonov } 33782eefd878SDmitry Safonov 3379a62c34bdSAndy Lutomirski /* 3380c1e8d7c6SMichel Lespinasse * Called with mm->mmap_lock held for writing. 3381a62c34bdSAndy Lutomirski * Insert a new vma covering the given region, with the given flags. 3382a62c34bdSAndy Lutomirski * Its pages are supplied by the given array of struct page *. 3383a62c34bdSAndy Lutomirski * The array can be shorter than len >> PAGE_SHIFT if it's null-terminated. 3384a62c34bdSAndy Lutomirski * The region past the last page supplied will always produce SIGBUS. 3385a62c34bdSAndy Lutomirski * The array pointer and the pages it points to are assumed to stay alive 3386a62c34bdSAndy Lutomirski * for as long as this mapping might exist. 3387a62c34bdSAndy Lutomirski */ 3388a62c34bdSAndy Lutomirski struct vm_area_struct *_install_special_mapping( 3389a62c34bdSAndy Lutomirski struct mm_struct *mm, 3390a62c34bdSAndy Lutomirski unsigned long addr, unsigned long len, 3391a62c34bdSAndy Lutomirski unsigned long vm_flags, const struct vm_special_mapping *spec) 3392a62c34bdSAndy Lutomirski { 339327f28b97SChen Gang return __install_special_mapping(mm, addr, len, vm_flags, (void *)spec, 339427f28b97SChen Gang &special_mapping_vmops); 3395a62c34bdSAndy Lutomirski } 3396a62c34bdSAndy Lutomirski 33973935ed6aSStefani Seibold int install_special_mapping(struct mm_struct *mm, 33983935ed6aSStefani Seibold unsigned long addr, unsigned long len, 33993935ed6aSStefani Seibold unsigned long vm_flags, struct page **pages) 34003935ed6aSStefani Seibold { 3401a62c34bdSAndy Lutomirski struct vm_area_struct *vma = __install_special_mapping( 340227f28b97SChen Gang mm, addr, len, vm_flags, (void *)pages, 340327f28b97SChen Gang &legacy_special_mapping_vmops); 34043935ed6aSStefani Seibold 340514bd5b45SDuan Jiong return PTR_ERR_OR_ZERO(vma); 3406fa5dc22fSRoland McGrath } 34077906d00cSAndrea Arcangeli 34087906d00cSAndrea Arcangeli static DEFINE_MUTEX(mm_all_locks_mutex); 34097906d00cSAndrea Arcangeli 3410454ed842SPeter Zijlstra static void vm_lock_anon_vma(struct mm_struct *mm, struct anon_vma *anon_vma) 34117906d00cSAndrea Arcangeli { 3412f808c13fSDavidlohr Bueso if (!test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_root.rb_node)) { 34137906d00cSAndrea Arcangeli /* 34147906d00cSAndrea Arcangeli * The LSB of head.next can't change from under us 34157906d00cSAndrea Arcangeli * because we hold the mm_all_locks_mutex. 34167906d00cSAndrea Arcangeli */ 3417da1c55f1SMichel Lespinasse down_write_nest_lock(&anon_vma->root->rwsem, &mm->mmap_lock); 34187906d00cSAndrea Arcangeli /* 34197906d00cSAndrea Arcangeli * We can safely modify head.next after taking the 34205a505085SIngo Molnar * anon_vma->root->rwsem. If some other vma in this mm shares 34217906d00cSAndrea Arcangeli * the same anon_vma we won't take it again. 34227906d00cSAndrea Arcangeli * 34237906d00cSAndrea Arcangeli * No need of atomic instructions here, head.next 34247906d00cSAndrea Arcangeli * can't change from under us thanks to the 34255a505085SIngo Molnar * anon_vma->root->rwsem. 34267906d00cSAndrea Arcangeli */ 34277906d00cSAndrea Arcangeli if (__test_and_set_bit(0, (unsigned long *) 3428f808c13fSDavidlohr Bueso &anon_vma->root->rb_root.rb_root.rb_node)) 34297906d00cSAndrea Arcangeli BUG(); 34307906d00cSAndrea Arcangeli } 34317906d00cSAndrea Arcangeli } 34327906d00cSAndrea Arcangeli 3433454ed842SPeter Zijlstra static void vm_lock_mapping(struct mm_struct *mm, struct address_space *mapping) 34347906d00cSAndrea Arcangeli { 34357906d00cSAndrea Arcangeli if (!test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) { 34367906d00cSAndrea Arcangeli /* 34377906d00cSAndrea Arcangeli * AS_MM_ALL_LOCKS can't change from under us because 34387906d00cSAndrea Arcangeli * we hold the mm_all_locks_mutex. 34397906d00cSAndrea Arcangeli * 34407906d00cSAndrea Arcangeli * Operations on ->flags have to be atomic because 34417906d00cSAndrea Arcangeli * even if AS_MM_ALL_LOCKS is stable thanks to the 34427906d00cSAndrea Arcangeli * mm_all_locks_mutex, there may be other cpus 34437906d00cSAndrea Arcangeli * changing other bitflags in parallel to us. 34447906d00cSAndrea Arcangeli */ 34457906d00cSAndrea Arcangeli if (test_and_set_bit(AS_MM_ALL_LOCKS, &mapping->flags)) 34467906d00cSAndrea Arcangeli BUG(); 3447da1c55f1SMichel Lespinasse down_write_nest_lock(&mapping->i_mmap_rwsem, &mm->mmap_lock); 34487906d00cSAndrea Arcangeli } 34497906d00cSAndrea Arcangeli } 34507906d00cSAndrea Arcangeli 34517906d00cSAndrea Arcangeli /* 34527906d00cSAndrea Arcangeli * This operation locks against the VM for all pte/vma/mm related 34537906d00cSAndrea Arcangeli * operations that could ever happen on a certain mm. This includes 34547906d00cSAndrea Arcangeli * vmtruncate, try_to_unmap, and all page faults. 34557906d00cSAndrea Arcangeli * 3456c1e8d7c6SMichel Lespinasse * The caller must take the mmap_lock in write mode before calling 34577906d00cSAndrea Arcangeli * mm_take_all_locks(). The caller isn't allowed to release the 3458c1e8d7c6SMichel Lespinasse * mmap_lock until mm_drop_all_locks() returns. 34597906d00cSAndrea Arcangeli * 3460c1e8d7c6SMichel Lespinasse * mmap_lock in write mode is required in order to block all operations 34617906d00cSAndrea Arcangeli * that could modify pagetables and free pages without need of 346227ba0644SKirill A. Shutemov * altering the vma layout. It's also needed in write mode to avoid new 34637906d00cSAndrea Arcangeli * anon_vmas to be associated with existing vmas. 34647906d00cSAndrea Arcangeli * 34657906d00cSAndrea Arcangeli * A single task can't take more than one mm_take_all_locks() in a row 34667906d00cSAndrea Arcangeli * or it would deadlock. 34677906d00cSAndrea Arcangeli * 3468bf181b9fSMichel Lespinasse * The LSB in anon_vma->rb_root.rb_node and the AS_MM_ALL_LOCKS bitflag in 34697906d00cSAndrea Arcangeli * mapping->flags avoid to take the same lock twice, if more than one 34707906d00cSAndrea Arcangeli * vma in this mm is backed by the same anon_vma or address_space. 34717906d00cSAndrea Arcangeli * 347288f306b6SKirill A. Shutemov * We take locks in following order, accordingly to comment at beginning 347388f306b6SKirill A. Shutemov * of mm/rmap.c: 347488f306b6SKirill A. Shutemov * - all hugetlbfs_i_mmap_rwsem_key locks (aka mapping->i_mmap_rwsem for 347588f306b6SKirill A. Shutemov * hugetlb mapping); 347688f306b6SKirill A. Shutemov * - all i_mmap_rwsem locks; 347788f306b6SKirill A. Shutemov * - all anon_vma->rwseml 347888f306b6SKirill A. Shutemov * 347988f306b6SKirill A. Shutemov * We can take all locks within these types randomly because the VM code 348088f306b6SKirill A. Shutemov * doesn't nest them and we protected from parallel mm_take_all_locks() by 348188f306b6SKirill A. Shutemov * mm_all_locks_mutex. 34827906d00cSAndrea Arcangeli * 34837906d00cSAndrea Arcangeli * mm_take_all_locks() and mm_drop_all_locks are expensive operations 34847906d00cSAndrea Arcangeli * that may have to take thousand of locks. 34857906d00cSAndrea Arcangeli * 34867906d00cSAndrea Arcangeli * mm_take_all_locks() can fail if it's interrupted by signals. 34877906d00cSAndrea Arcangeli */ 34887906d00cSAndrea Arcangeli int mm_take_all_locks(struct mm_struct *mm) 34897906d00cSAndrea Arcangeli { 34907906d00cSAndrea Arcangeli struct vm_area_struct *vma; 34915beb4930SRik van Riel struct anon_vma_chain *avc; 3492763ecb03SLiam R. Howlett MA_STATE(mas, &mm->mm_mt, 0, 0); 34937906d00cSAndrea Arcangeli 3494325bca1fSRolf Eike Beer mmap_assert_write_locked(mm); 34957906d00cSAndrea Arcangeli 34967906d00cSAndrea Arcangeli mutex_lock(&mm_all_locks_mutex); 34977906d00cSAndrea Arcangeli 3498763ecb03SLiam R. Howlett mas_for_each(&mas, vma, ULONG_MAX) { 34997906d00cSAndrea Arcangeli if (signal_pending(current)) 35007906d00cSAndrea Arcangeli goto out_unlock; 350188f306b6SKirill A. Shutemov if (vma->vm_file && vma->vm_file->f_mapping && 350288f306b6SKirill A. Shutemov is_vm_hugetlb_page(vma)) 350388f306b6SKirill A. Shutemov vm_lock_mapping(mm, vma->vm_file->f_mapping); 350488f306b6SKirill A. Shutemov } 350588f306b6SKirill A. Shutemov 3506763ecb03SLiam R. Howlett mas_set(&mas, 0); 3507763ecb03SLiam R. Howlett mas_for_each(&mas, vma, ULONG_MAX) { 350888f306b6SKirill A. Shutemov if (signal_pending(current)) 350988f306b6SKirill A. Shutemov goto out_unlock; 351088f306b6SKirill A. Shutemov if (vma->vm_file && vma->vm_file->f_mapping && 351188f306b6SKirill A. Shutemov !is_vm_hugetlb_page(vma)) 3512454ed842SPeter Zijlstra vm_lock_mapping(mm, vma->vm_file->f_mapping); 35137906d00cSAndrea Arcangeli } 35147cd5a02fSPeter Zijlstra 3515763ecb03SLiam R. Howlett mas_set(&mas, 0); 3516763ecb03SLiam R. Howlett mas_for_each(&mas, vma, ULONG_MAX) { 35177cd5a02fSPeter Zijlstra if (signal_pending(current)) 35187cd5a02fSPeter Zijlstra goto out_unlock; 35197cd5a02fSPeter Zijlstra if (vma->anon_vma) 35205beb4930SRik van Riel list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) 35215beb4930SRik van Riel vm_lock_anon_vma(mm, avc->anon_vma); 35227cd5a02fSPeter Zijlstra } 35237cd5a02fSPeter Zijlstra 3524584cff54SKautuk Consul return 0; 35257906d00cSAndrea Arcangeli 35267906d00cSAndrea Arcangeli out_unlock: 35277906d00cSAndrea Arcangeli mm_drop_all_locks(mm); 3528584cff54SKautuk Consul return -EINTR; 35297906d00cSAndrea Arcangeli } 35307906d00cSAndrea Arcangeli 35317906d00cSAndrea Arcangeli static void vm_unlock_anon_vma(struct anon_vma *anon_vma) 35327906d00cSAndrea Arcangeli { 3533f808c13fSDavidlohr Bueso if (test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_root.rb_node)) { 35347906d00cSAndrea Arcangeli /* 35357906d00cSAndrea Arcangeli * The LSB of head.next can't change to 0 from under 35367906d00cSAndrea Arcangeli * us because we hold the mm_all_locks_mutex. 35377906d00cSAndrea Arcangeli * 35387906d00cSAndrea Arcangeli * We must however clear the bitflag before unlocking 3539bf181b9fSMichel Lespinasse * the vma so the users using the anon_vma->rb_root will 35407906d00cSAndrea Arcangeli * never see our bitflag. 35417906d00cSAndrea Arcangeli * 35427906d00cSAndrea Arcangeli * No need of atomic instructions here, head.next 35437906d00cSAndrea Arcangeli * can't change from under us until we release the 35445a505085SIngo Molnar * anon_vma->root->rwsem. 35457906d00cSAndrea Arcangeli */ 35467906d00cSAndrea Arcangeli if (!__test_and_clear_bit(0, (unsigned long *) 3547f808c13fSDavidlohr Bueso &anon_vma->root->rb_root.rb_root.rb_node)) 35487906d00cSAndrea Arcangeli BUG(); 354908b52706SKonstantin Khlebnikov anon_vma_unlock_write(anon_vma); 35507906d00cSAndrea Arcangeli } 35517906d00cSAndrea Arcangeli } 35527906d00cSAndrea Arcangeli 35537906d00cSAndrea Arcangeli static void vm_unlock_mapping(struct address_space *mapping) 35547906d00cSAndrea Arcangeli { 35557906d00cSAndrea Arcangeli if (test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) { 35567906d00cSAndrea Arcangeli /* 35577906d00cSAndrea Arcangeli * AS_MM_ALL_LOCKS can't change to 0 from under us 35587906d00cSAndrea Arcangeli * because we hold the mm_all_locks_mutex. 35597906d00cSAndrea Arcangeli */ 356083cde9e8SDavidlohr Bueso i_mmap_unlock_write(mapping); 35617906d00cSAndrea Arcangeli if (!test_and_clear_bit(AS_MM_ALL_LOCKS, 35627906d00cSAndrea Arcangeli &mapping->flags)) 35637906d00cSAndrea Arcangeli BUG(); 35647906d00cSAndrea Arcangeli } 35657906d00cSAndrea Arcangeli } 35667906d00cSAndrea Arcangeli 35677906d00cSAndrea Arcangeli /* 3568c1e8d7c6SMichel Lespinasse * The mmap_lock cannot be released by the caller until 35697906d00cSAndrea Arcangeli * mm_drop_all_locks() returns. 35707906d00cSAndrea Arcangeli */ 35717906d00cSAndrea Arcangeli void mm_drop_all_locks(struct mm_struct *mm) 35727906d00cSAndrea Arcangeli { 35737906d00cSAndrea Arcangeli struct vm_area_struct *vma; 35745beb4930SRik van Riel struct anon_vma_chain *avc; 3575763ecb03SLiam R. Howlett MA_STATE(mas, &mm->mm_mt, 0, 0); 35767906d00cSAndrea Arcangeli 3577325bca1fSRolf Eike Beer mmap_assert_write_locked(mm); 35787906d00cSAndrea Arcangeli BUG_ON(!mutex_is_locked(&mm_all_locks_mutex)); 35797906d00cSAndrea Arcangeli 3580763ecb03SLiam R. Howlett mas_for_each(&mas, vma, ULONG_MAX) { 35817906d00cSAndrea Arcangeli if (vma->anon_vma) 35825beb4930SRik van Riel list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) 35835beb4930SRik van Riel vm_unlock_anon_vma(avc->anon_vma); 35847906d00cSAndrea Arcangeli if (vma->vm_file && vma->vm_file->f_mapping) 35857906d00cSAndrea Arcangeli vm_unlock_mapping(vma->vm_file->f_mapping); 35867906d00cSAndrea Arcangeli } 35877906d00cSAndrea Arcangeli 35887906d00cSAndrea Arcangeli mutex_unlock(&mm_all_locks_mutex); 35897906d00cSAndrea Arcangeli } 35908feae131SDavid Howells 35918feae131SDavid Howells /* 35923edf41d8Sseokhoon.yoon * initialise the percpu counter for VM 35938feae131SDavid Howells */ 35948feae131SDavid Howells void __init mmap_init(void) 35958feae131SDavid Howells { 359600a62ce9SKOSAKI Motohiro int ret; 359700a62ce9SKOSAKI Motohiro 3598908c7f19STejun Heo ret = percpu_counter_init(&vm_committed_as, 0, GFP_KERNEL); 359900a62ce9SKOSAKI Motohiro VM_BUG_ON(ret); 36008feae131SDavid Howells } 3601c9b1d098SAndrew Shewmaker 3602c9b1d098SAndrew Shewmaker /* 3603c9b1d098SAndrew Shewmaker * Initialise sysctl_user_reserve_kbytes. 3604c9b1d098SAndrew Shewmaker * 3605c9b1d098SAndrew Shewmaker * This is intended to prevent a user from starting a single memory hogging 3606c9b1d098SAndrew Shewmaker * process, such that they cannot recover (kill the hog) in OVERCOMMIT_NEVER 3607c9b1d098SAndrew Shewmaker * mode. 3608c9b1d098SAndrew Shewmaker * 3609c9b1d098SAndrew Shewmaker * The default value is min(3% of free memory, 128MB) 3610c9b1d098SAndrew Shewmaker * 128MB is enough to recover with sshd/login, bash, and top/kill. 3611c9b1d098SAndrew Shewmaker */ 36121640879aSAndrew Shewmaker static int init_user_reserve(void) 3613c9b1d098SAndrew Shewmaker { 3614c9b1d098SAndrew Shewmaker unsigned long free_kbytes; 3615c9b1d098SAndrew Shewmaker 3616c41f012aSMichal Hocko free_kbytes = global_zone_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10); 3617c9b1d098SAndrew Shewmaker 3618c9b1d098SAndrew Shewmaker sysctl_user_reserve_kbytes = min(free_kbytes / 32, 1UL << 17); 3619c9b1d098SAndrew Shewmaker return 0; 3620c9b1d098SAndrew Shewmaker } 3621a64fb3cdSPaul Gortmaker subsys_initcall(init_user_reserve); 36224eeab4f5SAndrew Shewmaker 36234eeab4f5SAndrew Shewmaker /* 36244eeab4f5SAndrew Shewmaker * Initialise sysctl_admin_reserve_kbytes. 36254eeab4f5SAndrew Shewmaker * 36264eeab4f5SAndrew Shewmaker * The purpose of sysctl_admin_reserve_kbytes is to allow the sys admin 36274eeab4f5SAndrew Shewmaker * to log in and kill a memory hogging process. 36284eeab4f5SAndrew Shewmaker * 36294eeab4f5SAndrew Shewmaker * Systems with more than 256MB will reserve 8MB, enough to recover 36304eeab4f5SAndrew Shewmaker * with sshd, bash, and top in OVERCOMMIT_GUESS. Smaller systems will 36314eeab4f5SAndrew Shewmaker * only reserve 3% of free pages by default. 36324eeab4f5SAndrew Shewmaker */ 36331640879aSAndrew Shewmaker static int init_admin_reserve(void) 36344eeab4f5SAndrew Shewmaker { 36354eeab4f5SAndrew Shewmaker unsigned long free_kbytes; 36364eeab4f5SAndrew Shewmaker 3637c41f012aSMichal Hocko free_kbytes = global_zone_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10); 36384eeab4f5SAndrew Shewmaker 36394eeab4f5SAndrew Shewmaker sysctl_admin_reserve_kbytes = min(free_kbytes / 32, 1UL << 13); 36404eeab4f5SAndrew Shewmaker return 0; 36414eeab4f5SAndrew Shewmaker } 3642a64fb3cdSPaul Gortmaker subsys_initcall(init_admin_reserve); 36431640879aSAndrew Shewmaker 36441640879aSAndrew Shewmaker /* 36451640879aSAndrew Shewmaker * Reinititalise user and admin reserves if memory is added or removed. 36461640879aSAndrew Shewmaker * 36471640879aSAndrew Shewmaker * The default user reserve max is 128MB, and the default max for the 36481640879aSAndrew Shewmaker * admin reserve is 8MB. These are usually, but not always, enough to 36491640879aSAndrew Shewmaker * enable recovery from a memory hogging process using login/sshd, a shell, 36501640879aSAndrew Shewmaker * and tools like top. It may make sense to increase or even disable the 36511640879aSAndrew Shewmaker * reserve depending on the existence of swap or variations in the recovery 36521640879aSAndrew Shewmaker * tools. So, the admin may have changed them. 36531640879aSAndrew Shewmaker * 36541640879aSAndrew Shewmaker * If memory is added and the reserves have been eliminated or increased above 36551640879aSAndrew Shewmaker * the default max, then we'll trust the admin. 36561640879aSAndrew Shewmaker * 36571640879aSAndrew Shewmaker * If memory is removed and there isn't enough free memory, then we 36581640879aSAndrew Shewmaker * need to reset the reserves. 36591640879aSAndrew Shewmaker * 36601640879aSAndrew Shewmaker * Otherwise keep the reserve set by the admin. 36611640879aSAndrew Shewmaker */ 36621640879aSAndrew Shewmaker static int reserve_mem_notifier(struct notifier_block *nb, 36631640879aSAndrew Shewmaker unsigned long action, void *data) 36641640879aSAndrew Shewmaker { 36651640879aSAndrew Shewmaker unsigned long tmp, free_kbytes; 36661640879aSAndrew Shewmaker 36671640879aSAndrew Shewmaker switch (action) { 36681640879aSAndrew Shewmaker case MEM_ONLINE: 36691640879aSAndrew Shewmaker /* Default max is 128MB. Leave alone if modified by operator. */ 36701640879aSAndrew Shewmaker tmp = sysctl_user_reserve_kbytes; 36711640879aSAndrew Shewmaker if (0 < tmp && tmp < (1UL << 17)) 36721640879aSAndrew Shewmaker init_user_reserve(); 36731640879aSAndrew Shewmaker 36741640879aSAndrew Shewmaker /* Default max is 8MB. Leave alone if modified by operator. */ 36751640879aSAndrew Shewmaker tmp = sysctl_admin_reserve_kbytes; 36761640879aSAndrew Shewmaker if (0 < tmp && tmp < (1UL << 13)) 36771640879aSAndrew Shewmaker init_admin_reserve(); 36781640879aSAndrew Shewmaker 36791640879aSAndrew Shewmaker break; 36801640879aSAndrew Shewmaker case MEM_OFFLINE: 3681c41f012aSMichal Hocko free_kbytes = global_zone_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10); 36821640879aSAndrew Shewmaker 36831640879aSAndrew Shewmaker if (sysctl_user_reserve_kbytes > free_kbytes) { 36841640879aSAndrew Shewmaker init_user_reserve(); 36851640879aSAndrew Shewmaker pr_info("vm.user_reserve_kbytes reset to %lu\n", 36861640879aSAndrew Shewmaker sysctl_user_reserve_kbytes); 36871640879aSAndrew Shewmaker } 36881640879aSAndrew Shewmaker 36891640879aSAndrew Shewmaker if (sysctl_admin_reserve_kbytes > free_kbytes) { 36901640879aSAndrew Shewmaker init_admin_reserve(); 36911640879aSAndrew Shewmaker pr_info("vm.admin_reserve_kbytes reset to %lu\n", 36921640879aSAndrew Shewmaker sysctl_admin_reserve_kbytes); 36931640879aSAndrew Shewmaker } 36941640879aSAndrew Shewmaker break; 36951640879aSAndrew Shewmaker default: 36961640879aSAndrew Shewmaker break; 36971640879aSAndrew Shewmaker } 36981640879aSAndrew Shewmaker return NOTIFY_OK; 36991640879aSAndrew Shewmaker } 37001640879aSAndrew Shewmaker 37011640879aSAndrew Shewmaker static int __meminit init_reserve_notifier(void) 37021640879aSAndrew Shewmaker { 37031eeaa4fdSLiu Shixin if (hotplug_memory_notifier(reserve_mem_notifier, DEFAULT_CALLBACK_PRI)) 3704b1de0d13SMitchel Humpherys pr_err("Failed registering memory add/remove notifier for admin reserve\n"); 37051640879aSAndrew Shewmaker 37061640879aSAndrew Shewmaker return 0; 37071640879aSAndrew Shewmaker } 3708a64fb3cdSPaul Gortmaker subsys_initcall(init_reserve_notifier); 3709