1457c8996SThomas Gleixner // SPDX-License-Identifier: GPL-2.0-only 21da177e4SLinus Torvalds /* 31da177e4SLinus Torvalds * mm/mmap.c 41da177e4SLinus Torvalds * 51da177e4SLinus Torvalds * Written by obz. 61da177e4SLinus Torvalds * 7046c6884SAlan Cox * Address space accounting code <alan@lxorguk.ukuu.org.uk> 81da177e4SLinus Torvalds */ 91da177e4SLinus Torvalds 10b1de0d13SMitchel Humpherys #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt 11b1de0d13SMitchel Humpherys 12e8420a8eSCyril Hrubis #include <linux/kernel.h> 131da177e4SLinus Torvalds #include <linux/slab.h> 144af3c9ccSAlexey Dobriyan #include <linux/backing-dev.h> 151da177e4SLinus Torvalds #include <linux/mm.h> 1617fca131SArnd Bergmann #include <linux/mm_inline.h> 171da177e4SLinus Torvalds #include <linux/shm.h> 181da177e4SLinus Torvalds #include <linux/mman.h> 191da177e4SLinus Torvalds #include <linux/pagemap.h> 201da177e4SLinus Torvalds #include <linux/swap.h> 211da177e4SLinus Torvalds #include <linux/syscalls.h> 22c59ede7bSRandy.Dunlap #include <linux/capability.h> 231da177e4SLinus Torvalds #include <linux/init.h> 241da177e4SLinus Torvalds #include <linux/file.h> 251da177e4SLinus Torvalds #include <linux/fs.h> 261da177e4SLinus Torvalds #include <linux/personality.h> 271da177e4SLinus Torvalds #include <linux/security.h> 281da177e4SLinus Torvalds #include <linux/hugetlb.h> 29c01d5b30SHugh Dickins #include <linux/shmem_fs.h> 301da177e4SLinus Torvalds #include <linux/profile.h> 31b95f1b31SPaul Gortmaker #include <linux/export.h> 321da177e4SLinus Torvalds #include <linux/mount.h> 331da177e4SLinus Torvalds #include <linux/mempolicy.h> 341da177e4SLinus Torvalds #include <linux/rmap.h> 35cddb8a5cSAndrea Arcangeli #include <linux/mmu_notifier.h> 3682f71ae4SKonstantin Khlebnikov #include <linux/mmdebug.h> 37cdd6c482SIngo Molnar #include <linux/perf_event.h> 38120a795dSAl Viro #include <linux/audit.h> 39b15d00b6SAndrea Arcangeli #include <linux/khugepaged.h> 402b144498SSrikar Dronamraju #include <linux/uprobes.h> 411640879aSAndrew Shewmaker #include <linux/notifier.h> 421640879aSAndrew Shewmaker #include <linux/memory.h> 43b1de0d13SMitchel Humpherys #include <linux/printk.h> 4419a809afSAndrea Arcangeli #include <linux/userfaultfd_k.h> 45d977d56cSKonstantin Khlebnikov #include <linux/moduleparam.h> 4662b5f7d0SDave Hansen #include <linux/pkeys.h> 4721292580SAndrea Arcangeli #include <linux/oom.h> 4804f5866eSAndrea Arcangeli #include <linux/sched/mm.h> 491da177e4SLinus Torvalds 507c0f6ba6SLinus Torvalds #include <linux/uaccess.h> 511da177e4SLinus Torvalds #include <asm/cacheflush.h> 521da177e4SLinus Torvalds #include <asm/tlb.h> 53d6dd61c8SJeremy Fitzhardinge #include <asm/mmu_context.h> 541da177e4SLinus Torvalds 55df529cabSJaewon Kim #define CREATE_TRACE_POINTS 56df529cabSJaewon Kim #include <trace/events/mmap.h> 57df529cabSJaewon Kim 5842b77728SJan Beulich #include "internal.h" 5942b77728SJan Beulich 603a459756SKirill Korotaev #ifndef arch_mmap_check 613a459756SKirill Korotaev #define arch_mmap_check(addr, len, flags) (0) 623a459756SKirill Korotaev #endif 633a459756SKirill Korotaev 64d07e2259SDaniel Cashman #ifdef CONFIG_HAVE_ARCH_MMAP_RND_BITS 65d07e2259SDaniel Cashman const int mmap_rnd_bits_min = CONFIG_ARCH_MMAP_RND_BITS_MIN; 66d07e2259SDaniel Cashman const int mmap_rnd_bits_max = CONFIG_ARCH_MMAP_RND_BITS_MAX; 67d07e2259SDaniel Cashman int mmap_rnd_bits __read_mostly = CONFIG_ARCH_MMAP_RND_BITS; 68d07e2259SDaniel Cashman #endif 69d07e2259SDaniel Cashman #ifdef CONFIG_HAVE_ARCH_MMAP_RND_COMPAT_BITS 70d07e2259SDaniel Cashman const int mmap_rnd_compat_bits_min = CONFIG_ARCH_MMAP_RND_COMPAT_BITS_MIN; 71d07e2259SDaniel Cashman const int mmap_rnd_compat_bits_max = CONFIG_ARCH_MMAP_RND_COMPAT_BITS_MAX; 72d07e2259SDaniel Cashman int mmap_rnd_compat_bits __read_mostly = CONFIG_ARCH_MMAP_RND_COMPAT_BITS; 73d07e2259SDaniel Cashman #endif 74d07e2259SDaniel Cashman 75f4fcd558SKonstantin Khlebnikov static bool ignore_rlimit_data; 76d977d56cSKonstantin Khlebnikov core_param(ignore_rlimit_data, ignore_rlimit_data, bool, 0644); 77d07e2259SDaniel Cashman 78763ecb03SLiam R. Howlett static void unmap_region(struct mm_struct *mm, struct maple_tree *mt, 79e0da382cSHugh Dickins struct vm_area_struct *vma, struct vm_area_struct *prev, 80763ecb03SLiam R. Howlett struct vm_area_struct *next, unsigned long start, 81763ecb03SLiam R. Howlett unsigned long end); 82e0da382cSHugh Dickins 8364e45507SPeter Feiner static pgprot_t vm_pgprot_modify(pgprot_t oldprot, unsigned long vm_flags) 8464e45507SPeter Feiner { 8564e45507SPeter Feiner return pgprot_modify(oldprot, vm_get_page_prot(vm_flags)); 8664e45507SPeter Feiner } 8764e45507SPeter Feiner 8864e45507SPeter Feiner /* Update vma->vm_page_prot to reflect vma->vm_flags. */ 8964e45507SPeter Feiner void vma_set_page_prot(struct vm_area_struct *vma) 9064e45507SPeter Feiner { 9164e45507SPeter Feiner unsigned long vm_flags = vma->vm_flags; 926d2329f8SAndrea Arcangeli pgprot_t vm_page_prot; 9364e45507SPeter Feiner 946d2329f8SAndrea Arcangeli vm_page_prot = vm_pgprot_modify(vma->vm_page_prot, vm_flags); 956d2329f8SAndrea Arcangeli if (vma_wants_writenotify(vma, vm_page_prot)) { 9664e45507SPeter Feiner vm_flags &= ~VM_SHARED; 976d2329f8SAndrea Arcangeli vm_page_prot = vm_pgprot_modify(vm_page_prot, vm_flags); 9864e45507SPeter Feiner } 99c1e8d7c6SMichel Lespinasse /* remove_protection_ptes reads vma->vm_page_prot without mmap_lock */ 1006d2329f8SAndrea Arcangeli WRITE_ONCE(vma->vm_page_prot, vm_page_prot); 10164e45507SPeter Feiner } 10264e45507SPeter Feiner 1031da177e4SLinus Torvalds /* 104c8c06efaSDavidlohr Bueso * Requires inode->i_mapping->i_mmap_rwsem 1051da177e4SLinus Torvalds */ 1061da177e4SLinus Torvalds static void __remove_shared_vm_struct(struct vm_area_struct *vma, 1071da177e4SLinus Torvalds struct file *file, struct address_space *mapping) 1081da177e4SLinus Torvalds { 1091da177e4SLinus Torvalds if (vma->vm_flags & VM_SHARED) 1104bb5f5d9SDavid Herrmann mapping_unmap_writable(mapping); 1111da177e4SLinus Torvalds 1121da177e4SLinus Torvalds flush_dcache_mmap_lock(mapping); 1136b2dbba8SMichel Lespinasse vma_interval_tree_remove(vma, &mapping->i_mmap); 1141da177e4SLinus Torvalds flush_dcache_mmap_unlock(mapping); 1151da177e4SLinus Torvalds } 1161da177e4SLinus Torvalds 1171da177e4SLinus Torvalds /* 1186b2dbba8SMichel Lespinasse * Unlink a file-based vm structure from its interval tree, to hide 119a8fb5618SHugh Dickins * vma from rmap and vmtruncate before freeing its page tables. 1201da177e4SLinus Torvalds */ 121a8fb5618SHugh Dickins void unlink_file_vma(struct vm_area_struct *vma) 1221da177e4SLinus Torvalds { 1231da177e4SLinus Torvalds struct file *file = vma->vm_file; 1241da177e4SLinus Torvalds 1251da177e4SLinus Torvalds if (file) { 1261da177e4SLinus Torvalds struct address_space *mapping = file->f_mapping; 12783cde9e8SDavidlohr Bueso i_mmap_lock_write(mapping); 1281da177e4SLinus Torvalds __remove_shared_vm_struct(vma, file, mapping); 12983cde9e8SDavidlohr Bueso i_mmap_unlock_write(mapping); 1301da177e4SLinus Torvalds } 131a8fb5618SHugh Dickins } 132a8fb5618SHugh Dickins 133a8fb5618SHugh Dickins /* 134763ecb03SLiam R. Howlett * Close a vm structure and free it. 135a8fb5618SHugh Dickins */ 136763ecb03SLiam R. Howlett static void remove_vma(struct vm_area_struct *vma) 137a8fb5618SHugh Dickins { 138a8fb5618SHugh Dickins might_sleep(); 1391da177e4SLinus Torvalds if (vma->vm_ops && vma->vm_ops->close) 1401da177e4SLinus Torvalds vma->vm_ops->close(vma); 141e9714acfSKonstantin Khlebnikov if (vma->vm_file) 142a8fb5618SHugh Dickins fput(vma->vm_file); 143f0be3d32SLee Schermerhorn mpol_put(vma_policy(vma)); 1443928d4f5SLinus Torvalds vm_area_free(vma); 1451da177e4SLinus Torvalds } 1461da177e4SLinus Torvalds 147b62b633eSLiam R. Howlett static inline struct vm_area_struct *vma_prev_limit(struct vma_iterator *vmi, 148b62b633eSLiam R. Howlett unsigned long min) 149b62b633eSLiam R. Howlett { 150b62b633eSLiam R. Howlett return mas_prev(&vmi->mas, min); 151b62b633eSLiam R. Howlett } 152b62b633eSLiam R. Howlett 153b62b633eSLiam R. Howlett static inline int vma_iter_clear_gfp(struct vma_iterator *vmi, 154b62b633eSLiam R. Howlett unsigned long start, unsigned long end, gfp_t gfp) 155b62b633eSLiam R. Howlett { 156b62b633eSLiam R. Howlett vmi->mas.index = start; 157b62b633eSLiam R. Howlett vmi->mas.last = end - 1; 158b62b633eSLiam R. Howlett mas_store_gfp(&vmi->mas, NULL, gfp); 159b62b633eSLiam R. Howlett if (unlikely(mas_is_err(&vmi->mas))) 160b62b633eSLiam R. Howlett return -ENOMEM; 161b62b633eSLiam R. Howlett 162b62b633eSLiam R. Howlett return 0; 163b62b633eSLiam R. Howlett } 164b62b633eSLiam R. Howlett 1652e7ce7d3SLiam R. Howlett /* 1662e7ce7d3SLiam R. Howlett * check_brk_limits() - Use platform specific check of range & verify mlock 1672e7ce7d3SLiam R. Howlett * limits. 1682e7ce7d3SLiam R. Howlett * @addr: The address to check 1692e7ce7d3SLiam R. Howlett * @len: The size of increase. 1702e7ce7d3SLiam R. Howlett * 1712e7ce7d3SLiam R. Howlett * Return: 0 on success. 1722e7ce7d3SLiam R. Howlett */ 1732e7ce7d3SLiam R. Howlett static int check_brk_limits(unsigned long addr, unsigned long len) 1742e7ce7d3SLiam R. Howlett { 1752e7ce7d3SLiam R. Howlett unsigned long mapped_addr; 1762e7ce7d3SLiam R. Howlett 1772e7ce7d3SLiam R. Howlett mapped_addr = get_unmapped_area(NULL, addr, len, 0, MAP_FIXED); 1782e7ce7d3SLiam R. Howlett if (IS_ERR_VALUE(mapped_addr)) 1792e7ce7d3SLiam R. Howlett return mapped_addr; 1802e7ce7d3SLiam R. Howlett 1812e7ce7d3SLiam R. Howlett return mlock_future_check(current->mm, current->mm->def_flags, len); 1822e7ce7d3SLiam R. Howlett } 18392fed820SLiam R. Howlett static int do_brk_flags(struct vma_iterator *vmi, struct vm_area_struct *brkvma, 184763ecb03SLiam R. Howlett unsigned long addr, unsigned long request, unsigned long flags); 1856a6160a7SHeiko Carstens SYSCALL_DEFINE1(brk, unsigned long, brk) 1861da177e4SLinus Torvalds { 1879bc8039eSYang Shi unsigned long newbrk, oldbrk, origbrk; 1881da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 1892e7ce7d3SLiam R. Howlett struct vm_area_struct *brkvma, *next = NULL; 190a5b4592cSJiri Kosina unsigned long min_brk; 191128557ffSMichel Lespinasse bool populate; 1929bc8039eSYang Shi bool downgraded = false; 193897ab3e0SMike Rapoport LIST_HEAD(uf); 19492fed820SLiam R. Howlett struct vma_iterator vmi; 1951da177e4SLinus Torvalds 196d8ed45c5SMichel Lespinasse if (mmap_write_lock_killable(mm)) 197dc0ef0dfSMichal Hocko return -EINTR; 1981da177e4SLinus Torvalds 1999bc8039eSYang Shi origbrk = mm->brk; 2009bc8039eSYang Shi 201a5b4592cSJiri Kosina #ifdef CONFIG_COMPAT_BRK 2025520e894SJiri Kosina /* 2035520e894SJiri Kosina * CONFIG_COMPAT_BRK can still be overridden by setting 2045520e894SJiri Kosina * randomize_va_space to 2, which will still cause mm->start_brk 2055520e894SJiri Kosina * to be arbitrarily shifted 2065520e894SJiri Kosina */ 2074471a675SJiri Kosina if (current->brk_randomized) 2085520e894SJiri Kosina min_brk = mm->start_brk; 2095520e894SJiri Kosina else 2105520e894SJiri Kosina min_brk = mm->end_data; 211a5b4592cSJiri Kosina #else 212a5b4592cSJiri Kosina min_brk = mm->start_brk; 213a5b4592cSJiri Kosina #endif 214a5b4592cSJiri Kosina if (brk < min_brk) 2151da177e4SLinus Torvalds goto out; 2161e624196SRam Gupta 2171e624196SRam Gupta /* 2181e624196SRam Gupta * Check against rlimit here. If this check is done later after the test 2191e624196SRam Gupta * of oldbrk with newbrk then it can escape the test and let the data 2201e624196SRam Gupta * segment grow beyond its set limit the in case where the limit is 2211e624196SRam Gupta * not page aligned -Ram Gupta 2221e624196SRam Gupta */ 2238764b338SCyrill Gorcunov if (check_data_rlimit(rlimit(RLIMIT_DATA), brk, mm->start_brk, 2248764b338SCyrill Gorcunov mm->end_data, mm->start_data)) 2251e624196SRam Gupta goto out; 2261e624196SRam Gupta 2271da177e4SLinus Torvalds newbrk = PAGE_ALIGN(brk); 2281da177e4SLinus Torvalds oldbrk = PAGE_ALIGN(mm->brk); 2299bc8039eSYang Shi if (oldbrk == newbrk) { 2309bc8039eSYang Shi mm->brk = brk; 2319bc8039eSYang Shi goto success; 2329bc8039eSYang Shi } 2331da177e4SLinus Torvalds 2349bc8039eSYang Shi /* 2359bc8039eSYang Shi * Always allow shrinking brk. 23627b26701SLiam R. Howlett * do_vma_munmap() may downgrade mmap_lock to read. 2379bc8039eSYang Shi */ 2381da177e4SLinus Torvalds if (brk <= mm->brk) { 2399bc8039eSYang Shi int ret; 2409bc8039eSYang Shi 2412e7ce7d3SLiam R. Howlett /* Search one past newbrk */ 24292fed820SLiam R. Howlett vma_iter_init(&vmi, mm, newbrk); 24392fed820SLiam R. Howlett brkvma = vma_find(&vmi, oldbrk); 244f5ad5083SJason A. Donenfeld if (!brkvma || brkvma->vm_start >= oldbrk) 2452e7ce7d3SLiam R. Howlett goto out; /* mapping intersects with an existing non-brk vma. */ 2469bc8039eSYang Shi /* 2472e7ce7d3SLiam R. Howlett * mm->brk must be protected by write mmap_lock. 24827b26701SLiam R. Howlett * do_vma_munmap() may downgrade the lock, so update it 24927b26701SLiam R. Howlett * before calling do_vma_munmap(). 2509bc8039eSYang Shi */ 2519bc8039eSYang Shi mm->brk = brk; 25227b26701SLiam R. Howlett ret = do_vma_munmap(&vmi, brkvma, newbrk, oldbrk, &uf, true); 2532e7ce7d3SLiam R. Howlett if (ret == 1) { 2542e7ce7d3SLiam R. Howlett downgraded = true; 2552e7ce7d3SLiam R. Howlett goto success; 2562e7ce7d3SLiam R. Howlett } else if (!ret) 2572e7ce7d3SLiam R. Howlett goto success; 2582e7ce7d3SLiam R. Howlett 2599bc8039eSYang Shi mm->brk = origbrk; 2601da177e4SLinus Torvalds goto out; 2611da177e4SLinus Torvalds } 2621da177e4SLinus Torvalds 2632e7ce7d3SLiam R. Howlett if (check_brk_limits(oldbrk, newbrk - oldbrk)) 2642e7ce7d3SLiam R. Howlett goto out; 2652e7ce7d3SLiam R. Howlett 2662e7ce7d3SLiam R. Howlett /* 2672e7ce7d3SLiam R. Howlett * Only check if the next VMA is within the stack_guard_gap of the 2682e7ce7d3SLiam R. Howlett * expansion area 2692e7ce7d3SLiam R. Howlett */ 27092fed820SLiam R. Howlett vma_iter_init(&vmi, mm, oldbrk); 27192fed820SLiam R. Howlett next = vma_find(&vmi, newbrk + PAGE_SIZE + stack_guard_gap); 2721be7107fSHugh Dickins if (next && newbrk + PAGE_SIZE > vm_start_gap(next)) 2731da177e4SLinus Torvalds goto out; 2741da177e4SLinus Torvalds 27592fed820SLiam R. Howlett brkvma = vma_prev_limit(&vmi, mm->start_brk); 2761da177e4SLinus Torvalds /* Ok, looks good - let it rip. */ 27792fed820SLiam R. Howlett if (do_brk_flags(&vmi, brkvma, oldbrk, newbrk - oldbrk, 0) < 0) 2781da177e4SLinus Torvalds goto out; 2792e7ce7d3SLiam R. Howlett 2801da177e4SLinus Torvalds mm->brk = brk; 2819bc8039eSYang Shi 2829bc8039eSYang Shi success: 283128557ffSMichel Lespinasse populate = newbrk > oldbrk && (mm->def_flags & VM_LOCKED) != 0; 2849bc8039eSYang Shi if (downgraded) 285d8ed45c5SMichel Lespinasse mmap_read_unlock(mm); 2869bc8039eSYang Shi else 287d8ed45c5SMichel Lespinasse mmap_write_unlock(mm); 288897ab3e0SMike Rapoport userfaultfd_unmap_complete(mm, &uf); 289128557ffSMichel Lespinasse if (populate) 290128557ffSMichel Lespinasse mm_populate(oldbrk, newbrk - oldbrk); 291128557ffSMichel Lespinasse return brk; 292128557ffSMichel Lespinasse 2931da177e4SLinus Torvalds out: 294d8ed45c5SMichel Lespinasse mmap_write_unlock(mm); 295b7204006SAdrian Huang return origbrk; 2961da177e4SLinus Torvalds } 2971da177e4SLinus Torvalds 298d4af56c5SLiam R. Howlett #if defined(CONFIG_DEBUG_VM_MAPLE_TREE) 299d4af56c5SLiam R. Howlett extern void mt_validate(struct maple_tree *mt); 300d4af56c5SLiam R. Howlett extern void mt_dump(const struct maple_tree *mt); 3011da177e4SLinus Torvalds 302d4af56c5SLiam R. Howlett /* Validate the maple tree */ 303d4af56c5SLiam R. Howlett static void validate_mm_mt(struct mm_struct *mm) 304d4af56c5SLiam R. Howlett { 305d4af56c5SLiam R. Howlett struct maple_tree *mt = &mm->mm_mt; 306763ecb03SLiam R. Howlett struct vm_area_struct *vma_mt; 307d4af56c5SLiam R. Howlett 308d4af56c5SLiam R. Howlett MA_STATE(mas, mt, 0, 0); 309d4af56c5SLiam R. Howlett 310d4af56c5SLiam R. Howlett mt_validate(&mm->mm_mt); 311d4af56c5SLiam R. Howlett mas_for_each(&mas, vma_mt, ULONG_MAX) { 312763ecb03SLiam R. Howlett if ((vma_mt->vm_start != mas.index) || 313763ecb03SLiam R. Howlett (vma_mt->vm_end - 1 != mas.last)) { 314d4af56c5SLiam R. Howlett pr_emerg("issue in %s\n", current->comm); 315d4af56c5SLiam R. Howlett dump_stack(); 316d4af56c5SLiam R. Howlett dump_vma(vma_mt); 317d4af56c5SLiam R. Howlett pr_emerg("mt piv: %p %lu - %lu\n", vma_mt, 318d4af56c5SLiam R. Howlett mas.index, mas.last); 319d4af56c5SLiam R. Howlett pr_emerg("mt vma: %p %lu - %lu\n", vma_mt, 320d4af56c5SLiam R. Howlett vma_mt->vm_start, vma_mt->vm_end); 321d4af56c5SLiam R. Howlett 322d4af56c5SLiam R. Howlett mt_dump(mas.tree); 323d4af56c5SLiam R. Howlett if (vma_mt->vm_end != mas.last + 1) { 324d4af56c5SLiam R. Howlett pr_err("vma: %p vma_mt %lu-%lu\tmt %lu-%lu\n", 325d4af56c5SLiam R. Howlett mm, vma_mt->vm_start, vma_mt->vm_end, 326d4af56c5SLiam R. Howlett mas.index, mas.last); 327d4af56c5SLiam R. Howlett mt_dump(mas.tree); 328d4af56c5SLiam R. Howlett } 329d4af56c5SLiam R. Howlett VM_BUG_ON_MM(vma_mt->vm_end != mas.last + 1, mm); 330d4af56c5SLiam R. Howlett if (vma_mt->vm_start != mas.index) { 331d4af56c5SLiam R. Howlett pr_err("vma: %p vma_mt %p %lu - %lu doesn't match\n", 332d4af56c5SLiam R. Howlett mm, vma_mt, vma_mt->vm_start, vma_mt->vm_end); 333d4af56c5SLiam R. Howlett mt_dump(mas.tree); 334d4af56c5SLiam R. Howlett } 335d4af56c5SLiam R. Howlett VM_BUG_ON_MM(vma_mt->vm_start != mas.index, mm); 336d4af56c5SLiam R. Howlett } 337d4af56c5SLiam R. Howlett } 338d4af56c5SLiam R. Howlett } 3391da177e4SLinus Torvalds 340eafd4dc4SRashika Kheria static void validate_mm(struct mm_struct *mm) 3411da177e4SLinus Torvalds { 3421da177e4SLinus Torvalds int bug = 0; 3431da177e4SLinus Torvalds int i = 0; 344763ecb03SLiam R. Howlett struct vm_area_struct *vma; 345763ecb03SLiam R. Howlett MA_STATE(mas, &mm->mm_mt, 0, 0); 346ff26f70fSAndrew Morton 347524e00b3SLiam R. Howlett validate_mm_mt(mm); 348524e00b3SLiam R. Howlett 349763ecb03SLiam R. Howlett mas_for_each(&mas, vma, ULONG_MAX) { 350524e00b3SLiam R. Howlett #ifdef CONFIG_DEBUG_VM_RB 35112352d3cSKonstantin Khlebnikov struct anon_vma *anon_vma = vma->anon_vma; 352ed8ea815SMichel Lespinasse struct anon_vma_chain *avc; 353ff26f70fSAndrew Morton 35412352d3cSKonstantin Khlebnikov if (anon_vma) { 35512352d3cSKonstantin Khlebnikov anon_vma_lock_read(anon_vma); 356ed8ea815SMichel Lespinasse list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) 357ed8ea815SMichel Lespinasse anon_vma_interval_tree_verify(avc); 35812352d3cSKonstantin Khlebnikov anon_vma_unlock_read(anon_vma); 35912352d3cSKonstantin Khlebnikov } 360524e00b3SLiam R. Howlett #endif 3611da177e4SLinus Torvalds i++; 3621da177e4SLinus Torvalds } 3635a0768f6SMichel Lespinasse if (i != mm->map_count) { 364763ecb03SLiam R. Howlett pr_emerg("map_count %d mas_for_each %d\n", mm->map_count, i); 3655a0768f6SMichel Lespinasse bug = 1; 3665a0768f6SMichel Lespinasse } 36796dad67fSSasha Levin VM_BUG_ON_MM(bug, mm); 3681da177e4SLinus Torvalds } 369524e00b3SLiam R. Howlett 370524e00b3SLiam R. Howlett #else /* !CONFIG_DEBUG_VM_MAPLE_TREE */ 371d4af56c5SLiam R. Howlett #define validate_mm_mt(root) do { } while (0) 3721da177e4SLinus Torvalds #define validate_mm(mm) do { } while (0) 373524e00b3SLiam R. Howlett #endif /* CONFIG_DEBUG_VM_MAPLE_TREE */ 374d3737187SMichel Lespinasse 375bf181b9fSMichel Lespinasse /* 376bf181b9fSMichel Lespinasse * vma has some anon_vma assigned, and is already inserted on that 377bf181b9fSMichel Lespinasse * anon_vma's interval trees. 378bf181b9fSMichel Lespinasse * 379bf181b9fSMichel Lespinasse * Before updating the vma's vm_start / vm_end / vm_pgoff fields, the 380bf181b9fSMichel Lespinasse * vma must be removed from the anon_vma's interval trees using 381bf181b9fSMichel Lespinasse * anon_vma_interval_tree_pre_update_vma(). 382bf181b9fSMichel Lespinasse * 383bf181b9fSMichel Lespinasse * After the update, the vma will be reinserted using 384bf181b9fSMichel Lespinasse * anon_vma_interval_tree_post_update_vma(). 385bf181b9fSMichel Lespinasse * 386c1e8d7c6SMichel Lespinasse * The entire update must be protected by exclusive mmap_lock and by 387bf181b9fSMichel Lespinasse * the root anon_vma's mutex. 388bf181b9fSMichel Lespinasse */ 389bf181b9fSMichel Lespinasse static inline void 390bf181b9fSMichel Lespinasse anon_vma_interval_tree_pre_update_vma(struct vm_area_struct *vma) 391bf181b9fSMichel Lespinasse { 392bf181b9fSMichel Lespinasse struct anon_vma_chain *avc; 393bf181b9fSMichel Lespinasse 394bf181b9fSMichel Lespinasse list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) 395bf181b9fSMichel Lespinasse anon_vma_interval_tree_remove(avc, &avc->anon_vma->rb_root); 396bf181b9fSMichel Lespinasse } 397bf181b9fSMichel Lespinasse 398bf181b9fSMichel Lespinasse static inline void 399bf181b9fSMichel Lespinasse anon_vma_interval_tree_post_update_vma(struct vm_area_struct *vma) 400bf181b9fSMichel Lespinasse { 401bf181b9fSMichel Lespinasse struct anon_vma_chain *avc; 402bf181b9fSMichel Lespinasse 403bf181b9fSMichel Lespinasse list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) 404bf181b9fSMichel Lespinasse anon_vma_interval_tree_insert(avc, &avc->anon_vma->rb_root); 405bf181b9fSMichel Lespinasse } 406bf181b9fSMichel Lespinasse 407e8420a8eSCyril Hrubis static unsigned long count_vma_pages_range(struct mm_struct *mm, 408e8420a8eSCyril Hrubis unsigned long addr, unsigned long end) 409e8420a8eSCyril Hrubis { 4102e3af1dbSMatthew Wilcox (Oracle) VMA_ITERATOR(vmi, mm, addr); 411e8420a8eSCyril Hrubis struct vm_area_struct *vma; 4122e3af1dbSMatthew Wilcox (Oracle) unsigned long nr_pages = 0; 413e8420a8eSCyril Hrubis 4142e3af1dbSMatthew Wilcox (Oracle) for_each_vma_range(vmi, vma, end) { 4152e3af1dbSMatthew Wilcox (Oracle) unsigned long vm_start = max(addr, vma->vm_start); 4162e3af1dbSMatthew Wilcox (Oracle) unsigned long vm_end = min(end, vma->vm_end); 417e8420a8eSCyril Hrubis 4182e3af1dbSMatthew Wilcox (Oracle) nr_pages += PHYS_PFN(vm_end - vm_start); 419e8420a8eSCyril Hrubis } 420e8420a8eSCyril Hrubis 421e8420a8eSCyril Hrubis return nr_pages; 422e8420a8eSCyril Hrubis } 423e8420a8eSCyril Hrubis 424c154124fSLiam R. Howlett static void __vma_link_file(struct vm_area_struct *vma, 425c154124fSLiam R. Howlett struct address_space *mapping) 4261da177e4SLinus Torvalds { 4271da177e4SLinus Torvalds if (vma->vm_flags & VM_SHARED) 428cf508b58SMiaohe Lin mapping_allow_writable(mapping); 4291da177e4SLinus Torvalds 4301da177e4SLinus Torvalds flush_dcache_mmap_lock(mapping); 4316b2dbba8SMichel Lespinasse vma_interval_tree_insert(vma, &mapping->i_mmap); 4321da177e4SLinus Torvalds flush_dcache_mmap_unlock(mapping); 4331da177e4SLinus Torvalds } 4341da177e4SLinus Torvalds 435763ecb03SLiam R. Howlett static int vma_link(struct mm_struct *mm, struct vm_area_struct *vma) 4361da177e4SLinus Torvalds { 43779e4f2caSLiam R. Howlett VMA_ITERATOR(vmi, mm, 0); 4381da177e4SLinus Torvalds struct address_space *mapping = NULL; 4391da177e4SLinus Torvalds 44079e4f2caSLiam R. Howlett if (vma_iter_prealloc(&vmi)) 441d4af56c5SLiam R. Howlett return -ENOMEM; 442d4af56c5SLiam R. Howlett 44364ac4940SHuang Shijie if (vma->vm_file) { 4441da177e4SLinus Torvalds mapping = vma->vm_file->f_mapping; 44583cde9e8SDavidlohr Bueso i_mmap_lock_write(mapping); 44664ac4940SHuang Shijie } 4471da177e4SLinus Torvalds 44879e4f2caSLiam R. Howlett vma_iter_store(&vmi, vma); 4491da177e4SLinus Torvalds 450c154124fSLiam R. Howlett if (mapping) { 451c154124fSLiam R. Howlett __vma_link_file(vma, mapping); 45283cde9e8SDavidlohr Bueso i_mmap_unlock_write(mapping); 453c154124fSLiam R. Howlett } 4541da177e4SLinus Torvalds 4551da177e4SLinus Torvalds mm->map_count++; 4561da177e4SLinus Torvalds validate_mm(mm); 457d4af56c5SLiam R. Howlett return 0; 4581da177e4SLinus Torvalds } 4591da177e4SLinus Torvalds 4601da177e4SLinus Torvalds /* 46168cefec5SLiam R. Howlett * init_multi_vma_prep() - Initializer for struct vma_prepare 46268cefec5SLiam R. Howlett * @vp: The vma_prepare struct 46368cefec5SLiam R. Howlett * @vma: The vma that will be altered once locked 46468cefec5SLiam R. Howlett * @next: The next vma if it is to be adjusted 46568cefec5SLiam R. Howlett * @remove: The first vma to be removed 46668cefec5SLiam R. Howlett * @remove2: The second vma to be removed 46768cefec5SLiam R. Howlett */ 46868cefec5SLiam R. Howlett static inline void init_multi_vma_prep(struct vma_prepare *vp, 46968cefec5SLiam R. Howlett struct vm_area_struct *vma, struct vm_area_struct *next, 47068cefec5SLiam R. Howlett struct vm_area_struct *remove, struct vm_area_struct *remove2) 47168cefec5SLiam R. Howlett { 47268cefec5SLiam R. Howlett memset(vp, 0, sizeof(struct vma_prepare)); 47368cefec5SLiam R. Howlett vp->vma = vma; 47468cefec5SLiam R. Howlett vp->anon_vma = vma->anon_vma; 47568cefec5SLiam R. Howlett vp->remove = remove; 47668cefec5SLiam R. Howlett vp->remove2 = remove2; 47768cefec5SLiam R. Howlett vp->adj_next = next; 47868cefec5SLiam R. Howlett if (!vp->anon_vma && next) 47968cefec5SLiam R. Howlett vp->anon_vma = next->anon_vma; 48068cefec5SLiam R. Howlett 48168cefec5SLiam R. Howlett vp->file = vma->vm_file; 48268cefec5SLiam R. Howlett if (vp->file) 48368cefec5SLiam R. Howlett vp->mapping = vma->vm_file->f_mapping; 48468cefec5SLiam R. Howlett 48568cefec5SLiam R. Howlett } 48668cefec5SLiam R. Howlett 48768cefec5SLiam R. Howlett /* 48868cefec5SLiam R. Howlett * init_vma_prep() - Initializer wrapper for vma_prepare struct 48968cefec5SLiam R. Howlett * @vp: The vma_prepare struct 49068cefec5SLiam R. Howlett * @vma: The vma that will be altered once locked 49168cefec5SLiam R. Howlett */ 49268cefec5SLiam R. Howlett static inline void init_vma_prep(struct vma_prepare *vp, 49368cefec5SLiam R. Howlett struct vm_area_struct *vma) 49468cefec5SLiam R. Howlett { 49568cefec5SLiam R. Howlett init_multi_vma_prep(vp, vma, NULL, NULL, NULL); 49668cefec5SLiam R. Howlett } 49768cefec5SLiam R. Howlett 49868cefec5SLiam R. Howlett 49968cefec5SLiam R. Howlett /* 500440703e0SLiam R. Howlett * vma_prepare() - Helper function for handling locking VMAs prior to altering 501440703e0SLiam R. Howlett * @vp: The initialized vma_prepare struct 502440703e0SLiam R. Howlett */ 503440703e0SLiam R. Howlett static inline void vma_prepare(struct vma_prepare *vp) 504440703e0SLiam R. Howlett { 505440703e0SLiam R. Howlett if (vp->file) { 506440703e0SLiam R. Howlett uprobe_munmap(vp->vma, vp->vma->vm_start, vp->vma->vm_end); 507440703e0SLiam R. Howlett 508440703e0SLiam R. Howlett if (vp->adj_next) 509440703e0SLiam R. Howlett uprobe_munmap(vp->adj_next, vp->adj_next->vm_start, 510440703e0SLiam R. Howlett vp->adj_next->vm_end); 511440703e0SLiam R. Howlett 512440703e0SLiam R. Howlett i_mmap_lock_write(vp->mapping); 513440703e0SLiam R. Howlett if (vp->insert && vp->insert->vm_file) { 514440703e0SLiam R. Howlett /* 515440703e0SLiam R. Howlett * Put into interval tree now, so instantiated pages 516440703e0SLiam R. Howlett * are visible to arm/parisc __flush_dcache_page 517440703e0SLiam R. Howlett * throughout; but we cannot insert into address 518440703e0SLiam R. Howlett * space until vma start or end is updated. 519440703e0SLiam R. Howlett */ 520440703e0SLiam R. Howlett __vma_link_file(vp->insert, 521440703e0SLiam R. Howlett vp->insert->vm_file->f_mapping); 522440703e0SLiam R. Howlett } 523440703e0SLiam R. Howlett } 524440703e0SLiam R. Howlett 525440703e0SLiam R. Howlett if (vp->anon_vma) { 526440703e0SLiam R. Howlett anon_vma_lock_write(vp->anon_vma); 527440703e0SLiam R. Howlett anon_vma_interval_tree_pre_update_vma(vp->vma); 528440703e0SLiam R. Howlett if (vp->adj_next) 529440703e0SLiam R. Howlett anon_vma_interval_tree_pre_update_vma(vp->adj_next); 530440703e0SLiam R. Howlett } 531440703e0SLiam R. Howlett 532440703e0SLiam R. Howlett if (vp->file) { 533440703e0SLiam R. Howlett flush_dcache_mmap_lock(vp->mapping); 534440703e0SLiam R. Howlett vma_interval_tree_remove(vp->vma, &vp->mapping->i_mmap); 535440703e0SLiam R. Howlett if (vp->adj_next) 536440703e0SLiam R. Howlett vma_interval_tree_remove(vp->adj_next, 537440703e0SLiam R. Howlett &vp->mapping->i_mmap); 538440703e0SLiam R. Howlett } 539440703e0SLiam R. Howlett 540440703e0SLiam R. Howlett } 541440703e0SLiam R. Howlett 542440703e0SLiam R. Howlett /* 543440703e0SLiam R. Howlett * vma_complete- Helper function for handling the unlocking after altering VMAs, 544440703e0SLiam R. Howlett * or for inserting a VMA. 545440703e0SLiam R. Howlett * 546440703e0SLiam R. Howlett * @vp: The vma_prepare struct 547440703e0SLiam R. Howlett * @vmi: The vma iterator 548440703e0SLiam R. Howlett * @mm: The mm_struct 549440703e0SLiam R. Howlett */ 550440703e0SLiam R. Howlett static inline void vma_complete(struct vma_prepare *vp, 551440703e0SLiam R. Howlett struct vma_iterator *vmi, struct mm_struct *mm) 552440703e0SLiam R. Howlett { 553440703e0SLiam R. Howlett if (vp->file) { 554440703e0SLiam R. Howlett if (vp->adj_next) 555440703e0SLiam R. Howlett vma_interval_tree_insert(vp->adj_next, 556440703e0SLiam R. Howlett &vp->mapping->i_mmap); 557440703e0SLiam R. Howlett vma_interval_tree_insert(vp->vma, &vp->mapping->i_mmap); 558440703e0SLiam R. Howlett flush_dcache_mmap_unlock(vp->mapping); 559440703e0SLiam R. Howlett } 560440703e0SLiam R. Howlett 561440703e0SLiam R. Howlett if (vp->remove && vp->file) { 562440703e0SLiam R. Howlett __remove_shared_vm_struct(vp->remove, vp->file, vp->mapping); 563440703e0SLiam R. Howlett if (vp->remove2) 564440703e0SLiam R. Howlett __remove_shared_vm_struct(vp->remove2, vp->file, 565440703e0SLiam R. Howlett vp->mapping); 566440703e0SLiam R. Howlett } else if (vp->insert) { 567440703e0SLiam R. Howlett /* 568440703e0SLiam R. Howlett * split_vma has split insert from vma, and needs 569440703e0SLiam R. Howlett * us to insert it before dropping the locks 570440703e0SLiam R. Howlett * (it may either follow vma or precede it). 571440703e0SLiam R. Howlett */ 572440703e0SLiam R. Howlett vma_iter_store(vmi, vp->insert); 573440703e0SLiam R. Howlett mm->map_count++; 574440703e0SLiam R. Howlett } 575440703e0SLiam R. Howlett 576440703e0SLiam R. Howlett if (vp->anon_vma) { 577440703e0SLiam R. Howlett anon_vma_interval_tree_post_update_vma(vp->vma); 578440703e0SLiam R. Howlett if (vp->adj_next) 579440703e0SLiam R. Howlett anon_vma_interval_tree_post_update_vma(vp->adj_next); 580440703e0SLiam R. Howlett anon_vma_unlock_write(vp->anon_vma); 581440703e0SLiam R. Howlett } 582440703e0SLiam R. Howlett 583440703e0SLiam R. Howlett if (vp->file) { 584440703e0SLiam R. Howlett i_mmap_unlock_write(vp->mapping); 585440703e0SLiam R. Howlett uprobe_mmap(vp->vma); 586440703e0SLiam R. Howlett 587440703e0SLiam R. Howlett if (vp->adj_next) 588440703e0SLiam R. Howlett uprobe_mmap(vp->adj_next); 589440703e0SLiam R. Howlett } 590440703e0SLiam R. Howlett 591440703e0SLiam R. Howlett if (vp->remove) { 592440703e0SLiam R. Howlett again: 593440703e0SLiam R. Howlett if (vp->file) { 594440703e0SLiam R. Howlett uprobe_munmap(vp->remove, vp->remove->vm_start, 595440703e0SLiam R. Howlett vp->remove->vm_end); 596440703e0SLiam R. Howlett fput(vp->file); 597440703e0SLiam R. Howlett } 598440703e0SLiam R. Howlett if (vp->remove->anon_vma) 599440703e0SLiam R. Howlett anon_vma_merge(vp->vma, vp->remove); 600440703e0SLiam R. Howlett mm->map_count--; 601440703e0SLiam R. Howlett mpol_put(vma_policy(vp->remove)); 602440703e0SLiam R. Howlett if (!vp->remove2) 603440703e0SLiam R. Howlett WARN_ON_ONCE(vp->vma->vm_end < vp->remove->vm_end); 604440703e0SLiam R. Howlett vm_area_free(vp->remove); 605440703e0SLiam R. Howlett 606440703e0SLiam R. Howlett /* 607440703e0SLiam R. Howlett * In mprotect's case 6 (see comments on vma_merge), 60868cefec5SLiam R. Howlett * we must remove the one after next as well. 609440703e0SLiam R. Howlett */ 610440703e0SLiam R. Howlett if (vp->remove2) { 611440703e0SLiam R. Howlett vp->remove = vp->remove2; 612440703e0SLiam R. Howlett vp->remove2 = NULL; 613440703e0SLiam R. Howlett goto again; 614440703e0SLiam R. Howlett } 615440703e0SLiam R. Howlett } 616440703e0SLiam R. Howlett if (vp->insert && vp->file) 617440703e0SLiam R. Howlett uprobe_mmap(vp->insert); 618440703e0SLiam R. Howlett } 619440703e0SLiam R. Howlett 620440703e0SLiam R. Howlett /* 621*04241ffeSLiam R. Howlett * dup_anon_vma() - Helper function to duplicate anon_vma 622*04241ffeSLiam R. Howlett * @dst: The destination VMA 623*04241ffeSLiam R. Howlett * @src: The source VMA 624*04241ffeSLiam R. Howlett * 625*04241ffeSLiam R. Howlett * Returns: 0 on success. 626*04241ffeSLiam R. Howlett */ 627*04241ffeSLiam R. Howlett static inline int dup_anon_vma(struct vm_area_struct *dst, 628*04241ffeSLiam R. Howlett struct vm_area_struct *src) 629*04241ffeSLiam R. Howlett { 630*04241ffeSLiam R. Howlett /* 631*04241ffeSLiam R. Howlett * Easily overlooked: when mprotect shifts the boundary, make sure the 632*04241ffeSLiam R. Howlett * expanding vma has anon_vma set if the shrinking vma had, to cover any 633*04241ffeSLiam R. Howlett * anon pages imported. 634*04241ffeSLiam R. Howlett */ 635*04241ffeSLiam R. Howlett if (src->anon_vma && !dst->anon_vma) { 636*04241ffeSLiam R. Howlett dst->anon_vma = src->anon_vma; 637*04241ffeSLiam R. Howlett return anon_vma_clone(dst, src); 638*04241ffeSLiam R. Howlett } 639*04241ffeSLiam R. Howlett 640*04241ffeSLiam R. Howlett return 0; 641*04241ffeSLiam R. Howlett } 642*04241ffeSLiam R. Howlett 643*04241ffeSLiam 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)) { 668*04241ffeSLiam R. Howlett int ret; 6699303d3e1SLiam R. Howlett 670*04241ffeSLiam R. Howlett remove_next = true; 671*04241ffeSLiam R. Howlett ret = dup_anon_vma(vma, next); 672*04241ffeSLiam R. Howlett if (ret) 673*04241ffeSLiam 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 * We cannot adjust vm_start, vm_end, vm_pgoff fields of a vma that 7451da177e4SLinus Torvalds * is already present in an i_mmap tree without adjusting the tree. 7461da177e4SLinus Torvalds * The following helper function should be used when such adjustments 7471da177e4SLinus Torvalds * are necessary. The "insert" vma (if any) is to be inserted 7481da177e4SLinus Torvalds * before we drop the necessary locks. 7491da177e4SLinus Torvalds */ 7509e560446SLiam R. Howlett int __vma_adjust(struct vma_iterator *vmi, struct vm_area_struct *vma, 7519e560446SLiam R. Howlett unsigned long start, unsigned long end, pgoff_t pgoff, 752b2b3b886SLiam R. Howlett struct vm_area_struct *expand) 7531da177e4SLinus Torvalds { 7541da177e4SLinus Torvalds struct mm_struct *mm = vma->vm_mm; 75568cefec5SLiam R. Howlett struct vm_area_struct *remove2 = NULL; 75668cefec5SLiam R. Howlett struct vm_area_struct *remove = NULL; 7571cd916d0SAndrew Morton struct vm_area_struct *next = find_vma(mm, vma->vm_end); 758524e00b3SLiam R. Howlett struct vm_area_struct *orig_vma = vma; 7591da177e4SLinus Torvalds struct file *file = vma->vm_file; 760524e00b3SLiam R. Howlett bool vma_changed = false; 7611da177e4SLinus Torvalds long adjust_next = 0; 762440703e0SLiam R. Howlett struct vma_prepare vma_prep; 763287d97acSLinus Torvalds 764b2b3b886SLiam R. Howlett if (next) { 765*04241ffeSLiam R. Howlett int error = 0; 766*04241ffeSLiam R. Howlett 7671da177e4SLinus Torvalds if (end >= next->vm_end) { 7681da177e4SLinus Torvalds /* 7691da177e4SLinus Torvalds * vma expands, overlapping all the next, and 7701da177e4SLinus Torvalds * perhaps the one after too (mprotect case 6). 77186d12e47SAndrea Arcangeli * The only other cases that gets here are 772e86f15eeSAndrea Arcangeli * case 1, case 7 and case 8. 773e86f15eeSAndrea Arcangeli */ 774e86f15eeSAndrea Arcangeli if (next == expand) { 775e86f15eeSAndrea Arcangeli /* 776e86f15eeSAndrea Arcangeli * The only case where we don't expand "vma" 777e86f15eeSAndrea Arcangeli * and we expand "next" instead is case 8. 778e86f15eeSAndrea Arcangeli */ 779e86f15eeSAndrea Arcangeli VM_WARN_ON(end != next->vm_end); 780e86f15eeSAndrea Arcangeli /* 78168cefec5SLiam R. Howlett * we're removing "vma" and that to do so we 782e86f15eeSAndrea Arcangeli * swapped "vma" and "next". 783e86f15eeSAndrea Arcangeli */ 784e86f15eeSAndrea Arcangeli VM_WARN_ON(file != next->vm_file); 785e86f15eeSAndrea Arcangeli swap(vma, next); 78668cefec5SLiam R. Howlett remove = next; 787e86f15eeSAndrea Arcangeli } else { 788e86f15eeSAndrea Arcangeli VM_WARN_ON(expand != vma); 789e86f15eeSAndrea Arcangeli /* 79068cefec5SLiam R. Howlett * case 1, 6, 7, remove next. 79168cefec5SLiam R. Howlett * case 6 also removes the one beyond next 7921da177e4SLinus Torvalds */ 79368cefec5SLiam R. Howlett remove = next; 79468cefec5SLiam R. Howlett if (end > next->vm_end) 79568cefec5SLiam R. Howlett remove2 = find_vma(mm, next->vm_end); 796d4af56c5SLiam R. Howlett 79768cefec5SLiam R. Howlett VM_WARN_ON(remove2 != NULL && 79868cefec5SLiam R. Howlett end != remove2->vm_end); 799e86f15eeSAndrea Arcangeli } 800e86f15eeSAndrea Arcangeli 801734537c9SKirill A. Shutemov /* 802734537c9SKirill A. Shutemov * If next doesn't have anon_vma, import from vma after 803734537c9SKirill A. Shutemov * next, if the vma overlaps with it. 804734537c9SKirill A. Shutemov */ 805*04241ffeSLiam R. Howlett if (remove != NULL && !next->anon_vma) 806*04241ffeSLiam R. Howlett error = dup_anon_vma(vma, remove2); 807*04241ffeSLiam R. Howlett else 808*04241ffeSLiam R. Howlett error = dup_anon_vma(vma, remove); 809734537c9SKirill A. Shutemov 8101da177e4SLinus Torvalds } else if (end > next->vm_start) { 8111da177e4SLinus Torvalds /* 8121da177e4SLinus Torvalds * vma expands, overlapping part of the next: 8131da177e4SLinus Torvalds * mprotect case 5 shifting the boundary up. 8141da177e4SLinus Torvalds */ 815f9d86a60SWei Yang adjust_next = (end - next->vm_start); 816*04241ffeSLiam R. Howlett VM_WARN_ON(expand != vma); 817*04241ffeSLiam R. Howlett error = dup_anon_vma(vma, next); 8181da177e4SLinus Torvalds } else if (end < vma->vm_end) { 8191da177e4SLinus Torvalds /* 8201da177e4SLinus Torvalds * vma shrinks, and !insert tells it's not 8211da177e4SLinus Torvalds * split_vma inserting another: so it must be 8221da177e4SLinus Torvalds * mprotect case 4 shifting the boundary down. 8231da177e4SLinus Torvalds */ 824f9d86a60SWei Yang adjust_next = -(vma->vm_end - end); 825*04241ffeSLiam R. Howlett VM_WARN_ON(expand != next); 826*04241ffeSLiam R. Howlett error = dup_anon_vma(next, vma); 8271da177e4SLinus Torvalds } 8283fe89b3eSLeon Yu if (error) 829b800c91aSKonstantin Khlebnikov return error; 830b800c91aSKonstantin Khlebnikov } 83137f9f559SKirill A. Shutemov 8329e560446SLiam R. Howlett if (vma_iter_prealloc(vmi)) 833d4af56c5SLiam R. Howlett return -ENOMEM; 834d4af56c5SLiam R. Howlett 835d4af56c5SLiam R. Howlett vma_adjust_trans_huge(orig_vma, start, end, adjust_next); 836682968e0SSrikar Dronamraju 83768cefec5SLiam R. Howlett init_multi_vma_prep(&vma_prep, vma, adjust_next ? next : NULL, remove, 83868cefec5SLiam R. Howlett remove2); 83968cefec5SLiam R. Howlett VM_WARN_ON(vma_prep.anon_vma && adjust_next && next->anon_vma && 84068cefec5SLiam R. Howlett vma_prep.anon_vma != next->anon_vma); 8411da177e4SLinus Torvalds 842440703e0SLiam R. Howlett vma_prepare(&vma_prep); 8431da177e4SLinus Torvalds 844cf51e86dSLiam R. Howlett if (start < vma->vm_start || end > vma->vm_end) 845524e00b3SLiam R. Howlett vma_changed = true; 846b2b3b886SLiam R. Howlett 847b2b3b886SLiam R. Howlett vma->vm_start = start; 8481da177e4SLinus Torvalds vma->vm_end = end; 849b2b3b886SLiam R. Howlett vma->vm_pgoff = pgoff; 850d4af56c5SLiam R. Howlett 851524e00b3SLiam R. Howlett if (vma_changed) 8529e560446SLiam R. Howlett vma_iter_store(vmi, vma); 853d4af56c5SLiam R. Howlett 8541da177e4SLinus Torvalds if (adjust_next) { 855f9d86a60SWei Yang next->vm_start += adjust_next; 856f9d86a60SWei Yang next->vm_pgoff += adjust_next >> PAGE_SHIFT; 857cf51e86dSLiam R. Howlett if (adjust_next < 0) { 858cf51e86dSLiam R. Howlett WARN_ON_ONCE(vma_changed); 8599e560446SLiam R. Howlett vma_iter_store(vmi, next); 8601da177e4SLinus Torvalds } 861cf51e86dSLiam R. Howlett } 8621da177e4SLinus Torvalds 863440703e0SLiam R. Howlett vma_complete(&vma_prep, vmi, mm); 8649e560446SLiam R. Howlett vma_iter_free(vmi); 8651da177e4SLinus Torvalds validate_mm(mm); 866763ecb03SLiam R. Howlett 8675beb4930SRik van Riel return 0; 8681da177e4SLinus Torvalds } 8691da177e4SLinus Torvalds 8701da177e4SLinus Torvalds /* 8711da177e4SLinus Torvalds * If the vma has a ->close operation then the driver probably needs to release 8721da177e4SLinus Torvalds * per-vma resources, so we don't attempt to merge those. 8731da177e4SLinus Torvalds */ 8741da177e4SLinus Torvalds static inline int is_mergeable_vma(struct vm_area_struct *vma, 87519a809afSAndrea Arcangeli struct file *file, unsigned long vm_flags, 8769a10064fSColin Cross struct vm_userfaultfd_ctx vm_userfaultfd_ctx, 8775c26f6acSSuren Baghdasaryan struct anon_vma_name *anon_name) 8781da177e4SLinus Torvalds { 87934228d47SCyrill Gorcunov /* 88034228d47SCyrill Gorcunov * VM_SOFTDIRTY should not prevent from VMA merging, if we 88134228d47SCyrill Gorcunov * match the flags but dirty bit -- the caller should mark 88234228d47SCyrill Gorcunov * merged VMA as dirty. If dirty bit won't be excluded from 8838bb4e7a2SWei Yang * comparison, we increase pressure on the memory system forcing 88434228d47SCyrill Gorcunov * the kernel to generate new VMAs when old one could be 88534228d47SCyrill Gorcunov * extended instead. 88634228d47SCyrill Gorcunov */ 88734228d47SCyrill Gorcunov if ((vma->vm_flags ^ vm_flags) & ~VM_SOFTDIRTY) 8881da177e4SLinus Torvalds return 0; 8891da177e4SLinus Torvalds if (vma->vm_file != file) 8901da177e4SLinus Torvalds return 0; 8911da177e4SLinus Torvalds if (vma->vm_ops && vma->vm_ops->close) 8921da177e4SLinus Torvalds return 0; 89319a809afSAndrea Arcangeli if (!is_mergeable_vm_userfaultfd_ctx(vma, vm_userfaultfd_ctx)) 89419a809afSAndrea Arcangeli return 0; 8955c26f6acSSuren Baghdasaryan if (!anon_vma_name_eq(anon_vma_name(vma), anon_name)) 8969a10064fSColin Cross return 0; 8971da177e4SLinus Torvalds return 1; 8981da177e4SLinus Torvalds } 8991da177e4SLinus Torvalds 9001da177e4SLinus Torvalds static inline int is_mergeable_anon_vma(struct anon_vma *anon_vma1, 901965f55deSShaohua Li struct anon_vma *anon_vma2, 902965f55deSShaohua Li struct vm_area_struct *vma) 9031da177e4SLinus Torvalds { 904965f55deSShaohua Li /* 905965f55deSShaohua Li * The list_is_singular() test is to avoid merging VMA cloned from 906965f55deSShaohua Li * parents. This can improve scalability caused by anon_vma lock. 907965f55deSShaohua Li */ 908965f55deSShaohua Li if ((!anon_vma1 || !anon_vma2) && (!vma || 909965f55deSShaohua Li list_is_singular(&vma->anon_vma_chain))) 910965f55deSShaohua Li return 1; 911965f55deSShaohua Li return anon_vma1 == anon_vma2; 9121da177e4SLinus Torvalds } 9131da177e4SLinus Torvalds 9141da177e4SLinus Torvalds /* 9151da177e4SLinus Torvalds * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff) 9161da177e4SLinus Torvalds * in front of (at a lower virtual address and file offset than) the vma. 9171da177e4SLinus Torvalds * 9181da177e4SLinus Torvalds * We cannot merge two vmas if they have differently assigned (non-NULL) 9191da177e4SLinus Torvalds * anon_vmas, nor if same anon_vma is assigned but offsets incompatible. 9201da177e4SLinus Torvalds * 9211da177e4SLinus Torvalds * We don't check here for the merged mmap wrapping around the end of pagecache 92245e55300SPeter Collingbourne * indices (16TB on ia32) because do_mmap() does not permit mmap's which 9231da177e4SLinus Torvalds * wrap, nor mmaps which cover the final page at index -1UL. 9241da177e4SLinus Torvalds */ 9251da177e4SLinus Torvalds static int 9261da177e4SLinus Torvalds can_vma_merge_before(struct vm_area_struct *vma, unsigned long vm_flags, 92719a809afSAndrea Arcangeli struct anon_vma *anon_vma, struct file *file, 92819a809afSAndrea Arcangeli pgoff_t vm_pgoff, 9299a10064fSColin Cross struct vm_userfaultfd_ctx vm_userfaultfd_ctx, 9305c26f6acSSuren Baghdasaryan struct anon_vma_name *anon_name) 9311da177e4SLinus Torvalds { 9329a10064fSColin Cross if (is_mergeable_vma(vma, file, vm_flags, vm_userfaultfd_ctx, anon_name) && 933965f55deSShaohua Li is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) { 9341da177e4SLinus Torvalds if (vma->vm_pgoff == vm_pgoff) 9351da177e4SLinus Torvalds return 1; 9361da177e4SLinus Torvalds } 9371da177e4SLinus Torvalds return 0; 9381da177e4SLinus Torvalds } 9391da177e4SLinus Torvalds 9401da177e4SLinus Torvalds /* 9411da177e4SLinus Torvalds * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff) 9421da177e4SLinus Torvalds * beyond (at a higher virtual address and file offset than) the vma. 9431da177e4SLinus Torvalds * 9441da177e4SLinus Torvalds * We cannot merge two vmas if they have differently assigned (non-NULL) 9451da177e4SLinus Torvalds * anon_vmas, nor if same anon_vma is assigned but offsets incompatible. 9461da177e4SLinus Torvalds */ 9471da177e4SLinus Torvalds static int 9481da177e4SLinus Torvalds can_vma_merge_after(struct vm_area_struct *vma, unsigned long vm_flags, 94919a809afSAndrea Arcangeli struct anon_vma *anon_vma, struct file *file, 95019a809afSAndrea Arcangeli pgoff_t vm_pgoff, 9519a10064fSColin Cross struct vm_userfaultfd_ctx vm_userfaultfd_ctx, 9525c26f6acSSuren Baghdasaryan struct anon_vma_name *anon_name) 9531da177e4SLinus Torvalds { 9549a10064fSColin Cross if (is_mergeable_vma(vma, file, vm_flags, vm_userfaultfd_ctx, anon_name) && 955965f55deSShaohua Li is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) { 9561da177e4SLinus Torvalds pgoff_t vm_pglen; 957d6e93217SLibin vm_pglen = vma_pages(vma); 9581da177e4SLinus Torvalds if (vma->vm_pgoff + vm_pglen == vm_pgoff) 9591da177e4SLinus Torvalds return 1; 9601da177e4SLinus Torvalds } 9611da177e4SLinus Torvalds return 0; 9621da177e4SLinus Torvalds } 9631da177e4SLinus Torvalds 9641da177e4SLinus Torvalds /* 9659a10064fSColin Cross * Given a mapping request (addr,end,vm_flags,file,pgoff,anon_name), 9669a10064fSColin Cross * figure out whether that can be merged with its predecessor or its 9679a10064fSColin Cross * successor. Or both (it neatly fills a hole). 9681da177e4SLinus Torvalds * 9691da177e4SLinus Torvalds * In most cases - when called for mmap, brk or mremap - [addr,end) is 9701da177e4SLinus Torvalds * certain not to be mapped by the time vma_merge is called; but when 9711da177e4SLinus Torvalds * called for mprotect, it is certain to be already mapped (either at 9721da177e4SLinus Torvalds * an offset within prev, or at the start of next), and the flags of 9731da177e4SLinus Torvalds * this area are about to be changed to vm_flags - and the no-change 9741da177e4SLinus Torvalds * case has already been eliminated. 9751da177e4SLinus Torvalds * 9761da177e4SLinus Torvalds * The following mprotect cases have to be considered, where AAAA is 9771da177e4SLinus Torvalds * the area passed down from mprotect_fixup, never extending beyond one 9781da177e4SLinus Torvalds * vma, PPPPPP is the prev vma specified, and NNNNNN the next vma after: 9791da177e4SLinus Torvalds * 9805d42ab29SWei Yang * AAAA AAAA AAAA 9815d42ab29SWei Yang * PPPPPPNNNNNN PPPPPPNNNNNN PPPPPPNNNNNN 9825d42ab29SWei Yang * cannot merge might become might become 9835d42ab29SWei Yang * PPNNNNNNNNNN PPPPPPPPPPNN 9845d42ab29SWei Yang * mmap, brk or case 4 below case 5 below 9855d42ab29SWei Yang * mremap move: 9865d42ab29SWei Yang * AAAA AAAA 9875d42ab29SWei Yang * PPPP NNNN PPPPNNNNXXXX 9885d42ab29SWei Yang * might become might become 9895d42ab29SWei Yang * PPPPPPPPPPPP 1 or PPPPPPPPPPPP 6 or 9905d42ab29SWei Yang * PPPPPPPPNNNN 2 or PPPPPPPPXXXX 7 or 9915d42ab29SWei Yang * PPPPNNNNNNNN 3 PPPPXXXXXXXX 8 9921da177e4SLinus Torvalds * 9938bb4e7a2SWei Yang * It is important for case 8 that the vma NNNN overlapping the 994e86f15eeSAndrea Arcangeli * region AAAA is never going to extended over XXXX. Instead XXXX must 995e86f15eeSAndrea Arcangeli * be extended in region AAAA and NNNN must be removed. This way in 996e86f15eeSAndrea Arcangeli * all cases where vma_merge succeeds, the moment vma_adjust drops the 997e86f15eeSAndrea Arcangeli * rmap_locks, the properties of the merged vma will be already 998e86f15eeSAndrea Arcangeli * correct for the whole merged range. Some of those properties like 999e86f15eeSAndrea Arcangeli * vm_page_prot/vm_flags may be accessed by rmap_walks and they must 1000e86f15eeSAndrea Arcangeli * be correct for the whole merged range immediately after the 1001e86f15eeSAndrea Arcangeli * rmap_locks are released. Otherwise if XXXX would be removed and 1002e86f15eeSAndrea Arcangeli * NNNN would be extended over the XXXX range, remove_migration_ptes 1003e86f15eeSAndrea Arcangeli * or other rmap walkers (if working on addresses beyond the "end" 1004e86f15eeSAndrea Arcangeli * parameter) may establish ptes with the wrong permissions of NNNN 1005e86f15eeSAndrea Arcangeli * instead of the right permissions of XXXX. 10061da177e4SLinus Torvalds */ 10079760ebffSLiam R. Howlett struct vm_area_struct *vma_merge(struct vma_iterator *vmi, struct mm_struct *mm, 10081da177e4SLinus Torvalds struct vm_area_struct *prev, unsigned long addr, 10091da177e4SLinus Torvalds unsigned long end, unsigned long vm_flags, 10101da177e4SLinus Torvalds struct anon_vma *anon_vma, struct file *file, 101119a809afSAndrea Arcangeli pgoff_t pgoff, struct mempolicy *policy, 10129a10064fSColin Cross struct vm_userfaultfd_ctx vm_userfaultfd_ctx, 10135c26f6acSSuren Baghdasaryan struct anon_vma_name *anon_name) 10141da177e4SLinus Torvalds { 10151da177e4SLinus Torvalds pgoff_t pglen = (end - addr) >> PAGE_SHIFT; 10169760ebffSLiam R. Howlett struct vm_area_struct *mid, *next, *res = NULL; 1017eef19944SJakub Matěna int err = -1; 1018eef19944SJakub Matěna bool merge_prev = false; 1019eef19944SJakub Matěna bool merge_next = false; 10201da177e4SLinus Torvalds 10211da177e4SLinus Torvalds /* 10221da177e4SLinus Torvalds * We later require that vma->vm_flags == vm_flags, 10231da177e4SLinus Torvalds * so this tests vma->vm_flags & VM_SPECIAL, too. 10241da177e4SLinus Torvalds */ 10251da177e4SLinus Torvalds if (vm_flags & VM_SPECIAL) 10261da177e4SLinus Torvalds return NULL; 10271da177e4SLinus Torvalds 1028763ecb03SLiam R. Howlett next = find_vma(mm, prev ? prev->vm_end : 0); 1029eef19944SJakub Matěna mid = next; 1030eef19944SJakub Matěna if (next && next->vm_end == end) /* cases 6, 7, 8 */ 1031763ecb03SLiam R. Howlett next = find_vma(mm, next->vm_end); 10321da177e4SLinus Torvalds 1033e86f15eeSAndrea Arcangeli /* verify some invariant that must be enforced by the caller */ 1034e86f15eeSAndrea Arcangeli VM_WARN_ON(prev && addr <= prev->vm_start); 1035eef19944SJakub Matěna VM_WARN_ON(mid && end > mid->vm_end); 1036e86f15eeSAndrea Arcangeli VM_WARN_ON(addr >= end); 1037e86f15eeSAndrea Arcangeli 1038eef19944SJakub Matěna /* Can we merge the predecessor? */ 10391da177e4SLinus Torvalds if (prev && prev->vm_end == addr && 10401da177e4SLinus Torvalds mpol_equal(vma_policy(prev), policy) && 10411da177e4SLinus Torvalds can_vma_merge_after(prev, vm_flags, 104219a809afSAndrea Arcangeli anon_vma, file, pgoff, 10439a10064fSColin Cross vm_userfaultfd_ctx, anon_name)) { 1044eef19944SJakub Matěna merge_prev = true; 10451da177e4SLinus Torvalds } 1046eef19944SJakub Matěna /* Can we merge the successor? */ 10471da177e4SLinus Torvalds if (next && end == next->vm_start && 10481da177e4SLinus Torvalds mpol_equal(policy, vma_policy(next)) && 10491da177e4SLinus Torvalds can_vma_merge_before(next, vm_flags, 105019a809afSAndrea Arcangeli anon_vma, file, pgoff+pglen, 10519a10064fSColin Cross vm_userfaultfd_ctx, anon_name)) { 1052eef19944SJakub Matěna merge_next = true; 1053eef19944SJakub Matěna } 1054eef19944SJakub Matěna /* Can we merge both the predecessor and the successor? */ 1055eef19944SJakub Matěna if (merge_prev && merge_next && 1056eef19944SJakub Matěna is_mergeable_anon_vma(prev->anon_vma, 1057eef19944SJakub Matěna next->anon_vma, NULL)) { /* cases 1, 6 */ 10589e560446SLiam R. Howlett err = __vma_adjust(vmi, prev, prev->vm_start, 1059b2b3b886SLiam R. Howlett next->vm_end, prev->vm_pgoff, prev); 1060eef19944SJakub Matěna res = prev; 1061eef19944SJakub Matěna } else if (merge_prev) { /* cases 2, 5, 7 */ 10629e560446SLiam R. Howlett err = __vma_adjust(vmi, prev, prev->vm_start, 1063b2b3b886SLiam R. Howlett end, prev->vm_pgoff, prev); 1064eef19944SJakub Matěna res = prev; 1065eef19944SJakub Matěna } else if (merge_next) { 10661da177e4SLinus Torvalds if (prev && addr < prev->vm_end) /* case 4 */ 10679e560446SLiam R. Howlett err = __vma_adjust(vmi, prev, prev->vm_start, 1068b2b3b886SLiam R. Howlett addr, prev->vm_pgoff, next); 1069eef19944SJakub Matěna else /* cases 3, 8 */ 10709e560446SLiam R. Howlett err = __vma_adjust(vmi, mid, addr, next->vm_end, 1071b2b3b886SLiam R. Howlett next->vm_pgoff - pglen, next); 1072eef19944SJakub Matěna res = next; 10731da177e4SLinus Torvalds } 10741da177e4SLinus Torvalds 1075eef19944SJakub Matěna /* 1076eef19944SJakub Matěna * Cannot merge with predecessor or successor or error in __vma_adjust? 1077eef19944SJakub Matěna */ 1078eef19944SJakub Matěna if (err) 10791da177e4SLinus Torvalds return NULL; 1080eef19944SJakub Matěna khugepaged_enter_vma(res, vm_flags); 10811da177e4SLinus Torvalds 10829760ebffSLiam R. Howlett if (res) 1083f2ebfe43SLiam R. Howlett vma_iter_set(vmi, end); 1084f2ebfe43SLiam R. Howlett 10859760ebffSLiam R. Howlett return res; 1086f2ebfe43SLiam R. Howlett } 1087f2ebfe43SLiam R. Howlett 10881da177e4SLinus Torvalds /* 1089b4f315b4SEthon Paul * Rough compatibility check to quickly see if it's even worth looking 1090d0e9fe17SLinus Torvalds * at sharing an anon_vma. 1091d0e9fe17SLinus Torvalds * 1092d0e9fe17SLinus Torvalds * They need to have the same vm_file, and the flags can only differ 1093d0e9fe17SLinus Torvalds * in things that mprotect may change. 1094d0e9fe17SLinus Torvalds * 1095d0e9fe17SLinus Torvalds * NOTE! The fact that we share an anon_vma doesn't _have_ to mean that 1096d0e9fe17SLinus Torvalds * we can merge the two vma's. For example, we refuse to merge a vma if 1097d0e9fe17SLinus Torvalds * there is a vm_ops->close() function, because that indicates that the 1098d0e9fe17SLinus Torvalds * driver is doing some kind of reference counting. But that doesn't 1099d0e9fe17SLinus Torvalds * really matter for the anon_vma sharing case. 1100d0e9fe17SLinus Torvalds */ 1101d0e9fe17SLinus Torvalds static int anon_vma_compatible(struct vm_area_struct *a, struct vm_area_struct *b) 1102d0e9fe17SLinus Torvalds { 1103d0e9fe17SLinus Torvalds return a->vm_end == b->vm_start && 1104d0e9fe17SLinus Torvalds mpol_equal(vma_policy(a), vma_policy(b)) && 1105d0e9fe17SLinus Torvalds a->vm_file == b->vm_file && 11066cb4d9a2SAnshuman Khandual !((a->vm_flags ^ b->vm_flags) & ~(VM_ACCESS_FLAGS | VM_SOFTDIRTY)) && 1107d0e9fe17SLinus Torvalds b->vm_pgoff == a->vm_pgoff + ((b->vm_start - a->vm_start) >> PAGE_SHIFT); 1108d0e9fe17SLinus Torvalds } 1109d0e9fe17SLinus Torvalds 1110d0e9fe17SLinus Torvalds /* 1111d0e9fe17SLinus Torvalds * Do some basic sanity checking to see if we can re-use the anon_vma 1112d0e9fe17SLinus Torvalds * from 'old'. The 'a'/'b' vma's are in VM order - one of them will be 1113d0e9fe17SLinus Torvalds * the same as 'old', the other will be the new one that is trying 1114d0e9fe17SLinus Torvalds * to share the anon_vma. 1115d0e9fe17SLinus Torvalds * 11165b449489SFlorian Rommel * NOTE! This runs with mmap_lock held for reading, so it is possible that 1117d0e9fe17SLinus Torvalds * the anon_vma of 'old' is concurrently in the process of being set up 1118d0e9fe17SLinus Torvalds * by another page fault trying to merge _that_. But that's ok: if it 1119d0e9fe17SLinus Torvalds * is being set up, that automatically means that it will be a singleton 1120d0e9fe17SLinus Torvalds * acceptable for merging, so we can do all of this optimistically. But 11214db0c3c2SJason Low * we do that READ_ONCE() to make sure that we never re-load the pointer. 1122d0e9fe17SLinus Torvalds * 1123d0e9fe17SLinus Torvalds * IOW: that the "list_is_singular()" test on the anon_vma_chain only 1124d0e9fe17SLinus Torvalds * matters for the 'stable anon_vma' case (ie the thing we want to avoid 1125d0e9fe17SLinus Torvalds * is to return an anon_vma that is "complex" due to having gone through 1126d0e9fe17SLinus Torvalds * a fork). 1127d0e9fe17SLinus Torvalds * 1128d0e9fe17SLinus Torvalds * We also make sure that the two vma's are compatible (adjacent, 1129d0e9fe17SLinus Torvalds * and with the same memory policies). That's all stable, even with just 11305b449489SFlorian Rommel * a read lock on the mmap_lock. 1131d0e9fe17SLinus Torvalds */ 1132d0e9fe17SLinus Torvalds static struct anon_vma *reusable_anon_vma(struct vm_area_struct *old, struct vm_area_struct *a, struct vm_area_struct *b) 1133d0e9fe17SLinus Torvalds { 1134d0e9fe17SLinus Torvalds if (anon_vma_compatible(a, b)) { 11354db0c3c2SJason Low struct anon_vma *anon_vma = READ_ONCE(old->anon_vma); 1136d0e9fe17SLinus Torvalds 1137d0e9fe17SLinus Torvalds if (anon_vma && list_is_singular(&old->anon_vma_chain)) 1138d0e9fe17SLinus Torvalds return anon_vma; 1139d0e9fe17SLinus Torvalds } 1140d0e9fe17SLinus Torvalds return NULL; 1141d0e9fe17SLinus Torvalds } 1142d0e9fe17SLinus Torvalds 1143d0e9fe17SLinus Torvalds /* 11441da177e4SLinus Torvalds * find_mergeable_anon_vma is used by anon_vma_prepare, to check 11451da177e4SLinus Torvalds * neighbouring vmas for a suitable anon_vma, before it goes off 11461da177e4SLinus Torvalds * to allocate a new anon_vma. It checks because a repetitive 11471da177e4SLinus Torvalds * sequence of mprotects and faults may otherwise lead to distinct 11481da177e4SLinus Torvalds * anon_vmas being allocated, preventing vma merge in subsequent 11491da177e4SLinus Torvalds * mprotect. 11501da177e4SLinus Torvalds */ 11511da177e4SLinus Torvalds struct anon_vma *find_mergeable_anon_vma(struct vm_area_struct *vma) 11521da177e4SLinus Torvalds { 1153763ecb03SLiam R. Howlett MA_STATE(mas, &vma->vm_mm->mm_mt, vma->vm_end, vma->vm_end); 1154a67c8caaSMiaohe Lin struct anon_vma *anon_vma = NULL; 1155763ecb03SLiam R. Howlett struct vm_area_struct *prev, *next; 11561da177e4SLinus Torvalds 1157a67c8caaSMiaohe Lin /* Try next first. */ 1158763ecb03SLiam R. Howlett next = mas_walk(&mas); 1159763ecb03SLiam R. Howlett if (next) { 1160763ecb03SLiam R. Howlett anon_vma = reusable_anon_vma(next, vma, next); 1161d0e9fe17SLinus Torvalds if (anon_vma) 1162d0e9fe17SLinus Torvalds return anon_vma; 1163a67c8caaSMiaohe Lin } 11641da177e4SLinus Torvalds 1165763ecb03SLiam R. Howlett prev = mas_prev(&mas, 0); 1166763ecb03SLiam R. Howlett VM_BUG_ON_VMA(prev != vma, vma); 1167763ecb03SLiam R. Howlett prev = mas_prev(&mas, 0); 1168a67c8caaSMiaohe Lin /* Try prev next. */ 1169763ecb03SLiam R. Howlett if (prev) 1170763ecb03SLiam R. Howlett anon_vma = reusable_anon_vma(prev, prev, vma); 1171a67c8caaSMiaohe Lin 11721da177e4SLinus Torvalds /* 1173a67c8caaSMiaohe Lin * We might reach here with anon_vma == NULL if we can't find 1174a67c8caaSMiaohe Lin * any reusable anon_vma. 11751da177e4SLinus Torvalds * There's no absolute need to look only at touching neighbours: 11761da177e4SLinus Torvalds * we could search further afield for "compatible" anon_vmas. 11771da177e4SLinus Torvalds * But it would probably just be a waste of time searching, 11781da177e4SLinus Torvalds * or lead to too many vmas hanging off the same anon_vma. 11791da177e4SLinus Torvalds * We're trying to allow mprotect remerging later on, 11801da177e4SLinus Torvalds * not trying to minimize memory used for anon_vmas. 11811da177e4SLinus Torvalds */ 1182a67c8caaSMiaohe Lin return anon_vma; 11831da177e4SLinus Torvalds } 11841da177e4SLinus Torvalds 11851da177e4SLinus Torvalds /* 118640401530SAl Viro * If a hint addr is less than mmap_min_addr change hint to be as 118740401530SAl Viro * low as possible but still greater than mmap_min_addr 118840401530SAl Viro */ 118940401530SAl Viro static inline unsigned long round_hint_to_min(unsigned long hint) 119040401530SAl Viro { 119140401530SAl Viro hint &= PAGE_MASK; 119240401530SAl Viro if (((void *)hint != NULL) && 119340401530SAl Viro (hint < mmap_min_addr)) 119440401530SAl Viro return PAGE_ALIGN(mmap_min_addr); 119540401530SAl Viro return hint; 119640401530SAl Viro } 119740401530SAl Viro 11986aeb2542SMike Rapoport int mlock_future_check(struct mm_struct *mm, unsigned long flags, 1199363ee17fSDavidlohr Bueso unsigned long len) 1200363ee17fSDavidlohr Bueso { 1201363ee17fSDavidlohr Bueso unsigned long locked, lock_limit; 1202363ee17fSDavidlohr Bueso 1203363ee17fSDavidlohr Bueso /* mlock MCL_FUTURE? */ 1204363ee17fSDavidlohr Bueso if (flags & VM_LOCKED) { 1205363ee17fSDavidlohr Bueso locked = len >> PAGE_SHIFT; 1206363ee17fSDavidlohr Bueso locked += mm->locked_vm; 1207363ee17fSDavidlohr Bueso lock_limit = rlimit(RLIMIT_MEMLOCK); 1208363ee17fSDavidlohr Bueso lock_limit >>= PAGE_SHIFT; 1209363ee17fSDavidlohr Bueso if (locked > lock_limit && !capable(CAP_IPC_LOCK)) 1210363ee17fSDavidlohr Bueso return -EAGAIN; 1211363ee17fSDavidlohr Bueso } 1212363ee17fSDavidlohr Bueso return 0; 1213363ee17fSDavidlohr Bueso } 1214363ee17fSDavidlohr Bueso 1215be83bbf8SLinus Torvalds static inline u64 file_mmap_size_max(struct file *file, struct inode *inode) 1216be83bbf8SLinus Torvalds { 1217be83bbf8SLinus Torvalds if (S_ISREG(inode->i_mode)) 1218423913adSLinus Torvalds return MAX_LFS_FILESIZE; 1219be83bbf8SLinus Torvalds 1220be83bbf8SLinus Torvalds if (S_ISBLK(inode->i_mode)) 1221be83bbf8SLinus Torvalds return MAX_LFS_FILESIZE; 1222be83bbf8SLinus Torvalds 122376f34950SIvan Khoronzhuk if (S_ISSOCK(inode->i_mode)) 122476f34950SIvan Khoronzhuk return MAX_LFS_FILESIZE; 122576f34950SIvan Khoronzhuk 1226be83bbf8SLinus Torvalds /* Special "we do even unsigned file positions" case */ 1227be83bbf8SLinus Torvalds if (file->f_mode & FMODE_UNSIGNED_OFFSET) 1228be83bbf8SLinus Torvalds return 0; 1229be83bbf8SLinus Torvalds 1230be83bbf8SLinus Torvalds /* Yes, random drivers might want more. But I'm tired of buggy drivers */ 1231be83bbf8SLinus Torvalds return ULONG_MAX; 1232be83bbf8SLinus Torvalds } 1233be83bbf8SLinus Torvalds 1234be83bbf8SLinus Torvalds static inline bool file_mmap_ok(struct file *file, struct inode *inode, 1235be83bbf8SLinus Torvalds unsigned long pgoff, unsigned long len) 1236be83bbf8SLinus Torvalds { 1237be83bbf8SLinus Torvalds u64 maxsize = file_mmap_size_max(file, inode); 1238be83bbf8SLinus Torvalds 1239be83bbf8SLinus Torvalds if (maxsize && len > maxsize) 1240be83bbf8SLinus Torvalds return false; 1241be83bbf8SLinus Torvalds maxsize -= len; 1242be83bbf8SLinus Torvalds if (pgoff > maxsize >> PAGE_SHIFT) 1243be83bbf8SLinus Torvalds return false; 1244be83bbf8SLinus Torvalds return true; 1245be83bbf8SLinus Torvalds } 1246be83bbf8SLinus Torvalds 124740401530SAl Viro /* 12483e4e28c5SMichel Lespinasse * The caller must write-lock current->mm->mmap_lock. 12491da177e4SLinus Torvalds */ 12501fcfd8dbSOleg Nesterov unsigned long do_mmap(struct file *file, unsigned long addr, 12511da177e4SLinus Torvalds unsigned long len, unsigned long prot, 125245e55300SPeter Collingbourne unsigned long flags, unsigned long pgoff, 125345e55300SPeter Collingbourne unsigned long *populate, struct list_head *uf) 12541da177e4SLinus Torvalds { 12551da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 125645e55300SPeter Collingbourne vm_flags_t vm_flags; 125762b5f7d0SDave Hansen int pkey = 0; 12581da177e4SLinus Torvalds 1259524e00b3SLiam R. Howlett validate_mm(mm); 126041badc15SMichel Lespinasse *populate = 0; 1261bebeb3d6SMichel Lespinasse 1262e37609bbSPiotr Kwapulinski if (!len) 1263e37609bbSPiotr Kwapulinski return -EINVAL; 1264e37609bbSPiotr Kwapulinski 12651da177e4SLinus Torvalds /* 12661da177e4SLinus Torvalds * Does the application expect PROT_READ to imply PROT_EXEC? 12671da177e4SLinus Torvalds * 12681da177e4SLinus Torvalds * (the exception is when the underlying filesystem is noexec 12691da177e4SLinus Torvalds * mounted, in which case we dont add PROT_EXEC.) 12701da177e4SLinus Torvalds */ 12711da177e4SLinus Torvalds if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC)) 127290f8572bSEric W. Biederman if (!(file && path_noexec(&file->f_path))) 12731da177e4SLinus Torvalds prot |= PROT_EXEC; 12741da177e4SLinus Torvalds 1275a4ff8e86SMichal Hocko /* force arch specific MAP_FIXED handling in get_unmapped_area */ 1276a4ff8e86SMichal Hocko if (flags & MAP_FIXED_NOREPLACE) 1277a4ff8e86SMichal Hocko flags |= MAP_FIXED; 1278a4ff8e86SMichal Hocko 12797cd94146SEric Paris if (!(flags & MAP_FIXED)) 12807cd94146SEric Paris addr = round_hint_to_min(addr); 12817cd94146SEric Paris 12821da177e4SLinus Torvalds /* Careful about overflows.. */ 12831da177e4SLinus Torvalds len = PAGE_ALIGN(len); 12849206de95SAl Viro if (!len) 12851da177e4SLinus Torvalds return -ENOMEM; 12861da177e4SLinus Torvalds 12871da177e4SLinus Torvalds /* offset overflow? */ 12881da177e4SLinus Torvalds if ((pgoff + (len >> PAGE_SHIFT)) < pgoff) 12891da177e4SLinus Torvalds return -EOVERFLOW; 12901da177e4SLinus Torvalds 12911da177e4SLinus Torvalds /* Too many mappings? */ 12921da177e4SLinus Torvalds if (mm->map_count > sysctl_max_map_count) 12931da177e4SLinus Torvalds return -ENOMEM; 12941da177e4SLinus Torvalds 12951da177e4SLinus Torvalds /* Obtain the address to map to. we verify (or select) it and ensure 12961da177e4SLinus Torvalds * that it represents a valid section of the address space. 12971da177e4SLinus Torvalds */ 12981da177e4SLinus Torvalds addr = get_unmapped_area(file, addr, len, pgoff, flags); 1299ff68dac6SGaowei Pu if (IS_ERR_VALUE(addr)) 13001da177e4SLinus Torvalds return addr; 13011da177e4SLinus Torvalds 1302a4ff8e86SMichal Hocko if (flags & MAP_FIXED_NOREPLACE) { 130335e43c5fSLiam Howlett if (find_vma_intersection(mm, addr, addr + len)) 1304a4ff8e86SMichal Hocko return -EEXIST; 1305a4ff8e86SMichal Hocko } 1306a4ff8e86SMichal Hocko 130762b5f7d0SDave Hansen if (prot == PROT_EXEC) { 130862b5f7d0SDave Hansen pkey = execute_only_pkey(mm); 130962b5f7d0SDave Hansen if (pkey < 0) 131062b5f7d0SDave Hansen pkey = 0; 131162b5f7d0SDave Hansen } 131262b5f7d0SDave Hansen 13131da177e4SLinus Torvalds /* Do simple checking here so the lower-level routines won't have 13141da177e4SLinus Torvalds * to. we assume access permissions have been handled by the open 13151da177e4SLinus Torvalds * of the memory object, so we don't do any here. 13161da177e4SLinus Torvalds */ 131745e55300SPeter Collingbourne vm_flags = calc_vm_prot_bits(prot, pkey) | calc_vm_flag_bits(flags) | 13181da177e4SLinus Torvalds mm->def_flags | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC; 13191da177e4SLinus Torvalds 1320cdf7b341SHuang Shijie if (flags & MAP_LOCKED) 13211da177e4SLinus Torvalds if (!can_do_mlock()) 13221da177e4SLinus Torvalds return -EPERM; 1323ba470de4SRik van Riel 1324363ee17fSDavidlohr Bueso if (mlock_future_check(mm, vm_flags, len)) 13251da177e4SLinus Torvalds return -EAGAIN; 13261da177e4SLinus Torvalds 13271da177e4SLinus Torvalds if (file) { 1328077bf22bSOleg Nesterov struct inode *inode = file_inode(file); 13291c972597SDan Williams unsigned long flags_mask; 13301c972597SDan Williams 1331be83bbf8SLinus Torvalds if (!file_mmap_ok(file, inode, pgoff, len)) 1332be83bbf8SLinus Torvalds return -EOVERFLOW; 1333be83bbf8SLinus Torvalds 13341c972597SDan Williams flags_mask = LEGACY_MAP_MASK | file->f_op->mmap_supported_flags; 1335077bf22bSOleg Nesterov 13361da177e4SLinus Torvalds switch (flags & MAP_TYPE) { 13371da177e4SLinus Torvalds case MAP_SHARED: 13381c972597SDan Williams /* 13391c972597SDan Williams * Force use of MAP_SHARED_VALIDATE with non-legacy 13401c972597SDan Williams * flags. E.g. MAP_SYNC is dangerous to use with 13411c972597SDan Williams * MAP_SHARED as you don't know which consistency model 13421c972597SDan Williams * you will get. We silently ignore unsupported flags 13431c972597SDan Williams * with MAP_SHARED to preserve backward compatibility. 13441c972597SDan Williams */ 13451c972597SDan Williams flags &= LEGACY_MAP_MASK; 1346e4a9bc58SJoe Perches fallthrough; 13471c972597SDan Williams case MAP_SHARED_VALIDATE: 13481c972597SDan Williams if (flags & ~flags_mask) 13491c972597SDan Williams return -EOPNOTSUPP; 1350dc617f29SDarrick J. Wong if (prot & PROT_WRITE) { 1351dc617f29SDarrick J. Wong if (!(file->f_mode & FMODE_WRITE)) 13521da177e4SLinus Torvalds return -EACCES; 1353dc617f29SDarrick J. Wong if (IS_SWAPFILE(file->f_mapping->host)) 1354dc617f29SDarrick J. Wong return -ETXTBSY; 1355dc617f29SDarrick J. Wong } 13561da177e4SLinus Torvalds 13571da177e4SLinus Torvalds /* 13581da177e4SLinus Torvalds * Make sure we don't allow writing to an append-only 13591da177e4SLinus Torvalds * file.. 13601da177e4SLinus Torvalds */ 13611da177e4SLinus Torvalds if (IS_APPEND(inode) && (file->f_mode & FMODE_WRITE)) 13621da177e4SLinus Torvalds return -EACCES; 13631da177e4SLinus Torvalds 13641da177e4SLinus Torvalds vm_flags |= VM_SHARED | VM_MAYSHARE; 13651da177e4SLinus Torvalds if (!(file->f_mode & FMODE_WRITE)) 13661da177e4SLinus Torvalds vm_flags &= ~(VM_MAYWRITE | VM_SHARED); 1367e4a9bc58SJoe Perches fallthrough; 13681da177e4SLinus Torvalds case MAP_PRIVATE: 13691da177e4SLinus Torvalds if (!(file->f_mode & FMODE_READ)) 13701da177e4SLinus Torvalds return -EACCES; 137190f8572bSEric W. Biederman if (path_noexec(&file->f_path)) { 137280c5606cSLinus Torvalds if (vm_flags & VM_EXEC) 137380c5606cSLinus Torvalds return -EPERM; 137480c5606cSLinus Torvalds vm_flags &= ~VM_MAYEXEC; 137580c5606cSLinus Torvalds } 137680c5606cSLinus Torvalds 137772c2d531SAl Viro if (!file->f_op->mmap) 137880c5606cSLinus Torvalds return -ENODEV; 1379b2c56e4fSOleg Nesterov if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP)) 1380b2c56e4fSOleg Nesterov return -EINVAL; 13811da177e4SLinus Torvalds break; 13821da177e4SLinus Torvalds 13831da177e4SLinus Torvalds default: 13841da177e4SLinus Torvalds return -EINVAL; 13851da177e4SLinus Torvalds } 13861da177e4SLinus Torvalds } else { 13871da177e4SLinus Torvalds switch (flags & MAP_TYPE) { 13881da177e4SLinus Torvalds case MAP_SHARED: 1389b2c56e4fSOleg Nesterov if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP)) 1390b2c56e4fSOleg Nesterov return -EINVAL; 1391ce363942STejun Heo /* 1392ce363942STejun Heo * Ignore pgoff. 1393ce363942STejun Heo */ 1394ce363942STejun Heo pgoff = 0; 13951da177e4SLinus Torvalds vm_flags |= VM_SHARED | VM_MAYSHARE; 13961da177e4SLinus Torvalds break; 13971da177e4SLinus Torvalds case MAP_PRIVATE: 13981da177e4SLinus Torvalds /* 13991da177e4SLinus Torvalds * Set pgoff according to addr for anon_vma. 14001da177e4SLinus Torvalds */ 14011da177e4SLinus Torvalds pgoff = addr >> PAGE_SHIFT; 14021da177e4SLinus Torvalds break; 14031da177e4SLinus Torvalds default: 14041da177e4SLinus Torvalds return -EINVAL; 14051da177e4SLinus Torvalds } 14061da177e4SLinus Torvalds } 14071da177e4SLinus Torvalds 1408c22c0d63SMichel Lespinasse /* 1409c22c0d63SMichel Lespinasse * Set 'VM_NORESERVE' if we should not account for the 1410c22c0d63SMichel Lespinasse * memory use of this mapping. 1411c22c0d63SMichel Lespinasse */ 1412c22c0d63SMichel Lespinasse if (flags & MAP_NORESERVE) { 1413c22c0d63SMichel Lespinasse /* We honor MAP_NORESERVE if allowed to overcommit */ 1414c22c0d63SMichel Lespinasse if (sysctl_overcommit_memory != OVERCOMMIT_NEVER) 1415c22c0d63SMichel Lespinasse vm_flags |= VM_NORESERVE; 1416c22c0d63SMichel Lespinasse 1417c22c0d63SMichel Lespinasse /* hugetlb applies strict overcommit unless MAP_NORESERVE */ 1418c22c0d63SMichel Lespinasse if (file && is_file_hugepages(file)) 1419c22c0d63SMichel Lespinasse vm_flags |= VM_NORESERVE; 1420c22c0d63SMichel Lespinasse } 1421c22c0d63SMichel Lespinasse 1422897ab3e0SMike Rapoport addr = mmap_region(file, addr, len, vm_flags, pgoff, uf); 142309a9f1d2SMichel Lespinasse if (!IS_ERR_VALUE(addr) && 142409a9f1d2SMichel Lespinasse ((vm_flags & VM_LOCKED) || 142509a9f1d2SMichel Lespinasse (flags & (MAP_POPULATE | MAP_NONBLOCK)) == MAP_POPULATE)) 142641badc15SMichel Lespinasse *populate = len; 1427bebeb3d6SMichel Lespinasse return addr; 14280165ab44SMiklos Szeredi } 14296be5ceb0SLinus Torvalds 1430a90f590aSDominik Brodowski unsigned long ksys_mmap_pgoff(unsigned long addr, unsigned long len, 1431a90f590aSDominik Brodowski unsigned long prot, unsigned long flags, 1432a90f590aSDominik Brodowski unsigned long fd, unsigned long pgoff) 143366f0dc48SHugh Dickins { 143466f0dc48SHugh Dickins struct file *file = NULL; 14351e3ee14bSChen Gang unsigned long retval; 143666f0dc48SHugh Dickins 143766f0dc48SHugh Dickins if (!(flags & MAP_ANONYMOUS)) { 1438120a795dSAl Viro audit_mmap_fd(fd, flags); 143966f0dc48SHugh Dickins file = fget(fd); 144066f0dc48SHugh Dickins if (!file) 14411e3ee14bSChen Gang return -EBADF; 14427bba8f0eSZhen Lei if (is_file_hugepages(file)) { 1443af73e4d9SNaoya Horiguchi len = ALIGN(len, huge_page_size(hstate_file(file))); 14447bba8f0eSZhen Lei } else if (unlikely(flags & MAP_HUGETLB)) { 1445493af578SJörn Engel retval = -EINVAL; 1446493af578SJörn Engel goto out_fput; 14477bba8f0eSZhen Lei } 144866f0dc48SHugh Dickins } else if (flags & MAP_HUGETLB) { 1449c103a4dcSAndrew Morton struct hstate *hs; 1450af73e4d9SNaoya Horiguchi 145120ac2893SAnshuman Khandual hs = hstate_sizelog((flags >> MAP_HUGE_SHIFT) & MAP_HUGE_MASK); 1452091d0d55SLi Zefan if (!hs) 1453091d0d55SLi Zefan return -EINVAL; 1454091d0d55SLi Zefan 1455091d0d55SLi Zefan len = ALIGN(len, huge_page_size(hs)); 145666f0dc48SHugh Dickins /* 145766f0dc48SHugh Dickins * VM_NORESERVE is used because the reservations will be 145866f0dc48SHugh Dickins * taken when vm_ops->mmap() is called 145966f0dc48SHugh Dickins */ 1460af73e4d9SNaoya Horiguchi file = hugetlb_file_setup(HUGETLB_ANON_FILE, len, 146142d7395fSAndi Kleen VM_NORESERVE, 146283c1fd76Szhangyiru HUGETLB_ANONHUGE_INODE, 146342d7395fSAndi Kleen (flags >> MAP_HUGE_SHIFT) & MAP_HUGE_MASK); 146466f0dc48SHugh Dickins if (IS_ERR(file)) 146566f0dc48SHugh Dickins return PTR_ERR(file); 146666f0dc48SHugh Dickins } 146766f0dc48SHugh Dickins 14689fbeb5abSMichal Hocko retval = vm_mmap_pgoff(file, addr, len, prot, flags, pgoff); 1469493af578SJörn Engel out_fput: 147066f0dc48SHugh Dickins if (file) 147166f0dc48SHugh Dickins fput(file); 147266f0dc48SHugh Dickins return retval; 147366f0dc48SHugh Dickins } 147466f0dc48SHugh Dickins 1475a90f590aSDominik Brodowski SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len, 1476a90f590aSDominik Brodowski unsigned long, prot, unsigned long, flags, 1477a90f590aSDominik Brodowski unsigned long, fd, unsigned long, pgoff) 1478a90f590aSDominik Brodowski { 1479a90f590aSDominik Brodowski return ksys_mmap_pgoff(addr, len, prot, flags, fd, pgoff); 1480a90f590aSDominik Brodowski } 1481a90f590aSDominik Brodowski 1482a4679373SChristoph Hellwig #ifdef __ARCH_WANT_SYS_OLD_MMAP 1483a4679373SChristoph Hellwig struct mmap_arg_struct { 1484a4679373SChristoph Hellwig unsigned long addr; 1485a4679373SChristoph Hellwig unsigned long len; 1486a4679373SChristoph Hellwig unsigned long prot; 1487a4679373SChristoph Hellwig unsigned long flags; 1488a4679373SChristoph Hellwig unsigned long fd; 1489a4679373SChristoph Hellwig unsigned long offset; 1490a4679373SChristoph Hellwig }; 1491a4679373SChristoph Hellwig 1492a4679373SChristoph Hellwig SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg) 1493a4679373SChristoph Hellwig { 1494a4679373SChristoph Hellwig struct mmap_arg_struct a; 1495a4679373SChristoph Hellwig 1496a4679373SChristoph Hellwig if (copy_from_user(&a, arg, sizeof(a))) 1497a4679373SChristoph Hellwig return -EFAULT; 1498de1741a1SAlexander Kuleshov if (offset_in_page(a.offset)) 1499a4679373SChristoph Hellwig return -EINVAL; 1500a4679373SChristoph Hellwig 1501a90f590aSDominik Brodowski return ksys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd, 1502a4679373SChristoph Hellwig a.offset >> PAGE_SHIFT); 1503a4679373SChristoph Hellwig } 1504a4679373SChristoph Hellwig #endif /* __ARCH_WANT_SYS_OLD_MMAP */ 1505a4679373SChristoph Hellwig 15064e950f6fSAlexey Dobriyan /* 15078bb4e7a2SWei Yang * Some shared mappings will want the pages marked read-only 15084e950f6fSAlexey Dobriyan * to track write events. If so, we'll downgrade vm_page_prot 15094e950f6fSAlexey Dobriyan * to the private version (using protection_map[] without the 15104e950f6fSAlexey Dobriyan * VM_SHARED bit). 15114e950f6fSAlexey Dobriyan */ 15126d2329f8SAndrea Arcangeli int vma_wants_writenotify(struct vm_area_struct *vma, pgprot_t vm_page_prot) 15134e950f6fSAlexey Dobriyan { 1514ca16d140SKOSAKI Motohiro vm_flags_t vm_flags = vma->vm_flags; 15158a04446aSKirill A. Shutemov const struct vm_operations_struct *vm_ops = vma->vm_ops; 15164e950f6fSAlexey Dobriyan 15174e950f6fSAlexey Dobriyan /* If it was private or non-writable, the write bit is already clear */ 15184e950f6fSAlexey Dobriyan if ((vm_flags & (VM_WRITE|VM_SHARED)) != ((VM_WRITE|VM_SHARED))) 15194e950f6fSAlexey Dobriyan return 0; 15204e950f6fSAlexey Dobriyan 15214e950f6fSAlexey Dobriyan /* The backer wishes to know when pages are first written to? */ 15228a04446aSKirill A. Shutemov if (vm_ops && (vm_ops->page_mkwrite || vm_ops->pfn_mkwrite)) 15234e950f6fSAlexey Dobriyan return 1; 15244e950f6fSAlexey Dobriyan 152564e45507SPeter Feiner /* The open routine did something to the protections that pgprot_modify 152664e45507SPeter Feiner * won't preserve? */ 15276d2329f8SAndrea Arcangeli if (pgprot_val(vm_page_prot) != 15286d2329f8SAndrea Arcangeli pgprot_val(vm_pgprot_modify(vm_page_prot, vm_flags))) 15294e950f6fSAlexey Dobriyan return 0; 15304e950f6fSAlexey Dobriyan 1531f96f7a40SDavid Hildenbrand /* 1532f96f7a40SDavid Hildenbrand * Do we need to track softdirty? hugetlb does not support softdirty 1533f96f7a40SDavid Hildenbrand * tracking yet. 1534f96f7a40SDavid Hildenbrand */ 1535f96f7a40SDavid Hildenbrand if (vma_soft_dirty_enabled(vma) && !is_vm_hugetlb_page(vma)) 153664e45507SPeter Feiner return 1; 153764e45507SPeter Feiner 153851d3d5ebSDavid Hildenbrand /* Do we need write faults for uffd-wp tracking? */ 153951d3d5ebSDavid Hildenbrand if (userfaultfd_wp(vma)) 154051d3d5ebSDavid Hildenbrand return 1; 154151d3d5ebSDavid Hildenbrand 15424e950f6fSAlexey Dobriyan /* Specialty mapping? */ 15434b6e1e37SKonstantin Khlebnikov if (vm_flags & VM_PFNMAP) 15444e950f6fSAlexey Dobriyan return 0; 15454e950f6fSAlexey Dobriyan 15464e950f6fSAlexey Dobriyan /* Can the mapping track the dirty pages? */ 15474e950f6fSAlexey Dobriyan return vma->vm_file && vma->vm_file->f_mapping && 1548f56753acSChristoph Hellwig mapping_can_writeback(vma->vm_file->f_mapping); 15494e950f6fSAlexey Dobriyan } 15504e950f6fSAlexey Dobriyan 1551fc8744adSLinus Torvalds /* 1552fc8744adSLinus Torvalds * We account for memory if it's a private writeable mapping, 15535a6fe125SMel Gorman * not hugepages and VM_NORESERVE wasn't set. 1554fc8744adSLinus Torvalds */ 1555ca16d140SKOSAKI Motohiro static inline int accountable_mapping(struct file *file, vm_flags_t vm_flags) 1556fc8744adSLinus Torvalds { 15575a6fe125SMel Gorman /* 15585a6fe125SMel Gorman * hugetlb has its own accounting separate from the core VM 15595a6fe125SMel Gorman * VM_HUGETLB may not be set yet so we cannot check for that flag. 15605a6fe125SMel Gorman */ 15615a6fe125SMel Gorman if (file && is_file_hugepages(file)) 15625a6fe125SMel Gorman return 0; 15635a6fe125SMel Gorman 1564fc8744adSLinus Torvalds return (vm_flags & (VM_NORESERVE | VM_SHARED | VM_WRITE)) == VM_WRITE; 1565fc8744adSLinus Torvalds } 1566fc8744adSLinus Torvalds 15673499a131SLiam R. Howlett /** 15683499a131SLiam R. Howlett * unmapped_area() - Find an area between the low_limit and the high_limit with 15693499a131SLiam R. Howlett * the correct alignment and offset, all from @info. Note: current->mm is used 15703499a131SLiam R. Howlett * for the search. 15713499a131SLiam R. Howlett * 157282b24936SVernon Yang * @info: The unmapped area information including the range [low_limit - 157382b24936SVernon Yang * high_limit), the alignment offset and mask. 15743499a131SLiam R. Howlett * 15753499a131SLiam R. Howlett * Return: A memory address or -ENOMEM. 15763499a131SLiam R. Howlett */ 1577baceaf1cSJaewon Kim static unsigned long unmapped_area(struct vm_unmapped_area_info *info) 1578db4fbfb9SMichel Lespinasse { 15793499a131SLiam R. Howlett unsigned long length, gap; 1580db4fbfb9SMichel Lespinasse 15813499a131SLiam R. Howlett MA_STATE(mas, ¤t->mm->mm_mt, 0, 0); 1582db4fbfb9SMichel Lespinasse 1583db4fbfb9SMichel Lespinasse /* Adjust search length to account for worst case alignment overhead */ 1584db4fbfb9SMichel Lespinasse length = info->length + info->align_mask; 1585db4fbfb9SMichel Lespinasse if (length < info->length) 1586db4fbfb9SMichel Lespinasse return -ENOMEM; 1587db4fbfb9SMichel Lespinasse 15883499a131SLiam R. Howlett if (mas_empty_area(&mas, info->low_limit, info->high_limit - 1, 15893499a131SLiam R. Howlett length)) 15903499a131SLiam R. Howlett return -ENOMEM; 15913499a131SLiam R. Howlett 1592d4af56c5SLiam R. Howlett gap = mas.index; 1593d4af56c5SLiam R. Howlett gap += (info->align_offset - gap) & info->align_mask; 15943499a131SLiam R. Howlett return gap; 1595db4fbfb9SMichel Lespinasse } 1596db4fbfb9SMichel Lespinasse 15973499a131SLiam R. Howlett /** 15983499a131SLiam R. Howlett * unmapped_area_topdown() - Find an area between the low_limit and the 159982b24936SVernon Yang * high_limit with the correct alignment and offset at the highest available 16003499a131SLiam R. Howlett * address, all from @info. Note: current->mm is used for the search. 16013499a131SLiam R. Howlett * 160282b24936SVernon Yang * @info: The unmapped area information including the range [low_limit - 160382b24936SVernon Yang * high_limit), the alignment offset and mask. 16043499a131SLiam R. Howlett * 16053499a131SLiam R. Howlett * Return: A memory address or -ENOMEM. 16063499a131SLiam R. Howlett */ 1607baceaf1cSJaewon Kim static unsigned long unmapped_area_topdown(struct vm_unmapped_area_info *info) 1608db4fbfb9SMichel Lespinasse { 16093499a131SLiam R. Howlett unsigned long length, gap; 1610d4af56c5SLiam R. Howlett 16113499a131SLiam R. Howlett MA_STATE(mas, ¤t->mm->mm_mt, 0, 0); 1612db4fbfb9SMichel Lespinasse /* Adjust search length to account for worst case alignment overhead */ 1613db4fbfb9SMichel Lespinasse length = info->length + info->align_mask; 1614db4fbfb9SMichel Lespinasse if (length < info->length) 1615db4fbfb9SMichel Lespinasse return -ENOMEM; 1616db4fbfb9SMichel Lespinasse 16173499a131SLiam R. Howlett if (mas_empty_area_rev(&mas, info->low_limit, info->high_limit - 1, 16183499a131SLiam R. Howlett length)) 16193499a131SLiam R. Howlett return -ENOMEM; 16203499a131SLiam R. Howlett 1621d4af56c5SLiam R. Howlett gap = mas.last + 1 - info->length; 1622d4af56c5SLiam R. Howlett gap -= (gap - info->align_offset) & info->align_mask; 16233499a131SLiam R. Howlett return gap; 1624db4fbfb9SMichel Lespinasse } 1625db4fbfb9SMichel Lespinasse 1626baceaf1cSJaewon Kim /* 1627baceaf1cSJaewon Kim * Search for an unmapped address range. 1628baceaf1cSJaewon Kim * 1629baceaf1cSJaewon Kim * We are looking for a range that: 1630baceaf1cSJaewon Kim * - does not intersect with any VMA; 1631baceaf1cSJaewon Kim * - is contained within the [low_limit, high_limit) interval; 1632baceaf1cSJaewon Kim * - is at least the desired size. 1633baceaf1cSJaewon Kim * - satisfies (begin_addr & align_mask) == (align_offset & align_mask) 1634baceaf1cSJaewon Kim */ 1635baceaf1cSJaewon Kim unsigned long vm_unmapped_area(struct vm_unmapped_area_info *info) 1636baceaf1cSJaewon Kim { 1637df529cabSJaewon Kim unsigned long addr; 1638df529cabSJaewon Kim 1639baceaf1cSJaewon Kim if (info->flags & VM_UNMAPPED_AREA_TOPDOWN) 1640df529cabSJaewon Kim addr = unmapped_area_topdown(info); 1641baceaf1cSJaewon Kim else 1642df529cabSJaewon Kim addr = unmapped_area(info); 1643df529cabSJaewon Kim 1644df529cabSJaewon Kim trace_vm_unmapped_area(addr, info); 1645df529cabSJaewon Kim return addr; 1646baceaf1cSJaewon Kim } 1647f6795053SSteve Capper 16481da177e4SLinus Torvalds /* Get an address range which is currently unmapped. 16491da177e4SLinus Torvalds * For shmat() with addr=0. 16501da177e4SLinus Torvalds * 16511da177e4SLinus Torvalds * Ugly calling convention alert: 16521da177e4SLinus Torvalds * Return value with the low bits set means error value, 16531da177e4SLinus Torvalds * ie 16541da177e4SLinus Torvalds * if (ret & ~PAGE_MASK) 16551da177e4SLinus Torvalds * error = ret; 16561da177e4SLinus Torvalds * 16571da177e4SLinus Torvalds * This function "knows" that -ENOMEM has the bits set. 16581da177e4SLinus Torvalds */ 16591da177e4SLinus Torvalds unsigned long 16604b439e25SChristophe Leroy generic_get_unmapped_area(struct file *filp, unsigned long addr, 16614b439e25SChristophe Leroy unsigned long len, unsigned long pgoff, 16624b439e25SChristophe Leroy unsigned long flags) 16631da177e4SLinus Torvalds { 16641da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 16651be7107fSHugh Dickins struct vm_area_struct *vma, *prev; 1666db4fbfb9SMichel Lespinasse struct vm_unmapped_area_info info; 16672cb4de08SChristophe Leroy const unsigned long mmap_end = arch_get_mmap_end(addr, len, flags); 16681da177e4SLinus Torvalds 1669f6795053SSteve Capper if (len > mmap_end - mmap_min_addr) 16701da177e4SLinus Torvalds return -ENOMEM; 16711da177e4SLinus Torvalds 167206abdfb4SBenjamin Herrenschmidt if (flags & MAP_FIXED) 167306abdfb4SBenjamin Herrenschmidt return addr; 167406abdfb4SBenjamin Herrenschmidt 16751da177e4SLinus Torvalds if (addr) { 16761da177e4SLinus Torvalds addr = PAGE_ALIGN(addr); 16771be7107fSHugh Dickins vma = find_vma_prev(mm, addr, &prev); 1678f6795053SSteve Capper if (mmap_end - len >= addr && addr >= mmap_min_addr && 16791be7107fSHugh Dickins (!vma || addr + len <= vm_start_gap(vma)) && 16801be7107fSHugh Dickins (!prev || addr >= vm_end_gap(prev))) 16811da177e4SLinus Torvalds return addr; 16821da177e4SLinus Torvalds } 16831da177e4SLinus Torvalds 1684db4fbfb9SMichel Lespinasse info.flags = 0; 1685db4fbfb9SMichel Lespinasse info.length = len; 16864e99b021SHeiko Carstens info.low_limit = mm->mmap_base; 1687f6795053SSteve Capper info.high_limit = mmap_end; 1688db4fbfb9SMichel Lespinasse info.align_mask = 0; 168909ef5283SJaewon Kim info.align_offset = 0; 1690db4fbfb9SMichel Lespinasse return vm_unmapped_area(&info); 16911da177e4SLinus Torvalds } 16924b439e25SChristophe Leroy 16934b439e25SChristophe Leroy #ifndef HAVE_ARCH_UNMAPPED_AREA 16944b439e25SChristophe Leroy unsigned long 16954b439e25SChristophe Leroy arch_get_unmapped_area(struct file *filp, unsigned long addr, 16964b439e25SChristophe Leroy unsigned long len, unsigned long pgoff, 16974b439e25SChristophe Leroy unsigned long flags) 16984b439e25SChristophe Leroy { 16994b439e25SChristophe Leroy return generic_get_unmapped_area(filp, addr, len, pgoff, flags); 17004b439e25SChristophe Leroy } 17011da177e4SLinus Torvalds #endif 17021da177e4SLinus Torvalds 17031da177e4SLinus Torvalds /* 17041da177e4SLinus Torvalds * This mmap-allocator allocates new areas top-down from below the 17051da177e4SLinus Torvalds * stack's low limit (the base): 17061da177e4SLinus Torvalds */ 17071da177e4SLinus Torvalds unsigned long 17084b439e25SChristophe Leroy generic_get_unmapped_area_topdown(struct file *filp, unsigned long addr, 170943cca0b1SYang Fan unsigned long len, unsigned long pgoff, 171043cca0b1SYang Fan unsigned long flags) 17111da177e4SLinus Torvalds { 17121be7107fSHugh Dickins struct vm_area_struct *vma, *prev; 17131da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 1714db4fbfb9SMichel Lespinasse struct vm_unmapped_area_info info; 17152cb4de08SChristophe Leroy const unsigned long mmap_end = arch_get_mmap_end(addr, len, flags); 17161da177e4SLinus Torvalds 17171da177e4SLinus Torvalds /* requested length too big for entire address space */ 1718f6795053SSteve Capper if (len > mmap_end - mmap_min_addr) 17191da177e4SLinus Torvalds return -ENOMEM; 17201da177e4SLinus Torvalds 172106abdfb4SBenjamin Herrenschmidt if (flags & MAP_FIXED) 172206abdfb4SBenjamin Herrenschmidt return addr; 172306abdfb4SBenjamin Herrenschmidt 17241da177e4SLinus Torvalds /* requesting a specific address */ 17251da177e4SLinus Torvalds if (addr) { 17261da177e4SLinus Torvalds addr = PAGE_ALIGN(addr); 17271be7107fSHugh Dickins vma = find_vma_prev(mm, addr, &prev); 1728f6795053SSteve Capper if (mmap_end - len >= addr && addr >= mmap_min_addr && 17291be7107fSHugh Dickins (!vma || addr + len <= vm_start_gap(vma)) && 17301be7107fSHugh Dickins (!prev || addr >= vm_end_gap(prev))) 17311da177e4SLinus Torvalds return addr; 17321da177e4SLinus Torvalds } 17331da177e4SLinus Torvalds 1734db4fbfb9SMichel Lespinasse info.flags = VM_UNMAPPED_AREA_TOPDOWN; 1735db4fbfb9SMichel Lespinasse info.length = len; 17362afc745fSAkira Takeuchi info.low_limit = max(PAGE_SIZE, mmap_min_addr); 1737f6795053SSteve Capper info.high_limit = arch_get_mmap_base(addr, mm->mmap_base); 1738db4fbfb9SMichel Lespinasse info.align_mask = 0; 173909ef5283SJaewon Kim info.align_offset = 0; 1740db4fbfb9SMichel Lespinasse addr = vm_unmapped_area(&info); 1741b716ad95SXiao Guangrong 17421da177e4SLinus Torvalds /* 17431da177e4SLinus Torvalds * A failed mmap() very likely causes application failure, 17441da177e4SLinus Torvalds * so fall back to the bottom-up function here. This scenario 17451da177e4SLinus Torvalds * can happen with large stack limits and large mmap() 17461da177e4SLinus Torvalds * allocations. 17471da177e4SLinus Torvalds */ 1748de1741a1SAlexander Kuleshov if (offset_in_page(addr)) { 1749db4fbfb9SMichel Lespinasse VM_BUG_ON(addr != -ENOMEM); 1750db4fbfb9SMichel Lespinasse info.flags = 0; 1751db4fbfb9SMichel Lespinasse info.low_limit = TASK_UNMAPPED_BASE; 1752f6795053SSteve Capper info.high_limit = mmap_end; 1753db4fbfb9SMichel Lespinasse addr = vm_unmapped_area(&info); 1754db4fbfb9SMichel Lespinasse } 17551da177e4SLinus Torvalds 17561da177e4SLinus Torvalds return addr; 17571da177e4SLinus Torvalds } 17584b439e25SChristophe Leroy 17594b439e25SChristophe Leroy #ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN 17604b439e25SChristophe Leroy unsigned long 17614b439e25SChristophe Leroy arch_get_unmapped_area_topdown(struct file *filp, unsigned long addr, 17624b439e25SChristophe Leroy unsigned long len, unsigned long pgoff, 17634b439e25SChristophe Leroy unsigned long flags) 17644b439e25SChristophe Leroy { 17654b439e25SChristophe Leroy return generic_get_unmapped_area_topdown(filp, addr, len, pgoff, flags); 17664b439e25SChristophe Leroy } 17671da177e4SLinus Torvalds #endif 17681da177e4SLinus Torvalds 17691da177e4SLinus Torvalds unsigned long 17701da177e4SLinus Torvalds get_unmapped_area(struct file *file, unsigned long addr, unsigned long len, 17711da177e4SLinus Torvalds unsigned long pgoff, unsigned long flags) 17721da177e4SLinus Torvalds { 177306abdfb4SBenjamin Herrenschmidt unsigned long (*get_area)(struct file *, unsigned long, 177406abdfb4SBenjamin Herrenschmidt unsigned long, unsigned long, unsigned long); 177507ab67c8SLinus Torvalds 17769206de95SAl Viro unsigned long error = arch_mmap_check(addr, len, flags); 17779206de95SAl Viro if (error) 17789206de95SAl Viro return error; 17799206de95SAl Viro 17809206de95SAl Viro /* Careful about overflows.. */ 17819206de95SAl Viro if (len > TASK_SIZE) 17829206de95SAl Viro return -ENOMEM; 17839206de95SAl Viro 178407ab67c8SLinus Torvalds get_area = current->mm->get_unmapped_area; 1785c01d5b30SHugh Dickins if (file) { 1786c01d5b30SHugh Dickins if (file->f_op->get_unmapped_area) 178707ab67c8SLinus Torvalds get_area = file->f_op->get_unmapped_area; 1788c01d5b30SHugh Dickins } else if (flags & MAP_SHARED) { 1789c01d5b30SHugh Dickins /* 1790c01d5b30SHugh Dickins * mmap_region() will call shmem_zero_setup() to create a file, 1791c01d5b30SHugh Dickins * so use shmem's get_unmapped_area in case it can be huge. 179245e55300SPeter Collingbourne * do_mmap() will clear pgoff, so match alignment. 1793c01d5b30SHugh Dickins */ 1794c01d5b30SHugh Dickins pgoff = 0; 1795c01d5b30SHugh Dickins get_area = shmem_get_unmapped_area; 1796c01d5b30SHugh Dickins } 1797c01d5b30SHugh Dickins 179807ab67c8SLinus Torvalds addr = get_area(file, addr, len, pgoff, flags); 179907ab67c8SLinus Torvalds if (IS_ERR_VALUE(addr)) 180007ab67c8SLinus Torvalds return addr; 180107ab67c8SLinus Torvalds 18021da177e4SLinus Torvalds if (addr > TASK_SIZE - len) 18031da177e4SLinus Torvalds return -ENOMEM; 1804de1741a1SAlexander Kuleshov if (offset_in_page(addr)) 18051da177e4SLinus Torvalds return -EINVAL; 180606abdfb4SBenjamin Herrenschmidt 18079ac4ed4bSAl Viro error = security_mmap_addr(addr); 18089ac4ed4bSAl Viro return error ? error : addr; 18091da177e4SLinus Torvalds } 18101da177e4SLinus Torvalds 18111da177e4SLinus Torvalds EXPORT_SYMBOL(get_unmapped_area); 18121da177e4SLinus Torvalds 1813be8432e7SLiam R. Howlett /** 1814abdba2ddSLiam R. Howlett * find_vma_intersection() - Look up the first VMA which intersects the interval 1815abdba2ddSLiam R. Howlett * @mm: The process address space. 1816abdba2ddSLiam R. Howlett * @start_addr: The inclusive start user address. 1817abdba2ddSLiam R. Howlett * @end_addr: The exclusive end user address. 1818abdba2ddSLiam R. Howlett * 1819abdba2ddSLiam R. Howlett * Returns: The first VMA within the provided range, %NULL otherwise. Assumes 1820abdba2ddSLiam R. Howlett * start_addr < end_addr. 1821abdba2ddSLiam R. Howlett */ 1822abdba2ddSLiam R. Howlett struct vm_area_struct *find_vma_intersection(struct mm_struct *mm, 1823abdba2ddSLiam R. Howlett unsigned long start_addr, 1824abdba2ddSLiam R. Howlett unsigned long end_addr) 1825abdba2ddSLiam R. Howlett { 1826abdba2ddSLiam R. Howlett unsigned long index = start_addr; 1827abdba2ddSLiam R. Howlett 1828abdba2ddSLiam R. Howlett mmap_assert_locked(mm); 18297964cf8cSLiam R. Howlett return mt_find(&mm->mm_mt, &index, end_addr - 1); 1830abdba2ddSLiam R. Howlett } 1831abdba2ddSLiam R. Howlett EXPORT_SYMBOL(find_vma_intersection); 1832abdba2ddSLiam R. Howlett 1833abdba2ddSLiam R. Howlett /** 1834be8432e7SLiam R. Howlett * find_vma() - Find the VMA for a given address, or the next VMA. 1835be8432e7SLiam R. Howlett * @mm: The mm_struct to check 1836be8432e7SLiam R. Howlett * @addr: The address 1837be8432e7SLiam R. Howlett * 1838be8432e7SLiam R. Howlett * Returns: The VMA associated with addr, or the next VMA. 1839be8432e7SLiam R. Howlett * May return %NULL in the case of no VMA at addr or above. 1840be8432e7SLiam R. Howlett */ 18411da177e4SLinus Torvalds struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr) 18421da177e4SLinus Torvalds { 1843be8432e7SLiam R. Howlett unsigned long index = addr; 18441da177e4SLinus Torvalds 18455b78ed24SLuigi Rizzo mmap_assert_locked(mm); 18467964cf8cSLiam R. Howlett return mt_find(&mm->mm_mt, &index, ULONG_MAX); 18471da177e4SLinus Torvalds } 18481da177e4SLinus Torvalds EXPORT_SYMBOL(find_vma); 18491da177e4SLinus Torvalds 18507fdbd37dSLiam R. Howlett /** 18517fdbd37dSLiam R. Howlett * find_vma_prev() - Find the VMA for a given address, or the next vma and 18527fdbd37dSLiam R. Howlett * set %pprev to the previous VMA, if any. 18537fdbd37dSLiam R. Howlett * @mm: The mm_struct to check 18547fdbd37dSLiam R. Howlett * @addr: The address 18557fdbd37dSLiam R. Howlett * @pprev: The pointer to set to the previous VMA 18567fdbd37dSLiam R. Howlett * 18577fdbd37dSLiam R. Howlett * Note that RCU lock is missing here since the external mmap_lock() is used 18587fdbd37dSLiam R. Howlett * instead. 18597fdbd37dSLiam R. Howlett * 18607fdbd37dSLiam R. Howlett * Returns: The VMA associated with @addr, or the next vma. 18617fdbd37dSLiam R. Howlett * May return %NULL in the case of no vma at addr or above. 18626bd4837dSKOSAKI Motohiro */ 18631da177e4SLinus Torvalds struct vm_area_struct * 18641da177e4SLinus Torvalds find_vma_prev(struct mm_struct *mm, unsigned long addr, 18651da177e4SLinus Torvalds struct vm_area_struct **pprev) 18661da177e4SLinus Torvalds { 18676bd4837dSKOSAKI Motohiro struct vm_area_struct *vma; 18687fdbd37dSLiam R. Howlett MA_STATE(mas, &mm->mm_mt, addr, addr); 18691da177e4SLinus Torvalds 18707fdbd37dSLiam R. Howlett vma = mas_walk(&mas); 18717fdbd37dSLiam R. Howlett *pprev = mas_prev(&mas, 0); 18727fdbd37dSLiam R. Howlett if (!vma) 18737fdbd37dSLiam R. Howlett vma = mas_next(&mas, ULONG_MAX); 18746bd4837dSKOSAKI Motohiro return vma; 18751da177e4SLinus Torvalds } 18761da177e4SLinus Torvalds 18771da177e4SLinus Torvalds /* 18781da177e4SLinus Torvalds * Verify that the stack growth is acceptable and 18791da177e4SLinus Torvalds * update accounting. This is shared with both the 18801da177e4SLinus Torvalds * grow-up and grow-down cases. 18811da177e4SLinus Torvalds */ 18821be7107fSHugh Dickins static int acct_stack_growth(struct vm_area_struct *vma, 18831be7107fSHugh Dickins unsigned long size, unsigned long grow) 18841da177e4SLinus Torvalds { 18851da177e4SLinus Torvalds struct mm_struct *mm = vma->vm_mm; 18861be7107fSHugh Dickins unsigned long new_start; 18871da177e4SLinus Torvalds 18881da177e4SLinus Torvalds /* address space limit tests */ 188984638335SKonstantin Khlebnikov if (!may_expand_vm(mm, vma->vm_flags, grow)) 18901da177e4SLinus Torvalds return -ENOMEM; 18911da177e4SLinus Torvalds 18921da177e4SLinus Torvalds /* Stack limit test */ 189324c79d8eSKrzysztof Opasiak if (size > rlimit(RLIMIT_STACK)) 18941da177e4SLinus Torvalds return -ENOMEM; 18951da177e4SLinus Torvalds 18961da177e4SLinus Torvalds /* mlock limit tests */ 1897c5d8a364SMiaohe Lin if (mlock_future_check(mm, vma->vm_flags, grow << PAGE_SHIFT)) 18981da177e4SLinus Torvalds return -ENOMEM; 18991da177e4SLinus Torvalds 19000d59a01bSAdam Litke /* Check to ensure the stack will not grow into a hugetlb-only region */ 19010d59a01bSAdam Litke new_start = (vma->vm_flags & VM_GROWSUP) ? vma->vm_start : 19020d59a01bSAdam Litke vma->vm_end - size; 19030d59a01bSAdam Litke if (is_hugepage_only_range(vma->vm_mm, new_start, size)) 19040d59a01bSAdam Litke return -EFAULT; 19050d59a01bSAdam Litke 19061da177e4SLinus Torvalds /* 19071da177e4SLinus Torvalds * Overcommit.. This must be the final test, as it will 19081da177e4SLinus Torvalds * update security statistics. 19091da177e4SLinus Torvalds */ 191005fa199dSHugh Dickins if (security_vm_enough_memory_mm(mm, grow)) 19111da177e4SLinus Torvalds return -ENOMEM; 19121da177e4SLinus Torvalds 19131da177e4SLinus Torvalds return 0; 19141da177e4SLinus Torvalds } 19151da177e4SLinus Torvalds 191646dea3d0SHugh Dickins #if defined(CONFIG_STACK_GROWSUP) || defined(CONFIG_IA64) 19171da177e4SLinus Torvalds /* 191846dea3d0SHugh Dickins * PA-RISC uses this for its stack; IA64 for its Register Backing Store. 191946dea3d0SHugh Dickins * vma is the last one with address > vma->vm_end. Have to extend vma. 19201da177e4SLinus Torvalds */ 192146dea3d0SHugh Dickins int expand_upwards(struct vm_area_struct *vma, unsigned long address) 19221da177e4SLinus Torvalds { 192309357814SOleg Nesterov struct mm_struct *mm = vma->vm_mm; 19241be7107fSHugh Dickins struct vm_area_struct *next; 19251be7107fSHugh Dickins unsigned long gap_addr; 192612352d3cSKonstantin Khlebnikov int error = 0; 1927d4af56c5SLiam R. Howlett MA_STATE(mas, &mm->mm_mt, 0, 0); 19281da177e4SLinus Torvalds 19291da177e4SLinus Torvalds if (!(vma->vm_flags & VM_GROWSUP)) 19301da177e4SLinus Torvalds return -EFAULT; 19311da177e4SLinus Torvalds 1932bd726c90SHelge Deller /* Guard against exceeding limits of the address space. */ 19331be7107fSHugh Dickins address &= PAGE_MASK; 193437511fb5SHelge Deller if (address >= (TASK_SIZE & PAGE_MASK)) 193512352d3cSKonstantin Khlebnikov return -ENOMEM; 1936bd726c90SHelge Deller address += PAGE_SIZE; 193712352d3cSKonstantin Khlebnikov 19381be7107fSHugh Dickins /* Enforce stack_guard_gap */ 19391be7107fSHugh Dickins gap_addr = address + stack_guard_gap; 1940bd726c90SHelge Deller 1941bd726c90SHelge Deller /* Guard against overflow */ 1942bd726c90SHelge Deller if (gap_addr < address || gap_addr > TASK_SIZE) 1943bd726c90SHelge Deller gap_addr = TASK_SIZE; 1944bd726c90SHelge Deller 1945763ecb03SLiam R. Howlett next = find_vma_intersection(mm, vma->vm_end, gap_addr); 1946763ecb03SLiam R. Howlett if (next && vma_is_accessible(next)) { 19471be7107fSHugh Dickins if (!(next->vm_flags & VM_GROWSUP)) 19481be7107fSHugh Dickins return -ENOMEM; 19491be7107fSHugh Dickins /* Check that both stack segments have the same anon_vma? */ 19501be7107fSHugh Dickins } 19511be7107fSHugh Dickins 1952c5d5546eSVernon Yang if (mas_preallocate(&mas, GFP_KERNEL)) 19531da177e4SLinus Torvalds return -ENOMEM; 19541da177e4SLinus Torvalds 1955d4af56c5SLiam R. Howlett /* We must make sure the anon_vma is allocated. */ 1956d4af56c5SLiam R. Howlett if (unlikely(anon_vma_prepare(vma))) { 1957d4af56c5SLiam R. Howlett mas_destroy(&mas); 1958d4af56c5SLiam R. Howlett return -ENOMEM; 1959d4af56c5SLiam R. Howlett } 1960d4af56c5SLiam R. Howlett 19611da177e4SLinus Torvalds /* 19621da177e4SLinus Torvalds * vma->vm_start/vm_end cannot change under us because the caller 1963c1e8d7c6SMichel Lespinasse * is required to hold the mmap_lock in read mode. We need the 19641da177e4SLinus Torvalds * anon_vma lock to serialize against concurrent expand_stacks. 19651da177e4SLinus Torvalds */ 196612352d3cSKonstantin Khlebnikov anon_vma_lock_write(vma->anon_vma); 19671da177e4SLinus Torvalds 19681da177e4SLinus Torvalds /* Somebody else might have raced and expanded it already */ 19691da177e4SLinus Torvalds if (address > vma->vm_end) { 19701da177e4SLinus Torvalds unsigned long size, grow; 19711da177e4SLinus Torvalds 19721da177e4SLinus Torvalds size = address - vma->vm_start; 19731da177e4SLinus Torvalds grow = (address - vma->vm_end) >> PAGE_SHIFT; 19741da177e4SLinus Torvalds 197542c36f63SHugh Dickins error = -ENOMEM; 197642c36f63SHugh Dickins if (vma->vm_pgoff + (size >> PAGE_SHIFT) >= vma->vm_pgoff) { 19771da177e4SLinus Torvalds error = acct_stack_growth(vma, size, grow); 19783af9e859SEric B Munson if (!error) { 19794128997bSMichel Lespinasse /* 1980524e00b3SLiam R. Howlett * We only hold a shared mmap_lock lock here, so 1981524e00b3SLiam R. Howlett * we need to protect against concurrent vma 1982524e00b3SLiam R. Howlett * expansions. anon_vma_lock_write() doesn't 1983524e00b3SLiam R. Howlett * help here, as we don't guarantee that all 1984524e00b3SLiam R. Howlett * growable vmas in a mm share the same root 1985524e00b3SLiam R. Howlett * anon vma. So, we reuse mm->page_table_lock 1986524e00b3SLiam R. Howlett * to guard against concurrent vma expansions. 19874128997bSMichel Lespinasse */ 198809357814SOleg Nesterov spin_lock(&mm->page_table_lock); 198987e8827bSOleg Nesterov if (vma->vm_flags & VM_LOCKED) 199009357814SOleg Nesterov mm->locked_vm += grow; 199184638335SKonstantin Khlebnikov vm_stat_account(mm, vma->vm_flags, grow); 1992bf181b9fSMichel Lespinasse anon_vma_interval_tree_pre_update_vma(vma); 19931da177e4SLinus Torvalds vma->vm_end = address; 1994d4af56c5SLiam R. Howlett /* Overwrite old entry in mtree. */ 1995fbcc3104SLiam R. Howlett mas_set_range(&mas, vma->vm_start, address - 1); 1996fbcc3104SLiam R. Howlett mas_store_prealloc(&mas, vma); 1997bf181b9fSMichel Lespinasse anon_vma_interval_tree_post_update_vma(vma); 199809357814SOleg Nesterov spin_unlock(&mm->page_table_lock); 19994128997bSMichel Lespinasse 20003af9e859SEric B Munson perf_event_mmap(vma); 20013af9e859SEric B Munson } 20021da177e4SLinus Torvalds } 200342c36f63SHugh Dickins } 200412352d3cSKonstantin Khlebnikov anon_vma_unlock_write(vma->anon_vma); 2005c791576cSYang Shi khugepaged_enter_vma(vma, vma->vm_flags); 2006d4af56c5SLiam R. Howlett mas_destroy(&mas); 20071da177e4SLinus Torvalds return error; 20081da177e4SLinus Torvalds } 200946dea3d0SHugh Dickins #endif /* CONFIG_STACK_GROWSUP || CONFIG_IA64 */ 201046dea3d0SHugh Dickins 20111da177e4SLinus Torvalds /* 20121da177e4SLinus Torvalds * vma is the first one with address < vma->vm_start. Have to extend vma. 20131da177e4SLinus Torvalds */ 2014524e00b3SLiam R. Howlett int expand_downwards(struct vm_area_struct *vma, unsigned long address) 20151da177e4SLinus Torvalds { 201609357814SOleg Nesterov struct mm_struct *mm = vma->vm_mm; 2017763ecb03SLiam R. Howlett MA_STATE(mas, &mm->mm_mt, vma->vm_start, vma->vm_start); 20181be7107fSHugh Dickins struct vm_area_struct *prev; 20190a1d5299SJann Horn int error = 0; 20201da177e4SLinus Torvalds 20218869477aSEric Paris address &= PAGE_MASK; 20220a1d5299SJann Horn if (address < mmap_min_addr) 20230a1d5299SJann Horn return -EPERM; 20248869477aSEric Paris 20251be7107fSHugh Dickins /* Enforce stack_guard_gap */ 2026763ecb03SLiam R. Howlett prev = mas_prev(&mas, 0); 20271be7107fSHugh Dickins /* Check that both stack segments have the same anon_vma? */ 202832e4e6d5SOleg Nesterov if (prev && !(prev->vm_flags & VM_GROWSDOWN) && 20293122e80eSAnshuman Khandual vma_is_accessible(prev)) { 203032e4e6d5SOleg Nesterov if (address - prev->vm_end < stack_guard_gap) 203132e4e6d5SOleg Nesterov return -ENOMEM; 20321be7107fSHugh Dickins } 20331be7107fSHugh Dickins 2034c5d5546eSVernon Yang if (mas_preallocate(&mas, GFP_KERNEL)) 203512352d3cSKonstantin Khlebnikov return -ENOMEM; 20361da177e4SLinus Torvalds 2037d4af56c5SLiam R. Howlett /* We must make sure the anon_vma is allocated. */ 2038d4af56c5SLiam R. Howlett if (unlikely(anon_vma_prepare(vma))) { 2039d4af56c5SLiam R. Howlett mas_destroy(&mas); 2040d4af56c5SLiam R. Howlett return -ENOMEM; 2041d4af56c5SLiam R. Howlett } 2042d4af56c5SLiam R. Howlett 20431da177e4SLinus Torvalds /* 20441da177e4SLinus Torvalds * vma->vm_start/vm_end cannot change under us because the caller 2045c1e8d7c6SMichel Lespinasse * is required to hold the mmap_lock in read mode. We need the 20461da177e4SLinus Torvalds * anon_vma lock to serialize against concurrent expand_stacks. 20471da177e4SLinus Torvalds */ 204812352d3cSKonstantin Khlebnikov anon_vma_lock_write(vma->anon_vma); 20491da177e4SLinus Torvalds 20501da177e4SLinus Torvalds /* Somebody else might have raced and expanded it already */ 20511da177e4SLinus Torvalds if (address < vma->vm_start) { 20521da177e4SLinus Torvalds unsigned long size, grow; 20531da177e4SLinus Torvalds 20541da177e4SLinus Torvalds size = vma->vm_end - address; 20551da177e4SLinus Torvalds grow = (vma->vm_start - address) >> PAGE_SHIFT; 20561da177e4SLinus Torvalds 2057a626ca6aSLinus Torvalds error = -ENOMEM; 2058a626ca6aSLinus Torvalds if (grow <= vma->vm_pgoff) { 20591da177e4SLinus Torvalds error = acct_stack_growth(vma, size, grow); 20601da177e4SLinus Torvalds if (!error) { 20614128997bSMichel Lespinasse /* 2062524e00b3SLiam R. Howlett * We only hold a shared mmap_lock lock here, so 2063524e00b3SLiam R. Howlett * we need to protect against concurrent vma 2064524e00b3SLiam R. Howlett * expansions. anon_vma_lock_write() doesn't 2065524e00b3SLiam R. Howlett * help here, as we don't guarantee that all 2066524e00b3SLiam R. Howlett * growable vmas in a mm share the same root 2067524e00b3SLiam R. Howlett * anon vma. So, we reuse mm->page_table_lock 2068524e00b3SLiam R. Howlett * to guard against concurrent vma expansions. 20694128997bSMichel Lespinasse */ 207009357814SOleg Nesterov spin_lock(&mm->page_table_lock); 207187e8827bSOleg Nesterov if (vma->vm_flags & VM_LOCKED) 207209357814SOleg Nesterov mm->locked_vm += grow; 207384638335SKonstantin Khlebnikov vm_stat_account(mm, vma->vm_flags, grow); 2074bf181b9fSMichel Lespinasse anon_vma_interval_tree_pre_update_vma(vma); 20751da177e4SLinus Torvalds vma->vm_start = address; 20761da177e4SLinus Torvalds vma->vm_pgoff -= grow; 2077d4af56c5SLiam R. Howlett /* Overwrite old entry in mtree. */ 2078fbcc3104SLiam R. Howlett mas_set_range(&mas, address, vma->vm_end - 1); 2079fbcc3104SLiam R. Howlett mas_store_prealloc(&mas, vma); 2080bf181b9fSMichel Lespinasse anon_vma_interval_tree_post_update_vma(vma); 208109357814SOleg Nesterov spin_unlock(&mm->page_table_lock); 20824128997bSMichel Lespinasse 20833af9e859SEric B Munson perf_event_mmap(vma); 20841da177e4SLinus Torvalds } 20851da177e4SLinus Torvalds } 2086a626ca6aSLinus Torvalds } 208712352d3cSKonstantin Khlebnikov anon_vma_unlock_write(vma->anon_vma); 2088c791576cSYang Shi khugepaged_enter_vma(vma, vma->vm_flags); 2089d4af56c5SLiam R. Howlett mas_destroy(&mas); 20901da177e4SLinus Torvalds return error; 20911da177e4SLinus Torvalds } 20921da177e4SLinus Torvalds 20931be7107fSHugh Dickins /* enforced gap between the expanding stack and other mappings. */ 20941be7107fSHugh Dickins unsigned long stack_guard_gap = 256UL<<PAGE_SHIFT; 20951be7107fSHugh Dickins 20961be7107fSHugh Dickins static int __init cmdline_parse_stack_guard_gap(char *p) 20971be7107fSHugh Dickins { 20981be7107fSHugh Dickins unsigned long val; 20991be7107fSHugh Dickins char *endptr; 21001be7107fSHugh Dickins 21011be7107fSHugh Dickins val = simple_strtoul(p, &endptr, 10); 21021be7107fSHugh Dickins if (!*endptr) 21031be7107fSHugh Dickins stack_guard_gap = val << PAGE_SHIFT; 21041be7107fSHugh Dickins 2105e6d09493SRandy Dunlap return 1; 21061be7107fSHugh Dickins } 21071be7107fSHugh Dickins __setup("stack_guard_gap=", cmdline_parse_stack_guard_gap); 21081be7107fSHugh Dickins 2109b6a2fea3SOllie Wild #ifdef CONFIG_STACK_GROWSUP 2110b6a2fea3SOllie Wild int expand_stack(struct vm_area_struct *vma, unsigned long address) 2111b6a2fea3SOllie Wild { 2112b6a2fea3SOllie Wild return expand_upwards(vma, address); 2113b6a2fea3SOllie Wild } 2114b6a2fea3SOllie Wild 2115b6a2fea3SOllie Wild struct vm_area_struct * 2116b6a2fea3SOllie Wild find_extend_vma(struct mm_struct *mm, unsigned long addr) 2117b6a2fea3SOllie Wild { 2118b6a2fea3SOllie Wild struct vm_area_struct *vma, *prev; 2119b6a2fea3SOllie Wild 2120b6a2fea3SOllie Wild addr &= PAGE_MASK; 2121b6a2fea3SOllie Wild vma = find_vma_prev(mm, addr, &prev); 2122b6a2fea3SOllie Wild if (vma && (vma->vm_start <= addr)) 2123b6a2fea3SOllie Wild return vma; 21244d45e75aSJann Horn if (!prev || expand_stack(prev, addr)) 2125b6a2fea3SOllie Wild return NULL; 2126cea10a19SMichel Lespinasse if (prev->vm_flags & VM_LOCKED) 2127fc05f566SKirill A. Shutemov populate_vma_page_range(prev, addr, prev->vm_end, NULL); 2128b6a2fea3SOllie Wild return prev; 2129b6a2fea3SOllie Wild } 2130b6a2fea3SOllie Wild #else 2131b6a2fea3SOllie Wild int expand_stack(struct vm_area_struct *vma, unsigned long address) 2132b6a2fea3SOllie Wild { 2133b6a2fea3SOllie Wild return expand_downwards(vma, address); 2134b6a2fea3SOllie Wild } 2135b6a2fea3SOllie Wild 21361da177e4SLinus Torvalds struct vm_area_struct * 21371da177e4SLinus Torvalds find_extend_vma(struct mm_struct *mm, unsigned long addr) 21381da177e4SLinus Torvalds { 21391da177e4SLinus Torvalds struct vm_area_struct *vma; 21401da177e4SLinus Torvalds unsigned long start; 21411da177e4SLinus Torvalds 21421da177e4SLinus Torvalds addr &= PAGE_MASK; 21431da177e4SLinus Torvalds vma = find_vma(mm, addr); 21441da177e4SLinus Torvalds if (!vma) 21451da177e4SLinus Torvalds return NULL; 21461da177e4SLinus Torvalds if (vma->vm_start <= addr) 21471da177e4SLinus Torvalds return vma; 21481da177e4SLinus Torvalds if (!(vma->vm_flags & VM_GROWSDOWN)) 21491da177e4SLinus Torvalds return NULL; 21501da177e4SLinus Torvalds start = vma->vm_start; 21511da177e4SLinus Torvalds if (expand_stack(vma, addr)) 21521da177e4SLinus Torvalds return NULL; 2153cea10a19SMichel Lespinasse if (vma->vm_flags & VM_LOCKED) 2154fc05f566SKirill A. Shutemov populate_vma_page_range(vma, addr, start, NULL); 21551da177e4SLinus Torvalds return vma; 21561da177e4SLinus Torvalds } 21571da177e4SLinus Torvalds #endif 21581da177e4SLinus Torvalds 2159e1d6d01aSJesse Barnes EXPORT_SYMBOL_GPL(find_extend_vma); 2160e1d6d01aSJesse Barnes 21612c0b3814SHugh Dickins /* 2162763ecb03SLiam R. Howlett * Ok - we have the memory areas we should free on a maple tree so release them, 2163763ecb03SLiam R. Howlett * and do the vma updates. 21641da177e4SLinus Torvalds * 21652c0b3814SHugh Dickins * Called with the mm semaphore held. 21661da177e4SLinus Torvalds */ 2167763ecb03SLiam R. Howlett static inline void remove_mt(struct mm_struct *mm, struct ma_state *mas) 21681da177e4SLinus Torvalds { 21694f74d2c8SLinus Torvalds unsigned long nr_accounted = 0; 2170763ecb03SLiam R. Howlett struct vm_area_struct *vma; 21714f74d2c8SLinus Torvalds 2172365e9c87SHugh Dickins /* Update high watermark before we lower total_vm */ 2173365e9c87SHugh Dickins update_hiwater_vm(mm); 2174763ecb03SLiam R. Howlett mas_for_each(mas, vma, ULONG_MAX) { 2175ab50b8edSHugh Dickins long nrpages = vma_pages(vma); 21761da177e4SLinus Torvalds 21774f74d2c8SLinus Torvalds if (vma->vm_flags & VM_ACCOUNT) 21784f74d2c8SLinus Torvalds nr_accounted += nrpages; 217984638335SKonstantin Khlebnikov vm_stat_account(mm, vma->vm_flags, -nrpages); 2180763ecb03SLiam R. Howlett remove_vma(vma); 2181763ecb03SLiam R. Howlett } 21824f74d2c8SLinus Torvalds vm_unacct_memory(nr_accounted); 21831da177e4SLinus Torvalds validate_mm(mm); 21841da177e4SLinus Torvalds } 21851da177e4SLinus Torvalds 21861da177e4SLinus Torvalds /* 21871da177e4SLinus Torvalds * Get rid of page table information in the indicated region. 21881da177e4SLinus Torvalds * 2189f10df686SPaolo 'Blaisorblade' Giarrusso * Called with the mm semaphore held. 21901da177e4SLinus Torvalds */ 2191763ecb03SLiam R. Howlett static void unmap_region(struct mm_struct *mm, struct maple_tree *mt, 2192e0da382cSHugh Dickins struct vm_area_struct *vma, struct vm_area_struct *prev, 2193763ecb03SLiam R. Howlett struct vm_area_struct *next, 2194e0da382cSHugh Dickins unsigned long start, unsigned long end) 21951da177e4SLinus Torvalds { 2196d16dfc55SPeter Zijlstra struct mmu_gather tlb; 21971da177e4SLinus Torvalds 21981da177e4SLinus Torvalds lru_add_drain(); 2199a72afd87SWill Deacon tlb_gather_mmu(&tlb, mm); 2200365e9c87SHugh Dickins update_hiwater_rss(mm); 2201763ecb03SLiam R. Howlett unmap_vmas(&tlb, mt, vma, start, end); 2202763ecb03SLiam R. Howlett free_pgtables(&tlb, mt, vma, prev ? prev->vm_end : FIRST_USER_ADDRESS, 22036ee8630eSHugh Dickins next ? next->vm_start : USER_PGTABLES_CEILING); 2204ae8eba8bSWill Deacon tlb_finish_mmu(&tlb); 22051da177e4SLinus Torvalds } 22061da177e4SLinus Torvalds 22071da177e4SLinus Torvalds /* 2208def5efe0SDavid Rientjes * __split_vma() bypasses sysctl_max_map_count checking. We use this where it 2209def5efe0SDavid Rientjes * has already been checked or doesn't make sense to fail. 22100fd5a9e2SLiam R. Howlett * VMA Iterator will point to the end VMA. 22111da177e4SLinus Torvalds */ 22129760ebffSLiam R. Howlett int __split_vma(struct vma_iterator *vmi, struct vm_area_struct *vma, 22131da177e4SLinus Torvalds unsigned long addr, int new_below) 22141da177e4SLinus Torvalds { 2215b2b3b886SLiam R. Howlett struct vma_prepare vp; 22161da177e4SLinus Torvalds struct vm_area_struct *new; 2217e3975891SChen Gang int err; 22189760ebffSLiam R. Howlett 22199760ebffSLiam R. Howlett validate_mm_mt(vma->vm_mm); 22201da177e4SLinus Torvalds 2221b2b3b886SLiam R. Howlett WARN_ON(vma->vm_start >= addr); 2222b2b3b886SLiam R. Howlett WARN_ON(vma->vm_end <= addr); 2223b2b3b886SLiam R. Howlett 2224dd3b614fSDmitry Safonov if (vma->vm_ops && vma->vm_ops->may_split) { 2225dd3b614fSDmitry Safonov err = vma->vm_ops->may_split(vma, addr); 222631383c68SDan Williams if (err) 222731383c68SDan Williams return err; 222831383c68SDan Williams } 22291da177e4SLinus Torvalds 22303928d4f5SLinus Torvalds new = vm_area_dup(vma); 22311da177e4SLinus Torvalds if (!new) 2232e3975891SChen Gang return -ENOMEM; 22331da177e4SLinus Torvalds 2234b2b3b886SLiam R. Howlett err = -ENOMEM; 2235b2b3b886SLiam R. Howlett if (vma_iter_prealloc(vmi)) 2236b2b3b886SLiam R. Howlett goto out_free_vma; 2237b2b3b886SLiam R. Howlett 2238b2b3b886SLiam R. Howlett if (new_below) { 22391da177e4SLinus Torvalds new->vm_end = addr; 2240b2b3b886SLiam R. Howlett } else { 22411da177e4SLinus Torvalds new->vm_start = addr; 22421da177e4SLinus Torvalds new->vm_pgoff += ((addr - vma->vm_start) >> PAGE_SHIFT); 22431da177e4SLinus Torvalds } 22441da177e4SLinus Torvalds 2245ef0855d3SOleg Nesterov err = vma_dup_policy(vma, new); 2246ef0855d3SOleg Nesterov if (err) 2247b2b3b886SLiam R. Howlett goto out_free_vmi; 22481da177e4SLinus Torvalds 2249c4ea95d7SDaniel Forrest err = anon_vma_clone(new, vma); 2250c4ea95d7SDaniel Forrest if (err) 22515beb4930SRik van Riel goto out_free_mpol; 22525beb4930SRik van Riel 2253e9714acfSKonstantin Khlebnikov if (new->vm_file) 22541da177e4SLinus Torvalds get_file(new->vm_file); 22551da177e4SLinus Torvalds 22561da177e4SLinus Torvalds if (new->vm_ops && new->vm_ops->open) 22571da177e4SLinus Torvalds new->vm_ops->open(new); 22581da177e4SLinus Torvalds 2259b2b3b886SLiam R. Howlett vma_adjust_trans_huge(vma, vma->vm_start, addr, 0); 2260b2b3b886SLiam R. Howlett init_vma_prep(&vp, vma); 2261b2b3b886SLiam R. Howlett vp.insert = new; 2262b2b3b886SLiam R. Howlett vma_prepare(&vp); 22631da177e4SLinus Torvalds 2264b2b3b886SLiam R. Howlett if (new_below) { 2265b2b3b886SLiam R. Howlett vma->vm_start = addr; 2266b2b3b886SLiam R. Howlett vma->vm_pgoff += (addr - new->vm_start) >> PAGE_SHIFT; 2267b2b3b886SLiam R. Howlett } else { 2268b2b3b886SLiam R. Howlett vma->vm_end = addr; 22699760ebffSLiam R. Howlett } 22705beb4930SRik van Riel 2271b2b3b886SLiam R. Howlett /* vma_complete stores the new vma */ 2272b2b3b886SLiam R. Howlett vma_complete(&vp, vmi, vma->vm_mm); 2273b2b3b886SLiam R. Howlett 2274b2b3b886SLiam R. Howlett /* Success. */ 2275b2b3b886SLiam R. Howlett if (new_below) 2276b2b3b886SLiam R. Howlett vma_next(vmi); 2277b2b3b886SLiam R. Howlett validate_mm_mt(vma->vm_mm); 2278b2b3b886SLiam R. Howlett return 0; 2279b2b3b886SLiam R. Howlett 22805beb4930SRik van Riel out_free_mpol: 2281ef0855d3SOleg Nesterov mpol_put(vma_policy(new)); 2282b2b3b886SLiam R. Howlett out_free_vmi: 2283b2b3b886SLiam R. Howlett vma_iter_free(vmi); 22845beb4930SRik van Riel out_free_vma: 22853928d4f5SLinus Torvalds vm_area_free(new); 22869760ebffSLiam R. Howlett validate_mm_mt(vma->vm_mm); 22875beb4930SRik van Riel return err; 22881da177e4SLinus Torvalds } 22891da177e4SLinus Torvalds 2290659ace58SKOSAKI Motohiro /* 2291659ace58SKOSAKI Motohiro * Split a vma into two pieces at address 'addr', a new vma is allocated 2292659ace58SKOSAKI Motohiro * either for the first part or the tail. 2293659ace58SKOSAKI Motohiro */ 22949760ebffSLiam R. Howlett int split_vma(struct vma_iterator *vmi, struct vm_area_struct *vma, 2295659ace58SKOSAKI Motohiro unsigned long addr, int new_below) 2296659ace58SKOSAKI Motohiro { 22979760ebffSLiam R. Howlett if (vma->vm_mm->map_count >= sysctl_max_map_count) 2298659ace58SKOSAKI Motohiro return -ENOMEM; 2299659ace58SKOSAKI Motohiro 23009760ebffSLiam R. Howlett return __split_vma(vmi, vma, addr, new_below); 2301f2ebfe43SLiam R. Howlett } 2302f2ebfe43SLiam R. Howlett 2303763ecb03SLiam R. Howlett static inline int munmap_sidetree(struct vm_area_struct *vma, 2304763ecb03SLiam R. Howlett struct ma_state *mas_detach) 23051da177e4SLinus Torvalds { 2306763ecb03SLiam R. Howlett mas_set_range(mas_detach, vma->vm_start, vma->vm_end - 1); 2307763ecb03SLiam R. Howlett if (mas_store_gfp(mas_detach, vma, GFP_KERNEL)) 2308763ecb03SLiam R. Howlett return -ENOMEM; 230911f9a21aSLiam R. Howlett 2310763ecb03SLiam R. Howlett if (vma->vm_flags & VM_LOCKED) 2311763ecb03SLiam R. Howlett vma->vm_mm->locked_vm -= vma_pages(vma); 231211f9a21aSLiam R. Howlett 2313763ecb03SLiam R. Howlett return 0; 231411f9a21aSLiam R. Howlett } 231511f9a21aSLiam R. Howlett 231611f9a21aSLiam R. Howlett /* 2317183654ceSLiam R. Howlett * do_vmi_align_munmap() - munmap the aligned region from @start to @end. 2318183654ceSLiam R. Howlett * @vmi: The vma iterator 231911f9a21aSLiam R. Howlett * @vma: The starting vm_area_struct 232011f9a21aSLiam R. Howlett * @mm: The mm_struct 232111f9a21aSLiam R. Howlett * @start: The aligned start address to munmap. 232211f9a21aSLiam R. Howlett * @end: The aligned end address to munmap. 232311f9a21aSLiam R. Howlett * @uf: The userfaultfd list_head 23248651a137SLorenzo Stoakes * @downgrade: Set to true to attempt a write downgrade of the mmap_lock 232511f9a21aSLiam R. Howlett * 232611f9a21aSLiam R. Howlett * If @downgrade is true, check return code for potential release of the lock. 232711f9a21aSLiam R. Howlett */ 232811f9a21aSLiam R. Howlett static int 2329183654ceSLiam R. Howlett do_vmi_align_munmap(struct vma_iterator *vmi, struct vm_area_struct *vma, 233011f9a21aSLiam R. Howlett struct mm_struct *mm, unsigned long start, 233111f9a21aSLiam R. Howlett unsigned long end, struct list_head *uf, bool downgrade) 233211f9a21aSLiam R. Howlett { 2333763ecb03SLiam R. Howlett struct vm_area_struct *prev, *next = NULL; 2334763ecb03SLiam R. Howlett struct maple_tree mt_detach; 2335763ecb03SLiam R. Howlett int count = 0; 2336d4af56c5SLiam R. Howlett int error = -ENOMEM; 2337763ecb03SLiam R. Howlett MA_STATE(mas_detach, &mt_detach, 0, 0); 2338763ecb03SLiam R. Howlett mt_init_flags(&mt_detach, MT_FLAGS_LOCK_EXTERN); 2339763ecb03SLiam R. Howlett mt_set_external_lock(&mt_detach, &mm->mmap_lock); 2340524e00b3SLiam R. Howlett 23411da177e4SLinus Torvalds /* 23421da177e4SLinus Torvalds * If we need to split any vma, do it now to save pain later. 23431da177e4SLinus Torvalds * 23441da177e4SLinus Torvalds * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially 23451da177e4SLinus Torvalds * unmapped vm_area_struct will remain in use: so lower split_vma 23461da177e4SLinus Torvalds * places tmp vma above, and higher split_vma places tmp vma below. 23471da177e4SLinus Torvalds */ 2348763ecb03SLiam R. Howlett 2349763ecb03SLiam R. Howlett /* Does it split the first one? */ 2350146425a3SHugh Dickins if (start > vma->vm_start) { 2351659ace58SKOSAKI Motohiro 2352659ace58SKOSAKI Motohiro /* 2353659ace58SKOSAKI Motohiro * Make sure that map_count on return from munmap() will 2354659ace58SKOSAKI Motohiro * not exceed its limit; but let map_count go just above 2355659ace58SKOSAKI Motohiro * its limit temporarily, to help free resources as expected. 2356659ace58SKOSAKI Motohiro */ 2357659ace58SKOSAKI Motohiro if (end < vma->vm_end && mm->map_count >= sysctl_max_map_count) 2358d4af56c5SLiam R. Howlett goto map_count_exceeded; 2359659ace58SKOSAKI Motohiro 23609760ebffSLiam R. Howlett error = __split_vma(vmi, vma, start, 0); 23611da177e4SLinus Torvalds if (error) 2362763ecb03SLiam R. Howlett goto start_split_failed; 236311f9a21aSLiam R. Howlett 23640fd5a9e2SLiam R. Howlett vma = vma_iter_load(vmi); 23651da177e4SLinus Torvalds } 23661da177e4SLinus Torvalds 2367183654ceSLiam R. Howlett prev = vma_prev(vmi); 2368763ecb03SLiam R. Howlett if (unlikely((!prev))) 2369183654ceSLiam R. Howlett vma_iter_set(vmi, start); 237011f9a21aSLiam R. Howlett 2371763ecb03SLiam R. Howlett /* 2372763ecb03SLiam R. Howlett * Detach a range of VMAs from the mm. Using next as a temp variable as 2373763ecb03SLiam R. Howlett * it is always overwritten. 2374763ecb03SLiam R. Howlett */ 2375183654ceSLiam R. Howlett for_each_vma_range(*vmi, next, end) { 2376763ecb03SLiam R. Howlett /* Does it split the end? */ 2377763ecb03SLiam R. Howlett if (next->vm_end > end) { 23786b73cff2SLiam R. Howlett error = __split_vma(vmi, next, end, 0); 23791da177e4SLinus Torvalds if (error) 2380763ecb03SLiam R. Howlett goto end_split_failed; 238111f9a21aSLiam R. Howlett } 2382763ecb03SLiam R. Howlett error = munmap_sidetree(next, &mas_detach); 2383763ecb03SLiam R. Howlett if (error) 2384763ecb03SLiam R. Howlett goto munmap_sidetree_failed; 2385763ecb03SLiam R. Howlett 2386763ecb03SLiam R. Howlett count++; 2387763ecb03SLiam R. Howlett #ifdef CONFIG_DEBUG_VM_MAPLE_TREE 2388763ecb03SLiam R. Howlett BUG_ON(next->vm_start < start); 2389763ecb03SLiam R. Howlett BUG_ON(next->vm_start > end); 2390763ecb03SLiam R. Howlett #endif 2391763ecb03SLiam R. Howlett } 2392763ecb03SLiam R. Howlett 2393183654ceSLiam R. Howlett next = vma_next(vmi); 23942376dd7cSAndrea Arcangeli if (unlikely(uf)) { 23952376dd7cSAndrea Arcangeli /* 23962376dd7cSAndrea Arcangeli * If userfaultfd_unmap_prep returns an error the vmas 2397f0953a1bSIngo Molnar * will remain split, but userland will get a 23982376dd7cSAndrea Arcangeli * highly unexpected error anyway. This is no 23992376dd7cSAndrea Arcangeli * different than the case where the first of the two 24002376dd7cSAndrea Arcangeli * __split_vma fails, but we don't undo the first 24012376dd7cSAndrea Arcangeli * split, despite we could. This is unlikely enough 24022376dd7cSAndrea Arcangeli * failure that it's not worth optimizing it for. 24032376dd7cSAndrea Arcangeli */ 240469dbe6daSLiam R. Howlett error = userfaultfd_unmap_prep(mm, start, end, uf); 240511f9a21aSLiam R. Howlett 24062376dd7cSAndrea Arcangeli if (error) 2407d4af56c5SLiam R. Howlett goto userfaultfd_error; 24082376dd7cSAndrea Arcangeli } 24092376dd7cSAndrea Arcangeli 2410763ecb03SLiam R. Howlett #if defined(CONFIG_DEBUG_VM_MAPLE_TREE) 2411763ecb03SLiam R. Howlett /* Make sure no VMAs are about to be lost. */ 2412763ecb03SLiam R. Howlett { 2413763ecb03SLiam R. Howlett MA_STATE(test, &mt_detach, start, end - 1); 2414763ecb03SLiam R. Howlett struct vm_area_struct *vma_mas, *vma_test; 2415763ecb03SLiam R. Howlett int test_count = 0; 2416763ecb03SLiam R. Howlett 2417183654ceSLiam R. Howlett vma_iter_set(vmi, start); 2418763ecb03SLiam R. Howlett rcu_read_lock(); 2419763ecb03SLiam R. Howlett vma_test = mas_find(&test, end - 1); 2420183654ceSLiam R. Howlett for_each_vma_range(*vmi, vma_mas, end) { 2421763ecb03SLiam R. Howlett BUG_ON(vma_mas != vma_test); 2422763ecb03SLiam R. Howlett test_count++; 2423763ecb03SLiam R. Howlett vma_test = mas_next(&test, end - 1); 2424763ecb03SLiam R. Howlett } 2425763ecb03SLiam R. Howlett rcu_read_unlock(); 2426763ecb03SLiam R. Howlett BUG_ON(count != test_count); 2427763ecb03SLiam R. Howlett } 2428763ecb03SLiam R. Howlett #endif 24290378c0a0SLiam R. Howlett /* Point of no return */ 2430183654ceSLiam R. Howlett vma_iter_set(vmi, start); 2431183654ceSLiam R. Howlett if (vma_iter_clear_gfp(vmi, start, end, GFP_KERNEL)) 24320378c0a0SLiam R. Howlett return -ENOMEM; 24330378c0a0SLiam R. Howlett 2434763ecb03SLiam R. Howlett mm->map_count -= count; 243511f9a21aSLiam R. Howlett /* 243611f9a21aSLiam R. Howlett * Do not downgrade mmap_lock if we are next to VM_GROWSDOWN or 243711f9a21aSLiam R. Howlett * VM_GROWSUP VMA. Such VMAs can change their size under 243811f9a21aSLiam R. Howlett * down_read(mmap_lock) and collide with the VMA we are about to unmap. 243911f9a21aSLiam R. Howlett */ 244011f9a21aSLiam R. Howlett if (downgrade) { 2441763ecb03SLiam R. Howlett if (next && (next->vm_flags & VM_GROWSDOWN)) 244211f9a21aSLiam R. Howlett downgrade = false; 244311f9a21aSLiam R. Howlett else if (prev && (prev->vm_flags & VM_GROWSUP)) 244411f9a21aSLiam R. Howlett downgrade = false; 244511f9a21aSLiam R. Howlett else 2446d8ed45c5SMichel Lespinasse mmap_write_downgrade(mm); 244711f9a21aSLiam R. Howlett } 2448dd2283f2SYang Shi 2449763ecb03SLiam R. Howlett unmap_region(mm, &mt_detach, vma, prev, next, start, end); 2450763ecb03SLiam R. Howlett /* Statistics and freeing VMAs */ 2451763ecb03SLiam R. Howlett mas_set(&mas_detach, start); 2452763ecb03SLiam R. Howlett remove_mt(mm, &mas_detach); 2453763ecb03SLiam R. Howlett __mt_destroy(&mt_detach); 24541da177e4SLinus Torvalds 2455524e00b3SLiam R. Howlett 2456524e00b3SLiam R. Howlett validate_mm(mm); 2457dd2283f2SYang Shi return downgrade ? 1 : 0; 2458d4af56c5SLiam R. Howlett 2459d4af56c5SLiam R. Howlett userfaultfd_error: 2460763ecb03SLiam R. Howlett munmap_sidetree_failed: 2461763ecb03SLiam R. Howlett end_split_failed: 2462763ecb03SLiam R. Howlett __mt_destroy(&mt_detach); 2463763ecb03SLiam R. Howlett start_split_failed: 2464763ecb03SLiam R. Howlett map_count_exceeded: 2465d4af56c5SLiam R. Howlett return error; 24661da177e4SLinus Torvalds } 24671da177e4SLinus Torvalds 246811f9a21aSLiam R. Howlett /* 2469183654ceSLiam R. Howlett * do_vmi_munmap() - munmap a given range. 2470183654ceSLiam R. Howlett * @vmi: The vma iterator 247111f9a21aSLiam R. Howlett * @mm: The mm_struct 247211f9a21aSLiam R. Howlett * @start: The start address to munmap 247311f9a21aSLiam R. Howlett * @len: The length of the range to munmap 247411f9a21aSLiam R. Howlett * @uf: The userfaultfd list_head 247511f9a21aSLiam R. Howlett * @downgrade: set to true if the user wants to attempt to write_downgrade the 24768651a137SLorenzo Stoakes * mmap_lock 247711f9a21aSLiam R. Howlett * 247811f9a21aSLiam R. Howlett * This function takes a @mas that is either pointing to the previous VMA or set 247911f9a21aSLiam R. Howlett * to MA_START and sets it up to remove the mapping(s). The @len will be 248011f9a21aSLiam R. Howlett * aligned and any arch_unmap work will be preformed. 248111f9a21aSLiam R. Howlett * 248211f9a21aSLiam R. Howlett * Returns: -EINVAL on failure, 1 on success and unlock, 0 otherwise. 248311f9a21aSLiam R. Howlett */ 2484183654ceSLiam R. Howlett int do_vmi_munmap(struct vma_iterator *vmi, struct mm_struct *mm, 248511f9a21aSLiam R. Howlett unsigned long start, size_t len, struct list_head *uf, 248611f9a21aSLiam R. Howlett bool downgrade) 248711f9a21aSLiam R. Howlett { 248811f9a21aSLiam R. Howlett unsigned long end; 248911f9a21aSLiam R. Howlett struct vm_area_struct *vma; 249011f9a21aSLiam R. Howlett 249111f9a21aSLiam R. Howlett if ((offset_in_page(start)) || start > TASK_SIZE || len > TASK_SIZE-start) 249211f9a21aSLiam R. Howlett return -EINVAL; 249311f9a21aSLiam R. Howlett 249411f9a21aSLiam R. Howlett end = start + PAGE_ALIGN(len); 249511f9a21aSLiam R. Howlett if (end == start) 249611f9a21aSLiam R. Howlett return -EINVAL; 249711f9a21aSLiam R. Howlett 249811f9a21aSLiam R. Howlett /* arch_unmap() might do unmaps itself. */ 249911f9a21aSLiam R. Howlett arch_unmap(mm, start, end); 250011f9a21aSLiam R. Howlett 250111f9a21aSLiam R. Howlett /* Find the first overlapping VMA */ 2502183654ceSLiam R. Howlett vma = vma_find(vmi, end); 250311f9a21aSLiam R. Howlett if (!vma) 250411f9a21aSLiam R. Howlett return 0; 250511f9a21aSLiam R. Howlett 2506183654ceSLiam R. Howlett return do_vmi_align_munmap(vmi, vma, mm, start, end, uf, downgrade); 250711f9a21aSLiam R. Howlett } 250811f9a21aSLiam R. Howlett 250911f9a21aSLiam R. Howlett /* do_munmap() - Wrapper function for non-maple tree aware do_munmap() calls. 251011f9a21aSLiam R. Howlett * @mm: The mm_struct 251111f9a21aSLiam R. Howlett * @start: The start address to munmap 251211f9a21aSLiam R. Howlett * @len: The length to be munmapped. 251311f9a21aSLiam R. Howlett * @uf: The userfaultfd list_head 251411f9a21aSLiam R. Howlett */ 2515dd2283f2SYang Shi int do_munmap(struct mm_struct *mm, unsigned long start, size_t len, 2516dd2283f2SYang Shi struct list_head *uf) 2517dd2283f2SYang Shi { 2518183654ceSLiam R. Howlett VMA_ITERATOR(vmi, mm, start); 251911f9a21aSLiam R. Howlett 2520183654ceSLiam R. Howlett return do_vmi_munmap(&vmi, mm, start, len, uf, false); 2521dd2283f2SYang Shi } 2522dd2283f2SYang Shi 2523e99668a5SLiam R. Howlett unsigned long mmap_region(struct file *file, unsigned long addr, 2524e99668a5SLiam R. Howlett unsigned long len, vm_flags_t vm_flags, unsigned long pgoff, 2525e99668a5SLiam R. Howlett struct list_head *uf) 2526e99668a5SLiam R. Howlett { 2527e99668a5SLiam R. Howlett struct mm_struct *mm = current->mm; 2528e99668a5SLiam R. Howlett struct vm_area_struct *vma = NULL; 2529e99668a5SLiam R. Howlett struct vm_area_struct *next, *prev, *merge; 2530e99668a5SLiam R. Howlett pgoff_t pglen = len >> PAGE_SHIFT; 2531e99668a5SLiam R. Howlett unsigned long charged = 0; 2532e99668a5SLiam R. Howlett unsigned long end = addr + len; 2533e99668a5SLiam R. Howlett unsigned long merge_start = addr, merge_end = end; 2534e99668a5SLiam R. Howlett pgoff_t vm_pgoff; 2535e99668a5SLiam R. Howlett int error; 2536183654ceSLiam R. Howlett VMA_ITERATOR(vmi, mm, addr); 2537e99668a5SLiam R. Howlett 2538e99668a5SLiam R. Howlett /* Check against address space limit. */ 2539e99668a5SLiam R. Howlett if (!may_expand_vm(mm, vm_flags, len >> PAGE_SHIFT)) { 2540e99668a5SLiam R. Howlett unsigned long nr_pages; 2541e99668a5SLiam R. Howlett 2542e99668a5SLiam R. Howlett /* 2543e99668a5SLiam R. Howlett * MAP_FIXED may remove pages of mappings that intersects with 2544e99668a5SLiam R. Howlett * requested mapping. Account for the pages it would unmap. 2545e99668a5SLiam R. Howlett */ 2546e99668a5SLiam R. Howlett nr_pages = count_vma_pages_range(mm, addr, end); 2547e99668a5SLiam R. Howlett 2548e99668a5SLiam R. Howlett if (!may_expand_vm(mm, vm_flags, 2549e99668a5SLiam R. Howlett (len >> PAGE_SHIFT) - nr_pages)) 2550e99668a5SLiam R. Howlett return -ENOMEM; 2551e99668a5SLiam R. Howlett } 2552e99668a5SLiam R. Howlett 2553e99668a5SLiam R. Howlett /* Unmap any existing mapping in the area */ 2554183654ceSLiam R. Howlett if (do_vmi_munmap(&vmi, mm, addr, len, uf, false)) 2555e99668a5SLiam R. Howlett return -ENOMEM; 2556e99668a5SLiam R. Howlett 2557e99668a5SLiam R. Howlett /* 2558e99668a5SLiam R. Howlett * Private writable mapping: check memory availability 2559e99668a5SLiam R. Howlett */ 2560e99668a5SLiam R. Howlett if (accountable_mapping(file, vm_flags)) { 2561e99668a5SLiam R. Howlett charged = len >> PAGE_SHIFT; 2562e99668a5SLiam R. Howlett if (security_vm_enough_memory_mm(mm, charged)) 2563e99668a5SLiam R. Howlett return -ENOMEM; 2564e99668a5SLiam R. Howlett vm_flags |= VM_ACCOUNT; 2565e99668a5SLiam R. Howlett } 2566e99668a5SLiam R. Howlett 2567183654ceSLiam R. Howlett next = vma_next(&vmi); 2568183654ceSLiam R. Howlett prev = vma_prev(&vmi); 2569e99668a5SLiam R. Howlett if (vm_flags & VM_SPECIAL) 2570e99668a5SLiam R. Howlett goto cannot_expand; 2571e99668a5SLiam R. Howlett 2572e99668a5SLiam R. Howlett /* Attempt to expand an old mapping */ 2573e99668a5SLiam R. Howlett /* Check next */ 2574e99668a5SLiam R. Howlett if (next && next->vm_start == end && !vma_policy(next) && 2575e99668a5SLiam R. Howlett can_vma_merge_before(next, vm_flags, NULL, file, pgoff+pglen, 2576e99668a5SLiam R. Howlett NULL_VM_UFFD_CTX, NULL)) { 2577e99668a5SLiam R. Howlett merge_end = next->vm_end; 2578e99668a5SLiam R. Howlett vma = next; 2579e99668a5SLiam R. Howlett vm_pgoff = next->vm_pgoff - pglen; 2580e99668a5SLiam R. Howlett } 2581e99668a5SLiam R. Howlett 2582e99668a5SLiam R. Howlett /* Check prev */ 2583e99668a5SLiam R. Howlett if (prev && prev->vm_end == addr && !vma_policy(prev) && 2584e99668a5SLiam R. Howlett (vma ? can_vma_merge_after(prev, vm_flags, vma->anon_vma, file, 2585e99668a5SLiam R. Howlett pgoff, vma->vm_userfaultfd_ctx, NULL) : 2586e99668a5SLiam R. Howlett can_vma_merge_after(prev, vm_flags, NULL, file, pgoff, 2587e99668a5SLiam R. Howlett NULL_VM_UFFD_CTX, NULL))) { 2588e99668a5SLiam R. Howlett merge_start = prev->vm_start; 2589e99668a5SLiam R. Howlett vma = prev; 2590e99668a5SLiam R. Howlett vm_pgoff = prev->vm_pgoff; 2591e99668a5SLiam R. Howlett } 2592e99668a5SLiam R. Howlett 2593e99668a5SLiam R. Howlett 2594e99668a5SLiam R. Howlett /* Actually expand, if possible */ 2595e99668a5SLiam R. Howlett if (vma && 25963c441ab7SLiam R. Howlett !vma_expand(&vmi, vma, merge_start, merge_end, vm_pgoff, next)) { 2597e99668a5SLiam R. Howlett khugepaged_enter_vma(vma, vm_flags); 2598e99668a5SLiam R. Howlett goto expanded; 2599e99668a5SLiam R. Howlett } 2600e99668a5SLiam R. Howlett 2601e99668a5SLiam R. Howlett cannot_expand: 2602e99668a5SLiam R. Howlett /* 2603e99668a5SLiam R. Howlett * Determine the object being mapped and call the appropriate 2604e99668a5SLiam R. Howlett * specific mapper. the address has already been validated, but 2605e99668a5SLiam R. Howlett * not unmapped, but the maps are removed from the list. 2606e99668a5SLiam R. Howlett */ 2607e99668a5SLiam R. Howlett vma = vm_area_alloc(mm); 2608e99668a5SLiam R. Howlett if (!vma) { 2609e99668a5SLiam R. Howlett error = -ENOMEM; 2610e99668a5SLiam R. Howlett goto unacct_error; 2611e99668a5SLiam R. Howlett } 2612e99668a5SLiam R. Howlett 26130fd5a9e2SLiam R. Howlett vma_iter_set(&vmi, addr); 2614e99668a5SLiam R. Howlett vma->vm_start = addr; 2615e99668a5SLiam R. Howlett vma->vm_end = end; 2616e99668a5SLiam R. Howlett vma->vm_flags = vm_flags; 2617e99668a5SLiam R. Howlett vma->vm_page_prot = vm_get_page_prot(vm_flags); 2618e99668a5SLiam R. Howlett vma->vm_pgoff = pgoff; 2619e99668a5SLiam R. Howlett 2620e99668a5SLiam R. Howlett if (file) { 2621e99668a5SLiam R. Howlett if (vm_flags & VM_SHARED) { 2622e99668a5SLiam R. Howlett error = mapping_map_writable(file->f_mapping); 2623e99668a5SLiam R. Howlett if (error) 2624e99668a5SLiam R. Howlett goto free_vma; 2625e99668a5SLiam R. Howlett } 2626e99668a5SLiam R. Howlett 2627e99668a5SLiam R. Howlett vma->vm_file = get_file(file); 2628e99668a5SLiam R. Howlett error = call_mmap(file, vma); 2629e99668a5SLiam R. Howlett if (error) 2630e99668a5SLiam R. Howlett goto unmap_and_free_vma; 2631e99668a5SLiam R. Howlett 2632a57b7051SLiam Howlett /* 2633a57b7051SLiam Howlett * Expansion is handled above, merging is handled below. 2634a57b7051SLiam Howlett * Drivers should not alter the address of the VMA. 2635e99668a5SLiam R. Howlett */ 2636a57b7051SLiam Howlett error = -EINVAL; 2637cc8d1b09SLiam R. Howlett if (WARN_ON((addr != vma->vm_start))) 2638a57b7051SLiam Howlett goto close_and_free_vma; 2639e99668a5SLiam R. Howlett 2640cc8d1b09SLiam R. Howlett vma_iter_set(&vmi, addr); 2641e99668a5SLiam R. Howlett /* 2642e99668a5SLiam R. Howlett * If vm_flags changed after call_mmap(), we should try merge 2643e99668a5SLiam R. Howlett * vma again as we may succeed this time. 2644e99668a5SLiam R. Howlett */ 2645e99668a5SLiam R. Howlett if (unlikely(vm_flags != vma->vm_flags && prev)) { 26469760ebffSLiam R. Howlett merge = vma_merge(&vmi, mm, prev, vma->vm_start, 26479760ebffSLiam R. Howlett vma->vm_end, vma->vm_flags, NULL, 26489760ebffSLiam R. Howlett vma->vm_file, vma->vm_pgoff, NULL, 26499760ebffSLiam R. Howlett NULL_VM_UFFD_CTX, NULL); 2650e99668a5SLiam R. Howlett if (merge) { 2651e99668a5SLiam R. Howlett /* 2652e99668a5SLiam R. Howlett * ->mmap() can change vma->vm_file and fput 2653e99668a5SLiam R. Howlett * the original file. So fput the vma->vm_file 2654e99668a5SLiam R. Howlett * here or we would add an extra fput for file 2655e99668a5SLiam R. Howlett * and cause general protection fault 2656e99668a5SLiam R. Howlett * ultimately. 2657e99668a5SLiam R. Howlett */ 2658e99668a5SLiam R. Howlett fput(vma->vm_file); 2659e99668a5SLiam R. Howlett vm_area_free(vma); 2660e99668a5SLiam R. Howlett vma = merge; 2661e99668a5SLiam R. Howlett /* Update vm_flags to pick up the change. */ 2662e99668a5SLiam R. Howlett vm_flags = vma->vm_flags; 2663e99668a5SLiam R. Howlett goto unmap_writable; 2664e99668a5SLiam R. Howlett } 2665e99668a5SLiam R. Howlett } 2666e99668a5SLiam R. Howlett 2667e99668a5SLiam R. Howlett vm_flags = vma->vm_flags; 2668e99668a5SLiam R. Howlett } else if (vm_flags & VM_SHARED) { 2669e99668a5SLiam R. Howlett error = shmem_zero_setup(vma); 2670e99668a5SLiam R. Howlett if (error) 2671e99668a5SLiam R. Howlett goto free_vma; 2672e99668a5SLiam R. Howlett } else { 2673e99668a5SLiam R. Howlett vma_set_anonymous(vma); 2674e99668a5SLiam R. Howlett } 2675e99668a5SLiam R. Howlett 2676b507808eSJoey Gouly if (map_deny_write_exec(vma, vma->vm_flags)) { 2677b507808eSJoey Gouly error = -EACCES; 2678b507808eSJoey Gouly if (file) 2679b507808eSJoey Gouly goto close_and_free_vma; 2680b507808eSJoey Gouly else if (vma->vm_file) 2681b507808eSJoey Gouly goto unmap_and_free_vma; 2682b507808eSJoey Gouly else 2683b507808eSJoey Gouly goto free_vma; 2684b507808eSJoey Gouly } 2685b507808eSJoey Gouly 2686e99668a5SLiam R. Howlett /* Allow architectures to sanity-check the vm_flags */ 2687e99668a5SLiam R. Howlett error = -EINVAL; 2688cc8d1b09SLiam R. Howlett if (!arch_validate_flags(vma->vm_flags)) 2689deb0f656SCarlos Llamas goto close_and_free_vma; 2690e99668a5SLiam R. Howlett 2691e99668a5SLiam R. Howlett error = -ENOMEM; 2692cc8d1b09SLiam R. Howlett if (vma_iter_prealloc(&vmi)) 26935789151eSMike Kravetz goto close_and_free_vma; 2694e99668a5SLiam R. Howlett 2695e99668a5SLiam R. Howlett if (vma->vm_file) 2696e99668a5SLiam R. Howlett i_mmap_lock_write(vma->vm_file->f_mapping); 2697e99668a5SLiam R. Howlett 2698183654ceSLiam R. Howlett vma_iter_store(&vmi, vma); 2699e99668a5SLiam R. Howlett mm->map_count++; 2700e99668a5SLiam R. Howlett if (vma->vm_file) { 2701e99668a5SLiam R. Howlett if (vma->vm_flags & VM_SHARED) 2702e99668a5SLiam R. Howlett mapping_allow_writable(vma->vm_file->f_mapping); 2703e99668a5SLiam R. Howlett 2704e99668a5SLiam R. Howlett flush_dcache_mmap_lock(vma->vm_file->f_mapping); 2705e99668a5SLiam R. Howlett vma_interval_tree_insert(vma, &vma->vm_file->f_mapping->i_mmap); 2706e99668a5SLiam R. Howlett flush_dcache_mmap_unlock(vma->vm_file->f_mapping); 2707e99668a5SLiam R. Howlett i_mmap_unlock_write(vma->vm_file->f_mapping); 2708e99668a5SLiam R. Howlett } 2709e99668a5SLiam R. Howlett 2710e99668a5SLiam R. Howlett /* 2711e99668a5SLiam R. Howlett * vma_merge() calls khugepaged_enter_vma() either, the below 2712e99668a5SLiam R. Howlett * call covers the non-merge case. 2713e99668a5SLiam R. Howlett */ 2714e99668a5SLiam R. Howlett khugepaged_enter_vma(vma, vma->vm_flags); 2715e99668a5SLiam R. Howlett 2716e99668a5SLiam R. Howlett /* Once vma denies write, undo our temporary denial count */ 2717e99668a5SLiam R. Howlett unmap_writable: 2718e99668a5SLiam R. Howlett if (file && vm_flags & VM_SHARED) 2719e99668a5SLiam R. Howlett mapping_unmap_writable(file->f_mapping); 2720e99668a5SLiam R. Howlett file = vma->vm_file; 2721e99668a5SLiam R. Howlett expanded: 2722e99668a5SLiam R. Howlett perf_event_mmap(vma); 2723e99668a5SLiam R. Howlett 2724e99668a5SLiam R. Howlett vm_stat_account(mm, vm_flags, len >> PAGE_SHIFT); 2725e99668a5SLiam R. Howlett if (vm_flags & VM_LOCKED) { 2726e99668a5SLiam R. Howlett if ((vm_flags & VM_SPECIAL) || vma_is_dax(vma) || 2727e99668a5SLiam R. Howlett is_vm_hugetlb_page(vma) || 2728e99668a5SLiam R. Howlett vma == get_gate_vma(current->mm)) 2729e99668a5SLiam R. Howlett vma->vm_flags &= VM_LOCKED_CLEAR_MASK; 2730e99668a5SLiam R. Howlett else 2731e99668a5SLiam R. Howlett mm->locked_vm += (len >> PAGE_SHIFT); 2732e99668a5SLiam R. Howlett } 2733e99668a5SLiam R. Howlett 2734e99668a5SLiam R. Howlett if (file) 2735e99668a5SLiam R. Howlett uprobe_mmap(vma); 2736e99668a5SLiam R. Howlett 2737e99668a5SLiam R. Howlett /* 2738e99668a5SLiam R. Howlett * New (or expanded) vma always get soft dirty status. 2739e99668a5SLiam R. Howlett * Otherwise user-space soft-dirty page tracker won't 2740e99668a5SLiam R. Howlett * be able to distinguish situation when vma area unmapped, 2741e99668a5SLiam R. Howlett * then new mapped in-place (which must be aimed as 2742e99668a5SLiam R. Howlett * a completely new data area). 2743e99668a5SLiam R. Howlett */ 2744e99668a5SLiam R. Howlett vma->vm_flags |= VM_SOFTDIRTY; 2745e99668a5SLiam R. Howlett 2746e99668a5SLiam R. Howlett vma_set_page_prot(vma); 2747e99668a5SLiam R. Howlett 2748e99668a5SLiam R. Howlett validate_mm(mm); 2749e99668a5SLiam R. Howlett return addr; 2750e99668a5SLiam R. Howlett 2751deb0f656SCarlos Llamas close_and_free_vma: 2752cc8d1b09SLiam R. Howlett if (file && vma->vm_ops && vma->vm_ops->close) 2753deb0f656SCarlos Llamas vma->vm_ops->close(vma); 2754cc8d1b09SLiam R. Howlett 2755cc8d1b09SLiam R. Howlett if (file || vma->vm_file) { 2756e99668a5SLiam R. Howlett unmap_and_free_vma: 2757e99668a5SLiam R. Howlett fput(vma->vm_file); 2758e99668a5SLiam R. Howlett vma->vm_file = NULL; 2759e99668a5SLiam R. Howlett 2760e99668a5SLiam R. Howlett /* Undo any partial mapping done by a device driver. */ 2761cc8d1b09SLiam R. Howlett unmap_region(mm, &mm->mm_mt, vma, prev, next, vma->vm_start, 2762cc8d1b09SLiam R. Howlett vma->vm_end); 2763cc8d1b09SLiam R. Howlett } 2764cc674ab3SLi Zetao if (file && (vm_flags & VM_SHARED)) 2765e99668a5SLiam R. Howlett mapping_unmap_writable(file->f_mapping); 2766e99668a5SLiam R. Howlett free_vma: 2767e99668a5SLiam R. Howlett vm_area_free(vma); 2768e99668a5SLiam R. Howlett unacct_error: 2769e99668a5SLiam R. Howlett if (charged) 2770e99668a5SLiam R. Howlett vm_unacct_memory(charged); 2771e99668a5SLiam R. Howlett validate_mm(mm); 2772e99668a5SLiam R. Howlett return error; 2773e99668a5SLiam R. Howlett } 2774e99668a5SLiam R. Howlett 2775dd2283f2SYang Shi static int __vm_munmap(unsigned long start, size_t len, bool downgrade) 2776a46ef99dSLinus Torvalds { 2777a46ef99dSLinus Torvalds int ret; 2778bfce281cSAl Viro struct mm_struct *mm = current->mm; 2779897ab3e0SMike Rapoport LIST_HEAD(uf); 2780183654ceSLiam R. Howlett VMA_ITERATOR(vmi, mm, start); 2781a46ef99dSLinus Torvalds 2782d8ed45c5SMichel Lespinasse if (mmap_write_lock_killable(mm)) 2783ae798783SMichal Hocko return -EINTR; 2784ae798783SMichal Hocko 2785183654ceSLiam R. Howlett ret = do_vmi_munmap(&vmi, mm, start, len, &uf, downgrade); 2786dd2283f2SYang Shi /* 2787c1e8d7c6SMichel Lespinasse * Returning 1 indicates mmap_lock is downgraded. 2788dd2283f2SYang Shi * But 1 is not legal return value of vm_munmap() and munmap(), reset 2789dd2283f2SYang Shi * it to 0 before return. 2790dd2283f2SYang Shi */ 2791dd2283f2SYang Shi if (ret == 1) { 2792d8ed45c5SMichel Lespinasse mmap_read_unlock(mm); 2793dd2283f2SYang Shi ret = 0; 2794dd2283f2SYang Shi } else 2795d8ed45c5SMichel Lespinasse mmap_write_unlock(mm); 2796dd2283f2SYang Shi 2797897ab3e0SMike Rapoport userfaultfd_unmap_complete(mm, &uf); 2798a46ef99dSLinus Torvalds return ret; 2799a46ef99dSLinus Torvalds } 2800dd2283f2SYang Shi 2801dd2283f2SYang Shi int vm_munmap(unsigned long start, size_t len) 2802dd2283f2SYang Shi { 2803dd2283f2SYang Shi return __vm_munmap(start, len, false); 2804dd2283f2SYang Shi } 2805a46ef99dSLinus Torvalds EXPORT_SYMBOL(vm_munmap); 2806a46ef99dSLinus Torvalds 28076a6160a7SHeiko Carstens SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len) 28081da177e4SLinus Torvalds { 2809ce18d171SCatalin Marinas addr = untagged_addr(addr); 2810dd2283f2SYang Shi return __vm_munmap(addr, len, true); 28111da177e4SLinus Torvalds } 28121da177e4SLinus Torvalds 2813c8d78c18SKirill A. Shutemov 2814c8d78c18SKirill A. Shutemov /* 2815c8d78c18SKirill A. Shutemov * Emulation of deprecated remap_file_pages() syscall. 2816c8d78c18SKirill A. Shutemov */ 2817c8d78c18SKirill A. Shutemov SYSCALL_DEFINE5(remap_file_pages, unsigned long, start, unsigned long, size, 2818c8d78c18SKirill A. Shutemov unsigned long, prot, unsigned long, pgoff, unsigned long, flags) 2819c8d78c18SKirill A. Shutemov { 2820c8d78c18SKirill A. Shutemov 2821c8d78c18SKirill A. Shutemov struct mm_struct *mm = current->mm; 2822c8d78c18SKirill A. Shutemov struct vm_area_struct *vma; 2823c8d78c18SKirill A. Shutemov unsigned long populate = 0; 2824c8d78c18SKirill A. Shutemov unsigned long ret = -EINVAL; 2825c8d78c18SKirill A. Shutemov struct file *file; 2826c8d78c18SKirill A. Shutemov 2827ee65728eSMike Rapoport pr_warn_once("%s (%d) uses deprecated remap_file_pages() syscall. See Documentation/mm/remap_file_pages.rst.\n", 2828c8d78c18SKirill A. Shutemov current->comm, current->pid); 2829c8d78c18SKirill A. Shutemov 2830c8d78c18SKirill A. Shutemov if (prot) 2831c8d78c18SKirill A. Shutemov return ret; 2832c8d78c18SKirill A. Shutemov start = start & PAGE_MASK; 2833c8d78c18SKirill A. Shutemov size = size & PAGE_MASK; 2834c8d78c18SKirill A. Shutemov 2835c8d78c18SKirill A. Shutemov if (start + size <= start) 2836c8d78c18SKirill A. Shutemov return ret; 2837c8d78c18SKirill A. Shutemov 2838c8d78c18SKirill A. Shutemov /* Does pgoff wrap? */ 2839c8d78c18SKirill A. Shutemov if (pgoff + (size >> PAGE_SHIFT) < pgoff) 2840c8d78c18SKirill A. Shutemov return ret; 2841c8d78c18SKirill A. Shutemov 2842d8ed45c5SMichel Lespinasse if (mmap_write_lock_killable(mm)) 2843dc0ef0dfSMichal Hocko return -EINTR; 2844dc0ef0dfSMichal Hocko 28459b593cb2SLiam R. Howlett vma = vma_lookup(mm, start); 2846c8d78c18SKirill A. Shutemov 2847c8d78c18SKirill A. Shutemov if (!vma || !(vma->vm_flags & VM_SHARED)) 2848c8d78c18SKirill A. Shutemov goto out; 2849c8d78c18SKirill A. Shutemov 285048f7df32SKirill A. Shutemov if (start + size > vma->vm_end) { 2851763ecb03SLiam R. Howlett VMA_ITERATOR(vmi, mm, vma->vm_end); 2852763ecb03SLiam R. Howlett struct vm_area_struct *next, *prev = vma; 285348f7df32SKirill A. Shutemov 2854763ecb03SLiam R. Howlett for_each_vma_range(vmi, next, start + size) { 285548f7df32SKirill A. Shutemov /* hole between vmas ? */ 2856763ecb03SLiam R. Howlett if (next->vm_start != prev->vm_end) 285748f7df32SKirill A. Shutemov goto out; 285848f7df32SKirill A. Shutemov 285948f7df32SKirill A. Shutemov if (next->vm_file != vma->vm_file) 286048f7df32SKirill A. Shutemov goto out; 286148f7df32SKirill A. Shutemov 286248f7df32SKirill A. Shutemov if (next->vm_flags != vma->vm_flags) 286348f7df32SKirill A. Shutemov goto out; 286448f7df32SKirill A. Shutemov 28651db43d3fSLiam Howlett if (start + size <= next->vm_end) 28661db43d3fSLiam Howlett break; 28671db43d3fSLiam Howlett 2868763ecb03SLiam R. Howlett prev = next; 286948f7df32SKirill A. Shutemov } 287048f7df32SKirill A. Shutemov 287148f7df32SKirill A. Shutemov if (!next) 2872c8d78c18SKirill A. Shutemov goto out; 2873c8d78c18SKirill A. Shutemov } 2874c8d78c18SKirill A. Shutemov 2875c8d78c18SKirill A. Shutemov prot |= vma->vm_flags & VM_READ ? PROT_READ : 0; 2876c8d78c18SKirill A. Shutemov prot |= vma->vm_flags & VM_WRITE ? PROT_WRITE : 0; 2877c8d78c18SKirill A. Shutemov prot |= vma->vm_flags & VM_EXEC ? PROT_EXEC : 0; 2878c8d78c18SKirill A. Shutemov 2879c8d78c18SKirill A. Shutemov flags &= MAP_NONBLOCK; 2880c8d78c18SKirill A. Shutemov flags |= MAP_SHARED | MAP_FIXED | MAP_POPULATE; 2881fce000b1SLiam Howlett if (vma->vm_flags & VM_LOCKED) 2882c8d78c18SKirill A. Shutemov flags |= MAP_LOCKED; 288348f7df32SKirill A. Shutemov 2884c8d78c18SKirill A. Shutemov file = get_file(vma->vm_file); 288545e55300SPeter Collingbourne ret = do_mmap(vma->vm_file, start, size, 2886897ab3e0SMike Rapoport prot, flags, pgoff, &populate, NULL); 2887c8d78c18SKirill A. Shutemov fput(file); 2888c8d78c18SKirill A. Shutemov out: 2889d8ed45c5SMichel Lespinasse mmap_write_unlock(mm); 2890c8d78c18SKirill A. Shutemov if (populate) 2891c8d78c18SKirill A. Shutemov mm_populate(ret, populate); 2892c8d78c18SKirill A. Shutemov if (!IS_ERR_VALUE(ret)) 2893c8d78c18SKirill A. Shutemov ret = 0; 2894c8d78c18SKirill A. Shutemov return ret; 2895c8d78c18SKirill A. Shutemov } 2896c8d78c18SKirill A. Shutemov 28971da177e4SLinus Torvalds /* 289827b26701SLiam R. Howlett * do_vma_munmap() - Unmap a full or partial vma. 289927b26701SLiam R. Howlett * @vmi: The vma iterator pointing at the vma 290027b26701SLiam R. Howlett * @vma: The first vma to be munmapped 290127b26701SLiam R. Howlett * @start: the start of the address to unmap 290227b26701SLiam R. Howlett * @end: The end of the address to unmap 29032e7ce7d3SLiam R. Howlett * @uf: The userfaultfd list_head 290427b26701SLiam R. Howlett * @downgrade: Attempt to downgrade or not 29052e7ce7d3SLiam R. Howlett * 290627b26701SLiam R. Howlett * Returns: 0 on success and not downgraded, 1 on success and downgraded. 290727b26701SLiam R. Howlett * unmaps a VMA mapping when the vma iterator is already in position. 290827b26701SLiam R. Howlett * Does not handle alignment. 29091da177e4SLinus Torvalds */ 291027b26701SLiam R. Howlett int do_vma_munmap(struct vma_iterator *vmi, struct vm_area_struct *vma, 291127b26701SLiam R. Howlett unsigned long start, unsigned long end, 291227b26701SLiam R. Howlett struct list_head *uf, bool downgrade) 29132e7ce7d3SLiam R. Howlett { 29142e7ce7d3SLiam R. Howlett struct mm_struct *mm = vma->vm_mm; 29152e7ce7d3SLiam R. Howlett int ret; 29162e7ce7d3SLiam R. Howlett 291727b26701SLiam R. Howlett arch_unmap(mm, start, end); 291827b26701SLiam R. Howlett ret = do_vmi_align_munmap(vmi, vma, mm, start, end, uf, downgrade); 29192e7ce7d3SLiam R. Howlett validate_mm_mt(mm); 29202e7ce7d3SLiam R. Howlett return ret; 29212e7ce7d3SLiam R. Howlett } 29222e7ce7d3SLiam R. Howlett 29232e7ce7d3SLiam R. Howlett /* 29242e7ce7d3SLiam R. Howlett * do_brk_flags() - Increase the brk vma if the flags match. 292592fed820SLiam R. Howlett * @vmi: The vma iterator 29262e7ce7d3SLiam R. Howlett * @addr: The start address 29272e7ce7d3SLiam R. Howlett * @len: The length of the increase 29282e7ce7d3SLiam R. Howlett * @vma: The vma, 29292e7ce7d3SLiam R. Howlett * @flags: The VMA Flags 29302e7ce7d3SLiam R. Howlett * 29312e7ce7d3SLiam R. Howlett * Extend the brk VMA from addr to addr + len. If the VMA is NULL or the flags 29322e7ce7d3SLiam R. Howlett * do not match then create a new anonymous VMA. Eventually we may be able to 29332e7ce7d3SLiam R. Howlett * do some brk-specific accounting here. 29342e7ce7d3SLiam R. Howlett */ 293592fed820SLiam R. Howlett static int do_brk_flags(struct vma_iterator *vmi, struct vm_area_struct *vma, 2936763ecb03SLiam R. Howlett unsigned long addr, unsigned long len, unsigned long flags) 29371da177e4SLinus Torvalds { 29381da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 29392e7ce7d3SLiam R. Howlett 2940d4af56c5SLiam R. Howlett validate_mm_mt(mm); 29412e7ce7d3SLiam R. Howlett /* 29422e7ce7d3SLiam R. Howlett * Check against address space limits by the changed size 29432e7ce7d3SLiam R. Howlett * Note: This happens *after* clearing old mappings in some code paths. 29442e7ce7d3SLiam R. Howlett */ 294516e72e9bSDenys Vlasenko flags |= VM_DATA_DEFAULT_FLAGS | VM_ACCOUNT | mm->def_flags; 294684638335SKonstantin Khlebnikov if (!may_expand_vm(mm, flags, len >> PAGE_SHIFT)) 29471da177e4SLinus Torvalds return -ENOMEM; 29481da177e4SLinus Torvalds 29491da177e4SLinus Torvalds if (mm->map_count > sysctl_max_map_count) 29501da177e4SLinus Torvalds return -ENOMEM; 29511da177e4SLinus Torvalds 2952191c5424SAl Viro if (security_vm_enough_memory_mm(mm, len >> PAGE_SHIFT)) 29531da177e4SLinus Torvalds return -ENOMEM; 29541da177e4SLinus Torvalds 29551da177e4SLinus Torvalds /* 29562e7ce7d3SLiam R. Howlett * Expand the existing vma if possible; Note that singular lists do not 29572e7ce7d3SLiam R. Howlett * occur after forking, so the expand will only happen on new VMAs. 29581da177e4SLinus Torvalds */ 29596c28ca64SLiam Howlett if (vma && vma->vm_end == addr && !vma_policy(vma) && 29606c28ca64SLiam Howlett can_vma_merge_after(vma, flags, NULL, NULL, 29616c28ca64SLiam Howlett addr >> PAGE_SHIFT, NULL_VM_UFFD_CTX, NULL)) { 296292fed820SLiam R. Howlett if (vma_iter_prealloc(vmi)) 2963675eaca1SAlistair Popple goto unacct_fail; 296428c5609fSLiam Howlett 296528c5609fSLiam Howlett vma_adjust_trans_huge(vma, vma->vm_start, addr + len, 0); 29662e7ce7d3SLiam R. Howlett if (vma->anon_vma) { 29672e7ce7d3SLiam R. Howlett anon_vma_lock_write(vma->anon_vma); 29682e7ce7d3SLiam R. Howlett anon_vma_interval_tree_pre_update_vma(vma); 29691da177e4SLinus Torvalds } 29702e7ce7d3SLiam R. Howlett vma->vm_end = addr + len; 29712e7ce7d3SLiam R. Howlett vma->vm_flags |= VM_SOFTDIRTY; 297292fed820SLiam R. Howlett vma_iter_store(vmi, vma); 29732e7ce7d3SLiam R. Howlett 29742e7ce7d3SLiam R. Howlett if (vma->anon_vma) { 29752e7ce7d3SLiam R. Howlett anon_vma_interval_tree_post_update_vma(vma); 29762e7ce7d3SLiam R. Howlett anon_vma_unlock_write(vma->anon_vma); 29772e7ce7d3SLiam R. Howlett } 29782e7ce7d3SLiam R. Howlett khugepaged_enter_vma(vma, flags); 29792e7ce7d3SLiam R. Howlett goto out; 29802e7ce7d3SLiam R. Howlett } 29812e7ce7d3SLiam R. Howlett 29822e7ce7d3SLiam R. Howlett /* create a vma struct for an anonymous mapping */ 29832e7ce7d3SLiam R. Howlett vma = vm_area_alloc(mm); 29842e7ce7d3SLiam R. Howlett if (!vma) 2985675eaca1SAlistair Popple goto unacct_fail; 29861da177e4SLinus Torvalds 2987bfd40eafSKirill A. Shutemov vma_set_anonymous(vma); 29881da177e4SLinus Torvalds vma->vm_start = addr; 29891da177e4SLinus Torvalds vma->vm_end = addr + len; 29902e7ce7d3SLiam R. Howlett vma->vm_pgoff = addr >> PAGE_SHIFT; 29911da177e4SLinus Torvalds vma->vm_flags = flags; 29923ed75eb8SColy Li vma->vm_page_prot = vm_get_page_prot(flags); 299392fed820SLiam R. Howlett if (vma_iter_store_gfp(vmi, vma, GFP_KERNEL)) 29942e7ce7d3SLiam R. Howlett goto mas_store_fail; 2995d4af56c5SLiam R. Howlett 29962e7ce7d3SLiam R. Howlett mm->map_count++; 29971da177e4SLinus Torvalds out: 29983af9e859SEric B Munson perf_event_mmap(vma); 29991da177e4SLinus Torvalds mm->total_vm += len >> PAGE_SHIFT; 300084638335SKonstantin Khlebnikov mm->data_vm += len >> PAGE_SHIFT; 3001128557ffSMichel Lespinasse if (flags & VM_LOCKED) 3002ba470de4SRik van Riel mm->locked_vm += (len >> PAGE_SHIFT); 3003d9104d1cSCyrill Gorcunov vma->vm_flags |= VM_SOFTDIRTY; 3004763ecb03SLiam R. Howlett validate_mm(mm); 30055d22fc25SLinus Torvalds return 0; 3006d4af56c5SLiam R. Howlett 30072e7ce7d3SLiam R. Howlett mas_store_fail: 3008d4af56c5SLiam R. Howlett vm_area_free(vma); 3009675eaca1SAlistair Popple unacct_fail: 30102e7ce7d3SLiam R. Howlett vm_unacct_memory(len >> PAGE_SHIFT); 30112e7ce7d3SLiam R. Howlett return -ENOMEM; 30121da177e4SLinus Torvalds } 30131da177e4SLinus Torvalds 3014bb177a73SMichal Hocko int vm_brk_flags(unsigned long addr, unsigned long request, unsigned long flags) 3015e4eb1ff6SLinus Torvalds { 3016e4eb1ff6SLinus Torvalds struct mm_struct *mm = current->mm; 30172e7ce7d3SLiam R. Howlett struct vm_area_struct *vma = NULL; 3018bb177a73SMichal Hocko unsigned long len; 30195d22fc25SLinus Torvalds int ret; 3020128557ffSMichel Lespinasse bool populate; 3021897ab3e0SMike Rapoport LIST_HEAD(uf); 302292fed820SLiam R. Howlett VMA_ITERATOR(vmi, mm, addr); 3023e4eb1ff6SLinus Torvalds 3024bb177a73SMichal Hocko len = PAGE_ALIGN(request); 3025bb177a73SMichal Hocko if (len < request) 3026bb177a73SMichal Hocko return -ENOMEM; 3027bb177a73SMichal Hocko if (!len) 3028bb177a73SMichal Hocko return 0; 3029bb177a73SMichal Hocko 3030d8ed45c5SMichel Lespinasse if (mmap_write_lock_killable(mm)) 30312d6c9282SMichal Hocko return -EINTR; 30322d6c9282SMichal Hocko 30332e7ce7d3SLiam R. Howlett /* Until we need other flags, refuse anything except VM_EXEC. */ 30342e7ce7d3SLiam R. Howlett if ((flags & (~VM_EXEC)) != 0) 30352e7ce7d3SLiam R. Howlett return -EINVAL; 30362e7ce7d3SLiam R. Howlett 30372e7ce7d3SLiam R. Howlett ret = check_brk_limits(addr, len); 30382e7ce7d3SLiam R. Howlett if (ret) 30392e7ce7d3SLiam R. Howlett goto limits_failed; 30402e7ce7d3SLiam R. Howlett 3041183654ceSLiam R. Howlett ret = do_vmi_munmap(&vmi, mm, addr, len, &uf, 0); 30422e7ce7d3SLiam R. Howlett if (ret) 30432e7ce7d3SLiam R. Howlett goto munmap_failed; 30442e7ce7d3SLiam R. Howlett 304592fed820SLiam R. Howlett vma = vma_prev(&vmi); 304692fed820SLiam R. Howlett ret = do_brk_flags(&vmi, vma, addr, len, flags); 3047128557ffSMichel Lespinasse populate = ((mm->def_flags & VM_LOCKED) != 0); 3048d8ed45c5SMichel Lespinasse mmap_write_unlock(mm); 3049897ab3e0SMike Rapoport userfaultfd_unmap_complete(mm, &uf); 30505d22fc25SLinus Torvalds if (populate && !ret) 3051128557ffSMichel Lespinasse mm_populate(addr, len); 3052e4eb1ff6SLinus Torvalds return ret; 30532e7ce7d3SLiam R. Howlett 30542e7ce7d3SLiam R. Howlett munmap_failed: 30552e7ce7d3SLiam R. Howlett limits_failed: 30562e7ce7d3SLiam R. Howlett mmap_write_unlock(mm); 30572e7ce7d3SLiam R. Howlett return ret; 3058e4eb1ff6SLinus Torvalds } 305916e72e9bSDenys Vlasenko EXPORT_SYMBOL(vm_brk_flags); 306016e72e9bSDenys Vlasenko 306116e72e9bSDenys Vlasenko int vm_brk(unsigned long addr, unsigned long len) 306216e72e9bSDenys Vlasenko { 306316e72e9bSDenys Vlasenko return vm_brk_flags(addr, len, 0); 306416e72e9bSDenys Vlasenko } 3065e4eb1ff6SLinus Torvalds EXPORT_SYMBOL(vm_brk); 30661da177e4SLinus Torvalds 30671da177e4SLinus Torvalds /* Release all mmaps. */ 30681da177e4SLinus Torvalds void exit_mmap(struct mm_struct *mm) 30691da177e4SLinus Torvalds { 3070d16dfc55SPeter Zijlstra struct mmu_gather tlb; 3071ba470de4SRik van Riel struct vm_area_struct *vma; 30721da177e4SLinus Torvalds unsigned long nr_accounted = 0; 3073763ecb03SLiam R. Howlett MA_STATE(mas, &mm->mm_mt, 0, 0); 3074763ecb03SLiam R. Howlett int count = 0; 30751da177e4SLinus Torvalds 3076d6dd61c8SJeremy Fitzhardinge /* mm's last user has gone, and its about to be pulled down */ 3077cddb8a5cSAndrea Arcangeli mmu_notifier_release(mm); 3078d6dd61c8SJeremy Fitzhardinge 3079bf3980c8SSuren Baghdasaryan mmap_read_lock(mm); 30809480c53eSJeremy Fitzhardinge arch_exit_mmap(mm); 30819480c53eSJeremy Fitzhardinge 3082763ecb03SLiam R. Howlett vma = mas_find(&mas, ULONG_MAX); 308364591e86SSuren Baghdasaryan if (!vma) { 308464591e86SSuren Baghdasaryan /* Can happen if dup_mmap() received an OOM */ 3085bf3980c8SSuren Baghdasaryan mmap_read_unlock(mm); 30869480c53eSJeremy Fitzhardinge return; 308764591e86SSuren Baghdasaryan } 30889480c53eSJeremy Fitzhardinge 30891da177e4SLinus Torvalds lru_add_drain(); 30901da177e4SLinus Torvalds flush_cache_mm(mm); 3091d8b45053SWill Deacon tlb_gather_mmu_fullmm(&tlb, mm); 3092901608d9SOleg Nesterov /* update_hiwater_rss(mm) here? but nobody should be looking */ 3093763ecb03SLiam R. Howlett /* Use ULONG_MAX here to ensure all VMAs in the mm are unmapped */ 3094763ecb03SLiam R. Howlett unmap_vmas(&tlb, &mm->mm_mt, vma, 0, ULONG_MAX); 3095bf3980c8SSuren Baghdasaryan mmap_read_unlock(mm); 3096bf3980c8SSuren Baghdasaryan 3097bf3980c8SSuren Baghdasaryan /* 3098bf3980c8SSuren Baghdasaryan * Set MMF_OOM_SKIP to hide this task from the oom killer/reaper 3099b3541d91SSuren Baghdasaryan * because the memory has been already freed. 3100bf3980c8SSuren Baghdasaryan */ 3101bf3980c8SSuren Baghdasaryan set_bit(MMF_OOM_SKIP, &mm->flags); 3102bf3980c8SSuren Baghdasaryan mmap_write_lock(mm); 3103763ecb03SLiam R. Howlett free_pgtables(&tlb, &mm->mm_mt, vma, FIRST_USER_ADDRESS, 3104763ecb03SLiam R. Howlett USER_PGTABLES_CEILING); 3105ae8eba8bSWill Deacon tlb_finish_mmu(&tlb); 31061da177e4SLinus Torvalds 3107763ecb03SLiam R. Howlett /* 3108763ecb03SLiam R. Howlett * Walk the list again, actually closing and freeing it, with preemption 3109763ecb03SLiam R. Howlett * enabled, without holding any MM locks besides the unreachable 3110763ecb03SLiam R. Howlett * mmap_write_lock. 3111763ecb03SLiam R. Howlett */ 3112763ecb03SLiam R. Howlett do { 31134f74d2c8SLinus Torvalds if (vma->vm_flags & VM_ACCOUNT) 31144f74d2c8SLinus Torvalds nr_accounted += vma_pages(vma); 3115763ecb03SLiam R. Howlett remove_vma(vma); 3116763ecb03SLiam R. Howlett count++; 31170a3b3c25SPaul E. McKenney cond_resched(); 3118763ecb03SLiam R. Howlett } while ((vma = mas_find(&mas, ULONG_MAX)) != NULL); 3119763ecb03SLiam R. Howlett 3120763ecb03SLiam R. Howlett BUG_ON(count != mm->map_count); 3121d4af56c5SLiam R. Howlett 3122d4af56c5SLiam R. Howlett trace_exit_mmap(mm); 3123d4af56c5SLiam R. Howlett __mt_destroy(&mm->mm_mt); 312464591e86SSuren Baghdasaryan mmap_write_unlock(mm); 31254f74d2c8SLinus Torvalds vm_unacct_memory(nr_accounted); 31261da177e4SLinus Torvalds } 31271da177e4SLinus Torvalds 31281da177e4SLinus Torvalds /* Insert vm structure into process list sorted by address 31291da177e4SLinus Torvalds * and into the inode's i_mmap tree. If vm_file is non-NULL 3130c8c06efaSDavidlohr Bueso * then i_mmap_rwsem is taken here. 31311da177e4SLinus Torvalds */ 31321da177e4SLinus Torvalds int insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma) 31331da177e4SLinus Torvalds { 3134d4af56c5SLiam R. Howlett unsigned long charged = vma_pages(vma); 31351da177e4SLinus Torvalds 3136d4af56c5SLiam R. Howlett 3137d0601a50SLiam R. Howlett if (find_vma_intersection(mm, vma->vm_start, vma->vm_end)) 3138c9d13f5fSChen Gang return -ENOMEM; 3139d4af56c5SLiam R. Howlett 3140c9d13f5fSChen Gang if ((vma->vm_flags & VM_ACCOUNT) && 3141d4af56c5SLiam R. Howlett security_vm_enough_memory_mm(mm, charged)) 3142c9d13f5fSChen Gang return -ENOMEM; 3143c9d13f5fSChen Gang 31441da177e4SLinus Torvalds /* 31451da177e4SLinus Torvalds * The vm_pgoff of a purely anonymous vma should be irrelevant 31461da177e4SLinus Torvalds * until its first write fault, when page's anon_vma and index 31471da177e4SLinus Torvalds * are set. But now set the vm_pgoff it will almost certainly 31481da177e4SLinus Torvalds * end up with (unless mremap moves it elsewhere before that 31491da177e4SLinus Torvalds * first wfault), so /proc/pid/maps tells a consistent story. 31501da177e4SLinus Torvalds * 31511da177e4SLinus Torvalds * By setting it to reflect the virtual start address of the 31521da177e4SLinus Torvalds * vma, merges and splits can happen in a seamless way, just 31531da177e4SLinus Torvalds * using the existing file pgoff checks and manipulations. 31548332326eSLiao Pingfang * Similarly in do_mmap and in do_brk_flags. 31551da177e4SLinus Torvalds */ 31568a9cc3b5SOleg Nesterov if (vma_is_anonymous(vma)) { 31571da177e4SLinus Torvalds BUG_ON(vma->anon_vma); 31581da177e4SLinus Torvalds vma->vm_pgoff = vma->vm_start >> PAGE_SHIFT; 31591da177e4SLinus Torvalds } 31602b144498SSrikar Dronamraju 3161763ecb03SLiam R. Howlett if (vma_link(mm, vma)) { 3162d4af56c5SLiam R. Howlett vm_unacct_memory(charged); 3163d4af56c5SLiam R. Howlett return -ENOMEM; 3164d4af56c5SLiam R. Howlett } 3165d4af56c5SLiam R. Howlett 31661da177e4SLinus Torvalds return 0; 31671da177e4SLinus Torvalds } 31681da177e4SLinus Torvalds 31691da177e4SLinus Torvalds /* 31701da177e4SLinus Torvalds * Copy the vma structure to a new location in the same mm, 31711da177e4SLinus Torvalds * prior to moving page table entries, to effect an mremap move. 31721da177e4SLinus Torvalds */ 31731da177e4SLinus Torvalds struct vm_area_struct *copy_vma(struct vm_area_struct **vmap, 317438a76013SMichel Lespinasse unsigned long addr, unsigned long len, pgoff_t pgoff, 317538a76013SMichel Lespinasse bool *need_rmap_locks) 31761da177e4SLinus Torvalds { 31771da177e4SLinus Torvalds struct vm_area_struct *vma = *vmap; 31781da177e4SLinus Torvalds unsigned long vma_start = vma->vm_start; 31791da177e4SLinus Torvalds struct mm_struct *mm = vma->vm_mm; 31801da177e4SLinus Torvalds struct vm_area_struct *new_vma, *prev; 3181948f017bSAndrea Arcangeli bool faulted_in_anon_vma = true; 3182076f16bfSLiam R. Howlett VMA_ITERATOR(vmi, mm, addr); 31831da177e4SLinus Torvalds 3184d4af56c5SLiam R. Howlett validate_mm_mt(mm); 31851da177e4SLinus Torvalds /* 31861da177e4SLinus Torvalds * If anonymous vma has not yet been faulted, update new pgoff 31871da177e4SLinus Torvalds * to match new location, to increase its chance of merging. 31881da177e4SLinus Torvalds */ 3189ce75799bSOleg Nesterov if (unlikely(vma_is_anonymous(vma) && !vma->anon_vma)) { 31901da177e4SLinus Torvalds pgoff = addr >> PAGE_SHIFT; 3191948f017bSAndrea Arcangeli faulted_in_anon_vma = false; 3192948f017bSAndrea Arcangeli } 31931da177e4SLinus Torvalds 3194763ecb03SLiam R. Howlett new_vma = find_vma_prev(mm, addr, &prev); 3195763ecb03SLiam R. Howlett if (new_vma && new_vma->vm_start < addr + len) 31966597d783SHugh Dickins return NULL; /* should never get here */ 3197524e00b3SLiam R. Howlett 31989760ebffSLiam R. Howlett new_vma = vma_merge(&vmi, mm, prev, addr, addr + len, vma->vm_flags, 319919a809afSAndrea Arcangeli vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma), 32005c26f6acSSuren Baghdasaryan vma->vm_userfaultfd_ctx, anon_vma_name(vma)); 32011da177e4SLinus Torvalds if (new_vma) { 32021da177e4SLinus Torvalds /* 32031da177e4SLinus Torvalds * Source vma may have been merged into new_vma 32041da177e4SLinus Torvalds */ 3205948f017bSAndrea Arcangeli if (unlikely(vma_start >= new_vma->vm_start && 3206948f017bSAndrea Arcangeli vma_start < new_vma->vm_end)) { 3207948f017bSAndrea Arcangeli /* 3208948f017bSAndrea Arcangeli * The only way we can get a vma_merge with 3209948f017bSAndrea Arcangeli * self during an mremap is if the vma hasn't 3210948f017bSAndrea Arcangeli * been faulted in yet and we were allowed to 3211948f017bSAndrea Arcangeli * reset the dst vma->vm_pgoff to the 3212948f017bSAndrea Arcangeli * destination address of the mremap to allow 3213948f017bSAndrea Arcangeli * the merge to happen. mremap must change the 3214948f017bSAndrea Arcangeli * vm_pgoff linearity between src and dst vmas 3215948f017bSAndrea Arcangeli * (in turn preventing a vma_merge) to be 3216948f017bSAndrea Arcangeli * safe. It is only safe to keep the vm_pgoff 3217948f017bSAndrea Arcangeli * linear if there are no pages mapped yet. 3218948f017bSAndrea Arcangeli */ 321981d1b09cSSasha Levin VM_BUG_ON_VMA(faulted_in_anon_vma, new_vma); 322038a76013SMichel Lespinasse *vmap = vma = new_vma; 3221108d6642SMichel Lespinasse } 322238a76013SMichel Lespinasse *need_rmap_locks = (new_vma->vm_pgoff <= vma->vm_pgoff); 32231da177e4SLinus Torvalds } else { 32243928d4f5SLinus Torvalds new_vma = vm_area_dup(vma); 3225e3975891SChen Gang if (!new_vma) 3226e3975891SChen Gang goto out; 32271da177e4SLinus Torvalds new_vma->vm_start = addr; 32281da177e4SLinus Torvalds new_vma->vm_end = addr + len; 32291da177e4SLinus Torvalds new_vma->vm_pgoff = pgoff; 3230ef0855d3SOleg Nesterov if (vma_dup_policy(vma, new_vma)) 3231523d4e20SMichel Lespinasse goto out_free_vma; 3232523d4e20SMichel Lespinasse if (anon_vma_clone(new_vma, vma)) 3233523d4e20SMichel Lespinasse goto out_free_mempol; 3234e9714acfSKonstantin Khlebnikov if (new_vma->vm_file) 32351da177e4SLinus Torvalds get_file(new_vma->vm_file); 32361da177e4SLinus Torvalds if (new_vma->vm_ops && new_vma->vm_ops->open) 32371da177e4SLinus Torvalds new_vma->vm_ops->open(new_vma); 3238763ecb03SLiam R. Howlett if (vma_link(mm, new_vma)) 3239524e00b3SLiam R. Howlett goto out_vma_link; 324038a76013SMichel Lespinasse *need_rmap_locks = false; 32411da177e4SLinus Torvalds } 3242d4af56c5SLiam R. Howlett validate_mm_mt(mm); 32431da177e4SLinus Torvalds return new_vma; 32445beb4930SRik van Riel 3245524e00b3SLiam R. Howlett out_vma_link: 3246524e00b3SLiam R. Howlett if (new_vma->vm_ops && new_vma->vm_ops->close) 3247524e00b3SLiam R. Howlett new_vma->vm_ops->close(new_vma); 324892b73996SLiam Howlett 324992b73996SLiam Howlett if (new_vma->vm_file) 325092b73996SLiam Howlett fput(new_vma->vm_file); 325192b73996SLiam Howlett 325292b73996SLiam Howlett unlink_anon_vmas(new_vma); 32535beb4930SRik van Riel out_free_mempol: 3254ef0855d3SOleg Nesterov mpol_put(vma_policy(new_vma)); 32555beb4930SRik van Riel out_free_vma: 32563928d4f5SLinus Torvalds vm_area_free(new_vma); 3257e3975891SChen Gang out: 3258d4af56c5SLiam R. Howlett validate_mm_mt(mm); 32595beb4930SRik van Riel return NULL; 32601da177e4SLinus Torvalds } 3261119f657cSakpm@osdl.org 3262119f657cSakpm@osdl.org /* 3263119f657cSakpm@osdl.org * Return true if the calling process may expand its vm space by the passed 3264119f657cSakpm@osdl.org * number of pages 3265119f657cSakpm@osdl.org */ 326684638335SKonstantin Khlebnikov bool may_expand_vm(struct mm_struct *mm, vm_flags_t flags, unsigned long npages) 3267119f657cSakpm@osdl.org { 326884638335SKonstantin Khlebnikov if (mm->total_vm + npages > rlimit(RLIMIT_AS) >> PAGE_SHIFT) 326984638335SKonstantin Khlebnikov return false; 3270119f657cSakpm@osdl.org 3271d977d56cSKonstantin Khlebnikov if (is_data_mapping(flags) && 3272d977d56cSKonstantin Khlebnikov mm->data_vm + npages > rlimit(RLIMIT_DATA) >> PAGE_SHIFT) { 3273f4fcd558SKonstantin Khlebnikov /* Workaround for Valgrind */ 3274f4fcd558SKonstantin Khlebnikov if (rlimit(RLIMIT_DATA) == 0 && 3275f4fcd558SKonstantin Khlebnikov mm->data_vm + npages <= rlimit_max(RLIMIT_DATA) >> PAGE_SHIFT) 3276f4fcd558SKonstantin Khlebnikov return true; 327757a7702bSDavid Woodhouse 327857a7702bSDavid Woodhouse pr_warn_once("%s (%d): VmData %lu exceed data ulimit %lu. Update limits%s.\n", 3279d977d56cSKonstantin Khlebnikov current->comm, current->pid, 3280d977d56cSKonstantin Khlebnikov (mm->data_vm + npages) << PAGE_SHIFT, 328157a7702bSDavid Woodhouse rlimit(RLIMIT_DATA), 328257a7702bSDavid Woodhouse ignore_rlimit_data ? "" : " or use boot option ignore_rlimit_data"); 328357a7702bSDavid Woodhouse 328457a7702bSDavid Woodhouse if (!ignore_rlimit_data) 3285d977d56cSKonstantin Khlebnikov return false; 3286d977d56cSKonstantin Khlebnikov } 3287119f657cSakpm@osdl.org 328884638335SKonstantin Khlebnikov return true; 328984638335SKonstantin Khlebnikov } 329084638335SKonstantin Khlebnikov 329184638335SKonstantin Khlebnikov void vm_stat_account(struct mm_struct *mm, vm_flags_t flags, long npages) 329284638335SKonstantin Khlebnikov { 32937866076bSPeng Liu WRITE_ONCE(mm->total_vm, READ_ONCE(mm->total_vm)+npages); 329484638335SKonstantin Khlebnikov 3295d977d56cSKonstantin Khlebnikov if (is_exec_mapping(flags)) 329684638335SKonstantin Khlebnikov mm->exec_vm += npages; 3297d977d56cSKonstantin Khlebnikov else if (is_stack_mapping(flags)) 329884638335SKonstantin Khlebnikov mm->stack_vm += npages; 3299d977d56cSKonstantin Khlebnikov else if (is_data_mapping(flags)) 330084638335SKonstantin Khlebnikov mm->data_vm += npages; 3301119f657cSakpm@osdl.org } 3302fa5dc22fSRoland McGrath 3303b3ec9f33SSouptick Joarder static vm_fault_t special_mapping_fault(struct vm_fault *vmf); 3304a62c34bdSAndy Lutomirski 3305a62c34bdSAndy Lutomirski /* 3306a62c34bdSAndy Lutomirski * Having a close hook prevents vma merging regardless of flags. 3307a62c34bdSAndy Lutomirski */ 3308a62c34bdSAndy Lutomirski static void special_mapping_close(struct vm_area_struct *vma) 3309a62c34bdSAndy Lutomirski { 3310a62c34bdSAndy Lutomirski } 3311a62c34bdSAndy Lutomirski 3312a62c34bdSAndy Lutomirski static const char *special_mapping_name(struct vm_area_struct *vma) 3313a62c34bdSAndy Lutomirski { 3314a62c34bdSAndy Lutomirski return ((struct vm_special_mapping *)vma->vm_private_data)->name; 3315a62c34bdSAndy Lutomirski } 3316a62c34bdSAndy Lutomirski 331714d07113SBrian Geffon static int special_mapping_mremap(struct vm_area_struct *new_vma) 3318b059a453SDmitry Safonov { 3319b059a453SDmitry Safonov struct vm_special_mapping *sm = new_vma->vm_private_data; 3320b059a453SDmitry Safonov 3321280e87e9SDmitry Safonov if (WARN_ON_ONCE(current->mm != new_vma->vm_mm)) 3322280e87e9SDmitry Safonov return -EFAULT; 3323280e87e9SDmitry Safonov 3324b059a453SDmitry Safonov if (sm->mremap) 3325b059a453SDmitry Safonov return sm->mremap(sm, new_vma); 3326280e87e9SDmitry Safonov 3327b059a453SDmitry Safonov return 0; 3328b059a453SDmitry Safonov } 3329b059a453SDmitry Safonov 3330871402e0SDmitry Safonov static int special_mapping_split(struct vm_area_struct *vma, unsigned long addr) 3331871402e0SDmitry Safonov { 3332871402e0SDmitry Safonov /* 3333871402e0SDmitry Safonov * Forbid splitting special mappings - kernel has expectations over 3334871402e0SDmitry Safonov * the number of pages in mapping. Together with VM_DONTEXPAND 3335871402e0SDmitry Safonov * the size of vma should stay the same over the special mapping's 3336871402e0SDmitry Safonov * lifetime. 3337871402e0SDmitry Safonov */ 3338871402e0SDmitry Safonov return -EINVAL; 3339871402e0SDmitry Safonov } 3340871402e0SDmitry Safonov 3341a62c34bdSAndy Lutomirski static const struct vm_operations_struct special_mapping_vmops = { 3342a62c34bdSAndy Lutomirski .close = special_mapping_close, 3343a62c34bdSAndy Lutomirski .fault = special_mapping_fault, 3344b059a453SDmitry Safonov .mremap = special_mapping_mremap, 3345a62c34bdSAndy Lutomirski .name = special_mapping_name, 3346af34ebebSDmitry Safonov /* vDSO code relies that VVAR can't be accessed remotely */ 3347af34ebebSDmitry Safonov .access = NULL, 3348871402e0SDmitry Safonov .may_split = special_mapping_split, 3349a62c34bdSAndy Lutomirski }; 3350a62c34bdSAndy Lutomirski 3351a62c34bdSAndy Lutomirski static const struct vm_operations_struct legacy_special_mapping_vmops = { 3352a62c34bdSAndy Lutomirski .close = special_mapping_close, 3353a62c34bdSAndy Lutomirski .fault = special_mapping_fault, 3354a62c34bdSAndy Lutomirski }; 3355fa5dc22fSRoland McGrath 3356b3ec9f33SSouptick Joarder static vm_fault_t special_mapping_fault(struct vm_fault *vmf) 3357fa5dc22fSRoland McGrath { 335811bac800SDave Jiang struct vm_area_struct *vma = vmf->vma; 3359b1d0e4f5SNick Piggin pgoff_t pgoff; 3360fa5dc22fSRoland McGrath struct page **pages; 3361fa5dc22fSRoland McGrath 3362f872f540SAndy Lutomirski if (vma->vm_ops == &legacy_special_mapping_vmops) { 3363a62c34bdSAndy Lutomirski pages = vma->vm_private_data; 3364f872f540SAndy Lutomirski } else { 3365f872f540SAndy Lutomirski struct vm_special_mapping *sm = vma->vm_private_data; 3366f872f540SAndy Lutomirski 3367f872f540SAndy Lutomirski if (sm->fault) 336811bac800SDave Jiang return sm->fault(sm, vmf->vma, vmf); 3369f872f540SAndy Lutomirski 3370f872f540SAndy Lutomirski pages = sm->pages; 3371f872f540SAndy Lutomirski } 3372a62c34bdSAndy Lutomirski 33738a9cc3b5SOleg Nesterov for (pgoff = vmf->pgoff; pgoff && *pages; ++pages) 3374b1d0e4f5SNick Piggin pgoff--; 3375fa5dc22fSRoland McGrath 3376fa5dc22fSRoland McGrath if (*pages) { 3377fa5dc22fSRoland McGrath struct page *page = *pages; 3378fa5dc22fSRoland McGrath get_page(page); 3379b1d0e4f5SNick Piggin vmf->page = page; 3380b1d0e4f5SNick Piggin return 0; 3381fa5dc22fSRoland McGrath } 3382fa5dc22fSRoland McGrath 3383b1d0e4f5SNick Piggin return VM_FAULT_SIGBUS; 3384fa5dc22fSRoland McGrath } 3385fa5dc22fSRoland McGrath 3386a62c34bdSAndy Lutomirski static struct vm_area_struct *__install_special_mapping( 3387a62c34bdSAndy Lutomirski struct mm_struct *mm, 3388fa5dc22fSRoland McGrath unsigned long addr, unsigned long len, 338927f28b97SChen Gang unsigned long vm_flags, void *priv, 339027f28b97SChen Gang const struct vm_operations_struct *ops) 3391fa5dc22fSRoland McGrath { 3392462e635eSTavis Ormandy int ret; 3393fa5dc22fSRoland McGrath struct vm_area_struct *vma; 3394fa5dc22fSRoland McGrath 3395d4af56c5SLiam R. Howlett validate_mm_mt(mm); 3396490fc053SLinus Torvalds vma = vm_area_alloc(mm); 3397fa5dc22fSRoland McGrath if (unlikely(vma == NULL)) 33983935ed6aSStefani Seibold return ERR_PTR(-ENOMEM); 3399fa5dc22fSRoland McGrath 3400fa5dc22fSRoland McGrath vma->vm_start = addr; 3401fa5dc22fSRoland McGrath vma->vm_end = addr + len; 3402fa5dc22fSRoland McGrath 3403d9104d1cSCyrill Gorcunov vma->vm_flags = vm_flags | mm->def_flags | VM_DONTEXPAND | VM_SOFTDIRTY; 34041fc09228SHugh Dickins vma->vm_flags &= VM_LOCKED_CLEAR_MASK; 34053ed75eb8SColy Li vma->vm_page_prot = vm_get_page_prot(vma->vm_flags); 3406fa5dc22fSRoland McGrath 3407a62c34bdSAndy Lutomirski vma->vm_ops = ops; 3408a62c34bdSAndy Lutomirski vma->vm_private_data = priv; 3409fa5dc22fSRoland McGrath 3410462e635eSTavis Ormandy ret = insert_vm_struct(mm, vma); 3411462e635eSTavis Ormandy if (ret) 3412462e635eSTavis Ormandy goto out; 3413fa5dc22fSRoland McGrath 341484638335SKonstantin Khlebnikov vm_stat_account(mm, vma->vm_flags, len >> PAGE_SHIFT); 3415fa5dc22fSRoland McGrath 3416cdd6c482SIngo Molnar perf_event_mmap(vma); 3417089dd79dSPeter Zijlstra 3418d4af56c5SLiam R. Howlett validate_mm_mt(mm); 34193935ed6aSStefani Seibold return vma; 3420462e635eSTavis Ormandy 3421462e635eSTavis Ormandy out: 34223928d4f5SLinus Torvalds vm_area_free(vma); 3423d4af56c5SLiam R. Howlett validate_mm_mt(mm); 34243935ed6aSStefani Seibold return ERR_PTR(ret); 34253935ed6aSStefani Seibold } 34263935ed6aSStefani Seibold 34272eefd878SDmitry Safonov bool vma_is_special_mapping(const struct vm_area_struct *vma, 34282eefd878SDmitry Safonov const struct vm_special_mapping *sm) 34292eefd878SDmitry Safonov { 34302eefd878SDmitry Safonov return vma->vm_private_data == sm && 34312eefd878SDmitry Safonov (vma->vm_ops == &special_mapping_vmops || 34322eefd878SDmitry Safonov vma->vm_ops == &legacy_special_mapping_vmops); 34332eefd878SDmitry Safonov } 34342eefd878SDmitry Safonov 3435a62c34bdSAndy Lutomirski /* 3436c1e8d7c6SMichel Lespinasse * Called with mm->mmap_lock held for writing. 3437a62c34bdSAndy Lutomirski * Insert a new vma covering the given region, with the given flags. 3438a62c34bdSAndy Lutomirski * Its pages are supplied by the given array of struct page *. 3439a62c34bdSAndy Lutomirski * The array can be shorter than len >> PAGE_SHIFT if it's null-terminated. 3440a62c34bdSAndy Lutomirski * The region past the last page supplied will always produce SIGBUS. 3441a62c34bdSAndy Lutomirski * The array pointer and the pages it points to are assumed to stay alive 3442a62c34bdSAndy Lutomirski * for as long as this mapping might exist. 3443a62c34bdSAndy Lutomirski */ 3444a62c34bdSAndy Lutomirski struct vm_area_struct *_install_special_mapping( 3445a62c34bdSAndy Lutomirski struct mm_struct *mm, 3446a62c34bdSAndy Lutomirski unsigned long addr, unsigned long len, 3447a62c34bdSAndy Lutomirski unsigned long vm_flags, const struct vm_special_mapping *spec) 3448a62c34bdSAndy Lutomirski { 344927f28b97SChen Gang return __install_special_mapping(mm, addr, len, vm_flags, (void *)spec, 345027f28b97SChen Gang &special_mapping_vmops); 3451a62c34bdSAndy Lutomirski } 3452a62c34bdSAndy Lutomirski 34533935ed6aSStefani Seibold int install_special_mapping(struct mm_struct *mm, 34543935ed6aSStefani Seibold unsigned long addr, unsigned long len, 34553935ed6aSStefani Seibold unsigned long vm_flags, struct page **pages) 34563935ed6aSStefani Seibold { 3457a62c34bdSAndy Lutomirski struct vm_area_struct *vma = __install_special_mapping( 345827f28b97SChen Gang mm, addr, len, vm_flags, (void *)pages, 345927f28b97SChen Gang &legacy_special_mapping_vmops); 34603935ed6aSStefani Seibold 346114bd5b45SDuan Jiong return PTR_ERR_OR_ZERO(vma); 3462fa5dc22fSRoland McGrath } 34637906d00cSAndrea Arcangeli 34647906d00cSAndrea Arcangeli static DEFINE_MUTEX(mm_all_locks_mutex); 34657906d00cSAndrea Arcangeli 3466454ed842SPeter Zijlstra static void vm_lock_anon_vma(struct mm_struct *mm, struct anon_vma *anon_vma) 34677906d00cSAndrea Arcangeli { 3468f808c13fSDavidlohr Bueso if (!test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_root.rb_node)) { 34697906d00cSAndrea Arcangeli /* 34707906d00cSAndrea Arcangeli * The LSB of head.next can't change from under us 34717906d00cSAndrea Arcangeli * because we hold the mm_all_locks_mutex. 34727906d00cSAndrea Arcangeli */ 3473da1c55f1SMichel Lespinasse down_write_nest_lock(&anon_vma->root->rwsem, &mm->mmap_lock); 34747906d00cSAndrea Arcangeli /* 34757906d00cSAndrea Arcangeli * We can safely modify head.next after taking the 34765a505085SIngo Molnar * anon_vma->root->rwsem. If some other vma in this mm shares 34777906d00cSAndrea Arcangeli * the same anon_vma we won't take it again. 34787906d00cSAndrea Arcangeli * 34797906d00cSAndrea Arcangeli * No need of atomic instructions here, head.next 34807906d00cSAndrea Arcangeli * can't change from under us thanks to the 34815a505085SIngo Molnar * anon_vma->root->rwsem. 34827906d00cSAndrea Arcangeli */ 34837906d00cSAndrea Arcangeli if (__test_and_set_bit(0, (unsigned long *) 3484f808c13fSDavidlohr Bueso &anon_vma->root->rb_root.rb_root.rb_node)) 34857906d00cSAndrea Arcangeli BUG(); 34867906d00cSAndrea Arcangeli } 34877906d00cSAndrea Arcangeli } 34887906d00cSAndrea Arcangeli 3489454ed842SPeter Zijlstra static void vm_lock_mapping(struct mm_struct *mm, struct address_space *mapping) 34907906d00cSAndrea Arcangeli { 34917906d00cSAndrea Arcangeli if (!test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) { 34927906d00cSAndrea Arcangeli /* 34937906d00cSAndrea Arcangeli * AS_MM_ALL_LOCKS can't change from under us because 34947906d00cSAndrea Arcangeli * we hold the mm_all_locks_mutex. 34957906d00cSAndrea Arcangeli * 34967906d00cSAndrea Arcangeli * Operations on ->flags have to be atomic because 34977906d00cSAndrea Arcangeli * even if AS_MM_ALL_LOCKS is stable thanks to the 34987906d00cSAndrea Arcangeli * mm_all_locks_mutex, there may be other cpus 34997906d00cSAndrea Arcangeli * changing other bitflags in parallel to us. 35007906d00cSAndrea Arcangeli */ 35017906d00cSAndrea Arcangeli if (test_and_set_bit(AS_MM_ALL_LOCKS, &mapping->flags)) 35027906d00cSAndrea Arcangeli BUG(); 3503da1c55f1SMichel Lespinasse down_write_nest_lock(&mapping->i_mmap_rwsem, &mm->mmap_lock); 35047906d00cSAndrea Arcangeli } 35057906d00cSAndrea Arcangeli } 35067906d00cSAndrea Arcangeli 35077906d00cSAndrea Arcangeli /* 35087906d00cSAndrea Arcangeli * This operation locks against the VM for all pte/vma/mm related 35097906d00cSAndrea Arcangeli * operations that could ever happen on a certain mm. This includes 35107906d00cSAndrea Arcangeli * vmtruncate, try_to_unmap, and all page faults. 35117906d00cSAndrea Arcangeli * 3512c1e8d7c6SMichel Lespinasse * The caller must take the mmap_lock in write mode before calling 35137906d00cSAndrea Arcangeli * mm_take_all_locks(). The caller isn't allowed to release the 3514c1e8d7c6SMichel Lespinasse * mmap_lock until mm_drop_all_locks() returns. 35157906d00cSAndrea Arcangeli * 3516c1e8d7c6SMichel Lespinasse * mmap_lock in write mode is required in order to block all operations 35177906d00cSAndrea Arcangeli * that could modify pagetables and free pages without need of 351827ba0644SKirill A. Shutemov * altering the vma layout. It's also needed in write mode to avoid new 35197906d00cSAndrea Arcangeli * anon_vmas to be associated with existing vmas. 35207906d00cSAndrea Arcangeli * 35217906d00cSAndrea Arcangeli * A single task can't take more than one mm_take_all_locks() in a row 35227906d00cSAndrea Arcangeli * or it would deadlock. 35237906d00cSAndrea Arcangeli * 3524bf181b9fSMichel Lespinasse * The LSB in anon_vma->rb_root.rb_node and the AS_MM_ALL_LOCKS bitflag in 35257906d00cSAndrea Arcangeli * mapping->flags avoid to take the same lock twice, if more than one 35267906d00cSAndrea Arcangeli * vma in this mm is backed by the same anon_vma or address_space. 35277906d00cSAndrea Arcangeli * 352888f306b6SKirill A. Shutemov * We take locks in following order, accordingly to comment at beginning 352988f306b6SKirill A. Shutemov * of mm/rmap.c: 353088f306b6SKirill A. Shutemov * - all hugetlbfs_i_mmap_rwsem_key locks (aka mapping->i_mmap_rwsem for 353188f306b6SKirill A. Shutemov * hugetlb mapping); 353288f306b6SKirill A. Shutemov * - all i_mmap_rwsem locks; 353388f306b6SKirill A. Shutemov * - all anon_vma->rwseml 353488f306b6SKirill A. Shutemov * 353588f306b6SKirill A. Shutemov * We can take all locks within these types randomly because the VM code 353688f306b6SKirill A. Shutemov * doesn't nest them and we protected from parallel mm_take_all_locks() by 353788f306b6SKirill A. Shutemov * mm_all_locks_mutex. 35387906d00cSAndrea Arcangeli * 35397906d00cSAndrea Arcangeli * mm_take_all_locks() and mm_drop_all_locks are expensive operations 35407906d00cSAndrea Arcangeli * that may have to take thousand of locks. 35417906d00cSAndrea Arcangeli * 35427906d00cSAndrea Arcangeli * mm_take_all_locks() can fail if it's interrupted by signals. 35437906d00cSAndrea Arcangeli */ 35447906d00cSAndrea Arcangeli int mm_take_all_locks(struct mm_struct *mm) 35457906d00cSAndrea Arcangeli { 35467906d00cSAndrea Arcangeli struct vm_area_struct *vma; 35475beb4930SRik van Riel struct anon_vma_chain *avc; 3548763ecb03SLiam R. Howlett MA_STATE(mas, &mm->mm_mt, 0, 0); 35497906d00cSAndrea Arcangeli 3550325bca1fSRolf Eike Beer mmap_assert_write_locked(mm); 35517906d00cSAndrea Arcangeli 35527906d00cSAndrea Arcangeli mutex_lock(&mm_all_locks_mutex); 35537906d00cSAndrea Arcangeli 3554763ecb03SLiam R. Howlett mas_for_each(&mas, vma, ULONG_MAX) { 35557906d00cSAndrea Arcangeli if (signal_pending(current)) 35567906d00cSAndrea Arcangeli goto out_unlock; 355788f306b6SKirill A. Shutemov if (vma->vm_file && vma->vm_file->f_mapping && 355888f306b6SKirill A. Shutemov is_vm_hugetlb_page(vma)) 355988f306b6SKirill A. Shutemov vm_lock_mapping(mm, vma->vm_file->f_mapping); 356088f306b6SKirill A. Shutemov } 356188f306b6SKirill A. Shutemov 3562763ecb03SLiam R. Howlett mas_set(&mas, 0); 3563763ecb03SLiam R. Howlett mas_for_each(&mas, vma, ULONG_MAX) { 356488f306b6SKirill A. Shutemov if (signal_pending(current)) 356588f306b6SKirill A. Shutemov goto out_unlock; 356688f306b6SKirill A. Shutemov if (vma->vm_file && vma->vm_file->f_mapping && 356788f306b6SKirill A. Shutemov !is_vm_hugetlb_page(vma)) 3568454ed842SPeter Zijlstra vm_lock_mapping(mm, vma->vm_file->f_mapping); 35697906d00cSAndrea Arcangeli } 35707cd5a02fSPeter Zijlstra 3571763ecb03SLiam R. Howlett mas_set(&mas, 0); 3572763ecb03SLiam R. Howlett mas_for_each(&mas, vma, ULONG_MAX) { 35737cd5a02fSPeter Zijlstra if (signal_pending(current)) 35747cd5a02fSPeter Zijlstra goto out_unlock; 35757cd5a02fSPeter Zijlstra if (vma->anon_vma) 35765beb4930SRik van Riel list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) 35775beb4930SRik van Riel vm_lock_anon_vma(mm, avc->anon_vma); 35787cd5a02fSPeter Zijlstra } 35797cd5a02fSPeter Zijlstra 3580584cff54SKautuk Consul return 0; 35817906d00cSAndrea Arcangeli 35827906d00cSAndrea Arcangeli out_unlock: 35837906d00cSAndrea Arcangeli mm_drop_all_locks(mm); 3584584cff54SKautuk Consul return -EINTR; 35857906d00cSAndrea Arcangeli } 35867906d00cSAndrea Arcangeli 35877906d00cSAndrea Arcangeli static void vm_unlock_anon_vma(struct anon_vma *anon_vma) 35887906d00cSAndrea Arcangeli { 3589f808c13fSDavidlohr Bueso if (test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_root.rb_node)) { 35907906d00cSAndrea Arcangeli /* 35917906d00cSAndrea Arcangeli * The LSB of head.next can't change to 0 from under 35927906d00cSAndrea Arcangeli * us because we hold the mm_all_locks_mutex. 35937906d00cSAndrea Arcangeli * 35947906d00cSAndrea Arcangeli * We must however clear the bitflag before unlocking 3595bf181b9fSMichel Lespinasse * the vma so the users using the anon_vma->rb_root will 35967906d00cSAndrea Arcangeli * never see our bitflag. 35977906d00cSAndrea Arcangeli * 35987906d00cSAndrea Arcangeli * No need of atomic instructions here, head.next 35997906d00cSAndrea Arcangeli * can't change from under us until we release the 36005a505085SIngo Molnar * anon_vma->root->rwsem. 36017906d00cSAndrea Arcangeli */ 36027906d00cSAndrea Arcangeli if (!__test_and_clear_bit(0, (unsigned long *) 3603f808c13fSDavidlohr Bueso &anon_vma->root->rb_root.rb_root.rb_node)) 36047906d00cSAndrea Arcangeli BUG(); 360508b52706SKonstantin Khlebnikov anon_vma_unlock_write(anon_vma); 36067906d00cSAndrea Arcangeli } 36077906d00cSAndrea Arcangeli } 36087906d00cSAndrea Arcangeli 36097906d00cSAndrea Arcangeli static void vm_unlock_mapping(struct address_space *mapping) 36107906d00cSAndrea Arcangeli { 36117906d00cSAndrea Arcangeli if (test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) { 36127906d00cSAndrea Arcangeli /* 36137906d00cSAndrea Arcangeli * AS_MM_ALL_LOCKS can't change to 0 from under us 36147906d00cSAndrea Arcangeli * because we hold the mm_all_locks_mutex. 36157906d00cSAndrea Arcangeli */ 361683cde9e8SDavidlohr Bueso i_mmap_unlock_write(mapping); 36177906d00cSAndrea Arcangeli if (!test_and_clear_bit(AS_MM_ALL_LOCKS, 36187906d00cSAndrea Arcangeli &mapping->flags)) 36197906d00cSAndrea Arcangeli BUG(); 36207906d00cSAndrea Arcangeli } 36217906d00cSAndrea Arcangeli } 36227906d00cSAndrea Arcangeli 36237906d00cSAndrea Arcangeli /* 3624c1e8d7c6SMichel Lespinasse * The mmap_lock cannot be released by the caller until 36257906d00cSAndrea Arcangeli * mm_drop_all_locks() returns. 36267906d00cSAndrea Arcangeli */ 36277906d00cSAndrea Arcangeli void mm_drop_all_locks(struct mm_struct *mm) 36287906d00cSAndrea Arcangeli { 36297906d00cSAndrea Arcangeli struct vm_area_struct *vma; 36305beb4930SRik van Riel struct anon_vma_chain *avc; 3631763ecb03SLiam R. Howlett MA_STATE(mas, &mm->mm_mt, 0, 0); 36327906d00cSAndrea Arcangeli 3633325bca1fSRolf Eike Beer mmap_assert_write_locked(mm); 36347906d00cSAndrea Arcangeli BUG_ON(!mutex_is_locked(&mm_all_locks_mutex)); 36357906d00cSAndrea Arcangeli 3636763ecb03SLiam R. Howlett mas_for_each(&mas, vma, ULONG_MAX) { 36377906d00cSAndrea Arcangeli if (vma->anon_vma) 36385beb4930SRik van Riel list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) 36395beb4930SRik van Riel vm_unlock_anon_vma(avc->anon_vma); 36407906d00cSAndrea Arcangeli if (vma->vm_file && vma->vm_file->f_mapping) 36417906d00cSAndrea Arcangeli vm_unlock_mapping(vma->vm_file->f_mapping); 36427906d00cSAndrea Arcangeli } 36437906d00cSAndrea Arcangeli 36447906d00cSAndrea Arcangeli mutex_unlock(&mm_all_locks_mutex); 36457906d00cSAndrea Arcangeli } 36468feae131SDavid Howells 36478feae131SDavid Howells /* 36483edf41d8Sseokhoon.yoon * initialise the percpu counter for VM 36498feae131SDavid Howells */ 36508feae131SDavid Howells void __init mmap_init(void) 36518feae131SDavid Howells { 365200a62ce9SKOSAKI Motohiro int ret; 365300a62ce9SKOSAKI Motohiro 3654908c7f19STejun Heo ret = percpu_counter_init(&vm_committed_as, 0, GFP_KERNEL); 365500a62ce9SKOSAKI Motohiro VM_BUG_ON(ret); 36568feae131SDavid Howells } 3657c9b1d098SAndrew Shewmaker 3658c9b1d098SAndrew Shewmaker /* 3659c9b1d098SAndrew Shewmaker * Initialise sysctl_user_reserve_kbytes. 3660c9b1d098SAndrew Shewmaker * 3661c9b1d098SAndrew Shewmaker * This is intended to prevent a user from starting a single memory hogging 3662c9b1d098SAndrew Shewmaker * process, such that they cannot recover (kill the hog) in OVERCOMMIT_NEVER 3663c9b1d098SAndrew Shewmaker * mode. 3664c9b1d098SAndrew Shewmaker * 3665c9b1d098SAndrew Shewmaker * The default value is min(3% of free memory, 128MB) 3666c9b1d098SAndrew Shewmaker * 128MB is enough to recover with sshd/login, bash, and top/kill. 3667c9b1d098SAndrew Shewmaker */ 36681640879aSAndrew Shewmaker static int init_user_reserve(void) 3669c9b1d098SAndrew Shewmaker { 3670c9b1d098SAndrew Shewmaker unsigned long free_kbytes; 3671c9b1d098SAndrew Shewmaker 3672c41f012aSMichal Hocko free_kbytes = global_zone_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10); 3673c9b1d098SAndrew Shewmaker 3674c9b1d098SAndrew Shewmaker sysctl_user_reserve_kbytes = min(free_kbytes / 32, 1UL << 17); 3675c9b1d098SAndrew Shewmaker return 0; 3676c9b1d098SAndrew Shewmaker } 3677a64fb3cdSPaul Gortmaker subsys_initcall(init_user_reserve); 36784eeab4f5SAndrew Shewmaker 36794eeab4f5SAndrew Shewmaker /* 36804eeab4f5SAndrew Shewmaker * Initialise sysctl_admin_reserve_kbytes. 36814eeab4f5SAndrew Shewmaker * 36824eeab4f5SAndrew Shewmaker * The purpose of sysctl_admin_reserve_kbytes is to allow the sys admin 36834eeab4f5SAndrew Shewmaker * to log in and kill a memory hogging process. 36844eeab4f5SAndrew Shewmaker * 36854eeab4f5SAndrew Shewmaker * Systems with more than 256MB will reserve 8MB, enough to recover 36864eeab4f5SAndrew Shewmaker * with sshd, bash, and top in OVERCOMMIT_GUESS. Smaller systems will 36874eeab4f5SAndrew Shewmaker * only reserve 3% of free pages by default. 36884eeab4f5SAndrew Shewmaker */ 36891640879aSAndrew Shewmaker static int init_admin_reserve(void) 36904eeab4f5SAndrew Shewmaker { 36914eeab4f5SAndrew Shewmaker unsigned long free_kbytes; 36924eeab4f5SAndrew Shewmaker 3693c41f012aSMichal Hocko free_kbytes = global_zone_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10); 36944eeab4f5SAndrew Shewmaker 36954eeab4f5SAndrew Shewmaker sysctl_admin_reserve_kbytes = min(free_kbytes / 32, 1UL << 13); 36964eeab4f5SAndrew Shewmaker return 0; 36974eeab4f5SAndrew Shewmaker } 3698a64fb3cdSPaul Gortmaker subsys_initcall(init_admin_reserve); 36991640879aSAndrew Shewmaker 37001640879aSAndrew Shewmaker /* 37011640879aSAndrew Shewmaker * Reinititalise user and admin reserves if memory is added or removed. 37021640879aSAndrew Shewmaker * 37031640879aSAndrew Shewmaker * The default user reserve max is 128MB, and the default max for the 37041640879aSAndrew Shewmaker * admin reserve is 8MB. These are usually, but not always, enough to 37051640879aSAndrew Shewmaker * enable recovery from a memory hogging process using login/sshd, a shell, 37061640879aSAndrew Shewmaker * and tools like top. It may make sense to increase or even disable the 37071640879aSAndrew Shewmaker * reserve depending on the existence of swap or variations in the recovery 37081640879aSAndrew Shewmaker * tools. So, the admin may have changed them. 37091640879aSAndrew Shewmaker * 37101640879aSAndrew Shewmaker * If memory is added and the reserves have been eliminated or increased above 37111640879aSAndrew Shewmaker * the default max, then we'll trust the admin. 37121640879aSAndrew Shewmaker * 37131640879aSAndrew Shewmaker * If memory is removed and there isn't enough free memory, then we 37141640879aSAndrew Shewmaker * need to reset the reserves. 37151640879aSAndrew Shewmaker * 37161640879aSAndrew Shewmaker * Otherwise keep the reserve set by the admin. 37171640879aSAndrew Shewmaker */ 37181640879aSAndrew Shewmaker static int reserve_mem_notifier(struct notifier_block *nb, 37191640879aSAndrew Shewmaker unsigned long action, void *data) 37201640879aSAndrew Shewmaker { 37211640879aSAndrew Shewmaker unsigned long tmp, free_kbytes; 37221640879aSAndrew Shewmaker 37231640879aSAndrew Shewmaker switch (action) { 37241640879aSAndrew Shewmaker case MEM_ONLINE: 37251640879aSAndrew Shewmaker /* Default max is 128MB. Leave alone if modified by operator. */ 37261640879aSAndrew Shewmaker tmp = sysctl_user_reserve_kbytes; 37271640879aSAndrew Shewmaker if (0 < tmp && tmp < (1UL << 17)) 37281640879aSAndrew Shewmaker init_user_reserve(); 37291640879aSAndrew Shewmaker 37301640879aSAndrew Shewmaker /* Default max is 8MB. Leave alone if modified by operator. */ 37311640879aSAndrew Shewmaker tmp = sysctl_admin_reserve_kbytes; 37321640879aSAndrew Shewmaker if (0 < tmp && tmp < (1UL << 13)) 37331640879aSAndrew Shewmaker init_admin_reserve(); 37341640879aSAndrew Shewmaker 37351640879aSAndrew Shewmaker break; 37361640879aSAndrew Shewmaker case MEM_OFFLINE: 3737c41f012aSMichal Hocko free_kbytes = global_zone_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10); 37381640879aSAndrew Shewmaker 37391640879aSAndrew Shewmaker if (sysctl_user_reserve_kbytes > free_kbytes) { 37401640879aSAndrew Shewmaker init_user_reserve(); 37411640879aSAndrew Shewmaker pr_info("vm.user_reserve_kbytes reset to %lu\n", 37421640879aSAndrew Shewmaker sysctl_user_reserve_kbytes); 37431640879aSAndrew Shewmaker } 37441640879aSAndrew Shewmaker 37451640879aSAndrew Shewmaker if (sysctl_admin_reserve_kbytes > free_kbytes) { 37461640879aSAndrew Shewmaker init_admin_reserve(); 37471640879aSAndrew Shewmaker pr_info("vm.admin_reserve_kbytes reset to %lu\n", 37481640879aSAndrew Shewmaker sysctl_admin_reserve_kbytes); 37491640879aSAndrew Shewmaker } 37501640879aSAndrew Shewmaker break; 37511640879aSAndrew Shewmaker default: 37521640879aSAndrew Shewmaker break; 37531640879aSAndrew Shewmaker } 37541640879aSAndrew Shewmaker return NOTIFY_OK; 37551640879aSAndrew Shewmaker } 37561640879aSAndrew Shewmaker 37571640879aSAndrew Shewmaker static int __meminit init_reserve_notifier(void) 37581640879aSAndrew Shewmaker { 37591eeaa4fdSLiu Shixin if (hotplug_memory_notifier(reserve_mem_notifier, DEFAULT_CALLBACK_PRI)) 3760b1de0d13SMitchel Humpherys pr_err("Failed registering memory add/remove notifier for admin reserve\n"); 37611640879aSAndrew Shewmaker 37621640879aSAndrew Shewmaker return 0; 37631640879aSAndrew Shewmaker } 3764a64fb3cdSPaul Gortmaker subsys_initcall(init_reserve_notifier); 3765