1457c8996SThomas Gleixner // SPDX-License-Identifier: GPL-2.0-only 21da177e4SLinus Torvalds /* 31da177e4SLinus Torvalds * mm/mmap.c 41da177e4SLinus Torvalds * 51da177e4SLinus Torvalds * Written by obz. 61da177e4SLinus Torvalds * 7046c6884SAlan Cox * Address space accounting code <alan@lxorguk.ukuu.org.uk> 81da177e4SLinus Torvalds */ 91da177e4SLinus Torvalds 10b1de0d13SMitchel Humpherys #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt 11b1de0d13SMitchel Humpherys 12e8420a8eSCyril Hrubis #include <linux/kernel.h> 131da177e4SLinus Torvalds #include <linux/slab.h> 144af3c9ccSAlexey Dobriyan #include <linux/backing-dev.h> 151da177e4SLinus Torvalds #include <linux/mm.h> 1617fca131SArnd Bergmann #include <linux/mm_inline.h> 171da177e4SLinus Torvalds #include <linux/shm.h> 181da177e4SLinus Torvalds #include <linux/mman.h> 191da177e4SLinus Torvalds #include <linux/pagemap.h> 201da177e4SLinus Torvalds #include <linux/swap.h> 211da177e4SLinus Torvalds #include <linux/syscalls.h> 22c59ede7bSRandy.Dunlap #include <linux/capability.h> 231da177e4SLinus Torvalds #include <linux/init.h> 241da177e4SLinus Torvalds #include <linux/file.h> 251da177e4SLinus Torvalds #include <linux/fs.h> 261da177e4SLinus Torvalds #include <linux/personality.h> 271da177e4SLinus Torvalds #include <linux/security.h> 281da177e4SLinus Torvalds #include <linux/hugetlb.h> 29c01d5b30SHugh Dickins #include <linux/shmem_fs.h> 301da177e4SLinus Torvalds #include <linux/profile.h> 31b95f1b31SPaul Gortmaker #include <linux/export.h> 321da177e4SLinus Torvalds #include <linux/mount.h> 331da177e4SLinus Torvalds #include <linux/mempolicy.h> 341da177e4SLinus Torvalds #include <linux/rmap.h> 35cddb8a5cSAndrea Arcangeli #include <linux/mmu_notifier.h> 3682f71ae4SKonstantin Khlebnikov #include <linux/mmdebug.h> 37cdd6c482SIngo Molnar #include <linux/perf_event.h> 38120a795dSAl Viro #include <linux/audit.h> 39b15d00b6SAndrea Arcangeli #include <linux/khugepaged.h> 402b144498SSrikar Dronamraju #include <linux/uprobes.h> 411640879aSAndrew Shewmaker #include <linux/notifier.h> 421640879aSAndrew Shewmaker #include <linux/memory.h> 43b1de0d13SMitchel Humpherys #include <linux/printk.h> 4419a809afSAndrea Arcangeli #include <linux/userfaultfd_k.h> 45d977d56cSKonstantin Khlebnikov #include <linux/moduleparam.h> 4662b5f7d0SDave Hansen #include <linux/pkeys.h> 4721292580SAndrea Arcangeli #include <linux/oom.h> 4804f5866eSAndrea Arcangeli #include <linux/sched/mm.h> 49d7597f59SStefan Roesch #include <linux/ksm.h> 501da177e4SLinus Torvalds 517c0f6ba6SLinus Torvalds #include <linux/uaccess.h> 521da177e4SLinus Torvalds #include <asm/cacheflush.h> 531da177e4SLinus Torvalds #include <asm/tlb.h> 54d6dd61c8SJeremy Fitzhardinge #include <asm/mmu_context.h> 551da177e4SLinus Torvalds 56df529cabSJaewon Kim #define CREATE_TRACE_POINTS 57df529cabSJaewon Kim #include <trace/events/mmap.h> 58df529cabSJaewon Kim 5942b77728SJan Beulich #include "internal.h" 6042b77728SJan Beulich 613a459756SKirill Korotaev #ifndef arch_mmap_check 623a459756SKirill Korotaev #define arch_mmap_check(addr, len, flags) (0) 633a459756SKirill Korotaev #endif 643a459756SKirill Korotaev 65d07e2259SDaniel Cashman #ifdef CONFIG_HAVE_ARCH_MMAP_RND_BITS 66d07e2259SDaniel Cashman const int mmap_rnd_bits_min = CONFIG_ARCH_MMAP_RND_BITS_MIN; 67d07e2259SDaniel Cashman const int mmap_rnd_bits_max = CONFIG_ARCH_MMAP_RND_BITS_MAX; 68d07e2259SDaniel Cashman int mmap_rnd_bits __read_mostly = CONFIG_ARCH_MMAP_RND_BITS; 69d07e2259SDaniel Cashman #endif 70d07e2259SDaniel Cashman #ifdef CONFIG_HAVE_ARCH_MMAP_RND_COMPAT_BITS 71d07e2259SDaniel Cashman const int mmap_rnd_compat_bits_min = CONFIG_ARCH_MMAP_RND_COMPAT_BITS_MIN; 72d07e2259SDaniel Cashman const int mmap_rnd_compat_bits_max = CONFIG_ARCH_MMAP_RND_COMPAT_BITS_MAX; 73d07e2259SDaniel Cashman int mmap_rnd_compat_bits __read_mostly = CONFIG_ARCH_MMAP_RND_COMPAT_BITS; 74d07e2259SDaniel Cashman #endif 75d07e2259SDaniel Cashman 76f4fcd558SKonstantin Khlebnikov static bool ignore_rlimit_data; 77d977d56cSKonstantin Khlebnikov core_param(ignore_rlimit_data, ignore_rlimit_data, bool, 0644); 78d07e2259SDaniel Cashman 79fd892593SLiam R. Howlett static void unmap_region(struct mm_struct *mm, struct ma_state *mas, 80e0da382cSHugh Dickins struct vm_area_struct *vma, struct vm_area_struct *prev, 81763ecb03SLiam R. Howlett struct vm_area_struct *next, unsigned long start, 82fd892593SLiam R. Howlett unsigned long end, unsigned long tree_end, bool mm_wr_locked); 83e0da382cSHugh Dickins 8464e45507SPeter Feiner static pgprot_t vm_pgprot_modify(pgprot_t oldprot, unsigned long vm_flags) 8564e45507SPeter Feiner { 8664e45507SPeter Feiner return pgprot_modify(oldprot, vm_get_page_prot(vm_flags)); 8764e45507SPeter Feiner } 8864e45507SPeter Feiner 8964e45507SPeter Feiner /* Update vma->vm_page_prot to reflect vma->vm_flags. */ 9064e45507SPeter Feiner void vma_set_page_prot(struct vm_area_struct *vma) 9164e45507SPeter Feiner { 9264e45507SPeter Feiner unsigned long vm_flags = vma->vm_flags; 936d2329f8SAndrea Arcangeli pgprot_t vm_page_prot; 9464e45507SPeter Feiner 956d2329f8SAndrea Arcangeli vm_page_prot = vm_pgprot_modify(vma->vm_page_prot, vm_flags); 966d2329f8SAndrea Arcangeli if (vma_wants_writenotify(vma, vm_page_prot)) { 9764e45507SPeter Feiner vm_flags &= ~VM_SHARED; 986d2329f8SAndrea Arcangeli vm_page_prot = vm_pgprot_modify(vm_page_prot, vm_flags); 9964e45507SPeter Feiner } 100c1e8d7c6SMichel Lespinasse /* remove_protection_ptes reads vma->vm_page_prot without mmap_lock */ 1016d2329f8SAndrea Arcangeli WRITE_ONCE(vma->vm_page_prot, vm_page_prot); 10264e45507SPeter Feiner } 10364e45507SPeter Feiner 1041da177e4SLinus Torvalds /* 105c8c06efaSDavidlohr Bueso * Requires inode->i_mapping->i_mmap_rwsem 1061da177e4SLinus Torvalds */ 1071da177e4SLinus Torvalds static void __remove_shared_vm_struct(struct vm_area_struct *vma, 1081da177e4SLinus Torvalds struct file *file, struct address_space *mapping) 1091da177e4SLinus Torvalds { 110e8e17ee9SLorenzo Stoakes if (vma_is_shared_maywrite(vma)) 1114bb5f5d9SDavid Herrmann mapping_unmap_writable(mapping); 1121da177e4SLinus Torvalds 1131da177e4SLinus Torvalds flush_dcache_mmap_lock(mapping); 1146b2dbba8SMichel Lespinasse vma_interval_tree_remove(vma, &mapping->i_mmap); 1151da177e4SLinus Torvalds flush_dcache_mmap_unlock(mapping); 1161da177e4SLinus Torvalds } 1171da177e4SLinus Torvalds 1181da177e4SLinus Torvalds /* 1196b2dbba8SMichel Lespinasse * Unlink a file-based vm structure from its interval tree, to hide 120a8fb5618SHugh Dickins * vma from rmap and vmtruncate before freeing its page tables. 1211da177e4SLinus Torvalds */ 122a8fb5618SHugh Dickins void unlink_file_vma(struct vm_area_struct *vma) 1231da177e4SLinus Torvalds { 1241da177e4SLinus Torvalds struct file *file = vma->vm_file; 1251da177e4SLinus Torvalds 1261da177e4SLinus Torvalds if (file) { 1271da177e4SLinus Torvalds struct address_space *mapping = file->f_mapping; 12883cde9e8SDavidlohr Bueso i_mmap_lock_write(mapping); 1291da177e4SLinus Torvalds __remove_shared_vm_struct(vma, file, mapping); 13083cde9e8SDavidlohr Bueso i_mmap_unlock_write(mapping); 1311da177e4SLinus Torvalds } 132a8fb5618SHugh Dickins } 133a8fb5618SHugh Dickins 134a8fb5618SHugh Dickins /* 135763ecb03SLiam R. Howlett * Close a vm structure and free it. 136a8fb5618SHugh Dickins */ 1370d2ebf9cSSuren Baghdasaryan static void remove_vma(struct vm_area_struct *vma, bool unreachable) 138a8fb5618SHugh Dickins { 139a8fb5618SHugh Dickins might_sleep(); 1401da177e4SLinus Torvalds if (vma->vm_ops && vma->vm_ops->close) 1411da177e4SLinus Torvalds vma->vm_ops->close(vma); 142e9714acfSKonstantin Khlebnikov if (vma->vm_file) 143a8fb5618SHugh Dickins fput(vma->vm_file); 144f0be3d32SLee Schermerhorn mpol_put(vma_policy(vma)); 1450d2ebf9cSSuren Baghdasaryan if (unreachable) 1460d2ebf9cSSuren Baghdasaryan __vm_area_free(vma); 1470d2ebf9cSSuren Baghdasaryan else 1483928d4f5SLinus Torvalds vm_area_free(vma); 1491da177e4SLinus Torvalds } 1501da177e4SLinus Torvalds 151b62b633eSLiam R. Howlett static inline struct vm_area_struct *vma_prev_limit(struct vma_iterator *vmi, 152b62b633eSLiam R. Howlett unsigned long min) 153b62b633eSLiam R. Howlett { 154b62b633eSLiam R. Howlett return mas_prev(&vmi->mas, min); 155b62b633eSLiam R. Howlett } 156b62b633eSLiam R. Howlett 1572e7ce7d3SLiam R. Howlett /* 1582e7ce7d3SLiam R. Howlett * check_brk_limits() - Use platform specific check of range & verify mlock 1592e7ce7d3SLiam R. Howlett * limits. 1602e7ce7d3SLiam R. Howlett * @addr: The address to check 1612e7ce7d3SLiam R. Howlett * @len: The size of increase. 1622e7ce7d3SLiam R. Howlett * 1632e7ce7d3SLiam R. Howlett * Return: 0 on success. 1642e7ce7d3SLiam R. Howlett */ 1652e7ce7d3SLiam R. Howlett static int check_brk_limits(unsigned long addr, unsigned long len) 1662e7ce7d3SLiam R. Howlett { 1672e7ce7d3SLiam R. Howlett unsigned long mapped_addr; 1682e7ce7d3SLiam R. Howlett 1692e7ce7d3SLiam R. Howlett mapped_addr = get_unmapped_area(NULL, addr, len, 0, MAP_FIXED); 1702e7ce7d3SLiam R. Howlett if (IS_ERR_VALUE(mapped_addr)) 1712e7ce7d3SLiam R. Howlett return mapped_addr; 1722e7ce7d3SLiam R. Howlett 173b0cc5e89SAndrew Morton return mlock_future_ok(current->mm, current->mm->def_flags, len) 1743c54a298SLorenzo Stoakes ? 0 : -EAGAIN; 1752e7ce7d3SLiam R. Howlett } 17692fed820SLiam R. Howlett static int do_brk_flags(struct vma_iterator *vmi, struct vm_area_struct *brkvma, 177763ecb03SLiam R. Howlett unsigned long addr, unsigned long request, unsigned long flags); 1786a6160a7SHeiko Carstens SYSCALL_DEFINE1(brk, unsigned long, brk) 1791da177e4SLinus Torvalds { 1809bc8039eSYang Shi unsigned long newbrk, oldbrk, origbrk; 1811da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 1822e7ce7d3SLiam R. Howlett struct vm_area_struct *brkvma, *next = NULL; 183a5b4592cSJiri Kosina unsigned long min_brk; 184408579cdSLiam R. Howlett bool populate = false; 185897ab3e0SMike Rapoport LIST_HEAD(uf); 18692fed820SLiam R. Howlett struct vma_iterator vmi; 1871da177e4SLinus Torvalds 188d8ed45c5SMichel Lespinasse if (mmap_write_lock_killable(mm)) 189dc0ef0dfSMichal Hocko return -EINTR; 1901da177e4SLinus Torvalds 1919bc8039eSYang Shi origbrk = mm->brk; 1929bc8039eSYang Shi 193a5b4592cSJiri Kosina #ifdef CONFIG_COMPAT_BRK 1945520e894SJiri Kosina /* 1955520e894SJiri Kosina * CONFIG_COMPAT_BRK can still be overridden by setting 1965520e894SJiri Kosina * randomize_va_space to 2, which will still cause mm->start_brk 1975520e894SJiri Kosina * to be arbitrarily shifted 1985520e894SJiri Kosina */ 1994471a675SJiri Kosina if (current->brk_randomized) 2005520e894SJiri Kosina min_brk = mm->start_brk; 2015520e894SJiri Kosina else 2025520e894SJiri Kosina min_brk = mm->end_data; 203a5b4592cSJiri Kosina #else 204a5b4592cSJiri Kosina min_brk = mm->start_brk; 205a5b4592cSJiri Kosina #endif 206a5b4592cSJiri Kosina if (brk < min_brk) 2071da177e4SLinus Torvalds goto out; 2081e624196SRam Gupta 2091e624196SRam Gupta /* 2101e624196SRam Gupta * Check against rlimit here. If this check is done later after the test 2111e624196SRam Gupta * of oldbrk with newbrk then it can escape the test and let the data 2121e624196SRam Gupta * segment grow beyond its set limit the in case where the limit is 2131e624196SRam Gupta * not page aligned -Ram Gupta 2141e624196SRam Gupta */ 2158764b338SCyrill Gorcunov if (check_data_rlimit(rlimit(RLIMIT_DATA), brk, mm->start_brk, 2168764b338SCyrill Gorcunov mm->end_data, mm->start_data)) 2171e624196SRam Gupta goto out; 2181e624196SRam Gupta 2191da177e4SLinus Torvalds newbrk = PAGE_ALIGN(brk); 2201da177e4SLinus Torvalds oldbrk = PAGE_ALIGN(mm->brk); 2219bc8039eSYang Shi if (oldbrk == newbrk) { 2229bc8039eSYang Shi mm->brk = brk; 2239bc8039eSYang Shi goto success; 2249bc8039eSYang Shi } 2251da177e4SLinus Torvalds 226408579cdSLiam R. Howlett /* Always allow shrinking brk. */ 2271da177e4SLinus Torvalds if (brk <= mm->brk) { 2282e7ce7d3SLiam R. Howlett /* Search one past newbrk */ 22992fed820SLiam R. Howlett vma_iter_init(&vmi, mm, newbrk); 23092fed820SLiam R. Howlett brkvma = vma_find(&vmi, oldbrk); 231f5ad5083SJason A. Donenfeld if (!brkvma || brkvma->vm_start >= oldbrk) 2322e7ce7d3SLiam R. Howlett goto out; /* mapping intersects with an existing non-brk vma. */ 2339bc8039eSYang Shi /* 2342e7ce7d3SLiam R. Howlett * mm->brk must be protected by write mmap_lock. 235408579cdSLiam R. Howlett * do_vma_munmap() will drop the lock on success, so update it 23627b26701SLiam R. Howlett * before calling do_vma_munmap(). 2379bc8039eSYang Shi */ 2389bc8039eSYang Shi mm->brk = brk; 239408579cdSLiam R. Howlett if (do_vma_munmap(&vmi, brkvma, newbrk, oldbrk, &uf, true)) 2401da177e4SLinus Torvalds goto out; 241408579cdSLiam R. Howlett 242408579cdSLiam R. Howlett goto success_unlocked; 2431da177e4SLinus Torvalds } 2441da177e4SLinus Torvalds 2452e7ce7d3SLiam R. Howlett if (check_brk_limits(oldbrk, newbrk - oldbrk)) 2462e7ce7d3SLiam R. Howlett goto out; 2472e7ce7d3SLiam R. Howlett 2482e7ce7d3SLiam R. Howlett /* 2492e7ce7d3SLiam R. Howlett * Only check if the next VMA is within the stack_guard_gap of the 2502e7ce7d3SLiam R. Howlett * expansion area 2512e7ce7d3SLiam R. Howlett */ 25292fed820SLiam R. Howlett vma_iter_init(&vmi, mm, oldbrk); 25392fed820SLiam R. Howlett next = vma_find(&vmi, newbrk + PAGE_SIZE + stack_guard_gap); 2541be7107fSHugh Dickins if (next && newbrk + PAGE_SIZE > vm_start_gap(next)) 2551da177e4SLinus Torvalds goto out; 2561da177e4SLinus Torvalds 25792fed820SLiam R. Howlett brkvma = vma_prev_limit(&vmi, mm->start_brk); 2581da177e4SLinus Torvalds /* Ok, looks good - let it rip. */ 25992fed820SLiam R. Howlett if (do_brk_flags(&vmi, brkvma, oldbrk, newbrk - oldbrk, 0) < 0) 2601da177e4SLinus Torvalds goto out; 2612e7ce7d3SLiam R. Howlett 2621da177e4SLinus Torvalds mm->brk = brk; 263408579cdSLiam R. Howlett if (mm->def_flags & VM_LOCKED) 264408579cdSLiam R. Howlett populate = true; 2659bc8039eSYang Shi 2669bc8039eSYang Shi success: 267d8ed45c5SMichel Lespinasse mmap_write_unlock(mm); 268408579cdSLiam R. Howlett success_unlocked: 269897ab3e0SMike Rapoport userfaultfd_unmap_complete(mm, &uf); 270128557ffSMichel Lespinasse if (populate) 271128557ffSMichel Lespinasse mm_populate(oldbrk, newbrk - oldbrk); 272128557ffSMichel Lespinasse return brk; 273128557ffSMichel Lespinasse 2741da177e4SLinus Torvalds out: 275408579cdSLiam R. Howlett mm->brk = origbrk; 276d8ed45c5SMichel Lespinasse mmap_write_unlock(mm); 277b7204006SAdrian Huang return origbrk; 2781da177e4SLinus Torvalds } 2791da177e4SLinus Torvalds 280d4af56c5SLiam R. Howlett #if defined(CONFIG_DEBUG_VM_MAPLE_TREE) 281eafd4dc4SRashika Kheria static void validate_mm(struct mm_struct *mm) 2821da177e4SLinus Torvalds { 2831da177e4SLinus Torvalds int bug = 0; 2841da177e4SLinus Torvalds int i = 0; 285763ecb03SLiam R. Howlett struct vm_area_struct *vma; 286b50e195fSLiam R. Howlett VMA_ITERATOR(vmi, mm, 0); 287ff26f70fSAndrew Morton 288b50e195fSLiam R. Howlett mt_validate(&mm->mm_mt); 289b50e195fSLiam R. Howlett for_each_vma(vmi, vma) { 290524e00b3SLiam R. Howlett #ifdef CONFIG_DEBUG_VM_RB 29112352d3cSKonstantin Khlebnikov struct anon_vma *anon_vma = vma->anon_vma; 292ed8ea815SMichel Lespinasse struct anon_vma_chain *avc; 293b50e195fSLiam R. Howlett #endif 294b50e195fSLiam R. Howlett unsigned long vmi_start, vmi_end; 295b50e195fSLiam R. Howlett bool warn = 0; 296ff26f70fSAndrew Morton 297b50e195fSLiam R. Howlett vmi_start = vma_iter_addr(&vmi); 298b50e195fSLiam R. Howlett vmi_end = vma_iter_end(&vmi); 299b50e195fSLiam R. Howlett if (VM_WARN_ON_ONCE_MM(vma->vm_end != vmi_end, mm)) 300b50e195fSLiam R. Howlett warn = 1; 301b50e195fSLiam R. Howlett 302b50e195fSLiam R. Howlett if (VM_WARN_ON_ONCE_MM(vma->vm_start != vmi_start, mm)) 303b50e195fSLiam R. Howlett warn = 1; 304b50e195fSLiam R. Howlett 305b50e195fSLiam R. Howlett if (warn) { 306b50e195fSLiam R. Howlett pr_emerg("issue in %s\n", current->comm); 307b50e195fSLiam R. Howlett dump_stack(); 308b50e195fSLiam R. Howlett dump_vma(vma); 309b50e195fSLiam R. Howlett pr_emerg("tree range: %px start %lx end %lx\n", vma, 310b50e195fSLiam R. Howlett vmi_start, vmi_end - 1); 311b50e195fSLiam R. Howlett vma_iter_dump_tree(&vmi); 312b50e195fSLiam R. Howlett } 313b50e195fSLiam R. Howlett 314b50e195fSLiam R. Howlett #ifdef CONFIG_DEBUG_VM_RB 31512352d3cSKonstantin Khlebnikov if (anon_vma) { 31612352d3cSKonstantin Khlebnikov anon_vma_lock_read(anon_vma); 317ed8ea815SMichel Lespinasse list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) 318ed8ea815SMichel Lespinasse anon_vma_interval_tree_verify(avc); 31912352d3cSKonstantin Khlebnikov anon_vma_unlock_read(anon_vma); 32012352d3cSKonstantin Khlebnikov } 321524e00b3SLiam R. Howlett #endif 3221da177e4SLinus Torvalds i++; 3231da177e4SLinus Torvalds } 3245a0768f6SMichel Lespinasse if (i != mm->map_count) { 325b50e195fSLiam R. Howlett pr_emerg("map_count %d vma iterator %d\n", mm->map_count, i); 3265a0768f6SMichel Lespinasse bug = 1; 3275a0768f6SMichel Lespinasse } 32896dad67fSSasha Levin VM_BUG_ON_MM(bug, mm); 3291da177e4SLinus Torvalds } 330524e00b3SLiam R. Howlett 331524e00b3SLiam R. Howlett #else /* !CONFIG_DEBUG_VM_MAPLE_TREE */ 3321da177e4SLinus Torvalds #define validate_mm(mm) do { } while (0) 333524e00b3SLiam R. Howlett #endif /* CONFIG_DEBUG_VM_MAPLE_TREE */ 334d3737187SMichel Lespinasse 335bf181b9fSMichel Lespinasse /* 336bf181b9fSMichel Lespinasse * vma has some anon_vma assigned, and is already inserted on that 337bf181b9fSMichel Lespinasse * anon_vma's interval trees. 338bf181b9fSMichel Lespinasse * 339bf181b9fSMichel Lespinasse * Before updating the vma's vm_start / vm_end / vm_pgoff fields, the 340bf181b9fSMichel Lespinasse * vma must be removed from the anon_vma's interval trees using 341bf181b9fSMichel Lespinasse * anon_vma_interval_tree_pre_update_vma(). 342bf181b9fSMichel Lespinasse * 343bf181b9fSMichel Lespinasse * After the update, the vma will be reinserted using 344bf181b9fSMichel Lespinasse * anon_vma_interval_tree_post_update_vma(). 345bf181b9fSMichel Lespinasse * 346c1e8d7c6SMichel Lespinasse * The entire update must be protected by exclusive mmap_lock and by 347bf181b9fSMichel Lespinasse * the root anon_vma's mutex. 348bf181b9fSMichel Lespinasse */ 349bf181b9fSMichel Lespinasse static inline void 350bf181b9fSMichel Lespinasse anon_vma_interval_tree_pre_update_vma(struct vm_area_struct *vma) 351bf181b9fSMichel Lespinasse { 352bf181b9fSMichel Lespinasse struct anon_vma_chain *avc; 353bf181b9fSMichel Lespinasse 354bf181b9fSMichel Lespinasse list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) 355bf181b9fSMichel Lespinasse anon_vma_interval_tree_remove(avc, &avc->anon_vma->rb_root); 356bf181b9fSMichel Lespinasse } 357bf181b9fSMichel Lespinasse 358bf181b9fSMichel Lespinasse static inline void 359bf181b9fSMichel Lespinasse anon_vma_interval_tree_post_update_vma(struct vm_area_struct *vma) 360bf181b9fSMichel Lespinasse { 361bf181b9fSMichel Lespinasse struct anon_vma_chain *avc; 362bf181b9fSMichel Lespinasse 363bf181b9fSMichel Lespinasse list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) 364bf181b9fSMichel Lespinasse anon_vma_interval_tree_insert(avc, &avc->anon_vma->rb_root); 365bf181b9fSMichel Lespinasse } 366bf181b9fSMichel Lespinasse 367e8420a8eSCyril Hrubis static unsigned long count_vma_pages_range(struct mm_struct *mm, 368e8420a8eSCyril Hrubis unsigned long addr, unsigned long end) 369e8420a8eSCyril Hrubis { 3702e3af1dbSMatthew Wilcox (Oracle) VMA_ITERATOR(vmi, mm, addr); 371e8420a8eSCyril Hrubis struct vm_area_struct *vma; 3722e3af1dbSMatthew Wilcox (Oracle) unsigned long nr_pages = 0; 373e8420a8eSCyril Hrubis 3742e3af1dbSMatthew Wilcox (Oracle) for_each_vma_range(vmi, vma, end) { 3752e3af1dbSMatthew Wilcox (Oracle) unsigned long vm_start = max(addr, vma->vm_start); 3762e3af1dbSMatthew Wilcox (Oracle) unsigned long vm_end = min(end, vma->vm_end); 377e8420a8eSCyril Hrubis 3782e3af1dbSMatthew Wilcox (Oracle) nr_pages += PHYS_PFN(vm_end - vm_start); 379e8420a8eSCyril Hrubis } 380e8420a8eSCyril Hrubis 381e8420a8eSCyril Hrubis return nr_pages; 382e8420a8eSCyril Hrubis } 383e8420a8eSCyril Hrubis 384c154124fSLiam R. Howlett static void __vma_link_file(struct vm_area_struct *vma, 385c154124fSLiam R. Howlett struct address_space *mapping) 3861da177e4SLinus Torvalds { 387e8e17ee9SLorenzo Stoakes if (vma_is_shared_maywrite(vma)) 388cf508b58SMiaohe Lin mapping_allow_writable(mapping); 3891da177e4SLinus Torvalds 3901da177e4SLinus Torvalds flush_dcache_mmap_lock(mapping); 3916b2dbba8SMichel Lespinasse vma_interval_tree_insert(vma, &mapping->i_mmap); 3921da177e4SLinus Torvalds flush_dcache_mmap_unlock(mapping); 3931da177e4SLinus Torvalds } 3941da177e4SLinus Torvalds 395763ecb03SLiam R. Howlett static int vma_link(struct mm_struct *mm, struct vm_area_struct *vma) 3961da177e4SLinus Torvalds { 39779e4f2caSLiam R. Howlett VMA_ITERATOR(vmi, mm, 0); 3981da177e4SLinus Torvalds struct address_space *mapping = NULL; 3991da177e4SLinus Torvalds 400b5df0922SLiam R. Howlett vma_iter_config(&vmi, vma->vm_start, vma->vm_end); 401b5df0922SLiam R. Howlett if (vma_iter_prealloc(&vmi, vma)) 402d4af56c5SLiam R. Howlett return -ENOMEM; 403d4af56c5SLiam R. Howlett 404ad9f0063SSuren Baghdasaryan vma_start_write(vma); 405ad9f0063SSuren Baghdasaryan 4066852c46cSYu Ma vma_iter_store(&vmi, vma); 4076852c46cSYu Ma 40864ac4940SHuang Shijie if (vma->vm_file) { 4091da177e4SLinus Torvalds mapping = vma->vm_file->f_mapping; 41083cde9e8SDavidlohr Bueso i_mmap_lock_write(mapping); 411c154124fSLiam R. Howlett __vma_link_file(vma, mapping); 41283cde9e8SDavidlohr Bueso i_mmap_unlock_write(mapping); 413c154124fSLiam R. Howlett } 4141da177e4SLinus Torvalds 4151da177e4SLinus Torvalds mm->map_count++; 4161da177e4SLinus Torvalds validate_mm(mm); 417d4af56c5SLiam R. Howlett return 0; 4181da177e4SLinus Torvalds } 4191da177e4SLinus Torvalds 4201da177e4SLinus Torvalds /* 42168cefec5SLiam R. Howlett * init_multi_vma_prep() - Initializer for struct vma_prepare 42268cefec5SLiam R. Howlett * @vp: The vma_prepare struct 42368cefec5SLiam R. Howlett * @vma: The vma that will be altered once locked 42468cefec5SLiam R. Howlett * @next: The next vma if it is to be adjusted 42568cefec5SLiam R. Howlett * @remove: The first vma to be removed 42668cefec5SLiam R. Howlett * @remove2: The second vma to be removed 42768cefec5SLiam R. Howlett */ 42868cefec5SLiam R. Howlett static inline void init_multi_vma_prep(struct vma_prepare *vp, 42968cefec5SLiam R. Howlett struct vm_area_struct *vma, struct vm_area_struct *next, 43068cefec5SLiam R. Howlett struct vm_area_struct *remove, struct vm_area_struct *remove2) 43168cefec5SLiam R. Howlett { 43268cefec5SLiam R. Howlett memset(vp, 0, sizeof(struct vma_prepare)); 43368cefec5SLiam R. Howlett vp->vma = vma; 43468cefec5SLiam R. Howlett vp->anon_vma = vma->anon_vma; 43568cefec5SLiam R. Howlett vp->remove = remove; 43668cefec5SLiam R. Howlett vp->remove2 = remove2; 43768cefec5SLiam R. Howlett vp->adj_next = next; 43868cefec5SLiam R. Howlett if (!vp->anon_vma && next) 43968cefec5SLiam R. Howlett vp->anon_vma = next->anon_vma; 44068cefec5SLiam R. Howlett 44168cefec5SLiam R. Howlett vp->file = vma->vm_file; 44268cefec5SLiam R. Howlett if (vp->file) 44368cefec5SLiam R. Howlett vp->mapping = vma->vm_file->f_mapping; 44468cefec5SLiam R. Howlett 44568cefec5SLiam R. Howlett } 44668cefec5SLiam R. Howlett 44768cefec5SLiam R. Howlett /* 44868cefec5SLiam R. Howlett * init_vma_prep() - Initializer wrapper for vma_prepare struct 44968cefec5SLiam R. Howlett * @vp: The vma_prepare struct 45068cefec5SLiam R. Howlett * @vma: The vma that will be altered once locked 45168cefec5SLiam R. Howlett */ 45268cefec5SLiam R. Howlett static inline void init_vma_prep(struct vma_prepare *vp, 45368cefec5SLiam R. Howlett struct vm_area_struct *vma) 45468cefec5SLiam R. Howlett { 45568cefec5SLiam R. Howlett init_multi_vma_prep(vp, vma, NULL, NULL, NULL); 45668cefec5SLiam R. Howlett } 45768cefec5SLiam R. Howlett 45868cefec5SLiam R. Howlett 45968cefec5SLiam R. Howlett /* 460440703e0SLiam R. Howlett * vma_prepare() - Helper function for handling locking VMAs prior to altering 461440703e0SLiam R. Howlett * @vp: The initialized vma_prepare struct 462440703e0SLiam R. Howlett */ 463440703e0SLiam R. Howlett static inline void vma_prepare(struct vma_prepare *vp) 464440703e0SLiam R. Howlett { 465440703e0SLiam R. Howlett if (vp->file) { 466440703e0SLiam R. Howlett uprobe_munmap(vp->vma, vp->vma->vm_start, vp->vma->vm_end); 467440703e0SLiam R. Howlett 468440703e0SLiam R. Howlett if (vp->adj_next) 469440703e0SLiam R. Howlett uprobe_munmap(vp->adj_next, vp->adj_next->vm_start, 470440703e0SLiam R. Howlett vp->adj_next->vm_end); 471440703e0SLiam R. Howlett 472440703e0SLiam R. Howlett i_mmap_lock_write(vp->mapping); 473440703e0SLiam R. Howlett if (vp->insert && vp->insert->vm_file) { 474440703e0SLiam R. Howlett /* 475440703e0SLiam R. Howlett * Put into interval tree now, so instantiated pages 476440703e0SLiam R. Howlett * are visible to arm/parisc __flush_dcache_page 477440703e0SLiam R. Howlett * throughout; but we cannot insert into address 478440703e0SLiam R. Howlett * space until vma start or end is updated. 479440703e0SLiam R. Howlett */ 480440703e0SLiam R. Howlett __vma_link_file(vp->insert, 481440703e0SLiam R. Howlett vp->insert->vm_file->f_mapping); 482440703e0SLiam R. Howlett } 483440703e0SLiam R. Howlett } 484440703e0SLiam R. Howlett 485440703e0SLiam R. Howlett if (vp->anon_vma) { 486440703e0SLiam R. Howlett anon_vma_lock_write(vp->anon_vma); 487440703e0SLiam R. Howlett anon_vma_interval_tree_pre_update_vma(vp->vma); 488440703e0SLiam R. Howlett if (vp->adj_next) 489440703e0SLiam R. Howlett anon_vma_interval_tree_pre_update_vma(vp->adj_next); 490440703e0SLiam R. Howlett } 491440703e0SLiam R. Howlett 492440703e0SLiam R. Howlett if (vp->file) { 493440703e0SLiam R. Howlett flush_dcache_mmap_lock(vp->mapping); 494440703e0SLiam R. Howlett vma_interval_tree_remove(vp->vma, &vp->mapping->i_mmap); 495440703e0SLiam R. Howlett if (vp->adj_next) 496440703e0SLiam R. Howlett vma_interval_tree_remove(vp->adj_next, 497440703e0SLiam R. Howlett &vp->mapping->i_mmap); 498440703e0SLiam R. Howlett } 499440703e0SLiam R. Howlett 500440703e0SLiam R. Howlett } 501440703e0SLiam R. Howlett 502440703e0SLiam R. Howlett /* 503440703e0SLiam R. Howlett * vma_complete- Helper function for handling the unlocking after altering VMAs, 504440703e0SLiam R. Howlett * or for inserting a VMA. 505440703e0SLiam R. Howlett * 506440703e0SLiam R. Howlett * @vp: The vma_prepare struct 507440703e0SLiam R. Howlett * @vmi: The vma iterator 508440703e0SLiam R. Howlett * @mm: The mm_struct 509440703e0SLiam R. Howlett */ 510440703e0SLiam R. Howlett static inline void vma_complete(struct vma_prepare *vp, 511440703e0SLiam R. Howlett struct vma_iterator *vmi, struct mm_struct *mm) 512440703e0SLiam R. Howlett { 513440703e0SLiam R. Howlett if (vp->file) { 514440703e0SLiam R. Howlett if (vp->adj_next) 515440703e0SLiam R. Howlett vma_interval_tree_insert(vp->adj_next, 516440703e0SLiam R. Howlett &vp->mapping->i_mmap); 517440703e0SLiam R. Howlett vma_interval_tree_insert(vp->vma, &vp->mapping->i_mmap); 518440703e0SLiam R. Howlett flush_dcache_mmap_unlock(vp->mapping); 519440703e0SLiam R. Howlett } 520440703e0SLiam R. Howlett 521440703e0SLiam R. Howlett if (vp->remove && vp->file) { 522440703e0SLiam R. Howlett __remove_shared_vm_struct(vp->remove, vp->file, vp->mapping); 523440703e0SLiam R. Howlett if (vp->remove2) 524440703e0SLiam R. Howlett __remove_shared_vm_struct(vp->remove2, vp->file, 525440703e0SLiam R. Howlett vp->mapping); 526440703e0SLiam R. Howlett } else if (vp->insert) { 527440703e0SLiam R. Howlett /* 528440703e0SLiam R. Howlett * split_vma has split insert from vma, and needs 529440703e0SLiam R. Howlett * us to insert it before dropping the locks 530440703e0SLiam R. Howlett * (it may either follow vma or precede it). 531440703e0SLiam R. Howlett */ 532440703e0SLiam R. Howlett vma_iter_store(vmi, vp->insert); 533440703e0SLiam R. Howlett mm->map_count++; 534440703e0SLiam R. Howlett } 535440703e0SLiam R. Howlett 536440703e0SLiam R. Howlett if (vp->anon_vma) { 537440703e0SLiam R. Howlett anon_vma_interval_tree_post_update_vma(vp->vma); 538440703e0SLiam R. Howlett if (vp->adj_next) 539440703e0SLiam R. Howlett anon_vma_interval_tree_post_update_vma(vp->adj_next); 540440703e0SLiam R. Howlett anon_vma_unlock_write(vp->anon_vma); 541440703e0SLiam R. Howlett } 542440703e0SLiam R. Howlett 543440703e0SLiam R. Howlett if (vp->file) { 544440703e0SLiam R. Howlett i_mmap_unlock_write(vp->mapping); 545440703e0SLiam R. Howlett uprobe_mmap(vp->vma); 546440703e0SLiam R. Howlett 547440703e0SLiam R. Howlett if (vp->adj_next) 548440703e0SLiam R. Howlett uprobe_mmap(vp->adj_next); 549440703e0SLiam R. Howlett } 550440703e0SLiam R. Howlett 551440703e0SLiam R. Howlett if (vp->remove) { 552440703e0SLiam R. Howlett again: 553457f67beSSuren Baghdasaryan vma_mark_detached(vp->remove, true); 554440703e0SLiam R. Howlett if (vp->file) { 555440703e0SLiam R. Howlett uprobe_munmap(vp->remove, vp->remove->vm_start, 556440703e0SLiam R. Howlett vp->remove->vm_end); 557440703e0SLiam R. Howlett fput(vp->file); 558440703e0SLiam R. Howlett } 559440703e0SLiam R. Howlett if (vp->remove->anon_vma) 560440703e0SLiam R. Howlett anon_vma_merge(vp->vma, vp->remove); 561440703e0SLiam R. Howlett mm->map_count--; 562440703e0SLiam R. Howlett mpol_put(vma_policy(vp->remove)); 563440703e0SLiam R. Howlett if (!vp->remove2) 564440703e0SLiam R. Howlett WARN_ON_ONCE(vp->vma->vm_end < vp->remove->vm_end); 565440703e0SLiam R. Howlett vm_area_free(vp->remove); 566440703e0SLiam R. Howlett 567440703e0SLiam R. Howlett /* 568440703e0SLiam R. Howlett * In mprotect's case 6 (see comments on vma_merge), 5695ff783f1SVlastimil Babka * we are removing both mid and next vmas 570440703e0SLiam R. Howlett */ 571440703e0SLiam R. Howlett if (vp->remove2) { 572440703e0SLiam R. Howlett vp->remove = vp->remove2; 573440703e0SLiam R. Howlett vp->remove2 = NULL; 574440703e0SLiam R. Howlett goto again; 575440703e0SLiam R. Howlett } 576440703e0SLiam R. Howlett } 577440703e0SLiam R. Howlett if (vp->insert && vp->file) 578440703e0SLiam R. Howlett uprobe_mmap(vp->insert); 5792574d5e4SLiam R. Howlett validate_mm(mm); 580440703e0SLiam R. Howlett } 581440703e0SLiam R. Howlett 582440703e0SLiam R. Howlett /* 58304241ffeSLiam R. Howlett * dup_anon_vma() - Helper function to duplicate anon_vma 58404241ffeSLiam R. Howlett * @dst: The destination VMA 58504241ffeSLiam R. Howlett * @src: The source VMA 586824135c4SLiam R. Howlett * @dup: Pointer to the destination VMA when successful. 58704241ffeSLiam R. Howlett * 58804241ffeSLiam R. Howlett * Returns: 0 on success. 58904241ffeSLiam R. Howlett */ 59004241ffeSLiam R. Howlett static inline int dup_anon_vma(struct vm_area_struct *dst, 591824135c4SLiam R. Howlett struct vm_area_struct *src, struct vm_area_struct **dup) 59204241ffeSLiam R. Howlett { 59304241ffeSLiam R. Howlett /* 59404241ffeSLiam R. Howlett * Easily overlooked: when mprotect shifts the boundary, make sure the 59504241ffeSLiam R. Howlett * expanding vma has anon_vma set if the shrinking vma had, to cover any 59604241ffeSLiam R. Howlett * anon pages imported. 59704241ffeSLiam R. Howlett */ 59804241ffeSLiam R. Howlett if (src->anon_vma && !dst->anon_vma) { 599824135c4SLiam R. Howlett int ret; 600824135c4SLiam R. Howlett 601c9d6e982SSuren Baghdasaryan vma_assert_write_locked(dst); 60204241ffeSLiam R. Howlett dst->anon_vma = src->anon_vma; 603824135c4SLiam R. Howlett ret = anon_vma_clone(dst, src); 604824135c4SLiam R. Howlett if (ret) 605824135c4SLiam R. Howlett return ret; 606824135c4SLiam R. Howlett 607824135c4SLiam R. Howlett *dup = dst; 60804241ffeSLiam R. Howlett } 60904241ffeSLiam R. Howlett 61004241ffeSLiam R. Howlett return 0; 61104241ffeSLiam R. Howlett } 61204241ffeSLiam R. Howlett 61304241ffeSLiam R. Howlett /* 6149303d3e1SLiam R. Howlett * vma_expand - Expand an existing VMA 6159303d3e1SLiam R. Howlett * 6169303d3e1SLiam R. Howlett * @vmi: The vma iterator 6179303d3e1SLiam R. Howlett * @vma: The vma to expand 6189303d3e1SLiam R. Howlett * @start: The start of the vma 6199303d3e1SLiam R. Howlett * @end: The exclusive end of the vma 6209303d3e1SLiam R. Howlett * @pgoff: The page offset of vma 6219303d3e1SLiam R. Howlett * @next: The current of next vma. 6229303d3e1SLiam R. Howlett * 6239303d3e1SLiam R. Howlett * Expand @vma to @start and @end. Can expand off the start and end. Will 6249303d3e1SLiam R. Howlett * expand over @next if it's different from @vma and @end == @next->vm_end. 6259303d3e1SLiam R. Howlett * Checking if the @vma can expand and merge with @next needs to be handled by 6269303d3e1SLiam R. Howlett * the caller. 6279303d3e1SLiam R. Howlett * 6289303d3e1SLiam R. Howlett * Returns: 0 on success 6299303d3e1SLiam R. Howlett */ 6307c9813e8SLiam R. Howlett int vma_expand(struct vma_iterator *vmi, struct vm_area_struct *vma, 6319303d3e1SLiam R. Howlett unsigned long start, unsigned long end, pgoff_t pgoff, 6329303d3e1SLiam R. Howlett struct vm_area_struct *next) 6339303d3e1SLiam R. Howlett { 634824135c4SLiam R. Howlett struct vm_area_struct *anon_dup = NULL; 63568cefec5SLiam R. Howlett bool remove_next = false; 6369303d3e1SLiam R. Howlett struct vma_prepare vp; 6379303d3e1SLiam R. Howlett 638c9d6e982SSuren Baghdasaryan vma_start_write(vma); 6399303d3e1SLiam R. Howlett if (next && (vma != next) && (end == next->vm_end)) { 64004241ffeSLiam R. Howlett int ret; 6419303d3e1SLiam R. Howlett 64204241ffeSLiam R. Howlett remove_next = true; 643c9d6e982SSuren Baghdasaryan vma_start_write(next); 644824135c4SLiam R. Howlett ret = dup_anon_vma(vma, next, &anon_dup); 64504241ffeSLiam R. Howlett if (ret) 64604241ffeSLiam R. Howlett return ret; 6479303d3e1SLiam R. Howlett } 6489303d3e1SLiam R. Howlett 64968cefec5SLiam R. Howlett init_multi_vma_prep(&vp, vma, NULL, remove_next ? next : NULL, NULL); 6509303d3e1SLiam R. Howlett /* Not merging but overwriting any part of next is not handled. */ 6519303d3e1SLiam R. Howlett VM_WARN_ON(next && !vp.remove && 6529303d3e1SLiam R. Howlett next != vma && end > next->vm_start); 6539303d3e1SLiam R. Howlett /* Only handles expanding */ 6549303d3e1SLiam R. Howlett VM_WARN_ON(vma->vm_start < start || vma->vm_end > end); 6559303d3e1SLiam R. Howlett 656b5df0922SLiam R. Howlett /* Note: vma iterator must be pointing to 'start' */ 657b5df0922SLiam R. Howlett vma_iter_config(vmi, start, end); 658b5df0922SLiam R. Howlett if (vma_iter_prealloc(vmi, vma)) 6599303d3e1SLiam R. Howlett goto nomem; 6609303d3e1SLiam R. Howlett 661ccf1d78dSSuren Baghdasaryan vma_prepare(&vp); 6629303d3e1SLiam R. Howlett vma_adjust_trans_huge(vma, start, end, 0); 6639303d3e1SLiam R. Howlett vma->vm_start = start; 6649303d3e1SLiam R. Howlett vma->vm_end = end; 6659303d3e1SLiam R. Howlett vma->vm_pgoff = pgoff; 6669303d3e1SLiam R. Howlett vma_iter_store(vmi, vma); 6679303d3e1SLiam R. Howlett 6689303d3e1SLiam R. Howlett vma_complete(&vp, vmi, vma->vm_mm); 6699303d3e1SLiam R. Howlett return 0; 6709303d3e1SLiam R. Howlett 6719303d3e1SLiam R. Howlett nomem: 672824135c4SLiam R. Howlett if (anon_dup) 673824135c4SLiam R. Howlett unlink_anon_vmas(anon_dup); 6749303d3e1SLiam R. Howlett return -ENOMEM; 6759303d3e1SLiam R. Howlett } 676cf51e86dSLiam R. Howlett 677cf51e86dSLiam R. Howlett /* 678cf51e86dSLiam R. Howlett * vma_shrink() - Reduce an existing VMAs memory area 679cf51e86dSLiam R. Howlett * @vmi: The vma iterator 680cf51e86dSLiam R. Howlett * @vma: The VMA to modify 681cf51e86dSLiam R. Howlett * @start: The new start 682cf51e86dSLiam R. Howlett * @end: The new end 683cf51e86dSLiam R. Howlett * 684cf51e86dSLiam R. Howlett * Returns: 0 on success, -ENOMEM otherwise 685cf51e86dSLiam R. Howlett */ 686cf51e86dSLiam R. Howlett int vma_shrink(struct vma_iterator *vmi, struct vm_area_struct *vma, 687cf51e86dSLiam R. Howlett unsigned long start, unsigned long end, pgoff_t pgoff) 688cf51e86dSLiam R. Howlett { 689cf51e86dSLiam R. Howlett struct vma_prepare vp; 690cf51e86dSLiam R. Howlett 691cf51e86dSLiam R. Howlett WARN_ON((vma->vm_start != start) && (vma->vm_end != end)); 692cf51e86dSLiam R. Howlett 693b5df0922SLiam R. Howlett if (vma->vm_start < start) 694b5df0922SLiam R. Howlett vma_iter_config(vmi, vma->vm_start, start); 695b5df0922SLiam R. Howlett else 696b5df0922SLiam R. Howlett vma_iter_config(vmi, end, vma->vm_end); 697b5df0922SLiam R. Howlett 698b5df0922SLiam R. Howlett if (vma_iter_prealloc(vmi, NULL)) 699cf51e86dSLiam R. Howlett return -ENOMEM; 700cf51e86dSLiam R. Howlett 701c9d6e982SSuren Baghdasaryan vma_start_write(vma); 702c9d6e982SSuren Baghdasaryan 703cf51e86dSLiam R. Howlett init_vma_prep(&vp, vma); 704cf51e86dSLiam R. Howlett vma_prepare(&vp); 705ccf1d78dSSuren Baghdasaryan vma_adjust_trans_huge(vma, start, end, 0); 706cf51e86dSLiam R. Howlett 707b5df0922SLiam R. Howlett vma_iter_clear(vmi); 708cf51e86dSLiam R. Howlett vma->vm_start = start; 709cf51e86dSLiam R. Howlett vma->vm_end = end; 710cf51e86dSLiam R. Howlett vma->vm_pgoff = pgoff; 711cf51e86dSLiam R. Howlett vma_complete(&vp, vmi, vma->vm_mm); 712cf51e86dSLiam R. Howlett return 0; 713cf51e86dSLiam R. Howlett } 714cf51e86dSLiam R. Howlett 7159303d3e1SLiam R. Howlett /* 7161da177e4SLinus Torvalds * If the vma has a ->close operation then the driver probably needs to release 717714965caSVlastimil Babka * per-vma resources, so we don't attempt to merge those if the caller indicates 718714965caSVlastimil Babka * the current vma may be removed as part of the merge. 7191da177e4SLinus Torvalds */ 7202dbf4010SVlastimil Babka static inline bool is_mergeable_vma(struct vm_area_struct *vma, 72119a809afSAndrea Arcangeli struct file *file, unsigned long vm_flags, 7229a10064fSColin Cross struct vm_userfaultfd_ctx vm_userfaultfd_ctx, 723714965caSVlastimil Babka struct anon_vma_name *anon_name, bool may_remove_vma) 7241da177e4SLinus Torvalds { 72534228d47SCyrill Gorcunov /* 72634228d47SCyrill Gorcunov * VM_SOFTDIRTY should not prevent from VMA merging, if we 72734228d47SCyrill Gorcunov * match the flags but dirty bit -- the caller should mark 72834228d47SCyrill Gorcunov * merged VMA as dirty. If dirty bit won't be excluded from 7298bb4e7a2SWei Yang * comparison, we increase pressure on the memory system forcing 73034228d47SCyrill Gorcunov * the kernel to generate new VMAs when old one could be 73134228d47SCyrill Gorcunov * extended instead. 73234228d47SCyrill Gorcunov */ 73334228d47SCyrill Gorcunov if ((vma->vm_flags ^ vm_flags) & ~VM_SOFTDIRTY) 7342dbf4010SVlastimil Babka return false; 7351da177e4SLinus Torvalds if (vma->vm_file != file) 7362dbf4010SVlastimil Babka return false; 737714965caSVlastimil Babka if (may_remove_vma && vma->vm_ops && vma->vm_ops->close) 7382dbf4010SVlastimil Babka return false; 73919a809afSAndrea Arcangeli if (!is_mergeable_vm_userfaultfd_ctx(vma, vm_userfaultfd_ctx)) 7402dbf4010SVlastimil Babka return false; 7415c26f6acSSuren Baghdasaryan if (!anon_vma_name_eq(anon_vma_name(vma), anon_name)) 7422dbf4010SVlastimil Babka return false; 7432dbf4010SVlastimil Babka return true; 7441da177e4SLinus Torvalds } 7451da177e4SLinus Torvalds 7462dbf4010SVlastimil Babka static inline bool is_mergeable_anon_vma(struct anon_vma *anon_vma1, 7472dbf4010SVlastimil Babka struct anon_vma *anon_vma2, struct vm_area_struct *vma) 7481da177e4SLinus Torvalds { 749965f55deSShaohua Li /* 750965f55deSShaohua Li * The list_is_singular() test is to avoid merging VMA cloned from 751965f55deSShaohua Li * parents. This can improve scalability caused by anon_vma lock. 752965f55deSShaohua Li */ 753965f55deSShaohua Li if ((!anon_vma1 || !anon_vma2) && (!vma || 754965f55deSShaohua Li list_is_singular(&vma->anon_vma_chain))) 7552dbf4010SVlastimil Babka return true; 756965f55deSShaohua Li return anon_vma1 == anon_vma2; 7571da177e4SLinus Torvalds } 7581da177e4SLinus Torvalds 7591da177e4SLinus Torvalds /* 7601da177e4SLinus Torvalds * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff) 7611da177e4SLinus Torvalds * in front of (at a lower virtual address and file offset than) the vma. 7621da177e4SLinus Torvalds * 7631da177e4SLinus Torvalds * We cannot merge two vmas if they have differently assigned (non-NULL) 7641da177e4SLinus Torvalds * anon_vmas, nor if same anon_vma is assigned but offsets incompatible. 7651da177e4SLinus Torvalds * 7661da177e4SLinus Torvalds * We don't check here for the merged mmap wrapping around the end of pagecache 76745e55300SPeter Collingbourne * indices (16TB on ia32) because do_mmap() does not permit mmap's which 7681da177e4SLinus Torvalds * wrap, nor mmaps which cover the final page at index -1UL. 769714965caSVlastimil Babka * 770714965caSVlastimil Babka * We assume the vma may be removed as part of the merge. 7711da177e4SLinus Torvalds */ 7722dbf4010SVlastimil Babka static bool 7731da177e4SLinus Torvalds can_vma_merge_before(struct vm_area_struct *vma, unsigned long vm_flags, 77419a809afSAndrea Arcangeli struct anon_vma *anon_vma, struct file *file, 7752dbf4010SVlastimil Babka pgoff_t vm_pgoff, struct vm_userfaultfd_ctx vm_userfaultfd_ctx, 7765c26f6acSSuren Baghdasaryan struct anon_vma_name *anon_name) 7771da177e4SLinus Torvalds { 778714965caSVlastimil Babka if (is_mergeable_vma(vma, file, vm_flags, vm_userfaultfd_ctx, anon_name, true) && 779965f55deSShaohua Li is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) { 7801da177e4SLinus Torvalds if (vma->vm_pgoff == vm_pgoff) 7812dbf4010SVlastimil Babka return true; 7821da177e4SLinus Torvalds } 7832dbf4010SVlastimil Babka return false; 7841da177e4SLinus Torvalds } 7851da177e4SLinus Torvalds 7861da177e4SLinus Torvalds /* 7871da177e4SLinus Torvalds * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff) 7881da177e4SLinus Torvalds * beyond (at a higher virtual address and file offset than) the vma. 7891da177e4SLinus Torvalds * 7901da177e4SLinus Torvalds * We cannot merge two vmas if they have differently assigned (non-NULL) 7911da177e4SLinus Torvalds * anon_vmas, nor if same anon_vma is assigned but offsets incompatible. 792714965caSVlastimil Babka * 793714965caSVlastimil Babka * We assume that vma is not removed as part of the merge. 7941da177e4SLinus Torvalds */ 7952dbf4010SVlastimil Babka static bool 7961da177e4SLinus Torvalds can_vma_merge_after(struct vm_area_struct *vma, unsigned long vm_flags, 79719a809afSAndrea Arcangeli struct anon_vma *anon_vma, struct file *file, 7982dbf4010SVlastimil Babka pgoff_t vm_pgoff, struct vm_userfaultfd_ctx vm_userfaultfd_ctx, 7995c26f6acSSuren Baghdasaryan struct anon_vma_name *anon_name) 8001da177e4SLinus Torvalds { 801714965caSVlastimil Babka if (is_mergeable_vma(vma, file, vm_flags, vm_userfaultfd_ctx, anon_name, false) && 802965f55deSShaohua Li is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) { 8031da177e4SLinus Torvalds pgoff_t vm_pglen; 804d6e93217SLibin vm_pglen = vma_pages(vma); 8051da177e4SLinus Torvalds if (vma->vm_pgoff + vm_pglen == vm_pgoff) 8062dbf4010SVlastimil Babka return true; 8071da177e4SLinus Torvalds } 8082dbf4010SVlastimil Babka return false; 8091da177e4SLinus Torvalds } 8101da177e4SLinus Torvalds 8111da177e4SLinus Torvalds /* 8129a10064fSColin Cross * Given a mapping request (addr,end,vm_flags,file,pgoff,anon_name), 8139a10064fSColin Cross * figure out whether that can be merged with its predecessor or its 8149a10064fSColin Cross * successor. Or both (it neatly fills a hole). 8151da177e4SLinus Torvalds * 8161da177e4SLinus Torvalds * In most cases - when called for mmap, brk or mremap - [addr,end) is 8171da177e4SLinus Torvalds * certain not to be mapped by the time vma_merge is called; but when 8181da177e4SLinus Torvalds * called for mprotect, it is certain to be already mapped (either at 8191da177e4SLinus Torvalds * an offset within prev, or at the start of next), and the flags of 8201da177e4SLinus Torvalds * this area are about to be changed to vm_flags - and the no-change 8211da177e4SLinus Torvalds * case has already been eliminated. 8221da177e4SLinus Torvalds * 823fcfccd91SLorenzo Stoakes * The following mprotect cases have to be considered, where **** is 8241da177e4SLinus Torvalds * the area passed down from mprotect_fixup, never extending beyond one 825fcfccd91SLorenzo Stoakes * vma, PPPP is the previous vma, CCCC is a concurrent vma that starts 826fcfccd91SLorenzo Stoakes * at the same address as **** and is of the same or larger span, and 827fcfccd91SLorenzo Stoakes * NNNN the next vma after ****: 8281da177e4SLinus Torvalds * 829fcfccd91SLorenzo Stoakes * **** **** **** 830fcfccd91SLorenzo Stoakes * PPPPPPNNNNNN PPPPPPNNNNNN PPPPPPCCCCCC 8315d42ab29SWei Yang * cannot merge might become might become 832fcfccd91SLorenzo Stoakes * PPNNNNNNNNNN PPPPPPPPPPCC 8335d42ab29SWei Yang * mmap, brk or case 4 below case 5 below 8345d42ab29SWei Yang * mremap move: 835fcfccd91SLorenzo Stoakes * **** **** 836fcfccd91SLorenzo Stoakes * PPPP NNNN PPPPCCCCNNNN 8375d42ab29SWei Yang * might become might become 8385d42ab29SWei Yang * PPPPPPPPPPPP 1 or PPPPPPPPPPPP 6 or 839fcfccd91SLorenzo Stoakes * PPPPPPPPNNNN 2 or PPPPPPPPNNNN 7 or 840fcfccd91SLorenzo Stoakes * PPPPNNNNNNNN 3 PPPPNNNNNNNN 8 8411da177e4SLinus Torvalds * 842fcfccd91SLorenzo Stoakes * It is important for case 8 that the vma CCCC overlapping the 843fcfccd91SLorenzo Stoakes * region **** is never going to extended over NNNN. Instead NNNN must 844fcfccd91SLorenzo Stoakes * be extended in region **** and CCCC must be removed. This way in 8450503ea8fSLiam R. Howlett * all cases where vma_merge succeeds, the moment vma_merge drops the 846e86f15eeSAndrea Arcangeli * rmap_locks, the properties of the merged vma will be already 847e86f15eeSAndrea Arcangeli * correct for the whole merged range. Some of those properties like 848e86f15eeSAndrea Arcangeli * vm_page_prot/vm_flags may be accessed by rmap_walks and they must 849e86f15eeSAndrea Arcangeli * be correct for the whole merged range immediately after the 850fcfccd91SLorenzo Stoakes * rmap_locks are released. Otherwise if NNNN would be removed and 851fcfccd91SLorenzo Stoakes * CCCC would be extended over the NNNN range, remove_migration_ptes 852e86f15eeSAndrea Arcangeli * or other rmap walkers (if working on addresses beyond the "end" 853fcfccd91SLorenzo Stoakes * parameter) may establish ptes with the wrong permissions of CCCC 854fcfccd91SLorenzo Stoakes * instead of the right permissions of NNNN. 8550503ea8fSLiam R. Howlett * 8560503ea8fSLiam R. Howlett * In the code below: 8570503ea8fSLiam R. Howlett * PPPP is represented by *prev 858fcfccd91SLorenzo Stoakes * CCCC is represented by *curr or not represented at all (NULL) 859fcfccd91SLorenzo Stoakes * NNNN is represented by *next or not represented at all (NULL) 860fcfccd91SLorenzo Stoakes * **** is not represented - it will be merged and the vma containing the 8619e8a39d2SVlastimil Babka * area is returned, or the function will return NULL 8621da177e4SLinus Torvalds */ 86393bf5d4aSLorenzo Stoakes static struct vm_area_struct 86493bf5d4aSLorenzo Stoakes *vma_merge(struct vma_iterator *vmi, struct mm_struct *mm, 86593bf5d4aSLorenzo Stoakes struct vm_area_struct *prev, unsigned long addr, unsigned long end, 86693bf5d4aSLorenzo Stoakes unsigned long vm_flags, struct anon_vma *anon_vma, struct file *file, 86719a809afSAndrea Arcangeli pgoff_t pgoff, struct mempolicy *policy, 8689a10064fSColin Cross struct vm_userfaultfd_ctx vm_userfaultfd_ctx, 8695c26f6acSSuren Baghdasaryan struct anon_vma_name *anon_name) 8701da177e4SLinus Torvalds { 871b0729ae0SLorenzo Stoakes struct vm_area_struct *curr, *next, *res; 8720503ea8fSLiam R. Howlett struct vm_area_struct *vma, *adjust, *remove, *remove2; 873824135c4SLiam R. Howlett struct vm_area_struct *anon_dup = NULL; 8740173db4fSLorenzo Stoakes struct vma_prepare vp; 8750173db4fSLorenzo Stoakes pgoff_t vma_pgoff; 8760173db4fSLorenzo Stoakes int err = 0; 877eef19944SJakub Matěna bool merge_prev = false; 878eef19944SJakub Matěna bool merge_next = false; 8790503ea8fSLiam R. Howlett bool vma_expanded = false; 8800503ea8fSLiam R. Howlett unsigned long vma_start = addr; 8810173db4fSLorenzo Stoakes unsigned long vma_end = end; 8820173db4fSLorenzo Stoakes pgoff_t pglen = (end - addr) >> PAGE_SHIFT; 8830173db4fSLorenzo Stoakes long adj_start = 0; 8841da177e4SLinus Torvalds 8851da177e4SLinus Torvalds /* 8861da177e4SLinus Torvalds * We later require that vma->vm_flags == vm_flags, 8871da177e4SLinus Torvalds * so this tests vma->vm_flags & VM_SPECIAL, too. 8881da177e4SLinus Torvalds */ 8891da177e4SLinus Torvalds if (vm_flags & VM_SPECIAL) 8901da177e4SLinus Torvalds return NULL; 8911da177e4SLinus Torvalds 89200cd00a6SLorenzo Stoakes /* Does the input range span an existing VMA? (cases 5 - 8) */ 89300cd00a6SLorenzo Stoakes curr = find_vma_intersection(mm, prev ? prev->vm_end : 0, end); 8941da177e4SLinus Torvalds 89500cd00a6SLorenzo Stoakes if (!curr || /* cases 1 - 4 */ 89600cd00a6SLorenzo Stoakes end == curr->vm_end) /* cases 6 - 8, adjacent VMA */ 89700cd00a6SLorenzo Stoakes next = vma_lookup(mm, end); 89800cd00a6SLorenzo Stoakes else 89900cd00a6SLorenzo Stoakes next = NULL; /* case 5 */ 900e86f15eeSAndrea Arcangeli 9010503ea8fSLiam R. Howlett if (prev) { 9020503ea8fSLiam R. Howlett vma_start = prev->vm_start; 9030503ea8fSLiam R. Howlett vma_pgoff = prev->vm_pgoff; 9040173db4fSLorenzo Stoakes 905eef19944SJakub Matěna /* Can we merge the predecessor? */ 9060173db4fSLorenzo Stoakes if (addr == prev->vm_end && mpol_equal(vma_policy(prev), policy) 9070503ea8fSLiam R. Howlett && can_vma_merge_after(prev, vm_flags, anon_vma, file, 9080503ea8fSLiam R. Howlett pgoff, vm_userfaultfd_ctx, anon_name)) { 909eef19944SJakub Matěna merge_prev = true; 91018b098afSLiam R. Howlett vma_prev(vmi); 9111da177e4SLinus Torvalds } 9120503ea8fSLiam R. Howlett } 913b0729ae0SLorenzo Stoakes 914eef19944SJakub Matěna /* Can we merge the successor? */ 91500cd00a6SLorenzo Stoakes if (next && mpol_equal(policy, vma_policy(next)) && 9160173db4fSLorenzo Stoakes can_vma_merge_before(next, vm_flags, anon_vma, file, pgoff+pglen, 9179a10064fSColin Cross vm_userfaultfd_ctx, anon_name)) { 918eef19944SJakub Matěna merge_next = true; 919eef19944SJakub Matěna } 9200503ea8fSLiam R. Howlett 92129417d29SLorenzo Stoakes /* Verify some invariant that must be enforced by the caller. */ 92229417d29SLorenzo Stoakes VM_WARN_ON(prev && addr <= prev->vm_start); 92329417d29SLorenzo Stoakes VM_WARN_ON(curr && (addr != curr->vm_start || end > curr->vm_end)); 92429417d29SLorenzo Stoakes VM_WARN_ON(addr >= end); 92529417d29SLorenzo Stoakes 9260173db4fSLorenzo Stoakes if (!merge_prev && !merge_next) 9270173db4fSLorenzo Stoakes return NULL; /* Not mergeable. */ 9280173db4fSLorenzo Stoakes 929c9d6e982SSuren Baghdasaryan if (merge_prev) 930c9d6e982SSuren Baghdasaryan vma_start_write(prev); 931c9d6e982SSuren Baghdasaryan 9320173db4fSLorenzo Stoakes res = vma = prev; 9330503ea8fSLiam R. Howlett remove = remove2 = adjust = NULL; 9340173db4fSLorenzo Stoakes 935eef19944SJakub Matěna /* Can we merge both the predecessor and the successor? */ 936eef19944SJakub Matěna if (merge_prev && merge_next && 9370503ea8fSLiam R. Howlett is_mergeable_anon_vma(prev->anon_vma, next->anon_vma, NULL)) { 938c9d6e982SSuren Baghdasaryan vma_start_write(next); 9395ff783f1SVlastimil Babka remove = next; /* case 1 */ 9400503ea8fSLiam R. Howlett vma_end = next->vm_end; 941824135c4SLiam R. Howlett err = dup_anon_vma(prev, next, &anon_dup); 942fcfccd91SLorenzo Stoakes if (curr) { /* case 6 */ 943c9d6e982SSuren Baghdasaryan vma_start_write(curr); 944fcfccd91SLorenzo Stoakes remove = curr; 9450503ea8fSLiam R. Howlett remove2 = next; 9469a12d103SLiam R. Howlett /* 9479a12d103SLiam R. Howlett * Note that the dup_anon_vma below cannot overwrite err 9489a12d103SLiam R. Howlett * since the first caller would do nothing unless next 9499a12d103SLiam R. Howlett * has an anon_vma. 9509a12d103SLiam R. Howlett */ 9515ff783f1SVlastimil Babka if (!next->anon_vma) 952824135c4SLiam R. Howlett err = dup_anon_vma(prev, curr, &anon_dup); 9530503ea8fSLiam R. Howlett } 9540173db4fSLorenzo Stoakes } else if (merge_prev) { /* case 2 */ 955fcfccd91SLorenzo Stoakes if (curr) { 956c9d6e982SSuren Baghdasaryan vma_start_write(curr); 957824135c4SLiam R. Howlett err = dup_anon_vma(prev, curr, &anon_dup); 958fcfccd91SLorenzo Stoakes if (end == curr->vm_end) { /* case 7 */ 959fcfccd91SLorenzo Stoakes remove = curr; 9600503ea8fSLiam R. Howlett } else { /* case 5 */ 961fcfccd91SLorenzo Stoakes adjust = curr; 962fcfccd91SLorenzo Stoakes adj_start = (end - curr->vm_start); 9630503ea8fSLiam R. Howlett } 9640503ea8fSLiam R. Howlett } 9650173db4fSLorenzo Stoakes } else { /* merge_next */ 966c9d6e982SSuren Baghdasaryan vma_start_write(next); 967eef19944SJakub Matěna res = next; 9680503ea8fSLiam R. Howlett if (prev && addr < prev->vm_end) { /* case 4 */ 969c9d6e982SSuren Baghdasaryan vma_start_write(prev); 9700503ea8fSLiam R. Howlett vma_end = addr; 971183b7a60SVlastimil Babka adjust = next; 9721e76454fSVlastimil Babka adj_start = -(prev->vm_end - addr); 973824135c4SLiam R. Howlett err = dup_anon_vma(next, prev, &anon_dup); 9740503ea8fSLiam R. Howlett } else { 975b0729ae0SLorenzo Stoakes /* 976b0729ae0SLorenzo Stoakes * Note that cases 3 and 8 are the ONLY ones where prev 977b0729ae0SLorenzo Stoakes * is permitted to be (but is not necessarily) NULL. 978b0729ae0SLorenzo Stoakes */ 9790503ea8fSLiam R. Howlett vma = next; /* case 3 */ 9800503ea8fSLiam R. Howlett vma_start = addr; 9810503ea8fSLiam R. Howlett vma_end = next->vm_end; 9827e775787SVlastimil Babka vma_pgoff = next->vm_pgoff - pglen; 983fcfccd91SLorenzo Stoakes if (curr) { /* case 8 */ 984fcfccd91SLorenzo Stoakes vma_pgoff = curr->vm_pgoff; 985c9d6e982SSuren Baghdasaryan vma_start_write(curr); 986fcfccd91SLorenzo Stoakes remove = curr; 987824135c4SLiam R. Howlett err = dup_anon_vma(next, curr, &anon_dup); 9880503ea8fSLiam R. Howlett } 9890503ea8fSLiam R. Howlett } 9901da177e4SLinus Torvalds } 9911da177e4SLinus Torvalds 9920173db4fSLorenzo Stoakes /* Error in anon_vma clone. */ 993eef19944SJakub Matěna if (err) 9941419430cSLiam R. Howlett goto anon_vma_fail; 9950503ea8fSLiam R. Howlett 996b5df0922SLiam R. Howlett if (vma_start < vma->vm_start || vma_end > vma->vm_end) 997b5df0922SLiam R. Howlett vma_expanded = true; 998b5df0922SLiam R. Howlett 999b5df0922SLiam R. Howlett if (vma_expanded) { 1000b5df0922SLiam R. Howlett vma_iter_config(vmi, vma_start, vma_end); 1001b5df0922SLiam R. Howlett } else { 1002b5df0922SLiam R. Howlett vma_iter_config(vmi, adjust->vm_start + adj_start, 1003b5df0922SLiam R. Howlett adjust->vm_end); 1004b5df0922SLiam R. Howlett } 1005b5df0922SLiam R. Howlett 1006b5df0922SLiam R. Howlett if (vma_iter_prealloc(vmi, vma)) 10071419430cSLiam R. Howlett goto prealloc_fail; 10080503ea8fSLiam R. Howlett 10090503ea8fSLiam R. Howlett init_multi_vma_prep(&vp, vma, adjust, remove, remove2); 10100503ea8fSLiam R. Howlett VM_WARN_ON(vp.anon_vma && adjust && adjust->anon_vma && 10110503ea8fSLiam R. Howlett vp.anon_vma != adjust->anon_vma); 10120503ea8fSLiam R. Howlett 10130503ea8fSLiam R. Howlett vma_prepare(&vp); 1014ccf1d78dSSuren Baghdasaryan vma_adjust_trans_huge(vma, vma_start, vma_end, adj_start); 10150503ea8fSLiam R. Howlett 10160503ea8fSLiam R. Howlett vma->vm_start = vma_start; 10170503ea8fSLiam R. Howlett vma->vm_end = vma_end; 10180503ea8fSLiam R. Howlett vma->vm_pgoff = vma_pgoff; 10190503ea8fSLiam R. Howlett 10200503ea8fSLiam R. Howlett if (vma_expanded) 10210503ea8fSLiam R. Howlett vma_iter_store(vmi, vma); 10220503ea8fSLiam R. Howlett 10231e76454fSVlastimil Babka if (adj_start) { 10241e76454fSVlastimil Babka adjust->vm_start += adj_start; 10251e76454fSVlastimil Babka adjust->vm_pgoff += adj_start >> PAGE_SHIFT; 10261e76454fSVlastimil Babka if (adj_start < 0) { 10270503ea8fSLiam R. Howlett WARN_ON(vma_expanded); 10280503ea8fSLiam R. Howlett vma_iter_store(vmi, next); 10290503ea8fSLiam R. Howlett } 10300503ea8fSLiam R. Howlett } 10310503ea8fSLiam R. Howlett 10320503ea8fSLiam R. Howlett vma_complete(&vp, vmi, mm); 1033eef19944SJakub Matěna khugepaged_enter_vma(res, vm_flags); 10349760ebffSLiam R. Howlett return res; 10351419430cSLiam R. Howlett 10361419430cSLiam R. Howlett prealloc_fail: 1037824135c4SLiam R. Howlett if (anon_dup) 1038824135c4SLiam R. Howlett unlink_anon_vmas(anon_dup); 1039824135c4SLiam R. Howlett 10401419430cSLiam R. Howlett anon_vma_fail: 10411419430cSLiam R. Howlett vma_iter_set(vmi, addr); 10421419430cSLiam R. Howlett vma_iter_load(vmi); 10431419430cSLiam R. Howlett return NULL; 1044f2ebfe43SLiam R. Howlett } 1045f2ebfe43SLiam R. Howlett 10461da177e4SLinus Torvalds /* 1047b4f315b4SEthon Paul * Rough compatibility check to quickly see if it's even worth looking 1048d0e9fe17SLinus Torvalds * at sharing an anon_vma. 1049d0e9fe17SLinus Torvalds * 1050d0e9fe17SLinus Torvalds * They need to have the same vm_file, and the flags can only differ 1051d0e9fe17SLinus Torvalds * in things that mprotect may change. 1052d0e9fe17SLinus Torvalds * 1053d0e9fe17SLinus Torvalds * NOTE! The fact that we share an anon_vma doesn't _have_ to mean that 1054d0e9fe17SLinus Torvalds * we can merge the two vma's. For example, we refuse to merge a vma if 1055d0e9fe17SLinus Torvalds * there is a vm_ops->close() function, because that indicates that the 1056d0e9fe17SLinus Torvalds * driver is doing some kind of reference counting. But that doesn't 1057d0e9fe17SLinus Torvalds * really matter for the anon_vma sharing case. 1058d0e9fe17SLinus Torvalds */ 1059d0e9fe17SLinus Torvalds static int anon_vma_compatible(struct vm_area_struct *a, struct vm_area_struct *b) 1060d0e9fe17SLinus Torvalds { 1061d0e9fe17SLinus Torvalds return a->vm_end == b->vm_start && 1062d0e9fe17SLinus Torvalds mpol_equal(vma_policy(a), vma_policy(b)) && 1063d0e9fe17SLinus Torvalds a->vm_file == b->vm_file && 10646cb4d9a2SAnshuman Khandual !((a->vm_flags ^ b->vm_flags) & ~(VM_ACCESS_FLAGS | VM_SOFTDIRTY)) && 1065d0e9fe17SLinus Torvalds b->vm_pgoff == a->vm_pgoff + ((b->vm_start - a->vm_start) >> PAGE_SHIFT); 1066d0e9fe17SLinus Torvalds } 1067d0e9fe17SLinus Torvalds 1068d0e9fe17SLinus Torvalds /* 1069d0e9fe17SLinus Torvalds * Do some basic sanity checking to see if we can re-use the anon_vma 1070d0e9fe17SLinus Torvalds * from 'old'. The 'a'/'b' vma's are in VM order - one of them will be 1071d0e9fe17SLinus Torvalds * the same as 'old', the other will be the new one that is trying 1072d0e9fe17SLinus Torvalds * to share the anon_vma. 1073d0e9fe17SLinus Torvalds * 10745b449489SFlorian Rommel * NOTE! This runs with mmap_lock held for reading, so it is possible that 1075d0e9fe17SLinus Torvalds * the anon_vma of 'old' is concurrently in the process of being set up 1076d0e9fe17SLinus Torvalds * by another page fault trying to merge _that_. But that's ok: if it 1077d0e9fe17SLinus Torvalds * is being set up, that automatically means that it will be a singleton 1078d0e9fe17SLinus Torvalds * acceptable for merging, so we can do all of this optimistically. But 10794db0c3c2SJason Low * we do that READ_ONCE() to make sure that we never re-load the pointer. 1080d0e9fe17SLinus Torvalds * 1081d0e9fe17SLinus Torvalds * IOW: that the "list_is_singular()" test on the anon_vma_chain only 1082d0e9fe17SLinus Torvalds * matters for the 'stable anon_vma' case (ie the thing we want to avoid 1083d0e9fe17SLinus Torvalds * is to return an anon_vma that is "complex" due to having gone through 1084d0e9fe17SLinus Torvalds * a fork). 1085d0e9fe17SLinus Torvalds * 1086d0e9fe17SLinus Torvalds * We also make sure that the two vma's are compatible (adjacent, 1087d0e9fe17SLinus Torvalds * and with the same memory policies). That's all stable, even with just 10885b449489SFlorian Rommel * a read lock on the mmap_lock. 1089d0e9fe17SLinus Torvalds */ 1090d0e9fe17SLinus Torvalds static struct anon_vma *reusable_anon_vma(struct vm_area_struct *old, struct vm_area_struct *a, struct vm_area_struct *b) 1091d0e9fe17SLinus Torvalds { 1092d0e9fe17SLinus Torvalds if (anon_vma_compatible(a, b)) { 10934db0c3c2SJason Low struct anon_vma *anon_vma = READ_ONCE(old->anon_vma); 1094d0e9fe17SLinus Torvalds 1095d0e9fe17SLinus Torvalds if (anon_vma && list_is_singular(&old->anon_vma_chain)) 1096d0e9fe17SLinus Torvalds return anon_vma; 1097d0e9fe17SLinus Torvalds } 1098d0e9fe17SLinus Torvalds return NULL; 1099d0e9fe17SLinus Torvalds } 1100d0e9fe17SLinus Torvalds 1101d0e9fe17SLinus Torvalds /* 11021da177e4SLinus Torvalds * find_mergeable_anon_vma is used by anon_vma_prepare, to check 11031da177e4SLinus Torvalds * neighbouring vmas for a suitable anon_vma, before it goes off 11041da177e4SLinus Torvalds * to allocate a new anon_vma. It checks because a repetitive 11051da177e4SLinus Torvalds * sequence of mprotects and faults may otherwise lead to distinct 11061da177e4SLinus Torvalds * anon_vmas being allocated, preventing vma merge in subsequent 11071da177e4SLinus Torvalds * mprotect. 11081da177e4SLinus Torvalds */ 11091da177e4SLinus Torvalds struct anon_vma *find_mergeable_anon_vma(struct vm_area_struct *vma) 11101da177e4SLinus Torvalds { 1111763ecb03SLiam R. Howlett MA_STATE(mas, &vma->vm_mm->mm_mt, vma->vm_end, vma->vm_end); 1112a67c8caaSMiaohe Lin struct anon_vma *anon_vma = NULL; 1113763ecb03SLiam R. Howlett struct vm_area_struct *prev, *next; 11141da177e4SLinus Torvalds 1115a67c8caaSMiaohe Lin /* Try next first. */ 1116763ecb03SLiam R. Howlett next = mas_walk(&mas); 1117763ecb03SLiam R. Howlett if (next) { 1118763ecb03SLiam R. Howlett anon_vma = reusable_anon_vma(next, vma, next); 1119d0e9fe17SLinus Torvalds if (anon_vma) 1120d0e9fe17SLinus Torvalds return anon_vma; 1121a67c8caaSMiaohe Lin } 11221da177e4SLinus Torvalds 1123763ecb03SLiam R. Howlett prev = mas_prev(&mas, 0); 1124763ecb03SLiam R. Howlett VM_BUG_ON_VMA(prev != vma, vma); 1125763ecb03SLiam R. Howlett prev = mas_prev(&mas, 0); 1126a67c8caaSMiaohe Lin /* Try prev next. */ 1127763ecb03SLiam R. Howlett if (prev) 1128763ecb03SLiam R. Howlett anon_vma = reusable_anon_vma(prev, prev, vma); 1129a67c8caaSMiaohe Lin 11301da177e4SLinus Torvalds /* 1131a67c8caaSMiaohe Lin * We might reach here with anon_vma == NULL if we can't find 1132a67c8caaSMiaohe Lin * any reusable anon_vma. 11331da177e4SLinus Torvalds * There's no absolute need to look only at touching neighbours: 11341da177e4SLinus Torvalds * we could search further afield for "compatible" anon_vmas. 11351da177e4SLinus Torvalds * But it would probably just be a waste of time searching, 11361da177e4SLinus Torvalds * or lead to too many vmas hanging off the same anon_vma. 11371da177e4SLinus Torvalds * We're trying to allow mprotect remerging later on, 11381da177e4SLinus Torvalds * not trying to minimize memory used for anon_vmas. 11391da177e4SLinus Torvalds */ 1140a67c8caaSMiaohe Lin return anon_vma; 11411da177e4SLinus Torvalds } 11421da177e4SLinus Torvalds 11431da177e4SLinus Torvalds /* 114440401530SAl Viro * If a hint addr is less than mmap_min_addr change hint to be as 114540401530SAl Viro * low as possible but still greater than mmap_min_addr 114640401530SAl Viro */ 114740401530SAl Viro static inline unsigned long round_hint_to_min(unsigned long hint) 114840401530SAl Viro { 114940401530SAl Viro hint &= PAGE_MASK; 115040401530SAl Viro if (((void *)hint != NULL) && 115140401530SAl Viro (hint < mmap_min_addr)) 115240401530SAl Viro return PAGE_ALIGN(mmap_min_addr); 115340401530SAl Viro return hint; 115440401530SAl Viro } 115540401530SAl Viro 1156b0cc5e89SAndrew Morton bool mlock_future_ok(struct mm_struct *mm, unsigned long flags, 11573c54a298SLorenzo Stoakes unsigned long bytes) 1158363ee17fSDavidlohr Bueso { 11593c54a298SLorenzo Stoakes unsigned long locked_pages, limit_pages; 1160363ee17fSDavidlohr Bueso 11613c54a298SLorenzo Stoakes if (!(flags & VM_LOCKED) || capable(CAP_IPC_LOCK)) 11623c54a298SLorenzo Stoakes return true; 11633c54a298SLorenzo Stoakes 11643c54a298SLorenzo Stoakes locked_pages = bytes >> PAGE_SHIFT; 11653c54a298SLorenzo Stoakes locked_pages += mm->locked_vm; 11663c54a298SLorenzo Stoakes 11673c54a298SLorenzo Stoakes limit_pages = rlimit(RLIMIT_MEMLOCK); 11683c54a298SLorenzo Stoakes limit_pages >>= PAGE_SHIFT; 11693c54a298SLorenzo Stoakes 11703c54a298SLorenzo Stoakes return locked_pages <= limit_pages; 1171363ee17fSDavidlohr Bueso } 1172363ee17fSDavidlohr Bueso 1173be83bbf8SLinus Torvalds static inline u64 file_mmap_size_max(struct file *file, struct inode *inode) 1174be83bbf8SLinus Torvalds { 1175be83bbf8SLinus Torvalds if (S_ISREG(inode->i_mode)) 1176423913adSLinus Torvalds return MAX_LFS_FILESIZE; 1177be83bbf8SLinus Torvalds 1178be83bbf8SLinus Torvalds if (S_ISBLK(inode->i_mode)) 1179be83bbf8SLinus Torvalds return MAX_LFS_FILESIZE; 1180be83bbf8SLinus Torvalds 118176f34950SIvan Khoronzhuk if (S_ISSOCK(inode->i_mode)) 118276f34950SIvan Khoronzhuk return MAX_LFS_FILESIZE; 118376f34950SIvan Khoronzhuk 1184be83bbf8SLinus Torvalds /* Special "we do even unsigned file positions" case */ 1185be83bbf8SLinus Torvalds if (file->f_mode & FMODE_UNSIGNED_OFFSET) 1186be83bbf8SLinus Torvalds return 0; 1187be83bbf8SLinus Torvalds 1188be83bbf8SLinus Torvalds /* Yes, random drivers might want more. But I'm tired of buggy drivers */ 1189be83bbf8SLinus Torvalds return ULONG_MAX; 1190be83bbf8SLinus Torvalds } 1191be83bbf8SLinus Torvalds 1192be83bbf8SLinus Torvalds static inline bool file_mmap_ok(struct file *file, struct inode *inode, 1193be83bbf8SLinus Torvalds unsigned long pgoff, unsigned long len) 1194be83bbf8SLinus Torvalds { 1195be83bbf8SLinus Torvalds u64 maxsize = file_mmap_size_max(file, inode); 1196be83bbf8SLinus Torvalds 1197be83bbf8SLinus Torvalds if (maxsize && len > maxsize) 1198be83bbf8SLinus Torvalds return false; 1199be83bbf8SLinus Torvalds maxsize -= len; 1200be83bbf8SLinus Torvalds if (pgoff > maxsize >> PAGE_SHIFT) 1201be83bbf8SLinus Torvalds return false; 1202be83bbf8SLinus Torvalds return true; 1203be83bbf8SLinus Torvalds } 1204be83bbf8SLinus Torvalds 120540401530SAl Viro /* 12063e4e28c5SMichel Lespinasse * The caller must write-lock current->mm->mmap_lock. 12071da177e4SLinus Torvalds */ 12081fcfd8dbSOleg Nesterov unsigned long do_mmap(struct file *file, unsigned long addr, 12091da177e4SLinus Torvalds unsigned long len, unsigned long prot, 1210592b5fadSYu-cheng Yu unsigned long flags, vm_flags_t vm_flags, 1211592b5fadSYu-cheng Yu unsigned long pgoff, unsigned long *populate, 1212592b5fadSYu-cheng Yu struct list_head *uf) 12131da177e4SLinus Torvalds { 12141da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 121562b5f7d0SDave Hansen int pkey = 0; 12161da177e4SLinus Torvalds 121741badc15SMichel Lespinasse *populate = 0; 1218bebeb3d6SMichel Lespinasse 1219e37609bbSPiotr Kwapulinski if (!len) 1220e37609bbSPiotr Kwapulinski return -EINVAL; 1221e37609bbSPiotr Kwapulinski 12221da177e4SLinus Torvalds /* 12231da177e4SLinus Torvalds * Does the application expect PROT_READ to imply PROT_EXEC? 12241da177e4SLinus Torvalds * 12251da177e4SLinus Torvalds * (the exception is when the underlying filesystem is noexec 1226be16dd76SMuhammad Muzammil * mounted, in which case we don't add PROT_EXEC.) 12271da177e4SLinus Torvalds */ 12281da177e4SLinus Torvalds if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC)) 122990f8572bSEric W. Biederman if (!(file && path_noexec(&file->f_path))) 12301da177e4SLinus Torvalds prot |= PROT_EXEC; 12311da177e4SLinus Torvalds 1232a4ff8e86SMichal Hocko /* force arch specific MAP_FIXED handling in get_unmapped_area */ 1233a4ff8e86SMichal Hocko if (flags & MAP_FIXED_NOREPLACE) 1234a4ff8e86SMichal Hocko flags |= MAP_FIXED; 1235a4ff8e86SMichal Hocko 12367cd94146SEric Paris if (!(flags & MAP_FIXED)) 12377cd94146SEric Paris addr = round_hint_to_min(addr); 12387cd94146SEric Paris 12391da177e4SLinus Torvalds /* Careful about overflows.. */ 12401da177e4SLinus Torvalds len = PAGE_ALIGN(len); 12419206de95SAl Viro if (!len) 12421da177e4SLinus Torvalds return -ENOMEM; 12431da177e4SLinus Torvalds 12441da177e4SLinus Torvalds /* offset overflow? */ 12451da177e4SLinus Torvalds if ((pgoff + (len >> PAGE_SHIFT)) < pgoff) 12461da177e4SLinus Torvalds return -EOVERFLOW; 12471da177e4SLinus Torvalds 12481da177e4SLinus Torvalds /* Too many mappings? */ 12491da177e4SLinus Torvalds if (mm->map_count > sysctl_max_map_count) 12501da177e4SLinus Torvalds return -ENOMEM; 12511da177e4SLinus Torvalds 12521da177e4SLinus Torvalds /* Obtain the address to map to. we verify (or select) it and ensure 12531da177e4SLinus Torvalds * that it represents a valid section of the address space. 12541da177e4SLinus Torvalds */ 12551da177e4SLinus Torvalds addr = get_unmapped_area(file, addr, len, pgoff, flags); 1256ff68dac6SGaowei Pu if (IS_ERR_VALUE(addr)) 12571da177e4SLinus Torvalds return addr; 12581da177e4SLinus Torvalds 1259a4ff8e86SMichal Hocko if (flags & MAP_FIXED_NOREPLACE) { 126035e43c5fSLiam Howlett if (find_vma_intersection(mm, addr, addr + len)) 1261a4ff8e86SMichal Hocko return -EEXIST; 1262a4ff8e86SMichal Hocko } 1263a4ff8e86SMichal Hocko 126462b5f7d0SDave Hansen if (prot == PROT_EXEC) { 126562b5f7d0SDave Hansen pkey = execute_only_pkey(mm); 126662b5f7d0SDave Hansen if (pkey < 0) 126762b5f7d0SDave Hansen pkey = 0; 126862b5f7d0SDave Hansen } 126962b5f7d0SDave Hansen 12701da177e4SLinus Torvalds /* Do simple checking here so the lower-level routines won't have 12711da177e4SLinus Torvalds * to. we assume access permissions have been handled by the open 12721da177e4SLinus Torvalds * of the memory object, so we don't do any here. 12731da177e4SLinus Torvalds */ 1274592b5fadSYu-cheng Yu vm_flags |= calc_vm_prot_bits(prot, pkey) | calc_vm_flag_bits(flags) | 12751da177e4SLinus Torvalds mm->def_flags | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC; 12761da177e4SLinus Torvalds 1277cdf7b341SHuang Shijie if (flags & MAP_LOCKED) 12781da177e4SLinus Torvalds if (!can_do_mlock()) 12791da177e4SLinus Torvalds return -EPERM; 1280ba470de4SRik van Riel 1281b0cc5e89SAndrew Morton if (!mlock_future_ok(mm, vm_flags, len)) 12821da177e4SLinus Torvalds return -EAGAIN; 12831da177e4SLinus Torvalds 12841da177e4SLinus Torvalds if (file) { 1285077bf22bSOleg Nesterov struct inode *inode = file_inode(file); 12861c972597SDan Williams unsigned long flags_mask; 12871c972597SDan Williams 1288be83bbf8SLinus Torvalds if (!file_mmap_ok(file, inode, pgoff, len)) 1289be83bbf8SLinus Torvalds return -EOVERFLOW; 1290be83bbf8SLinus Torvalds 12911c972597SDan Williams flags_mask = LEGACY_MAP_MASK | file->f_op->mmap_supported_flags; 1292077bf22bSOleg Nesterov 12931da177e4SLinus Torvalds switch (flags & MAP_TYPE) { 12941da177e4SLinus Torvalds case MAP_SHARED: 12951c972597SDan Williams /* 12961c972597SDan Williams * Force use of MAP_SHARED_VALIDATE with non-legacy 12971c972597SDan Williams * flags. E.g. MAP_SYNC is dangerous to use with 12981c972597SDan Williams * MAP_SHARED as you don't know which consistency model 12991c972597SDan Williams * you will get. We silently ignore unsupported flags 13001c972597SDan Williams * with MAP_SHARED to preserve backward compatibility. 13011c972597SDan Williams */ 13021c972597SDan Williams flags &= LEGACY_MAP_MASK; 1303e4a9bc58SJoe Perches fallthrough; 13041c972597SDan Williams case MAP_SHARED_VALIDATE: 13051c972597SDan Williams if (flags & ~flags_mask) 13061c972597SDan Williams return -EOPNOTSUPP; 1307dc617f29SDarrick J. Wong if (prot & PROT_WRITE) { 1308dc617f29SDarrick J. Wong if (!(file->f_mode & FMODE_WRITE)) 13091da177e4SLinus Torvalds return -EACCES; 1310dc617f29SDarrick J. Wong if (IS_SWAPFILE(file->f_mapping->host)) 1311dc617f29SDarrick J. Wong return -ETXTBSY; 1312dc617f29SDarrick J. Wong } 13131da177e4SLinus Torvalds 13141da177e4SLinus Torvalds /* 13151da177e4SLinus Torvalds * Make sure we don't allow writing to an append-only 13161da177e4SLinus Torvalds * file.. 13171da177e4SLinus Torvalds */ 13181da177e4SLinus Torvalds if (IS_APPEND(inode) && (file->f_mode & FMODE_WRITE)) 13191da177e4SLinus Torvalds return -EACCES; 13201da177e4SLinus Torvalds 13211da177e4SLinus Torvalds vm_flags |= VM_SHARED | VM_MAYSHARE; 13221da177e4SLinus Torvalds if (!(file->f_mode & FMODE_WRITE)) 13231da177e4SLinus Torvalds vm_flags &= ~(VM_MAYWRITE | VM_SHARED); 1324e4a9bc58SJoe Perches fallthrough; 13251da177e4SLinus Torvalds case MAP_PRIVATE: 13261da177e4SLinus Torvalds if (!(file->f_mode & FMODE_READ)) 13271da177e4SLinus Torvalds return -EACCES; 132890f8572bSEric W. Biederman if (path_noexec(&file->f_path)) { 132980c5606cSLinus Torvalds if (vm_flags & VM_EXEC) 133080c5606cSLinus Torvalds return -EPERM; 133180c5606cSLinus Torvalds vm_flags &= ~VM_MAYEXEC; 133280c5606cSLinus Torvalds } 133380c5606cSLinus Torvalds 133472c2d531SAl Viro if (!file->f_op->mmap) 133580c5606cSLinus Torvalds return -ENODEV; 1336b2c56e4fSOleg Nesterov if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP)) 1337b2c56e4fSOleg Nesterov return -EINVAL; 13381da177e4SLinus Torvalds break; 13391da177e4SLinus Torvalds 13401da177e4SLinus Torvalds default: 13411da177e4SLinus Torvalds return -EINVAL; 13421da177e4SLinus Torvalds } 13431da177e4SLinus Torvalds } else { 13441da177e4SLinus Torvalds switch (flags & MAP_TYPE) { 13451da177e4SLinus Torvalds case MAP_SHARED: 1346b2c56e4fSOleg Nesterov if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP)) 1347b2c56e4fSOleg Nesterov return -EINVAL; 1348ce363942STejun Heo /* 1349ce363942STejun Heo * Ignore pgoff. 1350ce363942STejun Heo */ 1351ce363942STejun Heo pgoff = 0; 13521da177e4SLinus Torvalds vm_flags |= VM_SHARED | VM_MAYSHARE; 13531da177e4SLinus Torvalds break; 13541da177e4SLinus Torvalds case MAP_PRIVATE: 13551da177e4SLinus Torvalds /* 13561da177e4SLinus Torvalds * Set pgoff according to addr for anon_vma. 13571da177e4SLinus Torvalds */ 13581da177e4SLinus Torvalds pgoff = addr >> PAGE_SHIFT; 13591da177e4SLinus Torvalds break; 13601da177e4SLinus Torvalds default: 13611da177e4SLinus Torvalds return -EINVAL; 13621da177e4SLinus Torvalds } 13631da177e4SLinus Torvalds } 13641da177e4SLinus Torvalds 1365c22c0d63SMichel Lespinasse /* 1366c22c0d63SMichel Lespinasse * Set 'VM_NORESERVE' if we should not account for the 1367c22c0d63SMichel Lespinasse * memory use of this mapping. 1368c22c0d63SMichel Lespinasse */ 1369c22c0d63SMichel Lespinasse if (flags & MAP_NORESERVE) { 1370c22c0d63SMichel Lespinasse /* We honor MAP_NORESERVE if allowed to overcommit */ 1371c22c0d63SMichel Lespinasse if (sysctl_overcommit_memory != OVERCOMMIT_NEVER) 1372c22c0d63SMichel Lespinasse vm_flags |= VM_NORESERVE; 1373c22c0d63SMichel Lespinasse 1374c22c0d63SMichel Lespinasse /* hugetlb applies strict overcommit unless MAP_NORESERVE */ 1375c22c0d63SMichel Lespinasse if (file && is_file_hugepages(file)) 1376c22c0d63SMichel Lespinasse vm_flags |= VM_NORESERVE; 1377c22c0d63SMichel Lespinasse } 1378c22c0d63SMichel Lespinasse 1379897ab3e0SMike Rapoport addr = mmap_region(file, addr, len, vm_flags, pgoff, uf); 138009a9f1d2SMichel Lespinasse if (!IS_ERR_VALUE(addr) && 138109a9f1d2SMichel Lespinasse ((vm_flags & VM_LOCKED) || 138209a9f1d2SMichel Lespinasse (flags & (MAP_POPULATE | MAP_NONBLOCK)) == MAP_POPULATE)) 138341badc15SMichel Lespinasse *populate = len; 1384bebeb3d6SMichel Lespinasse return addr; 13850165ab44SMiklos Szeredi } 13866be5ceb0SLinus Torvalds 1387a90f590aSDominik Brodowski unsigned long ksys_mmap_pgoff(unsigned long addr, unsigned long len, 1388a90f590aSDominik Brodowski unsigned long prot, unsigned long flags, 1389a90f590aSDominik Brodowski unsigned long fd, unsigned long pgoff) 139066f0dc48SHugh Dickins { 139166f0dc48SHugh Dickins struct file *file = NULL; 13921e3ee14bSChen Gang unsigned long retval; 139366f0dc48SHugh Dickins 139466f0dc48SHugh Dickins if (!(flags & MAP_ANONYMOUS)) { 1395120a795dSAl Viro audit_mmap_fd(fd, flags); 139666f0dc48SHugh Dickins file = fget(fd); 139766f0dc48SHugh Dickins if (!file) 13981e3ee14bSChen Gang return -EBADF; 13997bba8f0eSZhen Lei if (is_file_hugepages(file)) { 1400af73e4d9SNaoya Horiguchi len = ALIGN(len, huge_page_size(hstate_file(file))); 14017bba8f0eSZhen Lei } else if (unlikely(flags & MAP_HUGETLB)) { 1402493af578SJörn Engel retval = -EINVAL; 1403493af578SJörn Engel goto out_fput; 14047bba8f0eSZhen Lei } 140566f0dc48SHugh Dickins } else if (flags & MAP_HUGETLB) { 1406c103a4dcSAndrew Morton struct hstate *hs; 1407af73e4d9SNaoya Horiguchi 140820ac2893SAnshuman Khandual hs = hstate_sizelog((flags >> MAP_HUGE_SHIFT) & MAP_HUGE_MASK); 1409091d0d55SLi Zefan if (!hs) 1410091d0d55SLi Zefan return -EINVAL; 1411091d0d55SLi Zefan 1412091d0d55SLi Zefan len = ALIGN(len, huge_page_size(hs)); 141366f0dc48SHugh Dickins /* 141466f0dc48SHugh Dickins * VM_NORESERVE is used because the reservations will be 141566f0dc48SHugh Dickins * taken when vm_ops->mmap() is called 141666f0dc48SHugh Dickins */ 1417af73e4d9SNaoya Horiguchi file = hugetlb_file_setup(HUGETLB_ANON_FILE, len, 141842d7395fSAndi Kleen VM_NORESERVE, 141983c1fd76Szhangyiru HUGETLB_ANONHUGE_INODE, 142042d7395fSAndi Kleen (flags >> MAP_HUGE_SHIFT) & MAP_HUGE_MASK); 142166f0dc48SHugh Dickins if (IS_ERR(file)) 142266f0dc48SHugh Dickins return PTR_ERR(file); 142366f0dc48SHugh Dickins } 142466f0dc48SHugh Dickins 14259fbeb5abSMichal Hocko retval = vm_mmap_pgoff(file, addr, len, prot, flags, pgoff); 1426493af578SJörn Engel out_fput: 142766f0dc48SHugh Dickins if (file) 142866f0dc48SHugh Dickins fput(file); 142966f0dc48SHugh Dickins return retval; 143066f0dc48SHugh Dickins } 143166f0dc48SHugh Dickins 1432a90f590aSDominik Brodowski SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len, 1433a90f590aSDominik Brodowski unsigned long, prot, unsigned long, flags, 1434a90f590aSDominik Brodowski unsigned long, fd, unsigned long, pgoff) 1435a90f590aSDominik Brodowski { 1436a90f590aSDominik Brodowski return ksys_mmap_pgoff(addr, len, prot, flags, fd, pgoff); 1437a90f590aSDominik Brodowski } 1438a90f590aSDominik Brodowski 1439a4679373SChristoph Hellwig #ifdef __ARCH_WANT_SYS_OLD_MMAP 1440a4679373SChristoph Hellwig struct mmap_arg_struct { 1441a4679373SChristoph Hellwig unsigned long addr; 1442a4679373SChristoph Hellwig unsigned long len; 1443a4679373SChristoph Hellwig unsigned long prot; 1444a4679373SChristoph Hellwig unsigned long flags; 1445a4679373SChristoph Hellwig unsigned long fd; 1446a4679373SChristoph Hellwig unsigned long offset; 1447a4679373SChristoph Hellwig }; 1448a4679373SChristoph Hellwig 1449a4679373SChristoph Hellwig SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg) 1450a4679373SChristoph Hellwig { 1451a4679373SChristoph Hellwig struct mmap_arg_struct a; 1452a4679373SChristoph Hellwig 1453a4679373SChristoph Hellwig if (copy_from_user(&a, arg, sizeof(a))) 1454a4679373SChristoph Hellwig return -EFAULT; 1455de1741a1SAlexander Kuleshov if (offset_in_page(a.offset)) 1456a4679373SChristoph Hellwig return -EINVAL; 1457a4679373SChristoph Hellwig 1458a90f590aSDominik Brodowski return ksys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd, 1459a4679373SChristoph Hellwig a.offset >> PAGE_SHIFT); 1460a4679373SChristoph Hellwig } 1461a4679373SChristoph Hellwig #endif /* __ARCH_WANT_SYS_OLD_MMAP */ 1462a4679373SChristoph Hellwig 146354cbbbf3SLorenzo Stoakes static bool vm_ops_needs_writenotify(const struct vm_operations_struct *vm_ops) 146454cbbbf3SLorenzo Stoakes { 146554cbbbf3SLorenzo Stoakes return vm_ops && (vm_ops->page_mkwrite || vm_ops->pfn_mkwrite); 146654cbbbf3SLorenzo Stoakes } 146754cbbbf3SLorenzo Stoakes 146854cbbbf3SLorenzo Stoakes static bool vma_is_shared_writable(struct vm_area_struct *vma) 146954cbbbf3SLorenzo Stoakes { 147054cbbbf3SLorenzo Stoakes return (vma->vm_flags & (VM_WRITE | VM_SHARED)) == 147154cbbbf3SLorenzo Stoakes (VM_WRITE | VM_SHARED); 147254cbbbf3SLorenzo Stoakes } 147354cbbbf3SLorenzo Stoakes 147454cbbbf3SLorenzo Stoakes static bool vma_fs_can_writeback(struct vm_area_struct *vma) 147554cbbbf3SLorenzo Stoakes { 147654cbbbf3SLorenzo Stoakes /* No managed pages to writeback. */ 147754cbbbf3SLorenzo Stoakes if (vma->vm_flags & VM_PFNMAP) 147854cbbbf3SLorenzo Stoakes return false; 147954cbbbf3SLorenzo Stoakes 148054cbbbf3SLorenzo Stoakes return vma->vm_file && vma->vm_file->f_mapping && 148154cbbbf3SLorenzo Stoakes mapping_can_writeback(vma->vm_file->f_mapping); 148254cbbbf3SLorenzo Stoakes } 148354cbbbf3SLorenzo Stoakes 148454cbbbf3SLorenzo Stoakes /* 148554cbbbf3SLorenzo Stoakes * Does this VMA require the underlying folios to have their dirty state 148654cbbbf3SLorenzo Stoakes * tracked? 148754cbbbf3SLorenzo Stoakes */ 148854cbbbf3SLorenzo Stoakes bool vma_needs_dirty_tracking(struct vm_area_struct *vma) 148954cbbbf3SLorenzo Stoakes { 149054cbbbf3SLorenzo Stoakes /* Only shared, writable VMAs require dirty tracking. */ 149154cbbbf3SLorenzo Stoakes if (!vma_is_shared_writable(vma)) 149254cbbbf3SLorenzo Stoakes return false; 149354cbbbf3SLorenzo Stoakes 149454cbbbf3SLorenzo Stoakes /* Does the filesystem need to be notified? */ 149554cbbbf3SLorenzo Stoakes if (vm_ops_needs_writenotify(vma->vm_ops)) 149654cbbbf3SLorenzo Stoakes return true; 149754cbbbf3SLorenzo Stoakes 149854cbbbf3SLorenzo Stoakes /* 149954cbbbf3SLorenzo Stoakes * Even if the filesystem doesn't indicate a need for writenotify, if it 150054cbbbf3SLorenzo Stoakes * can writeback, dirty tracking is still required. 150154cbbbf3SLorenzo Stoakes */ 150254cbbbf3SLorenzo Stoakes return vma_fs_can_writeback(vma); 150354cbbbf3SLorenzo Stoakes } 150454cbbbf3SLorenzo Stoakes 15054e950f6fSAlexey Dobriyan /* 15068bb4e7a2SWei Yang * Some shared mappings will want the pages marked read-only 15074e950f6fSAlexey Dobriyan * to track write events. If so, we'll downgrade vm_page_prot 15084e950f6fSAlexey Dobriyan * to the private version (using protection_map[] without the 15094e950f6fSAlexey Dobriyan * VM_SHARED bit). 15104e950f6fSAlexey Dobriyan */ 15116d2329f8SAndrea Arcangeli int vma_wants_writenotify(struct vm_area_struct *vma, pgprot_t vm_page_prot) 15124e950f6fSAlexey Dobriyan { 15134e950f6fSAlexey Dobriyan /* If it was private or non-writable, the write bit is already clear */ 151454cbbbf3SLorenzo Stoakes if (!vma_is_shared_writable(vma)) 15154e950f6fSAlexey Dobriyan return 0; 15164e950f6fSAlexey Dobriyan 15174e950f6fSAlexey Dobriyan /* The backer wishes to know when pages are first written to? */ 151854cbbbf3SLorenzo Stoakes if (vm_ops_needs_writenotify(vma->vm_ops)) 15194e950f6fSAlexey Dobriyan return 1; 15204e950f6fSAlexey Dobriyan 152164e45507SPeter Feiner /* The open routine did something to the protections that pgprot_modify 152264e45507SPeter Feiner * won't preserve? */ 15236d2329f8SAndrea Arcangeli if (pgprot_val(vm_page_prot) != 152454cbbbf3SLorenzo Stoakes pgprot_val(vm_pgprot_modify(vm_page_prot, vma->vm_flags))) 15254e950f6fSAlexey Dobriyan return 0; 15264e950f6fSAlexey Dobriyan 1527f96f7a40SDavid Hildenbrand /* 1528f96f7a40SDavid Hildenbrand * Do we need to track softdirty? hugetlb does not support softdirty 1529f96f7a40SDavid Hildenbrand * tracking yet. 1530f96f7a40SDavid Hildenbrand */ 1531f96f7a40SDavid Hildenbrand if (vma_soft_dirty_enabled(vma) && !is_vm_hugetlb_page(vma)) 153264e45507SPeter Feiner return 1; 153364e45507SPeter Feiner 153451d3d5ebSDavid Hildenbrand /* Do we need write faults for uffd-wp tracking? */ 153551d3d5ebSDavid Hildenbrand if (userfaultfd_wp(vma)) 153651d3d5ebSDavid Hildenbrand return 1; 153751d3d5ebSDavid Hildenbrand 15384e950f6fSAlexey Dobriyan /* Can the mapping track the dirty pages? */ 153954cbbbf3SLorenzo Stoakes return vma_fs_can_writeback(vma); 15404e950f6fSAlexey Dobriyan } 15414e950f6fSAlexey Dobriyan 1542fc8744adSLinus Torvalds /* 1543fc8744adSLinus Torvalds * We account for memory if it's a private writeable mapping, 15445a6fe125SMel Gorman * not hugepages and VM_NORESERVE wasn't set. 1545fc8744adSLinus Torvalds */ 1546ca16d140SKOSAKI Motohiro static inline int accountable_mapping(struct file *file, vm_flags_t vm_flags) 1547fc8744adSLinus Torvalds { 15485a6fe125SMel Gorman /* 15495a6fe125SMel Gorman * hugetlb has its own accounting separate from the core VM 15505a6fe125SMel Gorman * VM_HUGETLB may not be set yet so we cannot check for that flag. 15515a6fe125SMel Gorman */ 15525a6fe125SMel Gorman if (file && is_file_hugepages(file)) 15535a6fe125SMel Gorman return 0; 15545a6fe125SMel Gorman 1555fc8744adSLinus Torvalds return (vm_flags & (VM_NORESERVE | VM_SHARED | VM_WRITE)) == VM_WRITE; 1556fc8744adSLinus Torvalds } 1557fc8744adSLinus Torvalds 15583499a131SLiam R. Howlett /** 15593499a131SLiam R. Howlett * unmapped_area() - Find an area between the low_limit and the high_limit with 15603499a131SLiam R. Howlett * the correct alignment and offset, all from @info. Note: current->mm is used 15613499a131SLiam R. Howlett * for the search. 15623499a131SLiam R. Howlett * 156382b24936SVernon Yang * @info: The unmapped area information including the range [low_limit - 156482b24936SVernon Yang * high_limit), the alignment offset and mask. 15653499a131SLiam R. Howlett * 15663499a131SLiam R. Howlett * Return: A memory address or -ENOMEM. 15673499a131SLiam R. Howlett */ 1568baceaf1cSJaewon Kim static unsigned long unmapped_area(struct vm_unmapped_area_info *info) 1569db4fbfb9SMichel Lespinasse { 15706b008640SLinus Torvalds unsigned long length, gap; 15716b008640SLinus Torvalds unsigned long low_limit, high_limit; 157258c5d0d6SLiam R. Howlett struct vm_area_struct *tmp; 1573db4fbfb9SMichel Lespinasse 15743499a131SLiam R. Howlett MA_STATE(mas, ¤t->mm->mm_mt, 0, 0); 1575db4fbfb9SMichel Lespinasse 1576db4fbfb9SMichel Lespinasse /* Adjust search length to account for worst case alignment overhead */ 1577db4fbfb9SMichel Lespinasse length = info->length + info->align_mask; 1578db4fbfb9SMichel Lespinasse if (length < info->length) 1579db4fbfb9SMichel Lespinasse return -ENOMEM; 1580db4fbfb9SMichel Lespinasse 158158c5d0d6SLiam R. Howlett low_limit = info->low_limit; 15826b008640SLinus Torvalds if (low_limit < mmap_min_addr) 15836b008640SLinus Torvalds low_limit = mmap_min_addr; 15846b008640SLinus Torvalds high_limit = info->high_limit; 158558c5d0d6SLiam R. Howlett retry: 15866b008640SLinus Torvalds if (mas_empty_area(&mas, low_limit, high_limit - 1, length)) 15873499a131SLiam R. Howlett return -ENOMEM; 15883499a131SLiam R. Howlett 1589d4af56c5SLiam R. Howlett gap = mas.index; 1590d4af56c5SLiam R. Howlett gap += (info->align_offset - gap) & info->align_mask; 159158c5d0d6SLiam R. Howlett tmp = mas_next(&mas, ULONG_MAX); 15920266e7c5SRick Edgecombe if (tmp && (tmp->vm_flags & VM_STARTGAP_FLAGS)) { /* Avoid prev check if possible */ 159358c5d0d6SLiam R. Howlett if (vm_start_gap(tmp) < gap + length - 1) { 159458c5d0d6SLiam R. Howlett low_limit = tmp->vm_end; 159558c5d0d6SLiam R. Howlett mas_reset(&mas); 159658c5d0d6SLiam R. Howlett goto retry; 159758c5d0d6SLiam R. Howlett } 159858c5d0d6SLiam R. Howlett } else { 159958c5d0d6SLiam R. Howlett tmp = mas_prev(&mas, 0); 160058c5d0d6SLiam R. Howlett if (tmp && vm_end_gap(tmp) > gap) { 160158c5d0d6SLiam R. Howlett low_limit = vm_end_gap(tmp); 160258c5d0d6SLiam R. Howlett mas_reset(&mas); 160358c5d0d6SLiam R. Howlett goto retry; 160458c5d0d6SLiam R. Howlett } 160558c5d0d6SLiam R. Howlett } 160658c5d0d6SLiam R. Howlett 16073499a131SLiam R. Howlett return gap; 1608db4fbfb9SMichel Lespinasse } 1609db4fbfb9SMichel Lespinasse 16103499a131SLiam R. Howlett /** 16113499a131SLiam R. Howlett * unmapped_area_topdown() - Find an area between the low_limit and the 161282b24936SVernon Yang * high_limit with the correct alignment and offset at the highest available 16133499a131SLiam R. Howlett * address, all from @info. Note: current->mm is used for the search. 16143499a131SLiam R. Howlett * 161582b24936SVernon Yang * @info: The unmapped area information including the range [low_limit - 161682b24936SVernon Yang * high_limit), the alignment offset and mask. 16173499a131SLiam R. Howlett * 16183499a131SLiam R. Howlett * Return: A memory address or -ENOMEM. 16193499a131SLiam R. Howlett */ 1620baceaf1cSJaewon Kim static unsigned long unmapped_area_topdown(struct vm_unmapped_area_info *info) 1621db4fbfb9SMichel Lespinasse { 16226b008640SLinus Torvalds unsigned long length, gap, gap_end; 16236b008640SLinus Torvalds unsigned long low_limit, high_limit; 162458c5d0d6SLiam R. Howlett struct vm_area_struct *tmp; 1625d4af56c5SLiam R. Howlett 16263499a131SLiam R. Howlett MA_STATE(mas, ¤t->mm->mm_mt, 0, 0); 1627db4fbfb9SMichel Lespinasse /* Adjust search length to account for worst case alignment overhead */ 1628db4fbfb9SMichel Lespinasse length = info->length + info->align_mask; 1629db4fbfb9SMichel Lespinasse if (length < info->length) 1630db4fbfb9SMichel Lespinasse return -ENOMEM; 1631db4fbfb9SMichel Lespinasse 16326b008640SLinus Torvalds low_limit = info->low_limit; 16336b008640SLinus Torvalds if (low_limit < mmap_min_addr) 16346b008640SLinus Torvalds low_limit = mmap_min_addr; 163558c5d0d6SLiam R. Howlett high_limit = info->high_limit; 163658c5d0d6SLiam R. Howlett retry: 16376b008640SLinus Torvalds if (mas_empty_area_rev(&mas, low_limit, high_limit - 1, length)) 16383499a131SLiam R. Howlett return -ENOMEM; 16393499a131SLiam R. Howlett 1640d4af56c5SLiam R. Howlett gap = mas.last + 1 - info->length; 1641d4af56c5SLiam R. Howlett gap -= (gap - info->align_offset) & info->align_mask; 164258c5d0d6SLiam R. Howlett gap_end = mas.last; 164358c5d0d6SLiam R. Howlett tmp = mas_next(&mas, ULONG_MAX); 16440266e7c5SRick Edgecombe if (tmp && (tmp->vm_flags & VM_STARTGAP_FLAGS)) { /* Avoid prev check if possible */ 164558c5d0d6SLiam R. Howlett if (vm_start_gap(tmp) <= gap_end) { 164658c5d0d6SLiam R. Howlett high_limit = vm_start_gap(tmp); 164758c5d0d6SLiam R. Howlett mas_reset(&mas); 164858c5d0d6SLiam R. Howlett goto retry; 164958c5d0d6SLiam R. Howlett } 165058c5d0d6SLiam R. Howlett } else { 165158c5d0d6SLiam R. Howlett tmp = mas_prev(&mas, 0); 165258c5d0d6SLiam R. Howlett if (tmp && vm_end_gap(tmp) > gap) { 165358c5d0d6SLiam R. Howlett high_limit = tmp->vm_start; 165458c5d0d6SLiam R. Howlett mas_reset(&mas); 165558c5d0d6SLiam R. Howlett goto retry; 165658c5d0d6SLiam R. Howlett } 165758c5d0d6SLiam R. Howlett } 165858c5d0d6SLiam R. Howlett 16593499a131SLiam R. Howlett return gap; 1660db4fbfb9SMichel Lespinasse } 1661db4fbfb9SMichel Lespinasse 1662baceaf1cSJaewon Kim /* 1663baceaf1cSJaewon Kim * Search for an unmapped address range. 1664baceaf1cSJaewon Kim * 1665baceaf1cSJaewon Kim * We are looking for a range that: 1666baceaf1cSJaewon Kim * - does not intersect with any VMA; 1667baceaf1cSJaewon Kim * - is contained within the [low_limit, high_limit) interval; 1668baceaf1cSJaewon Kim * - is at least the desired size. 1669baceaf1cSJaewon Kim * - satisfies (begin_addr & align_mask) == (align_offset & align_mask) 1670baceaf1cSJaewon Kim */ 1671baceaf1cSJaewon Kim unsigned long vm_unmapped_area(struct vm_unmapped_area_info *info) 1672baceaf1cSJaewon Kim { 1673df529cabSJaewon Kim unsigned long addr; 1674df529cabSJaewon Kim 1675baceaf1cSJaewon Kim if (info->flags & VM_UNMAPPED_AREA_TOPDOWN) 1676df529cabSJaewon Kim addr = unmapped_area_topdown(info); 1677baceaf1cSJaewon Kim else 1678df529cabSJaewon Kim addr = unmapped_area(info); 1679df529cabSJaewon Kim 1680df529cabSJaewon Kim trace_vm_unmapped_area(addr, info); 1681df529cabSJaewon Kim return addr; 1682baceaf1cSJaewon Kim } 1683f6795053SSteve Capper 16841da177e4SLinus Torvalds /* Get an address range which is currently unmapped. 16851da177e4SLinus Torvalds * For shmat() with addr=0. 16861da177e4SLinus Torvalds * 16871da177e4SLinus Torvalds * Ugly calling convention alert: 16881da177e4SLinus Torvalds * Return value with the low bits set means error value, 16891da177e4SLinus Torvalds * ie 16901da177e4SLinus Torvalds * if (ret & ~PAGE_MASK) 16911da177e4SLinus Torvalds * error = ret; 16921da177e4SLinus Torvalds * 16931da177e4SLinus Torvalds * This function "knows" that -ENOMEM has the bits set. 16941da177e4SLinus Torvalds */ 16951da177e4SLinus Torvalds unsigned long 16964b439e25SChristophe Leroy generic_get_unmapped_area(struct file *filp, unsigned long addr, 16974b439e25SChristophe Leroy unsigned long len, unsigned long pgoff, 16984b439e25SChristophe Leroy unsigned long flags) 16991da177e4SLinus Torvalds { 17001da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 17011be7107fSHugh Dickins struct vm_area_struct *vma, *prev; 1702db4fbfb9SMichel Lespinasse struct vm_unmapped_area_info info; 17032cb4de08SChristophe Leroy const unsigned long mmap_end = arch_get_mmap_end(addr, len, flags); 17041da177e4SLinus Torvalds 1705f6795053SSteve Capper if (len > mmap_end - mmap_min_addr) 17061da177e4SLinus Torvalds return -ENOMEM; 17071da177e4SLinus Torvalds 170806abdfb4SBenjamin Herrenschmidt if (flags & MAP_FIXED) 170906abdfb4SBenjamin Herrenschmidt return addr; 171006abdfb4SBenjamin Herrenschmidt 17111da177e4SLinus Torvalds if (addr) { 17121da177e4SLinus Torvalds addr = PAGE_ALIGN(addr); 17131be7107fSHugh Dickins vma = find_vma_prev(mm, addr, &prev); 1714f6795053SSteve Capper if (mmap_end - len >= addr && addr >= mmap_min_addr && 17151be7107fSHugh Dickins (!vma || addr + len <= vm_start_gap(vma)) && 17161be7107fSHugh Dickins (!prev || addr >= vm_end_gap(prev))) 17171da177e4SLinus Torvalds return addr; 17181da177e4SLinus Torvalds } 17191da177e4SLinus Torvalds 1720db4fbfb9SMichel Lespinasse info.flags = 0; 1721db4fbfb9SMichel Lespinasse info.length = len; 17224e99b021SHeiko Carstens info.low_limit = mm->mmap_base; 1723f6795053SSteve Capper info.high_limit = mmap_end; 1724db4fbfb9SMichel Lespinasse info.align_mask = 0; 172509ef5283SJaewon Kim info.align_offset = 0; 1726db4fbfb9SMichel Lespinasse return vm_unmapped_area(&info); 17271da177e4SLinus Torvalds } 17284b439e25SChristophe Leroy 17294b439e25SChristophe Leroy #ifndef HAVE_ARCH_UNMAPPED_AREA 17304b439e25SChristophe Leroy unsigned long 17314b439e25SChristophe Leroy arch_get_unmapped_area(struct file *filp, unsigned long addr, 17324b439e25SChristophe Leroy unsigned long len, unsigned long pgoff, 17334b439e25SChristophe Leroy unsigned long flags) 17344b439e25SChristophe Leroy { 17354b439e25SChristophe Leroy return generic_get_unmapped_area(filp, addr, len, pgoff, flags); 17364b439e25SChristophe Leroy } 17371da177e4SLinus Torvalds #endif 17381da177e4SLinus Torvalds 17391da177e4SLinus Torvalds /* 17401da177e4SLinus Torvalds * This mmap-allocator allocates new areas top-down from below the 17411da177e4SLinus Torvalds * stack's low limit (the base): 17421da177e4SLinus Torvalds */ 17431da177e4SLinus Torvalds unsigned long 17444b439e25SChristophe Leroy generic_get_unmapped_area_topdown(struct file *filp, unsigned long addr, 174543cca0b1SYang Fan unsigned long len, unsigned long pgoff, 174643cca0b1SYang Fan unsigned long flags) 17471da177e4SLinus Torvalds { 17481be7107fSHugh Dickins struct vm_area_struct *vma, *prev; 17491da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 1750db4fbfb9SMichel Lespinasse struct vm_unmapped_area_info info; 17512cb4de08SChristophe Leroy const unsigned long mmap_end = arch_get_mmap_end(addr, len, flags); 17521da177e4SLinus Torvalds 17531da177e4SLinus Torvalds /* requested length too big for entire address space */ 1754f6795053SSteve Capper if (len > mmap_end - mmap_min_addr) 17551da177e4SLinus Torvalds return -ENOMEM; 17561da177e4SLinus Torvalds 175706abdfb4SBenjamin Herrenschmidt if (flags & MAP_FIXED) 175806abdfb4SBenjamin Herrenschmidt return addr; 175906abdfb4SBenjamin Herrenschmidt 17601da177e4SLinus Torvalds /* requesting a specific address */ 17611da177e4SLinus Torvalds if (addr) { 17621da177e4SLinus Torvalds addr = PAGE_ALIGN(addr); 17631be7107fSHugh Dickins vma = find_vma_prev(mm, addr, &prev); 1764f6795053SSteve Capper if (mmap_end - len >= addr && addr >= mmap_min_addr && 17651be7107fSHugh Dickins (!vma || addr + len <= vm_start_gap(vma)) && 17661be7107fSHugh Dickins (!prev || addr >= vm_end_gap(prev))) 17671da177e4SLinus Torvalds return addr; 17681da177e4SLinus Torvalds } 17691da177e4SLinus Torvalds 1770db4fbfb9SMichel Lespinasse info.flags = VM_UNMAPPED_AREA_TOPDOWN; 1771db4fbfb9SMichel Lespinasse info.length = len; 17726b008640SLinus Torvalds info.low_limit = PAGE_SIZE; 1773f6795053SSteve Capper info.high_limit = arch_get_mmap_base(addr, mm->mmap_base); 1774db4fbfb9SMichel Lespinasse info.align_mask = 0; 177509ef5283SJaewon Kim info.align_offset = 0; 1776db4fbfb9SMichel Lespinasse addr = vm_unmapped_area(&info); 1777b716ad95SXiao Guangrong 17781da177e4SLinus Torvalds /* 17791da177e4SLinus Torvalds * A failed mmap() very likely causes application failure, 17801da177e4SLinus Torvalds * so fall back to the bottom-up function here. This scenario 17811da177e4SLinus Torvalds * can happen with large stack limits and large mmap() 17821da177e4SLinus Torvalds * allocations. 17831da177e4SLinus Torvalds */ 1784de1741a1SAlexander Kuleshov if (offset_in_page(addr)) { 1785db4fbfb9SMichel Lespinasse VM_BUG_ON(addr != -ENOMEM); 1786db4fbfb9SMichel Lespinasse info.flags = 0; 1787db4fbfb9SMichel Lespinasse info.low_limit = TASK_UNMAPPED_BASE; 1788f6795053SSteve Capper info.high_limit = mmap_end; 1789db4fbfb9SMichel Lespinasse addr = vm_unmapped_area(&info); 1790db4fbfb9SMichel Lespinasse } 17911da177e4SLinus Torvalds 17921da177e4SLinus Torvalds return addr; 17931da177e4SLinus Torvalds } 17944b439e25SChristophe Leroy 17954b439e25SChristophe Leroy #ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN 17964b439e25SChristophe Leroy unsigned long 17974b439e25SChristophe Leroy arch_get_unmapped_area_topdown(struct file *filp, unsigned long addr, 17984b439e25SChristophe Leroy unsigned long len, unsigned long pgoff, 17994b439e25SChristophe Leroy unsigned long flags) 18004b439e25SChristophe Leroy { 18014b439e25SChristophe Leroy return generic_get_unmapped_area_topdown(filp, addr, len, pgoff, flags); 18024b439e25SChristophe Leroy } 18031da177e4SLinus Torvalds #endif 18041da177e4SLinus Torvalds 18051da177e4SLinus Torvalds unsigned long 18061da177e4SLinus Torvalds get_unmapped_area(struct file *file, unsigned long addr, unsigned long len, 18071da177e4SLinus Torvalds unsigned long pgoff, unsigned long flags) 18081da177e4SLinus Torvalds { 180906abdfb4SBenjamin Herrenschmidt unsigned long (*get_area)(struct file *, unsigned long, 181006abdfb4SBenjamin Herrenschmidt unsigned long, unsigned long, unsigned long); 181107ab67c8SLinus Torvalds 18129206de95SAl Viro unsigned long error = arch_mmap_check(addr, len, flags); 18139206de95SAl Viro if (error) 18149206de95SAl Viro return error; 18159206de95SAl Viro 18169206de95SAl Viro /* Careful about overflows.. */ 18179206de95SAl Viro if (len > TASK_SIZE) 18189206de95SAl Viro return -ENOMEM; 18199206de95SAl Viro 182007ab67c8SLinus Torvalds get_area = current->mm->get_unmapped_area; 1821c01d5b30SHugh Dickins if (file) { 1822c01d5b30SHugh Dickins if (file->f_op->get_unmapped_area) 182307ab67c8SLinus Torvalds get_area = file->f_op->get_unmapped_area; 1824c01d5b30SHugh Dickins } else if (flags & MAP_SHARED) { 1825c01d5b30SHugh Dickins /* 1826c01d5b30SHugh Dickins * mmap_region() will call shmem_zero_setup() to create a file, 1827c01d5b30SHugh Dickins * so use shmem's get_unmapped_area in case it can be huge. 182845e55300SPeter Collingbourne * do_mmap() will clear pgoff, so match alignment. 1829c01d5b30SHugh Dickins */ 1830c01d5b30SHugh Dickins pgoff = 0; 1831c01d5b30SHugh Dickins get_area = shmem_get_unmapped_area; 1832*efa7df3eSRik van Riel } else if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE)) { 1833*efa7df3eSRik van Riel /* Ensures that larger anonymous mappings are THP aligned. */ 1834*efa7df3eSRik van Riel get_area = thp_get_unmapped_area; 1835c01d5b30SHugh Dickins } 1836c01d5b30SHugh Dickins 183707ab67c8SLinus Torvalds addr = get_area(file, addr, len, pgoff, flags); 183807ab67c8SLinus Torvalds if (IS_ERR_VALUE(addr)) 183907ab67c8SLinus Torvalds return addr; 184007ab67c8SLinus Torvalds 18411da177e4SLinus Torvalds if (addr > TASK_SIZE - len) 18421da177e4SLinus Torvalds return -ENOMEM; 1843de1741a1SAlexander Kuleshov if (offset_in_page(addr)) 18441da177e4SLinus Torvalds return -EINVAL; 184506abdfb4SBenjamin Herrenschmidt 18469ac4ed4bSAl Viro error = security_mmap_addr(addr); 18479ac4ed4bSAl Viro return error ? error : addr; 18481da177e4SLinus Torvalds } 18491da177e4SLinus Torvalds 18501da177e4SLinus Torvalds EXPORT_SYMBOL(get_unmapped_area); 18511da177e4SLinus Torvalds 1852be8432e7SLiam R. Howlett /** 1853abdba2ddSLiam R. Howlett * find_vma_intersection() - Look up the first VMA which intersects the interval 1854abdba2ddSLiam R. Howlett * @mm: The process address space. 1855abdba2ddSLiam R. Howlett * @start_addr: The inclusive start user address. 1856abdba2ddSLiam R. Howlett * @end_addr: The exclusive end user address. 1857abdba2ddSLiam R. Howlett * 1858abdba2ddSLiam R. Howlett * Returns: The first VMA within the provided range, %NULL otherwise. Assumes 1859abdba2ddSLiam R. Howlett * start_addr < end_addr. 1860abdba2ddSLiam R. Howlett */ 1861abdba2ddSLiam R. Howlett struct vm_area_struct *find_vma_intersection(struct mm_struct *mm, 1862abdba2ddSLiam R. Howlett unsigned long start_addr, 1863abdba2ddSLiam R. Howlett unsigned long end_addr) 1864abdba2ddSLiam R. Howlett { 1865abdba2ddSLiam R. Howlett unsigned long index = start_addr; 1866abdba2ddSLiam R. Howlett 1867abdba2ddSLiam R. Howlett mmap_assert_locked(mm); 18687964cf8cSLiam R. Howlett return mt_find(&mm->mm_mt, &index, end_addr - 1); 1869abdba2ddSLiam R. Howlett } 1870abdba2ddSLiam R. Howlett EXPORT_SYMBOL(find_vma_intersection); 1871abdba2ddSLiam R. Howlett 1872abdba2ddSLiam R. Howlett /** 1873be8432e7SLiam R. Howlett * find_vma() - Find the VMA for a given address, or the next VMA. 1874be8432e7SLiam R. Howlett * @mm: The mm_struct to check 1875be8432e7SLiam R. Howlett * @addr: The address 1876be8432e7SLiam R. Howlett * 1877be8432e7SLiam R. Howlett * Returns: The VMA associated with addr, or the next VMA. 1878be8432e7SLiam R. Howlett * May return %NULL in the case of no VMA at addr or above. 1879be8432e7SLiam R. Howlett */ 18801da177e4SLinus Torvalds struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr) 18811da177e4SLinus Torvalds { 1882be8432e7SLiam R. Howlett unsigned long index = addr; 18831da177e4SLinus Torvalds 18845b78ed24SLuigi Rizzo mmap_assert_locked(mm); 18857964cf8cSLiam R. Howlett return mt_find(&mm->mm_mt, &index, ULONG_MAX); 18861da177e4SLinus Torvalds } 18871da177e4SLinus Torvalds EXPORT_SYMBOL(find_vma); 18881da177e4SLinus Torvalds 18897fdbd37dSLiam R. Howlett /** 18907fdbd37dSLiam R. Howlett * find_vma_prev() - Find the VMA for a given address, or the next vma and 18917fdbd37dSLiam R. Howlett * set %pprev to the previous VMA, if any. 18927fdbd37dSLiam R. Howlett * @mm: The mm_struct to check 18937fdbd37dSLiam R. Howlett * @addr: The address 18947fdbd37dSLiam R. Howlett * @pprev: The pointer to set to the previous VMA 18957fdbd37dSLiam R. Howlett * 18967fdbd37dSLiam R. Howlett * Note that RCU lock is missing here since the external mmap_lock() is used 18977fdbd37dSLiam R. Howlett * instead. 18987fdbd37dSLiam R. Howlett * 18997fdbd37dSLiam R. Howlett * Returns: The VMA associated with @addr, or the next vma. 19007fdbd37dSLiam R. Howlett * May return %NULL in the case of no vma at addr or above. 19016bd4837dSKOSAKI Motohiro */ 19021da177e4SLinus Torvalds struct vm_area_struct * 19031da177e4SLinus Torvalds find_vma_prev(struct mm_struct *mm, unsigned long addr, 19041da177e4SLinus Torvalds struct vm_area_struct **pprev) 19051da177e4SLinus Torvalds { 19066bd4837dSKOSAKI Motohiro struct vm_area_struct *vma; 19077fdbd37dSLiam R. Howlett MA_STATE(mas, &mm->mm_mt, addr, addr); 19081da177e4SLinus Torvalds 19097fdbd37dSLiam R. Howlett vma = mas_walk(&mas); 19107fdbd37dSLiam R. Howlett *pprev = mas_prev(&mas, 0); 19117fdbd37dSLiam R. Howlett if (!vma) 19127fdbd37dSLiam R. Howlett vma = mas_next(&mas, ULONG_MAX); 19136bd4837dSKOSAKI Motohiro return vma; 19141da177e4SLinus Torvalds } 19151da177e4SLinus Torvalds 19161da177e4SLinus Torvalds /* 19171da177e4SLinus Torvalds * Verify that the stack growth is acceptable and 19181da177e4SLinus Torvalds * update accounting. This is shared with both the 19191da177e4SLinus Torvalds * grow-up and grow-down cases. 19201da177e4SLinus Torvalds */ 19211be7107fSHugh Dickins static int acct_stack_growth(struct vm_area_struct *vma, 19221be7107fSHugh Dickins unsigned long size, unsigned long grow) 19231da177e4SLinus Torvalds { 19241da177e4SLinus Torvalds struct mm_struct *mm = vma->vm_mm; 19251be7107fSHugh Dickins unsigned long new_start; 19261da177e4SLinus Torvalds 19271da177e4SLinus Torvalds /* address space limit tests */ 192884638335SKonstantin Khlebnikov if (!may_expand_vm(mm, vma->vm_flags, grow)) 19291da177e4SLinus Torvalds return -ENOMEM; 19301da177e4SLinus Torvalds 19311da177e4SLinus Torvalds /* Stack limit test */ 193224c79d8eSKrzysztof Opasiak if (size > rlimit(RLIMIT_STACK)) 19331da177e4SLinus Torvalds return -ENOMEM; 19341da177e4SLinus Torvalds 19351da177e4SLinus Torvalds /* mlock limit tests */ 1936b0cc5e89SAndrew Morton if (!mlock_future_ok(mm, vma->vm_flags, grow << PAGE_SHIFT)) 19371da177e4SLinus Torvalds return -ENOMEM; 19381da177e4SLinus Torvalds 19390d59a01bSAdam Litke /* Check to ensure the stack will not grow into a hugetlb-only region */ 19400d59a01bSAdam Litke new_start = (vma->vm_flags & VM_GROWSUP) ? vma->vm_start : 19410d59a01bSAdam Litke vma->vm_end - size; 19420d59a01bSAdam Litke if (is_hugepage_only_range(vma->vm_mm, new_start, size)) 19430d59a01bSAdam Litke return -EFAULT; 19440d59a01bSAdam Litke 19451da177e4SLinus Torvalds /* 19461da177e4SLinus Torvalds * Overcommit.. This must be the final test, as it will 19471da177e4SLinus Torvalds * update security statistics. 19481da177e4SLinus Torvalds */ 194905fa199dSHugh Dickins if (security_vm_enough_memory_mm(mm, grow)) 19501da177e4SLinus Torvalds return -ENOMEM; 19511da177e4SLinus Torvalds 19521da177e4SLinus Torvalds return 0; 19531da177e4SLinus Torvalds } 19541da177e4SLinus Torvalds 1955cf8e8658SArd Biesheuvel #if defined(CONFIG_STACK_GROWSUP) 19561da177e4SLinus Torvalds /* 1957cf8e8658SArd Biesheuvel * PA-RISC uses this for its stack. 195846dea3d0SHugh Dickins * vma is the last one with address > vma->vm_end. Have to extend vma. 19591da177e4SLinus Torvalds */ 19608d7071afSLinus Torvalds static int expand_upwards(struct vm_area_struct *vma, unsigned long address) 19611da177e4SLinus Torvalds { 196209357814SOleg Nesterov struct mm_struct *mm = vma->vm_mm; 19631be7107fSHugh Dickins struct vm_area_struct *next; 19641be7107fSHugh Dickins unsigned long gap_addr; 196512352d3cSKonstantin Khlebnikov int error = 0; 1966b5df0922SLiam R. Howlett MA_STATE(mas, &mm->mm_mt, vma->vm_start, address); 19671da177e4SLinus Torvalds 19681da177e4SLinus Torvalds if (!(vma->vm_flags & VM_GROWSUP)) 19691da177e4SLinus Torvalds return -EFAULT; 19701da177e4SLinus Torvalds 1971bd726c90SHelge Deller /* Guard against exceeding limits of the address space. */ 19721be7107fSHugh Dickins address &= PAGE_MASK; 197337511fb5SHelge Deller if (address >= (TASK_SIZE & PAGE_MASK)) 197412352d3cSKonstantin Khlebnikov return -ENOMEM; 1975bd726c90SHelge Deller address += PAGE_SIZE; 197612352d3cSKonstantin Khlebnikov 19771be7107fSHugh Dickins /* Enforce stack_guard_gap */ 19781be7107fSHugh Dickins gap_addr = address + stack_guard_gap; 1979bd726c90SHelge Deller 1980bd726c90SHelge Deller /* Guard against overflow */ 1981bd726c90SHelge Deller if (gap_addr < address || gap_addr > TASK_SIZE) 1982bd726c90SHelge Deller gap_addr = TASK_SIZE; 1983bd726c90SHelge Deller 1984763ecb03SLiam R. Howlett next = find_vma_intersection(mm, vma->vm_end, gap_addr); 1985763ecb03SLiam R. Howlett if (next && vma_is_accessible(next)) { 19861be7107fSHugh Dickins if (!(next->vm_flags & VM_GROWSUP)) 19871be7107fSHugh Dickins return -ENOMEM; 19881be7107fSHugh Dickins /* Check that both stack segments have the same anon_vma? */ 19891be7107fSHugh Dickins } 19901be7107fSHugh Dickins 1991b5df0922SLiam R. Howlett if (next) 1992b5df0922SLiam R. Howlett mas_prev_range(&mas, address); 1993b5df0922SLiam R. Howlett 1994b5df0922SLiam R. Howlett __mas_set_range(&mas, vma->vm_start, address - 1); 1995da089254SLiam R. Howlett if (mas_preallocate(&mas, vma, GFP_KERNEL)) 19961da177e4SLinus Torvalds return -ENOMEM; 19971da177e4SLinus Torvalds 1998d4af56c5SLiam R. Howlett /* We must make sure the anon_vma is allocated. */ 1999d4af56c5SLiam R. Howlett if (unlikely(anon_vma_prepare(vma))) { 2000d4af56c5SLiam R. Howlett mas_destroy(&mas); 2001d4af56c5SLiam R. Howlett return -ENOMEM; 2002d4af56c5SLiam R. Howlett } 2003d4af56c5SLiam R. Howlett 2004c137381fSSuren Baghdasaryan /* Lock the VMA before expanding to prevent concurrent page faults */ 2005c137381fSSuren Baghdasaryan vma_start_write(vma); 20061da177e4SLinus Torvalds /* 20071da177e4SLinus Torvalds * vma->vm_start/vm_end cannot change under us because the caller 2008c1e8d7c6SMichel Lespinasse * is required to hold the mmap_lock in read mode. We need the 20091da177e4SLinus Torvalds * anon_vma lock to serialize against concurrent expand_stacks. 20101da177e4SLinus Torvalds */ 201112352d3cSKonstantin Khlebnikov anon_vma_lock_write(vma->anon_vma); 20121da177e4SLinus Torvalds 20131da177e4SLinus Torvalds /* Somebody else might have raced and expanded it already */ 20141da177e4SLinus Torvalds if (address > vma->vm_end) { 20151da177e4SLinus Torvalds unsigned long size, grow; 20161da177e4SLinus Torvalds 20171da177e4SLinus Torvalds size = address - vma->vm_start; 20181da177e4SLinus Torvalds grow = (address - vma->vm_end) >> PAGE_SHIFT; 20191da177e4SLinus Torvalds 202042c36f63SHugh Dickins error = -ENOMEM; 202142c36f63SHugh Dickins if (vma->vm_pgoff + (size >> PAGE_SHIFT) >= vma->vm_pgoff) { 20221da177e4SLinus Torvalds error = acct_stack_growth(vma, size, grow); 20233af9e859SEric B Munson if (!error) { 20244128997bSMichel Lespinasse /* 2025524e00b3SLiam R. Howlett * We only hold a shared mmap_lock lock here, so 2026524e00b3SLiam R. Howlett * we need to protect against concurrent vma 2027524e00b3SLiam R. Howlett * expansions. anon_vma_lock_write() doesn't 2028524e00b3SLiam R. Howlett * help here, as we don't guarantee that all 2029524e00b3SLiam R. Howlett * growable vmas in a mm share the same root 2030524e00b3SLiam R. Howlett * anon vma. So, we reuse mm->page_table_lock 2031524e00b3SLiam R. Howlett * to guard against concurrent vma expansions. 20324128997bSMichel Lespinasse */ 203309357814SOleg Nesterov spin_lock(&mm->page_table_lock); 203487e8827bSOleg Nesterov if (vma->vm_flags & VM_LOCKED) 203509357814SOleg Nesterov mm->locked_vm += grow; 203684638335SKonstantin Khlebnikov vm_stat_account(mm, vma->vm_flags, grow); 2037bf181b9fSMichel Lespinasse anon_vma_interval_tree_pre_update_vma(vma); 20381da177e4SLinus Torvalds vma->vm_end = address; 2039d4af56c5SLiam R. Howlett /* Overwrite old entry in mtree. */ 2040fbcc3104SLiam R. Howlett mas_store_prealloc(&mas, vma); 2041bf181b9fSMichel Lespinasse anon_vma_interval_tree_post_update_vma(vma); 204209357814SOleg Nesterov spin_unlock(&mm->page_table_lock); 20434128997bSMichel Lespinasse 20443af9e859SEric B Munson perf_event_mmap(vma); 20453af9e859SEric B Munson } 20461da177e4SLinus Torvalds } 204742c36f63SHugh Dickins } 204812352d3cSKonstantin Khlebnikov anon_vma_unlock_write(vma->anon_vma); 2049c791576cSYang Shi khugepaged_enter_vma(vma, vma->vm_flags); 2050d4af56c5SLiam R. Howlett mas_destroy(&mas); 20512574d5e4SLiam R. Howlett validate_mm(mm); 20521da177e4SLinus Torvalds return error; 20531da177e4SLinus Torvalds } 2054cf8e8658SArd Biesheuvel #endif /* CONFIG_STACK_GROWSUP */ 205546dea3d0SHugh Dickins 20561da177e4SLinus Torvalds /* 20571da177e4SLinus Torvalds * vma is the first one with address < vma->vm_start. Have to extend vma. 20588d7071afSLinus Torvalds * mmap_lock held for writing. 20591da177e4SLinus Torvalds */ 2060524e00b3SLiam R. Howlett int expand_downwards(struct vm_area_struct *vma, unsigned long address) 20611da177e4SLinus Torvalds { 206209357814SOleg Nesterov struct mm_struct *mm = vma->vm_mm; 2063763ecb03SLiam R. Howlett MA_STATE(mas, &mm->mm_mt, vma->vm_start, vma->vm_start); 20641be7107fSHugh Dickins struct vm_area_struct *prev; 20650a1d5299SJann Horn int error = 0; 20661da177e4SLinus Torvalds 20678d7071afSLinus Torvalds if (!(vma->vm_flags & VM_GROWSDOWN)) 20688d7071afSLinus Torvalds return -EFAULT; 20698d7071afSLinus Torvalds 20708869477aSEric Paris address &= PAGE_MASK; 20718b35ca3eSBen Hutchings if (address < mmap_min_addr || address < FIRST_USER_ADDRESS) 20720a1d5299SJann Horn return -EPERM; 20738869477aSEric Paris 20741be7107fSHugh Dickins /* Enforce stack_guard_gap */ 2075763ecb03SLiam R. Howlett prev = mas_prev(&mas, 0); 20761be7107fSHugh Dickins /* Check that both stack segments have the same anon_vma? */ 2077f440fa1aSLiam R. Howlett if (prev) { 2078f440fa1aSLiam R. Howlett if (!(prev->vm_flags & VM_GROWSDOWN) && 2079f440fa1aSLiam R. Howlett vma_is_accessible(prev) && 2080f440fa1aSLiam R. Howlett (address - prev->vm_end < stack_guard_gap)) 208132e4e6d5SOleg Nesterov return -ENOMEM; 20821be7107fSHugh Dickins } 20831be7107fSHugh Dickins 2084b5df0922SLiam R. Howlett if (prev) 2085b5df0922SLiam R. Howlett mas_next_range(&mas, vma->vm_start); 2086b5df0922SLiam R. Howlett 2087b5df0922SLiam R. Howlett __mas_set_range(&mas, address, vma->vm_end - 1); 2088da089254SLiam R. Howlett if (mas_preallocate(&mas, vma, GFP_KERNEL)) 208912352d3cSKonstantin Khlebnikov return -ENOMEM; 20901da177e4SLinus Torvalds 2091d4af56c5SLiam R. Howlett /* We must make sure the anon_vma is allocated. */ 2092d4af56c5SLiam R. Howlett if (unlikely(anon_vma_prepare(vma))) { 2093d4af56c5SLiam R. Howlett mas_destroy(&mas); 2094d4af56c5SLiam R. Howlett return -ENOMEM; 2095d4af56c5SLiam R. Howlett } 2096d4af56c5SLiam R. Howlett 2097c137381fSSuren Baghdasaryan /* Lock the VMA before expanding to prevent concurrent page faults */ 2098c137381fSSuren Baghdasaryan vma_start_write(vma); 20991da177e4SLinus Torvalds /* 21001da177e4SLinus Torvalds * vma->vm_start/vm_end cannot change under us because the caller 2101c1e8d7c6SMichel Lespinasse * is required to hold the mmap_lock in read mode. We need the 21021da177e4SLinus Torvalds * anon_vma lock to serialize against concurrent expand_stacks. 21031da177e4SLinus Torvalds */ 210412352d3cSKonstantin Khlebnikov anon_vma_lock_write(vma->anon_vma); 21051da177e4SLinus Torvalds 21061da177e4SLinus Torvalds /* Somebody else might have raced and expanded it already */ 21071da177e4SLinus Torvalds if (address < vma->vm_start) { 21081da177e4SLinus Torvalds unsigned long size, grow; 21091da177e4SLinus Torvalds 21101da177e4SLinus Torvalds size = vma->vm_end - address; 21111da177e4SLinus Torvalds grow = (vma->vm_start - address) >> PAGE_SHIFT; 21121da177e4SLinus Torvalds 2113a626ca6aSLinus Torvalds error = -ENOMEM; 2114a626ca6aSLinus Torvalds if (grow <= vma->vm_pgoff) { 21151da177e4SLinus Torvalds error = acct_stack_growth(vma, size, grow); 21161da177e4SLinus Torvalds if (!error) { 21174128997bSMichel Lespinasse /* 2118524e00b3SLiam R. Howlett * We only hold a shared mmap_lock lock here, so 2119524e00b3SLiam R. Howlett * we need to protect against concurrent vma 2120524e00b3SLiam R. Howlett * expansions. anon_vma_lock_write() doesn't 2121524e00b3SLiam R. Howlett * help here, as we don't guarantee that all 2122524e00b3SLiam R. Howlett * growable vmas in a mm share the same root 2123524e00b3SLiam R. Howlett * anon vma. So, we reuse mm->page_table_lock 2124524e00b3SLiam R. Howlett * to guard against concurrent vma expansions. 21254128997bSMichel Lespinasse */ 212609357814SOleg Nesterov spin_lock(&mm->page_table_lock); 212787e8827bSOleg Nesterov if (vma->vm_flags & VM_LOCKED) 212809357814SOleg Nesterov mm->locked_vm += grow; 212984638335SKonstantin Khlebnikov vm_stat_account(mm, vma->vm_flags, grow); 2130bf181b9fSMichel Lespinasse anon_vma_interval_tree_pre_update_vma(vma); 21311da177e4SLinus Torvalds vma->vm_start = address; 21321da177e4SLinus Torvalds vma->vm_pgoff -= grow; 2133d4af56c5SLiam R. Howlett /* Overwrite old entry in mtree. */ 2134fbcc3104SLiam R. Howlett mas_store_prealloc(&mas, vma); 2135bf181b9fSMichel Lespinasse anon_vma_interval_tree_post_update_vma(vma); 213609357814SOleg Nesterov spin_unlock(&mm->page_table_lock); 21374128997bSMichel Lespinasse 21383af9e859SEric B Munson perf_event_mmap(vma); 21391da177e4SLinus Torvalds } 21401da177e4SLinus Torvalds } 2141a626ca6aSLinus Torvalds } 214212352d3cSKonstantin Khlebnikov anon_vma_unlock_write(vma->anon_vma); 2143c791576cSYang Shi khugepaged_enter_vma(vma, vma->vm_flags); 2144d4af56c5SLiam R. Howlett mas_destroy(&mas); 21452574d5e4SLiam R. Howlett validate_mm(mm); 21461da177e4SLinus Torvalds return error; 21471da177e4SLinus Torvalds } 21481da177e4SLinus Torvalds 21491be7107fSHugh Dickins /* enforced gap between the expanding stack and other mappings. */ 21501be7107fSHugh Dickins unsigned long stack_guard_gap = 256UL<<PAGE_SHIFT; 21511be7107fSHugh Dickins 21521be7107fSHugh Dickins static int __init cmdline_parse_stack_guard_gap(char *p) 21531be7107fSHugh Dickins { 21541be7107fSHugh Dickins unsigned long val; 21551be7107fSHugh Dickins char *endptr; 21561be7107fSHugh Dickins 21571be7107fSHugh Dickins val = simple_strtoul(p, &endptr, 10); 21581be7107fSHugh Dickins if (!*endptr) 21591be7107fSHugh Dickins stack_guard_gap = val << PAGE_SHIFT; 21601be7107fSHugh Dickins 2161e6d09493SRandy Dunlap return 1; 21621be7107fSHugh Dickins } 21631be7107fSHugh Dickins __setup("stack_guard_gap=", cmdline_parse_stack_guard_gap); 21641be7107fSHugh Dickins 2165b6a2fea3SOllie Wild #ifdef CONFIG_STACK_GROWSUP 21668d7071afSLinus Torvalds int expand_stack_locked(struct vm_area_struct *vma, unsigned long address) 2167b6a2fea3SOllie Wild { 2168b6a2fea3SOllie Wild return expand_upwards(vma, address); 2169b6a2fea3SOllie Wild } 2170b6a2fea3SOllie Wild 21718d7071afSLinus Torvalds struct vm_area_struct *find_extend_vma_locked(struct mm_struct *mm, unsigned long addr) 2172b6a2fea3SOllie Wild { 2173b6a2fea3SOllie Wild struct vm_area_struct *vma, *prev; 2174b6a2fea3SOllie Wild 2175b6a2fea3SOllie Wild addr &= PAGE_MASK; 2176b6a2fea3SOllie Wild vma = find_vma_prev(mm, addr, &prev); 2177b6a2fea3SOllie Wild if (vma && (vma->vm_start <= addr)) 2178b6a2fea3SOllie Wild return vma; 2179f440fa1aSLiam R. Howlett if (!prev) 2180f440fa1aSLiam R. Howlett return NULL; 21818d7071afSLinus Torvalds if (expand_stack_locked(prev, addr)) 2182b6a2fea3SOllie Wild return NULL; 2183cea10a19SMichel Lespinasse if (prev->vm_flags & VM_LOCKED) 2184fc05f566SKirill A. Shutemov populate_vma_page_range(prev, addr, prev->vm_end, NULL); 2185b6a2fea3SOllie Wild return prev; 2186b6a2fea3SOllie Wild } 2187b6a2fea3SOllie Wild #else 21888d7071afSLinus Torvalds int expand_stack_locked(struct vm_area_struct *vma, unsigned long address) 2189b6a2fea3SOllie Wild { 2190b6a2fea3SOllie Wild return expand_downwards(vma, address); 2191b6a2fea3SOllie Wild } 2192b6a2fea3SOllie Wild 21938d7071afSLinus Torvalds struct vm_area_struct *find_extend_vma_locked(struct mm_struct *mm, unsigned long addr) 21941da177e4SLinus Torvalds { 21951da177e4SLinus Torvalds struct vm_area_struct *vma; 21961da177e4SLinus Torvalds unsigned long start; 21971da177e4SLinus Torvalds 21981da177e4SLinus Torvalds addr &= PAGE_MASK; 21991da177e4SLinus Torvalds vma = find_vma(mm, addr); 22001da177e4SLinus Torvalds if (!vma) 22011da177e4SLinus Torvalds return NULL; 22021da177e4SLinus Torvalds if (vma->vm_start <= addr) 22031da177e4SLinus Torvalds return vma; 22041da177e4SLinus Torvalds start = vma->vm_start; 22058d7071afSLinus Torvalds if (expand_stack_locked(vma, addr)) 22061da177e4SLinus Torvalds return NULL; 2207cea10a19SMichel Lespinasse if (vma->vm_flags & VM_LOCKED) 2208fc05f566SKirill A. Shutemov populate_vma_page_range(vma, addr, start, NULL); 22091da177e4SLinus Torvalds return vma; 22101da177e4SLinus Torvalds } 22111da177e4SLinus Torvalds #endif 22121da177e4SLinus Torvalds 22138d7071afSLinus Torvalds /* 22148d7071afSLinus Torvalds * IA64 has some horrid mapping rules: it can expand both up and down, 22158d7071afSLinus Torvalds * but with various special rules. 22168d7071afSLinus Torvalds * 22178d7071afSLinus Torvalds * We'll get rid of this architecture eventually, so the ugliness is 22188d7071afSLinus Torvalds * temporary. 22198d7071afSLinus Torvalds */ 22208d7071afSLinus Torvalds #ifdef CONFIG_IA64 22218d7071afSLinus Torvalds static inline bool vma_expand_ok(struct vm_area_struct *vma, unsigned long addr) 2222f440fa1aSLiam R. Howlett { 22238d7071afSLinus Torvalds return REGION_NUMBER(addr) == REGION_NUMBER(vma->vm_start) && 22248d7071afSLinus Torvalds REGION_OFFSET(addr) < RGN_MAP_LIMIT; 2225f440fa1aSLiam R. Howlett } 22268d7071afSLinus Torvalds 22278d7071afSLinus Torvalds /* 22288d7071afSLinus Torvalds * IA64 stacks grow down, but there's a special register backing store 22298d7071afSLinus Torvalds * that can grow up. Only sequentially, though, so the new address must 22308d7071afSLinus Torvalds * match vm_end. 22318d7071afSLinus Torvalds */ 22328d7071afSLinus Torvalds static inline int vma_expand_up(struct vm_area_struct *vma, unsigned long addr) 22338d7071afSLinus Torvalds { 22348d7071afSLinus Torvalds if (!vma_expand_ok(vma, addr)) 22358d7071afSLinus Torvalds return -EFAULT; 22368d7071afSLinus Torvalds if (vma->vm_end != (addr & PAGE_MASK)) 22378d7071afSLinus Torvalds return -EFAULT; 22388d7071afSLinus Torvalds return expand_upwards(vma, addr); 22398d7071afSLinus Torvalds } 22408d7071afSLinus Torvalds 22418d7071afSLinus Torvalds static inline bool vma_expand_down(struct vm_area_struct *vma, unsigned long addr) 22428d7071afSLinus Torvalds { 22438d7071afSLinus Torvalds if (!vma_expand_ok(vma, addr)) 22448d7071afSLinus Torvalds return -EFAULT; 22458d7071afSLinus Torvalds return expand_downwards(vma, addr); 22468d7071afSLinus Torvalds } 22478d7071afSLinus Torvalds 22488d7071afSLinus Torvalds #elif defined(CONFIG_STACK_GROWSUP) 22498d7071afSLinus Torvalds 22508d7071afSLinus Torvalds #define vma_expand_up(vma,addr) expand_upwards(vma, addr) 22518d7071afSLinus Torvalds #define vma_expand_down(vma, addr) (-EFAULT) 22528d7071afSLinus Torvalds 22538d7071afSLinus Torvalds #else 22548d7071afSLinus Torvalds 22558d7071afSLinus Torvalds #define vma_expand_up(vma,addr) (-EFAULT) 22568d7071afSLinus Torvalds #define vma_expand_down(vma, addr) expand_downwards(vma, addr) 22578d7071afSLinus Torvalds 22588d7071afSLinus Torvalds #endif 22598d7071afSLinus Torvalds 22608d7071afSLinus Torvalds /* 22618d7071afSLinus Torvalds * expand_stack(): legacy interface for page faulting. Don't use unless 22628d7071afSLinus Torvalds * you have to. 22638d7071afSLinus Torvalds * 22648d7071afSLinus Torvalds * This is called with the mm locked for reading, drops the lock, takes 22658d7071afSLinus Torvalds * the lock for writing, tries to look up a vma again, expands it if 22668d7071afSLinus Torvalds * necessary, and downgrades the lock to reading again. 22678d7071afSLinus Torvalds * 22688d7071afSLinus Torvalds * If no vma is found or it can't be expanded, it returns NULL and has 22698d7071afSLinus Torvalds * dropped the lock. 22708d7071afSLinus Torvalds */ 22718d7071afSLinus Torvalds struct vm_area_struct *expand_stack(struct mm_struct *mm, unsigned long addr) 22728d7071afSLinus Torvalds { 22738d7071afSLinus Torvalds struct vm_area_struct *vma, *prev; 22748d7071afSLinus Torvalds 22758d7071afSLinus Torvalds mmap_read_unlock(mm); 22768d7071afSLinus Torvalds if (mmap_write_lock_killable(mm)) 22778d7071afSLinus Torvalds return NULL; 22788d7071afSLinus Torvalds 22798d7071afSLinus Torvalds vma = find_vma_prev(mm, addr, &prev); 22808d7071afSLinus Torvalds if (vma && vma->vm_start <= addr) 22818d7071afSLinus Torvalds goto success; 22828d7071afSLinus Torvalds 22838d7071afSLinus Torvalds if (prev && !vma_expand_up(prev, addr)) { 22848d7071afSLinus Torvalds vma = prev; 22858d7071afSLinus Torvalds goto success; 22868d7071afSLinus Torvalds } 22878d7071afSLinus Torvalds 22888d7071afSLinus Torvalds if (vma && !vma_expand_down(vma, addr)) 22898d7071afSLinus Torvalds goto success; 22908d7071afSLinus Torvalds 22918d7071afSLinus Torvalds mmap_write_unlock(mm); 22928d7071afSLinus Torvalds return NULL; 22938d7071afSLinus Torvalds 22948d7071afSLinus Torvalds success: 22958d7071afSLinus Torvalds mmap_write_downgrade(mm); 22968d7071afSLinus Torvalds return vma; 22978d7071afSLinus Torvalds } 2298e1d6d01aSJesse Barnes 22992c0b3814SHugh Dickins /* 2300763ecb03SLiam R. Howlett * Ok - we have the memory areas we should free on a maple tree so release them, 2301763ecb03SLiam R. Howlett * and do the vma updates. 23021da177e4SLinus Torvalds * 23032c0b3814SHugh Dickins * Called with the mm semaphore held. 23041da177e4SLinus Torvalds */ 2305763ecb03SLiam R. Howlett static inline void remove_mt(struct mm_struct *mm, struct ma_state *mas) 23061da177e4SLinus Torvalds { 23074f74d2c8SLinus Torvalds unsigned long nr_accounted = 0; 2308763ecb03SLiam R. Howlett struct vm_area_struct *vma; 23094f74d2c8SLinus Torvalds 2310365e9c87SHugh Dickins /* Update high watermark before we lower total_vm */ 2311365e9c87SHugh Dickins update_hiwater_vm(mm); 2312763ecb03SLiam R. Howlett mas_for_each(mas, vma, ULONG_MAX) { 2313ab50b8edSHugh Dickins long nrpages = vma_pages(vma); 23141da177e4SLinus Torvalds 23154f74d2c8SLinus Torvalds if (vma->vm_flags & VM_ACCOUNT) 23164f74d2c8SLinus Torvalds nr_accounted += nrpages; 231784638335SKonstantin Khlebnikov vm_stat_account(mm, vma->vm_flags, -nrpages); 23180d2ebf9cSSuren Baghdasaryan remove_vma(vma, false); 2319763ecb03SLiam R. Howlett } 23204f74d2c8SLinus Torvalds vm_unacct_memory(nr_accounted); 23211da177e4SLinus Torvalds } 23221da177e4SLinus Torvalds 23231da177e4SLinus Torvalds /* 23241da177e4SLinus Torvalds * Get rid of page table information in the indicated region. 23251da177e4SLinus Torvalds * 2326f10df686SPaolo 'Blaisorblade' Giarrusso * Called with the mm semaphore held. 23271da177e4SLinus Torvalds */ 2328fd892593SLiam R. Howlett static void unmap_region(struct mm_struct *mm, struct ma_state *mas, 2329e0da382cSHugh Dickins struct vm_area_struct *vma, struct vm_area_struct *prev, 2330fd892593SLiam R. Howlett struct vm_area_struct *next, unsigned long start, 2331fd892593SLiam R. Howlett unsigned long end, unsigned long tree_end, bool mm_wr_locked) 23321da177e4SLinus Torvalds { 2333d16dfc55SPeter Zijlstra struct mmu_gather tlb; 2334fd892593SLiam R. Howlett unsigned long mt_start = mas->index; 23351da177e4SLinus Torvalds 23361da177e4SLinus Torvalds lru_add_drain(); 2337a72afd87SWill Deacon tlb_gather_mmu(&tlb, mm); 2338365e9c87SHugh Dickins update_hiwater_rss(mm); 2339fd892593SLiam R. Howlett unmap_vmas(&tlb, mas, vma, start, end, tree_end, mm_wr_locked); 2340fd892593SLiam R. Howlett mas_set(mas, mt_start); 2341fd892593SLiam R. Howlett free_pgtables(&tlb, mas, vma, prev ? prev->vm_end : FIRST_USER_ADDRESS, 234298e51a22SSuren Baghdasaryan next ? next->vm_start : USER_PGTABLES_CEILING, 234398e51a22SSuren Baghdasaryan mm_wr_locked); 2344ae8eba8bSWill Deacon tlb_finish_mmu(&tlb); 23451da177e4SLinus Torvalds } 23461da177e4SLinus Torvalds 23471da177e4SLinus Torvalds /* 2348def5efe0SDavid Rientjes * __split_vma() bypasses sysctl_max_map_count checking. We use this where it 2349def5efe0SDavid Rientjes * has already been checked or doesn't make sense to fail. 23500fd5a9e2SLiam R. Howlett * VMA Iterator will point to the end VMA. 23511da177e4SLinus Torvalds */ 2352adb20b0cSLorenzo Stoakes static int __split_vma(struct vma_iterator *vmi, struct vm_area_struct *vma, 23531da177e4SLinus Torvalds unsigned long addr, int new_below) 23541da177e4SLinus Torvalds { 2355b2b3b886SLiam R. Howlett struct vma_prepare vp; 23561da177e4SLinus Torvalds struct vm_area_struct *new; 2357e3975891SChen Gang int err; 23589760ebffSLiam R. Howlett 2359b2b3b886SLiam R. Howlett WARN_ON(vma->vm_start >= addr); 2360b2b3b886SLiam R. Howlett WARN_ON(vma->vm_end <= addr); 2361b2b3b886SLiam R. Howlett 2362dd3b614fSDmitry Safonov if (vma->vm_ops && vma->vm_ops->may_split) { 2363dd3b614fSDmitry Safonov err = vma->vm_ops->may_split(vma, addr); 236431383c68SDan Williams if (err) 236531383c68SDan Williams return err; 236631383c68SDan Williams } 23671da177e4SLinus Torvalds 23683928d4f5SLinus Torvalds new = vm_area_dup(vma); 23691da177e4SLinus Torvalds if (!new) 2370e3975891SChen Gang return -ENOMEM; 23711da177e4SLinus Torvalds 2372b2b3b886SLiam R. Howlett if (new_below) { 23731da177e4SLinus Torvalds new->vm_end = addr; 2374b2b3b886SLiam R. Howlett } else { 23751da177e4SLinus Torvalds new->vm_start = addr; 23761da177e4SLinus Torvalds new->vm_pgoff += ((addr - vma->vm_start) >> PAGE_SHIFT); 23771da177e4SLinus Torvalds } 23781da177e4SLinus Torvalds 2379b5df0922SLiam R. Howlett err = -ENOMEM; 2380b5df0922SLiam R. Howlett vma_iter_config(vmi, new->vm_start, new->vm_end); 2381b5df0922SLiam R. Howlett if (vma_iter_prealloc(vmi, new)) 2382b5df0922SLiam R. Howlett goto out_free_vma; 2383b5df0922SLiam R. Howlett 2384ef0855d3SOleg Nesterov err = vma_dup_policy(vma, new); 2385ef0855d3SOleg Nesterov if (err) 2386b2b3b886SLiam R. Howlett goto out_free_vmi; 23871da177e4SLinus Torvalds 2388c4ea95d7SDaniel Forrest err = anon_vma_clone(new, vma); 2389c4ea95d7SDaniel Forrest if (err) 23905beb4930SRik van Riel goto out_free_mpol; 23915beb4930SRik van Riel 2392e9714acfSKonstantin Khlebnikov if (new->vm_file) 23931da177e4SLinus Torvalds get_file(new->vm_file); 23941da177e4SLinus Torvalds 23951da177e4SLinus Torvalds if (new->vm_ops && new->vm_ops->open) 23961da177e4SLinus Torvalds new->vm_ops->open(new); 23971da177e4SLinus Torvalds 2398c9d6e982SSuren Baghdasaryan vma_start_write(vma); 2399c9d6e982SSuren Baghdasaryan vma_start_write(new); 2400c9d6e982SSuren Baghdasaryan 2401b2b3b886SLiam R. Howlett init_vma_prep(&vp, vma); 2402b2b3b886SLiam R. Howlett vp.insert = new; 2403b2b3b886SLiam R. Howlett vma_prepare(&vp); 2404ccf1d78dSSuren Baghdasaryan vma_adjust_trans_huge(vma, vma->vm_start, addr, 0); 24051da177e4SLinus Torvalds 2406b2b3b886SLiam R. Howlett if (new_below) { 2407b2b3b886SLiam R. Howlett vma->vm_start = addr; 2408b2b3b886SLiam R. Howlett vma->vm_pgoff += (addr - new->vm_start) >> PAGE_SHIFT; 2409b2b3b886SLiam R. Howlett } else { 2410b2b3b886SLiam R. Howlett vma->vm_end = addr; 24119760ebffSLiam R. Howlett } 24125beb4930SRik van Riel 2413b2b3b886SLiam R. Howlett /* vma_complete stores the new vma */ 2414b2b3b886SLiam R. Howlett vma_complete(&vp, vmi, vma->vm_mm); 2415b2b3b886SLiam R. Howlett 2416b2b3b886SLiam R. Howlett /* Success. */ 2417b2b3b886SLiam R. Howlett if (new_below) 2418b2b3b886SLiam R. Howlett vma_next(vmi); 2419b2b3b886SLiam R. Howlett return 0; 2420b2b3b886SLiam R. Howlett 24215beb4930SRik van Riel out_free_mpol: 2422ef0855d3SOleg Nesterov mpol_put(vma_policy(new)); 2423b2b3b886SLiam R. Howlett out_free_vmi: 2424b2b3b886SLiam R. Howlett vma_iter_free(vmi); 24255beb4930SRik van Riel out_free_vma: 24263928d4f5SLinus Torvalds vm_area_free(new); 24275beb4930SRik van Riel return err; 24281da177e4SLinus Torvalds } 24291da177e4SLinus Torvalds 2430659ace58SKOSAKI Motohiro /* 2431659ace58SKOSAKI Motohiro * Split a vma into two pieces at address 'addr', a new vma is allocated 2432659ace58SKOSAKI Motohiro * either for the first part or the tail. 2433659ace58SKOSAKI Motohiro */ 2434adb20b0cSLorenzo Stoakes static int split_vma(struct vma_iterator *vmi, struct vm_area_struct *vma, 2435659ace58SKOSAKI Motohiro unsigned long addr, int new_below) 2436659ace58SKOSAKI Motohiro { 24379760ebffSLiam R. Howlett if (vma->vm_mm->map_count >= sysctl_max_map_count) 2438659ace58SKOSAKI Motohiro return -ENOMEM; 2439659ace58SKOSAKI Motohiro 24409760ebffSLiam R. Howlett return __split_vma(vmi, vma, addr, new_below); 2441f2ebfe43SLiam R. Howlett } 2442f2ebfe43SLiam R. Howlett 244311f9a21aSLiam R. Howlett /* 244494d7d923SLorenzo Stoakes * We are about to modify one or multiple of a VMA's flags, policy, userfaultfd 244594d7d923SLorenzo Stoakes * context and anonymous VMA name within the range [start, end). 244694d7d923SLorenzo Stoakes * 244794d7d923SLorenzo Stoakes * As a result, we might be able to merge the newly modified VMA range with an 244894d7d923SLorenzo Stoakes * adjacent VMA with identical properties. 244994d7d923SLorenzo Stoakes * 245094d7d923SLorenzo Stoakes * If no merge is possible and the range does not span the entirety of the VMA, 245194d7d923SLorenzo Stoakes * we then need to split the VMA to accommodate the change. 245294d7d923SLorenzo Stoakes * 245394d7d923SLorenzo Stoakes * The function returns either the merged VMA, the original VMA if a split was 245494d7d923SLorenzo Stoakes * required instead, or an error if the split failed. 245594d7d923SLorenzo Stoakes */ 245694d7d923SLorenzo Stoakes struct vm_area_struct *vma_modify(struct vma_iterator *vmi, 245794d7d923SLorenzo Stoakes struct vm_area_struct *prev, 245894d7d923SLorenzo Stoakes struct vm_area_struct *vma, 245994d7d923SLorenzo Stoakes unsigned long start, unsigned long end, 246094d7d923SLorenzo Stoakes unsigned long vm_flags, 246194d7d923SLorenzo Stoakes struct mempolicy *policy, 246294d7d923SLorenzo Stoakes struct vm_userfaultfd_ctx uffd_ctx, 246394d7d923SLorenzo Stoakes struct anon_vma_name *anon_name) 246494d7d923SLorenzo Stoakes { 246594d7d923SLorenzo Stoakes pgoff_t pgoff = vma->vm_pgoff + ((start - vma->vm_start) >> PAGE_SHIFT); 246694d7d923SLorenzo Stoakes struct vm_area_struct *merged; 246794d7d923SLorenzo Stoakes 246894d7d923SLorenzo Stoakes merged = vma_merge(vmi, vma->vm_mm, prev, start, end, vm_flags, 246994d7d923SLorenzo Stoakes vma->anon_vma, vma->vm_file, pgoff, policy, 247094d7d923SLorenzo Stoakes uffd_ctx, anon_name); 247194d7d923SLorenzo Stoakes if (merged) 247294d7d923SLorenzo Stoakes return merged; 247394d7d923SLorenzo Stoakes 247494d7d923SLorenzo Stoakes if (vma->vm_start < start) { 247594d7d923SLorenzo Stoakes int err = split_vma(vmi, vma, start, 1); 247694d7d923SLorenzo Stoakes 247794d7d923SLorenzo Stoakes if (err) 247894d7d923SLorenzo Stoakes return ERR_PTR(err); 247994d7d923SLorenzo Stoakes } 248094d7d923SLorenzo Stoakes 248194d7d923SLorenzo Stoakes if (vma->vm_end > end) { 248294d7d923SLorenzo Stoakes int err = split_vma(vmi, vma, end, 0); 248394d7d923SLorenzo Stoakes 248494d7d923SLorenzo Stoakes if (err) 248594d7d923SLorenzo Stoakes return ERR_PTR(err); 248694d7d923SLorenzo Stoakes } 248794d7d923SLorenzo Stoakes 248894d7d923SLorenzo Stoakes return vma; 248994d7d923SLorenzo Stoakes } 249094d7d923SLorenzo Stoakes 249194d7d923SLorenzo Stoakes /* 24924b5f2d20SLorenzo Stoakes * Attempt to merge a newly mapped VMA with those adjacent to it. The caller 24934b5f2d20SLorenzo Stoakes * must ensure that [start, end) does not overlap any existing VMA. 24944b5f2d20SLorenzo Stoakes */ 24954b5f2d20SLorenzo Stoakes static struct vm_area_struct 24964b5f2d20SLorenzo Stoakes *vma_merge_new_vma(struct vma_iterator *vmi, struct vm_area_struct *prev, 24974b5f2d20SLorenzo Stoakes struct vm_area_struct *vma, unsigned long start, 24984b5f2d20SLorenzo Stoakes unsigned long end, pgoff_t pgoff) 24994b5f2d20SLorenzo Stoakes { 25004b5f2d20SLorenzo Stoakes return vma_merge(vmi, vma->vm_mm, prev, start, end, vma->vm_flags, 25014b5f2d20SLorenzo Stoakes vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma), 25024b5f2d20SLorenzo Stoakes vma->vm_userfaultfd_ctx, anon_vma_name(vma)); 25034b5f2d20SLorenzo Stoakes } 25044b5f2d20SLorenzo Stoakes 25054b5f2d20SLorenzo Stoakes /* 250693bf5d4aSLorenzo Stoakes * Expand vma by delta bytes, potentially merging with an immediately adjacent 250793bf5d4aSLorenzo Stoakes * VMA with identical properties. 250893bf5d4aSLorenzo Stoakes */ 250993bf5d4aSLorenzo Stoakes struct vm_area_struct *vma_merge_extend(struct vma_iterator *vmi, 251093bf5d4aSLorenzo Stoakes struct vm_area_struct *vma, 251193bf5d4aSLorenzo Stoakes unsigned long delta) 251293bf5d4aSLorenzo Stoakes { 251393bf5d4aSLorenzo Stoakes pgoff_t pgoff = vma->vm_pgoff + vma_pages(vma); 251493bf5d4aSLorenzo Stoakes 251593bf5d4aSLorenzo Stoakes /* vma is specified as prev, so case 1 or 2 will apply. */ 251693bf5d4aSLorenzo Stoakes return vma_merge(vmi, vma->vm_mm, vma, vma->vm_end, vma->vm_end + delta, 251793bf5d4aSLorenzo Stoakes vma->vm_flags, vma->anon_vma, vma->vm_file, pgoff, 251893bf5d4aSLorenzo Stoakes vma_policy(vma), vma->vm_userfaultfd_ctx, 251993bf5d4aSLorenzo Stoakes anon_vma_name(vma)); 252093bf5d4aSLorenzo Stoakes } 252193bf5d4aSLorenzo Stoakes 252293bf5d4aSLorenzo Stoakes /* 2523183654ceSLiam R. Howlett * do_vmi_align_munmap() - munmap the aligned region from @start to @end. 2524183654ceSLiam R. Howlett * @vmi: The vma iterator 252511f9a21aSLiam R. Howlett * @vma: The starting vm_area_struct 252611f9a21aSLiam R. Howlett * @mm: The mm_struct 252711f9a21aSLiam R. Howlett * @start: The aligned start address to munmap. 252811f9a21aSLiam R. Howlett * @end: The aligned end address to munmap. 252911f9a21aSLiam R. Howlett * @uf: The userfaultfd list_head 2530408579cdSLiam R. Howlett * @unlock: Set to true to drop the mmap_lock. unlocking only happens on 2531408579cdSLiam R. Howlett * success. 253211f9a21aSLiam R. Howlett * 2533408579cdSLiam R. Howlett * Return: 0 on success and drops the lock if so directed, error and leaves the 2534408579cdSLiam R. Howlett * lock held otherwise. 253511f9a21aSLiam R. Howlett */ 253611f9a21aSLiam R. Howlett static int 2537183654ceSLiam R. Howlett do_vmi_align_munmap(struct vma_iterator *vmi, struct vm_area_struct *vma, 253811f9a21aSLiam R. Howlett struct mm_struct *mm, unsigned long start, 2539408579cdSLiam R. Howlett unsigned long end, struct list_head *uf, bool unlock) 254011f9a21aSLiam R. Howlett { 2541763ecb03SLiam R. Howlett struct vm_area_struct *prev, *next = NULL; 2542763ecb03SLiam R. Howlett struct maple_tree mt_detach; 2543763ecb03SLiam R. Howlett int count = 0; 2544d4af56c5SLiam R. Howlett int error = -ENOMEM; 2545606c812eSLiam R. Howlett unsigned long locked_vm = 0; 2546763ecb03SLiam R. Howlett MA_STATE(mas_detach, &mt_detach, 0, 0); 25473dd44325SLiam R. Howlett mt_init_flags(&mt_detach, vmi->mas.tree->ma_flags & MT_FLAGS_LOCK_MASK); 254802fdb25fSLiam R. Howlett mt_on_stack(mt_detach); 2549524e00b3SLiam R. Howlett 25501da177e4SLinus Torvalds /* 25511da177e4SLinus Torvalds * If we need to split any vma, do it now to save pain later. 25521da177e4SLinus Torvalds * 25531da177e4SLinus Torvalds * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially 25541da177e4SLinus Torvalds * unmapped vm_area_struct will remain in use: so lower split_vma 25551da177e4SLinus Torvalds * places tmp vma above, and higher split_vma places tmp vma below. 25561da177e4SLinus Torvalds */ 2557763ecb03SLiam R. Howlett 2558763ecb03SLiam R. Howlett /* Does it split the first one? */ 2559146425a3SHugh Dickins if (start > vma->vm_start) { 2560659ace58SKOSAKI Motohiro 2561659ace58SKOSAKI Motohiro /* 2562659ace58SKOSAKI Motohiro * Make sure that map_count on return from munmap() will 2563659ace58SKOSAKI Motohiro * not exceed its limit; but let map_count go just above 2564659ace58SKOSAKI Motohiro * its limit temporarily, to help free resources as expected. 2565659ace58SKOSAKI Motohiro */ 2566659ace58SKOSAKI Motohiro if (end < vma->vm_end && mm->map_count >= sysctl_max_map_count) 2567d4af56c5SLiam R. Howlett goto map_count_exceeded; 2568659ace58SKOSAKI Motohiro 25696935e052SLiam R. Howlett error = __split_vma(vmi, vma, start, 1); 25701da177e4SLinus Torvalds if (error) 2571763ecb03SLiam R. Howlett goto start_split_failed; 25721da177e4SLinus Torvalds } 25731da177e4SLinus Torvalds 2574763ecb03SLiam R. Howlett /* 2575763ecb03SLiam R. Howlett * Detach a range of VMAs from the mm. Using next as a temp variable as 2576763ecb03SLiam R. Howlett * it is always overwritten. 2577763ecb03SLiam R. Howlett */ 25786935e052SLiam R. Howlett next = vma; 25796935e052SLiam R. Howlett do { 2580763ecb03SLiam R. Howlett /* Does it split the end? */ 2581763ecb03SLiam R. Howlett if (next->vm_end > end) { 25826b73cff2SLiam R. Howlett error = __split_vma(vmi, next, end, 0); 25831da177e4SLinus Torvalds if (error) 2584763ecb03SLiam R. Howlett goto end_split_failed; 258511f9a21aSLiam R. Howlett } 2586606c812eSLiam R. Howlett vma_start_write(next); 2587fd892593SLiam R. Howlett mas_set(&mas_detach, count); 25886c26bd43SDavid Woodhouse error = mas_store_gfp(&mas_detach, next, GFP_KERNEL); 25896c26bd43SDavid Woodhouse if (error) 2590606c812eSLiam R. Howlett goto munmap_gather_failed; 2591606c812eSLiam R. Howlett vma_mark_detached(next, true); 2592606c812eSLiam R. Howlett if (next->vm_flags & VM_LOCKED) 2593606c812eSLiam R. Howlett locked_vm += vma_pages(next); 2594763ecb03SLiam R. Howlett 2595763ecb03SLiam R. Howlett count++; 25962376dd7cSAndrea Arcangeli if (unlikely(uf)) { 25972376dd7cSAndrea Arcangeli /* 25982376dd7cSAndrea Arcangeli * If userfaultfd_unmap_prep returns an error the vmas 2599f0953a1bSIngo Molnar * will remain split, but userland will get a 26002376dd7cSAndrea Arcangeli * highly unexpected error anyway. This is no 26012376dd7cSAndrea Arcangeli * different than the case where the first of the two 26022376dd7cSAndrea Arcangeli * __split_vma fails, but we don't undo the first 26032376dd7cSAndrea Arcangeli * split, despite we could. This is unlikely enough 26042376dd7cSAndrea Arcangeli * failure that it's not worth optimizing it for. 26052376dd7cSAndrea Arcangeli */ 260665ac1320SLiam R. Howlett error = userfaultfd_unmap_prep(next, start, end, uf); 260711f9a21aSLiam R. Howlett 26082376dd7cSAndrea Arcangeli if (error) 2609d4af56c5SLiam R. Howlett goto userfaultfd_error; 26102376dd7cSAndrea Arcangeli } 2611ba470de4SRik van Riel #ifdef CONFIG_DEBUG_VM_MAPLE_TREE 2612ba470de4SRik van Riel BUG_ON(next->vm_start < start); 2613ba470de4SRik van Riel BUG_ON(next->vm_start > end); 26141da177e4SLinus Torvalds #endif 26156935e052SLiam R. Howlett } for_each_vma_range(*vmi, next, end); 26162376dd7cSAndrea Arcangeli 2617763ecb03SLiam R. Howlett #if defined(CONFIG_DEBUG_VM_MAPLE_TREE) 2618763ecb03SLiam R. Howlett /* Make sure no VMAs are about to be lost. */ 2619763ecb03SLiam R. Howlett { 2620fd892593SLiam R. Howlett MA_STATE(test, &mt_detach, 0, 0); 2621763ecb03SLiam R. Howlett struct vm_area_struct *vma_mas, *vma_test; 2622763ecb03SLiam R. Howlett int test_count = 0; 2623763ecb03SLiam R. Howlett 2624183654ceSLiam R. Howlett vma_iter_set(vmi, start); 2625763ecb03SLiam R. Howlett rcu_read_lock(); 2626fd892593SLiam R. Howlett vma_test = mas_find(&test, count - 1); 2627183654ceSLiam R. Howlett for_each_vma_range(*vmi, vma_mas, end) { 2628763ecb03SLiam R. Howlett BUG_ON(vma_mas != vma_test); 2629763ecb03SLiam R. Howlett test_count++; 2630fd892593SLiam R. Howlett vma_test = mas_next(&test, count - 1); 2631763ecb03SLiam R. Howlett } 2632763ecb03SLiam R. Howlett rcu_read_unlock(); 2633763ecb03SLiam R. Howlett BUG_ON(count != test_count); 2634763ecb03SLiam R. Howlett } 2635763ecb03SLiam R. Howlett #endif 26366935e052SLiam R. Howlett 26376935e052SLiam R. Howlett while (vma_iter_addr(vmi) > start) 26386935e052SLiam R. Howlett vma_iter_prev_range(vmi); 26396935e052SLiam R. Howlett 26406c26bd43SDavid Woodhouse error = vma_iter_clear_gfp(vmi, start, end, GFP_KERNEL); 26416c26bd43SDavid Woodhouse if (error) 2642606c812eSLiam R. Howlett goto clear_tree_failed; 26430378c0a0SLiam R. Howlett 26446c26bd43SDavid Woodhouse /* Point of no return */ 2645606c812eSLiam R. Howlett mm->locked_vm -= locked_vm; 2646763ecb03SLiam R. Howlett mm->map_count -= count; 2647408579cdSLiam R. Howlett if (unlock) 2648d8ed45c5SMichel Lespinasse mmap_write_downgrade(mm); 2649dd2283f2SYang Shi 26506935e052SLiam R. Howlett prev = vma_iter_prev_range(vmi); 26516935e052SLiam R. Howlett next = vma_next(vmi); 26526935e052SLiam R. Howlett if (next) 26536935e052SLiam R. Howlett vma_iter_prev_range(vmi); 26546935e052SLiam R. Howlett 265568f48381SSuren Baghdasaryan /* 265668f48381SSuren Baghdasaryan * We can free page tables without write-locking mmap_lock because VMAs 265768f48381SSuren Baghdasaryan * were isolated before we downgraded mmap_lock. 265868f48381SSuren Baghdasaryan */ 2659fd892593SLiam R. Howlett mas_set(&mas_detach, 1); 2660fd892593SLiam R. Howlett unmap_region(mm, &mas_detach, vma, prev, next, start, end, count, 2661fd892593SLiam R. Howlett !unlock); 2662763ecb03SLiam R. Howlett /* Statistics and freeing VMAs */ 2663fd892593SLiam R. Howlett mas_set(&mas_detach, 0); 2664763ecb03SLiam R. Howlett remove_mt(mm, &mas_detach); 2665ae80b404SLinus Torvalds validate_mm(mm); 2666408579cdSLiam R. Howlett if (unlock) 2667408579cdSLiam R. Howlett mmap_read_unlock(mm); 26681da177e4SLinus Torvalds 266902fdb25fSLiam R. Howlett __mt_destroy(&mt_detach); 2670408579cdSLiam R. Howlett return 0; 2671d4af56c5SLiam R. Howlett 2672606c812eSLiam R. Howlett clear_tree_failed: 2673d4af56c5SLiam R. Howlett userfaultfd_error: 2674606c812eSLiam R. Howlett munmap_gather_failed: 2675763ecb03SLiam R. Howlett end_split_failed: 2676606c812eSLiam R. Howlett mas_set(&mas_detach, 0); 2677606c812eSLiam R. Howlett mas_for_each(&mas_detach, next, end) 2678606c812eSLiam R. Howlett vma_mark_detached(next, false); 2679606c812eSLiam R. Howlett 2680763ecb03SLiam R. Howlett __mt_destroy(&mt_detach); 2681763ecb03SLiam R. Howlett start_split_failed: 2682763ecb03SLiam R. Howlett map_count_exceeded: 2683b5641a5dSLinus Torvalds validate_mm(mm); 2684d4af56c5SLiam R. Howlett return error; 26851da177e4SLinus Torvalds } 26861da177e4SLinus Torvalds 268711f9a21aSLiam R. Howlett /* 2688183654ceSLiam R. Howlett * do_vmi_munmap() - munmap a given range. 2689183654ceSLiam R. Howlett * @vmi: The vma iterator 269011f9a21aSLiam R. Howlett * @mm: The mm_struct 269111f9a21aSLiam R. Howlett * @start: The start address to munmap 269211f9a21aSLiam R. Howlett * @len: The length of the range to munmap 269311f9a21aSLiam R. Howlett * @uf: The userfaultfd list_head 2694408579cdSLiam R. Howlett * @unlock: set to true if the user wants to drop the mmap_lock on success 269511f9a21aSLiam R. Howlett * 269611f9a21aSLiam R. Howlett * This function takes a @mas that is either pointing to the previous VMA or set 269711f9a21aSLiam R. Howlett * to MA_START and sets it up to remove the mapping(s). The @len will be 269811f9a21aSLiam R. Howlett * aligned and any arch_unmap work will be preformed. 269911f9a21aSLiam R. Howlett * 2700408579cdSLiam R. Howlett * Return: 0 on success and drops the lock if so directed, error and leaves the 2701408579cdSLiam R. Howlett * lock held otherwise. 270211f9a21aSLiam R. Howlett */ 2703183654ceSLiam R. Howlett int do_vmi_munmap(struct vma_iterator *vmi, struct mm_struct *mm, 270411f9a21aSLiam R. Howlett unsigned long start, size_t len, struct list_head *uf, 2705408579cdSLiam R. Howlett bool unlock) 270611f9a21aSLiam R. Howlett { 270711f9a21aSLiam R. Howlett unsigned long end; 270811f9a21aSLiam R. Howlett struct vm_area_struct *vma; 270911f9a21aSLiam R. Howlett 271011f9a21aSLiam R. Howlett if ((offset_in_page(start)) || start > TASK_SIZE || len > TASK_SIZE-start) 271111f9a21aSLiam R. Howlett return -EINVAL; 271211f9a21aSLiam R. Howlett 271311f9a21aSLiam R. Howlett end = start + PAGE_ALIGN(len); 271411f9a21aSLiam R. Howlett if (end == start) 271511f9a21aSLiam R. Howlett return -EINVAL; 271611f9a21aSLiam R. Howlett 271711f9a21aSLiam R. Howlett /* arch_unmap() might do unmaps itself. */ 271811f9a21aSLiam R. Howlett arch_unmap(mm, start, end); 271911f9a21aSLiam R. Howlett 272011f9a21aSLiam R. Howlett /* Find the first overlapping VMA */ 2721183654ceSLiam R. Howlett vma = vma_find(vmi, end); 2722408579cdSLiam R. Howlett if (!vma) { 2723408579cdSLiam R. Howlett if (unlock) 2724408579cdSLiam R. Howlett mmap_write_unlock(mm); 272511f9a21aSLiam R. Howlett return 0; 2726408579cdSLiam R. Howlett } 272711f9a21aSLiam R. Howlett 2728408579cdSLiam R. Howlett return do_vmi_align_munmap(vmi, vma, mm, start, end, uf, unlock); 272911f9a21aSLiam R. Howlett } 273011f9a21aSLiam R. Howlett 273111f9a21aSLiam R. Howlett /* do_munmap() - Wrapper function for non-maple tree aware do_munmap() calls. 273211f9a21aSLiam R. Howlett * @mm: The mm_struct 273311f9a21aSLiam R. Howlett * @start: The start address to munmap 273411f9a21aSLiam R. Howlett * @len: The length to be munmapped. 273511f9a21aSLiam R. Howlett * @uf: The userfaultfd list_head 2736408579cdSLiam R. Howlett * 2737408579cdSLiam R. Howlett * Return: 0 on success, error otherwise. 273811f9a21aSLiam R. Howlett */ 2739dd2283f2SYang Shi int do_munmap(struct mm_struct *mm, unsigned long start, size_t len, 2740dd2283f2SYang Shi struct list_head *uf) 2741dd2283f2SYang Shi { 2742183654ceSLiam R. Howlett VMA_ITERATOR(vmi, mm, start); 274311f9a21aSLiam R. Howlett 2744183654ceSLiam R. Howlett return do_vmi_munmap(&vmi, mm, start, len, uf, false); 2745dd2283f2SYang Shi } 2746dd2283f2SYang Shi 2747e99668a5SLiam R. Howlett unsigned long mmap_region(struct file *file, unsigned long addr, 2748e99668a5SLiam R. Howlett unsigned long len, vm_flags_t vm_flags, unsigned long pgoff, 2749e99668a5SLiam R. Howlett struct list_head *uf) 2750e99668a5SLiam R. Howlett { 2751e99668a5SLiam R. Howlett struct mm_struct *mm = current->mm; 2752e99668a5SLiam R. Howlett struct vm_area_struct *vma = NULL; 2753e99668a5SLiam R. Howlett struct vm_area_struct *next, *prev, *merge; 2754e99668a5SLiam R. Howlett pgoff_t pglen = len >> PAGE_SHIFT; 2755e99668a5SLiam R. Howlett unsigned long charged = 0; 2756e99668a5SLiam R. Howlett unsigned long end = addr + len; 2757e99668a5SLiam R. Howlett unsigned long merge_start = addr, merge_end = end; 275815897894SLorenzo Stoakes bool writable_file_mapping = false; 2759e99668a5SLiam R. Howlett pgoff_t vm_pgoff; 2760e99668a5SLiam R. Howlett int error; 2761183654ceSLiam R. Howlett VMA_ITERATOR(vmi, mm, addr); 2762e99668a5SLiam R. Howlett 2763e99668a5SLiam R. Howlett /* Check against address space limit. */ 2764e99668a5SLiam R. Howlett if (!may_expand_vm(mm, vm_flags, len >> PAGE_SHIFT)) { 2765e99668a5SLiam R. Howlett unsigned long nr_pages; 2766e99668a5SLiam R. Howlett 2767e99668a5SLiam R. Howlett /* 2768e99668a5SLiam R. Howlett * MAP_FIXED may remove pages of mappings that intersects with 2769e99668a5SLiam R. Howlett * requested mapping. Account for the pages it would unmap. 2770e99668a5SLiam R. Howlett */ 2771e99668a5SLiam R. Howlett nr_pages = count_vma_pages_range(mm, addr, end); 2772e99668a5SLiam R. Howlett 2773e99668a5SLiam R. Howlett if (!may_expand_vm(mm, vm_flags, 2774e99668a5SLiam R. Howlett (len >> PAGE_SHIFT) - nr_pages)) 2775e99668a5SLiam R. Howlett return -ENOMEM; 2776e99668a5SLiam R. Howlett } 2777e99668a5SLiam R. Howlett 2778e99668a5SLiam R. Howlett /* Unmap any existing mapping in the area */ 2779183654ceSLiam R. Howlett if (do_vmi_munmap(&vmi, mm, addr, len, uf, false)) 2780e99668a5SLiam R. Howlett return -ENOMEM; 2781e99668a5SLiam R. Howlett 2782e99668a5SLiam R. Howlett /* 2783e99668a5SLiam R. Howlett * Private writable mapping: check memory availability 2784e99668a5SLiam R. Howlett */ 2785e99668a5SLiam R. Howlett if (accountable_mapping(file, vm_flags)) { 2786e99668a5SLiam R. Howlett charged = len >> PAGE_SHIFT; 2787e99668a5SLiam R. Howlett if (security_vm_enough_memory_mm(mm, charged)) 2788e99668a5SLiam R. Howlett return -ENOMEM; 2789e99668a5SLiam R. Howlett vm_flags |= VM_ACCOUNT; 2790e99668a5SLiam R. Howlett } 2791e99668a5SLiam R. Howlett 2792183654ceSLiam R. Howlett next = vma_next(&vmi); 2793183654ceSLiam R. Howlett prev = vma_prev(&vmi); 279453bee98dSLiam R. Howlett if (vm_flags & VM_SPECIAL) { 279553bee98dSLiam R. Howlett if (prev) 279653bee98dSLiam R. Howlett vma_iter_next_range(&vmi); 2797e99668a5SLiam R. Howlett goto cannot_expand; 279853bee98dSLiam R. Howlett } 2799e99668a5SLiam R. Howlett 2800e99668a5SLiam R. Howlett /* Attempt to expand an old mapping */ 2801e99668a5SLiam R. Howlett /* Check next */ 2802e99668a5SLiam R. Howlett if (next && next->vm_start == end && !vma_policy(next) && 2803e99668a5SLiam R. Howlett can_vma_merge_before(next, vm_flags, NULL, file, pgoff+pglen, 2804e99668a5SLiam R. Howlett NULL_VM_UFFD_CTX, NULL)) { 2805e99668a5SLiam R. Howlett merge_end = next->vm_end; 2806e99668a5SLiam R. Howlett vma = next; 2807e99668a5SLiam R. Howlett vm_pgoff = next->vm_pgoff - pglen; 2808e99668a5SLiam R. Howlett } 2809e99668a5SLiam R. Howlett 2810e99668a5SLiam R. Howlett /* Check prev */ 2811e99668a5SLiam R. Howlett if (prev && prev->vm_end == addr && !vma_policy(prev) && 2812e99668a5SLiam R. Howlett (vma ? can_vma_merge_after(prev, vm_flags, vma->anon_vma, file, 2813e99668a5SLiam R. Howlett pgoff, vma->vm_userfaultfd_ctx, NULL) : 2814e99668a5SLiam R. Howlett can_vma_merge_after(prev, vm_flags, NULL, file, pgoff, 2815e99668a5SLiam R. Howlett NULL_VM_UFFD_CTX, NULL))) { 2816e99668a5SLiam R. Howlett merge_start = prev->vm_start; 2817e99668a5SLiam R. Howlett vma = prev; 2818e99668a5SLiam R. Howlett vm_pgoff = prev->vm_pgoff; 281953bee98dSLiam R. Howlett } else if (prev) { 282053bee98dSLiam R. Howlett vma_iter_next_range(&vmi); 2821e99668a5SLiam R. Howlett } 2822e99668a5SLiam R. Howlett 2823e99668a5SLiam R. Howlett /* Actually expand, if possible */ 2824e99668a5SLiam R. Howlett if (vma && 28253c441ab7SLiam R. Howlett !vma_expand(&vmi, vma, merge_start, merge_end, vm_pgoff, next)) { 2826e99668a5SLiam R. Howlett khugepaged_enter_vma(vma, vm_flags); 2827e99668a5SLiam R. Howlett goto expanded; 2828e99668a5SLiam R. Howlett } 2829e99668a5SLiam R. Howlett 283053bee98dSLiam R. Howlett if (vma == prev) 283153bee98dSLiam R. Howlett vma_iter_set(&vmi, addr); 2832e99668a5SLiam R. Howlett cannot_expand: 28335c1c03deSLiam R. Howlett 2834e99668a5SLiam R. Howlett /* 2835e99668a5SLiam R. Howlett * Determine the object being mapped and call the appropriate 2836e99668a5SLiam R. Howlett * specific mapper. the address has already been validated, but 2837e99668a5SLiam R. Howlett * not unmapped, but the maps are removed from the list. 2838e99668a5SLiam R. Howlett */ 2839e99668a5SLiam R. Howlett vma = vm_area_alloc(mm); 2840e99668a5SLiam R. Howlett if (!vma) { 2841e99668a5SLiam R. Howlett error = -ENOMEM; 2842e99668a5SLiam R. Howlett goto unacct_error; 2843e99668a5SLiam R. Howlett } 2844e99668a5SLiam R. Howlett 284553bee98dSLiam R. Howlett vma_iter_config(&vmi, addr, end); 2846e99668a5SLiam R. Howlett vma->vm_start = addr; 2847e99668a5SLiam R. Howlett vma->vm_end = end; 28481c71222eSSuren Baghdasaryan vm_flags_init(vma, vm_flags); 2849e99668a5SLiam R. Howlett vma->vm_page_prot = vm_get_page_prot(vm_flags); 2850e99668a5SLiam R. Howlett vma->vm_pgoff = pgoff; 2851e99668a5SLiam R. Howlett 2852e99668a5SLiam R. Howlett if (file) { 2853e99668a5SLiam R. Howlett vma->vm_file = get_file(file); 2854e99668a5SLiam R. Howlett error = call_mmap(file, vma); 2855e99668a5SLiam R. Howlett if (error) 2856e99668a5SLiam R. Howlett goto unmap_and_free_vma; 2857e99668a5SLiam R. Howlett 285815897894SLorenzo Stoakes if (vma_is_shared_maywrite(vma)) { 285915897894SLorenzo Stoakes error = mapping_map_writable(file->f_mapping); 286015897894SLorenzo Stoakes if (error) 286115897894SLorenzo Stoakes goto close_and_free_vma; 286215897894SLorenzo Stoakes 286315897894SLorenzo Stoakes writable_file_mapping = true; 286415897894SLorenzo Stoakes } 286515897894SLorenzo Stoakes 2866a57b7051SLiam Howlett /* 2867a57b7051SLiam Howlett * Expansion is handled above, merging is handled below. 2868a57b7051SLiam Howlett * Drivers should not alter the address of the VMA. 2869e99668a5SLiam R. Howlett */ 2870a57b7051SLiam Howlett error = -EINVAL; 2871cc8d1b09SLiam R. Howlett if (WARN_ON((addr != vma->vm_start))) 2872a57b7051SLiam Howlett goto close_and_free_vma; 2873e99668a5SLiam R. Howlett 287453bee98dSLiam R. Howlett vma_iter_config(&vmi, addr, end); 2875e99668a5SLiam R. Howlett /* 2876e99668a5SLiam R. Howlett * If vm_flags changed after call_mmap(), we should try merge 2877e99668a5SLiam R. Howlett * vma again as we may succeed this time. 2878e99668a5SLiam R. Howlett */ 2879e99668a5SLiam R. Howlett if (unlikely(vm_flags != vma->vm_flags && prev)) { 28804b5f2d20SLorenzo Stoakes merge = vma_merge_new_vma(&vmi, prev, vma, 28814b5f2d20SLorenzo Stoakes vma->vm_start, vma->vm_end, 28824b5f2d20SLorenzo Stoakes vma->vm_pgoff); 2883e99668a5SLiam R. Howlett if (merge) { 2884e99668a5SLiam R. Howlett /* 2885e99668a5SLiam R. Howlett * ->mmap() can change vma->vm_file and fput 2886e99668a5SLiam R. Howlett * the original file. So fput the vma->vm_file 2887e99668a5SLiam R. Howlett * here or we would add an extra fput for file 2888e99668a5SLiam R. Howlett * and cause general protection fault 2889e99668a5SLiam R. Howlett * ultimately. 2890e99668a5SLiam R. Howlett */ 2891e99668a5SLiam R. Howlett fput(vma->vm_file); 2892e99668a5SLiam R. Howlett vm_area_free(vma); 2893e99668a5SLiam R. Howlett vma = merge; 2894e99668a5SLiam R. Howlett /* Update vm_flags to pick up the change. */ 2895e99668a5SLiam R. Howlett vm_flags = vma->vm_flags; 2896e99668a5SLiam R. Howlett goto unmap_writable; 2897e99668a5SLiam R. Howlett } 2898e99668a5SLiam R. Howlett } 2899e99668a5SLiam R. Howlett 2900e99668a5SLiam R. Howlett vm_flags = vma->vm_flags; 2901e99668a5SLiam R. Howlett } else if (vm_flags & VM_SHARED) { 2902e99668a5SLiam R. Howlett error = shmem_zero_setup(vma); 2903e99668a5SLiam R. Howlett if (error) 2904e99668a5SLiam R. Howlett goto free_vma; 2905e99668a5SLiam R. Howlett } else { 2906e99668a5SLiam R. Howlett vma_set_anonymous(vma); 2907e99668a5SLiam R. Howlett } 2908e99668a5SLiam R. Howlett 2909b507808eSJoey Gouly if (map_deny_write_exec(vma, vma->vm_flags)) { 2910b507808eSJoey Gouly error = -EACCES; 2911b507808eSJoey Gouly goto close_and_free_vma; 2912b507808eSJoey Gouly } 2913b507808eSJoey Gouly 2914e99668a5SLiam R. Howlett /* Allow architectures to sanity-check the vm_flags */ 2915e99668a5SLiam R. Howlett error = -EINVAL; 2916cc8d1b09SLiam R. Howlett if (!arch_validate_flags(vma->vm_flags)) 2917deb0f656SCarlos Llamas goto close_and_free_vma; 2918e99668a5SLiam R. Howlett 2919e99668a5SLiam R. Howlett error = -ENOMEM; 2920b5df0922SLiam R. Howlett if (vma_iter_prealloc(&vmi, vma)) 29215789151eSMike Kravetz goto close_and_free_vma; 2922e99668a5SLiam R. Howlett 29231c7873e3SHugh Dickins /* Lock the VMA since it is modified after insertion into VMA tree */ 29241c7873e3SHugh Dickins vma_start_write(vma); 2925183654ceSLiam R. Howlett vma_iter_store(&vmi, vma); 2926e99668a5SLiam R. Howlett mm->map_count++; 2927e99668a5SLiam R. Howlett if (vma->vm_file) { 29286852c46cSYu Ma i_mmap_lock_write(vma->vm_file->f_mapping); 2929e8e17ee9SLorenzo Stoakes if (vma_is_shared_maywrite(vma)) 2930e99668a5SLiam R. Howlett mapping_allow_writable(vma->vm_file->f_mapping); 2931e99668a5SLiam R. Howlett 2932e99668a5SLiam R. Howlett flush_dcache_mmap_lock(vma->vm_file->f_mapping); 2933e99668a5SLiam R. Howlett vma_interval_tree_insert(vma, &vma->vm_file->f_mapping->i_mmap); 2934e99668a5SLiam R. Howlett flush_dcache_mmap_unlock(vma->vm_file->f_mapping); 2935e99668a5SLiam R. Howlett i_mmap_unlock_write(vma->vm_file->f_mapping); 2936e99668a5SLiam R. Howlett } 2937e99668a5SLiam R. Howlett 2938e99668a5SLiam R. Howlett /* 2939e99668a5SLiam R. Howlett * vma_merge() calls khugepaged_enter_vma() either, the below 2940e99668a5SLiam R. Howlett * call covers the non-merge case. 2941e99668a5SLiam R. Howlett */ 2942e99668a5SLiam R. Howlett khugepaged_enter_vma(vma, vma->vm_flags); 2943e99668a5SLiam R. Howlett 2944e99668a5SLiam R. Howlett /* Once vma denies write, undo our temporary denial count */ 2945e99668a5SLiam R. Howlett unmap_writable: 294615897894SLorenzo Stoakes if (writable_file_mapping) 2947e99668a5SLiam R. Howlett mapping_unmap_writable(file->f_mapping); 2948e99668a5SLiam R. Howlett file = vma->vm_file; 2949d7597f59SStefan Roesch ksm_add_vma(vma); 2950e99668a5SLiam R. Howlett expanded: 2951e99668a5SLiam R. Howlett perf_event_mmap(vma); 2952e99668a5SLiam R. Howlett 2953e99668a5SLiam R. Howlett vm_stat_account(mm, vm_flags, len >> PAGE_SHIFT); 2954e99668a5SLiam R. Howlett if (vm_flags & VM_LOCKED) { 2955e99668a5SLiam R. Howlett if ((vm_flags & VM_SPECIAL) || vma_is_dax(vma) || 2956e99668a5SLiam R. Howlett is_vm_hugetlb_page(vma) || 2957e99668a5SLiam R. Howlett vma == get_gate_vma(current->mm)) 2958e430a95aSSuren Baghdasaryan vm_flags_clear(vma, VM_LOCKED_MASK); 2959e99668a5SLiam R. Howlett else 2960e99668a5SLiam R. Howlett mm->locked_vm += (len >> PAGE_SHIFT); 2961e99668a5SLiam R. Howlett } 2962e99668a5SLiam R. Howlett 2963e99668a5SLiam R. Howlett if (file) 2964e99668a5SLiam R. Howlett uprobe_mmap(vma); 2965e99668a5SLiam R. Howlett 2966e99668a5SLiam R. Howlett /* 2967e99668a5SLiam R. Howlett * New (or expanded) vma always get soft dirty status. 2968e99668a5SLiam R. Howlett * Otherwise user-space soft-dirty page tracker won't 2969e99668a5SLiam R. Howlett * be able to distinguish situation when vma area unmapped, 2970e99668a5SLiam R. Howlett * then new mapped in-place (which must be aimed as 2971e99668a5SLiam R. Howlett * a completely new data area). 2972e99668a5SLiam R. Howlett */ 29731c71222eSSuren Baghdasaryan vm_flags_set(vma, VM_SOFTDIRTY); 2974e99668a5SLiam R. Howlett 2975e99668a5SLiam R. Howlett vma_set_page_prot(vma); 2976e99668a5SLiam R. Howlett 2977e99668a5SLiam R. Howlett validate_mm(mm); 2978e99668a5SLiam R. Howlett return addr; 2979e99668a5SLiam R. Howlett 2980deb0f656SCarlos Llamas close_and_free_vma: 2981cc8d1b09SLiam R. Howlett if (file && vma->vm_ops && vma->vm_ops->close) 2982deb0f656SCarlos Llamas vma->vm_ops->close(vma); 2983cc8d1b09SLiam R. Howlett 2984cc8d1b09SLiam R. Howlett if (file || vma->vm_file) { 2985e99668a5SLiam R. Howlett unmap_and_free_vma: 2986e99668a5SLiam R. Howlett fput(vma->vm_file); 2987e99668a5SLiam R. Howlett vma->vm_file = NULL; 2988e99668a5SLiam R. Howlett 2989fd892593SLiam R. Howlett vma_iter_set(&vmi, vma->vm_end); 2990e99668a5SLiam R. Howlett /* Undo any partial mapping done by a device driver. */ 2991fd892593SLiam R. Howlett unmap_region(mm, &vmi.mas, vma, prev, next, vma->vm_start, 2992fd892593SLiam R. Howlett vma->vm_end, vma->vm_end, true); 2993cc8d1b09SLiam R. Howlett } 299415897894SLorenzo Stoakes if (writable_file_mapping) 2995e99668a5SLiam R. Howlett mapping_unmap_writable(file->f_mapping); 2996e99668a5SLiam R. Howlett free_vma: 2997e99668a5SLiam R. Howlett vm_area_free(vma); 2998e99668a5SLiam R. Howlett unacct_error: 2999e99668a5SLiam R. Howlett if (charged) 3000e99668a5SLiam R. Howlett vm_unacct_memory(charged); 3001e99668a5SLiam R. Howlett validate_mm(mm); 3002e99668a5SLiam R. Howlett return error; 3003e99668a5SLiam R. Howlett } 3004e99668a5SLiam R. Howlett 3005408579cdSLiam R. Howlett static int __vm_munmap(unsigned long start, size_t len, bool unlock) 3006a46ef99dSLinus Torvalds { 3007a46ef99dSLinus Torvalds int ret; 3008bfce281cSAl Viro struct mm_struct *mm = current->mm; 3009897ab3e0SMike Rapoport LIST_HEAD(uf); 3010183654ceSLiam R. Howlett VMA_ITERATOR(vmi, mm, start); 3011a46ef99dSLinus Torvalds 3012d8ed45c5SMichel Lespinasse if (mmap_write_lock_killable(mm)) 3013ae798783SMichal Hocko return -EINTR; 3014ae798783SMichal Hocko 3015408579cdSLiam R. Howlett ret = do_vmi_munmap(&vmi, mm, start, len, &uf, unlock); 3016408579cdSLiam R. Howlett if (ret || !unlock) 3017d8ed45c5SMichel Lespinasse mmap_write_unlock(mm); 3018dd2283f2SYang Shi 3019897ab3e0SMike Rapoport userfaultfd_unmap_complete(mm, &uf); 3020a46ef99dSLinus Torvalds return ret; 3021a46ef99dSLinus Torvalds } 3022dd2283f2SYang Shi 3023dd2283f2SYang Shi int vm_munmap(unsigned long start, size_t len) 3024dd2283f2SYang Shi { 3025dd2283f2SYang Shi return __vm_munmap(start, len, false); 3026dd2283f2SYang Shi } 3027a46ef99dSLinus Torvalds EXPORT_SYMBOL(vm_munmap); 3028a46ef99dSLinus Torvalds 30296a6160a7SHeiko Carstens SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len) 30301da177e4SLinus Torvalds { 3031ce18d171SCatalin Marinas addr = untagged_addr(addr); 3032dd2283f2SYang Shi return __vm_munmap(addr, len, true); 30331da177e4SLinus Torvalds } 30341da177e4SLinus Torvalds 3035c8d78c18SKirill A. Shutemov 3036c8d78c18SKirill A. Shutemov /* 3037c8d78c18SKirill A. Shutemov * Emulation of deprecated remap_file_pages() syscall. 3038c8d78c18SKirill A. Shutemov */ 3039c8d78c18SKirill A. Shutemov SYSCALL_DEFINE5(remap_file_pages, unsigned long, start, unsigned long, size, 3040c8d78c18SKirill A. Shutemov unsigned long, prot, unsigned long, pgoff, unsigned long, flags) 3041c8d78c18SKirill A. Shutemov { 3042c8d78c18SKirill A. Shutemov 3043c8d78c18SKirill A. Shutemov struct mm_struct *mm = current->mm; 3044c8d78c18SKirill A. Shutemov struct vm_area_struct *vma; 3045c8d78c18SKirill A. Shutemov unsigned long populate = 0; 3046c8d78c18SKirill A. Shutemov unsigned long ret = -EINVAL; 3047c8d78c18SKirill A. Shutemov struct file *file; 3048c8d78c18SKirill A. Shutemov 3049ee65728eSMike Rapoport pr_warn_once("%s (%d) uses deprecated remap_file_pages() syscall. See Documentation/mm/remap_file_pages.rst.\n", 3050c8d78c18SKirill A. Shutemov current->comm, current->pid); 3051c8d78c18SKirill A. Shutemov 3052c8d78c18SKirill A. Shutemov if (prot) 3053c8d78c18SKirill A. Shutemov return ret; 3054c8d78c18SKirill A. Shutemov start = start & PAGE_MASK; 3055c8d78c18SKirill A. Shutemov size = size & PAGE_MASK; 3056c8d78c18SKirill A. Shutemov 3057c8d78c18SKirill A. Shutemov if (start + size <= start) 3058c8d78c18SKirill A. Shutemov return ret; 3059c8d78c18SKirill A. Shutemov 3060c8d78c18SKirill A. Shutemov /* Does pgoff wrap? */ 3061c8d78c18SKirill A. Shutemov if (pgoff + (size >> PAGE_SHIFT) < pgoff) 3062c8d78c18SKirill A. Shutemov return ret; 3063c8d78c18SKirill A. Shutemov 3064d8ed45c5SMichel Lespinasse if (mmap_write_lock_killable(mm)) 3065dc0ef0dfSMichal Hocko return -EINTR; 3066dc0ef0dfSMichal Hocko 30679b593cb2SLiam R. Howlett vma = vma_lookup(mm, start); 3068c8d78c18SKirill A. Shutemov 3069c8d78c18SKirill A. Shutemov if (!vma || !(vma->vm_flags & VM_SHARED)) 3070c8d78c18SKirill A. Shutemov goto out; 3071c8d78c18SKirill A. Shutemov 307248f7df32SKirill A. Shutemov if (start + size > vma->vm_end) { 3073763ecb03SLiam R. Howlett VMA_ITERATOR(vmi, mm, vma->vm_end); 3074763ecb03SLiam R. Howlett struct vm_area_struct *next, *prev = vma; 307548f7df32SKirill A. Shutemov 3076763ecb03SLiam R. Howlett for_each_vma_range(vmi, next, start + size) { 307748f7df32SKirill A. Shutemov /* hole between vmas ? */ 3078763ecb03SLiam R. Howlett if (next->vm_start != prev->vm_end) 307948f7df32SKirill A. Shutemov goto out; 308048f7df32SKirill A. Shutemov 308148f7df32SKirill A. Shutemov if (next->vm_file != vma->vm_file) 308248f7df32SKirill A. Shutemov goto out; 308348f7df32SKirill A. Shutemov 308448f7df32SKirill A. Shutemov if (next->vm_flags != vma->vm_flags) 308548f7df32SKirill A. Shutemov goto out; 308648f7df32SKirill A. Shutemov 30871db43d3fSLiam Howlett if (start + size <= next->vm_end) 30881db43d3fSLiam Howlett break; 30891db43d3fSLiam Howlett 3090763ecb03SLiam R. Howlett prev = next; 309148f7df32SKirill A. Shutemov } 309248f7df32SKirill A. Shutemov 309348f7df32SKirill A. Shutemov if (!next) 3094c8d78c18SKirill A. Shutemov goto out; 3095c8d78c18SKirill A. Shutemov } 3096c8d78c18SKirill A. Shutemov 3097c8d78c18SKirill A. Shutemov prot |= vma->vm_flags & VM_READ ? PROT_READ : 0; 3098c8d78c18SKirill A. Shutemov prot |= vma->vm_flags & VM_WRITE ? PROT_WRITE : 0; 3099c8d78c18SKirill A. Shutemov prot |= vma->vm_flags & VM_EXEC ? PROT_EXEC : 0; 3100c8d78c18SKirill A. Shutemov 3101c8d78c18SKirill A. Shutemov flags &= MAP_NONBLOCK; 3102c8d78c18SKirill A. Shutemov flags |= MAP_SHARED | MAP_FIXED | MAP_POPULATE; 3103fce000b1SLiam Howlett if (vma->vm_flags & VM_LOCKED) 3104c8d78c18SKirill A. Shutemov flags |= MAP_LOCKED; 310548f7df32SKirill A. Shutemov 3106c8d78c18SKirill A. Shutemov file = get_file(vma->vm_file); 310745e55300SPeter Collingbourne ret = do_mmap(vma->vm_file, start, size, 3108592b5fadSYu-cheng Yu prot, flags, 0, pgoff, &populate, NULL); 3109c8d78c18SKirill A. Shutemov fput(file); 3110c8d78c18SKirill A. Shutemov out: 3111d8ed45c5SMichel Lespinasse mmap_write_unlock(mm); 3112c8d78c18SKirill A. Shutemov if (populate) 3113c8d78c18SKirill A. Shutemov mm_populate(ret, populate); 3114c8d78c18SKirill A. Shutemov if (!IS_ERR_VALUE(ret)) 3115c8d78c18SKirill A. Shutemov ret = 0; 3116c8d78c18SKirill A. Shutemov return ret; 3117c8d78c18SKirill A. Shutemov } 3118c8d78c18SKirill A. Shutemov 31191da177e4SLinus Torvalds /* 312027b26701SLiam R. Howlett * do_vma_munmap() - Unmap a full or partial vma. 312127b26701SLiam R. Howlett * @vmi: The vma iterator pointing at the vma 312227b26701SLiam R. Howlett * @vma: The first vma to be munmapped 312327b26701SLiam R. Howlett * @start: the start of the address to unmap 312427b26701SLiam R. Howlett * @end: The end of the address to unmap 31252e7ce7d3SLiam R. Howlett * @uf: The userfaultfd list_head 3126408579cdSLiam R. Howlett * @unlock: Drop the lock on success 31272e7ce7d3SLiam R. Howlett * 312827b26701SLiam R. Howlett * unmaps a VMA mapping when the vma iterator is already in position. 312927b26701SLiam R. Howlett * Does not handle alignment. 3130408579cdSLiam R. Howlett * 3131408579cdSLiam R. Howlett * Return: 0 on success drops the lock of so directed, error on failure and will 3132408579cdSLiam R. Howlett * still hold the lock. 31331da177e4SLinus Torvalds */ 313427b26701SLiam R. Howlett int do_vma_munmap(struct vma_iterator *vmi, struct vm_area_struct *vma, 3135408579cdSLiam R. Howlett unsigned long start, unsigned long end, struct list_head *uf, 3136408579cdSLiam R. Howlett bool unlock) 31372e7ce7d3SLiam R. Howlett { 31382e7ce7d3SLiam R. Howlett struct mm_struct *mm = vma->vm_mm; 31392e7ce7d3SLiam R. Howlett 314027b26701SLiam R. Howlett arch_unmap(mm, start, end); 3141b5641a5dSLinus Torvalds return do_vmi_align_munmap(vmi, vma, mm, start, end, uf, unlock); 31422e7ce7d3SLiam R. Howlett } 31432e7ce7d3SLiam R. Howlett 31442e7ce7d3SLiam R. Howlett /* 31452e7ce7d3SLiam R. Howlett * do_brk_flags() - Increase the brk vma if the flags match. 314692fed820SLiam R. Howlett * @vmi: The vma iterator 31472e7ce7d3SLiam R. Howlett * @addr: The start address 31482e7ce7d3SLiam R. Howlett * @len: The length of the increase 31492e7ce7d3SLiam R. Howlett * @vma: The vma, 31502e7ce7d3SLiam R. Howlett * @flags: The VMA Flags 31512e7ce7d3SLiam R. Howlett * 31522e7ce7d3SLiam R. Howlett * Extend the brk VMA from addr to addr + len. If the VMA is NULL or the flags 31532e7ce7d3SLiam R. Howlett * do not match then create a new anonymous VMA. Eventually we may be able to 31542e7ce7d3SLiam R. Howlett * do some brk-specific accounting here. 31552e7ce7d3SLiam R. Howlett */ 315692fed820SLiam R. Howlett static int do_brk_flags(struct vma_iterator *vmi, struct vm_area_struct *vma, 3157763ecb03SLiam R. Howlett unsigned long addr, unsigned long len, unsigned long flags) 31581da177e4SLinus Torvalds { 31591da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 3160287051b1SLiam R. Howlett struct vma_prepare vp; 31612e7ce7d3SLiam R. Howlett 31622e7ce7d3SLiam R. Howlett /* 31632e7ce7d3SLiam R. Howlett * Check against address space limits by the changed size 31642e7ce7d3SLiam R. Howlett * Note: This happens *after* clearing old mappings in some code paths. 31652e7ce7d3SLiam R. Howlett */ 316616e72e9bSDenys Vlasenko flags |= VM_DATA_DEFAULT_FLAGS | VM_ACCOUNT | mm->def_flags; 316784638335SKonstantin Khlebnikov if (!may_expand_vm(mm, flags, len >> PAGE_SHIFT)) 31681da177e4SLinus Torvalds return -ENOMEM; 31691da177e4SLinus Torvalds 31701da177e4SLinus Torvalds if (mm->map_count > sysctl_max_map_count) 31711da177e4SLinus Torvalds return -ENOMEM; 31721da177e4SLinus Torvalds 3173191c5424SAl Viro if (security_vm_enough_memory_mm(mm, len >> PAGE_SHIFT)) 31741da177e4SLinus Torvalds return -ENOMEM; 31751da177e4SLinus Torvalds 31761da177e4SLinus Torvalds /* 31772e7ce7d3SLiam R. Howlett * Expand the existing vma if possible; Note that singular lists do not 31782e7ce7d3SLiam R. Howlett * occur after forking, so the expand will only happen on new VMAs. 31791da177e4SLinus Torvalds */ 31806c28ca64SLiam Howlett if (vma && vma->vm_end == addr && !vma_policy(vma) && 31816c28ca64SLiam Howlett can_vma_merge_after(vma, flags, NULL, NULL, 31826c28ca64SLiam Howlett addr >> PAGE_SHIFT, NULL_VM_UFFD_CTX, NULL)) { 3183b5df0922SLiam R. Howlett vma_iter_config(vmi, vma->vm_start, addr + len); 3184b5df0922SLiam R. Howlett if (vma_iter_prealloc(vmi, vma)) 3185675eaca1SAlistair Popple goto unacct_fail; 318628c5609fSLiam Howlett 3187c9d6e982SSuren Baghdasaryan vma_start_write(vma); 3188c9d6e982SSuren Baghdasaryan 3189287051b1SLiam R. Howlett init_vma_prep(&vp, vma); 3190287051b1SLiam R. Howlett vma_prepare(&vp); 3191ccf1d78dSSuren Baghdasaryan vma_adjust_trans_huge(vma, vma->vm_start, addr + len, 0); 31922e7ce7d3SLiam R. Howlett vma->vm_end = addr + len; 31931c71222eSSuren Baghdasaryan vm_flags_set(vma, VM_SOFTDIRTY); 319492fed820SLiam R. Howlett vma_iter_store(vmi, vma); 31952e7ce7d3SLiam R. Howlett 3196287051b1SLiam R. Howlett vma_complete(&vp, vmi, mm); 31972e7ce7d3SLiam R. Howlett khugepaged_enter_vma(vma, flags); 31982e7ce7d3SLiam R. Howlett goto out; 31992e7ce7d3SLiam R. Howlett } 32002e7ce7d3SLiam R. Howlett 3201b5df0922SLiam R. Howlett if (vma) 3202b5df0922SLiam R. Howlett vma_iter_next_range(vmi); 32032e7ce7d3SLiam R. Howlett /* create a vma struct for an anonymous mapping */ 32042e7ce7d3SLiam R. Howlett vma = vm_area_alloc(mm); 32052e7ce7d3SLiam R. Howlett if (!vma) 3206675eaca1SAlistair Popple goto unacct_fail; 32071da177e4SLinus Torvalds 3208bfd40eafSKirill A. Shutemov vma_set_anonymous(vma); 32091da177e4SLinus Torvalds vma->vm_start = addr; 32101da177e4SLinus Torvalds vma->vm_end = addr + len; 32112e7ce7d3SLiam R. Howlett vma->vm_pgoff = addr >> PAGE_SHIFT; 32121c71222eSSuren Baghdasaryan vm_flags_init(vma, flags); 32133ed75eb8SColy Li vma->vm_page_prot = vm_get_page_prot(flags); 3214ad9f0063SSuren Baghdasaryan vma_start_write(vma); 321592fed820SLiam R. Howlett if (vma_iter_store_gfp(vmi, vma, GFP_KERNEL)) 32162e7ce7d3SLiam R. Howlett goto mas_store_fail; 3217d4af56c5SLiam R. Howlett 32182e7ce7d3SLiam R. Howlett mm->map_count++; 32192574d5e4SLiam R. Howlett validate_mm(mm); 3220d7597f59SStefan Roesch ksm_add_vma(vma); 32211da177e4SLinus Torvalds out: 32223af9e859SEric B Munson perf_event_mmap(vma); 32231da177e4SLinus Torvalds mm->total_vm += len >> PAGE_SHIFT; 322484638335SKonstantin Khlebnikov mm->data_vm += len >> PAGE_SHIFT; 3225128557ffSMichel Lespinasse if (flags & VM_LOCKED) 3226ba470de4SRik van Riel mm->locked_vm += (len >> PAGE_SHIFT); 32271c71222eSSuren Baghdasaryan vm_flags_set(vma, VM_SOFTDIRTY); 32285d22fc25SLinus Torvalds return 0; 3229d4af56c5SLiam R. Howlett 32302e7ce7d3SLiam R. Howlett mas_store_fail: 3231d4af56c5SLiam R. Howlett vm_area_free(vma); 3232675eaca1SAlistair Popple unacct_fail: 32332e7ce7d3SLiam R. Howlett vm_unacct_memory(len >> PAGE_SHIFT); 32342e7ce7d3SLiam R. Howlett return -ENOMEM; 32351da177e4SLinus Torvalds } 32361da177e4SLinus Torvalds 3237bb177a73SMichal Hocko int vm_brk_flags(unsigned long addr, unsigned long request, unsigned long flags) 3238e4eb1ff6SLinus Torvalds { 3239e4eb1ff6SLinus Torvalds struct mm_struct *mm = current->mm; 32402e7ce7d3SLiam R. Howlett struct vm_area_struct *vma = NULL; 3241bb177a73SMichal Hocko unsigned long len; 32425d22fc25SLinus Torvalds int ret; 3243128557ffSMichel Lespinasse bool populate; 3244897ab3e0SMike Rapoport LIST_HEAD(uf); 324592fed820SLiam R. Howlett VMA_ITERATOR(vmi, mm, addr); 3246e4eb1ff6SLinus Torvalds 3247bb177a73SMichal Hocko len = PAGE_ALIGN(request); 3248bb177a73SMichal Hocko if (len < request) 3249bb177a73SMichal Hocko return -ENOMEM; 3250bb177a73SMichal Hocko if (!len) 3251bb177a73SMichal Hocko return 0; 3252bb177a73SMichal Hocko 32532e7ce7d3SLiam R. Howlett /* Until we need other flags, refuse anything except VM_EXEC. */ 32542e7ce7d3SLiam R. Howlett if ((flags & (~VM_EXEC)) != 0) 32552e7ce7d3SLiam R. Howlett return -EINVAL; 32562e7ce7d3SLiam R. Howlett 3257e0f81ab1SSebastian Ott if (mmap_write_lock_killable(mm)) 3258e0f81ab1SSebastian Ott return -EINTR; 3259e0f81ab1SSebastian Ott 32602e7ce7d3SLiam R. Howlett ret = check_brk_limits(addr, len); 32612e7ce7d3SLiam R. Howlett if (ret) 32622e7ce7d3SLiam R. Howlett goto limits_failed; 32632e7ce7d3SLiam R. Howlett 3264183654ceSLiam R. Howlett ret = do_vmi_munmap(&vmi, mm, addr, len, &uf, 0); 32652e7ce7d3SLiam R. Howlett if (ret) 32662e7ce7d3SLiam R. Howlett goto munmap_failed; 32672e7ce7d3SLiam R. Howlett 326892fed820SLiam R. Howlett vma = vma_prev(&vmi); 326992fed820SLiam R. Howlett ret = do_brk_flags(&vmi, vma, addr, len, flags); 3270128557ffSMichel Lespinasse populate = ((mm->def_flags & VM_LOCKED) != 0); 3271d8ed45c5SMichel Lespinasse mmap_write_unlock(mm); 3272897ab3e0SMike Rapoport userfaultfd_unmap_complete(mm, &uf); 32735d22fc25SLinus Torvalds if (populate && !ret) 3274128557ffSMichel Lespinasse mm_populate(addr, len); 3275e4eb1ff6SLinus Torvalds return ret; 32762e7ce7d3SLiam R. Howlett 32772e7ce7d3SLiam R. Howlett munmap_failed: 32782e7ce7d3SLiam R. Howlett limits_failed: 32792e7ce7d3SLiam R. Howlett mmap_write_unlock(mm); 32802e7ce7d3SLiam R. Howlett return ret; 3281e4eb1ff6SLinus Torvalds } 328216e72e9bSDenys Vlasenko EXPORT_SYMBOL(vm_brk_flags); 328316e72e9bSDenys Vlasenko 32841da177e4SLinus Torvalds /* Release all mmaps. */ 32851da177e4SLinus Torvalds void exit_mmap(struct mm_struct *mm) 32861da177e4SLinus Torvalds { 3287d16dfc55SPeter Zijlstra struct mmu_gather tlb; 3288ba470de4SRik van Riel struct vm_area_struct *vma; 32891da177e4SLinus Torvalds unsigned long nr_accounted = 0; 3290763ecb03SLiam R. Howlett MA_STATE(mas, &mm->mm_mt, 0, 0); 3291763ecb03SLiam R. Howlett int count = 0; 32921da177e4SLinus Torvalds 3293d6dd61c8SJeremy Fitzhardinge /* mm's last user has gone, and its about to be pulled down */ 3294cddb8a5cSAndrea Arcangeli mmu_notifier_release(mm); 3295d6dd61c8SJeremy Fitzhardinge 3296bf3980c8SSuren Baghdasaryan mmap_read_lock(mm); 32979480c53eSJeremy Fitzhardinge arch_exit_mmap(mm); 32989480c53eSJeremy Fitzhardinge 3299763ecb03SLiam R. Howlett vma = mas_find(&mas, ULONG_MAX); 330064591e86SSuren Baghdasaryan if (!vma) { 330164591e86SSuren Baghdasaryan /* Can happen if dup_mmap() received an OOM */ 3302bf3980c8SSuren Baghdasaryan mmap_read_unlock(mm); 33039480c53eSJeremy Fitzhardinge return; 330464591e86SSuren Baghdasaryan } 33059480c53eSJeremy Fitzhardinge 33061da177e4SLinus Torvalds lru_add_drain(); 33071da177e4SLinus Torvalds flush_cache_mm(mm); 3308d8b45053SWill Deacon tlb_gather_mmu_fullmm(&tlb, mm); 3309901608d9SOleg Nesterov /* update_hiwater_rss(mm) here? but nobody should be looking */ 3310763ecb03SLiam R. Howlett /* Use ULONG_MAX here to ensure all VMAs in the mm are unmapped */ 3311fd892593SLiam R. Howlett unmap_vmas(&tlb, &mas, vma, 0, ULONG_MAX, ULONG_MAX, false); 3312bf3980c8SSuren Baghdasaryan mmap_read_unlock(mm); 3313bf3980c8SSuren Baghdasaryan 3314bf3980c8SSuren Baghdasaryan /* 3315bf3980c8SSuren Baghdasaryan * Set MMF_OOM_SKIP to hide this task from the oom killer/reaper 3316b3541d91SSuren Baghdasaryan * because the memory has been already freed. 3317bf3980c8SSuren Baghdasaryan */ 3318bf3980c8SSuren Baghdasaryan set_bit(MMF_OOM_SKIP, &mm->flags); 3319bf3980c8SSuren Baghdasaryan mmap_write_lock(mm); 33203dd44325SLiam R. Howlett mt_clear_in_rcu(&mm->mm_mt); 3321fd892593SLiam R. Howlett mas_set(&mas, vma->vm_end); 3322fd892593SLiam R. Howlett free_pgtables(&tlb, &mas, vma, FIRST_USER_ADDRESS, 332398e51a22SSuren Baghdasaryan USER_PGTABLES_CEILING, true); 3324ae8eba8bSWill Deacon tlb_finish_mmu(&tlb); 33251da177e4SLinus Torvalds 3326763ecb03SLiam R. Howlett /* 3327763ecb03SLiam R. Howlett * Walk the list again, actually closing and freeing it, with preemption 3328763ecb03SLiam R. Howlett * enabled, without holding any MM locks besides the unreachable 3329763ecb03SLiam R. Howlett * mmap_write_lock. 3330763ecb03SLiam R. Howlett */ 3331fd892593SLiam R. Howlett mas_set(&mas, vma->vm_end); 3332763ecb03SLiam R. Howlett do { 33334f74d2c8SLinus Torvalds if (vma->vm_flags & VM_ACCOUNT) 33344f74d2c8SLinus Torvalds nr_accounted += vma_pages(vma); 33350d2ebf9cSSuren Baghdasaryan remove_vma(vma, true); 3336763ecb03SLiam R. Howlett count++; 33370a3b3c25SPaul E. McKenney cond_resched(); 3338763ecb03SLiam R. Howlett } while ((vma = mas_find(&mas, ULONG_MAX)) != NULL); 3339763ecb03SLiam R. Howlett 3340763ecb03SLiam R. Howlett BUG_ON(count != mm->map_count); 3341d4af56c5SLiam R. Howlett 3342d4af56c5SLiam R. Howlett trace_exit_mmap(mm); 3343d4af56c5SLiam R. Howlett __mt_destroy(&mm->mm_mt); 334464591e86SSuren Baghdasaryan mmap_write_unlock(mm); 33454f74d2c8SLinus Torvalds vm_unacct_memory(nr_accounted); 33461da177e4SLinus Torvalds } 33471da177e4SLinus Torvalds 33481da177e4SLinus Torvalds /* Insert vm structure into process list sorted by address 33491da177e4SLinus Torvalds * and into the inode's i_mmap tree. If vm_file is non-NULL 3350c8c06efaSDavidlohr Bueso * then i_mmap_rwsem is taken here. 33511da177e4SLinus Torvalds */ 33521da177e4SLinus Torvalds int insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma) 33531da177e4SLinus Torvalds { 3354d4af56c5SLiam R. Howlett unsigned long charged = vma_pages(vma); 33551da177e4SLinus Torvalds 3356d4af56c5SLiam R. Howlett 3357d0601a50SLiam R. Howlett if (find_vma_intersection(mm, vma->vm_start, vma->vm_end)) 3358c9d13f5fSChen Gang return -ENOMEM; 3359d4af56c5SLiam R. Howlett 3360c9d13f5fSChen Gang if ((vma->vm_flags & VM_ACCOUNT) && 3361d4af56c5SLiam R. Howlett security_vm_enough_memory_mm(mm, charged)) 3362c9d13f5fSChen Gang return -ENOMEM; 3363c9d13f5fSChen Gang 33641da177e4SLinus Torvalds /* 33651da177e4SLinus Torvalds * The vm_pgoff of a purely anonymous vma should be irrelevant 33661da177e4SLinus Torvalds * until its first write fault, when page's anon_vma and index 33671da177e4SLinus Torvalds * are set. But now set the vm_pgoff it will almost certainly 33681da177e4SLinus Torvalds * end up with (unless mremap moves it elsewhere before that 33691da177e4SLinus Torvalds * first wfault), so /proc/pid/maps tells a consistent story. 33701da177e4SLinus Torvalds * 33711da177e4SLinus Torvalds * By setting it to reflect the virtual start address of the 33721da177e4SLinus Torvalds * vma, merges and splits can happen in a seamless way, just 33731da177e4SLinus Torvalds * using the existing file pgoff checks and manipulations. 33748332326eSLiao Pingfang * Similarly in do_mmap and in do_brk_flags. 33751da177e4SLinus Torvalds */ 33768a9cc3b5SOleg Nesterov if (vma_is_anonymous(vma)) { 33771da177e4SLinus Torvalds BUG_ON(vma->anon_vma); 33781da177e4SLinus Torvalds vma->vm_pgoff = vma->vm_start >> PAGE_SHIFT; 33791da177e4SLinus Torvalds } 33802b144498SSrikar Dronamraju 3381763ecb03SLiam R. Howlett if (vma_link(mm, vma)) { 3382dd34d9feSAnthony Yznaga if (vma->vm_flags & VM_ACCOUNT) 3383d4af56c5SLiam R. Howlett vm_unacct_memory(charged); 3384d4af56c5SLiam R. Howlett return -ENOMEM; 3385d4af56c5SLiam R. Howlett } 3386d4af56c5SLiam R. Howlett 33871da177e4SLinus Torvalds return 0; 33881da177e4SLinus Torvalds } 33891da177e4SLinus Torvalds 33901da177e4SLinus Torvalds /* 33911da177e4SLinus Torvalds * Copy the vma structure to a new location in the same mm, 33921da177e4SLinus Torvalds * prior to moving page table entries, to effect an mremap move. 33931da177e4SLinus Torvalds */ 33941da177e4SLinus Torvalds struct vm_area_struct *copy_vma(struct vm_area_struct **vmap, 339538a76013SMichel Lespinasse unsigned long addr, unsigned long len, pgoff_t pgoff, 339638a76013SMichel Lespinasse bool *need_rmap_locks) 33971da177e4SLinus Torvalds { 33981da177e4SLinus Torvalds struct vm_area_struct *vma = *vmap; 33991da177e4SLinus Torvalds unsigned long vma_start = vma->vm_start; 34001da177e4SLinus Torvalds struct mm_struct *mm = vma->vm_mm; 34011da177e4SLinus Torvalds struct vm_area_struct *new_vma, *prev; 3402948f017bSAndrea Arcangeli bool faulted_in_anon_vma = true; 3403076f16bfSLiam R. Howlett VMA_ITERATOR(vmi, mm, addr); 34041da177e4SLinus Torvalds 34051da177e4SLinus Torvalds /* 34061da177e4SLinus Torvalds * If anonymous vma has not yet been faulted, update new pgoff 34071da177e4SLinus Torvalds * to match new location, to increase its chance of merging. 34081da177e4SLinus Torvalds */ 3409ce75799bSOleg Nesterov if (unlikely(vma_is_anonymous(vma) && !vma->anon_vma)) { 34101da177e4SLinus Torvalds pgoff = addr >> PAGE_SHIFT; 3411948f017bSAndrea Arcangeli faulted_in_anon_vma = false; 3412948f017bSAndrea Arcangeli } 34131da177e4SLinus Torvalds 3414763ecb03SLiam R. Howlett new_vma = find_vma_prev(mm, addr, &prev); 3415763ecb03SLiam R. Howlett if (new_vma && new_vma->vm_start < addr + len) 34166597d783SHugh Dickins return NULL; /* should never get here */ 3417524e00b3SLiam R. Howlett 34184b5f2d20SLorenzo Stoakes new_vma = vma_merge_new_vma(&vmi, prev, vma, addr, addr + len, pgoff); 34191da177e4SLinus Torvalds if (new_vma) { 34201da177e4SLinus Torvalds /* 34211da177e4SLinus Torvalds * Source vma may have been merged into new_vma 34221da177e4SLinus Torvalds */ 3423948f017bSAndrea Arcangeli if (unlikely(vma_start >= new_vma->vm_start && 3424948f017bSAndrea Arcangeli vma_start < new_vma->vm_end)) { 3425948f017bSAndrea Arcangeli /* 3426948f017bSAndrea Arcangeli * The only way we can get a vma_merge with 3427948f017bSAndrea Arcangeli * self during an mremap is if the vma hasn't 3428948f017bSAndrea Arcangeli * been faulted in yet and we were allowed to 3429948f017bSAndrea Arcangeli * reset the dst vma->vm_pgoff to the 3430948f017bSAndrea Arcangeli * destination address of the mremap to allow 3431948f017bSAndrea Arcangeli * the merge to happen. mremap must change the 3432948f017bSAndrea Arcangeli * vm_pgoff linearity between src and dst vmas 3433948f017bSAndrea Arcangeli * (in turn preventing a vma_merge) to be 3434948f017bSAndrea Arcangeli * safe. It is only safe to keep the vm_pgoff 3435948f017bSAndrea Arcangeli * linear if there are no pages mapped yet. 3436948f017bSAndrea Arcangeli */ 343781d1b09cSSasha Levin VM_BUG_ON_VMA(faulted_in_anon_vma, new_vma); 343838a76013SMichel Lespinasse *vmap = vma = new_vma; 3439108d6642SMichel Lespinasse } 344038a76013SMichel Lespinasse *need_rmap_locks = (new_vma->vm_pgoff <= vma->vm_pgoff); 34411da177e4SLinus Torvalds } else { 34423928d4f5SLinus Torvalds new_vma = vm_area_dup(vma); 3443e3975891SChen Gang if (!new_vma) 3444e3975891SChen Gang goto out; 34451da177e4SLinus Torvalds new_vma->vm_start = addr; 34461da177e4SLinus Torvalds new_vma->vm_end = addr + len; 34471da177e4SLinus Torvalds new_vma->vm_pgoff = pgoff; 3448ef0855d3SOleg Nesterov if (vma_dup_policy(vma, new_vma)) 3449523d4e20SMichel Lespinasse goto out_free_vma; 3450523d4e20SMichel Lespinasse if (anon_vma_clone(new_vma, vma)) 3451523d4e20SMichel Lespinasse goto out_free_mempol; 3452e9714acfSKonstantin Khlebnikov if (new_vma->vm_file) 34531da177e4SLinus Torvalds get_file(new_vma->vm_file); 34541da177e4SLinus Torvalds if (new_vma->vm_ops && new_vma->vm_ops->open) 34551da177e4SLinus Torvalds new_vma->vm_ops->open(new_vma); 3456763ecb03SLiam R. Howlett if (vma_link(mm, new_vma)) 3457524e00b3SLiam R. Howlett goto out_vma_link; 345838a76013SMichel Lespinasse *need_rmap_locks = false; 34591da177e4SLinus Torvalds } 34601da177e4SLinus Torvalds return new_vma; 34615beb4930SRik van Riel 3462524e00b3SLiam R. Howlett out_vma_link: 3463524e00b3SLiam R. Howlett if (new_vma->vm_ops && new_vma->vm_ops->close) 3464524e00b3SLiam R. Howlett new_vma->vm_ops->close(new_vma); 346592b73996SLiam Howlett 346692b73996SLiam Howlett if (new_vma->vm_file) 346792b73996SLiam Howlett fput(new_vma->vm_file); 346892b73996SLiam Howlett 346992b73996SLiam Howlett unlink_anon_vmas(new_vma); 34705beb4930SRik van Riel out_free_mempol: 3471ef0855d3SOleg Nesterov mpol_put(vma_policy(new_vma)); 34725beb4930SRik van Riel out_free_vma: 34733928d4f5SLinus Torvalds vm_area_free(new_vma); 3474e3975891SChen Gang out: 34755beb4930SRik van Riel return NULL; 34761da177e4SLinus Torvalds } 3477119f657cSakpm@osdl.org 3478119f657cSakpm@osdl.org /* 3479119f657cSakpm@osdl.org * Return true if the calling process may expand its vm space by the passed 3480119f657cSakpm@osdl.org * number of pages 3481119f657cSakpm@osdl.org */ 348284638335SKonstantin Khlebnikov bool may_expand_vm(struct mm_struct *mm, vm_flags_t flags, unsigned long npages) 3483119f657cSakpm@osdl.org { 348484638335SKonstantin Khlebnikov if (mm->total_vm + npages > rlimit(RLIMIT_AS) >> PAGE_SHIFT) 348584638335SKonstantin Khlebnikov return false; 3486119f657cSakpm@osdl.org 3487d977d56cSKonstantin Khlebnikov if (is_data_mapping(flags) && 3488d977d56cSKonstantin Khlebnikov mm->data_vm + npages > rlimit(RLIMIT_DATA) >> PAGE_SHIFT) { 3489f4fcd558SKonstantin Khlebnikov /* Workaround for Valgrind */ 3490f4fcd558SKonstantin Khlebnikov if (rlimit(RLIMIT_DATA) == 0 && 3491f4fcd558SKonstantin Khlebnikov mm->data_vm + npages <= rlimit_max(RLIMIT_DATA) >> PAGE_SHIFT) 3492f4fcd558SKonstantin Khlebnikov return true; 349357a7702bSDavid Woodhouse 349457a7702bSDavid Woodhouse pr_warn_once("%s (%d): VmData %lu exceed data ulimit %lu. Update limits%s.\n", 3495d977d56cSKonstantin Khlebnikov current->comm, current->pid, 3496d977d56cSKonstantin Khlebnikov (mm->data_vm + npages) << PAGE_SHIFT, 349757a7702bSDavid Woodhouse rlimit(RLIMIT_DATA), 349857a7702bSDavid Woodhouse ignore_rlimit_data ? "" : " or use boot option ignore_rlimit_data"); 349957a7702bSDavid Woodhouse 350057a7702bSDavid Woodhouse if (!ignore_rlimit_data) 3501d977d56cSKonstantin Khlebnikov return false; 3502d977d56cSKonstantin Khlebnikov } 3503119f657cSakpm@osdl.org 350484638335SKonstantin Khlebnikov return true; 350584638335SKonstantin Khlebnikov } 350684638335SKonstantin Khlebnikov 350784638335SKonstantin Khlebnikov void vm_stat_account(struct mm_struct *mm, vm_flags_t flags, long npages) 350884638335SKonstantin Khlebnikov { 35097866076bSPeng Liu WRITE_ONCE(mm->total_vm, READ_ONCE(mm->total_vm)+npages); 351084638335SKonstantin Khlebnikov 3511d977d56cSKonstantin Khlebnikov if (is_exec_mapping(flags)) 351284638335SKonstantin Khlebnikov mm->exec_vm += npages; 3513d977d56cSKonstantin Khlebnikov else if (is_stack_mapping(flags)) 351484638335SKonstantin Khlebnikov mm->stack_vm += npages; 3515d977d56cSKonstantin Khlebnikov else if (is_data_mapping(flags)) 351684638335SKonstantin Khlebnikov mm->data_vm += npages; 3517119f657cSakpm@osdl.org } 3518fa5dc22fSRoland McGrath 3519b3ec9f33SSouptick Joarder static vm_fault_t special_mapping_fault(struct vm_fault *vmf); 3520a62c34bdSAndy Lutomirski 3521a62c34bdSAndy Lutomirski /* 3522a62c34bdSAndy Lutomirski * Having a close hook prevents vma merging regardless of flags. 3523a62c34bdSAndy Lutomirski */ 3524a62c34bdSAndy Lutomirski static void special_mapping_close(struct vm_area_struct *vma) 3525a62c34bdSAndy Lutomirski { 3526a62c34bdSAndy Lutomirski } 3527a62c34bdSAndy Lutomirski 3528a62c34bdSAndy Lutomirski static const char *special_mapping_name(struct vm_area_struct *vma) 3529a62c34bdSAndy Lutomirski { 3530a62c34bdSAndy Lutomirski return ((struct vm_special_mapping *)vma->vm_private_data)->name; 3531a62c34bdSAndy Lutomirski } 3532a62c34bdSAndy Lutomirski 353314d07113SBrian Geffon static int special_mapping_mremap(struct vm_area_struct *new_vma) 3534b059a453SDmitry Safonov { 3535b059a453SDmitry Safonov struct vm_special_mapping *sm = new_vma->vm_private_data; 3536b059a453SDmitry Safonov 3537280e87e9SDmitry Safonov if (WARN_ON_ONCE(current->mm != new_vma->vm_mm)) 3538280e87e9SDmitry Safonov return -EFAULT; 3539280e87e9SDmitry Safonov 3540b059a453SDmitry Safonov if (sm->mremap) 3541b059a453SDmitry Safonov return sm->mremap(sm, new_vma); 3542280e87e9SDmitry Safonov 3543b059a453SDmitry Safonov return 0; 3544b059a453SDmitry Safonov } 3545b059a453SDmitry Safonov 3546871402e0SDmitry Safonov static int special_mapping_split(struct vm_area_struct *vma, unsigned long addr) 3547871402e0SDmitry Safonov { 3548871402e0SDmitry Safonov /* 3549871402e0SDmitry Safonov * Forbid splitting special mappings - kernel has expectations over 3550871402e0SDmitry Safonov * the number of pages in mapping. Together with VM_DONTEXPAND 3551871402e0SDmitry Safonov * the size of vma should stay the same over the special mapping's 3552871402e0SDmitry Safonov * lifetime. 3553871402e0SDmitry Safonov */ 3554871402e0SDmitry Safonov return -EINVAL; 3555871402e0SDmitry Safonov } 3556871402e0SDmitry Safonov 3557a62c34bdSAndy Lutomirski static const struct vm_operations_struct special_mapping_vmops = { 3558a62c34bdSAndy Lutomirski .close = special_mapping_close, 3559a62c34bdSAndy Lutomirski .fault = special_mapping_fault, 3560b059a453SDmitry Safonov .mremap = special_mapping_mremap, 3561a62c34bdSAndy Lutomirski .name = special_mapping_name, 3562af34ebebSDmitry Safonov /* vDSO code relies that VVAR can't be accessed remotely */ 3563af34ebebSDmitry Safonov .access = NULL, 3564871402e0SDmitry Safonov .may_split = special_mapping_split, 3565a62c34bdSAndy Lutomirski }; 3566a62c34bdSAndy Lutomirski 3567a62c34bdSAndy Lutomirski static const struct vm_operations_struct legacy_special_mapping_vmops = { 3568a62c34bdSAndy Lutomirski .close = special_mapping_close, 3569a62c34bdSAndy Lutomirski .fault = special_mapping_fault, 3570a62c34bdSAndy Lutomirski }; 3571fa5dc22fSRoland McGrath 3572b3ec9f33SSouptick Joarder static vm_fault_t special_mapping_fault(struct vm_fault *vmf) 3573fa5dc22fSRoland McGrath { 357411bac800SDave Jiang struct vm_area_struct *vma = vmf->vma; 3575b1d0e4f5SNick Piggin pgoff_t pgoff; 3576fa5dc22fSRoland McGrath struct page **pages; 3577fa5dc22fSRoland McGrath 3578f872f540SAndy Lutomirski if (vma->vm_ops == &legacy_special_mapping_vmops) { 3579a62c34bdSAndy Lutomirski pages = vma->vm_private_data; 3580f872f540SAndy Lutomirski } else { 3581f872f540SAndy Lutomirski struct vm_special_mapping *sm = vma->vm_private_data; 3582f872f540SAndy Lutomirski 3583f872f540SAndy Lutomirski if (sm->fault) 358411bac800SDave Jiang return sm->fault(sm, vmf->vma, vmf); 3585f872f540SAndy Lutomirski 3586f872f540SAndy Lutomirski pages = sm->pages; 3587f872f540SAndy Lutomirski } 3588a62c34bdSAndy Lutomirski 35898a9cc3b5SOleg Nesterov for (pgoff = vmf->pgoff; pgoff && *pages; ++pages) 3590b1d0e4f5SNick Piggin pgoff--; 3591fa5dc22fSRoland McGrath 3592fa5dc22fSRoland McGrath if (*pages) { 3593fa5dc22fSRoland McGrath struct page *page = *pages; 3594fa5dc22fSRoland McGrath get_page(page); 3595b1d0e4f5SNick Piggin vmf->page = page; 3596b1d0e4f5SNick Piggin return 0; 3597fa5dc22fSRoland McGrath } 3598fa5dc22fSRoland McGrath 3599b1d0e4f5SNick Piggin return VM_FAULT_SIGBUS; 3600fa5dc22fSRoland McGrath } 3601fa5dc22fSRoland McGrath 3602a62c34bdSAndy Lutomirski static struct vm_area_struct *__install_special_mapping( 3603a62c34bdSAndy Lutomirski struct mm_struct *mm, 3604fa5dc22fSRoland McGrath unsigned long addr, unsigned long len, 360527f28b97SChen Gang unsigned long vm_flags, void *priv, 360627f28b97SChen Gang const struct vm_operations_struct *ops) 3607fa5dc22fSRoland McGrath { 3608462e635eSTavis Ormandy int ret; 3609fa5dc22fSRoland McGrath struct vm_area_struct *vma; 3610fa5dc22fSRoland McGrath 3611490fc053SLinus Torvalds vma = vm_area_alloc(mm); 3612fa5dc22fSRoland McGrath if (unlikely(vma == NULL)) 36133935ed6aSStefani Seibold return ERR_PTR(-ENOMEM); 3614fa5dc22fSRoland McGrath 3615fa5dc22fSRoland McGrath vma->vm_start = addr; 3616fa5dc22fSRoland McGrath vma->vm_end = addr + len; 3617fa5dc22fSRoland McGrath 3618e430a95aSSuren Baghdasaryan vm_flags_init(vma, (vm_flags | mm->def_flags | 3619e430a95aSSuren Baghdasaryan VM_DONTEXPAND | VM_SOFTDIRTY) & ~VM_LOCKED_MASK); 36203ed75eb8SColy Li vma->vm_page_prot = vm_get_page_prot(vma->vm_flags); 3621fa5dc22fSRoland McGrath 3622a62c34bdSAndy Lutomirski vma->vm_ops = ops; 3623a62c34bdSAndy Lutomirski vma->vm_private_data = priv; 3624fa5dc22fSRoland McGrath 3625462e635eSTavis Ormandy ret = insert_vm_struct(mm, vma); 3626462e635eSTavis Ormandy if (ret) 3627462e635eSTavis Ormandy goto out; 3628fa5dc22fSRoland McGrath 362984638335SKonstantin Khlebnikov vm_stat_account(mm, vma->vm_flags, len >> PAGE_SHIFT); 3630fa5dc22fSRoland McGrath 3631cdd6c482SIngo Molnar perf_event_mmap(vma); 3632089dd79dSPeter Zijlstra 36333935ed6aSStefani Seibold return vma; 3634462e635eSTavis Ormandy 3635462e635eSTavis Ormandy out: 36363928d4f5SLinus Torvalds vm_area_free(vma); 36373935ed6aSStefani Seibold return ERR_PTR(ret); 36383935ed6aSStefani Seibold } 36393935ed6aSStefani Seibold 36402eefd878SDmitry Safonov bool vma_is_special_mapping(const struct vm_area_struct *vma, 36412eefd878SDmitry Safonov const struct vm_special_mapping *sm) 36422eefd878SDmitry Safonov { 36432eefd878SDmitry Safonov return vma->vm_private_data == sm && 36442eefd878SDmitry Safonov (vma->vm_ops == &special_mapping_vmops || 36452eefd878SDmitry Safonov vma->vm_ops == &legacy_special_mapping_vmops); 36462eefd878SDmitry Safonov } 36472eefd878SDmitry Safonov 3648a62c34bdSAndy Lutomirski /* 3649c1e8d7c6SMichel Lespinasse * Called with mm->mmap_lock held for writing. 3650a62c34bdSAndy Lutomirski * Insert a new vma covering the given region, with the given flags. 3651a62c34bdSAndy Lutomirski * Its pages are supplied by the given array of struct page *. 3652a62c34bdSAndy Lutomirski * The array can be shorter than len >> PAGE_SHIFT if it's null-terminated. 3653a62c34bdSAndy Lutomirski * The region past the last page supplied will always produce SIGBUS. 3654a62c34bdSAndy Lutomirski * The array pointer and the pages it points to are assumed to stay alive 3655a62c34bdSAndy Lutomirski * for as long as this mapping might exist. 3656a62c34bdSAndy Lutomirski */ 3657a62c34bdSAndy Lutomirski struct vm_area_struct *_install_special_mapping( 3658a62c34bdSAndy Lutomirski struct mm_struct *mm, 3659a62c34bdSAndy Lutomirski unsigned long addr, unsigned long len, 3660a62c34bdSAndy Lutomirski unsigned long vm_flags, const struct vm_special_mapping *spec) 3661a62c34bdSAndy Lutomirski { 366227f28b97SChen Gang return __install_special_mapping(mm, addr, len, vm_flags, (void *)spec, 366327f28b97SChen Gang &special_mapping_vmops); 3664a62c34bdSAndy Lutomirski } 3665a62c34bdSAndy Lutomirski 36663935ed6aSStefani Seibold int install_special_mapping(struct mm_struct *mm, 36673935ed6aSStefani Seibold unsigned long addr, unsigned long len, 36683935ed6aSStefani Seibold unsigned long vm_flags, struct page **pages) 36693935ed6aSStefani Seibold { 3670a62c34bdSAndy Lutomirski struct vm_area_struct *vma = __install_special_mapping( 367127f28b97SChen Gang mm, addr, len, vm_flags, (void *)pages, 367227f28b97SChen Gang &legacy_special_mapping_vmops); 36733935ed6aSStefani Seibold 367414bd5b45SDuan Jiong return PTR_ERR_OR_ZERO(vma); 3675fa5dc22fSRoland McGrath } 36767906d00cSAndrea Arcangeli 36777906d00cSAndrea Arcangeli static DEFINE_MUTEX(mm_all_locks_mutex); 36787906d00cSAndrea Arcangeli 3679454ed842SPeter Zijlstra static void vm_lock_anon_vma(struct mm_struct *mm, struct anon_vma *anon_vma) 36807906d00cSAndrea Arcangeli { 3681f808c13fSDavidlohr Bueso if (!test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_root.rb_node)) { 36827906d00cSAndrea Arcangeli /* 36837906d00cSAndrea Arcangeli * The LSB of head.next can't change from under us 36847906d00cSAndrea Arcangeli * because we hold the mm_all_locks_mutex. 36857906d00cSAndrea Arcangeli */ 3686da1c55f1SMichel Lespinasse down_write_nest_lock(&anon_vma->root->rwsem, &mm->mmap_lock); 36877906d00cSAndrea Arcangeli /* 36887906d00cSAndrea Arcangeli * We can safely modify head.next after taking the 36895a505085SIngo Molnar * anon_vma->root->rwsem. If some other vma in this mm shares 36907906d00cSAndrea Arcangeli * the same anon_vma we won't take it again. 36917906d00cSAndrea Arcangeli * 36927906d00cSAndrea Arcangeli * No need of atomic instructions here, head.next 36937906d00cSAndrea Arcangeli * can't change from under us thanks to the 36945a505085SIngo Molnar * anon_vma->root->rwsem. 36957906d00cSAndrea Arcangeli */ 36967906d00cSAndrea Arcangeli if (__test_and_set_bit(0, (unsigned long *) 3697f808c13fSDavidlohr Bueso &anon_vma->root->rb_root.rb_root.rb_node)) 36987906d00cSAndrea Arcangeli BUG(); 36997906d00cSAndrea Arcangeli } 37007906d00cSAndrea Arcangeli } 37017906d00cSAndrea Arcangeli 3702454ed842SPeter Zijlstra static void vm_lock_mapping(struct mm_struct *mm, struct address_space *mapping) 37037906d00cSAndrea Arcangeli { 37047906d00cSAndrea Arcangeli if (!test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) { 37057906d00cSAndrea Arcangeli /* 37067906d00cSAndrea Arcangeli * AS_MM_ALL_LOCKS can't change from under us because 37077906d00cSAndrea Arcangeli * we hold the mm_all_locks_mutex. 37087906d00cSAndrea Arcangeli * 37097906d00cSAndrea Arcangeli * Operations on ->flags have to be atomic because 37107906d00cSAndrea Arcangeli * even if AS_MM_ALL_LOCKS is stable thanks to the 37117906d00cSAndrea Arcangeli * mm_all_locks_mutex, there may be other cpus 37127906d00cSAndrea Arcangeli * changing other bitflags in parallel to us. 37137906d00cSAndrea Arcangeli */ 37147906d00cSAndrea Arcangeli if (test_and_set_bit(AS_MM_ALL_LOCKS, &mapping->flags)) 37157906d00cSAndrea Arcangeli BUG(); 3716da1c55f1SMichel Lespinasse down_write_nest_lock(&mapping->i_mmap_rwsem, &mm->mmap_lock); 37177906d00cSAndrea Arcangeli } 37187906d00cSAndrea Arcangeli } 37197906d00cSAndrea Arcangeli 37207906d00cSAndrea Arcangeli /* 37217906d00cSAndrea Arcangeli * This operation locks against the VM for all pte/vma/mm related 37227906d00cSAndrea Arcangeli * operations that could ever happen on a certain mm. This includes 37237906d00cSAndrea Arcangeli * vmtruncate, try_to_unmap, and all page faults. 37247906d00cSAndrea Arcangeli * 3725c1e8d7c6SMichel Lespinasse * The caller must take the mmap_lock in write mode before calling 37267906d00cSAndrea Arcangeli * mm_take_all_locks(). The caller isn't allowed to release the 3727c1e8d7c6SMichel Lespinasse * mmap_lock until mm_drop_all_locks() returns. 37287906d00cSAndrea Arcangeli * 3729c1e8d7c6SMichel Lespinasse * mmap_lock in write mode is required in order to block all operations 37307906d00cSAndrea Arcangeli * that could modify pagetables and free pages without need of 373127ba0644SKirill A. Shutemov * altering the vma layout. It's also needed in write mode to avoid new 37327906d00cSAndrea Arcangeli * anon_vmas to be associated with existing vmas. 37337906d00cSAndrea Arcangeli * 37347906d00cSAndrea Arcangeli * A single task can't take more than one mm_take_all_locks() in a row 37357906d00cSAndrea Arcangeli * or it would deadlock. 37367906d00cSAndrea Arcangeli * 3737bf181b9fSMichel Lespinasse * The LSB in anon_vma->rb_root.rb_node and the AS_MM_ALL_LOCKS bitflag in 37387906d00cSAndrea Arcangeli * mapping->flags avoid to take the same lock twice, if more than one 37397906d00cSAndrea Arcangeli * vma in this mm is backed by the same anon_vma or address_space. 37407906d00cSAndrea Arcangeli * 374188f306b6SKirill A. Shutemov * We take locks in following order, accordingly to comment at beginning 374288f306b6SKirill A. Shutemov * of mm/rmap.c: 374388f306b6SKirill A. Shutemov * - all hugetlbfs_i_mmap_rwsem_key locks (aka mapping->i_mmap_rwsem for 374488f306b6SKirill A. Shutemov * hugetlb mapping); 3745eeff9a5dSSuren Baghdasaryan * - all vmas marked locked 374688f306b6SKirill A. Shutemov * - all i_mmap_rwsem locks; 374788f306b6SKirill A. Shutemov * - all anon_vma->rwseml 374888f306b6SKirill A. Shutemov * 374988f306b6SKirill A. Shutemov * We can take all locks within these types randomly because the VM code 375088f306b6SKirill A. Shutemov * doesn't nest them and we protected from parallel mm_take_all_locks() by 375188f306b6SKirill A. Shutemov * mm_all_locks_mutex. 37527906d00cSAndrea Arcangeli * 37537906d00cSAndrea Arcangeli * mm_take_all_locks() and mm_drop_all_locks are expensive operations 37547906d00cSAndrea Arcangeli * that may have to take thousand of locks. 37557906d00cSAndrea Arcangeli * 37567906d00cSAndrea Arcangeli * mm_take_all_locks() can fail if it's interrupted by signals. 37577906d00cSAndrea Arcangeli */ 37587906d00cSAndrea Arcangeli int mm_take_all_locks(struct mm_struct *mm) 37597906d00cSAndrea Arcangeli { 37607906d00cSAndrea Arcangeli struct vm_area_struct *vma; 37615beb4930SRik van Riel struct anon_vma_chain *avc; 3762763ecb03SLiam R. Howlett MA_STATE(mas, &mm->mm_mt, 0, 0); 37637906d00cSAndrea Arcangeli 3764325bca1fSRolf Eike Beer mmap_assert_write_locked(mm); 37657906d00cSAndrea Arcangeli 37667906d00cSAndrea Arcangeli mutex_lock(&mm_all_locks_mutex); 37677906d00cSAndrea Arcangeli 376890717566SJann Horn /* 376990717566SJann Horn * vma_start_write() does not have a complement in mm_drop_all_locks() 377090717566SJann Horn * because vma_start_write() is always asymmetrical; it marks a VMA as 377190717566SJann Horn * being written to until mmap_write_unlock() or mmap_write_downgrade() 377290717566SJann Horn * is reached. 377390717566SJann Horn */ 3774763ecb03SLiam R. Howlett mas_for_each(&mas, vma, ULONG_MAX) { 37757906d00cSAndrea Arcangeli if (signal_pending(current)) 37767906d00cSAndrea Arcangeli goto out_unlock; 3777eeff9a5dSSuren Baghdasaryan vma_start_write(vma); 3778eeff9a5dSSuren Baghdasaryan } 3779eeff9a5dSSuren Baghdasaryan 3780eeff9a5dSSuren Baghdasaryan mas_set(&mas, 0); 3781eeff9a5dSSuren Baghdasaryan mas_for_each(&mas, vma, ULONG_MAX) { 3782eeff9a5dSSuren Baghdasaryan if (signal_pending(current)) 3783eeff9a5dSSuren Baghdasaryan goto out_unlock; 378488f306b6SKirill A. Shutemov if (vma->vm_file && vma->vm_file->f_mapping && 378588f306b6SKirill A. Shutemov is_vm_hugetlb_page(vma)) 378688f306b6SKirill A. Shutemov vm_lock_mapping(mm, vma->vm_file->f_mapping); 378788f306b6SKirill A. Shutemov } 378888f306b6SKirill A. Shutemov 3789763ecb03SLiam R. Howlett mas_set(&mas, 0); 3790763ecb03SLiam R. Howlett mas_for_each(&mas, vma, ULONG_MAX) { 379188f306b6SKirill A. Shutemov if (signal_pending(current)) 379288f306b6SKirill A. Shutemov goto out_unlock; 379388f306b6SKirill A. Shutemov if (vma->vm_file && vma->vm_file->f_mapping && 379488f306b6SKirill A. Shutemov !is_vm_hugetlb_page(vma)) 3795454ed842SPeter Zijlstra vm_lock_mapping(mm, vma->vm_file->f_mapping); 37967906d00cSAndrea Arcangeli } 37977cd5a02fSPeter Zijlstra 3798763ecb03SLiam R. Howlett mas_set(&mas, 0); 3799763ecb03SLiam R. Howlett mas_for_each(&mas, vma, ULONG_MAX) { 38007cd5a02fSPeter Zijlstra if (signal_pending(current)) 38017cd5a02fSPeter Zijlstra goto out_unlock; 38027cd5a02fSPeter Zijlstra if (vma->anon_vma) 38035beb4930SRik van Riel list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) 38045beb4930SRik van Riel vm_lock_anon_vma(mm, avc->anon_vma); 38057cd5a02fSPeter Zijlstra } 38067cd5a02fSPeter Zijlstra 3807584cff54SKautuk Consul return 0; 38087906d00cSAndrea Arcangeli 38097906d00cSAndrea Arcangeli out_unlock: 38107906d00cSAndrea Arcangeli mm_drop_all_locks(mm); 3811584cff54SKautuk Consul return -EINTR; 38127906d00cSAndrea Arcangeli } 38137906d00cSAndrea Arcangeli 38147906d00cSAndrea Arcangeli static void vm_unlock_anon_vma(struct anon_vma *anon_vma) 38157906d00cSAndrea Arcangeli { 3816f808c13fSDavidlohr Bueso if (test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_root.rb_node)) { 38177906d00cSAndrea Arcangeli /* 38187906d00cSAndrea Arcangeli * The LSB of head.next can't change to 0 from under 38197906d00cSAndrea Arcangeli * us because we hold the mm_all_locks_mutex. 38207906d00cSAndrea Arcangeli * 38217906d00cSAndrea Arcangeli * We must however clear the bitflag before unlocking 3822bf181b9fSMichel Lespinasse * the vma so the users using the anon_vma->rb_root will 38237906d00cSAndrea Arcangeli * never see our bitflag. 38247906d00cSAndrea Arcangeli * 38257906d00cSAndrea Arcangeli * No need of atomic instructions here, head.next 38267906d00cSAndrea Arcangeli * can't change from under us until we release the 38275a505085SIngo Molnar * anon_vma->root->rwsem. 38287906d00cSAndrea Arcangeli */ 38297906d00cSAndrea Arcangeli if (!__test_and_clear_bit(0, (unsigned long *) 3830f808c13fSDavidlohr Bueso &anon_vma->root->rb_root.rb_root.rb_node)) 38317906d00cSAndrea Arcangeli BUG(); 383208b52706SKonstantin Khlebnikov anon_vma_unlock_write(anon_vma); 38337906d00cSAndrea Arcangeli } 38347906d00cSAndrea Arcangeli } 38357906d00cSAndrea Arcangeli 38367906d00cSAndrea Arcangeli static void vm_unlock_mapping(struct address_space *mapping) 38377906d00cSAndrea Arcangeli { 38387906d00cSAndrea Arcangeli if (test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) { 38397906d00cSAndrea Arcangeli /* 38407906d00cSAndrea Arcangeli * AS_MM_ALL_LOCKS can't change to 0 from under us 38417906d00cSAndrea Arcangeli * because we hold the mm_all_locks_mutex. 38427906d00cSAndrea Arcangeli */ 384383cde9e8SDavidlohr Bueso i_mmap_unlock_write(mapping); 38447906d00cSAndrea Arcangeli if (!test_and_clear_bit(AS_MM_ALL_LOCKS, 38457906d00cSAndrea Arcangeli &mapping->flags)) 38467906d00cSAndrea Arcangeli BUG(); 38477906d00cSAndrea Arcangeli } 38487906d00cSAndrea Arcangeli } 38497906d00cSAndrea Arcangeli 38507906d00cSAndrea Arcangeli /* 3851c1e8d7c6SMichel Lespinasse * The mmap_lock cannot be released by the caller until 38527906d00cSAndrea Arcangeli * mm_drop_all_locks() returns. 38537906d00cSAndrea Arcangeli */ 38547906d00cSAndrea Arcangeli void mm_drop_all_locks(struct mm_struct *mm) 38557906d00cSAndrea Arcangeli { 38567906d00cSAndrea Arcangeli struct vm_area_struct *vma; 38575beb4930SRik van Riel struct anon_vma_chain *avc; 3858763ecb03SLiam R. Howlett MA_STATE(mas, &mm->mm_mt, 0, 0); 38597906d00cSAndrea Arcangeli 3860325bca1fSRolf Eike Beer mmap_assert_write_locked(mm); 38617906d00cSAndrea Arcangeli BUG_ON(!mutex_is_locked(&mm_all_locks_mutex)); 38627906d00cSAndrea Arcangeli 3863763ecb03SLiam R. Howlett mas_for_each(&mas, vma, ULONG_MAX) { 38647906d00cSAndrea Arcangeli if (vma->anon_vma) 38655beb4930SRik van Riel list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) 38665beb4930SRik van Riel vm_unlock_anon_vma(avc->anon_vma); 38677906d00cSAndrea Arcangeli if (vma->vm_file && vma->vm_file->f_mapping) 38687906d00cSAndrea Arcangeli vm_unlock_mapping(vma->vm_file->f_mapping); 38697906d00cSAndrea Arcangeli } 38707906d00cSAndrea Arcangeli 38717906d00cSAndrea Arcangeli mutex_unlock(&mm_all_locks_mutex); 38727906d00cSAndrea Arcangeli } 38738feae131SDavid Howells 38748feae131SDavid Howells /* 38753edf41d8Sseokhoon.yoon * initialise the percpu counter for VM 38768feae131SDavid Howells */ 38778feae131SDavid Howells void __init mmap_init(void) 38788feae131SDavid Howells { 387900a62ce9SKOSAKI Motohiro int ret; 388000a62ce9SKOSAKI Motohiro 3881908c7f19STejun Heo ret = percpu_counter_init(&vm_committed_as, 0, GFP_KERNEL); 388200a62ce9SKOSAKI Motohiro VM_BUG_ON(ret); 38838feae131SDavid Howells } 3884c9b1d098SAndrew Shewmaker 3885c9b1d098SAndrew Shewmaker /* 3886c9b1d098SAndrew Shewmaker * Initialise sysctl_user_reserve_kbytes. 3887c9b1d098SAndrew Shewmaker * 3888c9b1d098SAndrew Shewmaker * This is intended to prevent a user from starting a single memory hogging 3889c9b1d098SAndrew Shewmaker * process, such that they cannot recover (kill the hog) in OVERCOMMIT_NEVER 3890c9b1d098SAndrew Shewmaker * mode. 3891c9b1d098SAndrew Shewmaker * 3892c9b1d098SAndrew Shewmaker * The default value is min(3% of free memory, 128MB) 3893c9b1d098SAndrew Shewmaker * 128MB is enough to recover with sshd/login, bash, and top/kill. 3894c9b1d098SAndrew Shewmaker */ 38951640879aSAndrew Shewmaker static int init_user_reserve(void) 3896c9b1d098SAndrew Shewmaker { 3897c9b1d098SAndrew Shewmaker unsigned long free_kbytes; 3898c9b1d098SAndrew Shewmaker 3899b1773e0eSZhangPeng free_kbytes = K(global_zone_page_state(NR_FREE_PAGES)); 3900c9b1d098SAndrew Shewmaker 3901c9b1d098SAndrew Shewmaker sysctl_user_reserve_kbytes = min(free_kbytes / 32, 1UL << 17); 3902c9b1d098SAndrew Shewmaker return 0; 3903c9b1d098SAndrew Shewmaker } 3904a64fb3cdSPaul Gortmaker subsys_initcall(init_user_reserve); 39054eeab4f5SAndrew Shewmaker 39064eeab4f5SAndrew Shewmaker /* 39074eeab4f5SAndrew Shewmaker * Initialise sysctl_admin_reserve_kbytes. 39084eeab4f5SAndrew Shewmaker * 39094eeab4f5SAndrew Shewmaker * The purpose of sysctl_admin_reserve_kbytes is to allow the sys admin 39104eeab4f5SAndrew Shewmaker * to log in and kill a memory hogging process. 39114eeab4f5SAndrew Shewmaker * 39124eeab4f5SAndrew Shewmaker * Systems with more than 256MB will reserve 8MB, enough to recover 39134eeab4f5SAndrew Shewmaker * with sshd, bash, and top in OVERCOMMIT_GUESS. Smaller systems will 39144eeab4f5SAndrew Shewmaker * only reserve 3% of free pages by default. 39154eeab4f5SAndrew Shewmaker */ 39161640879aSAndrew Shewmaker static int init_admin_reserve(void) 39174eeab4f5SAndrew Shewmaker { 39184eeab4f5SAndrew Shewmaker unsigned long free_kbytes; 39194eeab4f5SAndrew Shewmaker 3920b1773e0eSZhangPeng free_kbytes = K(global_zone_page_state(NR_FREE_PAGES)); 39214eeab4f5SAndrew Shewmaker 39224eeab4f5SAndrew Shewmaker sysctl_admin_reserve_kbytes = min(free_kbytes / 32, 1UL << 13); 39234eeab4f5SAndrew Shewmaker return 0; 39244eeab4f5SAndrew Shewmaker } 3925a64fb3cdSPaul Gortmaker subsys_initcall(init_admin_reserve); 39261640879aSAndrew Shewmaker 39271640879aSAndrew Shewmaker /* 39281640879aSAndrew Shewmaker * Reinititalise user and admin reserves if memory is added or removed. 39291640879aSAndrew Shewmaker * 39301640879aSAndrew Shewmaker * The default user reserve max is 128MB, and the default max for the 39311640879aSAndrew Shewmaker * admin reserve is 8MB. These are usually, but not always, enough to 39321640879aSAndrew Shewmaker * enable recovery from a memory hogging process using login/sshd, a shell, 39331640879aSAndrew Shewmaker * and tools like top. It may make sense to increase or even disable the 39341640879aSAndrew Shewmaker * reserve depending on the existence of swap or variations in the recovery 39351640879aSAndrew Shewmaker * tools. So, the admin may have changed them. 39361640879aSAndrew Shewmaker * 39371640879aSAndrew Shewmaker * If memory is added and the reserves have been eliminated or increased above 39381640879aSAndrew Shewmaker * the default max, then we'll trust the admin. 39391640879aSAndrew Shewmaker * 39401640879aSAndrew Shewmaker * If memory is removed and there isn't enough free memory, then we 39411640879aSAndrew Shewmaker * need to reset the reserves. 39421640879aSAndrew Shewmaker * 39431640879aSAndrew Shewmaker * Otherwise keep the reserve set by the admin. 39441640879aSAndrew Shewmaker */ 39451640879aSAndrew Shewmaker static int reserve_mem_notifier(struct notifier_block *nb, 39461640879aSAndrew Shewmaker unsigned long action, void *data) 39471640879aSAndrew Shewmaker { 39481640879aSAndrew Shewmaker unsigned long tmp, free_kbytes; 39491640879aSAndrew Shewmaker 39501640879aSAndrew Shewmaker switch (action) { 39511640879aSAndrew Shewmaker case MEM_ONLINE: 39521640879aSAndrew Shewmaker /* Default max is 128MB. Leave alone if modified by operator. */ 39531640879aSAndrew Shewmaker tmp = sysctl_user_reserve_kbytes; 39541640879aSAndrew Shewmaker if (0 < tmp && tmp < (1UL << 17)) 39551640879aSAndrew Shewmaker init_user_reserve(); 39561640879aSAndrew Shewmaker 39571640879aSAndrew Shewmaker /* Default max is 8MB. Leave alone if modified by operator. */ 39581640879aSAndrew Shewmaker tmp = sysctl_admin_reserve_kbytes; 39591640879aSAndrew Shewmaker if (0 < tmp && tmp < (1UL << 13)) 39601640879aSAndrew Shewmaker init_admin_reserve(); 39611640879aSAndrew Shewmaker 39621640879aSAndrew Shewmaker break; 39631640879aSAndrew Shewmaker case MEM_OFFLINE: 3964b1773e0eSZhangPeng free_kbytes = K(global_zone_page_state(NR_FREE_PAGES)); 39651640879aSAndrew Shewmaker 39661640879aSAndrew Shewmaker if (sysctl_user_reserve_kbytes > free_kbytes) { 39671640879aSAndrew Shewmaker init_user_reserve(); 39681640879aSAndrew Shewmaker pr_info("vm.user_reserve_kbytes reset to %lu\n", 39691640879aSAndrew Shewmaker sysctl_user_reserve_kbytes); 39701640879aSAndrew Shewmaker } 39711640879aSAndrew Shewmaker 39721640879aSAndrew Shewmaker if (sysctl_admin_reserve_kbytes > free_kbytes) { 39731640879aSAndrew Shewmaker init_admin_reserve(); 39741640879aSAndrew Shewmaker pr_info("vm.admin_reserve_kbytes reset to %lu\n", 39751640879aSAndrew Shewmaker sysctl_admin_reserve_kbytes); 39761640879aSAndrew Shewmaker } 39771640879aSAndrew Shewmaker break; 39781640879aSAndrew Shewmaker default: 39791640879aSAndrew Shewmaker break; 39801640879aSAndrew Shewmaker } 39811640879aSAndrew Shewmaker return NOTIFY_OK; 39821640879aSAndrew Shewmaker } 39831640879aSAndrew Shewmaker 39841640879aSAndrew Shewmaker static int __meminit init_reserve_notifier(void) 39851640879aSAndrew Shewmaker { 39861eeaa4fdSLiu Shixin if (hotplug_memory_notifier(reserve_mem_notifier, DEFAULT_CALLBACK_PRI)) 3987b1de0d13SMitchel Humpherys pr_err("Failed registering memory add/remove notifier for admin reserve\n"); 39881640879aSAndrew Shewmaker 39891640879aSAndrew Shewmaker return 0; 39901640879aSAndrew Shewmaker } 3991a64fb3cdSPaul Gortmaker subsys_initcall(init_reserve_notifier); 3992