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 { 1101da177e4SLinus Torvalds if (vma->vm_flags & VM_SHARED) 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 { 3871da177e4SLinus Torvalds if (vma->vm_flags & VM_SHARED) 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 58604241ffeSLiam R. Howlett * 58704241ffeSLiam R. Howlett * Returns: 0 on success. 58804241ffeSLiam R. Howlett */ 58904241ffeSLiam R. Howlett static inline int dup_anon_vma(struct vm_area_struct *dst, 59004241ffeSLiam R. Howlett struct vm_area_struct *src) 59104241ffeSLiam R. Howlett { 59204241ffeSLiam R. Howlett /* 59304241ffeSLiam R. Howlett * Easily overlooked: when mprotect shifts the boundary, make sure the 59404241ffeSLiam R. Howlett * expanding vma has anon_vma set if the shrinking vma had, to cover any 59504241ffeSLiam R. Howlett * anon pages imported. 59604241ffeSLiam R. Howlett */ 59704241ffeSLiam R. Howlett if (src->anon_vma && !dst->anon_vma) { 598c9d6e982SSuren Baghdasaryan vma_assert_write_locked(dst); 59904241ffeSLiam R. Howlett dst->anon_vma = src->anon_vma; 60004241ffeSLiam R. Howlett return anon_vma_clone(dst, src); 60104241ffeSLiam R. Howlett } 60204241ffeSLiam R. Howlett 60304241ffeSLiam R. Howlett return 0; 60404241ffeSLiam R. Howlett } 60504241ffeSLiam R. Howlett 60604241ffeSLiam R. Howlett /* 6079303d3e1SLiam R. Howlett * vma_expand - Expand an existing VMA 6089303d3e1SLiam R. Howlett * 6099303d3e1SLiam R. Howlett * @vmi: The vma iterator 6109303d3e1SLiam R. Howlett * @vma: The vma to expand 6119303d3e1SLiam R. Howlett * @start: The start of the vma 6129303d3e1SLiam R. Howlett * @end: The exclusive end of the vma 6139303d3e1SLiam R. Howlett * @pgoff: The page offset of vma 6149303d3e1SLiam R. Howlett * @next: The current of next vma. 6159303d3e1SLiam R. Howlett * 6169303d3e1SLiam R. Howlett * Expand @vma to @start and @end. Can expand off the start and end. Will 6179303d3e1SLiam R. Howlett * expand over @next if it's different from @vma and @end == @next->vm_end. 6189303d3e1SLiam R. Howlett * Checking if the @vma can expand and merge with @next needs to be handled by 6199303d3e1SLiam R. Howlett * the caller. 6209303d3e1SLiam R. Howlett * 6219303d3e1SLiam R. Howlett * Returns: 0 on success 6229303d3e1SLiam R. Howlett */ 6237c9813e8SLiam R. Howlett int vma_expand(struct vma_iterator *vmi, struct vm_area_struct *vma, 6249303d3e1SLiam R. Howlett unsigned long start, unsigned long end, pgoff_t pgoff, 6259303d3e1SLiam R. Howlett struct vm_area_struct *next) 6269303d3e1SLiam R. Howlett { 62768cefec5SLiam R. Howlett bool remove_next = false; 6289303d3e1SLiam R. Howlett struct vma_prepare vp; 6299303d3e1SLiam R. Howlett 630c9d6e982SSuren Baghdasaryan vma_start_write(vma); 6319303d3e1SLiam R. Howlett if (next && (vma != next) && (end == next->vm_end)) { 63204241ffeSLiam R. Howlett int ret; 6339303d3e1SLiam R. Howlett 63404241ffeSLiam R. Howlett remove_next = true; 635c9d6e982SSuren Baghdasaryan vma_start_write(next); 63604241ffeSLiam R. Howlett ret = dup_anon_vma(vma, next); 63704241ffeSLiam R. Howlett if (ret) 63804241ffeSLiam R. Howlett return ret; 6399303d3e1SLiam R. Howlett } 6409303d3e1SLiam R. Howlett 64168cefec5SLiam R. Howlett init_multi_vma_prep(&vp, vma, NULL, remove_next ? next : NULL, NULL); 6429303d3e1SLiam R. Howlett /* Not merging but overwriting any part of next is not handled. */ 6439303d3e1SLiam R. Howlett VM_WARN_ON(next && !vp.remove && 6449303d3e1SLiam R. Howlett next != vma && end > next->vm_start); 6459303d3e1SLiam R. Howlett /* Only handles expanding */ 6469303d3e1SLiam R. Howlett VM_WARN_ON(vma->vm_start < start || vma->vm_end > end); 6479303d3e1SLiam R. Howlett 648b5df0922SLiam R. Howlett /* Note: vma iterator must be pointing to 'start' */ 649b5df0922SLiam R. Howlett vma_iter_config(vmi, start, end); 650b5df0922SLiam R. Howlett if (vma_iter_prealloc(vmi, vma)) 6519303d3e1SLiam R. Howlett goto nomem; 6529303d3e1SLiam R. Howlett 653ccf1d78dSSuren Baghdasaryan vma_prepare(&vp); 6549303d3e1SLiam R. Howlett vma_adjust_trans_huge(vma, start, end, 0); 6559303d3e1SLiam R. Howlett vma->vm_start = start; 6569303d3e1SLiam R. Howlett vma->vm_end = end; 6579303d3e1SLiam R. Howlett vma->vm_pgoff = pgoff; 6589303d3e1SLiam R. Howlett vma_iter_store(vmi, vma); 6599303d3e1SLiam R. Howlett 6609303d3e1SLiam R. Howlett vma_complete(&vp, vmi, vma->vm_mm); 6619303d3e1SLiam R. Howlett return 0; 6629303d3e1SLiam R. Howlett 6639303d3e1SLiam R. Howlett nomem: 6649303d3e1SLiam R. Howlett return -ENOMEM; 6659303d3e1SLiam R. Howlett } 666cf51e86dSLiam R. Howlett 667cf51e86dSLiam R. Howlett /* 668cf51e86dSLiam R. Howlett * vma_shrink() - Reduce an existing VMAs memory area 669cf51e86dSLiam R. Howlett * @vmi: The vma iterator 670cf51e86dSLiam R. Howlett * @vma: The VMA to modify 671cf51e86dSLiam R. Howlett * @start: The new start 672cf51e86dSLiam R. Howlett * @end: The new end 673cf51e86dSLiam R. Howlett * 674cf51e86dSLiam R. Howlett * Returns: 0 on success, -ENOMEM otherwise 675cf51e86dSLiam R. Howlett */ 676cf51e86dSLiam R. Howlett int vma_shrink(struct vma_iterator *vmi, struct vm_area_struct *vma, 677cf51e86dSLiam R. Howlett unsigned long start, unsigned long end, pgoff_t pgoff) 678cf51e86dSLiam R. Howlett { 679cf51e86dSLiam R. Howlett struct vma_prepare vp; 680cf51e86dSLiam R. Howlett 681cf51e86dSLiam R. Howlett WARN_ON((vma->vm_start != start) && (vma->vm_end != end)); 682cf51e86dSLiam R. Howlett 683b5df0922SLiam R. Howlett if (vma->vm_start < start) 684b5df0922SLiam R. Howlett vma_iter_config(vmi, vma->vm_start, start); 685b5df0922SLiam R. Howlett else 686b5df0922SLiam R. Howlett vma_iter_config(vmi, end, vma->vm_end); 687b5df0922SLiam R. Howlett 688b5df0922SLiam R. Howlett if (vma_iter_prealloc(vmi, NULL)) 689cf51e86dSLiam R. Howlett return -ENOMEM; 690cf51e86dSLiam R. Howlett 691c9d6e982SSuren Baghdasaryan vma_start_write(vma); 692c9d6e982SSuren Baghdasaryan 693cf51e86dSLiam R. Howlett init_vma_prep(&vp, vma); 694cf51e86dSLiam R. Howlett vma_prepare(&vp); 695ccf1d78dSSuren Baghdasaryan vma_adjust_trans_huge(vma, start, end, 0); 696cf51e86dSLiam R. Howlett 697b5df0922SLiam R. Howlett vma_iter_clear(vmi); 698cf51e86dSLiam R. Howlett vma->vm_start = start; 699cf51e86dSLiam R. Howlett vma->vm_end = end; 700cf51e86dSLiam R. Howlett vma->vm_pgoff = pgoff; 701cf51e86dSLiam R. Howlett vma_complete(&vp, vmi, vma->vm_mm); 702cf51e86dSLiam R. Howlett return 0; 703cf51e86dSLiam R. Howlett } 704cf51e86dSLiam R. Howlett 7059303d3e1SLiam R. Howlett /* 7061da177e4SLinus Torvalds * If the vma has a ->close operation then the driver probably needs to release 707714965caSVlastimil Babka * per-vma resources, so we don't attempt to merge those if the caller indicates 708714965caSVlastimil Babka * the current vma may be removed as part of the merge. 7091da177e4SLinus Torvalds */ 7102dbf4010SVlastimil Babka static inline bool is_mergeable_vma(struct vm_area_struct *vma, 71119a809afSAndrea Arcangeli struct file *file, unsigned long vm_flags, 7129a10064fSColin Cross struct vm_userfaultfd_ctx vm_userfaultfd_ctx, 713714965caSVlastimil Babka struct anon_vma_name *anon_name, bool may_remove_vma) 7141da177e4SLinus Torvalds { 71534228d47SCyrill Gorcunov /* 71634228d47SCyrill Gorcunov * VM_SOFTDIRTY should not prevent from VMA merging, if we 71734228d47SCyrill Gorcunov * match the flags but dirty bit -- the caller should mark 71834228d47SCyrill Gorcunov * merged VMA as dirty. If dirty bit won't be excluded from 7198bb4e7a2SWei Yang * comparison, we increase pressure on the memory system forcing 72034228d47SCyrill Gorcunov * the kernel to generate new VMAs when old one could be 72134228d47SCyrill Gorcunov * extended instead. 72234228d47SCyrill Gorcunov */ 72334228d47SCyrill Gorcunov if ((vma->vm_flags ^ vm_flags) & ~VM_SOFTDIRTY) 7242dbf4010SVlastimil Babka return false; 7251da177e4SLinus Torvalds if (vma->vm_file != file) 7262dbf4010SVlastimil Babka return false; 727714965caSVlastimil Babka if (may_remove_vma && vma->vm_ops && vma->vm_ops->close) 7282dbf4010SVlastimil Babka return false; 72919a809afSAndrea Arcangeli if (!is_mergeable_vm_userfaultfd_ctx(vma, vm_userfaultfd_ctx)) 7302dbf4010SVlastimil Babka return false; 7315c26f6acSSuren Baghdasaryan if (!anon_vma_name_eq(anon_vma_name(vma), anon_name)) 7322dbf4010SVlastimil Babka return false; 7332dbf4010SVlastimil Babka return true; 7341da177e4SLinus Torvalds } 7351da177e4SLinus Torvalds 7362dbf4010SVlastimil Babka static inline bool is_mergeable_anon_vma(struct anon_vma *anon_vma1, 7372dbf4010SVlastimil Babka struct anon_vma *anon_vma2, struct vm_area_struct *vma) 7381da177e4SLinus Torvalds { 739965f55deSShaohua Li /* 740965f55deSShaohua Li * The list_is_singular() test is to avoid merging VMA cloned from 741965f55deSShaohua Li * parents. This can improve scalability caused by anon_vma lock. 742965f55deSShaohua Li */ 743965f55deSShaohua Li if ((!anon_vma1 || !anon_vma2) && (!vma || 744965f55deSShaohua Li list_is_singular(&vma->anon_vma_chain))) 7452dbf4010SVlastimil Babka return true; 746965f55deSShaohua Li return anon_vma1 == anon_vma2; 7471da177e4SLinus Torvalds } 7481da177e4SLinus Torvalds 7491da177e4SLinus Torvalds /* 7501da177e4SLinus Torvalds * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff) 7511da177e4SLinus Torvalds * in front of (at a lower virtual address and file offset than) the vma. 7521da177e4SLinus Torvalds * 7531da177e4SLinus Torvalds * We cannot merge two vmas if they have differently assigned (non-NULL) 7541da177e4SLinus Torvalds * anon_vmas, nor if same anon_vma is assigned but offsets incompatible. 7551da177e4SLinus Torvalds * 7561da177e4SLinus Torvalds * We don't check here for the merged mmap wrapping around the end of pagecache 75745e55300SPeter Collingbourne * indices (16TB on ia32) because do_mmap() does not permit mmap's which 7581da177e4SLinus Torvalds * wrap, nor mmaps which cover the final page at index -1UL. 759714965caSVlastimil Babka * 760714965caSVlastimil Babka * We assume the vma may be removed as part of the merge. 7611da177e4SLinus Torvalds */ 7622dbf4010SVlastimil Babka static bool 7631da177e4SLinus Torvalds can_vma_merge_before(struct vm_area_struct *vma, unsigned long vm_flags, 76419a809afSAndrea Arcangeli struct anon_vma *anon_vma, struct file *file, 7652dbf4010SVlastimil Babka pgoff_t vm_pgoff, struct vm_userfaultfd_ctx vm_userfaultfd_ctx, 7665c26f6acSSuren Baghdasaryan struct anon_vma_name *anon_name) 7671da177e4SLinus Torvalds { 768714965caSVlastimil Babka if (is_mergeable_vma(vma, file, vm_flags, vm_userfaultfd_ctx, anon_name, true) && 769965f55deSShaohua Li is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) { 7701da177e4SLinus Torvalds if (vma->vm_pgoff == vm_pgoff) 7712dbf4010SVlastimil Babka return true; 7721da177e4SLinus Torvalds } 7732dbf4010SVlastimil Babka return false; 7741da177e4SLinus Torvalds } 7751da177e4SLinus Torvalds 7761da177e4SLinus Torvalds /* 7771da177e4SLinus Torvalds * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff) 7781da177e4SLinus Torvalds * beyond (at a higher virtual address and file offset than) the vma. 7791da177e4SLinus Torvalds * 7801da177e4SLinus Torvalds * We cannot merge two vmas if they have differently assigned (non-NULL) 7811da177e4SLinus Torvalds * anon_vmas, nor if same anon_vma is assigned but offsets incompatible. 782714965caSVlastimil Babka * 783714965caSVlastimil Babka * We assume that vma is not removed as part of the merge. 7841da177e4SLinus Torvalds */ 7852dbf4010SVlastimil Babka static bool 7861da177e4SLinus Torvalds can_vma_merge_after(struct vm_area_struct *vma, unsigned long vm_flags, 78719a809afSAndrea Arcangeli struct anon_vma *anon_vma, struct file *file, 7882dbf4010SVlastimil Babka pgoff_t vm_pgoff, struct vm_userfaultfd_ctx vm_userfaultfd_ctx, 7895c26f6acSSuren Baghdasaryan struct anon_vma_name *anon_name) 7901da177e4SLinus Torvalds { 791714965caSVlastimil Babka if (is_mergeable_vma(vma, file, vm_flags, vm_userfaultfd_ctx, anon_name, false) && 792965f55deSShaohua Li is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) { 7931da177e4SLinus Torvalds pgoff_t vm_pglen; 794d6e93217SLibin vm_pglen = vma_pages(vma); 7951da177e4SLinus Torvalds if (vma->vm_pgoff + vm_pglen == vm_pgoff) 7962dbf4010SVlastimil Babka return true; 7971da177e4SLinus Torvalds } 7982dbf4010SVlastimil Babka return false; 7991da177e4SLinus Torvalds } 8001da177e4SLinus Torvalds 8011da177e4SLinus Torvalds /* 8029a10064fSColin Cross * Given a mapping request (addr,end,vm_flags,file,pgoff,anon_name), 8039a10064fSColin Cross * figure out whether that can be merged with its predecessor or its 8049a10064fSColin Cross * successor. Or both (it neatly fills a hole). 8051da177e4SLinus Torvalds * 8061da177e4SLinus Torvalds * In most cases - when called for mmap, brk or mremap - [addr,end) is 8071da177e4SLinus Torvalds * certain not to be mapped by the time vma_merge is called; but when 8081da177e4SLinus Torvalds * called for mprotect, it is certain to be already mapped (either at 8091da177e4SLinus Torvalds * an offset within prev, or at the start of next), and the flags of 8101da177e4SLinus Torvalds * this area are about to be changed to vm_flags - and the no-change 8111da177e4SLinus Torvalds * case has already been eliminated. 8121da177e4SLinus Torvalds * 813fcfccd91SLorenzo Stoakes * The following mprotect cases have to be considered, where **** is 8141da177e4SLinus Torvalds * the area passed down from mprotect_fixup, never extending beyond one 815fcfccd91SLorenzo Stoakes * vma, PPPP is the previous vma, CCCC is a concurrent vma that starts 816fcfccd91SLorenzo Stoakes * at the same address as **** and is of the same or larger span, and 817fcfccd91SLorenzo Stoakes * NNNN the next vma after ****: 8181da177e4SLinus Torvalds * 819fcfccd91SLorenzo Stoakes * **** **** **** 820fcfccd91SLorenzo Stoakes * PPPPPPNNNNNN PPPPPPNNNNNN PPPPPPCCCCCC 8215d42ab29SWei Yang * cannot merge might become might become 822fcfccd91SLorenzo Stoakes * PPNNNNNNNNNN PPPPPPPPPPCC 8235d42ab29SWei Yang * mmap, brk or case 4 below case 5 below 8245d42ab29SWei Yang * mremap move: 825fcfccd91SLorenzo Stoakes * **** **** 826fcfccd91SLorenzo Stoakes * PPPP NNNN PPPPCCCCNNNN 8275d42ab29SWei Yang * might become might become 8285d42ab29SWei Yang * PPPPPPPPPPPP 1 or PPPPPPPPPPPP 6 or 829fcfccd91SLorenzo Stoakes * PPPPPPPPNNNN 2 or PPPPPPPPNNNN 7 or 830fcfccd91SLorenzo Stoakes * PPPPNNNNNNNN 3 PPPPNNNNNNNN 8 8311da177e4SLinus Torvalds * 832fcfccd91SLorenzo Stoakes * It is important for case 8 that the vma CCCC overlapping the 833fcfccd91SLorenzo Stoakes * region **** is never going to extended over NNNN. Instead NNNN must 834fcfccd91SLorenzo Stoakes * be extended in region **** and CCCC must be removed. This way in 8350503ea8fSLiam R. Howlett * all cases where vma_merge succeeds, the moment vma_merge drops the 836e86f15eeSAndrea Arcangeli * rmap_locks, the properties of the merged vma will be already 837e86f15eeSAndrea Arcangeli * correct for the whole merged range. Some of those properties like 838e86f15eeSAndrea Arcangeli * vm_page_prot/vm_flags may be accessed by rmap_walks and they must 839e86f15eeSAndrea Arcangeli * be correct for the whole merged range immediately after the 840fcfccd91SLorenzo Stoakes * rmap_locks are released. Otherwise if NNNN would be removed and 841fcfccd91SLorenzo Stoakes * CCCC would be extended over the NNNN range, remove_migration_ptes 842e86f15eeSAndrea Arcangeli * or other rmap walkers (if working on addresses beyond the "end" 843fcfccd91SLorenzo Stoakes * parameter) may establish ptes with the wrong permissions of CCCC 844fcfccd91SLorenzo Stoakes * instead of the right permissions of NNNN. 8450503ea8fSLiam R. Howlett * 8460503ea8fSLiam R. Howlett * In the code below: 8470503ea8fSLiam R. Howlett * PPPP is represented by *prev 848fcfccd91SLorenzo Stoakes * CCCC is represented by *curr or not represented at all (NULL) 849fcfccd91SLorenzo Stoakes * NNNN is represented by *next or not represented at all (NULL) 850fcfccd91SLorenzo Stoakes * **** is not represented - it will be merged and the vma containing the 8519e8a39d2SVlastimil Babka * area is returned, or the function will return NULL 8521da177e4SLinus Torvalds */ 8539760ebffSLiam R. Howlett struct vm_area_struct *vma_merge(struct vma_iterator *vmi, struct mm_struct *mm, 8541da177e4SLinus Torvalds struct vm_area_struct *prev, unsigned long addr, 8551da177e4SLinus Torvalds unsigned long end, unsigned long vm_flags, 8561da177e4SLinus Torvalds struct anon_vma *anon_vma, struct file *file, 85719a809afSAndrea Arcangeli pgoff_t pgoff, struct mempolicy *policy, 8589a10064fSColin Cross struct vm_userfaultfd_ctx vm_userfaultfd_ctx, 8595c26f6acSSuren Baghdasaryan struct anon_vma_name *anon_name) 8601da177e4SLinus Torvalds { 861b0729ae0SLorenzo Stoakes struct vm_area_struct *curr, *next, *res; 8620503ea8fSLiam R. Howlett struct vm_area_struct *vma, *adjust, *remove, *remove2; 8630173db4fSLorenzo Stoakes struct vma_prepare vp; 8640173db4fSLorenzo Stoakes pgoff_t vma_pgoff; 8650173db4fSLorenzo Stoakes int err = 0; 866eef19944SJakub Matěna bool merge_prev = false; 867eef19944SJakub Matěna bool merge_next = false; 8680503ea8fSLiam R. Howlett bool vma_expanded = false; 8690503ea8fSLiam R. Howlett unsigned long vma_start = addr; 8700173db4fSLorenzo Stoakes unsigned long vma_end = end; 8710173db4fSLorenzo Stoakes pgoff_t pglen = (end - addr) >> PAGE_SHIFT; 8720173db4fSLorenzo Stoakes long adj_start = 0; 8731da177e4SLinus Torvalds 8741da177e4SLinus Torvalds /* 8751da177e4SLinus Torvalds * We later require that vma->vm_flags == vm_flags, 8761da177e4SLinus Torvalds * so this tests vma->vm_flags & VM_SPECIAL, too. 8771da177e4SLinus Torvalds */ 8781da177e4SLinus Torvalds if (vm_flags & VM_SPECIAL) 8791da177e4SLinus Torvalds return NULL; 8801da177e4SLinus Torvalds 88100cd00a6SLorenzo Stoakes /* Does the input range span an existing VMA? (cases 5 - 8) */ 88200cd00a6SLorenzo Stoakes curr = find_vma_intersection(mm, prev ? prev->vm_end : 0, end); 8831da177e4SLinus Torvalds 88400cd00a6SLorenzo Stoakes if (!curr || /* cases 1 - 4 */ 88500cd00a6SLorenzo Stoakes end == curr->vm_end) /* cases 6 - 8, adjacent VMA */ 88600cd00a6SLorenzo Stoakes next = vma_lookup(mm, end); 88700cd00a6SLorenzo Stoakes else 88800cd00a6SLorenzo Stoakes next = NULL; /* case 5 */ 889e86f15eeSAndrea Arcangeli 8900503ea8fSLiam R. Howlett if (prev) { 8910503ea8fSLiam R. Howlett vma_start = prev->vm_start; 8920503ea8fSLiam R. Howlett vma_pgoff = prev->vm_pgoff; 8930173db4fSLorenzo Stoakes 894eef19944SJakub Matěna /* Can we merge the predecessor? */ 8950173db4fSLorenzo Stoakes if (addr == prev->vm_end && mpol_equal(vma_policy(prev), policy) 8960503ea8fSLiam R. Howlett && can_vma_merge_after(prev, vm_flags, anon_vma, file, 8970503ea8fSLiam R. Howlett pgoff, vm_userfaultfd_ctx, anon_name)) { 898eef19944SJakub Matěna merge_prev = true; 89918b098afSLiam R. Howlett vma_prev(vmi); 9001da177e4SLinus Torvalds } 9010503ea8fSLiam R. Howlett } 902b0729ae0SLorenzo Stoakes 903eef19944SJakub Matěna /* Can we merge the successor? */ 90400cd00a6SLorenzo Stoakes if (next && mpol_equal(policy, vma_policy(next)) && 9050173db4fSLorenzo Stoakes can_vma_merge_before(next, vm_flags, anon_vma, file, pgoff+pglen, 9069a10064fSColin Cross vm_userfaultfd_ctx, anon_name)) { 907eef19944SJakub Matěna merge_next = true; 908eef19944SJakub Matěna } 9090503ea8fSLiam R. Howlett 91029417d29SLorenzo Stoakes /* Verify some invariant that must be enforced by the caller. */ 91129417d29SLorenzo Stoakes VM_WARN_ON(prev && addr <= prev->vm_start); 91229417d29SLorenzo Stoakes VM_WARN_ON(curr && (addr != curr->vm_start || end > curr->vm_end)); 91329417d29SLorenzo Stoakes VM_WARN_ON(addr >= end); 91429417d29SLorenzo Stoakes 9150173db4fSLorenzo Stoakes if (!merge_prev && !merge_next) 9160173db4fSLorenzo Stoakes return NULL; /* Not mergeable. */ 9170173db4fSLorenzo Stoakes 918c9d6e982SSuren Baghdasaryan if (merge_prev) 919c9d6e982SSuren Baghdasaryan vma_start_write(prev); 920c9d6e982SSuren Baghdasaryan 9210173db4fSLorenzo Stoakes res = vma = prev; 9220503ea8fSLiam R. Howlett remove = remove2 = adjust = NULL; 9230173db4fSLorenzo Stoakes 924eef19944SJakub Matěna /* Can we merge both the predecessor and the successor? */ 925eef19944SJakub Matěna if (merge_prev && merge_next && 9260503ea8fSLiam R. Howlett is_mergeable_anon_vma(prev->anon_vma, next->anon_vma, NULL)) { 927c9d6e982SSuren Baghdasaryan vma_start_write(next); 9285ff783f1SVlastimil Babka remove = next; /* case 1 */ 9290503ea8fSLiam R. Howlett vma_end = next->vm_end; 93050dac011SVlastimil Babka err = dup_anon_vma(prev, next); 931fcfccd91SLorenzo Stoakes if (curr) { /* case 6 */ 932c9d6e982SSuren Baghdasaryan vma_start_write(curr); 933fcfccd91SLorenzo Stoakes remove = curr; 9340503ea8fSLiam R. Howlett remove2 = next; 9355ff783f1SVlastimil Babka if (!next->anon_vma) 936fcfccd91SLorenzo Stoakes err = dup_anon_vma(prev, curr); 9370503ea8fSLiam R. Howlett } 9380173db4fSLorenzo Stoakes } else if (merge_prev) { /* case 2 */ 939fcfccd91SLorenzo Stoakes if (curr) { 940c9d6e982SSuren Baghdasaryan vma_start_write(curr); 941fcfccd91SLorenzo Stoakes err = dup_anon_vma(prev, curr); 942fcfccd91SLorenzo Stoakes if (end == curr->vm_end) { /* case 7 */ 943fcfccd91SLorenzo Stoakes remove = curr; 9440503ea8fSLiam R. Howlett } else { /* case 5 */ 945fcfccd91SLorenzo Stoakes adjust = curr; 946fcfccd91SLorenzo Stoakes adj_start = (end - curr->vm_start); 9470503ea8fSLiam R. Howlett } 9480503ea8fSLiam R. Howlett } 9490173db4fSLorenzo Stoakes } else { /* merge_next */ 950c9d6e982SSuren Baghdasaryan vma_start_write(next); 951eef19944SJakub Matěna res = next; 9520503ea8fSLiam R. Howlett if (prev && addr < prev->vm_end) { /* case 4 */ 953c9d6e982SSuren Baghdasaryan vma_start_write(prev); 9540503ea8fSLiam R. Howlett vma_end = addr; 955183b7a60SVlastimil Babka adjust = next; 9561e76454fSVlastimil Babka adj_start = -(prev->vm_end - addr); 957183b7a60SVlastimil Babka err = dup_anon_vma(next, prev); 9580503ea8fSLiam R. Howlett } else { 959b0729ae0SLorenzo Stoakes /* 960b0729ae0SLorenzo Stoakes * Note that cases 3 and 8 are the ONLY ones where prev 961b0729ae0SLorenzo Stoakes * is permitted to be (but is not necessarily) NULL. 962b0729ae0SLorenzo Stoakes */ 9630503ea8fSLiam R. Howlett vma = next; /* case 3 */ 9640503ea8fSLiam R. Howlett vma_start = addr; 9650503ea8fSLiam R. Howlett vma_end = next->vm_end; 9667e775787SVlastimil Babka vma_pgoff = next->vm_pgoff - pglen; 967fcfccd91SLorenzo Stoakes if (curr) { /* case 8 */ 968fcfccd91SLorenzo Stoakes vma_pgoff = curr->vm_pgoff; 969c9d6e982SSuren Baghdasaryan vma_start_write(curr); 970fcfccd91SLorenzo Stoakes remove = curr; 971fcfccd91SLorenzo Stoakes err = dup_anon_vma(next, curr); 9720503ea8fSLiam R. Howlett } 9730503ea8fSLiam R. Howlett } 9741da177e4SLinus Torvalds } 9751da177e4SLinus Torvalds 9760173db4fSLorenzo Stoakes /* Error in anon_vma clone. */ 977eef19944SJakub Matěna if (err) 9781da177e4SLinus Torvalds return NULL; 9790503ea8fSLiam R. Howlett 980b5df0922SLiam R. Howlett if (vma_start < vma->vm_start || vma_end > vma->vm_end) 981b5df0922SLiam R. Howlett vma_expanded = true; 982b5df0922SLiam R. Howlett 983b5df0922SLiam R. Howlett if (vma_expanded) { 984b5df0922SLiam R. Howlett vma_iter_config(vmi, vma_start, vma_end); 985b5df0922SLiam R. Howlett } else { 986b5df0922SLiam R. Howlett vma_iter_config(vmi, adjust->vm_start + adj_start, 987b5df0922SLiam R. Howlett adjust->vm_end); 988b5df0922SLiam R. Howlett } 989b5df0922SLiam R. Howlett 990b5df0922SLiam R. Howlett if (vma_iter_prealloc(vmi, vma)) 9910503ea8fSLiam R. Howlett return NULL; 9920503ea8fSLiam R. Howlett 9930503ea8fSLiam R. Howlett init_multi_vma_prep(&vp, vma, adjust, remove, remove2); 9940503ea8fSLiam R. Howlett VM_WARN_ON(vp.anon_vma && adjust && adjust->anon_vma && 9950503ea8fSLiam R. Howlett vp.anon_vma != adjust->anon_vma); 9960503ea8fSLiam R. Howlett 9970503ea8fSLiam R. Howlett vma_prepare(&vp); 998ccf1d78dSSuren Baghdasaryan vma_adjust_trans_huge(vma, vma_start, vma_end, adj_start); 9990503ea8fSLiam R. Howlett 10000503ea8fSLiam R. Howlett vma->vm_start = vma_start; 10010503ea8fSLiam R. Howlett vma->vm_end = vma_end; 10020503ea8fSLiam R. Howlett vma->vm_pgoff = vma_pgoff; 10030503ea8fSLiam R. Howlett 10040503ea8fSLiam R. Howlett if (vma_expanded) 10050503ea8fSLiam R. Howlett vma_iter_store(vmi, vma); 10060503ea8fSLiam R. Howlett 10071e76454fSVlastimil Babka if (adj_start) { 10081e76454fSVlastimil Babka adjust->vm_start += adj_start; 10091e76454fSVlastimil Babka adjust->vm_pgoff += adj_start >> PAGE_SHIFT; 10101e76454fSVlastimil Babka if (adj_start < 0) { 10110503ea8fSLiam R. Howlett WARN_ON(vma_expanded); 10120503ea8fSLiam R. Howlett vma_iter_store(vmi, next); 10130503ea8fSLiam R. Howlett } 10140503ea8fSLiam R. Howlett } 10150503ea8fSLiam R. Howlett 10160503ea8fSLiam R. Howlett vma_complete(&vp, vmi, mm); 1017eef19944SJakub Matěna khugepaged_enter_vma(res, vm_flags); 10189760ebffSLiam R. Howlett return res; 1019f2ebfe43SLiam R. Howlett } 1020f2ebfe43SLiam R. Howlett 10211da177e4SLinus Torvalds /* 1022b4f315b4SEthon Paul * Rough compatibility check to quickly see if it's even worth looking 1023d0e9fe17SLinus Torvalds * at sharing an anon_vma. 1024d0e9fe17SLinus Torvalds * 1025d0e9fe17SLinus Torvalds * They need to have the same vm_file, and the flags can only differ 1026d0e9fe17SLinus Torvalds * in things that mprotect may change. 1027d0e9fe17SLinus Torvalds * 1028d0e9fe17SLinus Torvalds * NOTE! The fact that we share an anon_vma doesn't _have_ to mean that 1029d0e9fe17SLinus Torvalds * we can merge the two vma's. For example, we refuse to merge a vma if 1030d0e9fe17SLinus Torvalds * there is a vm_ops->close() function, because that indicates that the 1031d0e9fe17SLinus Torvalds * driver is doing some kind of reference counting. But that doesn't 1032d0e9fe17SLinus Torvalds * really matter for the anon_vma sharing case. 1033d0e9fe17SLinus Torvalds */ 1034d0e9fe17SLinus Torvalds static int anon_vma_compatible(struct vm_area_struct *a, struct vm_area_struct *b) 1035d0e9fe17SLinus Torvalds { 1036d0e9fe17SLinus Torvalds return a->vm_end == b->vm_start && 1037d0e9fe17SLinus Torvalds mpol_equal(vma_policy(a), vma_policy(b)) && 1038d0e9fe17SLinus Torvalds a->vm_file == b->vm_file && 10396cb4d9a2SAnshuman Khandual !((a->vm_flags ^ b->vm_flags) & ~(VM_ACCESS_FLAGS | VM_SOFTDIRTY)) && 1040d0e9fe17SLinus Torvalds b->vm_pgoff == a->vm_pgoff + ((b->vm_start - a->vm_start) >> PAGE_SHIFT); 1041d0e9fe17SLinus Torvalds } 1042d0e9fe17SLinus Torvalds 1043d0e9fe17SLinus Torvalds /* 1044d0e9fe17SLinus Torvalds * Do some basic sanity checking to see if we can re-use the anon_vma 1045d0e9fe17SLinus Torvalds * from 'old'. The 'a'/'b' vma's are in VM order - one of them will be 1046d0e9fe17SLinus Torvalds * the same as 'old', the other will be the new one that is trying 1047d0e9fe17SLinus Torvalds * to share the anon_vma. 1048d0e9fe17SLinus Torvalds * 10495b449489SFlorian Rommel * NOTE! This runs with mmap_lock held for reading, so it is possible that 1050d0e9fe17SLinus Torvalds * the anon_vma of 'old' is concurrently in the process of being set up 1051d0e9fe17SLinus Torvalds * by another page fault trying to merge _that_. But that's ok: if it 1052d0e9fe17SLinus Torvalds * is being set up, that automatically means that it will be a singleton 1053d0e9fe17SLinus Torvalds * acceptable for merging, so we can do all of this optimistically. But 10544db0c3c2SJason Low * we do that READ_ONCE() to make sure that we never re-load the pointer. 1055d0e9fe17SLinus Torvalds * 1056d0e9fe17SLinus Torvalds * IOW: that the "list_is_singular()" test on the anon_vma_chain only 1057d0e9fe17SLinus Torvalds * matters for the 'stable anon_vma' case (ie the thing we want to avoid 1058d0e9fe17SLinus Torvalds * is to return an anon_vma that is "complex" due to having gone through 1059d0e9fe17SLinus Torvalds * a fork). 1060d0e9fe17SLinus Torvalds * 1061d0e9fe17SLinus Torvalds * We also make sure that the two vma's are compatible (adjacent, 1062d0e9fe17SLinus Torvalds * and with the same memory policies). That's all stable, even with just 10635b449489SFlorian Rommel * a read lock on the mmap_lock. 1064d0e9fe17SLinus Torvalds */ 1065d0e9fe17SLinus Torvalds static struct anon_vma *reusable_anon_vma(struct vm_area_struct *old, struct vm_area_struct *a, struct vm_area_struct *b) 1066d0e9fe17SLinus Torvalds { 1067d0e9fe17SLinus Torvalds if (anon_vma_compatible(a, b)) { 10684db0c3c2SJason Low struct anon_vma *anon_vma = READ_ONCE(old->anon_vma); 1069d0e9fe17SLinus Torvalds 1070d0e9fe17SLinus Torvalds if (anon_vma && list_is_singular(&old->anon_vma_chain)) 1071d0e9fe17SLinus Torvalds return anon_vma; 1072d0e9fe17SLinus Torvalds } 1073d0e9fe17SLinus Torvalds return NULL; 1074d0e9fe17SLinus Torvalds } 1075d0e9fe17SLinus Torvalds 1076d0e9fe17SLinus Torvalds /* 10771da177e4SLinus Torvalds * find_mergeable_anon_vma is used by anon_vma_prepare, to check 10781da177e4SLinus Torvalds * neighbouring vmas for a suitable anon_vma, before it goes off 10791da177e4SLinus Torvalds * to allocate a new anon_vma. It checks because a repetitive 10801da177e4SLinus Torvalds * sequence of mprotects and faults may otherwise lead to distinct 10811da177e4SLinus Torvalds * anon_vmas being allocated, preventing vma merge in subsequent 10821da177e4SLinus Torvalds * mprotect. 10831da177e4SLinus Torvalds */ 10841da177e4SLinus Torvalds struct anon_vma *find_mergeable_anon_vma(struct vm_area_struct *vma) 10851da177e4SLinus Torvalds { 1086763ecb03SLiam R. Howlett MA_STATE(mas, &vma->vm_mm->mm_mt, vma->vm_end, vma->vm_end); 1087a67c8caaSMiaohe Lin struct anon_vma *anon_vma = NULL; 1088763ecb03SLiam R. Howlett struct vm_area_struct *prev, *next; 10891da177e4SLinus Torvalds 1090a67c8caaSMiaohe Lin /* Try next first. */ 1091763ecb03SLiam R. Howlett next = mas_walk(&mas); 1092763ecb03SLiam R. Howlett if (next) { 1093763ecb03SLiam R. Howlett anon_vma = reusable_anon_vma(next, vma, next); 1094d0e9fe17SLinus Torvalds if (anon_vma) 1095d0e9fe17SLinus Torvalds return anon_vma; 1096a67c8caaSMiaohe Lin } 10971da177e4SLinus Torvalds 1098763ecb03SLiam R. Howlett prev = mas_prev(&mas, 0); 1099763ecb03SLiam R. Howlett VM_BUG_ON_VMA(prev != vma, vma); 1100763ecb03SLiam R. Howlett prev = mas_prev(&mas, 0); 1101a67c8caaSMiaohe Lin /* Try prev next. */ 1102763ecb03SLiam R. Howlett if (prev) 1103763ecb03SLiam R. Howlett anon_vma = reusable_anon_vma(prev, prev, vma); 1104a67c8caaSMiaohe Lin 11051da177e4SLinus Torvalds /* 1106a67c8caaSMiaohe Lin * We might reach here with anon_vma == NULL if we can't find 1107a67c8caaSMiaohe Lin * any reusable anon_vma. 11081da177e4SLinus Torvalds * There's no absolute need to look only at touching neighbours: 11091da177e4SLinus Torvalds * we could search further afield for "compatible" anon_vmas. 11101da177e4SLinus Torvalds * But it would probably just be a waste of time searching, 11111da177e4SLinus Torvalds * or lead to too many vmas hanging off the same anon_vma. 11121da177e4SLinus Torvalds * We're trying to allow mprotect remerging later on, 11131da177e4SLinus Torvalds * not trying to minimize memory used for anon_vmas. 11141da177e4SLinus Torvalds */ 1115a67c8caaSMiaohe Lin return anon_vma; 11161da177e4SLinus Torvalds } 11171da177e4SLinus Torvalds 11181da177e4SLinus Torvalds /* 111940401530SAl Viro * If a hint addr is less than mmap_min_addr change hint to be as 112040401530SAl Viro * low as possible but still greater than mmap_min_addr 112140401530SAl Viro */ 112240401530SAl Viro static inline unsigned long round_hint_to_min(unsigned long hint) 112340401530SAl Viro { 112440401530SAl Viro hint &= PAGE_MASK; 112540401530SAl Viro if (((void *)hint != NULL) && 112640401530SAl Viro (hint < mmap_min_addr)) 112740401530SAl Viro return PAGE_ALIGN(mmap_min_addr); 112840401530SAl Viro return hint; 112940401530SAl Viro } 113040401530SAl Viro 1131b0cc5e89SAndrew Morton bool mlock_future_ok(struct mm_struct *mm, unsigned long flags, 11323c54a298SLorenzo Stoakes unsigned long bytes) 1133363ee17fSDavidlohr Bueso { 11343c54a298SLorenzo Stoakes unsigned long locked_pages, limit_pages; 1135363ee17fSDavidlohr Bueso 11363c54a298SLorenzo Stoakes if (!(flags & VM_LOCKED) || capable(CAP_IPC_LOCK)) 11373c54a298SLorenzo Stoakes return true; 11383c54a298SLorenzo Stoakes 11393c54a298SLorenzo Stoakes locked_pages = bytes >> PAGE_SHIFT; 11403c54a298SLorenzo Stoakes locked_pages += mm->locked_vm; 11413c54a298SLorenzo Stoakes 11423c54a298SLorenzo Stoakes limit_pages = rlimit(RLIMIT_MEMLOCK); 11433c54a298SLorenzo Stoakes limit_pages >>= PAGE_SHIFT; 11443c54a298SLorenzo Stoakes 11453c54a298SLorenzo Stoakes return locked_pages <= limit_pages; 1146363ee17fSDavidlohr Bueso } 1147363ee17fSDavidlohr Bueso 1148be83bbf8SLinus Torvalds static inline u64 file_mmap_size_max(struct file *file, struct inode *inode) 1149be83bbf8SLinus Torvalds { 1150be83bbf8SLinus Torvalds if (S_ISREG(inode->i_mode)) 1151423913adSLinus Torvalds return MAX_LFS_FILESIZE; 1152be83bbf8SLinus Torvalds 1153be83bbf8SLinus Torvalds if (S_ISBLK(inode->i_mode)) 1154be83bbf8SLinus Torvalds return MAX_LFS_FILESIZE; 1155be83bbf8SLinus Torvalds 115676f34950SIvan Khoronzhuk if (S_ISSOCK(inode->i_mode)) 115776f34950SIvan Khoronzhuk return MAX_LFS_FILESIZE; 115876f34950SIvan Khoronzhuk 1159be83bbf8SLinus Torvalds /* Special "we do even unsigned file positions" case */ 1160be83bbf8SLinus Torvalds if (file->f_mode & FMODE_UNSIGNED_OFFSET) 1161be83bbf8SLinus Torvalds return 0; 1162be83bbf8SLinus Torvalds 1163be83bbf8SLinus Torvalds /* Yes, random drivers might want more. But I'm tired of buggy drivers */ 1164be83bbf8SLinus Torvalds return ULONG_MAX; 1165be83bbf8SLinus Torvalds } 1166be83bbf8SLinus Torvalds 1167be83bbf8SLinus Torvalds static inline bool file_mmap_ok(struct file *file, struct inode *inode, 1168be83bbf8SLinus Torvalds unsigned long pgoff, unsigned long len) 1169be83bbf8SLinus Torvalds { 1170be83bbf8SLinus Torvalds u64 maxsize = file_mmap_size_max(file, inode); 1171be83bbf8SLinus Torvalds 1172be83bbf8SLinus Torvalds if (maxsize && len > maxsize) 1173be83bbf8SLinus Torvalds return false; 1174be83bbf8SLinus Torvalds maxsize -= len; 1175be83bbf8SLinus Torvalds if (pgoff > maxsize >> PAGE_SHIFT) 1176be83bbf8SLinus Torvalds return false; 1177be83bbf8SLinus Torvalds return true; 1178be83bbf8SLinus Torvalds } 1179be83bbf8SLinus Torvalds 118040401530SAl Viro /* 11813e4e28c5SMichel Lespinasse * The caller must write-lock current->mm->mmap_lock. 11821da177e4SLinus Torvalds */ 11831fcfd8dbSOleg Nesterov unsigned long do_mmap(struct file *file, unsigned long addr, 11841da177e4SLinus Torvalds unsigned long len, unsigned long prot, 1185592b5fadSYu-cheng Yu unsigned long flags, vm_flags_t vm_flags, 1186592b5fadSYu-cheng Yu unsigned long pgoff, unsigned long *populate, 1187592b5fadSYu-cheng Yu struct list_head *uf) 11881da177e4SLinus Torvalds { 11891da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 119062b5f7d0SDave Hansen int pkey = 0; 11911da177e4SLinus Torvalds 119241badc15SMichel Lespinasse *populate = 0; 1193bebeb3d6SMichel Lespinasse 1194e37609bbSPiotr Kwapulinski if (!len) 1195e37609bbSPiotr Kwapulinski return -EINVAL; 1196e37609bbSPiotr Kwapulinski 11971da177e4SLinus Torvalds /* 11981da177e4SLinus Torvalds * Does the application expect PROT_READ to imply PROT_EXEC? 11991da177e4SLinus Torvalds * 12001da177e4SLinus Torvalds * (the exception is when the underlying filesystem is noexec 12011da177e4SLinus Torvalds * mounted, in which case we dont add PROT_EXEC.) 12021da177e4SLinus Torvalds */ 12031da177e4SLinus Torvalds if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC)) 120490f8572bSEric W. Biederman if (!(file && path_noexec(&file->f_path))) 12051da177e4SLinus Torvalds prot |= PROT_EXEC; 12061da177e4SLinus Torvalds 1207a4ff8e86SMichal Hocko /* force arch specific MAP_FIXED handling in get_unmapped_area */ 1208a4ff8e86SMichal Hocko if (flags & MAP_FIXED_NOREPLACE) 1209a4ff8e86SMichal Hocko flags |= MAP_FIXED; 1210a4ff8e86SMichal Hocko 12117cd94146SEric Paris if (!(flags & MAP_FIXED)) 12127cd94146SEric Paris addr = round_hint_to_min(addr); 12137cd94146SEric Paris 12141da177e4SLinus Torvalds /* Careful about overflows.. */ 12151da177e4SLinus Torvalds len = PAGE_ALIGN(len); 12169206de95SAl Viro if (!len) 12171da177e4SLinus Torvalds return -ENOMEM; 12181da177e4SLinus Torvalds 12191da177e4SLinus Torvalds /* offset overflow? */ 12201da177e4SLinus Torvalds if ((pgoff + (len >> PAGE_SHIFT)) < pgoff) 12211da177e4SLinus Torvalds return -EOVERFLOW; 12221da177e4SLinus Torvalds 12231da177e4SLinus Torvalds /* Too many mappings? */ 12241da177e4SLinus Torvalds if (mm->map_count > sysctl_max_map_count) 12251da177e4SLinus Torvalds return -ENOMEM; 12261da177e4SLinus Torvalds 12271da177e4SLinus Torvalds /* Obtain the address to map to. we verify (or select) it and ensure 12281da177e4SLinus Torvalds * that it represents a valid section of the address space. 12291da177e4SLinus Torvalds */ 12301da177e4SLinus Torvalds addr = get_unmapped_area(file, addr, len, pgoff, flags); 1231ff68dac6SGaowei Pu if (IS_ERR_VALUE(addr)) 12321da177e4SLinus Torvalds return addr; 12331da177e4SLinus Torvalds 1234a4ff8e86SMichal Hocko if (flags & MAP_FIXED_NOREPLACE) { 123535e43c5fSLiam Howlett if (find_vma_intersection(mm, addr, addr + len)) 1236a4ff8e86SMichal Hocko return -EEXIST; 1237a4ff8e86SMichal Hocko } 1238a4ff8e86SMichal Hocko 123962b5f7d0SDave Hansen if (prot == PROT_EXEC) { 124062b5f7d0SDave Hansen pkey = execute_only_pkey(mm); 124162b5f7d0SDave Hansen if (pkey < 0) 124262b5f7d0SDave Hansen pkey = 0; 124362b5f7d0SDave Hansen } 124462b5f7d0SDave Hansen 12451da177e4SLinus Torvalds /* Do simple checking here so the lower-level routines won't have 12461da177e4SLinus Torvalds * to. we assume access permissions have been handled by the open 12471da177e4SLinus Torvalds * of the memory object, so we don't do any here. 12481da177e4SLinus Torvalds */ 1249592b5fadSYu-cheng Yu vm_flags |= calc_vm_prot_bits(prot, pkey) | calc_vm_flag_bits(flags) | 12501da177e4SLinus Torvalds mm->def_flags | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC; 12511da177e4SLinus Torvalds 1252cdf7b341SHuang Shijie if (flags & MAP_LOCKED) 12531da177e4SLinus Torvalds if (!can_do_mlock()) 12541da177e4SLinus Torvalds return -EPERM; 1255ba470de4SRik van Riel 1256b0cc5e89SAndrew Morton if (!mlock_future_ok(mm, vm_flags, len)) 12571da177e4SLinus Torvalds return -EAGAIN; 12581da177e4SLinus Torvalds 12591da177e4SLinus Torvalds if (file) { 1260077bf22bSOleg Nesterov struct inode *inode = file_inode(file); 12611c972597SDan Williams unsigned long flags_mask; 12621c972597SDan Williams 1263be83bbf8SLinus Torvalds if (!file_mmap_ok(file, inode, pgoff, len)) 1264be83bbf8SLinus Torvalds return -EOVERFLOW; 1265be83bbf8SLinus Torvalds 12661c972597SDan Williams flags_mask = LEGACY_MAP_MASK | file->f_op->mmap_supported_flags; 1267077bf22bSOleg Nesterov 12681da177e4SLinus Torvalds switch (flags & MAP_TYPE) { 12691da177e4SLinus Torvalds case MAP_SHARED: 12701c972597SDan Williams /* 12711c972597SDan Williams * Force use of MAP_SHARED_VALIDATE with non-legacy 12721c972597SDan Williams * flags. E.g. MAP_SYNC is dangerous to use with 12731c972597SDan Williams * MAP_SHARED as you don't know which consistency model 12741c972597SDan Williams * you will get. We silently ignore unsupported flags 12751c972597SDan Williams * with MAP_SHARED to preserve backward compatibility. 12761c972597SDan Williams */ 12771c972597SDan Williams flags &= LEGACY_MAP_MASK; 1278e4a9bc58SJoe Perches fallthrough; 12791c972597SDan Williams case MAP_SHARED_VALIDATE: 12801c972597SDan Williams if (flags & ~flags_mask) 12811c972597SDan Williams return -EOPNOTSUPP; 1282dc617f29SDarrick J. Wong if (prot & PROT_WRITE) { 1283dc617f29SDarrick J. Wong if (!(file->f_mode & FMODE_WRITE)) 12841da177e4SLinus Torvalds return -EACCES; 1285dc617f29SDarrick J. Wong if (IS_SWAPFILE(file->f_mapping->host)) 1286dc617f29SDarrick J. Wong return -ETXTBSY; 1287dc617f29SDarrick J. Wong } 12881da177e4SLinus Torvalds 12891da177e4SLinus Torvalds /* 12901da177e4SLinus Torvalds * Make sure we don't allow writing to an append-only 12911da177e4SLinus Torvalds * file.. 12921da177e4SLinus Torvalds */ 12931da177e4SLinus Torvalds if (IS_APPEND(inode) && (file->f_mode & FMODE_WRITE)) 12941da177e4SLinus Torvalds return -EACCES; 12951da177e4SLinus Torvalds 12961da177e4SLinus Torvalds vm_flags |= VM_SHARED | VM_MAYSHARE; 12971da177e4SLinus Torvalds if (!(file->f_mode & FMODE_WRITE)) 12981da177e4SLinus Torvalds vm_flags &= ~(VM_MAYWRITE | VM_SHARED); 1299e4a9bc58SJoe Perches fallthrough; 13001da177e4SLinus Torvalds case MAP_PRIVATE: 13011da177e4SLinus Torvalds if (!(file->f_mode & FMODE_READ)) 13021da177e4SLinus Torvalds return -EACCES; 130390f8572bSEric W. Biederman if (path_noexec(&file->f_path)) { 130480c5606cSLinus Torvalds if (vm_flags & VM_EXEC) 130580c5606cSLinus Torvalds return -EPERM; 130680c5606cSLinus Torvalds vm_flags &= ~VM_MAYEXEC; 130780c5606cSLinus Torvalds } 130880c5606cSLinus Torvalds 130972c2d531SAl Viro if (!file->f_op->mmap) 131080c5606cSLinus Torvalds return -ENODEV; 1311b2c56e4fSOleg Nesterov if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP)) 1312b2c56e4fSOleg Nesterov return -EINVAL; 13131da177e4SLinus Torvalds break; 13141da177e4SLinus Torvalds 13151da177e4SLinus Torvalds default: 13161da177e4SLinus Torvalds return -EINVAL; 13171da177e4SLinus Torvalds } 13181da177e4SLinus Torvalds } else { 13191da177e4SLinus Torvalds switch (flags & MAP_TYPE) { 13201da177e4SLinus Torvalds case MAP_SHARED: 1321b2c56e4fSOleg Nesterov if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP)) 1322b2c56e4fSOleg Nesterov return -EINVAL; 1323ce363942STejun Heo /* 1324ce363942STejun Heo * Ignore pgoff. 1325ce363942STejun Heo */ 1326ce363942STejun Heo pgoff = 0; 13271da177e4SLinus Torvalds vm_flags |= VM_SHARED | VM_MAYSHARE; 13281da177e4SLinus Torvalds break; 13291da177e4SLinus Torvalds case MAP_PRIVATE: 13301da177e4SLinus Torvalds /* 13311da177e4SLinus Torvalds * Set pgoff according to addr for anon_vma. 13321da177e4SLinus Torvalds */ 13331da177e4SLinus Torvalds pgoff = addr >> PAGE_SHIFT; 13341da177e4SLinus Torvalds break; 13351da177e4SLinus Torvalds default: 13361da177e4SLinus Torvalds return -EINVAL; 13371da177e4SLinus Torvalds } 13381da177e4SLinus Torvalds } 13391da177e4SLinus Torvalds 1340c22c0d63SMichel Lespinasse /* 1341c22c0d63SMichel Lespinasse * Set 'VM_NORESERVE' if we should not account for the 1342c22c0d63SMichel Lespinasse * memory use of this mapping. 1343c22c0d63SMichel Lespinasse */ 1344c22c0d63SMichel Lespinasse if (flags & MAP_NORESERVE) { 1345c22c0d63SMichel Lespinasse /* We honor MAP_NORESERVE if allowed to overcommit */ 1346c22c0d63SMichel Lespinasse if (sysctl_overcommit_memory != OVERCOMMIT_NEVER) 1347c22c0d63SMichel Lespinasse vm_flags |= VM_NORESERVE; 1348c22c0d63SMichel Lespinasse 1349c22c0d63SMichel Lespinasse /* hugetlb applies strict overcommit unless MAP_NORESERVE */ 1350c22c0d63SMichel Lespinasse if (file && is_file_hugepages(file)) 1351c22c0d63SMichel Lespinasse vm_flags |= VM_NORESERVE; 1352c22c0d63SMichel Lespinasse } 1353c22c0d63SMichel Lespinasse 1354897ab3e0SMike Rapoport addr = mmap_region(file, addr, len, vm_flags, pgoff, uf); 135509a9f1d2SMichel Lespinasse if (!IS_ERR_VALUE(addr) && 135609a9f1d2SMichel Lespinasse ((vm_flags & VM_LOCKED) || 135709a9f1d2SMichel Lespinasse (flags & (MAP_POPULATE | MAP_NONBLOCK)) == MAP_POPULATE)) 135841badc15SMichel Lespinasse *populate = len; 1359bebeb3d6SMichel Lespinasse return addr; 13600165ab44SMiklos Szeredi } 13616be5ceb0SLinus Torvalds 1362a90f590aSDominik Brodowski unsigned long ksys_mmap_pgoff(unsigned long addr, unsigned long len, 1363a90f590aSDominik Brodowski unsigned long prot, unsigned long flags, 1364a90f590aSDominik Brodowski unsigned long fd, unsigned long pgoff) 136566f0dc48SHugh Dickins { 136666f0dc48SHugh Dickins struct file *file = NULL; 13671e3ee14bSChen Gang unsigned long retval; 136866f0dc48SHugh Dickins 136966f0dc48SHugh Dickins if (!(flags & MAP_ANONYMOUS)) { 1370120a795dSAl Viro audit_mmap_fd(fd, flags); 137166f0dc48SHugh Dickins file = fget(fd); 137266f0dc48SHugh Dickins if (!file) 13731e3ee14bSChen Gang return -EBADF; 13747bba8f0eSZhen Lei if (is_file_hugepages(file)) { 1375af73e4d9SNaoya Horiguchi len = ALIGN(len, huge_page_size(hstate_file(file))); 13767bba8f0eSZhen Lei } else if (unlikely(flags & MAP_HUGETLB)) { 1377493af578SJörn Engel retval = -EINVAL; 1378493af578SJörn Engel goto out_fput; 13797bba8f0eSZhen Lei } 138066f0dc48SHugh Dickins } else if (flags & MAP_HUGETLB) { 1381c103a4dcSAndrew Morton struct hstate *hs; 1382af73e4d9SNaoya Horiguchi 138320ac2893SAnshuman Khandual hs = hstate_sizelog((flags >> MAP_HUGE_SHIFT) & MAP_HUGE_MASK); 1384091d0d55SLi Zefan if (!hs) 1385091d0d55SLi Zefan return -EINVAL; 1386091d0d55SLi Zefan 1387091d0d55SLi Zefan len = ALIGN(len, huge_page_size(hs)); 138866f0dc48SHugh Dickins /* 138966f0dc48SHugh Dickins * VM_NORESERVE is used because the reservations will be 139066f0dc48SHugh Dickins * taken when vm_ops->mmap() is called 139166f0dc48SHugh Dickins */ 1392af73e4d9SNaoya Horiguchi file = hugetlb_file_setup(HUGETLB_ANON_FILE, len, 139342d7395fSAndi Kleen VM_NORESERVE, 139483c1fd76Szhangyiru HUGETLB_ANONHUGE_INODE, 139542d7395fSAndi Kleen (flags >> MAP_HUGE_SHIFT) & MAP_HUGE_MASK); 139666f0dc48SHugh Dickins if (IS_ERR(file)) 139766f0dc48SHugh Dickins return PTR_ERR(file); 139866f0dc48SHugh Dickins } 139966f0dc48SHugh Dickins 14009fbeb5abSMichal Hocko retval = vm_mmap_pgoff(file, addr, len, prot, flags, pgoff); 1401493af578SJörn Engel out_fput: 140266f0dc48SHugh Dickins if (file) 140366f0dc48SHugh Dickins fput(file); 140466f0dc48SHugh Dickins return retval; 140566f0dc48SHugh Dickins } 140666f0dc48SHugh Dickins 1407a90f590aSDominik Brodowski SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len, 1408a90f590aSDominik Brodowski unsigned long, prot, unsigned long, flags, 1409a90f590aSDominik Brodowski unsigned long, fd, unsigned long, pgoff) 1410a90f590aSDominik Brodowski { 1411a90f590aSDominik Brodowski return ksys_mmap_pgoff(addr, len, prot, flags, fd, pgoff); 1412a90f590aSDominik Brodowski } 1413a90f590aSDominik Brodowski 1414a4679373SChristoph Hellwig #ifdef __ARCH_WANT_SYS_OLD_MMAP 1415a4679373SChristoph Hellwig struct mmap_arg_struct { 1416a4679373SChristoph Hellwig unsigned long addr; 1417a4679373SChristoph Hellwig unsigned long len; 1418a4679373SChristoph Hellwig unsigned long prot; 1419a4679373SChristoph Hellwig unsigned long flags; 1420a4679373SChristoph Hellwig unsigned long fd; 1421a4679373SChristoph Hellwig unsigned long offset; 1422a4679373SChristoph Hellwig }; 1423a4679373SChristoph Hellwig 1424a4679373SChristoph Hellwig SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg) 1425a4679373SChristoph Hellwig { 1426a4679373SChristoph Hellwig struct mmap_arg_struct a; 1427a4679373SChristoph Hellwig 1428a4679373SChristoph Hellwig if (copy_from_user(&a, arg, sizeof(a))) 1429a4679373SChristoph Hellwig return -EFAULT; 1430de1741a1SAlexander Kuleshov if (offset_in_page(a.offset)) 1431a4679373SChristoph Hellwig return -EINVAL; 1432a4679373SChristoph Hellwig 1433a90f590aSDominik Brodowski return ksys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd, 1434a4679373SChristoph Hellwig a.offset >> PAGE_SHIFT); 1435a4679373SChristoph Hellwig } 1436a4679373SChristoph Hellwig #endif /* __ARCH_WANT_SYS_OLD_MMAP */ 1437a4679373SChristoph Hellwig 143854cbbbf3SLorenzo Stoakes static bool vm_ops_needs_writenotify(const struct vm_operations_struct *vm_ops) 143954cbbbf3SLorenzo Stoakes { 144054cbbbf3SLorenzo Stoakes return vm_ops && (vm_ops->page_mkwrite || vm_ops->pfn_mkwrite); 144154cbbbf3SLorenzo Stoakes } 144254cbbbf3SLorenzo Stoakes 144354cbbbf3SLorenzo Stoakes static bool vma_is_shared_writable(struct vm_area_struct *vma) 144454cbbbf3SLorenzo Stoakes { 144554cbbbf3SLorenzo Stoakes return (vma->vm_flags & (VM_WRITE | VM_SHARED)) == 144654cbbbf3SLorenzo Stoakes (VM_WRITE | VM_SHARED); 144754cbbbf3SLorenzo Stoakes } 144854cbbbf3SLorenzo Stoakes 144954cbbbf3SLorenzo Stoakes static bool vma_fs_can_writeback(struct vm_area_struct *vma) 145054cbbbf3SLorenzo Stoakes { 145154cbbbf3SLorenzo Stoakes /* No managed pages to writeback. */ 145254cbbbf3SLorenzo Stoakes if (vma->vm_flags & VM_PFNMAP) 145354cbbbf3SLorenzo Stoakes return false; 145454cbbbf3SLorenzo Stoakes 145554cbbbf3SLorenzo Stoakes return vma->vm_file && vma->vm_file->f_mapping && 145654cbbbf3SLorenzo Stoakes mapping_can_writeback(vma->vm_file->f_mapping); 145754cbbbf3SLorenzo Stoakes } 145854cbbbf3SLorenzo Stoakes 145954cbbbf3SLorenzo Stoakes /* 146054cbbbf3SLorenzo Stoakes * Does this VMA require the underlying folios to have their dirty state 146154cbbbf3SLorenzo Stoakes * tracked? 146254cbbbf3SLorenzo Stoakes */ 146354cbbbf3SLorenzo Stoakes bool vma_needs_dirty_tracking(struct vm_area_struct *vma) 146454cbbbf3SLorenzo Stoakes { 146554cbbbf3SLorenzo Stoakes /* Only shared, writable VMAs require dirty tracking. */ 146654cbbbf3SLorenzo Stoakes if (!vma_is_shared_writable(vma)) 146754cbbbf3SLorenzo Stoakes return false; 146854cbbbf3SLorenzo Stoakes 146954cbbbf3SLorenzo Stoakes /* Does the filesystem need to be notified? */ 147054cbbbf3SLorenzo Stoakes if (vm_ops_needs_writenotify(vma->vm_ops)) 147154cbbbf3SLorenzo Stoakes return true; 147254cbbbf3SLorenzo Stoakes 147354cbbbf3SLorenzo Stoakes /* 147454cbbbf3SLorenzo Stoakes * Even if the filesystem doesn't indicate a need for writenotify, if it 147554cbbbf3SLorenzo Stoakes * can writeback, dirty tracking is still required. 147654cbbbf3SLorenzo Stoakes */ 147754cbbbf3SLorenzo Stoakes return vma_fs_can_writeback(vma); 147854cbbbf3SLorenzo Stoakes } 147954cbbbf3SLorenzo Stoakes 14804e950f6fSAlexey Dobriyan /* 14818bb4e7a2SWei Yang * Some shared mappings will want the pages marked read-only 14824e950f6fSAlexey Dobriyan * to track write events. If so, we'll downgrade vm_page_prot 14834e950f6fSAlexey Dobriyan * to the private version (using protection_map[] without the 14844e950f6fSAlexey Dobriyan * VM_SHARED bit). 14854e950f6fSAlexey Dobriyan */ 14866d2329f8SAndrea Arcangeli int vma_wants_writenotify(struct vm_area_struct *vma, pgprot_t vm_page_prot) 14874e950f6fSAlexey Dobriyan { 14884e950f6fSAlexey Dobriyan /* If it was private or non-writable, the write bit is already clear */ 148954cbbbf3SLorenzo Stoakes if (!vma_is_shared_writable(vma)) 14904e950f6fSAlexey Dobriyan return 0; 14914e950f6fSAlexey Dobriyan 14924e950f6fSAlexey Dobriyan /* The backer wishes to know when pages are first written to? */ 149354cbbbf3SLorenzo Stoakes if (vm_ops_needs_writenotify(vma->vm_ops)) 14944e950f6fSAlexey Dobriyan return 1; 14954e950f6fSAlexey Dobriyan 149664e45507SPeter Feiner /* The open routine did something to the protections that pgprot_modify 149764e45507SPeter Feiner * won't preserve? */ 14986d2329f8SAndrea Arcangeli if (pgprot_val(vm_page_prot) != 149954cbbbf3SLorenzo Stoakes pgprot_val(vm_pgprot_modify(vm_page_prot, vma->vm_flags))) 15004e950f6fSAlexey Dobriyan return 0; 15014e950f6fSAlexey Dobriyan 1502f96f7a40SDavid Hildenbrand /* 1503f96f7a40SDavid Hildenbrand * Do we need to track softdirty? hugetlb does not support softdirty 1504f96f7a40SDavid Hildenbrand * tracking yet. 1505f96f7a40SDavid Hildenbrand */ 1506f96f7a40SDavid Hildenbrand if (vma_soft_dirty_enabled(vma) && !is_vm_hugetlb_page(vma)) 150764e45507SPeter Feiner return 1; 150864e45507SPeter Feiner 150951d3d5ebSDavid Hildenbrand /* Do we need write faults for uffd-wp tracking? */ 151051d3d5ebSDavid Hildenbrand if (userfaultfd_wp(vma)) 151151d3d5ebSDavid Hildenbrand return 1; 151251d3d5ebSDavid Hildenbrand 15134e950f6fSAlexey Dobriyan /* Can the mapping track the dirty pages? */ 151454cbbbf3SLorenzo Stoakes return vma_fs_can_writeback(vma); 15154e950f6fSAlexey Dobriyan } 15164e950f6fSAlexey Dobriyan 1517fc8744adSLinus Torvalds /* 1518fc8744adSLinus Torvalds * We account for memory if it's a private writeable mapping, 15195a6fe125SMel Gorman * not hugepages and VM_NORESERVE wasn't set. 1520fc8744adSLinus Torvalds */ 1521ca16d140SKOSAKI Motohiro static inline int accountable_mapping(struct file *file, vm_flags_t vm_flags) 1522fc8744adSLinus Torvalds { 15235a6fe125SMel Gorman /* 15245a6fe125SMel Gorman * hugetlb has its own accounting separate from the core VM 15255a6fe125SMel Gorman * VM_HUGETLB may not be set yet so we cannot check for that flag. 15265a6fe125SMel Gorman */ 15275a6fe125SMel Gorman if (file && is_file_hugepages(file)) 15285a6fe125SMel Gorman return 0; 15295a6fe125SMel Gorman 1530fc8744adSLinus Torvalds return (vm_flags & (VM_NORESERVE | VM_SHARED | VM_WRITE)) == VM_WRITE; 1531fc8744adSLinus Torvalds } 1532fc8744adSLinus Torvalds 15333499a131SLiam R. Howlett /** 15343499a131SLiam R. Howlett * unmapped_area() - Find an area between the low_limit and the high_limit with 15353499a131SLiam R. Howlett * the correct alignment and offset, all from @info. Note: current->mm is used 15363499a131SLiam R. Howlett * for the search. 15373499a131SLiam R. Howlett * 153882b24936SVernon Yang * @info: The unmapped area information including the range [low_limit - 153982b24936SVernon Yang * high_limit), the alignment offset and mask. 15403499a131SLiam R. Howlett * 15413499a131SLiam R. Howlett * Return: A memory address or -ENOMEM. 15423499a131SLiam R. Howlett */ 1543baceaf1cSJaewon Kim static unsigned long unmapped_area(struct vm_unmapped_area_info *info) 1544db4fbfb9SMichel Lespinasse { 15456b008640SLinus Torvalds unsigned long length, gap; 15466b008640SLinus Torvalds unsigned long low_limit, high_limit; 154758c5d0d6SLiam R. Howlett struct vm_area_struct *tmp; 1548db4fbfb9SMichel Lespinasse 15493499a131SLiam R. Howlett MA_STATE(mas, ¤t->mm->mm_mt, 0, 0); 1550db4fbfb9SMichel Lespinasse 1551db4fbfb9SMichel Lespinasse /* Adjust search length to account for worst case alignment overhead */ 1552db4fbfb9SMichel Lespinasse length = info->length + info->align_mask; 1553db4fbfb9SMichel Lespinasse if (length < info->length) 1554db4fbfb9SMichel Lespinasse return -ENOMEM; 1555db4fbfb9SMichel Lespinasse 155658c5d0d6SLiam R. Howlett low_limit = info->low_limit; 15576b008640SLinus Torvalds if (low_limit < mmap_min_addr) 15586b008640SLinus Torvalds low_limit = mmap_min_addr; 15596b008640SLinus Torvalds high_limit = info->high_limit; 156058c5d0d6SLiam R. Howlett retry: 15616b008640SLinus Torvalds if (mas_empty_area(&mas, low_limit, high_limit - 1, length)) 15623499a131SLiam R. Howlett return -ENOMEM; 15633499a131SLiam R. Howlett 1564d4af56c5SLiam R. Howlett gap = mas.index; 1565d4af56c5SLiam R. Howlett gap += (info->align_offset - gap) & info->align_mask; 156658c5d0d6SLiam R. Howlett tmp = mas_next(&mas, ULONG_MAX); 15670266e7c5SRick Edgecombe if (tmp && (tmp->vm_flags & VM_STARTGAP_FLAGS)) { /* Avoid prev check if possible */ 156858c5d0d6SLiam R. Howlett if (vm_start_gap(tmp) < gap + length - 1) { 156958c5d0d6SLiam R. Howlett low_limit = tmp->vm_end; 157058c5d0d6SLiam R. Howlett mas_reset(&mas); 157158c5d0d6SLiam R. Howlett goto retry; 157258c5d0d6SLiam R. Howlett } 157358c5d0d6SLiam R. Howlett } else { 157458c5d0d6SLiam R. Howlett tmp = mas_prev(&mas, 0); 157558c5d0d6SLiam R. Howlett if (tmp && vm_end_gap(tmp) > gap) { 157658c5d0d6SLiam R. Howlett low_limit = vm_end_gap(tmp); 157758c5d0d6SLiam R. Howlett mas_reset(&mas); 157858c5d0d6SLiam R. Howlett goto retry; 157958c5d0d6SLiam R. Howlett } 158058c5d0d6SLiam R. Howlett } 158158c5d0d6SLiam R. Howlett 15823499a131SLiam R. Howlett return gap; 1583db4fbfb9SMichel Lespinasse } 1584db4fbfb9SMichel Lespinasse 15853499a131SLiam R. Howlett /** 15863499a131SLiam R. Howlett * unmapped_area_topdown() - Find an area between the low_limit and the 158782b24936SVernon Yang * high_limit with the correct alignment and offset at the highest available 15883499a131SLiam R. Howlett * address, all from @info. Note: current->mm is used for the search. 15893499a131SLiam R. Howlett * 159082b24936SVernon Yang * @info: The unmapped area information including the range [low_limit - 159182b24936SVernon Yang * high_limit), the alignment offset and mask. 15923499a131SLiam R. Howlett * 15933499a131SLiam R. Howlett * Return: A memory address or -ENOMEM. 15943499a131SLiam R. Howlett */ 1595baceaf1cSJaewon Kim static unsigned long unmapped_area_topdown(struct vm_unmapped_area_info *info) 1596db4fbfb9SMichel Lespinasse { 15976b008640SLinus Torvalds unsigned long length, gap, gap_end; 15986b008640SLinus Torvalds unsigned long low_limit, high_limit; 159958c5d0d6SLiam R. Howlett struct vm_area_struct *tmp; 1600d4af56c5SLiam R. Howlett 16013499a131SLiam R. Howlett MA_STATE(mas, ¤t->mm->mm_mt, 0, 0); 1602db4fbfb9SMichel Lespinasse /* Adjust search length to account for worst case alignment overhead */ 1603db4fbfb9SMichel Lespinasse length = info->length + info->align_mask; 1604db4fbfb9SMichel Lespinasse if (length < info->length) 1605db4fbfb9SMichel Lespinasse return -ENOMEM; 1606db4fbfb9SMichel Lespinasse 16076b008640SLinus Torvalds low_limit = info->low_limit; 16086b008640SLinus Torvalds if (low_limit < mmap_min_addr) 16096b008640SLinus Torvalds low_limit = mmap_min_addr; 161058c5d0d6SLiam R. Howlett high_limit = info->high_limit; 161158c5d0d6SLiam R. Howlett retry: 16126b008640SLinus Torvalds if (mas_empty_area_rev(&mas, low_limit, high_limit - 1, length)) 16133499a131SLiam R. Howlett return -ENOMEM; 16143499a131SLiam R. Howlett 1615d4af56c5SLiam R. Howlett gap = mas.last + 1 - info->length; 1616d4af56c5SLiam R. Howlett gap -= (gap - info->align_offset) & info->align_mask; 161758c5d0d6SLiam R. Howlett gap_end = mas.last; 161858c5d0d6SLiam R. Howlett tmp = mas_next(&mas, ULONG_MAX); 16190266e7c5SRick Edgecombe if (tmp && (tmp->vm_flags & VM_STARTGAP_FLAGS)) { /* Avoid prev check if possible */ 162058c5d0d6SLiam R. Howlett if (vm_start_gap(tmp) <= gap_end) { 162158c5d0d6SLiam R. Howlett high_limit = vm_start_gap(tmp); 162258c5d0d6SLiam R. Howlett mas_reset(&mas); 162358c5d0d6SLiam R. Howlett goto retry; 162458c5d0d6SLiam R. Howlett } 162558c5d0d6SLiam R. Howlett } else { 162658c5d0d6SLiam R. Howlett tmp = mas_prev(&mas, 0); 162758c5d0d6SLiam R. Howlett if (tmp && vm_end_gap(tmp) > gap) { 162858c5d0d6SLiam R. Howlett high_limit = tmp->vm_start; 162958c5d0d6SLiam R. Howlett mas_reset(&mas); 163058c5d0d6SLiam R. Howlett goto retry; 163158c5d0d6SLiam R. Howlett } 163258c5d0d6SLiam R. Howlett } 163358c5d0d6SLiam R. Howlett 16343499a131SLiam R. Howlett return gap; 1635db4fbfb9SMichel Lespinasse } 1636db4fbfb9SMichel Lespinasse 1637baceaf1cSJaewon Kim /* 1638baceaf1cSJaewon Kim * Search for an unmapped address range. 1639baceaf1cSJaewon Kim * 1640baceaf1cSJaewon Kim * We are looking for a range that: 1641baceaf1cSJaewon Kim * - does not intersect with any VMA; 1642baceaf1cSJaewon Kim * - is contained within the [low_limit, high_limit) interval; 1643baceaf1cSJaewon Kim * - is at least the desired size. 1644baceaf1cSJaewon Kim * - satisfies (begin_addr & align_mask) == (align_offset & align_mask) 1645baceaf1cSJaewon Kim */ 1646baceaf1cSJaewon Kim unsigned long vm_unmapped_area(struct vm_unmapped_area_info *info) 1647baceaf1cSJaewon Kim { 1648df529cabSJaewon Kim unsigned long addr; 1649df529cabSJaewon Kim 1650baceaf1cSJaewon Kim if (info->flags & VM_UNMAPPED_AREA_TOPDOWN) 1651df529cabSJaewon Kim addr = unmapped_area_topdown(info); 1652baceaf1cSJaewon Kim else 1653df529cabSJaewon Kim addr = unmapped_area(info); 1654df529cabSJaewon Kim 1655df529cabSJaewon Kim trace_vm_unmapped_area(addr, info); 1656df529cabSJaewon Kim return addr; 1657baceaf1cSJaewon Kim } 1658f6795053SSteve Capper 16591da177e4SLinus Torvalds /* Get an address range which is currently unmapped. 16601da177e4SLinus Torvalds * For shmat() with addr=0. 16611da177e4SLinus Torvalds * 16621da177e4SLinus Torvalds * Ugly calling convention alert: 16631da177e4SLinus Torvalds * Return value with the low bits set means error value, 16641da177e4SLinus Torvalds * ie 16651da177e4SLinus Torvalds * if (ret & ~PAGE_MASK) 16661da177e4SLinus Torvalds * error = ret; 16671da177e4SLinus Torvalds * 16681da177e4SLinus Torvalds * This function "knows" that -ENOMEM has the bits set. 16691da177e4SLinus Torvalds */ 16701da177e4SLinus Torvalds unsigned long 16714b439e25SChristophe Leroy generic_get_unmapped_area(struct file *filp, unsigned long addr, 16724b439e25SChristophe Leroy unsigned long len, unsigned long pgoff, 16734b439e25SChristophe Leroy unsigned long flags) 16741da177e4SLinus Torvalds { 16751da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 16761be7107fSHugh Dickins struct vm_area_struct *vma, *prev; 1677db4fbfb9SMichel Lespinasse struct vm_unmapped_area_info info; 16782cb4de08SChristophe Leroy const unsigned long mmap_end = arch_get_mmap_end(addr, len, flags); 16791da177e4SLinus Torvalds 1680f6795053SSteve Capper if (len > mmap_end - mmap_min_addr) 16811da177e4SLinus Torvalds return -ENOMEM; 16821da177e4SLinus Torvalds 168306abdfb4SBenjamin Herrenschmidt if (flags & MAP_FIXED) 168406abdfb4SBenjamin Herrenschmidt return addr; 168506abdfb4SBenjamin Herrenschmidt 16861da177e4SLinus Torvalds if (addr) { 16871da177e4SLinus Torvalds addr = PAGE_ALIGN(addr); 16881be7107fSHugh Dickins vma = find_vma_prev(mm, addr, &prev); 1689f6795053SSteve Capper if (mmap_end - len >= addr && addr >= mmap_min_addr && 16901be7107fSHugh Dickins (!vma || addr + len <= vm_start_gap(vma)) && 16911be7107fSHugh Dickins (!prev || addr >= vm_end_gap(prev))) 16921da177e4SLinus Torvalds return addr; 16931da177e4SLinus Torvalds } 16941da177e4SLinus Torvalds 1695db4fbfb9SMichel Lespinasse info.flags = 0; 1696db4fbfb9SMichel Lespinasse info.length = len; 16974e99b021SHeiko Carstens info.low_limit = mm->mmap_base; 1698f6795053SSteve Capper info.high_limit = mmap_end; 1699db4fbfb9SMichel Lespinasse info.align_mask = 0; 170009ef5283SJaewon Kim info.align_offset = 0; 1701db4fbfb9SMichel Lespinasse return vm_unmapped_area(&info); 17021da177e4SLinus Torvalds } 17034b439e25SChristophe Leroy 17044b439e25SChristophe Leroy #ifndef HAVE_ARCH_UNMAPPED_AREA 17054b439e25SChristophe Leroy unsigned long 17064b439e25SChristophe Leroy arch_get_unmapped_area(struct file *filp, unsigned long addr, 17074b439e25SChristophe Leroy unsigned long len, unsigned long pgoff, 17084b439e25SChristophe Leroy unsigned long flags) 17094b439e25SChristophe Leroy { 17104b439e25SChristophe Leroy return generic_get_unmapped_area(filp, addr, len, pgoff, flags); 17114b439e25SChristophe Leroy } 17121da177e4SLinus Torvalds #endif 17131da177e4SLinus Torvalds 17141da177e4SLinus Torvalds /* 17151da177e4SLinus Torvalds * This mmap-allocator allocates new areas top-down from below the 17161da177e4SLinus Torvalds * stack's low limit (the base): 17171da177e4SLinus Torvalds */ 17181da177e4SLinus Torvalds unsigned long 17194b439e25SChristophe Leroy generic_get_unmapped_area_topdown(struct file *filp, unsigned long addr, 172043cca0b1SYang Fan unsigned long len, unsigned long pgoff, 172143cca0b1SYang Fan unsigned long flags) 17221da177e4SLinus Torvalds { 17231be7107fSHugh Dickins struct vm_area_struct *vma, *prev; 17241da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 1725db4fbfb9SMichel Lespinasse struct vm_unmapped_area_info info; 17262cb4de08SChristophe Leroy const unsigned long mmap_end = arch_get_mmap_end(addr, len, flags); 17271da177e4SLinus Torvalds 17281da177e4SLinus Torvalds /* requested length too big for entire address space */ 1729f6795053SSteve Capper if (len > mmap_end - mmap_min_addr) 17301da177e4SLinus Torvalds return -ENOMEM; 17311da177e4SLinus Torvalds 173206abdfb4SBenjamin Herrenschmidt if (flags & MAP_FIXED) 173306abdfb4SBenjamin Herrenschmidt return addr; 173406abdfb4SBenjamin Herrenschmidt 17351da177e4SLinus Torvalds /* requesting a specific address */ 17361da177e4SLinus Torvalds if (addr) { 17371da177e4SLinus Torvalds addr = PAGE_ALIGN(addr); 17381be7107fSHugh Dickins vma = find_vma_prev(mm, addr, &prev); 1739f6795053SSteve Capper if (mmap_end - len >= addr && addr >= mmap_min_addr && 17401be7107fSHugh Dickins (!vma || addr + len <= vm_start_gap(vma)) && 17411be7107fSHugh Dickins (!prev || addr >= vm_end_gap(prev))) 17421da177e4SLinus Torvalds return addr; 17431da177e4SLinus Torvalds } 17441da177e4SLinus Torvalds 1745db4fbfb9SMichel Lespinasse info.flags = VM_UNMAPPED_AREA_TOPDOWN; 1746db4fbfb9SMichel Lespinasse info.length = len; 17476b008640SLinus Torvalds info.low_limit = PAGE_SIZE; 1748f6795053SSteve Capper info.high_limit = arch_get_mmap_base(addr, mm->mmap_base); 1749db4fbfb9SMichel Lespinasse info.align_mask = 0; 175009ef5283SJaewon Kim info.align_offset = 0; 1751db4fbfb9SMichel Lespinasse addr = vm_unmapped_area(&info); 1752b716ad95SXiao Guangrong 17531da177e4SLinus Torvalds /* 17541da177e4SLinus Torvalds * A failed mmap() very likely causes application failure, 17551da177e4SLinus Torvalds * so fall back to the bottom-up function here. This scenario 17561da177e4SLinus Torvalds * can happen with large stack limits and large mmap() 17571da177e4SLinus Torvalds * allocations. 17581da177e4SLinus Torvalds */ 1759de1741a1SAlexander Kuleshov if (offset_in_page(addr)) { 1760db4fbfb9SMichel Lespinasse VM_BUG_ON(addr != -ENOMEM); 1761db4fbfb9SMichel Lespinasse info.flags = 0; 1762db4fbfb9SMichel Lespinasse info.low_limit = TASK_UNMAPPED_BASE; 1763f6795053SSteve Capper info.high_limit = mmap_end; 1764db4fbfb9SMichel Lespinasse addr = vm_unmapped_area(&info); 1765db4fbfb9SMichel Lespinasse } 17661da177e4SLinus Torvalds 17671da177e4SLinus Torvalds return addr; 17681da177e4SLinus Torvalds } 17694b439e25SChristophe Leroy 17704b439e25SChristophe Leroy #ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN 17714b439e25SChristophe Leroy unsigned long 17724b439e25SChristophe Leroy arch_get_unmapped_area_topdown(struct file *filp, unsigned long addr, 17734b439e25SChristophe Leroy unsigned long len, unsigned long pgoff, 17744b439e25SChristophe Leroy unsigned long flags) 17754b439e25SChristophe Leroy { 17764b439e25SChristophe Leroy return generic_get_unmapped_area_topdown(filp, addr, len, pgoff, flags); 17774b439e25SChristophe Leroy } 17781da177e4SLinus Torvalds #endif 17791da177e4SLinus Torvalds 17801da177e4SLinus Torvalds unsigned long 17811da177e4SLinus Torvalds get_unmapped_area(struct file *file, unsigned long addr, unsigned long len, 17821da177e4SLinus Torvalds unsigned long pgoff, unsigned long flags) 17831da177e4SLinus Torvalds { 178406abdfb4SBenjamin Herrenschmidt unsigned long (*get_area)(struct file *, unsigned long, 178506abdfb4SBenjamin Herrenschmidt unsigned long, unsigned long, unsigned long); 178607ab67c8SLinus Torvalds 17879206de95SAl Viro unsigned long error = arch_mmap_check(addr, len, flags); 17889206de95SAl Viro if (error) 17899206de95SAl Viro return error; 17909206de95SAl Viro 17919206de95SAl Viro /* Careful about overflows.. */ 17929206de95SAl Viro if (len > TASK_SIZE) 17939206de95SAl Viro return -ENOMEM; 17949206de95SAl Viro 179507ab67c8SLinus Torvalds get_area = current->mm->get_unmapped_area; 1796c01d5b30SHugh Dickins if (file) { 1797c01d5b30SHugh Dickins if (file->f_op->get_unmapped_area) 179807ab67c8SLinus Torvalds get_area = file->f_op->get_unmapped_area; 1799c01d5b30SHugh Dickins } else if (flags & MAP_SHARED) { 1800c01d5b30SHugh Dickins /* 1801c01d5b30SHugh Dickins * mmap_region() will call shmem_zero_setup() to create a file, 1802c01d5b30SHugh Dickins * so use shmem's get_unmapped_area in case it can be huge. 180345e55300SPeter Collingbourne * do_mmap() will clear pgoff, so match alignment. 1804c01d5b30SHugh Dickins */ 1805c01d5b30SHugh Dickins pgoff = 0; 1806c01d5b30SHugh Dickins get_area = shmem_get_unmapped_area; 1807c01d5b30SHugh Dickins } 1808c01d5b30SHugh Dickins 180907ab67c8SLinus Torvalds addr = get_area(file, addr, len, pgoff, flags); 181007ab67c8SLinus Torvalds if (IS_ERR_VALUE(addr)) 181107ab67c8SLinus Torvalds return addr; 181207ab67c8SLinus Torvalds 18131da177e4SLinus Torvalds if (addr > TASK_SIZE - len) 18141da177e4SLinus Torvalds return -ENOMEM; 1815de1741a1SAlexander Kuleshov if (offset_in_page(addr)) 18161da177e4SLinus Torvalds return -EINVAL; 181706abdfb4SBenjamin Herrenschmidt 18189ac4ed4bSAl Viro error = security_mmap_addr(addr); 18199ac4ed4bSAl Viro return error ? error : addr; 18201da177e4SLinus Torvalds } 18211da177e4SLinus Torvalds 18221da177e4SLinus Torvalds EXPORT_SYMBOL(get_unmapped_area); 18231da177e4SLinus Torvalds 1824be8432e7SLiam R. Howlett /** 1825abdba2ddSLiam R. Howlett * find_vma_intersection() - Look up the first VMA which intersects the interval 1826abdba2ddSLiam R. Howlett * @mm: The process address space. 1827abdba2ddSLiam R. Howlett * @start_addr: The inclusive start user address. 1828abdba2ddSLiam R. Howlett * @end_addr: The exclusive end user address. 1829abdba2ddSLiam R. Howlett * 1830abdba2ddSLiam R. Howlett * Returns: The first VMA within the provided range, %NULL otherwise. Assumes 1831abdba2ddSLiam R. Howlett * start_addr < end_addr. 1832abdba2ddSLiam R. Howlett */ 1833abdba2ddSLiam R. Howlett struct vm_area_struct *find_vma_intersection(struct mm_struct *mm, 1834abdba2ddSLiam R. Howlett unsigned long start_addr, 1835abdba2ddSLiam R. Howlett unsigned long end_addr) 1836abdba2ddSLiam R. Howlett { 1837abdba2ddSLiam R. Howlett unsigned long index = start_addr; 1838abdba2ddSLiam R. Howlett 1839abdba2ddSLiam R. Howlett mmap_assert_locked(mm); 18407964cf8cSLiam R. Howlett return mt_find(&mm->mm_mt, &index, end_addr - 1); 1841abdba2ddSLiam R. Howlett } 1842abdba2ddSLiam R. Howlett EXPORT_SYMBOL(find_vma_intersection); 1843abdba2ddSLiam R. Howlett 1844abdba2ddSLiam R. Howlett /** 1845be8432e7SLiam R. Howlett * find_vma() - Find the VMA for a given address, or the next VMA. 1846be8432e7SLiam R. Howlett * @mm: The mm_struct to check 1847be8432e7SLiam R. Howlett * @addr: The address 1848be8432e7SLiam R. Howlett * 1849be8432e7SLiam R. Howlett * Returns: The VMA associated with addr, or the next VMA. 1850be8432e7SLiam R. Howlett * May return %NULL in the case of no VMA at addr or above. 1851be8432e7SLiam R. Howlett */ 18521da177e4SLinus Torvalds struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr) 18531da177e4SLinus Torvalds { 1854be8432e7SLiam R. Howlett unsigned long index = addr; 18551da177e4SLinus Torvalds 18565b78ed24SLuigi Rizzo mmap_assert_locked(mm); 18577964cf8cSLiam R. Howlett return mt_find(&mm->mm_mt, &index, ULONG_MAX); 18581da177e4SLinus Torvalds } 18591da177e4SLinus Torvalds EXPORT_SYMBOL(find_vma); 18601da177e4SLinus Torvalds 18617fdbd37dSLiam R. Howlett /** 18627fdbd37dSLiam R. Howlett * find_vma_prev() - Find the VMA for a given address, or the next vma and 18637fdbd37dSLiam R. Howlett * set %pprev to the previous VMA, if any. 18647fdbd37dSLiam R. Howlett * @mm: The mm_struct to check 18657fdbd37dSLiam R. Howlett * @addr: The address 18667fdbd37dSLiam R. Howlett * @pprev: The pointer to set to the previous VMA 18677fdbd37dSLiam R. Howlett * 18687fdbd37dSLiam R. Howlett * Note that RCU lock is missing here since the external mmap_lock() is used 18697fdbd37dSLiam R. Howlett * instead. 18707fdbd37dSLiam R. Howlett * 18717fdbd37dSLiam R. Howlett * Returns: The VMA associated with @addr, or the next vma. 18727fdbd37dSLiam R. Howlett * May return %NULL in the case of no vma at addr or above. 18736bd4837dSKOSAKI Motohiro */ 18741da177e4SLinus Torvalds struct vm_area_struct * 18751da177e4SLinus Torvalds find_vma_prev(struct mm_struct *mm, unsigned long addr, 18761da177e4SLinus Torvalds struct vm_area_struct **pprev) 18771da177e4SLinus Torvalds { 18786bd4837dSKOSAKI Motohiro struct vm_area_struct *vma; 18797fdbd37dSLiam R. Howlett MA_STATE(mas, &mm->mm_mt, addr, addr); 18801da177e4SLinus Torvalds 18817fdbd37dSLiam R. Howlett vma = mas_walk(&mas); 18827fdbd37dSLiam R. Howlett *pprev = mas_prev(&mas, 0); 18837fdbd37dSLiam R. Howlett if (!vma) 18847fdbd37dSLiam R. Howlett vma = mas_next(&mas, ULONG_MAX); 18856bd4837dSKOSAKI Motohiro return vma; 18861da177e4SLinus Torvalds } 18871da177e4SLinus Torvalds 18881da177e4SLinus Torvalds /* 18891da177e4SLinus Torvalds * Verify that the stack growth is acceptable and 18901da177e4SLinus Torvalds * update accounting. This is shared with both the 18911da177e4SLinus Torvalds * grow-up and grow-down cases. 18921da177e4SLinus Torvalds */ 18931be7107fSHugh Dickins static int acct_stack_growth(struct vm_area_struct *vma, 18941be7107fSHugh Dickins unsigned long size, unsigned long grow) 18951da177e4SLinus Torvalds { 18961da177e4SLinus Torvalds struct mm_struct *mm = vma->vm_mm; 18971be7107fSHugh Dickins unsigned long new_start; 18981da177e4SLinus Torvalds 18991da177e4SLinus Torvalds /* address space limit tests */ 190084638335SKonstantin Khlebnikov if (!may_expand_vm(mm, vma->vm_flags, grow)) 19011da177e4SLinus Torvalds return -ENOMEM; 19021da177e4SLinus Torvalds 19031da177e4SLinus Torvalds /* Stack limit test */ 190424c79d8eSKrzysztof Opasiak if (size > rlimit(RLIMIT_STACK)) 19051da177e4SLinus Torvalds return -ENOMEM; 19061da177e4SLinus Torvalds 19071da177e4SLinus Torvalds /* mlock limit tests */ 1908b0cc5e89SAndrew Morton if (!mlock_future_ok(mm, vma->vm_flags, grow << PAGE_SHIFT)) 19091da177e4SLinus Torvalds return -ENOMEM; 19101da177e4SLinus Torvalds 19110d59a01bSAdam Litke /* Check to ensure the stack will not grow into a hugetlb-only region */ 19120d59a01bSAdam Litke new_start = (vma->vm_flags & VM_GROWSUP) ? vma->vm_start : 19130d59a01bSAdam Litke vma->vm_end - size; 19140d59a01bSAdam Litke if (is_hugepage_only_range(vma->vm_mm, new_start, size)) 19150d59a01bSAdam Litke return -EFAULT; 19160d59a01bSAdam Litke 19171da177e4SLinus Torvalds /* 19181da177e4SLinus Torvalds * Overcommit.. This must be the final test, as it will 19191da177e4SLinus Torvalds * update security statistics. 19201da177e4SLinus Torvalds */ 192105fa199dSHugh Dickins if (security_vm_enough_memory_mm(mm, grow)) 19221da177e4SLinus Torvalds return -ENOMEM; 19231da177e4SLinus Torvalds 19241da177e4SLinus Torvalds return 0; 19251da177e4SLinus Torvalds } 19261da177e4SLinus Torvalds 192746dea3d0SHugh Dickins #if defined(CONFIG_STACK_GROWSUP) || defined(CONFIG_IA64) 19281da177e4SLinus Torvalds /* 192946dea3d0SHugh Dickins * PA-RISC uses this for its stack; IA64 for its Register Backing Store. 193046dea3d0SHugh Dickins * vma is the last one with address > vma->vm_end. Have to extend vma. 19311da177e4SLinus Torvalds */ 19328d7071afSLinus Torvalds static int expand_upwards(struct vm_area_struct *vma, unsigned long address) 19331da177e4SLinus Torvalds { 193409357814SOleg Nesterov struct mm_struct *mm = vma->vm_mm; 19351be7107fSHugh Dickins struct vm_area_struct *next; 19361be7107fSHugh Dickins unsigned long gap_addr; 193712352d3cSKonstantin Khlebnikov int error = 0; 1938b5df0922SLiam R. Howlett MA_STATE(mas, &mm->mm_mt, vma->vm_start, address); 19391da177e4SLinus Torvalds 19401da177e4SLinus Torvalds if (!(vma->vm_flags & VM_GROWSUP)) 19411da177e4SLinus Torvalds return -EFAULT; 19421da177e4SLinus Torvalds 1943bd726c90SHelge Deller /* Guard against exceeding limits of the address space. */ 19441be7107fSHugh Dickins address &= PAGE_MASK; 194537511fb5SHelge Deller if (address >= (TASK_SIZE & PAGE_MASK)) 194612352d3cSKonstantin Khlebnikov return -ENOMEM; 1947bd726c90SHelge Deller address += PAGE_SIZE; 194812352d3cSKonstantin Khlebnikov 19491be7107fSHugh Dickins /* Enforce stack_guard_gap */ 19501be7107fSHugh Dickins gap_addr = address + stack_guard_gap; 1951bd726c90SHelge Deller 1952bd726c90SHelge Deller /* Guard against overflow */ 1953bd726c90SHelge Deller if (gap_addr < address || gap_addr > TASK_SIZE) 1954bd726c90SHelge Deller gap_addr = TASK_SIZE; 1955bd726c90SHelge Deller 1956763ecb03SLiam R. Howlett next = find_vma_intersection(mm, vma->vm_end, gap_addr); 1957763ecb03SLiam R. Howlett if (next && vma_is_accessible(next)) { 19581be7107fSHugh Dickins if (!(next->vm_flags & VM_GROWSUP)) 19591be7107fSHugh Dickins return -ENOMEM; 19601be7107fSHugh Dickins /* Check that both stack segments have the same anon_vma? */ 19611be7107fSHugh Dickins } 19621be7107fSHugh Dickins 1963b5df0922SLiam R. Howlett if (next) 1964b5df0922SLiam R. Howlett mas_prev_range(&mas, address); 1965b5df0922SLiam R. Howlett 1966b5df0922SLiam R. Howlett __mas_set_range(&mas, vma->vm_start, address - 1); 1967da089254SLiam R. Howlett if (mas_preallocate(&mas, vma, GFP_KERNEL)) 19681da177e4SLinus Torvalds return -ENOMEM; 19691da177e4SLinus Torvalds 1970d4af56c5SLiam R. Howlett /* We must make sure the anon_vma is allocated. */ 1971d4af56c5SLiam R. Howlett if (unlikely(anon_vma_prepare(vma))) { 1972d4af56c5SLiam R. Howlett mas_destroy(&mas); 1973d4af56c5SLiam R. Howlett return -ENOMEM; 1974d4af56c5SLiam R. Howlett } 1975d4af56c5SLiam R. Howlett 1976c137381fSSuren Baghdasaryan /* Lock the VMA before expanding to prevent concurrent page faults */ 1977c137381fSSuren Baghdasaryan vma_start_write(vma); 19781da177e4SLinus Torvalds /* 19791da177e4SLinus Torvalds * vma->vm_start/vm_end cannot change under us because the caller 1980c1e8d7c6SMichel Lespinasse * is required to hold the mmap_lock in read mode. We need the 19811da177e4SLinus Torvalds * anon_vma lock to serialize against concurrent expand_stacks. 19821da177e4SLinus Torvalds */ 198312352d3cSKonstantin Khlebnikov anon_vma_lock_write(vma->anon_vma); 19841da177e4SLinus Torvalds 19851da177e4SLinus Torvalds /* Somebody else might have raced and expanded it already */ 19861da177e4SLinus Torvalds if (address > vma->vm_end) { 19871da177e4SLinus Torvalds unsigned long size, grow; 19881da177e4SLinus Torvalds 19891da177e4SLinus Torvalds size = address - vma->vm_start; 19901da177e4SLinus Torvalds grow = (address - vma->vm_end) >> PAGE_SHIFT; 19911da177e4SLinus Torvalds 199242c36f63SHugh Dickins error = -ENOMEM; 199342c36f63SHugh Dickins if (vma->vm_pgoff + (size >> PAGE_SHIFT) >= vma->vm_pgoff) { 19941da177e4SLinus Torvalds error = acct_stack_growth(vma, size, grow); 19953af9e859SEric B Munson if (!error) { 19964128997bSMichel Lespinasse /* 1997524e00b3SLiam R. Howlett * We only hold a shared mmap_lock lock here, so 1998524e00b3SLiam R. Howlett * we need to protect against concurrent vma 1999524e00b3SLiam R. Howlett * expansions. anon_vma_lock_write() doesn't 2000524e00b3SLiam R. Howlett * help here, as we don't guarantee that all 2001524e00b3SLiam R. Howlett * growable vmas in a mm share the same root 2002524e00b3SLiam R. Howlett * anon vma. So, we reuse mm->page_table_lock 2003524e00b3SLiam R. Howlett * to guard against concurrent vma expansions. 20044128997bSMichel Lespinasse */ 200509357814SOleg Nesterov spin_lock(&mm->page_table_lock); 200687e8827bSOleg Nesterov if (vma->vm_flags & VM_LOCKED) 200709357814SOleg Nesterov mm->locked_vm += grow; 200884638335SKonstantin Khlebnikov vm_stat_account(mm, vma->vm_flags, grow); 2009bf181b9fSMichel Lespinasse anon_vma_interval_tree_pre_update_vma(vma); 20101da177e4SLinus Torvalds vma->vm_end = address; 2011d4af56c5SLiam R. Howlett /* Overwrite old entry in mtree. */ 2012fbcc3104SLiam R. Howlett mas_store_prealloc(&mas, vma); 2013bf181b9fSMichel Lespinasse anon_vma_interval_tree_post_update_vma(vma); 201409357814SOleg Nesterov spin_unlock(&mm->page_table_lock); 20154128997bSMichel Lespinasse 20163af9e859SEric B Munson perf_event_mmap(vma); 20173af9e859SEric B Munson } 20181da177e4SLinus Torvalds } 201942c36f63SHugh Dickins } 202012352d3cSKonstantin Khlebnikov anon_vma_unlock_write(vma->anon_vma); 2021c791576cSYang Shi khugepaged_enter_vma(vma, vma->vm_flags); 2022d4af56c5SLiam R. Howlett mas_destroy(&mas); 20232574d5e4SLiam R. Howlett validate_mm(mm); 20241da177e4SLinus Torvalds return error; 20251da177e4SLinus Torvalds } 202646dea3d0SHugh Dickins #endif /* CONFIG_STACK_GROWSUP || CONFIG_IA64 */ 202746dea3d0SHugh Dickins 20281da177e4SLinus Torvalds /* 20291da177e4SLinus Torvalds * vma is the first one with address < vma->vm_start. Have to extend vma. 20308d7071afSLinus Torvalds * mmap_lock held for writing. 20311da177e4SLinus Torvalds */ 2032524e00b3SLiam R. Howlett int expand_downwards(struct vm_area_struct *vma, unsigned long address) 20331da177e4SLinus Torvalds { 203409357814SOleg Nesterov struct mm_struct *mm = vma->vm_mm; 2035763ecb03SLiam R. Howlett MA_STATE(mas, &mm->mm_mt, vma->vm_start, vma->vm_start); 20361be7107fSHugh Dickins struct vm_area_struct *prev; 20370a1d5299SJann Horn int error = 0; 20381da177e4SLinus Torvalds 20398d7071afSLinus Torvalds if (!(vma->vm_flags & VM_GROWSDOWN)) 20408d7071afSLinus Torvalds return -EFAULT; 20418d7071afSLinus Torvalds 20428869477aSEric Paris address &= PAGE_MASK; 20438b35ca3eSBen Hutchings if (address < mmap_min_addr || address < FIRST_USER_ADDRESS) 20440a1d5299SJann Horn return -EPERM; 20458869477aSEric Paris 20461be7107fSHugh Dickins /* Enforce stack_guard_gap */ 2047763ecb03SLiam R. Howlett prev = mas_prev(&mas, 0); 20481be7107fSHugh Dickins /* Check that both stack segments have the same anon_vma? */ 2049f440fa1aSLiam R. Howlett if (prev) { 2050f440fa1aSLiam R. Howlett if (!(prev->vm_flags & VM_GROWSDOWN) && 2051f440fa1aSLiam R. Howlett vma_is_accessible(prev) && 2052f440fa1aSLiam R. Howlett (address - prev->vm_end < stack_guard_gap)) 205332e4e6d5SOleg Nesterov return -ENOMEM; 20541be7107fSHugh Dickins } 20551be7107fSHugh Dickins 2056b5df0922SLiam R. Howlett if (prev) 2057b5df0922SLiam R. Howlett mas_next_range(&mas, vma->vm_start); 2058b5df0922SLiam R. Howlett 2059b5df0922SLiam R. Howlett __mas_set_range(&mas, address, vma->vm_end - 1); 2060da089254SLiam R. Howlett if (mas_preallocate(&mas, vma, GFP_KERNEL)) 206112352d3cSKonstantin Khlebnikov return -ENOMEM; 20621da177e4SLinus Torvalds 2063d4af56c5SLiam R. Howlett /* We must make sure the anon_vma is allocated. */ 2064d4af56c5SLiam R. Howlett if (unlikely(anon_vma_prepare(vma))) { 2065d4af56c5SLiam R. Howlett mas_destroy(&mas); 2066d4af56c5SLiam R. Howlett return -ENOMEM; 2067d4af56c5SLiam R. Howlett } 2068d4af56c5SLiam R. Howlett 2069c137381fSSuren Baghdasaryan /* Lock the VMA before expanding to prevent concurrent page faults */ 2070c137381fSSuren Baghdasaryan vma_start_write(vma); 20711da177e4SLinus Torvalds /* 20721da177e4SLinus Torvalds * vma->vm_start/vm_end cannot change under us because the caller 2073c1e8d7c6SMichel Lespinasse * is required to hold the mmap_lock in read mode. We need the 20741da177e4SLinus Torvalds * anon_vma lock to serialize against concurrent expand_stacks. 20751da177e4SLinus Torvalds */ 207612352d3cSKonstantin Khlebnikov anon_vma_lock_write(vma->anon_vma); 20771da177e4SLinus Torvalds 20781da177e4SLinus Torvalds /* Somebody else might have raced and expanded it already */ 20791da177e4SLinus Torvalds if (address < vma->vm_start) { 20801da177e4SLinus Torvalds unsigned long size, grow; 20811da177e4SLinus Torvalds 20821da177e4SLinus Torvalds size = vma->vm_end - address; 20831da177e4SLinus Torvalds grow = (vma->vm_start - address) >> PAGE_SHIFT; 20841da177e4SLinus Torvalds 2085a626ca6aSLinus Torvalds error = -ENOMEM; 2086a626ca6aSLinus Torvalds if (grow <= vma->vm_pgoff) { 20871da177e4SLinus Torvalds error = acct_stack_growth(vma, size, grow); 20881da177e4SLinus Torvalds if (!error) { 20894128997bSMichel Lespinasse /* 2090524e00b3SLiam R. Howlett * We only hold a shared mmap_lock lock here, so 2091524e00b3SLiam R. Howlett * we need to protect against concurrent vma 2092524e00b3SLiam R. Howlett * expansions. anon_vma_lock_write() doesn't 2093524e00b3SLiam R. Howlett * help here, as we don't guarantee that all 2094524e00b3SLiam R. Howlett * growable vmas in a mm share the same root 2095524e00b3SLiam R. Howlett * anon vma. So, we reuse mm->page_table_lock 2096524e00b3SLiam R. Howlett * to guard against concurrent vma expansions. 20974128997bSMichel Lespinasse */ 209809357814SOleg Nesterov spin_lock(&mm->page_table_lock); 209987e8827bSOleg Nesterov if (vma->vm_flags & VM_LOCKED) 210009357814SOleg Nesterov mm->locked_vm += grow; 210184638335SKonstantin Khlebnikov vm_stat_account(mm, vma->vm_flags, grow); 2102bf181b9fSMichel Lespinasse anon_vma_interval_tree_pre_update_vma(vma); 21031da177e4SLinus Torvalds vma->vm_start = address; 21041da177e4SLinus Torvalds vma->vm_pgoff -= grow; 2105d4af56c5SLiam R. Howlett /* Overwrite old entry in mtree. */ 2106fbcc3104SLiam R. Howlett mas_store_prealloc(&mas, vma); 2107bf181b9fSMichel Lespinasse anon_vma_interval_tree_post_update_vma(vma); 210809357814SOleg Nesterov spin_unlock(&mm->page_table_lock); 21094128997bSMichel Lespinasse 21103af9e859SEric B Munson perf_event_mmap(vma); 21111da177e4SLinus Torvalds } 21121da177e4SLinus Torvalds } 2113a626ca6aSLinus Torvalds } 211412352d3cSKonstantin Khlebnikov anon_vma_unlock_write(vma->anon_vma); 2115c791576cSYang Shi khugepaged_enter_vma(vma, vma->vm_flags); 2116d4af56c5SLiam R. Howlett mas_destroy(&mas); 21172574d5e4SLiam R. Howlett validate_mm(mm); 21181da177e4SLinus Torvalds return error; 21191da177e4SLinus Torvalds } 21201da177e4SLinus Torvalds 21211be7107fSHugh Dickins /* enforced gap between the expanding stack and other mappings. */ 21221be7107fSHugh Dickins unsigned long stack_guard_gap = 256UL<<PAGE_SHIFT; 21231be7107fSHugh Dickins 21241be7107fSHugh Dickins static int __init cmdline_parse_stack_guard_gap(char *p) 21251be7107fSHugh Dickins { 21261be7107fSHugh Dickins unsigned long val; 21271be7107fSHugh Dickins char *endptr; 21281be7107fSHugh Dickins 21291be7107fSHugh Dickins val = simple_strtoul(p, &endptr, 10); 21301be7107fSHugh Dickins if (!*endptr) 21311be7107fSHugh Dickins stack_guard_gap = val << PAGE_SHIFT; 21321be7107fSHugh Dickins 2133e6d09493SRandy Dunlap return 1; 21341be7107fSHugh Dickins } 21351be7107fSHugh Dickins __setup("stack_guard_gap=", cmdline_parse_stack_guard_gap); 21361be7107fSHugh Dickins 2137b6a2fea3SOllie Wild #ifdef CONFIG_STACK_GROWSUP 21388d7071afSLinus Torvalds int expand_stack_locked(struct vm_area_struct *vma, unsigned long address) 2139b6a2fea3SOllie Wild { 2140b6a2fea3SOllie Wild return expand_upwards(vma, address); 2141b6a2fea3SOllie Wild } 2142b6a2fea3SOllie Wild 21438d7071afSLinus Torvalds struct vm_area_struct *find_extend_vma_locked(struct mm_struct *mm, unsigned long addr) 2144b6a2fea3SOllie Wild { 2145b6a2fea3SOllie Wild struct vm_area_struct *vma, *prev; 2146b6a2fea3SOllie Wild 2147b6a2fea3SOllie Wild addr &= PAGE_MASK; 2148b6a2fea3SOllie Wild vma = find_vma_prev(mm, addr, &prev); 2149b6a2fea3SOllie Wild if (vma && (vma->vm_start <= addr)) 2150b6a2fea3SOllie Wild return vma; 2151f440fa1aSLiam R. Howlett if (!prev) 2152f440fa1aSLiam R. Howlett return NULL; 21538d7071afSLinus Torvalds if (expand_stack_locked(prev, addr)) 2154b6a2fea3SOllie Wild return NULL; 2155cea10a19SMichel Lespinasse if (prev->vm_flags & VM_LOCKED) 2156fc05f566SKirill A. Shutemov populate_vma_page_range(prev, addr, prev->vm_end, NULL); 2157b6a2fea3SOllie Wild return prev; 2158b6a2fea3SOllie Wild } 2159b6a2fea3SOllie Wild #else 21608d7071afSLinus Torvalds int expand_stack_locked(struct vm_area_struct *vma, unsigned long address) 2161b6a2fea3SOllie Wild { 2162f440fa1aSLiam R. Howlett if (unlikely(!(vma->vm_flags & VM_GROWSDOWN))) 2163f440fa1aSLiam R. Howlett return -EINVAL; 2164b6a2fea3SOllie Wild return expand_downwards(vma, address); 2165b6a2fea3SOllie Wild } 2166b6a2fea3SOllie Wild 21678d7071afSLinus Torvalds struct vm_area_struct *find_extend_vma_locked(struct mm_struct *mm, unsigned long addr) 21681da177e4SLinus Torvalds { 21691da177e4SLinus Torvalds struct vm_area_struct *vma; 21701da177e4SLinus Torvalds unsigned long start; 21711da177e4SLinus Torvalds 21721da177e4SLinus Torvalds addr &= PAGE_MASK; 21731da177e4SLinus Torvalds vma = find_vma(mm, addr); 21741da177e4SLinus Torvalds if (!vma) 21751da177e4SLinus Torvalds return NULL; 21761da177e4SLinus Torvalds if (vma->vm_start <= addr) 21771da177e4SLinus Torvalds return vma; 21781da177e4SLinus Torvalds start = vma->vm_start; 21798d7071afSLinus Torvalds if (expand_stack_locked(vma, addr)) 21801da177e4SLinus Torvalds return NULL; 2181cea10a19SMichel Lespinasse if (vma->vm_flags & VM_LOCKED) 2182fc05f566SKirill A. Shutemov populate_vma_page_range(vma, addr, start, NULL); 21831da177e4SLinus Torvalds return vma; 21841da177e4SLinus Torvalds } 21851da177e4SLinus Torvalds #endif 21861da177e4SLinus Torvalds 21878d7071afSLinus Torvalds /* 21888d7071afSLinus Torvalds * IA64 has some horrid mapping rules: it can expand both up and down, 21898d7071afSLinus Torvalds * but with various special rules. 21908d7071afSLinus Torvalds * 21918d7071afSLinus Torvalds * We'll get rid of this architecture eventually, so the ugliness is 21928d7071afSLinus Torvalds * temporary. 21938d7071afSLinus Torvalds */ 21948d7071afSLinus Torvalds #ifdef CONFIG_IA64 21958d7071afSLinus Torvalds static inline bool vma_expand_ok(struct vm_area_struct *vma, unsigned long addr) 2196f440fa1aSLiam R. Howlett { 21978d7071afSLinus Torvalds return REGION_NUMBER(addr) == REGION_NUMBER(vma->vm_start) && 21988d7071afSLinus Torvalds REGION_OFFSET(addr) < RGN_MAP_LIMIT; 2199f440fa1aSLiam R. Howlett } 22008d7071afSLinus Torvalds 22018d7071afSLinus Torvalds /* 22028d7071afSLinus Torvalds * IA64 stacks grow down, but there's a special register backing store 22038d7071afSLinus Torvalds * that can grow up. Only sequentially, though, so the new address must 22048d7071afSLinus Torvalds * match vm_end. 22058d7071afSLinus Torvalds */ 22068d7071afSLinus Torvalds static inline int vma_expand_up(struct vm_area_struct *vma, unsigned long addr) 22078d7071afSLinus Torvalds { 22088d7071afSLinus Torvalds if (!vma_expand_ok(vma, addr)) 22098d7071afSLinus Torvalds return -EFAULT; 22108d7071afSLinus Torvalds if (vma->vm_end != (addr & PAGE_MASK)) 22118d7071afSLinus Torvalds return -EFAULT; 22128d7071afSLinus Torvalds return expand_upwards(vma, addr); 22138d7071afSLinus Torvalds } 22148d7071afSLinus Torvalds 22158d7071afSLinus Torvalds static inline bool vma_expand_down(struct vm_area_struct *vma, unsigned long addr) 22168d7071afSLinus Torvalds { 22178d7071afSLinus Torvalds if (!vma_expand_ok(vma, addr)) 22188d7071afSLinus Torvalds return -EFAULT; 22198d7071afSLinus Torvalds return expand_downwards(vma, addr); 22208d7071afSLinus Torvalds } 22218d7071afSLinus Torvalds 22228d7071afSLinus Torvalds #elif defined(CONFIG_STACK_GROWSUP) 22238d7071afSLinus Torvalds 22248d7071afSLinus Torvalds #define vma_expand_up(vma,addr) expand_upwards(vma, addr) 22258d7071afSLinus Torvalds #define vma_expand_down(vma, addr) (-EFAULT) 22268d7071afSLinus Torvalds 22278d7071afSLinus Torvalds #else 22288d7071afSLinus Torvalds 22298d7071afSLinus Torvalds #define vma_expand_up(vma,addr) (-EFAULT) 22308d7071afSLinus Torvalds #define vma_expand_down(vma, addr) expand_downwards(vma, addr) 22318d7071afSLinus Torvalds 22328d7071afSLinus Torvalds #endif 22338d7071afSLinus Torvalds 22348d7071afSLinus Torvalds /* 22358d7071afSLinus Torvalds * expand_stack(): legacy interface for page faulting. Don't use unless 22368d7071afSLinus Torvalds * you have to. 22378d7071afSLinus Torvalds * 22388d7071afSLinus Torvalds * This is called with the mm locked for reading, drops the lock, takes 22398d7071afSLinus Torvalds * the lock for writing, tries to look up a vma again, expands it if 22408d7071afSLinus Torvalds * necessary, and downgrades the lock to reading again. 22418d7071afSLinus Torvalds * 22428d7071afSLinus Torvalds * If no vma is found or it can't be expanded, it returns NULL and has 22438d7071afSLinus Torvalds * dropped the lock. 22448d7071afSLinus Torvalds */ 22458d7071afSLinus Torvalds struct vm_area_struct *expand_stack(struct mm_struct *mm, unsigned long addr) 22468d7071afSLinus Torvalds { 22478d7071afSLinus Torvalds struct vm_area_struct *vma, *prev; 22488d7071afSLinus Torvalds 22498d7071afSLinus Torvalds mmap_read_unlock(mm); 22508d7071afSLinus Torvalds if (mmap_write_lock_killable(mm)) 22518d7071afSLinus Torvalds return NULL; 22528d7071afSLinus Torvalds 22538d7071afSLinus Torvalds vma = find_vma_prev(mm, addr, &prev); 22548d7071afSLinus Torvalds if (vma && vma->vm_start <= addr) 22558d7071afSLinus Torvalds goto success; 22568d7071afSLinus Torvalds 22578d7071afSLinus Torvalds if (prev && !vma_expand_up(prev, addr)) { 22588d7071afSLinus Torvalds vma = prev; 22598d7071afSLinus Torvalds goto success; 22608d7071afSLinus Torvalds } 22618d7071afSLinus Torvalds 22628d7071afSLinus Torvalds if (vma && !vma_expand_down(vma, addr)) 22638d7071afSLinus Torvalds goto success; 22648d7071afSLinus Torvalds 22658d7071afSLinus Torvalds mmap_write_unlock(mm); 22668d7071afSLinus Torvalds return NULL; 22678d7071afSLinus Torvalds 22688d7071afSLinus Torvalds success: 22698d7071afSLinus Torvalds mmap_write_downgrade(mm); 22708d7071afSLinus Torvalds return vma; 22718d7071afSLinus Torvalds } 2272e1d6d01aSJesse Barnes 22732c0b3814SHugh Dickins /* 2274763ecb03SLiam R. Howlett * Ok - we have the memory areas we should free on a maple tree so release them, 2275763ecb03SLiam R. Howlett * and do the vma updates. 22761da177e4SLinus Torvalds * 22772c0b3814SHugh Dickins * Called with the mm semaphore held. 22781da177e4SLinus Torvalds */ 2279763ecb03SLiam R. Howlett static inline void remove_mt(struct mm_struct *mm, struct ma_state *mas) 22801da177e4SLinus Torvalds { 22814f74d2c8SLinus Torvalds unsigned long nr_accounted = 0; 2282763ecb03SLiam R. Howlett struct vm_area_struct *vma; 22834f74d2c8SLinus Torvalds 2284365e9c87SHugh Dickins /* Update high watermark before we lower total_vm */ 2285365e9c87SHugh Dickins update_hiwater_vm(mm); 2286763ecb03SLiam R. Howlett mas_for_each(mas, vma, ULONG_MAX) { 2287ab50b8edSHugh Dickins long nrpages = vma_pages(vma); 22881da177e4SLinus Torvalds 22894f74d2c8SLinus Torvalds if (vma->vm_flags & VM_ACCOUNT) 22904f74d2c8SLinus Torvalds nr_accounted += nrpages; 229184638335SKonstantin Khlebnikov vm_stat_account(mm, vma->vm_flags, -nrpages); 22920d2ebf9cSSuren Baghdasaryan remove_vma(vma, false); 2293763ecb03SLiam R. Howlett } 22944f74d2c8SLinus Torvalds vm_unacct_memory(nr_accounted); 22951da177e4SLinus Torvalds } 22961da177e4SLinus Torvalds 22971da177e4SLinus Torvalds /* 22981da177e4SLinus Torvalds * Get rid of page table information in the indicated region. 22991da177e4SLinus Torvalds * 2300f10df686SPaolo 'Blaisorblade' Giarrusso * Called with the mm semaphore held. 23011da177e4SLinus Torvalds */ 2302fd892593SLiam R. Howlett static void unmap_region(struct mm_struct *mm, struct ma_state *mas, 2303e0da382cSHugh Dickins struct vm_area_struct *vma, struct vm_area_struct *prev, 2304fd892593SLiam R. Howlett struct vm_area_struct *next, unsigned long start, 2305fd892593SLiam R. Howlett unsigned long end, unsigned long tree_end, bool mm_wr_locked) 23061da177e4SLinus Torvalds { 2307d16dfc55SPeter Zijlstra struct mmu_gather tlb; 2308fd892593SLiam R. Howlett unsigned long mt_start = mas->index; 23091da177e4SLinus Torvalds 23101da177e4SLinus Torvalds lru_add_drain(); 2311a72afd87SWill Deacon tlb_gather_mmu(&tlb, mm); 2312365e9c87SHugh Dickins update_hiwater_rss(mm); 2313fd892593SLiam R. Howlett unmap_vmas(&tlb, mas, vma, start, end, tree_end, mm_wr_locked); 2314fd892593SLiam R. Howlett mas_set(mas, mt_start); 2315fd892593SLiam R. Howlett free_pgtables(&tlb, mas, vma, prev ? prev->vm_end : FIRST_USER_ADDRESS, 231698e51a22SSuren Baghdasaryan next ? next->vm_start : USER_PGTABLES_CEILING, 231798e51a22SSuren Baghdasaryan mm_wr_locked); 2318ae8eba8bSWill Deacon tlb_finish_mmu(&tlb); 23191da177e4SLinus Torvalds } 23201da177e4SLinus Torvalds 23211da177e4SLinus Torvalds /* 2322def5efe0SDavid Rientjes * __split_vma() bypasses sysctl_max_map_count checking. We use this where it 2323def5efe0SDavid Rientjes * has already been checked or doesn't make sense to fail. 23240fd5a9e2SLiam R. Howlett * VMA Iterator will point to the end VMA. 23251da177e4SLinus Torvalds */ 23269760ebffSLiam R. Howlett int __split_vma(struct vma_iterator *vmi, struct vm_area_struct *vma, 23271da177e4SLinus Torvalds unsigned long addr, int new_below) 23281da177e4SLinus Torvalds { 2329b2b3b886SLiam R. Howlett struct vma_prepare vp; 23301da177e4SLinus Torvalds struct vm_area_struct *new; 2331e3975891SChen Gang int err; 23329760ebffSLiam R. Howlett 2333b2b3b886SLiam R. Howlett WARN_ON(vma->vm_start >= addr); 2334b2b3b886SLiam R. Howlett WARN_ON(vma->vm_end <= addr); 2335b2b3b886SLiam R. Howlett 2336dd3b614fSDmitry Safonov if (vma->vm_ops && vma->vm_ops->may_split) { 2337dd3b614fSDmitry Safonov err = vma->vm_ops->may_split(vma, addr); 233831383c68SDan Williams if (err) 233931383c68SDan Williams return err; 234031383c68SDan Williams } 23411da177e4SLinus Torvalds 23423928d4f5SLinus Torvalds new = vm_area_dup(vma); 23431da177e4SLinus Torvalds if (!new) 2344e3975891SChen Gang return -ENOMEM; 23451da177e4SLinus Torvalds 2346b2b3b886SLiam R. Howlett if (new_below) { 23471da177e4SLinus Torvalds new->vm_end = addr; 2348b2b3b886SLiam R. Howlett } else { 23491da177e4SLinus Torvalds new->vm_start = addr; 23501da177e4SLinus Torvalds new->vm_pgoff += ((addr - vma->vm_start) >> PAGE_SHIFT); 23511da177e4SLinus Torvalds } 23521da177e4SLinus Torvalds 2353b5df0922SLiam R. Howlett err = -ENOMEM; 2354b5df0922SLiam R. Howlett vma_iter_config(vmi, new->vm_start, new->vm_end); 2355b5df0922SLiam R. Howlett if (vma_iter_prealloc(vmi, new)) 2356b5df0922SLiam R. Howlett goto out_free_vma; 2357b5df0922SLiam R. Howlett 2358ef0855d3SOleg Nesterov err = vma_dup_policy(vma, new); 2359ef0855d3SOleg Nesterov if (err) 2360b2b3b886SLiam R. Howlett goto out_free_vmi; 23611da177e4SLinus Torvalds 2362c4ea95d7SDaniel Forrest err = anon_vma_clone(new, vma); 2363c4ea95d7SDaniel Forrest if (err) 23645beb4930SRik van Riel goto out_free_mpol; 23655beb4930SRik van Riel 2366e9714acfSKonstantin Khlebnikov if (new->vm_file) 23671da177e4SLinus Torvalds get_file(new->vm_file); 23681da177e4SLinus Torvalds 23691da177e4SLinus Torvalds if (new->vm_ops && new->vm_ops->open) 23701da177e4SLinus Torvalds new->vm_ops->open(new); 23711da177e4SLinus Torvalds 2372c9d6e982SSuren Baghdasaryan vma_start_write(vma); 2373c9d6e982SSuren Baghdasaryan vma_start_write(new); 2374c9d6e982SSuren Baghdasaryan 2375b2b3b886SLiam R. Howlett init_vma_prep(&vp, vma); 2376b2b3b886SLiam R. Howlett vp.insert = new; 2377b2b3b886SLiam R. Howlett vma_prepare(&vp); 2378ccf1d78dSSuren Baghdasaryan vma_adjust_trans_huge(vma, vma->vm_start, addr, 0); 23791da177e4SLinus Torvalds 2380b2b3b886SLiam R. Howlett if (new_below) { 2381b2b3b886SLiam R. Howlett vma->vm_start = addr; 2382b2b3b886SLiam R. Howlett vma->vm_pgoff += (addr - new->vm_start) >> PAGE_SHIFT; 2383b2b3b886SLiam R. Howlett } else { 2384b2b3b886SLiam R. Howlett vma->vm_end = addr; 23859760ebffSLiam R. Howlett } 23865beb4930SRik van Riel 2387b2b3b886SLiam R. Howlett /* vma_complete stores the new vma */ 2388b2b3b886SLiam R. Howlett vma_complete(&vp, vmi, vma->vm_mm); 2389b2b3b886SLiam R. Howlett 2390b2b3b886SLiam R. Howlett /* Success. */ 2391b2b3b886SLiam R. Howlett if (new_below) 2392b2b3b886SLiam R. Howlett vma_next(vmi); 2393b2b3b886SLiam R. Howlett return 0; 2394b2b3b886SLiam R. Howlett 23955beb4930SRik van Riel out_free_mpol: 2396ef0855d3SOleg Nesterov mpol_put(vma_policy(new)); 2397b2b3b886SLiam R. Howlett out_free_vmi: 2398b2b3b886SLiam R. Howlett vma_iter_free(vmi); 23995beb4930SRik van Riel out_free_vma: 24003928d4f5SLinus Torvalds vm_area_free(new); 24015beb4930SRik van Riel return err; 24021da177e4SLinus Torvalds } 24031da177e4SLinus Torvalds 2404659ace58SKOSAKI Motohiro /* 2405659ace58SKOSAKI Motohiro * Split a vma into two pieces at address 'addr', a new vma is allocated 2406659ace58SKOSAKI Motohiro * either for the first part or the tail. 2407659ace58SKOSAKI Motohiro */ 24089760ebffSLiam R. Howlett int split_vma(struct vma_iterator *vmi, struct vm_area_struct *vma, 2409659ace58SKOSAKI Motohiro unsigned long addr, int new_below) 2410659ace58SKOSAKI Motohiro { 24119760ebffSLiam R. Howlett if (vma->vm_mm->map_count >= sysctl_max_map_count) 2412659ace58SKOSAKI Motohiro return -ENOMEM; 2413659ace58SKOSAKI Motohiro 24149760ebffSLiam R. Howlett return __split_vma(vmi, vma, addr, new_below); 2415f2ebfe43SLiam R. Howlett } 2416f2ebfe43SLiam R. Howlett 241711f9a21aSLiam R. Howlett /* 2418183654ceSLiam R. Howlett * do_vmi_align_munmap() - munmap the aligned region from @start to @end. 2419183654ceSLiam R. Howlett * @vmi: The vma iterator 242011f9a21aSLiam R. Howlett * @vma: The starting vm_area_struct 242111f9a21aSLiam R. Howlett * @mm: The mm_struct 242211f9a21aSLiam R. Howlett * @start: The aligned start address to munmap. 242311f9a21aSLiam R. Howlett * @end: The aligned end address to munmap. 242411f9a21aSLiam R. Howlett * @uf: The userfaultfd list_head 2425408579cdSLiam R. Howlett * @unlock: Set to true to drop the mmap_lock. unlocking only happens on 2426408579cdSLiam R. Howlett * success. 242711f9a21aSLiam R. Howlett * 2428408579cdSLiam R. Howlett * Return: 0 on success and drops the lock if so directed, error and leaves the 2429408579cdSLiam R. Howlett * lock held otherwise. 243011f9a21aSLiam R. Howlett */ 243111f9a21aSLiam R. Howlett static int 2432183654ceSLiam R. Howlett do_vmi_align_munmap(struct vma_iterator *vmi, struct vm_area_struct *vma, 243311f9a21aSLiam R. Howlett struct mm_struct *mm, unsigned long start, 2434408579cdSLiam R. Howlett unsigned long end, struct list_head *uf, bool unlock) 243511f9a21aSLiam R. Howlett { 2436763ecb03SLiam R. Howlett struct vm_area_struct *prev, *next = NULL; 2437763ecb03SLiam R. Howlett struct maple_tree mt_detach; 2438763ecb03SLiam R. Howlett int count = 0; 2439d4af56c5SLiam R. Howlett int error = -ENOMEM; 2440606c812eSLiam R. Howlett unsigned long locked_vm = 0; 2441763ecb03SLiam R. Howlett MA_STATE(mas_detach, &mt_detach, 0, 0); 24423dd44325SLiam R. Howlett mt_init_flags(&mt_detach, vmi->mas.tree->ma_flags & MT_FLAGS_LOCK_MASK); 244302fdb25fSLiam R. Howlett mt_on_stack(mt_detach); 2444524e00b3SLiam R. Howlett 24451da177e4SLinus Torvalds /* 24461da177e4SLinus Torvalds * If we need to split any vma, do it now to save pain later. 24471da177e4SLinus Torvalds * 24481da177e4SLinus Torvalds * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially 24491da177e4SLinus Torvalds * unmapped vm_area_struct will remain in use: so lower split_vma 24501da177e4SLinus Torvalds * places tmp vma above, and higher split_vma places tmp vma below. 24511da177e4SLinus Torvalds */ 2452763ecb03SLiam R. Howlett 2453763ecb03SLiam R. Howlett /* Does it split the first one? */ 2454146425a3SHugh Dickins if (start > vma->vm_start) { 2455659ace58SKOSAKI Motohiro 2456659ace58SKOSAKI Motohiro /* 2457659ace58SKOSAKI Motohiro * Make sure that map_count on return from munmap() will 2458659ace58SKOSAKI Motohiro * not exceed its limit; but let map_count go just above 2459659ace58SKOSAKI Motohiro * its limit temporarily, to help free resources as expected. 2460659ace58SKOSAKI Motohiro */ 2461659ace58SKOSAKI Motohiro if (end < vma->vm_end && mm->map_count >= sysctl_max_map_count) 2462d4af56c5SLiam R. Howlett goto map_count_exceeded; 2463659ace58SKOSAKI Motohiro 24646935e052SLiam R. Howlett error = __split_vma(vmi, vma, start, 1); 24651da177e4SLinus Torvalds if (error) 2466763ecb03SLiam R. Howlett goto start_split_failed; 24671da177e4SLinus Torvalds } 24681da177e4SLinus Torvalds 2469763ecb03SLiam R. Howlett /* 2470763ecb03SLiam R. Howlett * Detach a range of VMAs from the mm. Using next as a temp variable as 2471763ecb03SLiam R. Howlett * it is always overwritten. 2472763ecb03SLiam R. Howlett */ 24736935e052SLiam R. Howlett next = vma; 24746935e052SLiam R. Howlett do { 2475763ecb03SLiam R. Howlett /* Does it split the end? */ 2476763ecb03SLiam R. Howlett if (next->vm_end > end) { 24776b73cff2SLiam R. Howlett error = __split_vma(vmi, next, end, 0); 24781da177e4SLinus Torvalds if (error) 2479763ecb03SLiam R. Howlett goto end_split_failed; 248011f9a21aSLiam R. Howlett } 2481606c812eSLiam R. Howlett vma_start_write(next); 2482fd892593SLiam R. Howlett mas_set(&mas_detach, count); 24836c26bd43SDavid Woodhouse error = mas_store_gfp(&mas_detach, next, GFP_KERNEL); 24846c26bd43SDavid Woodhouse if (error) 2485606c812eSLiam R. Howlett goto munmap_gather_failed; 2486606c812eSLiam R. Howlett vma_mark_detached(next, true); 2487606c812eSLiam R. Howlett if (next->vm_flags & VM_LOCKED) 2488606c812eSLiam R. Howlett locked_vm += vma_pages(next); 2489763ecb03SLiam R. Howlett 2490763ecb03SLiam R. Howlett count++; 24912376dd7cSAndrea Arcangeli if (unlikely(uf)) { 24922376dd7cSAndrea Arcangeli /* 24932376dd7cSAndrea Arcangeli * If userfaultfd_unmap_prep returns an error the vmas 2494f0953a1bSIngo Molnar * will remain split, but userland will get a 24952376dd7cSAndrea Arcangeli * highly unexpected error anyway. This is no 24962376dd7cSAndrea Arcangeli * different than the case where the first of the two 24972376dd7cSAndrea Arcangeli * __split_vma fails, but we don't undo the first 24982376dd7cSAndrea Arcangeli * split, despite we could. This is unlikely enough 24992376dd7cSAndrea Arcangeli * failure that it's not worth optimizing it for. 25002376dd7cSAndrea Arcangeli */ 250165ac1320SLiam R. Howlett error = userfaultfd_unmap_prep(next, start, end, uf); 250211f9a21aSLiam R. Howlett 25032376dd7cSAndrea Arcangeli if (error) 2504d4af56c5SLiam R. Howlett goto userfaultfd_error; 25052376dd7cSAndrea Arcangeli } 2506ba470de4SRik van Riel #ifdef CONFIG_DEBUG_VM_MAPLE_TREE 2507ba470de4SRik van Riel BUG_ON(next->vm_start < start); 2508ba470de4SRik van Riel BUG_ON(next->vm_start > end); 25091da177e4SLinus Torvalds #endif 25106935e052SLiam R. Howlett } for_each_vma_range(*vmi, next, end); 25112376dd7cSAndrea Arcangeli 2512763ecb03SLiam R. Howlett #if defined(CONFIG_DEBUG_VM_MAPLE_TREE) 2513763ecb03SLiam R. Howlett /* Make sure no VMAs are about to be lost. */ 2514763ecb03SLiam R. Howlett { 2515fd892593SLiam R. Howlett MA_STATE(test, &mt_detach, 0, 0); 2516763ecb03SLiam R. Howlett struct vm_area_struct *vma_mas, *vma_test; 2517763ecb03SLiam R. Howlett int test_count = 0; 2518763ecb03SLiam R. Howlett 2519183654ceSLiam R. Howlett vma_iter_set(vmi, start); 2520763ecb03SLiam R. Howlett rcu_read_lock(); 2521fd892593SLiam R. Howlett vma_test = mas_find(&test, count - 1); 2522183654ceSLiam R. Howlett for_each_vma_range(*vmi, vma_mas, end) { 2523763ecb03SLiam R. Howlett BUG_ON(vma_mas != vma_test); 2524763ecb03SLiam R. Howlett test_count++; 2525fd892593SLiam R. Howlett vma_test = mas_next(&test, count - 1); 2526763ecb03SLiam R. Howlett } 2527763ecb03SLiam R. Howlett rcu_read_unlock(); 2528763ecb03SLiam R. Howlett BUG_ON(count != test_count); 2529763ecb03SLiam R. Howlett } 2530763ecb03SLiam R. Howlett #endif 25316935e052SLiam R. Howlett 25326935e052SLiam R. Howlett while (vma_iter_addr(vmi) > start) 25336935e052SLiam R. Howlett vma_iter_prev_range(vmi); 25346935e052SLiam R. Howlett 25356c26bd43SDavid Woodhouse error = vma_iter_clear_gfp(vmi, start, end, GFP_KERNEL); 25366c26bd43SDavid Woodhouse if (error) 2537606c812eSLiam R. Howlett goto clear_tree_failed; 25380378c0a0SLiam R. Howlett 25396c26bd43SDavid Woodhouse /* Point of no return */ 2540606c812eSLiam R. Howlett mm->locked_vm -= locked_vm; 2541763ecb03SLiam R. Howlett mm->map_count -= count; 2542408579cdSLiam R. Howlett if (unlock) 2543d8ed45c5SMichel Lespinasse mmap_write_downgrade(mm); 2544dd2283f2SYang Shi 25456935e052SLiam R. Howlett prev = vma_iter_prev_range(vmi); 25466935e052SLiam R. Howlett next = vma_next(vmi); 25476935e052SLiam R. Howlett if (next) 25486935e052SLiam R. Howlett vma_iter_prev_range(vmi); 25496935e052SLiam R. Howlett 255068f48381SSuren Baghdasaryan /* 255168f48381SSuren Baghdasaryan * We can free page tables without write-locking mmap_lock because VMAs 255268f48381SSuren Baghdasaryan * were isolated before we downgraded mmap_lock. 255368f48381SSuren Baghdasaryan */ 2554fd892593SLiam R. Howlett mas_set(&mas_detach, 1); 2555fd892593SLiam R. Howlett unmap_region(mm, &mas_detach, vma, prev, next, start, end, count, 2556fd892593SLiam R. Howlett !unlock); 2557763ecb03SLiam R. Howlett /* Statistics and freeing VMAs */ 2558fd892593SLiam R. Howlett mas_set(&mas_detach, 0); 2559763ecb03SLiam R. Howlett remove_mt(mm, &mas_detach); 2560ae80b404SLinus Torvalds validate_mm(mm); 2561408579cdSLiam R. Howlett if (unlock) 2562408579cdSLiam R. Howlett mmap_read_unlock(mm); 25631da177e4SLinus Torvalds 256402fdb25fSLiam R. Howlett __mt_destroy(&mt_detach); 2565408579cdSLiam R. Howlett return 0; 2566d4af56c5SLiam R. Howlett 2567606c812eSLiam R. Howlett clear_tree_failed: 2568d4af56c5SLiam R. Howlett userfaultfd_error: 2569606c812eSLiam R. Howlett munmap_gather_failed: 2570763ecb03SLiam R. Howlett end_split_failed: 2571606c812eSLiam R. Howlett mas_set(&mas_detach, 0); 2572606c812eSLiam R. Howlett mas_for_each(&mas_detach, next, end) 2573606c812eSLiam R. Howlett vma_mark_detached(next, false); 2574606c812eSLiam R. Howlett 2575763ecb03SLiam R. Howlett __mt_destroy(&mt_detach); 2576763ecb03SLiam R. Howlett start_split_failed: 2577763ecb03SLiam R. Howlett map_count_exceeded: 2578b5641a5dSLinus Torvalds validate_mm(mm); 2579d4af56c5SLiam R. Howlett return error; 25801da177e4SLinus Torvalds } 25811da177e4SLinus Torvalds 258211f9a21aSLiam R. Howlett /* 2583183654ceSLiam R. Howlett * do_vmi_munmap() - munmap a given range. 2584183654ceSLiam R. Howlett * @vmi: The vma iterator 258511f9a21aSLiam R. Howlett * @mm: The mm_struct 258611f9a21aSLiam R. Howlett * @start: The start address to munmap 258711f9a21aSLiam R. Howlett * @len: The length of the range to munmap 258811f9a21aSLiam R. Howlett * @uf: The userfaultfd list_head 2589408579cdSLiam R. Howlett * @unlock: set to true if the user wants to drop the mmap_lock on success 259011f9a21aSLiam R. Howlett * 259111f9a21aSLiam R. Howlett * This function takes a @mas that is either pointing to the previous VMA or set 259211f9a21aSLiam R. Howlett * to MA_START and sets it up to remove the mapping(s). The @len will be 259311f9a21aSLiam R. Howlett * aligned and any arch_unmap work will be preformed. 259411f9a21aSLiam R. Howlett * 2595408579cdSLiam R. Howlett * Return: 0 on success and drops the lock if so directed, error and leaves the 2596408579cdSLiam R. Howlett * lock held otherwise. 259711f9a21aSLiam R. Howlett */ 2598183654ceSLiam R. Howlett int do_vmi_munmap(struct vma_iterator *vmi, struct mm_struct *mm, 259911f9a21aSLiam R. Howlett unsigned long start, size_t len, struct list_head *uf, 2600408579cdSLiam R. Howlett bool unlock) 260111f9a21aSLiam R. Howlett { 260211f9a21aSLiam R. Howlett unsigned long end; 260311f9a21aSLiam R. Howlett struct vm_area_struct *vma; 260411f9a21aSLiam R. Howlett 260511f9a21aSLiam R. Howlett if ((offset_in_page(start)) || start > TASK_SIZE || len > TASK_SIZE-start) 260611f9a21aSLiam R. Howlett return -EINVAL; 260711f9a21aSLiam R. Howlett 260811f9a21aSLiam R. Howlett end = start + PAGE_ALIGN(len); 260911f9a21aSLiam R. Howlett if (end == start) 261011f9a21aSLiam R. Howlett return -EINVAL; 261111f9a21aSLiam R. Howlett 261211f9a21aSLiam R. Howlett /* arch_unmap() might do unmaps itself. */ 261311f9a21aSLiam R. Howlett arch_unmap(mm, start, end); 261411f9a21aSLiam R. Howlett 261511f9a21aSLiam R. Howlett /* Find the first overlapping VMA */ 2616183654ceSLiam R. Howlett vma = vma_find(vmi, end); 2617408579cdSLiam R. Howlett if (!vma) { 2618408579cdSLiam R. Howlett if (unlock) 2619408579cdSLiam R. Howlett mmap_write_unlock(mm); 262011f9a21aSLiam R. Howlett return 0; 2621408579cdSLiam R. Howlett } 262211f9a21aSLiam R. Howlett 2623408579cdSLiam R. Howlett return do_vmi_align_munmap(vmi, vma, mm, start, end, uf, unlock); 262411f9a21aSLiam R. Howlett } 262511f9a21aSLiam R. Howlett 262611f9a21aSLiam R. Howlett /* do_munmap() - Wrapper function for non-maple tree aware do_munmap() calls. 262711f9a21aSLiam R. Howlett * @mm: The mm_struct 262811f9a21aSLiam R. Howlett * @start: The start address to munmap 262911f9a21aSLiam R. Howlett * @len: The length to be munmapped. 263011f9a21aSLiam R. Howlett * @uf: The userfaultfd list_head 2631408579cdSLiam R. Howlett * 2632408579cdSLiam R. Howlett * Return: 0 on success, error otherwise. 263311f9a21aSLiam R. Howlett */ 2634dd2283f2SYang Shi int do_munmap(struct mm_struct *mm, unsigned long start, size_t len, 2635dd2283f2SYang Shi struct list_head *uf) 2636dd2283f2SYang Shi { 2637183654ceSLiam R. Howlett VMA_ITERATOR(vmi, mm, start); 263811f9a21aSLiam R. Howlett 2639183654ceSLiam R. Howlett return do_vmi_munmap(&vmi, mm, start, len, uf, false); 2640dd2283f2SYang Shi } 2641dd2283f2SYang Shi 2642e99668a5SLiam R. Howlett unsigned long mmap_region(struct file *file, unsigned long addr, 2643e99668a5SLiam R. Howlett unsigned long len, vm_flags_t vm_flags, unsigned long pgoff, 2644e99668a5SLiam R. Howlett struct list_head *uf) 2645e99668a5SLiam R. Howlett { 2646e99668a5SLiam R. Howlett struct mm_struct *mm = current->mm; 2647e99668a5SLiam R. Howlett struct vm_area_struct *vma = NULL; 2648e99668a5SLiam R. Howlett struct vm_area_struct *next, *prev, *merge; 2649e99668a5SLiam R. Howlett pgoff_t pglen = len >> PAGE_SHIFT; 2650e99668a5SLiam R. Howlett unsigned long charged = 0; 2651e99668a5SLiam R. Howlett unsigned long end = addr + len; 2652e99668a5SLiam R. Howlett unsigned long merge_start = addr, merge_end = end; 2653e99668a5SLiam R. Howlett pgoff_t vm_pgoff; 2654e99668a5SLiam R. Howlett int error; 2655183654ceSLiam R. Howlett VMA_ITERATOR(vmi, mm, addr); 2656e99668a5SLiam R. Howlett 2657e99668a5SLiam R. Howlett /* Check against address space limit. */ 2658e99668a5SLiam R. Howlett if (!may_expand_vm(mm, vm_flags, len >> PAGE_SHIFT)) { 2659e99668a5SLiam R. Howlett unsigned long nr_pages; 2660e99668a5SLiam R. Howlett 2661e99668a5SLiam R. Howlett /* 2662e99668a5SLiam R. Howlett * MAP_FIXED may remove pages of mappings that intersects with 2663e99668a5SLiam R. Howlett * requested mapping. Account for the pages it would unmap. 2664e99668a5SLiam R. Howlett */ 2665e99668a5SLiam R. Howlett nr_pages = count_vma_pages_range(mm, addr, end); 2666e99668a5SLiam R. Howlett 2667e99668a5SLiam R. Howlett if (!may_expand_vm(mm, vm_flags, 2668e99668a5SLiam R. Howlett (len >> PAGE_SHIFT) - nr_pages)) 2669e99668a5SLiam R. Howlett return -ENOMEM; 2670e99668a5SLiam R. Howlett } 2671e99668a5SLiam R. Howlett 2672e99668a5SLiam R. Howlett /* Unmap any existing mapping in the area */ 2673183654ceSLiam R. Howlett if (do_vmi_munmap(&vmi, mm, addr, len, uf, false)) 2674e99668a5SLiam R. Howlett return -ENOMEM; 2675e99668a5SLiam R. Howlett 2676e99668a5SLiam R. Howlett /* 2677e99668a5SLiam R. Howlett * Private writable mapping: check memory availability 2678e99668a5SLiam R. Howlett */ 2679e99668a5SLiam R. Howlett if (accountable_mapping(file, vm_flags)) { 2680e99668a5SLiam R. Howlett charged = len >> PAGE_SHIFT; 2681e99668a5SLiam R. Howlett if (security_vm_enough_memory_mm(mm, charged)) 2682e99668a5SLiam R. Howlett return -ENOMEM; 2683e99668a5SLiam R. Howlett vm_flags |= VM_ACCOUNT; 2684e99668a5SLiam R. Howlett } 2685e99668a5SLiam R. Howlett 2686183654ceSLiam R. Howlett next = vma_next(&vmi); 2687183654ceSLiam R. Howlett prev = vma_prev(&vmi); 268853bee98dSLiam R. Howlett if (vm_flags & VM_SPECIAL) { 268953bee98dSLiam R. Howlett if (prev) 269053bee98dSLiam R. Howlett vma_iter_next_range(&vmi); 2691e99668a5SLiam R. Howlett goto cannot_expand; 269253bee98dSLiam R. Howlett } 2693e99668a5SLiam R. Howlett 2694e99668a5SLiam R. Howlett /* Attempt to expand an old mapping */ 2695e99668a5SLiam R. Howlett /* Check next */ 2696e99668a5SLiam R. Howlett if (next && next->vm_start == end && !vma_policy(next) && 2697e99668a5SLiam R. Howlett can_vma_merge_before(next, vm_flags, NULL, file, pgoff+pglen, 2698e99668a5SLiam R. Howlett NULL_VM_UFFD_CTX, NULL)) { 2699e99668a5SLiam R. Howlett merge_end = next->vm_end; 2700e99668a5SLiam R. Howlett vma = next; 2701e99668a5SLiam R. Howlett vm_pgoff = next->vm_pgoff - pglen; 2702e99668a5SLiam R. Howlett } 2703e99668a5SLiam R. Howlett 2704e99668a5SLiam R. Howlett /* Check prev */ 2705e99668a5SLiam R. Howlett if (prev && prev->vm_end == addr && !vma_policy(prev) && 2706e99668a5SLiam R. Howlett (vma ? can_vma_merge_after(prev, vm_flags, vma->anon_vma, file, 2707e99668a5SLiam R. Howlett pgoff, vma->vm_userfaultfd_ctx, NULL) : 2708e99668a5SLiam R. Howlett can_vma_merge_after(prev, vm_flags, NULL, file, pgoff, 2709e99668a5SLiam R. Howlett NULL_VM_UFFD_CTX, NULL))) { 2710e99668a5SLiam R. Howlett merge_start = prev->vm_start; 2711e99668a5SLiam R. Howlett vma = prev; 2712e99668a5SLiam R. Howlett vm_pgoff = prev->vm_pgoff; 271353bee98dSLiam R. Howlett } else if (prev) { 271453bee98dSLiam R. Howlett vma_iter_next_range(&vmi); 2715e99668a5SLiam R. Howlett } 2716e99668a5SLiam R. Howlett 2717e99668a5SLiam R. Howlett /* Actually expand, if possible */ 2718e99668a5SLiam R. Howlett if (vma && 27193c441ab7SLiam R. Howlett !vma_expand(&vmi, vma, merge_start, merge_end, vm_pgoff, next)) { 2720e99668a5SLiam R. Howlett khugepaged_enter_vma(vma, vm_flags); 2721e99668a5SLiam R. Howlett goto expanded; 2722e99668a5SLiam R. Howlett } 2723e99668a5SLiam R. Howlett 272453bee98dSLiam R. Howlett if (vma == prev) 272553bee98dSLiam R. Howlett vma_iter_set(&vmi, addr); 2726e99668a5SLiam R. Howlett cannot_expand: 27275c1c03deSLiam R. Howlett 2728e99668a5SLiam R. Howlett /* 2729e99668a5SLiam R. Howlett * Determine the object being mapped and call the appropriate 2730e99668a5SLiam R. Howlett * specific mapper. the address has already been validated, but 2731e99668a5SLiam R. Howlett * not unmapped, but the maps are removed from the list. 2732e99668a5SLiam R. Howlett */ 2733e99668a5SLiam R. Howlett vma = vm_area_alloc(mm); 2734e99668a5SLiam R. Howlett if (!vma) { 2735e99668a5SLiam R. Howlett error = -ENOMEM; 2736e99668a5SLiam R. Howlett goto unacct_error; 2737e99668a5SLiam R. Howlett } 2738e99668a5SLiam R. Howlett 273953bee98dSLiam R. Howlett vma_iter_config(&vmi, addr, end); 2740e99668a5SLiam R. Howlett vma->vm_start = addr; 2741e99668a5SLiam R. Howlett vma->vm_end = end; 27421c71222eSSuren Baghdasaryan vm_flags_init(vma, vm_flags); 2743e99668a5SLiam R. Howlett vma->vm_page_prot = vm_get_page_prot(vm_flags); 2744e99668a5SLiam R. Howlett vma->vm_pgoff = pgoff; 2745e99668a5SLiam R. Howlett 2746e99668a5SLiam R. Howlett if (file) { 2747e99668a5SLiam R. Howlett if (vm_flags & VM_SHARED) { 2748e99668a5SLiam R. Howlett error = mapping_map_writable(file->f_mapping); 2749e99668a5SLiam R. Howlett if (error) 2750e99668a5SLiam R. Howlett goto free_vma; 2751e99668a5SLiam R. Howlett } 2752e99668a5SLiam R. Howlett 2753e99668a5SLiam R. Howlett vma->vm_file = get_file(file); 2754e99668a5SLiam R. Howlett error = call_mmap(file, vma); 2755e99668a5SLiam R. Howlett if (error) 2756e99668a5SLiam R. Howlett goto unmap_and_free_vma; 2757e99668a5SLiam R. Howlett 2758a57b7051SLiam Howlett /* 2759a57b7051SLiam Howlett * Expansion is handled above, merging is handled below. 2760a57b7051SLiam Howlett * Drivers should not alter the address of the VMA. 2761e99668a5SLiam R. Howlett */ 2762a57b7051SLiam Howlett error = -EINVAL; 2763cc8d1b09SLiam R. Howlett if (WARN_ON((addr != vma->vm_start))) 2764a57b7051SLiam Howlett goto close_and_free_vma; 2765e99668a5SLiam R. Howlett 276653bee98dSLiam R. Howlett vma_iter_config(&vmi, addr, end); 2767e99668a5SLiam R. Howlett /* 2768e99668a5SLiam R. Howlett * If vm_flags changed after call_mmap(), we should try merge 2769e99668a5SLiam R. Howlett * vma again as we may succeed this time. 2770e99668a5SLiam R. Howlett */ 2771e99668a5SLiam R. Howlett if (unlikely(vm_flags != vma->vm_flags && prev)) { 27729760ebffSLiam R. Howlett merge = vma_merge(&vmi, mm, prev, vma->vm_start, 27739760ebffSLiam R. Howlett vma->vm_end, vma->vm_flags, NULL, 27749760ebffSLiam R. Howlett vma->vm_file, vma->vm_pgoff, NULL, 27759760ebffSLiam R. Howlett NULL_VM_UFFD_CTX, NULL); 2776e99668a5SLiam R. Howlett if (merge) { 2777e99668a5SLiam R. Howlett /* 2778e99668a5SLiam R. Howlett * ->mmap() can change vma->vm_file and fput 2779e99668a5SLiam R. Howlett * the original file. So fput the vma->vm_file 2780e99668a5SLiam R. Howlett * here or we would add an extra fput for file 2781e99668a5SLiam R. Howlett * and cause general protection fault 2782e99668a5SLiam R. Howlett * ultimately. 2783e99668a5SLiam R. Howlett */ 2784e99668a5SLiam R. Howlett fput(vma->vm_file); 2785e99668a5SLiam R. Howlett vm_area_free(vma); 2786e99668a5SLiam R. Howlett vma = merge; 2787e99668a5SLiam R. Howlett /* Update vm_flags to pick up the change. */ 2788e99668a5SLiam R. Howlett vm_flags = vma->vm_flags; 2789e99668a5SLiam R. Howlett goto unmap_writable; 2790e99668a5SLiam R. Howlett } 2791e99668a5SLiam R. Howlett } 2792e99668a5SLiam R. Howlett 2793e99668a5SLiam R. Howlett vm_flags = vma->vm_flags; 2794e99668a5SLiam R. Howlett } else if (vm_flags & VM_SHARED) { 2795e99668a5SLiam R. Howlett error = shmem_zero_setup(vma); 2796e99668a5SLiam R. Howlett if (error) 2797e99668a5SLiam R. Howlett goto free_vma; 2798e99668a5SLiam R. Howlett } else { 2799e99668a5SLiam R. Howlett vma_set_anonymous(vma); 2800e99668a5SLiam R. Howlett } 2801e99668a5SLiam R. Howlett 2802b507808eSJoey Gouly if (map_deny_write_exec(vma, vma->vm_flags)) { 2803b507808eSJoey Gouly error = -EACCES; 2804b507808eSJoey Gouly goto close_and_free_vma; 2805b507808eSJoey Gouly } 2806b507808eSJoey Gouly 2807e99668a5SLiam R. Howlett /* Allow architectures to sanity-check the vm_flags */ 2808e99668a5SLiam R. Howlett error = -EINVAL; 2809cc8d1b09SLiam R. Howlett if (!arch_validate_flags(vma->vm_flags)) 2810deb0f656SCarlos Llamas goto close_and_free_vma; 2811e99668a5SLiam R. Howlett 2812e99668a5SLiam R. Howlett error = -ENOMEM; 2813b5df0922SLiam R. Howlett if (vma_iter_prealloc(&vmi, vma)) 28145789151eSMike Kravetz goto close_and_free_vma; 2815e99668a5SLiam R. Howlett 28161c7873e3SHugh Dickins /* Lock the VMA since it is modified after insertion into VMA tree */ 28171c7873e3SHugh Dickins vma_start_write(vma); 2818183654ceSLiam R. Howlett vma_iter_store(&vmi, vma); 2819e99668a5SLiam R. Howlett mm->map_count++; 2820e99668a5SLiam R. Howlett if (vma->vm_file) { 28216852c46cSYu Ma i_mmap_lock_write(vma->vm_file->f_mapping); 2822e99668a5SLiam R. Howlett if (vma->vm_flags & VM_SHARED) 2823e99668a5SLiam R. Howlett mapping_allow_writable(vma->vm_file->f_mapping); 2824e99668a5SLiam R. Howlett 2825e99668a5SLiam R. Howlett flush_dcache_mmap_lock(vma->vm_file->f_mapping); 2826e99668a5SLiam R. Howlett vma_interval_tree_insert(vma, &vma->vm_file->f_mapping->i_mmap); 2827e99668a5SLiam R. Howlett flush_dcache_mmap_unlock(vma->vm_file->f_mapping); 2828e99668a5SLiam R. Howlett i_mmap_unlock_write(vma->vm_file->f_mapping); 2829e99668a5SLiam R. Howlett } 2830e99668a5SLiam R. Howlett 2831e99668a5SLiam R. Howlett /* 2832e99668a5SLiam R. Howlett * vma_merge() calls khugepaged_enter_vma() either, the below 2833e99668a5SLiam R. Howlett * call covers the non-merge case. 2834e99668a5SLiam R. Howlett */ 2835e99668a5SLiam R. Howlett khugepaged_enter_vma(vma, vma->vm_flags); 2836e99668a5SLiam R. Howlett 2837e99668a5SLiam R. Howlett /* Once vma denies write, undo our temporary denial count */ 2838e99668a5SLiam R. Howlett unmap_writable: 2839e99668a5SLiam R. Howlett if (file && vm_flags & VM_SHARED) 2840e99668a5SLiam R. Howlett mapping_unmap_writable(file->f_mapping); 2841e99668a5SLiam R. Howlett file = vma->vm_file; 2842d7597f59SStefan Roesch ksm_add_vma(vma); 2843e99668a5SLiam R. Howlett expanded: 2844e99668a5SLiam R. Howlett perf_event_mmap(vma); 2845e99668a5SLiam R. Howlett 2846e99668a5SLiam R. Howlett vm_stat_account(mm, vm_flags, len >> PAGE_SHIFT); 2847e99668a5SLiam R. Howlett if (vm_flags & VM_LOCKED) { 2848e99668a5SLiam R. Howlett if ((vm_flags & VM_SPECIAL) || vma_is_dax(vma) || 2849e99668a5SLiam R. Howlett is_vm_hugetlb_page(vma) || 2850e99668a5SLiam R. Howlett vma == get_gate_vma(current->mm)) 2851e430a95aSSuren Baghdasaryan vm_flags_clear(vma, VM_LOCKED_MASK); 2852e99668a5SLiam R. Howlett else 2853e99668a5SLiam R. Howlett mm->locked_vm += (len >> PAGE_SHIFT); 2854e99668a5SLiam R. Howlett } 2855e99668a5SLiam R. Howlett 2856e99668a5SLiam R. Howlett if (file) 2857e99668a5SLiam R. Howlett uprobe_mmap(vma); 2858e99668a5SLiam R. Howlett 2859e99668a5SLiam R. Howlett /* 2860e99668a5SLiam R. Howlett * New (or expanded) vma always get soft dirty status. 2861e99668a5SLiam R. Howlett * Otherwise user-space soft-dirty page tracker won't 2862e99668a5SLiam R. Howlett * be able to distinguish situation when vma area unmapped, 2863e99668a5SLiam R. Howlett * then new mapped in-place (which must be aimed as 2864e99668a5SLiam R. Howlett * a completely new data area). 2865e99668a5SLiam R. Howlett */ 28661c71222eSSuren Baghdasaryan vm_flags_set(vma, VM_SOFTDIRTY); 2867e99668a5SLiam R. Howlett 2868e99668a5SLiam R. Howlett vma_set_page_prot(vma); 2869e99668a5SLiam R. Howlett 2870e99668a5SLiam R. Howlett validate_mm(mm); 2871e99668a5SLiam R. Howlett return addr; 2872e99668a5SLiam R. Howlett 2873deb0f656SCarlos Llamas close_and_free_vma: 2874cc8d1b09SLiam R. Howlett if (file && vma->vm_ops && vma->vm_ops->close) 2875deb0f656SCarlos Llamas vma->vm_ops->close(vma); 2876cc8d1b09SLiam R. Howlett 2877cc8d1b09SLiam R. Howlett if (file || vma->vm_file) { 2878e99668a5SLiam R. Howlett unmap_and_free_vma: 2879e99668a5SLiam R. Howlett fput(vma->vm_file); 2880e99668a5SLiam R. Howlett vma->vm_file = NULL; 2881e99668a5SLiam R. Howlett 2882fd892593SLiam R. Howlett vma_iter_set(&vmi, vma->vm_end); 2883e99668a5SLiam R. Howlett /* Undo any partial mapping done by a device driver. */ 2884fd892593SLiam R. Howlett unmap_region(mm, &vmi.mas, vma, prev, next, vma->vm_start, 2885fd892593SLiam R. Howlett vma->vm_end, vma->vm_end, true); 2886cc8d1b09SLiam R. Howlett } 2887cc674ab3SLi Zetao if (file && (vm_flags & VM_SHARED)) 2888e99668a5SLiam R. Howlett mapping_unmap_writable(file->f_mapping); 2889e99668a5SLiam R. Howlett free_vma: 2890e99668a5SLiam R. Howlett vm_area_free(vma); 2891e99668a5SLiam R. Howlett unacct_error: 2892e99668a5SLiam R. Howlett if (charged) 2893e99668a5SLiam R. Howlett vm_unacct_memory(charged); 2894e99668a5SLiam R. Howlett validate_mm(mm); 2895e99668a5SLiam R. Howlett return error; 2896e99668a5SLiam R. Howlett } 2897e99668a5SLiam R. Howlett 2898408579cdSLiam R. Howlett static int __vm_munmap(unsigned long start, size_t len, bool unlock) 2899a46ef99dSLinus Torvalds { 2900a46ef99dSLinus Torvalds int ret; 2901bfce281cSAl Viro struct mm_struct *mm = current->mm; 2902897ab3e0SMike Rapoport LIST_HEAD(uf); 2903183654ceSLiam R. Howlett VMA_ITERATOR(vmi, mm, start); 2904a46ef99dSLinus Torvalds 2905d8ed45c5SMichel Lespinasse if (mmap_write_lock_killable(mm)) 2906ae798783SMichal Hocko return -EINTR; 2907ae798783SMichal Hocko 2908408579cdSLiam R. Howlett ret = do_vmi_munmap(&vmi, mm, start, len, &uf, unlock); 2909408579cdSLiam R. Howlett if (ret || !unlock) 2910d8ed45c5SMichel Lespinasse mmap_write_unlock(mm); 2911dd2283f2SYang Shi 2912897ab3e0SMike Rapoport userfaultfd_unmap_complete(mm, &uf); 2913a46ef99dSLinus Torvalds return ret; 2914a46ef99dSLinus Torvalds } 2915dd2283f2SYang Shi 2916dd2283f2SYang Shi int vm_munmap(unsigned long start, size_t len) 2917dd2283f2SYang Shi { 2918dd2283f2SYang Shi return __vm_munmap(start, len, false); 2919dd2283f2SYang Shi } 2920a46ef99dSLinus Torvalds EXPORT_SYMBOL(vm_munmap); 2921a46ef99dSLinus Torvalds 29226a6160a7SHeiko Carstens SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len) 29231da177e4SLinus Torvalds { 2924ce18d171SCatalin Marinas addr = untagged_addr(addr); 2925dd2283f2SYang Shi return __vm_munmap(addr, len, true); 29261da177e4SLinus Torvalds } 29271da177e4SLinus Torvalds 2928c8d78c18SKirill A. Shutemov 2929c8d78c18SKirill A. Shutemov /* 2930c8d78c18SKirill A. Shutemov * Emulation of deprecated remap_file_pages() syscall. 2931c8d78c18SKirill A. Shutemov */ 2932c8d78c18SKirill A. Shutemov SYSCALL_DEFINE5(remap_file_pages, unsigned long, start, unsigned long, size, 2933c8d78c18SKirill A. Shutemov unsigned long, prot, unsigned long, pgoff, unsigned long, flags) 2934c8d78c18SKirill A. Shutemov { 2935c8d78c18SKirill A. Shutemov 2936c8d78c18SKirill A. Shutemov struct mm_struct *mm = current->mm; 2937c8d78c18SKirill A. Shutemov struct vm_area_struct *vma; 2938c8d78c18SKirill A. Shutemov unsigned long populate = 0; 2939c8d78c18SKirill A. Shutemov unsigned long ret = -EINVAL; 2940c8d78c18SKirill A. Shutemov struct file *file; 2941c8d78c18SKirill A. Shutemov 2942ee65728eSMike Rapoport pr_warn_once("%s (%d) uses deprecated remap_file_pages() syscall. See Documentation/mm/remap_file_pages.rst.\n", 2943c8d78c18SKirill A. Shutemov current->comm, current->pid); 2944c8d78c18SKirill A. Shutemov 2945c8d78c18SKirill A. Shutemov if (prot) 2946c8d78c18SKirill A. Shutemov return ret; 2947c8d78c18SKirill A. Shutemov start = start & PAGE_MASK; 2948c8d78c18SKirill A. Shutemov size = size & PAGE_MASK; 2949c8d78c18SKirill A. Shutemov 2950c8d78c18SKirill A. Shutemov if (start + size <= start) 2951c8d78c18SKirill A. Shutemov return ret; 2952c8d78c18SKirill A. Shutemov 2953c8d78c18SKirill A. Shutemov /* Does pgoff wrap? */ 2954c8d78c18SKirill A. Shutemov if (pgoff + (size >> PAGE_SHIFT) < pgoff) 2955c8d78c18SKirill A. Shutemov return ret; 2956c8d78c18SKirill A. Shutemov 2957d8ed45c5SMichel Lespinasse if (mmap_write_lock_killable(mm)) 2958dc0ef0dfSMichal Hocko return -EINTR; 2959dc0ef0dfSMichal Hocko 29609b593cb2SLiam R. Howlett vma = vma_lookup(mm, start); 2961c8d78c18SKirill A. Shutemov 2962c8d78c18SKirill A. Shutemov if (!vma || !(vma->vm_flags & VM_SHARED)) 2963c8d78c18SKirill A. Shutemov goto out; 2964c8d78c18SKirill A. Shutemov 296548f7df32SKirill A. Shutemov if (start + size > vma->vm_end) { 2966763ecb03SLiam R. Howlett VMA_ITERATOR(vmi, mm, vma->vm_end); 2967763ecb03SLiam R. Howlett struct vm_area_struct *next, *prev = vma; 296848f7df32SKirill A. Shutemov 2969763ecb03SLiam R. Howlett for_each_vma_range(vmi, next, start + size) { 297048f7df32SKirill A. Shutemov /* hole between vmas ? */ 2971763ecb03SLiam R. Howlett if (next->vm_start != prev->vm_end) 297248f7df32SKirill A. Shutemov goto out; 297348f7df32SKirill A. Shutemov 297448f7df32SKirill A. Shutemov if (next->vm_file != vma->vm_file) 297548f7df32SKirill A. Shutemov goto out; 297648f7df32SKirill A. Shutemov 297748f7df32SKirill A. Shutemov if (next->vm_flags != vma->vm_flags) 297848f7df32SKirill A. Shutemov goto out; 297948f7df32SKirill A. Shutemov 29801db43d3fSLiam Howlett if (start + size <= next->vm_end) 29811db43d3fSLiam Howlett break; 29821db43d3fSLiam Howlett 2983763ecb03SLiam R. Howlett prev = next; 298448f7df32SKirill A. Shutemov } 298548f7df32SKirill A. Shutemov 298648f7df32SKirill A. Shutemov if (!next) 2987c8d78c18SKirill A. Shutemov goto out; 2988c8d78c18SKirill A. Shutemov } 2989c8d78c18SKirill A. Shutemov 2990c8d78c18SKirill A. Shutemov prot |= vma->vm_flags & VM_READ ? PROT_READ : 0; 2991c8d78c18SKirill A. Shutemov prot |= vma->vm_flags & VM_WRITE ? PROT_WRITE : 0; 2992c8d78c18SKirill A. Shutemov prot |= vma->vm_flags & VM_EXEC ? PROT_EXEC : 0; 2993c8d78c18SKirill A. Shutemov 2994c8d78c18SKirill A. Shutemov flags &= MAP_NONBLOCK; 2995c8d78c18SKirill A. Shutemov flags |= MAP_SHARED | MAP_FIXED | MAP_POPULATE; 2996fce000b1SLiam Howlett if (vma->vm_flags & VM_LOCKED) 2997c8d78c18SKirill A. Shutemov flags |= MAP_LOCKED; 299848f7df32SKirill A. Shutemov 2999c8d78c18SKirill A. Shutemov file = get_file(vma->vm_file); 300045e55300SPeter Collingbourne ret = do_mmap(vma->vm_file, start, size, 3001592b5fadSYu-cheng Yu prot, flags, 0, pgoff, &populate, NULL); 3002c8d78c18SKirill A. Shutemov fput(file); 3003c8d78c18SKirill A. Shutemov out: 3004d8ed45c5SMichel Lespinasse mmap_write_unlock(mm); 3005c8d78c18SKirill A. Shutemov if (populate) 3006c8d78c18SKirill A. Shutemov mm_populate(ret, populate); 3007c8d78c18SKirill A. Shutemov if (!IS_ERR_VALUE(ret)) 3008c8d78c18SKirill A. Shutemov ret = 0; 3009c8d78c18SKirill A. Shutemov return ret; 3010c8d78c18SKirill A. Shutemov } 3011c8d78c18SKirill A. Shutemov 30121da177e4SLinus Torvalds /* 301327b26701SLiam R. Howlett * do_vma_munmap() - Unmap a full or partial vma. 301427b26701SLiam R. Howlett * @vmi: The vma iterator pointing at the vma 301527b26701SLiam R. Howlett * @vma: The first vma to be munmapped 301627b26701SLiam R. Howlett * @start: the start of the address to unmap 301727b26701SLiam R. Howlett * @end: The end of the address to unmap 30182e7ce7d3SLiam R. Howlett * @uf: The userfaultfd list_head 3019408579cdSLiam R. Howlett * @unlock: Drop the lock on success 30202e7ce7d3SLiam R. Howlett * 302127b26701SLiam R. Howlett * unmaps a VMA mapping when the vma iterator is already in position. 302227b26701SLiam R. Howlett * Does not handle alignment. 3023408579cdSLiam R. Howlett * 3024408579cdSLiam R. Howlett * Return: 0 on success drops the lock of so directed, error on failure and will 3025408579cdSLiam R. Howlett * still hold the lock. 30261da177e4SLinus Torvalds */ 302727b26701SLiam R. Howlett int do_vma_munmap(struct vma_iterator *vmi, struct vm_area_struct *vma, 3028408579cdSLiam R. Howlett unsigned long start, unsigned long end, struct list_head *uf, 3029408579cdSLiam R. Howlett bool unlock) 30302e7ce7d3SLiam R. Howlett { 30312e7ce7d3SLiam R. Howlett struct mm_struct *mm = vma->vm_mm; 30322e7ce7d3SLiam R. Howlett 303327b26701SLiam R. Howlett arch_unmap(mm, start, end); 3034b5641a5dSLinus Torvalds return do_vmi_align_munmap(vmi, vma, mm, start, end, uf, unlock); 30352e7ce7d3SLiam R. Howlett } 30362e7ce7d3SLiam R. Howlett 30372e7ce7d3SLiam R. Howlett /* 30382e7ce7d3SLiam R. Howlett * do_brk_flags() - Increase the brk vma if the flags match. 303992fed820SLiam R. Howlett * @vmi: The vma iterator 30402e7ce7d3SLiam R. Howlett * @addr: The start address 30412e7ce7d3SLiam R. Howlett * @len: The length of the increase 30422e7ce7d3SLiam R. Howlett * @vma: The vma, 30432e7ce7d3SLiam R. Howlett * @flags: The VMA Flags 30442e7ce7d3SLiam R. Howlett * 30452e7ce7d3SLiam R. Howlett * Extend the brk VMA from addr to addr + len. If the VMA is NULL or the flags 30462e7ce7d3SLiam R. Howlett * do not match then create a new anonymous VMA. Eventually we may be able to 30472e7ce7d3SLiam R. Howlett * do some brk-specific accounting here. 30482e7ce7d3SLiam R. Howlett */ 304992fed820SLiam R. Howlett static int do_brk_flags(struct vma_iterator *vmi, struct vm_area_struct *vma, 3050763ecb03SLiam R. Howlett unsigned long addr, unsigned long len, unsigned long flags) 30511da177e4SLinus Torvalds { 30521da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 3053287051b1SLiam R. Howlett struct vma_prepare vp; 30542e7ce7d3SLiam R. Howlett 30552e7ce7d3SLiam R. Howlett /* 30562e7ce7d3SLiam R. Howlett * Check against address space limits by the changed size 30572e7ce7d3SLiam R. Howlett * Note: This happens *after* clearing old mappings in some code paths. 30582e7ce7d3SLiam R. Howlett */ 305916e72e9bSDenys Vlasenko flags |= VM_DATA_DEFAULT_FLAGS | VM_ACCOUNT | mm->def_flags; 306084638335SKonstantin Khlebnikov if (!may_expand_vm(mm, flags, len >> PAGE_SHIFT)) 30611da177e4SLinus Torvalds return -ENOMEM; 30621da177e4SLinus Torvalds 30631da177e4SLinus Torvalds if (mm->map_count > sysctl_max_map_count) 30641da177e4SLinus Torvalds return -ENOMEM; 30651da177e4SLinus Torvalds 3066191c5424SAl Viro if (security_vm_enough_memory_mm(mm, len >> PAGE_SHIFT)) 30671da177e4SLinus Torvalds return -ENOMEM; 30681da177e4SLinus Torvalds 30691da177e4SLinus Torvalds /* 30702e7ce7d3SLiam R. Howlett * Expand the existing vma if possible; Note that singular lists do not 30712e7ce7d3SLiam R. Howlett * occur after forking, so the expand will only happen on new VMAs. 30721da177e4SLinus Torvalds */ 30736c28ca64SLiam Howlett if (vma && vma->vm_end == addr && !vma_policy(vma) && 30746c28ca64SLiam Howlett can_vma_merge_after(vma, flags, NULL, NULL, 30756c28ca64SLiam Howlett addr >> PAGE_SHIFT, NULL_VM_UFFD_CTX, NULL)) { 3076b5df0922SLiam R. Howlett vma_iter_config(vmi, vma->vm_start, addr + len); 3077b5df0922SLiam R. Howlett if (vma_iter_prealloc(vmi, vma)) 3078675eaca1SAlistair Popple goto unacct_fail; 307928c5609fSLiam Howlett 3080c9d6e982SSuren Baghdasaryan vma_start_write(vma); 3081c9d6e982SSuren Baghdasaryan 3082287051b1SLiam R. Howlett init_vma_prep(&vp, vma); 3083287051b1SLiam R. Howlett vma_prepare(&vp); 3084ccf1d78dSSuren Baghdasaryan vma_adjust_trans_huge(vma, vma->vm_start, addr + len, 0); 30852e7ce7d3SLiam R. Howlett vma->vm_end = addr + len; 30861c71222eSSuren Baghdasaryan vm_flags_set(vma, VM_SOFTDIRTY); 308792fed820SLiam R. Howlett vma_iter_store(vmi, vma); 30882e7ce7d3SLiam R. Howlett 3089287051b1SLiam R. Howlett vma_complete(&vp, vmi, mm); 30902e7ce7d3SLiam R. Howlett khugepaged_enter_vma(vma, flags); 30912e7ce7d3SLiam R. Howlett goto out; 30922e7ce7d3SLiam R. Howlett } 30932e7ce7d3SLiam R. Howlett 3094b5df0922SLiam R. Howlett if (vma) 3095b5df0922SLiam R. Howlett vma_iter_next_range(vmi); 30962e7ce7d3SLiam R. Howlett /* create a vma struct for an anonymous mapping */ 30972e7ce7d3SLiam R. Howlett vma = vm_area_alloc(mm); 30982e7ce7d3SLiam R. Howlett if (!vma) 3099675eaca1SAlistair Popple goto unacct_fail; 31001da177e4SLinus Torvalds 3101bfd40eafSKirill A. Shutemov vma_set_anonymous(vma); 31021da177e4SLinus Torvalds vma->vm_start = addr; 31031da177e4SLinus Torvalds vma->vm_end = addr + len; 31042e7ce7d3SLiam R. Howlett vma->vm_pgoff = addr >> PAGE_SHIFT; 31051c71222eSSuren Baghdasaryan vm_flags_init(vma, flags); 31063ed75eb8SColy Li vma->vm_page_prot = vm_get_page_prot(flags); 3107ad9f0063SSuren Baghdasaryan vma_start_write(vma); 310892fed820SLiam R. Howlett if (vma_iter_store_gfp(vmi, vma, GFP_KERNEL)) 31092e7ce7d3SLiam R. Howlett goto mas_store_fail; 3110d4af56c5SLiam R. Howlett 31112e7ce7d3SLiam R. Howlett mm->map_count++; 31122574d5e4SLiam R. Howlett validate_mm(mm); 3113d7597f59SStefan Roesch ksm_add_vma(vma); 31141da177e4SLinus Torvalds out: 31153af9e859SEric B Munson perf_event_mmap(vma); 31161da177e4SLinus Torvalds mm->total_vm += len >> PAGE_SHIFT; 311784638335SKonstantin Khlebnikov mm->data_vm += len >> PAGE_SHIFT; 3118128557ffSMichel Lespinasse if (flags & VM_LOCKED) 3119ba470de4SRik van Riel mm->locked_vm += (len >> PAGE_SHIFT); 31201c71222eSSuren Baghdasaryan vm_flags_set(vma, VM_SOFTDIRTY); 31215d22fc25SLinus Torvalds return 0; 3122d4af56c5SLiam R. Howlett 31232e7ce7d3SLiam R. Howlett mas_store_fail: 3124d4af56c5SLiam R. Howlett vm_area_free(vma); 3125675eaca1SAlistair Popple unacct_fail: 31262e7ce7d3SLiam R. Howlett vm_unacct_memory(len >> PAGE_SHIFT); 31272e7ce7d3SLiam R. Howlett return -ENOMEM; 31281da177e4SLinus Torvalds } 31291da177e4SLinus Torvalds 3130bb177a73SMichal Hocko int vm_brk_flags(unsigned long addr, unsigned long request, unsigned long flags) 3131e4eb1ff6SLinus Torvalds { 3132e4eb1ff6SLinus Torvalds struct mm_struct *mm = current->mm; 31332e7ce7d3SLiam R. Howlett struct vm_area_struct *vma = NULL; 3134bb177a73SMichal Hocko unsigned long len; 31355d22fc25SLinus Torvalds int ret; 3136128557ffSMichel Lespinasse bool populate; 3137897ab3e0SMike Rapoport LIST_HEAD(uf); 313892fed820SLiam R. Howlett VMA_ITERATOR(vmi, mm, addr); 3139e4eb1ff6SLinus Torvalds 3140bb177a73SMichal Hocko len = PAGE_ALIGN(request); 3141bb177a73SMichal Hocko if (len < request) 3142bb177a73SMichal Hocko return -ENOMEM; 3143bb177a73SMichal Hocko if (!len) 3144bb177a73SMichal Hocko return 0; 3145bb177a73SMichal Hocko 3146d8ed45c5SMichel Lespinasse if (mmap_write_lock_killable(mm)) 31472d6c9282SMichal Hocko return -EINTR; 31482d6c9282SMichal Hocko 31492e7ce7d3SLiam R. Howlett /* Until we need other flags, refuse anything except VM_EXEC. */ 31502e7ce7d3SLiam R. Howlett if ((flags & (~VM_EXEC)) != 0) 31512e7ce7d3SLiam R. Howlett return -EINVAL; 31522e7ce7d3SLiam R. Howlett 31532e7ce7d3SLiam R. Howlett ret = check_brk_limits(addr, len); 31542e7ce7d3SLiam R. Howlett if (ret) 31552e7ce7d3SLiam R. Howlett goto limits_failed; 31562e7ce7d3SLiam R. Howlett 3157183654ceSLiam R. Howlett ret = do_vmi_munmap(&vmi, mm, addr, len, &uf, 0); 31582e7ce7d3SLiam R. Howlett if (ret) 31592e7ce7d3SLiam R. Howlett goto munmap_failed; 31602e7ce7d3SLiam R. Howlett 316192fed820SLiam R. Howlett vma = vma_prev(&vmi); 316292fed820SLiam R. Howlett ret = do_brk_flags(&vmi, vma, addr, len, flags); 3163128557ffSMichel Lespinasse populate = ((mm->def_flags & VM_LOCKED) != 0); 3164d8ed45c5SMichel Lespinasse mmap_write_unlock(mm); 3165897ab3e0SMike Rapoport userfaultfd_unmap_complete(mm, &uf); 31665d22fc25SLinus Torvalds if (populate && !ret) 3167128557ffSMichel Lespinasse mm_populate(addr, len); 3168e4eb1ff6SLinus Torvalds return ret; 31692e7ce7d3SLiam R. Howlett 31702e7ce7d3SLiam R. Howlett munmap_failed: 31712e7ce7d3SLiam R. Howlett limits_failed: 31722e7ce7d3SLiam R. Howlett mmap_write_unlock(mm); 31732e7ce7d3SLiam R. Howlett return ret; 3174e4eb1ff6SLinus Torvalds } 317516e72e9bSDenys Vlasenko EXPORT_SYMBOL(vm_brk_flags); 317616e72e9bSDenys Vlasenko 317716e72e9bSDenys Vlasenko int vm_brk(unsigned long addr, unsigned long len) 317816e72e9bSDenys Vlasenko { 317916e72e9bSDenys Vlasenko return vm_brk_flags(addr, len, 0); 318016e72e9bSDenys Vlasenko } 3181e4eb1ff6SLinus Torvalds EXPORT_SYMBOL(vm_brk); 31821da177e4SLinus Torvalds 31831da177e4SLinus Torvalds /* Release all mmaps. */ 31841da177e4SLinus Torvalds void exit_mmap(struct mm_struct *mm) 31851da177e4SLinus Torvalds { 3186d16dfc55SPeter Zijlstra struct mmu_gather tlb; 3187ba470de4SRik van Riel struct vm_area_struct *vma; 31881da177e4SLinus Torvalds unsigned long nr_accounted = 0; 3189763ecb03SLiam R. Howlett MA_STATE(mas, &mm->mm_mt, 0, 0); 3190763ecb03SLiam R. Howlett int count = 0; 31911da177e4SLinus Torvalds 3192d6dd61c8SJeremy Fitzhardinge /* mm's last user has gone, and its about to be pulled down */ 3193cddb8a5cSAndrea Arcangeli mmu_notifier_release(mm); 3194d6dd61c8SJeremy Fitzhardinge 3195bf3980c8SSuren Baghdasaryan mmap_read_lock(mm); 31969480c53eSJeremy Fitzhardinge arch_exit_mmap(mm); 31979480c53eSJeremy Fitzhardinge 3198763ecb03SLiam R. Howlett vma = mas_find(&mas, ULONG_MAX); 319964591e86SSuren Baghdasaryan if (!vma) { 320064591e86SSuren Baghdasaryan /* Can happen if dup_mmap() received an OOM */ 3201bf3980c8SSuren Baghdasaryan mmap_read_unlock(mm); 32029480c53eSJeremy Fitzhardinge return; 320364591e86SSuren Baghdasaryan } 32049480c53eSJeremy Fitzhardinge 32051da177e4SLinus Torvalds lru_add_drain(); 32061da177e4SLinus Torvalds flush_cache_mm(mm); 3207d8b45053SWill Deacon tlb_gather_mmu_fullmm(&tlb, mm); 3208901608d9SOleg Nesterov /* update_hiwater_rss(mm) here? but nobody should be looking */ 3209763ecb03SLiam R. Howlett /* Use ULONG_MAX here to ensure all VMAs in the mm are unmapped */ 3210fd892593SLiam R. Howlett unmap_vmas(&tlb, &mas, vma, 0, ULONG_MAX, ULONG_MAX, false); 3211bf3980c8SSuren Baghdasaryan mmap_read_unlock(mm); 3212bf3980c8SSuren Baghdasaryan 3213bf3980c8SSuren Baghdasaryan /* 3214bf3980c8SSuren Baghdasaryan * Set MMF_OOM_SKIP to hide this task from the oom killer/reaper 3215b3541d91SSuren Baghdasaryan * because the memory has been already freed. 3216bf3980c8SSuren Baghdasaryan */ 3217bf3980c8SSuren Baghdasaryan set_bit(MMF_OOM_SKIP, &mm->flags); 3218bf3980c8SSuren Baghdasaryan mmap_write_lock(mm); 32193dd44325SLiam R. Howlett mt_clear_in_rcu(&mm->mm_mt); 3220fd892593SLiam R. Howlett mas_set(&mas, vma->vm_end); 3221fd892593SLiam R. Howlett free_pgtables(&tlb, &mas, vma, FIRST_USER_ADDRESS, 322298e51a22SSuren Baghdasaryan USER_PGTABLES_CEILING, true); 3223ae8eba8bSWill Deacon tlb_finish_mmu(&tlb); 32241da177e4SLinus Torvalds 3225763ecb03SLiam R. Howlett /* 3226763ecb03SLiam R. Howlett * Walk the list again, actually closing and freeing it, with preemption 3227763ecb03SLiam R. Howlett * enabled, without holding any MM locks besides the unreachable 3228763ecb03SLiam R. Howlett * mmap_write_lock. 3229763ecb03SLiam R. Howlett */ 3230fd892593SLiam R. Howlett mas_set(&mas, vma->vm_end); 3231763ecb03SLiam R. Howlett do { 32324f74d2c8SLinus Torvalds if (vma->vm_flags & VM_ACCOUNT) 32334f74d2c8SLinus Torvalds nr_accounted += vma_pages(vma); 32340d2ebf9cSSuren Baghdasaryan remove_vma(vma, true); 3235763ecb03SLiam R. Howlett count++; 32360a3b3c25SPaul E. McKenney cond_resched(); 3237763ecb03SLiam R. Howlett } while ((vma = mas_find(&mas, ULONG_MAX)) != NULL); 3238763ecb03SLiam R. Howlett 3239763ecb03SLiam R. Howlett BUG_ON(count != mm->map_count); 3240d4af56c5SLiam R. Howlett 3241d4af56c5SLiam R. Howlett trace_exit_mmap(mm); 3242d4af56c5SLiam R. Howlett __mt_destroy(&mm->mm_mt); 324364591e86SSuren Baghdasaryan mmap_write_unlock(mm); 32444f74d2c8SLinus Torvalds vm_unacct_memory(nr_accounted); 32451da177e4SLinus Torvalds } 32461da177e4SLinus Torvalds 32471da177e4SLinus Torvalds /* Insert vm structure into process list sorted by address 32481da177e4SLinus Torvalds * and into the inode's i_mmap tree. If vm_file is non-NULL 3249c8c06efaSDavidlohr Bueso * then i_mmap_rwsem is taken here. 32501da177e4SLinus Torvalds */ 32511da177e4SLinus Torvalds int insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma) 32521da177e4SLinus Torvalds { 3253d4af56c5SLiam R. Howlett unsigned long charged = vma_pages(vma); 32541da177e4SLinus Torvalds 3255d4af56c5SLiam R. Howlett 3256d0601a50SLiam R. Howlett if (find_vma_intersection(mm, vma->vm_start, vma->vm_end)) 3257c9d13f5fSChen Gang return -ENOMEM; 3258d4af56c5SLiam R. Howlett 3259c9d13f5fSChen Gang if ((vma->vm_flags & VM_ACCOUNT) && 3260d4af56c5SLiam R. Howlett security_vm_enough_memory_mm(mm, charged)) 3261c9d13f5fSChen Gang return -ENOMEM; 3262c9d13f5fSChen Gang 32631da177e4SLinus Torvalds /* 32641da177e4SLinus Torvalds * The vm_pgoff of a purely anonymous vma should be irrelevant 32651da177e4SLinus Torvalds * until its first write fault, when page's anon_vma and index 32661da177e4SLinus Torvalds * are set. But now set the vm_pgoff it will almost certainly 32671da177e4SLinus Torvalds * end up with (unless mremap moves it elsewhere before that 32681da177e4SLinus Torvalds * first wfault), so /proc/pid/maps tells a consistent story. 32691da177e4SLinus Torvalds * 32701da177e4SLinus Torvalds * By setting it to reflect the virtual start address of the 32711da177e4SLinus Torvalds * vma, merges and splits can happen in a seamless way, just 32721da177e4SLinus Torvalds * using the existing file pgoff checks and manipulations. 32738332326eSLiao Pingfang * Similarly in do_mmap and in do_brk_flags. 32741da177e4SLinus Torvalds */ 32758a9cc3b5SOleg Nesterov if (vma_is_anonymous(vma)) { 32761da177e4SLinus Torvalds BUG_ON(vma->anon_vma); 32771da177e4SLinus Torvalds vma->vm_pgoff = vma->vm_start >> PAGE_SHIFT; 32781da177e4SLinus Torvalds } 32792b144498SSrikar Dronamraju 3280763ecb03SLiam R. Howlett if (vma_link(mm, vma)) { 3281*dd34d9feSAnthony Yznaga if (vma->vm_flags & VM_ACCOUNT) 3282d4af56c5SLiam R. Howlett vm_unacct_memory(charged); 3283d4af56c5SLiam R. Howlett return -ENOMEM; 3284d4af56c5SLiam R. Howlett } 3285d4af56c5SLiam R. Howlett 32861da177e4SLinus Torvalds return 0; 32871da177e4SLinus Torvalds } 32881da177e4SLinus Torvalds 32891da177e4SLinus Torvalds /* 32901da177e4SLinus Torvalds * Copy the vma structure to a new location in the same mm, 32911da177e4SLinus Torvalds * prior to moving page table entries, to effect an mremap move. 32921da177e4SLinus Torvalds */ 32931da177e4SLinus Torvalds struct vm_area_struct *copy_vma(struct vm_area_struct **vmap, 329438a76013SMichel Lespinasse unsigned long addr, unsigned long len, pgoff_t pgoff, 329538a76013SMichel Lespinasse bool *need_rmap_locks) 32961da177e4SLinus Torvalds { 32971da177e4SLinus Torvalds struct vm_area_struct *vma = *vmap; 32981da177e4SLinus Torvalds unsigned long vma_start = vma->vm_start; 32991da177e4SLinus Torvalds struct mm_struct *mm = vma->vm_mm; 33001da177e4SLinus Torvalds struct vm_area_struct *new_vma, *prev; 3301948f017bSAndrea Arcangeli bool faulted_in_anon_vma = true; 3302076f16bfSLiam R. Howlett VMA_ITERATOR(vmi, mm, addr); 33031da177e4SLinus Torvalds 33041da177e4SLinus Torvalds /* 33051da177e4SLinus Torvalds * If anonymous vma has not yet been faulted, update new pgoff 33061da177e4SLinus Torvalds * to match new location, to increase its chance of merging. 33071da177e4SLinus Torvalds */ 3308ce75799bSOleg Nesterov if (unlikely(vma_is_anonymous(vma) && !vma->anon_vma)) { 33091da177e4SLinus Torvalds pgoff = addr >> PAGE_SHIFT; 3310948f017bSAndrea Arcangeli faulted_in_anon_vma = false; 3311948f017bSAndrea Arcangeli } 33121da177e4SLinus Torvalds 3313763ecb03SLiam R. Howlett new_vma = find_vma_prev(mm, addr, &prev); 3314763ecb03SLiam R. Howlett if (new_vma && new_vma->vm_start < addr + len) 33156597d783SHugh Dickins return NULL; /* should never get here */ 3316524e00b3SLiam R. Howlett 33179760ebffSLiam R. Howlett new_vma = vma_merge(&vmi, mm, prev, addr, addr + len, vma->vm_flags, 331819a809afSAndrea Arcangeli vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma), 33195c26f6acSSuren Baghdasaryan vma->vm_userfaultfd_ctx, anon_vma_name(vma)); 33201da177e4SLinus Torvalds if (new_vma) { 33211da177e4SLinus Torvalds /* 33221da177e4SLinus Torvalds * Source vma may have been merged into new_vma 33231da177e4SLinus Torvalds */ 3324948f017bSAndrea Arcangeli if (unlikely(vma_start >= new_vma->vm_start && 3325948f017bSAndrea Arcangeli vma_start < new_vma->vm_end)) { 3326948f017bSAndrea Arcangeli /* 3327948f017bSAndrea Arcangeli * The only way we can get a vma_merge with 3328948f017bSAndrea Arcangeli * self during an mremap is if the vma hasn't 3329948f017bSAndrea Arcangeli * been faulted in yet and we were allowed to 3330948f017bSAndrea Arcangeli * reset the dst vma->vm_pgoff to the 3331948f017bSAndrea Arcangeli * destination address of the mremap to allow 3332948f017bSAndrea Arcangeli * the merge to happen. mremap must change the 3333948f017bSAndrea Arcangeli * vm_pgoff linearity between src and dst vmas 3334948f017bSAndrea Arcangeli * (in turn preventing a vma_merge) to be 3335948f017bSAndrea Arcangeli * safe. It is only safe to keep the vm_pgoff 3336948f017bSAndrea Arcangeli * linear if there are no pages mapped yet. 3337948f017bSAndrea Arcangeli */ 333881d1b09cSSasha Levin VM_BUG_ON_VMA(faulted_in_anon_vma, new_vma); 333938a76013SMichel Lespinasse *vmap = vma = new_vma; 3340108d6642SMichel Lespinasse } 334138a76013SMichel Lespinasse *need_rmap_locks = (new_vma->vm_pgoff <= vma->vm_pgoff); 33421da177e4SLinus Torvalds } else { 33433928d4f5SLinus Torvalds new_vma = vm_area_dup(vma); 3344e3975891SChen Gang if (!new_vma) 3345e3975891SChen Gang goto out; 33461da177e4SLinus Torvalds new_vma->vm_start = addr; 33471da177e4SLinus Torvalds new_vma->vm_end = addr + len; 33481da177e4SLinus Torvalds new_vma->vm_pgoff = pgoff; 3349ef0855d3SOleg Nesterov if (vma_dup_policy(vma, new_vma)) 3350523d4e20SMichel Lespinasse goto out_free_vma; 3351523d4e20SMichel Lespinasse if (anon_vma_clone(new_vma, vma)) 3352523d4e20SMichel Lespinasse goto out_free_mempol; 3353e9714acfSKonstantin Khlebnikov if (new_vma->vm_file) 33541da177e4SLinus Torvalds get_file(new_vma->vm_file); 33551da177e4SLinus Torvalds if (new_vma->vm_ops && new_vma->vm_ops->open) 33561da177e4SLinus Torvalds new_vma->vm_ops->open(new_vma); 3357763ecb03SLiam R. Howlett if (vma_link(mm, new_vma)) 3358524e00b3SLiam R. Howlett goto out_vma_link; 335938a76013SMichel Lespinasse *need_rmap_locks = false; 33601da177e4SLinus Torvalds } 33611da177e4SLinus Torvalds return new_vma; 33625beb4930SRik van Riel 3363524e00b3SLiam R. Howlett out_vma_link: 3364524e00b3SLiam R. Howlett if (new_vma->vm_ops && new_vma->vm_ops->close) 3365524e00b3SLiam R. Howlett new_vma->vm_ops->close(new_vma); 336692b73996SLiam Howlett 336792b73996SLiam Howlett if (new_vma->vm_file) 336892b73996SLiam Howlett fput(new_vma->vm_file); 336992b73996SLiam Howlett 337092b73996SLiam Howlett unlink_anon_vmas(new_vma); 33715beb4930SRik van Riel out_free_mempol: 3372ef0855d3SOleg Nesterov mpol_put(vma_policy(new_vma)); 33735beb4930SRik van Riel out_free_vma: 33743928d4f5SLinus Torvalds vm_area_free(new_vma); 3375e3975891SChen Gang out: 33765beb4930SRik van Riel return NULL; 33771da177e4SLinus Torvalds } 3378119f657cSakpm@osdl.org 3379119f657cSakpm@osdl.org /* 3380119f657cSakpm@osdl.org * Return true if the calling process may expand its vm space by the passed 3381119f657cSakpm@osdl.org * number of pages 3382119f657cSakpm@osdl.org */ 338384638335SKonstantin Khlebnikov bool may_expand_vm(struct mm_struct *mm, vm_flags_t flags, unsigned long npages) 3384119f657cSakpm@osdl.org { 338584638335SKonstantin Khlebnikov if (mm->total_vm + npages > rlimit(RLIMIT_AS) >> PAGE_SHIFT) 338684638335SKonstantin Khlebnikov return false; 3387119f657cSakpm@osdl.org 3388d977d56cSKonstantin Khlebnikov if (is_data_mapping(flags) && 3389d977d56cSKonstantin Khlebnikov mm->data_vm + npages > rlimit(RLIMIT_DATA) >> PAGE_SHIFT) { 3390f4fcd558SKonstantin Khlebnikov /* Workaround for Valgrind */ 3391f4fcd558SKonstantin Khlebnikov if (rlimit(RLIMIT_DATA) == 0 && 3392f4fcd558SKonstantin Khlebnikov mm->data_vm + npages <= rlimit_max(RLIMIT_DATA) >> PAGE_SHIFT) 3393f4fcd558SKonstantin Khlebnikov return true; 339457a7702bSDavid Woodhouse 339557a7702bSDavid Woodhouse pr_warn_once("%s (%d): VmData %lu exceed data ulimit %lu. Update limits%s.\n", 3396d977d56cSKonstantin Khlebnikov current->comm, current->pid, 3397d977d56cSKonstantin Khlebnikov (mm->data_vm + npages) << PAGE_SHIFT, 339857a7702bSDavid Woodhouse rlimit(RLIMIT_DATA), 339957a7702bSDavid Woodhouse ignore_rlimit_data ? "" : " or use boot option ignore_rlimit_data"); 340057a7702bSDavid Woodhouse 340157a7702bSDavid Woodhouse if (!ignore_rlimit_data) 3402d977d56cSKonstantin Khlebnikov return false; 3403d977d56cSKonstantin Khlebnikov } 3404119f657cSakpm@osdl.org 340584638335SKonstantin Khlebnikov return true; 340684638335SKonstantin Khlebnikov } 340784638335SKonstantin Khlebnikov 340884638335SKonstantin Khlebnikov void vm_stat_account(struct mm_struct *mm, vm_flags_t flags, long npages) 340984638335SKonstantin Khlebnikov { 34107866076bSPeng Liu WRITE_ONCE(mm->total_vm, READ_ONCE(mm->total_vm)+npages); 341184638335SKonstantin Khlebnikov 3412d977d56cSKonstantin Khlebnikov if (is_exec_mapping(flags)) 341384638335SKonstantin Khlebnikov mm->exec_vm += npages; 3414d977d56cSKonstantin Khlebnikov else if (is_stack_mapping(flags)) 341584638335SKonstantin Khlebnikov mm->stack_vm += npages; 3416d977d56cSKonstantin Khlebnikov else if (is_data_mapping(flags)) 341784638335SKonstantin Khlebnikov mm->data_vm += npages; 3418119f657cSakpm@osdl.org } 3419fa5dc22fSRoland McGrath 3420b3ec9f33SSouptick Joarder static vm_fault_t special_mapping_fault(struct vm_fault *vmf); 3421a62c34bdSAndy Lutomirski 3422a62c34bdSAndy Lutomirski /* 3423a62c34bdSAndy Lutomirski * Having a close hook prevents vma merging regardless of flags. 3424a62c34bdSAndy Lutomirski */ 3425a62c34bdSAndy Lutomirski static void special_mapping_close(struct vm_area_struct *vma) 3426a62c34bdSAndy Lutomirski { 3427a62c34bdSAndy Lutomirski } 3428a62c34bdSAndy Lutomirski 3429a62c34bdSAndy Lutomirski static const char *special_mapping_name(struct vm_area_struct *vma) 3430a62c34bdSAndy Lutomirski { 3431a62c34bdSAndy Lutomirski return ((struct vm_special_mapping *)vma->vm_private_data)->name; 3432a62c34bdSAndy Lutomirski } 3433a62c34bdSAndy Lutomirski 343414d07113SBrian Geffon static int special_mapping_mremap(struct vm_area_struct *new_vma) 3435b059a453SDmitry Safonov { 3436b059a453SDmitry Safonov struct vm_special_mapping *sm = new_vma->vm_private_data; 3437b059a453SDmitry Safonov 3438280e87e9SDmitry Safonov if (WARN_ON_ONCE(current->mm != new_vma->vm_mm)) 3439280e87e9SDmitry Safonov return -EFAULT; 3440280e87e9SDmitry Safonov 3441b059a453SDmitry Safonov if (sm->mremap) 3442b059a453SDmitry Safonov return sm->mremap(sm, new_vma); 3443280e87e9SDmitry Safonov 3444b059a453SDmitry Safonov return 0; 3445b059a453SDmitry Safonov } 3446b059a453SDmitry Safonov 3447871402e0SDmitry Safonov static int special_mapping_split(struct vm_area_struct *vma, unsigned long addr) 3448871402e0SDmitry Safonov { 3449871402e0SDmitry Safonov /* 3450871402e0SDmitry Safonov * Forbid splitting special mappings - kernel has expectations over 3451871402e0SDmitry Safonov * the number of pages in mapping. Together with VM_DONTEXPAND 3452871402e0SDmitry Safonov * the size of vma should stay the same over the special mapping's 3453871402e0SDmitry Safonov * lifetime. 3454871402e0SDmitry Safonov */ 3455871402e0SDmitry Safonov return -EINVAL; 3456871402e0SDmitry Safonov } 3457871402e0SDmitry Safonov 3458a62c34bdSAndy Lutomirski static const struct vm_operations_struct special_mapping_vmops = { 3459a62c34bdSAndy Lutomirski .close = special_mapping_close, 3460a62c34bdSAndy Lutomirski .fault = special_mapping_fault, 3461b059a453SDmitry Safonov .mremap = special_mapping_mremap, 3462a62c34bdSAndy Lutomirski .name = special_mapping_name, 3463af34ebebSDmitry Safonov /* vDSO code relies that VVAR can't be accessed remotely */ 3464af34ebebSDmitry Safonov .access = NULL, 3465871402e0SDmitry Safonov .may_split = special_mapping_split, 3466a62c34bdSAndy Lutomirski }; 3467a62c34bdSAndy Lutomirski 3468a62c34bdSAndy Lutomirski static const struct vm_operations_struct legacy_special_mapping_vmops = { 3469a62c34bdSAndy Lutomirski .close = special_mapping_close, 3470a62c34bdSAndy Lutomirski .fault = special_mapping_fault, 3471a62c34bdSAndy Lutomirski }; 3472fa5dc22fSRoland McGrath 3473b3ec9f33SSouptick Joarder static vm_fault_t special_mapping_fault(struct vm_fault *vmf) 3474fa5dc22fSRoland McGrath { 347511bac800SDave Jiang struct vm_area_struct *vma = vmf->vma; 3476b1d0e4f5SNick Piggin pgoff_t pgoff; 3477fa5dc22fSRoland McGrath struct page **pages; 3478fa5dc22fSRoland McGrath 3479f872f540SAndy Lutomirski if (vma->vm_ops == &legacy_special_mapping_vmops) { 3480a62c34bdSAndy Lutomirski pages = vma->vm_private_data; 3481f872f540SAndy Lutomirski } else { 3482f872f540SAndy Lutomirski struct vm_special_mapping *sm = vma->vm_private_data; 3483f872f540SAndy Lutomirski 3484f872f540SAndy Lutomirski if (sm->fault) 348511bac800SDave Jiang return sm->fault(sm, vmf->vma, vmf); 3486f872f540SAndy Lutomirski 3487f872f540SAndy Lutomirski pages = sm->pages; 3488f872f540SAndy Lutomirski } 3489a62c34bdSAndy Lutomirski 34908a9cc3b5SOleg Nesterov for (pgoff = vmf->pgoff; pgoff && *pages; ++pages) 3491b1d0e4f5SNick Piggin pgoff--; 3492fa5dc22fSRoland McGrath 3493fa5dc22fSRoland McGrath if (*pages) { 3494fa5dc22fSRoland McGrath struct page *page = *pages; 3495fa5dc22fSRoland McGrath get_page(page); 3496b1d0e4f5SNick Piggin vmf->page = page; 3497b1d0e4f5SNick Piggin return 0; 3498fa5dc22fSRoland McGrath } 3499fa5dc22fSRoland McGrath 3500b1d0e4f5SNick Piggin return VM_FAULT_SIGBUS; 3501fa5dc22fSRoland McGrath } 3502fa5dc22fSRoland McGrath 3503a62c34bdSAndy Lutomirski static struct vm_area_struct *__install_special_mapping( 3504a62c34bdSAndy Lutomirski struct mm_struct *mm, 3505fa5dc22fSRoland McGrath unsigned long addr, unsigned long len, 350627f28b97SChen Gang unsigned long vm_flags, void *priv, 350727f28b97SChen Gang const struct vm_operations_struct *ops) 3508fa5dc22fSRoland McGrath { 3509462e635eSTavis Ormandy int ret; 3510fa5dc22fSRoland McGrath struct vm_area_struct *vma; 3511fa5dc22fSRoland McGrath 3512490fc053SLinus Torvalds vma = vm_area_alloc(mm); 3513fa5dc22fSRoland McGrath if (unlikely(vma == NULL)) 35143935ed6aSStefani Seibold return ERR_PTR(-ENOMEM); 3515fa5dc22fSRoland McGrath 3516fa5dc22fSRoland McGrath vma->vm_start = addr; 3517fa5dc22fSRoland McGrath vma->vm_end = addr + len; 3518fa5dc22fSRoland McGrath 3519e430a95aSSuren Baghdasaryan vm_flags_init(vma, (vm_flags | mm->def_flags | 3520e430a95aSSuren Baghdasaryan VM_DONTEXPAND | VM_SOFTDIRTY) & ~VM_LOCKED_MASK); 35213ed75eb8SColy Li vma->vm_page_prot = vm_get_page_prot(vma->vm_flags); 3522fa5dc22fSRoland McGrath 3523a62c34bdSAndy Lutomirski vma->vm_ops = ops; 3524a62c34bdSAndy Lutomirski vma->vm_private_data = priv; 3525fa5dc22fSRoland McGrath 3526462e635eSTavis Ormandy ret = insert_vm_struct(mm, vma); 3527462e635eSTavis Ormandy if (ret) 3528462e635eSTavis Ormandy goto out; 3529fa5dc22fSRoland McGrath 353084638335SKonstantin Khlebnikov vm_stat_account(mm, vma->vm_flags, len >> PAGE_SHIFT); 3531fa5dc22fSRoland McGrath 3532cdd6c482SIngo Molnar perf_event_mmap(vma); 3533089dd79dSPeter Zijlstra 35343935ed6aSStefani Seibold return vma; 3535462e635eSTavis Ormandy 3536462e635eSTavis Ormandy out: 35373928d4f5SLinus Torvalds vm_area_free(vma); 35383935ed6aSStefani Seibold return ERR_PTR(ret); 35393935ed6aSStefani Seibold } 35403935ed6aSStefani Seibold 35412eefd878SDmitry Safonov bool vma_is_special_mapping(const struct vm_area_struct *vma, 35422eefd878SDmitry Safonov const struct vm_special_mapping *sm) 35432eefd878SDmitry Safonov { 35442eefd878SDmitry Safonov return vma->vm_private_data == sm && 35452eefd878SDmitry Safonov (vma->vm_ops == &special_mapping_vmops || 35462eefd878SDmitry Safonov vma->vm_ops == &legacy_special_mapping_vmops); 35472eefd878SDmitry Safonov } 35482eefd878SDmitry Safonov 3549a62c34bdSAndy Lutomirski /* 3550c1e8d7c6SMichel Lespinasse * Called with mm->mmap_lock held for writing. 3551a62c34bdSAndy Lutomirski * Insert a new vma covering the given region, with the given flags. 3552a62c34bdSAndy Lutomirski * Its pages are supplied by the given array of struct page *. 3553a62c34bdSAndy Lutomirski * The array can be shorter than len >> PAGE_SHIFT if it's null-terminated. 3554a62c34bdSAndy Lutomirski * The region past the last page supplied will always produce SIGBUS. 3555a62c34bdSAndy Lutomirski * The array pointer and the pages it points to are assumed to stay alive 3556a62c34bdSAndy Lutomirski * for as long as this mapping might exist. 3557a62c34bdSAndy Lutomirski */ 3558a62c34bdSAndy Lutomirski struct vm_area_struct *_install_special_mapping( 3559a62c34bdSAndy Lutomirski struct mm_struct *mm, 3560a62c34bdSAndy Lutomirski unsigned long addr, unsigned long len, 3561a62c34bdSAndy Lutomirski unsigned long vm_flags, const struct vm_special_mapping *spec) 3562a62c34bdSAndy Lutomirski { 356327f28b97SChen Gang return __install_special_mapping(mm, addr, len, vm_flags, (void *)spec, 356427f28b97SChen Gang &special_mapping_vmops); 3565a62c34bdSAndy Lutomirski } 3566a62c34bdSAndy Lutomirski 35673935ed6aSStefani Seibold int install_special_mapping(struct mm_struct *mm, 35683935ed6aSStefani Seibold unsigned long addr, unsigned long len, 35693935ed6aSStefani Seibold unsigned long vm_flags, struct page **pages) 35703935ed6aSStefani Seibold { 3571a62c34bdSAndy Lutomirski struct vm_area_struct *vma = __install_special_mapping( 357227f28b97SChen Gang mm, addr, len, vm_flags, (void *)pages, 357327f28b97SChen Gang &legacy_special_mapping_vmops); 35743935ed6aSStefani Seibold 357514bd5b45SDuan Jiong return PTR_ERR_OR_ZERO(vma); 3576fa5dc22fSRoland McGrath } 35777906d00cSAndrea Arcangeli 35787906d00cSAndrea Arcangeli static DEFINE_MUTEX(mm_all_locks_mutex); 35797906d00cSAndrea Arcangeli 3580454ed842SPeter Zijlstra static void vm_lock_anon_vma(struct mm_struct *mm, struct anon_vma *anon_vma) 35817906d00cSAndrea Arcangeli { 3582f808c13fSDavidlohr Bueso if (!test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_root.rb_node)) { 35837906d00cSAndrea Arcangeli /* 35847906d00cSAndrea Arcangeli * The LSB of head.next can't change from under us 35857906d00cSAndrea Arcangeli * because we hold the mm_all_locks_mutex. 35867906d00cSAndrea Arcangeli */ 3587da1c55f1SMichel Lespinasse down_write_nest_lock(&anon_vma->root->rwsem, &mm->mmap_lock); 35887906d00cSAndrea Arcangeli /* 35897906d00cSAndrea Arcangeli * We can safely modify head.next after taking the 35905a505085SIngo Molnar * anon_vma->root->rwsem. If some other vma in this mm shares 35917906d00cSAndrea Arcangeli * the same anon_vma we won't take it again. 35927906d00cSAndrea Arcangeli * 35937906d00cSAndrea Arcangeli * No need of atomic instructions here, head.next 35947906d00cSAndrea Arcangeli * can't change from under us thanks to the 35955a505085SIngo Molnar * anon_vma->root->rwsem. 35967906d00cSAndrea Arcangeli */ 35977906d00cSAndrea Arcangeli if (__test_and_set_bit(0, (unsigned long *) 3598f808c13fSDavidlohr Bueso &anon_vma->root->rb_root.rb_root.rb_node)) 35997906d00cSAndrea Arcangeli BUG(); 36007906d00cSAndrea Arcangeli } 36017906d00cSAndrea Arcangeli } 36027906d00cSAndrea Arcangeli 3603454ed842SPeter Zijlstra static void vm_lock_mapping(struct mm_struct *mm, struct address_space *mapping) 36047906d00cSAndrea Arcangeli { 36057906d00cSAndrea Arcangeli if (!test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) { 36067906d00cSAndrea Arcangeli /* 36077906d00cSAndrea Arcangeli * AS_MM_ALL_LOCKS can't change from under us because 36087906d00cSAndrea Arcangeli * we hold the mm_all_locks_mutex. 36097906d00cSAndrea Arcangeli * 36107906d00cSAndrea Arcangeli * Operations on ->flags have to be atomic because 36117906d00cSAndrea Arcangeli * even if AS_MM_ALL_LOCKS is stable thanks to the 36127906d00cSAndrea Arcangeli * mm_all_locks_mutex, there may be other cpus 36137906d00cSAndrea Arcangeli * changing other bitflags in parallel to us. 36147906d00cSAndrea Arcangeli */ 36157906d00cSAndrea Arcangeli if (test_and_set_bit(AS_MM_ALL_LOCKS, &mapping->flags)) 36167906d00cSAndrea Arcangeli BUG(); 3617da1c55f1SMichel Lespinasse down_write_nest_lock(&mapping->i_mmap_rwsem, &mm->mmap_lock); 36187906d00cSAndrea Arcangeli } 36197906d00cSAndrea Arcangeli } 36207906d00cSAndrea Arcangeli 36217906d00cSAndrea Arcangeli /* 36227906d00cSAndrea Arcangeli * This operation locks against the VM for all pte/vma/mm related 36237906d00cSAndrea Arcangeli * operations that could ever happen on a certain mm. This includes 36247906d00cSAndrea Arcangeli * vmtruncate, try_to_unmap, and all page faults. 36257906d00cSAndrea Arcangeli * 3626c1e8d7c6SMichel Lespinasse * The caller must take the mmap_lock in write mode before calling 36277906d00cSAndrea Arcangeli * mm_take_all_locks(). The caller isn't allowed to release the 3628c1e8d7c6SMichel Lespinasse * mmap_lock until mm_drop_all_locks() returns. 36297906d00cSAndrea Arcangeli * 3630c1e8d7c6SMichel Lespinasse * mmap_lock in write mode is required in order to block all operations 36317906d00cSAndrea Arcangeli * that could modify pagetables and free pages without need of 363227ba0644SKirill A. Shutemov * altering the vma layout. It's also needed in write mode to avoid new 36337906d00cSAndrea Arcangeli * anon_vmas to be associated with existing vmas. 36347906d00cSAndrea Arcangeli * 36357906d00cSAndrea Arcangeli * A single task can't take more than one mm_take_all_locks() in a row 36367906d00cSAndrea Arcangeli * or it would deadlock. 36377906d00cSAndrea Arcangeli * 3638bf181b9fSMichel Lespinasse * The LSB in anon_vma->rb_root.rb_node and the AS_MM_ALL_LOCKS bitflag in 36397906d00cSAndrea Arcangeli * mapping->flags avoid to take the same lock twice, if more than one 36407906d00cSAndrea Arcangeli * vma in this mm is backed by the same anon_vma or address_space. 36417906d00cSAndrea Arcangeli * 364288f306b6SKirill A. Shutemov * We take locks in following order, accordingly to comment at beginning 364388f306b6SKirill A. Shutemov * of mm/rmap.c: 364488f306b6SKirill A. Shutemov * - all hugetlbfs_i_mmap_rwsem_key locks (aka mapping->i_mmap_rwsem for 364588f306b6SKirill A. Shutemov * hugetlb mapping); 3646eeff9a5dSSuren Baghdasaryan * - all vmas marked locked 364788f306b6SKirill A. Shutemov * - all i_mmap_rwsem locks; 364888f306b6SKirill A. Shutemov * - all anon_vma->rwseml 364988f306b6SKirill A. Shutemov * 365088f306b6SKirill A. Shutemov * We can take all locks within these types randomly because the VM code 365188f306b6SKirill A. Shutemov * doesn't nest them and we protected from parallel mm_take_all_locks() by 365288f306b6SKirill A. Shutemov * mm_all_locks_mutex. 36537906d00cSAndrea Arcangeli * 36547906d00cSAndrea Arcangeli * mm_take_all_locks() and mm_drop_all_locks are expensive operations 36557906d00cSAndrea Arcangeli * that may have to take thousand of locks. 36567906d00cSAndrea Arcangeli * 36577906d00cSAndrea Arcangeli * mm_take_all_locks() can fail if it's interrupted by signals. 36587906d00cSAndrea Arcangeli */ 36597906d00cSAndrea Arcangeli int mm_take_all_locks(struct mm_struct *mm) 36607906d00cSAndrea Arcangeli { 36617906d00cSAndrea Arcangeli struct vm_area_struct *vma; 36625beb4930SRik van Riel struct anon_vma_chain *avc; 3663763ecb03SLiam R. Howlett MA_STATE(mas, &mm->mm_mt, 0, 0); 36647906d00cSAndrea Arcangeli 3665325bca1fSRolf Eike Beer mmap_assert_write_locked(mm); 36667906d00cSAndrea Arcangeli 36677906d00cSAndrea Arcangeli mutex_lock(&mm_all_locks_mutex); 36687906d00cSAndrea Arcangeli 366990717566SJann Horn /* 367090717566SJann Horn * vma_start_write() does not have a complement in mm_drop_all_locks() 367190717566SJann Horn * because vma_start_write() is always asymmetrical; it marks a VMA as 367290717566SJann Horn * being written to until mmap_write_unlock() or mmap_write_downgrade() 367390717566SJann Horn * is reached. 367490717566SJann Horn */ 3675763ecb03SLiam R. Howlett mas_for_each(&mas, vma, ULONG_MAX) { 36767906d00cSAndrea Arcangeli if (signal_pending(current)) 36777906d00cSAndrea Arcangeli goto out_unlock; 3678eeff9a5dSSuren Baghdasaryan vma_start_write(vma); 3679eeff9a5dSSuren Baghdasaryan } 3680eeff9a5dSSuren Baghdasaryan 3681eeff9a5dSSuren Baghdasaryan mas_set(&mas, 0); 3682eeff9a5dSSuren Baghdasaryan mas_for_each(&mas, vma, ULONG_MAX) { 3683eeff9a5dSSuren Baghdasaryan if (signal_pending(current)) 3684eeff9a5dSSuren Baghdasaryan goto out_unlock; 368588f306b6SKirill A. Shutemov if (vma->vm_file && vma->vm_file->f_mapping && 368688f306b6SKirill A. Shutemov is_vm_hugetlb_page(vma)) 368788f306b6SKirill A. Shutemov vm_lock_mapping(mm, vma->vm_file->f_mapping); 368888f306b6SKirill A. Shutemov } 368988f306b6SKirill A. Shutemov 3690763ecb03SLiam R. Howlett mas_set(&mas, 0); 3691763ecb03SLiam R. Howlett mas_for_each(&mas, vma, ULONG_MAX) { 369288f306b6SKirill A. Shutemov if (signal_pending(current)) 369388f306b6SKirill A. Shutemov goto out_unlock; 369488f306b6SKirill A. Shutemov if (vma->vm_file && vma->vm_file->f_mapping && 369588f306b6SKirill A. Shutemov !is_vm_hugetlb_page(vma)) 3696454ed842SPeter Zijlstra vm_lock_mapping(mm, vma->vm_file->f_mapping); 36977906d00cSAndrea Arcangeli } 36987cd5a02fSPeter Zijlstra 3699763ecb03SLiam R. Howlett mas_set(&mas, 0); 3700763ecb03SLiam R. Howlett mas_for_each(&mas, vma, ULONG_MAX) { 37017cd5a02fSPeter Zijlstra if (signal_pending(current)) 37027cd5a02fSPeter Zijlstra goto out_unlock; 37037cd5a02fSPeter Zijlstra if (vma->anon_vma) 37045beb4930SRik van Riel list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) 37055beb4930SRik van Riel vm_lock_anon_vma(mm, avc->anon_vma); 37067cd5a02fSPeter Zijlstra } 37077cd5a02fSPeter Zijlstra 3708584cff54SKautuk Consul return 0; 37097906d00cSAndrea Arcangeli 37107906d00cSAndrea Arcangeli out_unlock: 37117906d00cSAndrea Arcangeli mm_drop_all_locks(mm); 3712584cff54SKautuk Consul return -EINTR; 37137906d00cSAndrea Arcangeli } 37147906d00cSAndrea Arcangeli 37157906d00cSAndrea Arcangeli static void vm_unlock_anon_vma(struct anon_vma *anon_vma) 37167906d00cSAndrea Arcangeli { 3717f808c13fSDavidlohr Bueso if (test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_root.rb_node)) { 37187906d00cSAndrea Arcangeli /* 37197906d00cSAndrea Arcangeli * The LSB of head.next can't change to 0 from under 37207906d00cSAndrea Arcangeli * us because we hold the mm_all_locks_mutex. 37217906d00cSAndrea Arcangeli * 37227906d00cSAndrea Arcangeli * We must however clear the bitflag before unlocking 3723bf181b9fSMichel Lespinasse * the vma so the users using the anon_vma->rb_root will 37247906d00cSAndrea Arcangeli * never see our bitflag. 37257906d00cSAndrea Arcangeli * 37267906d00cSAndrea Arcangeli * No need of atomic instructions here, head.next 37277906d00cSAndrea Arcangeli * can't change from under us until we release the 37285a505085SIngo Molnar * anon_vma->root->rwsem. 37297906d00cSAndrea Arcangeli */ 37307906d00cSAndrea Arcangeli if (!__test_and_clear_bit(0, (unsigned long *) 3731f808c13fSDavidlohr Bueso &anon_vma->root->rb_root.rb_root.rb_node)) 37327906d00cSAndrea Arcangeli BUG(); 373308b52706SKonstantin Khlebnikov anon_vma_unlock_write(anon_vma); 37347906d00cSAndrea Arcangeli } 37357906d00cSAndrea Arcangeli } 37367906d00cSAndrea Arcangeli 37377906d00cSAndrea Arcangeli static void vm_unlock_mapping(struct address_space *mapping) 37387906d00cSAndrea Arcangeli { 37397906d00cSAndrea Arcangeli if (test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) { 37407906d00cSAndrea Arcangeli /* 37417906d00cSAndrea Arcangeli * AS_MM_ALL_LOCKS can't change to 0 from under us 37427906d00cSAndrea Arcangeli * because we hold the mm_all_locks_mutex. 37437906d00cSAndrea Arcangeli */ 374483cde9e8SDavidlohr Bueso i_mmap_unlock_write(mapping); 37457906d00cSAndrea Arcangeli if (!test_and_clear_bit(AS_MM_ALL_LOCKS, 37467906d00cSAndrea Arcangeli &mapping->flags)) 37477906d00cSAndrea Arcangeli BUG(); 37487906d00cSAndrea Arcangeli } 37497906d00cSAndrea Arcangeli } 37507906d00cSAndrea Arcangeli 37517906d00cSAndrea Arcangeli /* 3752c1e8d7c6SMichel Lespinasse * The mmap_lock cannot be released by the caller until 37537906d00cSAndrea Arcangeli * mm_drop_all_locks() returns. 37547906d00cSAndrea Arcangeli */ 37557906d00cSAndrea Arcangeli void mm_drop_all_locks(struct mm_struct *mm) 37567906d00cSAndrea Arcangeli { 37577906d00cSAndrea Arcangeli struct vm_area_struct *vma; 37585beb4930SRik van Riel struct anon_vma_chain *avc; 3759763ecb03SLiam R. Howlett MA_STATE(mas, &mm->mm_mt, 0, 0); 37607906d00cSAndrea Arcangeli 3761325bca1fSRolf Eike Beer mmap_assert_write_locked(mm); 37627906d00cSAndrea Arcangeli BUG_ON(!mutex_is_locked(&mm_all_locks_mutex)); 37637906d00cSAndrea Arcangeli 3764763ecb03SLiam R. Howlett mas_for_each(&mas, vma, ULONG_MAX) { 37657906d00cSAndrea Arcangeli if (vma->anon_vma) 37665beb4930SRik van Riel list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) 37675beb4930SRik van Riel vm_unlock_anon_vma(avc->anon_vma); 37687906d00cSAndrea Arcangeli if (vma->vm_file && vma->vm_file->f_mapping) 37697906d00cSAndrea Arcangeli vm_unlock_mapping(vma->vm_file->f_mapping); 37707906d00cSAndrea Arcangeli } 37717906d00cSAndrea Arcangeli 37727906d00cSAndrea Arcangeli mutex_unlock(&mm_all_locks_mutex); 37737906d00cSAndrea Arcangeli } 37748feae131SDavid Howells 37758feae131SDavid Howells /* 37763edf41d8Sseokhoon.yoon * initialise the percpu counter for VM 37778feae131SDavid Howells */ 37788feae131SDavid Howells void __init mmap_init(void) 37798feae131SDavid Howells { 378000a62ce9SKOSAKI Motohiro int ret; 378100a62ce9SKOSAKI Motohiro 3782908c7f19STejun Heo ret = percpu_counter_init(&vm_committed_as, 0, GFP_KERNEL); 378300a62ce9SKOSAKI Motohiro VM_BUG_ON(ret); 37848feae131SDavid Howells } 3785c9b1d098SAndrew Shewmaker 3786c9b1d098SAndrew Shewmaker /* 3787c9b1d098SAndrew Shewmaker * Initialise sysctl_user_reserve_kbytes. 3788c9b1d098SAndrew Shewmaker * 3789c9b1d098SAndrew Shewmaker * This is intended to prevent a user from starting a single memory hogging 3790c9b1d098SAndrew Shewmaker * process, such that they cannot recover (kill the hog) in OVERCOMMIT_NEVER 3791c9b1d098SAndrew Shewmaker * mode. 3792c9b1d098SAndrew Shewmaker * 3793c9b1d098SAndrew Shewmaker * The default value is min(3% of free memory, 128MB) 3794c9b1d098SAndrew Shewmaker * 128MB is enough to recover with sshd/login, bash, and top/kill. 3795c9b1d098SAndrew Shewmaker */ 37961640879aSAndrew Shewmaker static int init_user_reserve(void) 3797c9b1d098SAndrew Shewmaker { 3798c9b1d098SAndrew Shewmaker unsigned long free_kbytes; 3799c9b1d098SAndrew Shewmaker 3800b1773e0eSZhangPeng free_kbytes = K(global_zone_page_state(NR_FREE_PAGES)); 3801c9b1d098SAndrew Shewmaker 3802c9b1d098SAndrew Shewmaker sysctl_user_reserve_kbytes = min(free_kbytes / 32, 1UL << 17); 3803c9b1d098SAndrew Shewmaker return 0; 3804c9b1d098SAndrew Shewmaker } 3805a64fb3cdSPaul Gortmaker subsys_initcall(init_user_reserve); 38064eeab4f5SAndrew Shewmaker 38074eeab4f5SAndrew Shewmaker /* 38084eeab4f5SAndrew Shewmaker * Initialise sysctl_admin_reserve_kbytes. 38094eeab4f5SAndrew Shewmaker * 38104eeab4f5SAndrew Shewmaker * The purpose of sysctl_admin_reserve_kbytes is to allow the sys admin 38114eeab4f5SAndrew Shewmaker * to log in and kill a memory hogging process. 38124eeab4f5SAndrew Shewmaker * 38134eeab4f5SAndrew Shewmaker * Systems with more than 256MB will reserve 8MB, enough to recover 38144eeab4f5SAndrew Shewmaker * with sshd, bash, and top in OVERCOMMIT_GUESS. Smaller systems will 38154eeab4f5SAndrew Shewmaker * only reserve 3% of free pages by default. 38164eeab4f5SAndrew Shewmaker */ 38171640879aSAndrew Shewmaker static int init_admin_reserve(void) 38184eeab4f5SAndrew Shewmaker { 38194eeab4f5SAndrew Shewmaker unsigned long free_kbytes; 38204eeab4f5SAndrew Shewmaker 3821b1773e0eSZhangPeng free_kbytes = K(global_zone_page_state(NR_FREE_PAGES)); 38224eeab4f5SAndrew Shewmaker 38234eeab4f5SAndrew Shewmaker sysctl_admin_reserve_kbytes = min(free_kbytes / 32, 1UL << 13); 38244eeab4f5SAndrew Shewmaker return 0; 38254eeab4f5SAndrew Shewmaker } 3826a64fb3cdSPaul Gortmaker subsys_initcall(init_admin_reserve); 38271640879aSAndrew Shewmaker 38281640879aSAndrew Shewmaker /* 38291640879aSAndrew Shewmaker * Reinititalise user and admin reserves if memory is added or removed. 38301640879aSAndrew Shewmaker * 38311640879aSAndrew Shewmaker * The default user reserve max is 128MB, and the default max for the 38321640879aSAndrew Shewmaker * admin reserve is 8MB. These are usually, but not always, enough to 38331640879aSAndrew Shewmaker * enable recovery from a memory hogging process using login/sshd, a shell, 38341640879aSAndrew Shewmaker * and tools like top. It may make sense to increase or even disable the 38351640879aSAndrew Shewmaker * reserve depending on the existence of swap or variations in the recovery 38361640879aSAndrew Shewmaker * tools. So, the admin may have changed them. 38371640879aSAndrew Shewmaker * 38381640879aSAndrew Shewmaker * If memory is added and the reserves have been eliminated or increased above 38391640879aSAndrew Shewmaker * the default max, then we'll trust the admin. 38401640879aSAndrew Shewmaker * 38411640879aSAndrew Shewmaker * If memory is removed and there isn't enough free memory, then we 38421640879aSAndrew Shewmaker * need to reset the reserves. 38431640879aSAndrew Shewmaker * 38441640879aSAndrew Shewmaker * Otherwise keep the reserve set by the admin. 38451640879aSAndrew Shewmaker */ 38461640879aSAndrew Shewmaker static int reserve_mem_notifier(struct notifier_block *nb, 38471640879aSAndrew Shewmaker unsigned long action, void *data) 38481640879aSAndrew Shewmaker { 38491640879aSAndrew Shewmaker unsigned long tmp, free_kbytes; 38501640879aSAndrew Shewmaker 38511640879aSAndrew Shewmaker switch (action) { 38521640879aSAndrew Shewmaker case MEM_ONLINE: 38531640879aSAndrew Shewmaker /* Default max is 128MB. Leave alone if modified by operator. */ 38541640879aSAndrew Shewmaker tmp = sysctl_user_reserve_kbytes; 38551640879aSAndrew Shewmaker if (0 < tmp && tmp < (1UL << 17)) 38561640879aSAndrew Shewmaker init_user_reserve(); 38571640879aSAndrew Shewmaker 38581640879aSAndrew Shewmaker /* Default max is 8MB. Leave alone if modified by operator. */ 38591640879aSAndrew Shewmaker tmp = sysctl_admin_reserve_kbytes; 38601640879aSAndrew Shewmaker if (0 < tmp && tmp < (1UL << 13)) 38611640879aSAndrew Shewmaker init_admin_reserve(); 38621640879aSAndrew Shewmaker 38631640879aSAndrew Shewmaker break; 38641640879aSAndrew Shewmaker case MEM_OFFLINE: 3865b1773e0eSZhangPeng free_kbytes = K(global_zone_page_state(NR_FREE_PAGES)); 38661640879aSAndrew Shewmaker 38671640879aSAndrew Shewmaker if (sysctl_user_reserve_kbytes > free_kbytes) { 38681640879aSAndrew Shewmaker init_user_reserve(); 38691640879aSAndrew Shewmaker pr_info("vm.user_reserve_kbytes reset to %lu\n", 38701640879aSAndrew Shewmaker sysctl_user_reserve_kbytes); 38711640879aSAndrew Shewmaker } 38721640879aSAndrew Shewmaker 38731640879aSAndrew Shewmaker if (sysctl_admin_reserve_kbytes > free_kbytes) { 38741640879aSAndrew Shewmaker init_admin_reserve(); 38751640879aSAndrew Shewmaker pr_info("vm.admin_reserve_kbytes reset to %lu\n", 38761640879aSAndrew Shewmaker sysctl_admin_reserve_kbytes); 38771640879aSAndrew Shewmaker } 38781640879aSAndrew Shewmaker break; 38791640879aSAndrew Shewmaker default: 38801640879aSAndrew Shewmaker break; 38811640879aSAndrew Shewmaker } 38821640879aSAndrew Shewmaker return NOTIFY_OK; 38831640879aSAndrew Shewmaker } 38841640879aSAndrew Shewmaker 38851640879aSAndrew Shewmaker static int __meminit init_reserve_notifier(void) 38861640879aSAndrew Shewmaker { 38871eeaa4fdSLiu Shixin if (hotplug_memory_notifier(reserve_mem_notifier, DEFAULT_CALLBACK_PRI)) 3888b1de0d13SMitchel Humpherys pr_err("Failed registering memory add/remove notifier for admin reserve\n"); 38891640879aSAndrew Shewmaker 38901640879aSAndrew Shewmaker return 0; 38911640879aSAndrew Shewmaker } 3892a64fb3cdSPaul Gortmaker subsys_initcall(init_reserve_notifier); 3893