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 404*ad9f0063SSuren Baghdasaryan vma_start_write(vma); 405*ad9f0063SSuren 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 { 46510fca64aSSuren Baghdasaryan vma_start_write(vp->vma); 46610fca64aSSuren Baghdasaryan if (vp->adj_next) 46710fca64aSSuren Baghdasaryan vma_start_write(vp->adj_next); 468*ad9f0063SSuren Baghdasaryan if (vp->insert) 469*ad9f0063SSuren Baghdasaryan vma_start_write(vp->insert); 47010fca64aSSuren Baghdasaryan if (vp->remove) 47110fca64aSSuren Baghdasaryan vma_start_write(vp->remove); 47210fca64aSSuren Baghdasaryan if (vp->remove2) 47310fca64aSSuren Baghdasaryan vma_start_write(vp->remove2); 47410fca64aSSuren Baghdasaryan 475440703e0SLiam R. Howlett if (vp->file) { 476440703e0SLiam R. Howlett uprobe_munmap(vp->vma, vp->vma->vm_start, vp->vma->vm_end); 477440703e0SLiam R. Howlett 478440703e0SLiam R. Howlett if (vp->adj_next) 479440703e0SLiam R. Howlett uprobe_munmap(vp->adj_next, vp->adj_next->vm_start, 480440703e0SLiam R. Howlett vp->adj_next->vm_end); 481440703e0SLiam R. Howlett 482440703e0SLiam R. Howlett i_mmap_lock_write(vp->mapping); 483440703e0SLiam R. Howlett if (vp->insert && vp->insert->vm_file) { 484440703e0SLiam R. Howlett /* 485440703e0SLiam R. Howlett * Put into interval tree now, so instantiated pages 486440703e0SLiam R. Howlett * are visible to arm/parisc __flush_dcache_page 487440703e0SLiam R. Howlett * throughout; but we cannot insert into address 488440703e0SLiam R. Howlett * space until vma start or end is updated. 489440703e0SLiam R. Howlett */ 490440703e0SLiam R. Howlett __vma_link_file(vp->insert, 491440703e0SLiam R. Howlett vp->insert->vm_file->f_mapping); 492440703e0SLiam R. Howlett } 493440703e0SLiam R. Howlett } 494440703e0SLiam R. Howlett 495440703e0SLiam R. Howlett if (vp->anon_vma) { 496440703e0SLiam R. Howlett anon_vma_lock_write(vp->anon_vma); 497440703e0SLiam R. Howlett anon_vma_interval_tree_pre_update_vma(vp->vma); 498440703e0SLiam R. Howlett if (vp->adj_next) 499440703e0SLiam R. Howlett anon_vma_interval_tree_pre_update_vma(vp->adj_next); 500440703e0SLiam R. Howlett } 501440703e0SLiam R. Howlett 502440703e0SLiam R. Howlett if (vp->file) { 503440703e0SLiam R. Howlett flush_dcache_mmap_lock(vp->mapping); 504440703e0SLiam R. Howlett vma_interval_tree_remove(vp->vma, &vp->mapping->i_mmap); 505440703e0SLiam R. Howlett if (vp->adj_next) 506440703e0SLiam R. Howlett vma_interval_tree_remove(vp->adj_next, 507440703e0SLiam R. Howlett &vp->mapping->i_mmap); 508440703e0SLiam R. Howlett } 509440703e0SLiam R. Howlett 510440703e0SLiam R. Howlett } 511440703e0SLiam R. Howlett 512440703e0SLiam R. Howlett /* 513440703e0SLiam R. Howlett * vma_complete- Helper function for handling the unlocking after altering VMAs, 514440703e0SLiam R. Howlett * or for inserting a VMA. 515440703e0SLiam R. Howlett * 516440703e0SLiam R. Howlett * @vp: The vma_prepare struct 517440703e0SLiam R. Howlett * @vmi: The vma iterator 518440703e0SLiam R. Howlett * @mm: The mm_struct 519440703e0SLiam R. Howlett */ 520440703e0SLiam R. Howlett static inline void vma_complete(struct vma_prepare *vp, 521440703e0SLiam R. Howlett struct vma_iterator *vmi, struct mm_struct *mm) 522440703e0SLiam R. Howlett { 523440703e0SLiam R. Howlett if (vp->file) { 524440703e0SLiam R. Howlett if (vp->adj_next) 525440703e0SLiam R. Howlett vma_interval_tree_insert(vp->adj_next, 526440703e0SLiam R. Howlett &vp->mapping->i_mmap); 527440703e0SLiam R. Howlett vma_interval_tree_insert(vp->vma, &vp->mapping->i_mmap); 528440703e0SLiam R. Howlett flush_dcache_mmap_unlock(vp->mapping); 529440703e0SLiam R. Howlett } 530440703e0SLiam R. Howlett 531440703e0SLiam R. Howlett if (vp->remove && vp->file) { 532440703e0SLiam R. Howlett __remove_shared_vm_struct(vp->remove, vp->file, vp->mapping); 533440703e0SLiam R. Howlett if (vp->remove2) 534440703e0SLiam R. Howlett __remove_shared_vm_struct(vp->remove2, vp->file, 535440703e0SLiam R. Howlett vp->mapping); 536440703e0SLiam R. Howlett } else if (vp->insert) { 537440703e0SLiam R. Howlett /* 538440703e0SLiam R. Howlett * split_vma has split insert from vma, and needs 539440703e0SLiam R. Howlett * us to insert it before dropping the locks 540440703e0SLiam R. Howlett * (it may either follow vma or precede it). 541440703e0SLiam R. Howlett */ 542440703e0SLiam R. Howlett vma_iter_store(vmi, vp->insert); 543440703e0SLiam R. Howlett mm->map_count++; 544440703e0SLiam R. Howlett } 545440703e0SLiam R. Howlett 546440703e0SLiam R. Howlett if (vp->anon_vma) { 547440703e0SLiam R. Howlett anon_vma_interval_tree_post_update_vma(vp->vma); 548440703e0SLiam R. Howlett if (vp->adj_next) 549440703e0SLiam R. Howlett anon_vma_interval_tree_post_update_vma(vp->adj_next); 550440703e0SLiam R. Howlett anon_vma_unlock_write(vp->anon_vma); 551440703e0SLiam R. Howlett } 552440703e0SLiam R. Howlett 553440703e0SLiam R. Howlett if (vp->file) { 554440703e0SLiam R. Howlett i_mmap_unlock_write(vp->mapping); 555440703e0SLiam R. Howlett uprobe_mmap(vp->vma); 556440703e0SLiam R. Howlett 557440703e0SLiam R. Howlett if (vp->adj_next) 558440703e0SLiam R. Howlett uprobe_mmap(vp->adj_next); 559440703e0SLiam R. Howlett } 560440703e0SLiam R. Howlett 561440703e0SLiam R. Howlett if (vp->remove) { 562440703e0SLiam R. Howlett again: 563457f67beSSuren Baghdasaryan vma_mark_detached(vp->remove, true); 564440703e0SLiam R. Howlett if (vp->file) { 565440703e0SLiam R. Howlett uprobe_munmap(vp->remove, vp->remove->vm_start, 566440703e0SLiam R. Howlett vp->remove->vm_end); 567440703e0SLiam R. Howlett fput(vp->file); 568440703e0SLiam R. Howlett } 569440703e0SLiam R. Howlett if (vp->remove->anon_vma) 570440703e0SLiam R. Howlett anon_vma_merge(vp->vma, vp->remove); 571440703e0SLiam R. Howlett mm->map_count--; 572440703e0SLiam R. Howlett mpol_put(vma_policy(vp->remove)); 573440703e0SLiam R. Howlett if (!vp->remove2) 574440703e0SLiam R. Howlett WARN_ON_ONCE(vp->vma->vm_end < vp->remove->vm_end); 575440703e0SLiam R. Howlett vm_area_free(vp->remove); 576440703e0SLiam R. Howlett 577440703e0SLiam R. Howlett /* 578440703e0SLiam R. Howlett * In mprotect's case 6 (see comments on vma_merge), 5795ff783f1SVlastimil Babka * we are removing both mid and next vmas 580440703e0SLiam R. Howlett */ 581440703e0SLiam R. Howlett if (vp->remove2) { 582440703e0SLiam R. Howlett vp->remove = vp->remove2; 583440703e0SLiam R. Howlett vp->remove2 = NULL; 584440703e0SLiam R. Howlett goto again; 585440703e0SLiam R. Howlett } 586440703e0SLiam R. Howlett } 587440703e0SLiam R. Howlett if (vp->insert && vp->file) 588440703e0SLiam R. Howlett uprobe_mmap(vp->insert); 5892574d5e4SLiam R. Howlett validate_mm(mm); 590440703e0SLiam R. Howlett } 591440703e0SLiam R. Howlett 592440703e0SLiam R. Howlett /* 59304241ffeSLiam R. Howlett * dup_anon_vma() - Helper function to duplicate anon_vma 59404241ffeSLiam R. Howlett * @dst: The destination VMA 59504241ffeSLiam R. Howlett * @src: The source VMA 59604241ffeSLiam R. Howlett * 59704241ffeSLiam R. Howlett * Returns: 0 on success. 59804241ffeSLiam R. Howlett */ 59904241ffeSLiam R. Howlett static inline int dup_anon_vma(struct vm_area_struct *dst, 60004241ffeSLiam R. Howlett struct vm_area_struct *src) 60104241ffeSLiam R. Howlett { 60204241ffeSLiam R. Howlett /* 60304241ffeSLiam R. Howlett * Easily overlooked: when mprotect shifts the boundary, make sure the 60404241ffeSLiam R. Howlett * expanding vma has anon_vma set if the shrinking vma had, to cover any 60504241ffeSLiam R. Howlett * anon pages imported. 60604241ffeSLiam R. Howlett */ 60704241ffeSLiam R. Howlett if (src->anon_vma && !dst->anon_vma) { 608d8ab9f7bSJann Horn vma_start_write(dst); 60904241ffeSLiam R. Howlett dst->anon_vma = src->anon_vma; 61004241ffeSLiam R. Howlett return anon_vma_clone(dst, src); 61104241ffeSLiam R. Howlett } 61204241ffeSLiam R. Howlett 61304241ffeSLiam R. Howlett return 0; 61404241ffeSLiam R. Howlett } 61504241ffeSLiam R. Howlett 61604241ffeSLiam R. Howlett /* 6179303d3e1SLiam R. Howlett * vma_expand - Expand an existing VMA 6189303d3e1SLiam R. Howlett * 6199303d3e1SLiam R. Howlett * @vmi: The vma iterator 6209303d3e1SLiam R. Howlett * @vma: The vma to expand 6219303d3e1SLiam R. Howlett * @start: The start of the vma 6229303d3e1SLiam R. Howlett * @end: The exclusive end of the vma 6239303d3e1SLiam R. Howlett * @pgoff: The page offset of vma 6249303d3e1SLiam R. Howlett * @next: The current of next vma. 6259303d3e1SLiam R. Howlett * 6269303d3e1SLiam R. Howlett * Expand @vma to @start and @end. Can expand off the start and end. Will 6279303d3e1SLiam R. Howlett * expand over @next if it's different from @vma and @end == @next->vm_end. 6289303d3e1SLiam R. Howlett * Checking if the @vma can expand and merge with @next needs to be handled by 6299303d3e1SLiam R. Howlett * the caller. 6309303d3e1SLiam R. Howlett * 6319303d3e1SLiam R. Howlett * Returns: 0 on success 6329303d3e1SLiam R. Howlett */ 6337c9813e8SLiam R. Howlett int vma_expand(struct vma_iterator *vmi, struct vm_area_struct *vma, 6349303d3e1SLiam R. Howlett unsigned long start, unsigned long end, pgoff_t pgoff, 6359303d3e1SLiam R. Howlett struct vm_area_struct *next) 6369303d3e1SLiam R. Howlett { 63768cefec5SLiam R. Howlett bool remove_next = false; 6389303d3e1SLiam R. Howlett struct vma_prepare vp; 6399303d3e1SLiam R. Howlett 6409303d3e1SLiam R. Howlett if (next && (vma != next) && (end == next->vm_end)) { 64104241ffeSLiam R. Howlett int ret; 6429303d3e1SLiam R. Howlett 64304241ffeSLiam R. Howlett remove_next = true; 64404241ffeSLiam R. Howlett ret = dup_anon_vma(vma, next); 64504241ffeSLiam R. Howlett if (ret) 64604241ffeSLiam R. Howlett return ret; 6479303d3e1SLiam R. Howlett } 6489303d3e1SLiam R. Howlett 64968cefec5SLiam R. Howlett init_multi_vma_prep(&vp, vma, NULL, remove_next ? next : NULL, NULL); 6509303d3e1SLiam R. Howlett /* Not merging but overwriting any part of next is not handled. */ 6519303d3e1SLiam R. Howlett VM_WARN_ON(next && !vp.remove && 6529303d3e1SLiam R. Howlett next != vma && end > next->vm_start); 6539303d3e1SLiam R. Howlett /* Only handles expanding */ 6549303d3e1SLiam R. Howlett VM_WARN_ON(vma->vm_start < start || vma->vm_end > end); 6559303d3e1SLiam R. Howlett 656b5df0922SLiam R. Howlett /* Note: vma iterator must be pointing to 'start' */ 657b5df0922SLiam R. Howlett vma_iter_config(vmi, start, end); 658b5df0922SLiam R. Howlett if (vma_iter_prealloc(vmi, vma)) 6599303d3e1SLiam R. Howlett goto nomem; 6609303d3e1SLiam R. Howlett 661ccf1d78dSSuren Baghdasaryan vma_prepare(&vp); 6629303d3e1SLiam R. Howlett vma_adjust_trans_huge(vma, start, end, 0); 6639303d3e1SLiam R. Howlett vma->vm_start = start; 6649303d3e1SLiam R. Howlett vma->vm_end = end; 6659303d3e1SLiam R. Howlett vma->vm_pgoff = pgoff; 6669303d3e1SLiam R. Howlett vma_iter_store(vmi, vma); 6679303d3e1SLiam R. Howlett 6689303d3e1SLiam R. Howlett vma_complete(&vp, vmi, vma->vm_mm); 6699303d3e1SLiam R. Howlett return 0; 6709303d3e1SLiam R. Howlett 6719303d3e1SLiam R. Howlett nomem: 6729303d3e1SLiam R. Howlett return -ENOMEM; 6739303d3e1SLiam R. Howlett } 674cf51e86dSLiam R. Howlett 675cf51e86dSLiam R. Howlett /* 676cf51e86dSLiam R. Howlett * vma_shrink() - Reduce an existing VMAs memory area 677cf51e86dSLiam R. Howlett * @vmi: The vma iterator 678cf51e86dSLiam R. Howlett * @vma: The VMA to modify 679cf51e86dSLiam R. Howlett * @start: The new start 680cf51e86dSLiam R. Howlett * @end: The new end 681cf51e86dSLiam R. Howlett * 682cf51e86dSLiam R. Howlett * Returns: 0 on success, -ENOMEM otherwise 683cf51e86dSLiam R. Howlett */ 684cf51e86dSLiam R. Howlett int vma_shrink(struct vma_iterator *vmi, struct vm_area_struct *vma, 685cf51e86dSLiam R. Howlett unsigned long start, unsigned long end, pgoff_t pgoff) 686cf51e86dSLiam R. Howlett { 687cf51e86dSLiam R. Howlett struct vma_prepare vp; 688cf51e86dSLiam R. Howlett 689cf51e86dSLiam R. Howlett WARN_ON((vma->vm_start != start) && (vma->vm_end != end)); 690cf51e86dSLiam R. Howlett 691b5df0922SLiam R. Howlett if (vma->vm_start < start) 692b5df0922SLiam R. Howlett vma_iter_config(vmi, vma->vm_start, start); 693b5df0922SLiam R. Howlett else 694b5df0922SLiam R. Howlett vma_iter_config(vmi, end, vma->vm_end); 695b5df0922SLiam R. Howlett 696b5df0922SLiam R. Howlett if (vma_iter_prealloc(vmi, NULL)) 697cf51e86dSLiam R. Howlett return -ENOMEM; 698cf51e86dSLiam R. Howlett 699cf51e86dSLiam R. Howlett init_vma_prep(&vp, vma); 700cf51e86dSLiam R. Howlett vma_prepare(&vp); 701ccf1d78dSSuren Baghdasaryan vma_adjust_trans_huge(vma, start, end, 0); 702cf51e86dSLiam R. Howlett 703b5df0922SLiam R. Howlett vma_iter_clear(vmi); 704cf51e86dSLiam R. Howlett vma->vm_start = start; 705cf51e86dSLiam R. Howlett vma->vm_end = end; 706cf51e86dSLiam R. Howlett vma->vm_pgoff = pgoff; 707cf51e86dSLiam R. Howlett vma_complete(&vp, vmi, vma->vm_mm); 708cf51e86dSLiam R. Howlett return 0; 709cf51e86dSLiam R. Howlett } 710cf51e86dSLiam R. Howlett 7119303d3e1SLiam R. Howlett /* 7121da177e4SLinus Torvalds * If the vma has a ->close operation then the driver probably needs to release 713714965caSVlastimil Babka * per-vma resources, so we don't attempt to merge those if the caller indicates 714714965caSVlastimil Babka * the current vma may be removed as part of the merge. 7151da177e4SLinus Torvalds */ 7162dbf4010SVlastimil Babka static inline bool is_mergeable_vma(struct vm_area_struct *vma, 71719a809afSAndrea Arcangeli struct file *file, unsigned long vm_flags, 7189a10064fSColin Cross struct vm_userfaultfd_ctx vm_userfaultfd_ctx, 719714965caSVlastimil Babka struct anon_vma_name *anon_name, bool may_remove_vma) 7201da177e4SLinus Torvalds { 72134228d47SCyrill Gorcunov /* 72234228d47SCyrill Gorcunov * VM_SOFTDIRTY should not prevent from VMA merging, if we 72334228d47SCyrill Gorcunov * match the flags but dirty bit -- the caller should mark 72434228d47SCyrill Gorcunov * merged VMA as dirty. If dirty bit won't be excluded from 7258bb4e7a2SWei Yang * comparison, we increase pressure on the memory system forcing 72634228d47SCyrill Gorcunov * the kernel to generate new VMAs when old one could be 72734228d47SCyrill Gorcunov * extended instead. 72834228d47SCyrill Gorcunov */ 72934228d47SCyrill Gorcunov if ((vma->vm_flags ^ vm_flags) & ~VM_SOFTDIRTY) 7302dbf4010SVlastimil Babka return false; 7311da177e4SLinus Torvalds if (vma->vm_file != file) 7322dbf4010SVlastimil Babka return false; 733714965caSVlastimil Babka if (may_remove_vma && vma->vm_ops && vma->vm_ops->close) 7342dbf4010SVlastimil Babka return false; 73519a809afSAndrea Arcangeli if (!is_mergeable_vm_userfaultfd_ctx(vma, vm_userfaultfd_ctx)) 7362dbf4010SVlastimil Babka return false; 7375c26f6acSSuren Baghdasaryan if (!anon_vma_name_eq(anon_vma_name(vma), anon_name)) 7382dbf4010SVlastimil Babka return false; 7392dbf4010SVlastimil Babka return true; 7401da177e4SLinus Torvalds } 7411da177e4SLinus Torvalds 7422dbf4010SVlastimil Babka static inline bool is_mergeable_anon_vma(struct anon_vma *anon_vma1, 7432dbf4010SVlastimil Babka struct anon_vma *anon_vma2, struct vm_area_struct *vma) 7441da177e4SLinus Torvalds { 745965f55deSShaohua Li /* 746965f55deSShaohua Li * The list_is_singular() test is to avoid merging VMA cloned from 747965f55deSShaohua Li * parents. This can improve scalability caused by anon_vma lock. 748965f55deSShaohua Li */ 749965f55deSShaohua Li if ((!anon_vma1 || !anon_vma2) && (!vma || 750965f55deSShaohua Li list_is_singular(&vma->anon_vma_chain))) 7512dbf4010SVlastimil Babka return true; 752965f55deSShaohua Li return anon_vma1 == anon_vma2; 7531da177e4SLinus Torvalds } 7541da177e4SLinus Torvalds 7551da177e4SLinus Torvalds /* 7561da177e4SLinus Torvalds * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff) 7571da177e4SLinus Torvalds * in front of (at a lower virtual address and file offset than) the vma. 7581da177e4SLinus Torvalds * 7591da177e4SLinus Torvalds * We cannot merge two vmas if they have differently assigned (non-NULL) 7601da177e4SLinus Torvalds * anon_vmas, nor if same anon_vma is assigned but offsets incompatible. 7611da177e4SLinus Torvalds * 7621da177e4SLinus Torvalds * We don't check here for the merged mmap wrapping around the end of pagecache 76345e55300SPeter Collingbourne * indices (16TB on ia32) because do_mmap() does not permit mmap's which 7641da177e4SLinus Torvalds * wrap, nor mmaps which cover the final page at index -1UL. 765714965caSVlastimil Babka * 766714965caSVlastimil Babka * We assume the vma may be removed as part of the merge. 7671da177e4SLinus Torvalds */ 7682dbf4010SVlastimil Babka static bool 7691da177e4SLinus Torvalds can_vma_merge_before(struct vm_area_struct *vma, unsigned long vm_flags, 77019a809afSAndrea Arcangeli struct anon_vma *anon_vma, struct file *file, 7712dbf4010SVlastimil Babka pgoff_t vm_pgoff, struct vm_userfaultfd_ctx vm_userfaultfd_ctx, 7725c26f6acSSuren Baghdasaryan struct anon_vma_name *anon_name) 7731da177e4SLinus Torvalds { 774714965caSVlastimil Babka if (is_mergeable_vma(vma, file, vm_flags, vm_userfaultfd_ctx, anon_name, true) && 775965f55deSShaohua Li is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) { 7761da177e4SLinus Torvalds if (vma->vm_pgoff == vm_pgoff) 7772dbf4010SVlastimil Babka return true; 7781da177e4SLinus Torvalds } 7792dbf4010SVlastimil Babka return false; 7801da177e4SLinus Torvalds } 7811da177e4SLinus Torvalds 7821da177e4SLinus Torvalds /* 7831da177e4SLinus Torvalds * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff) 7841da177e4SLinus Torvalds * beyond (at a higher virtual address and file offset than) the vma. 7851da177e4SLinus Torvalds * 7861da177e4SLinus Torvalds * We cannot merge two vmas if they have differently assigned (non-NULL) 7871da177e4SLinus Torvalds * anon_vmas, nor if same anon_vma is assigned but offsets incompatible. 788714965caSVlastimil Babka * 789714965caSVlastimil Babka * We assume that vma is not removed as part of the merge. 7901da177e4SLinus Torvalds */ 7912dbf4010SVlastimil Babka static bool 7921da177e4SLinus Torvalds can_vma_merge_after(struct vm_area_struct *vma, unsigned long vm_flags, 79319a809afSAndrea Arcangeli struct anon_vma *anon_vma, struct file *file, 7942dbf4010SVlastimil Babka pgoff_t vm_pgoff, struct vm_userfaultfd_ctx vm_userfaultfd_ctx, 7955c26f6acSSuren Baghdasaryan struct anon_vma_name *anon_name) 7961da177e4SLinus Torvalds { 797714965caSVlastimil Babka if (is_mergeable_vma(vma, file, vm_flags, vm_userfaultfd_ctx, anon_name, false) && 798965f55deSShaohua Li is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) { 7991da177e4SLinus Torvalds pgoff_t vm_pglen; 800d6e93217SLibin vm_pglen = vma_pages(vma); 8011da177e4SLinus Torvalds if (vma->vm_pgoff + vm_pglen == vm_pgoff) 8022dbf4010SVlastimil Babka return true; 8031da177e4SLinus Torvalds } 8042dbf4010SVlastimil Babka return false; 8051da177e4SLinus Torvalds } 8061da177e4SLinus Torvalds 8071da177e4SLinus Torvalds /* 8089a10064fSColin Cross * Given a mapping request (addr,end,vm_flags,file,pgoff,anon_name), 8099a10064fSColin Cross * figure out whether that can be merged with its predecessor or its 8109a10064fSColin Cross * successor. Or both (it neatly fills a hole). 8111da177e4SLinus Torvalds * 8121da177e4SLinus Torvalds * In most cases - when called for mmap, brk or mremap - [addr,end) is 8131da177e4SLinus Torvalds * certain not to be mapped by the time vma_merge is called; but when 8141da177e4SLinus Torvalds * called for mprotect, it is certain to be already mapped (either at 8151da177e4SLinus Torvalds * an offset within prev, or at the start of next), and the flags of 8161da177e4SLinus Torvalds * this area are about to be changed to vm_flags - and the no-change 8171da177e4SLinus Torvalds * case has already been eliminated. 8181da177e4SLinus Torvalds * 819fcfccd91SLorenzo Stoakes * The following mprotect cases have to be considered, where **** is 8201da177e4SLinus Torvalds * the area passed down from mprotect_fixup, never extending beyond one 821fcfccd91SLorenzo Stoakes * vma, PPPP is the previous vma, CCCC is a concurrent vma that starts 822fcfccd91SLorenzo Stoakes * at the same address as **** and is of the same or larger span, and 823fcfccd91SLorenzo Stoakes * NNNN the next vma after ****: 8241da177e4SLinus Torvalds * 825fcfccd91SLorenzo Stoakes * **** **** **** 826fcfccd91SLorenzo Stoakes * PPPPPPNNNNNN PPPPPPNNNNNN PPPPPPCCCCCC 8275d42ab29SWei Yang * cannot merge might become might become 828fcfccd91SLorenzo Stoakes * PPNNNNNNNNNN PPPPPPPPPPCC 8295d42ab29SWei Yang * mmap, brk or case 4 below case 5 below 8305d42ab29SWei Yang * mremap move: 831fcfccd91SLorenzo Stoakes * **** **** 832fcfccd91SLorenzo Stoakes * PPPP NNNN PPPPCCCCNNNN 8335d42ab29SWei Yang * might become might become 8345d42ab29SWei Yang * PPPPPPPPPPPP 1 or PPPPPPPPPPPP 6 or 835fcfccd91SLorenzo Stoakes * PPPPPPPPNNNN 2 or PPPPPPPPNNNN 7 or 836fcfccd91SLorenzo Stoakes * PPPPNNNNNNNN 3 PPPPNNNNNNNN 8 8371da177e4SLinus Torvalds * 838fcfccd91SLorenzo Stoakes * It is important for case 8 that the vma CCCC overlapping the 839fcfccd91SLorenzo Stoakes * region **** is never going to extended over NNNN. Instead NNNN must 840fcfccd91SLorenzo Stoakes * be extended in region **** and CCCC must be removed. This way in 8410503ea8fSLiam R. Howlett * all cases where vma_merge succeeds, the moment vma_merge drops the 842e86f15eeSAndrea Arcangeli * rmap_locks, the properties of the merged vma will be already 843e86f15eeSAndrea Arcangeli * correct for the whole merged range. Some of those properties like 844e86f15eeSAndrea Arcangeli * vm_page_prot/vm_flags may be accessed by rmap_walks and they must 845e86f15eeSAndrea Arcangeli * be correct for the whole merged range immediately after the 846fcfccd91SLorenzo Stoakes * rmap_locks are released. Otherwise if NNNN would be removed and 847fcfccd91SLorenzo Stoakes * CCCC would be extended over the NNNN range, remove_migration_ptes 848e86f15eeSAndrea Arcangeli * or other rmap walkers (if working on addresses beyond the "end" 849fcfccd91SLorenzo Stoakes * parameter) may establish ptes with the wrong permissions of CCCC 850fcfccd91SLorenzo Stoakes * instead of the right permissions of NNNN. 8510503ea8fSLiam R. Howlett * 8520503ea8fSLiam R. Howlett * In the code below: 8530503ea8fSLiam R. Howlett * PPPP is represented by *prev 854fcfccd91SLorenzo Stoakes * CCCC is represented by *curr or not represented at all (NULL) 855fcfccd91SLorenzo Stoakes * NNNN is represented by *next or not represented at all (NULL) 856fcfccd91SLorenzo Stoakes * **** is not represented - it will be merged and the vma containing the 8579e8a39d2SVlastimil Babka * area is returned, or the function will return NULL 8581da177e4SLinus Torvalds */ 8599760ebffSLiam R. Howlett struct vm_area_struct *vma_merge(struct vma_iterator *vmi, struct mm_struct *mm, 8601da177e4SLinus Torvalds struct vm_area_struct *prev, unsigned long addr, 8611da177e4SLinus Torvalds unsigned long end, unsigned long vm_flags, 8621da177e4SLinus Torvalds struct anon_vma *anon_vma, struct file *file, 86319a809afSAndrea Arcangeli pgoff_t pgoff, struct mempolicy *policy, 8649a10064fSColin Cross struct vm_userfaultfd_ctx vm_userfaultfd_ctx, 8655c26f6acSSuren Baghdasaryan struct anon_vma_name *anon_name) 8661da177e4SLinus Torvalds { 867b0729ae0SLorenzo Stoakes struct vm_area_struct *curr, *next, *res; 8680503ea8fSLiam R. Howlett struct vm_area_struct *vma, *adjust, *remove, *remove2; 8690173db4fSLorenzo Stoakes struct vma_prepare vp; 8700173db4fSLorenzo Stoakes pgoff_t vma_pgoff; 8710173db4fSLorenzo Stoakes int err = 0; 872eef19944SJakub Matěna bool merge_prev = false; 873eef19944SJakub Matěna bool merge_next = false; 8740503ea8fSLiam R. Howlett bool vma_expanded = false; 8750503ea8fSLiam R. Howlett unsigned long vma_start = addr; 8760173db4fSLorenzo Stoakes unsigned long vma_end = end; 8770173db4fSLorenzo Stoakes pgoff_t pglen = (end - addr) >> PAGE_SHIFT; 8780173db4fSLorenzo Stoakes long adj_start = 0; 8791da177e4SLinus Torvalds 8801da177e4SLinus Torvalds /* 8811da177e4SLinus Torvalds * We later require that vma->vm_flags == vm_flags, 8821da177e4SLinus Torvalds * so this tests vma->vm_flags & VM_SPECIAL, too. 8831da177e4SLinus Torvalds */ 8841da177e4SLinus Torvalds if (vm_flags & VM_SPECIAL) 8851da177e4SLinus Torvalds return NULL; 8861da177e4SLinus Torvalds 88700cd00a6SLorenzo Stoakes /* Does the input range span an existing VMA? (cases 5 - 8) */ 88800cd00a6SLorenzo Stoakes curr = find_vma_intersection(mm, prev ? prev->vm_end : 0, end); 8891da177e4SLinus Torvalds 89000cd00a6SLorenzo Stoakes if (!curr || /* cases 1 - 4 */ 89100cd00a6SLorenzo Stoakes end == curr->vm_end) /* cases 6 - 8, adjacent VMA */ 89200cd00a6SLorenzo Stoakes next = vma_lookup(mm, end); 89300cd00a6SLorenzo Stoakes else 89400cd00a6SLorenzo Stoakes next = NULL; /* case 5 */ 895e86f15eeSAndrea Arcangeli 8960503ea8fSLiam R. Howlett if (prev) { 8970503ea8fSLiam R. Howlett vma_start = prev->vm_start; 8980503ea8fSLiam R. Howlett vma_pgoff = prev->vm_pgoff; 8990173db4fSLorenzo Stoakes 900eef19944SJakub Matěna /* Can we merge the predecessor? */ 9010173db4fSLorenzo Stoakes if (addr == prev->vm_end && mpol_equal(vma_policy(prev), policy) 9020503ea8fSLiam R. Howlett && can_vma_merge_after(prev, vm_flags, anon_vma, file, 9030503ea8fSLiam R. Howlett pgoff, vm_userfaultfd_ctx, anon_name)) { 904eef19944SJakub Matěna merge_prev = true; 90518b098afSLiam R. Howlett vma_prev(vmi); 9061da177e4SLinus Torvalds } 9070503ea8fSLiam R. Howlett } 908b0729ae0SLorenzo Stoakes 909eef19944SJakub Matěna /* Can we merge the successor? */ 91000cd00a6SLorenzo Stoakes if (next && mpol_equal(policy, vma_policy(next)) && 9110173db4fSLorenzo Stoakes can_vma_merge_before(next, vm_flags, anon_vma, file, pgoff+pglen, 9129a10064fSColin Cross vm_userfaultfd_ctx, anon_name)) { 913eef19944SJakub Matěna merge_next = true; 914eef19944SJakub Matěna } 9150503ea8fSLiam R. Howlett 91629417d29SLorenzo Stoakes /* Verify some invariant that must be enforced by the caller. */ 91729417d29SLorenzo Stoakes VM_WARN_ON(prev && addr <= prev->vm_start); 91829417d29SLorenzo Stoakes VM_WARN_ON(curr && (addr != curr->vm_start || end > curr->vm_end)); 91929417d29SLorenzo Stoakes VM_WARN_ON(addr >= end); 92029417d29SLorenzo Stoakes 9210173db4fSLorenzo Stoakes if (!merge_prev && !merge_next) 9220173db4fSLorenzo Stoakes return NULL; /* Not mergeable. */ 9230173db4fSLorenzo Stoakes 9240173db4fSLorenzo Stoakes res = vma = prev; 9250503ea8fSLiam R. Howlett remove = remove2 = adjust = NULL; 9260173db4fSLorenzo Stoakes 927eef19944SJakub Matěna /* Can we merge both the predecessor and the successor? */ 928eef19944SJakub Matěna if (merge_prev && merge_next && 9290503ea8fSLiam R. Howlett is_mergeable_anon_vma(prev->anon_vma, next->anon_vma, NULL)) { 9305ff783f1SVlastimil Babka remove = next; /* case 1 */ 9310503ea8fSLiam R. Howlett vma_end = next->vm_end; 93250dac011SVlastimil Babka err = dup_anon_vma(prev, next); 933fcfccd91SLorenzo Stoakes if (curr) { /* case 6 */ 934fcfccd91SLorenzo Stoakes remove = curr; 9350503ea8fSLiam R. Howlett remove2 = next; 9365ff783f1SVlastimil Babka if (!next->anon_vma) 937fcfccd91SLorenzo Stoakes err = dup_anon_vma(prev, curr); 9380503ea8fSLiam R. Howlett } 9390173db4fSLorenzo Stoakes } else if (merge_prev) { /* case 2 */ 940fcfccd91SLorenzo Stoakes if (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 */ 950eef19944SJakub Matěna res = next; 9510503ea8fSLiam R. Howlett if (prev && addr < prev->vm_end) { /* case 4 */ 9520503ea8fSLiam R. Howlett vma_end = addr; 953183b7a60SVlastimil Babka adjust = next; 9541e76454fSVlastimil Babka adj_start = -(prev->vm_end - addr); 955183b7a60SVlastimil Babka err = dup_anon_vma(next, prev); 9560503ea8fSLiam R. Howlett } else { 957b0729ae0SLorenzo Stoakes /* 958b0729ae0SLorenzo Stoakes * Note that cases 3 and 8 are the ONLY ones where prev 959b0729ae0SLorenzo Stoakes * is permitted to be (but is not necessarily) NULL. 960b0729ae0SLorenzo Stoakes */ 9610503ea8fSLiam R. Howlett vma = next; /* case 3 */ 9620503ea8fSLiam R. Howlett vma_start = addr; 9630503ea8fSLiam R. Howlett vma_end = next->vm_end; 9647e775787SVlastimil Babka vma_pgoff = next->vm_pgoff - pglen; 965fcfccd91SLorenzo Stoakes if (curr) { /* case 8 */ 966fcfccd91SLorenzo Stoakes vma_pgoff = curr->vm_pgoff; 967fcfccd91SLorenzo Stoakes remove = curr; 968fcfccd91SLorenzo Stoakes err = dup_anon_vma(next, curr); 9690503ea8fSLiam R. Howlett } 9700503ea8fSLiam R. Howlett } 9711da177e4SLinus Torvalds } 9721da177e4SLinus Torvalds 9730173db4fSLorenzo Stoakes /* Error in anon_vma clone. */ 974eef19944SJakub Matěna if (err) 9751da177e4SLinus Torvalds return NULL; 9760503ea8fSLiam R. Howlett 977b5df0922SLiam R. Howlett if (vma_start < vma->vm_start || vma_end > vma->vm_end) 978b5df0922SLiam R. Howlett vma_expanded = true; 979b5df0922SLiam R. Howlett 980b5df0922SLiam R. Howlett if (vma_expanded) { 981b5df0922SLiam R. Howlett vma_iter_config(vmi, vma_start, vma_end); 982b5df0922SLiam R. Howlett } else { 983b5df0922SLiam R. Howlett vma_iter_config(vmi, adjust->vm_start + adj_start, 984b5df0922SLiam R. Howlett adjust->vm_end); 985b5df0922SLiam R. Howlett } 986b5df0922SLiam R. Howlett 987b5df0922SLiam R. Howlett if (vma_iter_prealloc(vmi, vma)) 9880503ea8fSLiam R. Howlett return NULL; 9890503ea8fSLiam R. Howlett 9900503ea8fSLiam R. Howlett init_multi_vma_prep(&vp, vma, adjust, remove, remove2); 9910503ea8fSLiam R. Howlett VM_WARN_ON(vp.anon_vma && adjust && adjust->anon_vma && 9920503ea8fSLiam R. Howlett vp.anon_vma != adjust->anon_vma); 9930503ea8fSLiam R. Howlett 9940503ea8fSLiam R. Howlett vma_prepare(&vp); 995ccf1d78dSSuren Baghdasaryan vma_adjust_trans_huge(vma, vma_start, vma_end, adj_start); 9960503ea8fSLiam R. Howlett 9970503ea8fSLiam R. Howlett vma->vm_start = vma_start; 9980503ea8fSLiam R. Howlett vma->vm_end = vma_end; 9990503ea8fSLiam R. Howlett vma->vm_pgoff = vma_pgoff; 10000503ea8fSLiam R. Howlett 10010503ea8fSLiam R. Howlett if (vma_expanded) 10020503ea8fSLiam R. Howlett vma_iter_store(vmi, vma); 10030503ea8fSLiam R. Howlett 10041e76454fSVlastimil Babka if (adj_start) { 10051e76454fSVlastimil Babka adjust->vm_start += adj_start; 10061e76454fSVlastimil Babka adjust->vm_pgoff += adj_start >> PAGE_SHIFT; 10071e76454fSVlastimil Babka if (adj_start < 0) { 10080503ea8fSLiam R. Howlett WARN_ON(vma_expanded); 10090503ea8fSLiam R. Howlett vma_iter_store(vmi, next); 10100503ea8fSLiam R. Howlett } 10110503ea8fSLiam R. Howlett } 10120503ea8fSLiam R. Howlett 10130503ea8fSLiam R. Howlett vma_complete(&vp, vmi, mm); 1014eef19944SJakub Matěna khugepaged_enter_vma(res, vm_flags); 10159760ebffSLiam R. Howlett return res; 1016f2ebfe43SLiam R. Howlett } 1017f2ebfe43SLiam R. Howlett 10181da177e4SLinus Torvalds /* 1019b4f315b4SEthon Paul * Rough compatibility check to quickly see if it's even worth looking 1020d0e9fe17SLinus Torvalds * at sharing an anon_vma. 1021d0e9fe17SLinus Torvalds * 1022d0e9fe17SLinus Torvalds * They need to have the same vm_file, and the flags can only differ 1023d0e9fe17SLinus Torvalds * in things that mprotect may change. 1024d0e9fe17SLinus Torvalds * 1025d0e9fe17SLinus Torvalds * NOTE! The fact that we share an anon_vma doesn't _have_ to mean that 1026d0e9fe17SLinus Torvalds * we can merge the two vma's. For example, we refuse to merge a vma if 1027d0e9fe17SLinus Torvalds * there is a vm_ops->close() function, because that indicates that the 1028d0e9fe17SLinus Torvalds * driver is doing some kind of reference counting. But that doesn't 1029d0e9fe17SLinus Torvalds * really matter for the anon_vma sharing case. 1030d0e9fe17SLinus Torvalds */ 1031d0e9fe17SLinus Torvalds static int anon_vma_compatible(struct vm_area_struct *a, struct vm_area_struct *b) 1032d0e9fe17SLinus Torvalds { 1033d0e9fe17SLinus Torvalds return a->vm_end == b->vm_start && 1034d0e9fe17SLinus Torvalds mpol_equal(vma_policy(a), vma_policy(b)) && 1035d0e9fe17SLinus Torvalds a->vm_file == b->vm_file && 10366cb4d9a2SAnshuman Khandual !((a->vm_flags ^ b->vm_flags) & ~(VM_ACCESS_FLAGS | VM_SOFTDIRTY)) && 1037d0e9fe17SLinus Torvalds b->vm_pgoff == a->vm_pgoff + ((b->vm_start - a->vm_start) >> PAGE_SHIFT); 1038d0e9fe17SLinus Torvalds } 1039d0e9fe17SLinus Torvalds 1040d0e9fe17SLinus Torvalds /* 1041d0e9fe17SLinus Torvalds * Do some basic sanity checking to see if we can re-use the anon_vma 1042d0e9fe17SLinus Torvalds * from 'old'. The 'a'/'b' vma's are in VM order - one of them will be 1043d0e9fe17SLinus Torvalds * the same as 'old', the other will be the new one that is trying 1044d0e9fe17SLinus Torvalds * to share the anon_vma. 1045d0e9fe17SLinus Torvalds * 10465b449489SFlorian Rommel * NOTE! This runs with mmap_lock held for reading, so it is possible that 1047d0e9fe17SLinus Torvalds * the anon_vma of 'old' is concurrently in the process of being set up 1048d0e9fe17SLinus Torvalds * by another page fault trying to merge _that_. But that's ok: if it 1049d0e9fe17SLinus Torvalds * is being set up, that automatically means that it will be a singleton 1050d0e9fe17SLinus Torvalds * acceptable for merging, so we can do all of this optimistically. But 10514db0c3c2SJason Low * we do that READ_ONCE() to make sure that we never re-load the pointer. 1052d0e9fe17SLinus Torvalds * 1053d0e9fe17SLinus Torvalds * IOW: that the "list_is_singular()" test on the anon_vma_chain only 1054d0e9fe17SLinus Torvalds * matters for the 'stable anon_vma' case (ie the thing we want to avoid 1055d0e9fe17SLinus Torvalds * is to return an anon_vma that is "complex" due to having gone through 1056d0e9fe17SLinus Torvalds * a fork). 1057d0e9fe17SLinus Torvalds * 1058d0e9fe17SLinus Torvalds * We also make sure that the two vma's are compatible (adjacent, 1059d0e9fe17SLinus Torvalds * and with the same memory policies). That's all stable, even with just 10605b449489SFlorian Rommel * a read lock on the mmap_lock. 1061d0e9fe17SLinus Torvalds */ 1062d0e9fe17SLinus Torvalds static struct anon_vma *reusable_anon_vma(struct vm_area_struct *old, struct vm_area_struct *a, struct vm_area_struct *b) 1063d0e9fe17SLinus Torvalds { 1064d0e9fe17SLinus Torvalds if (anon_vma_compatible(a, b)) { 10654db0c3c2SJason Low struct anon_vma *anon_vma = READ_ONCE(old->anon_vma); 1066d0e9fe17SLinus Torvalds 1067d0e9fe17SLinus Torvalds if (anon_vma && list_is_singular(&old->anon_vma_chain)) 1068d0e9fe17SLinus Torvalds return anon_vma; 1069d0e9fe17SLinus Torvalds } 1070d0e9fe17SLinus Torvalds return NULL; 1071d0e9fe17SLinus Torvalds } 1072d0e9fe17SLinus Torvalds 1073d0e9fe17SLinus Torvalds /* 10741da177e4SLinus Torvalds * find_mergeable_anon_vma is used by anon_vma_prepare, to check 10751da177e4SLinus Torvalds * neighbouring vmas for a suitable anon_vma, before it goes off 10761da177e4SLinus Torvalds * to allocate a new anon_vma. It checks because a repetitive 10771da177e4SLinus Torvalds * sequence of mprotects and faults may otherwise lead to distinct 10781da177e4SLinus Torvalds * anon_vmas being allocated, preventing vma merge in subsequent 10791da177e4SLinus Torvalds * mprotect. 10801da177e4SLinus Torvalds */ 10811da177e4SLinus Torvalds struct anon_vma *find_mergeable_anon_vma(struct vm_area_struct *vma) 10821da177e4SLinus Torvalds { 1083763ecb03SLiam R. Howlett MA_STATE(mas, &vma->vm_mm->mm_mt, vma->vm_end, vma->vm_end); 1084a67c8caaSMiaohe Lin struct anon_vma *anon_vma = NULL; 1085763ecb03SLiam R. Howlett struct vm_area_struct *prev, *next; 10861da177e4SLinus Torvalds 1087a67c8caaSMiaohe Lin /* Try next first. */ 1088763ecb03SLiam R. Howlett next = mas_walk(&mas); 1089763ecb03SLiam R. Howlett if (next) { 1090763ecb03SLiam R. Howlett anon_vma = reusable_anon_vma(next, vma, next); 1091d0e9fe17SLinus Torvalds if (anon_vma) 1092d0e9fe17SLinus Torvalds return anon_vma; 1093a67c8caaSMiaohe Lin } 10941da177e4SLinus Torvalds 1095763ecb03SLiam R. Howlett prev = mas_prev(&mas, 0); 1096763ecb03SLiam R. Howlett VM_BUG_ON_VMA(prev != vma, vma); 1097763ecb03SLiam R. Howlett prev = mas_prev(&mas, 0); 1098a67c8caaSMiaohe Lin /* Try prev next. */ 1099763ecb03SLiam R. Howlett if (prev) 1100763ecb03SLiam R. Howlett anon_vma = reusable_anon_vma(prev, prev, vma); 1101a67c8caaSMiaohe Lin 11021da177e4SLinus Torvalds /* 1103a67c8caaSMiaohe Lin * We might reach here with anon_vma == NULL if we can't find 1104a67c8caaSMiaohe Lin * any reusable anon_vma. 11051da177e4SLinus Torvalds * There's no absolute need to look only at touching neighbours: 11061da177e4SLinus Torvalds * we could search further afield for "compatible" anon_vmas. 11071da177e4SLinus Torvalds * But it would probably just be a waste of time searching, 11081da177e4SLinus Torvalds * or lead to too many vmas hanging off the same anon_vma. 11091da177e4SLinus Torvalds * We're trying to allow mprotect remerging later on, 11101da177e4SLinus Torvalds * not trying to minimize memory used for anon_vmas. 11111da177e4SLinus Torvalds */ 1112a67c8caaSMiaohe Lin return anon_vma; 11131da177e4SLinus Torvalds } 11141da177e4SLinus Torvalds 11151da177e4SLinus Torvalds /* 111640401530SAl Viro * If a hint addr is less than mmap_min_addr change hint to be as 111740401530SAl Viro * low as possible but still greater than mmap_min_addr 111840401530SAl Viro */ 111940401530SAl Viro static inline unsigned long round_hint_to_min(unsigned long hint) 112040401530SAl Viro { 112140401530SAl Viro hint &= PAGE_MASK; 112240401530SAl Viro if (((void *)hint != NULL) && 112340401530SAl Viro (hint < mmap_min_addr)) 112440401530SAl Viro return PAGE_ALIGN(mmap_min_addr); 112540401530SAl Viro return hint; 112640401530SAl Viro } 112740401530SAl Viro 1128b0cc5e89SAndrew Morton bool mlock_future_ok(struct mm_struct *mm, unsigned long flags, 11293c54a298SLorenzo Stoakes unsigned long bytes) 1130363ee17fSDavidlohr Bueso { 11313c54a298SLorenzo Stoakes unsigned long locked_pages, limit_pages; 1132363ee17fSDavidlohr Bueso 11333c54a298SLorenzo Stoakes if (!(flags & VM_LOCKED) || capable(CAP_IPC_LOCK)) 11343c54a298SLorenzo Stoakes return true; 11353c54a298SLorenzo Stoakes 11363c54a298SLorenzo Stoakes locked_pages = bytes >> PAGE_SHIFT; 11373c54a298SLorenzo Stoakes locked_pages += mm->locked_vm; 11383c54a298SLorenzo Stoakes 11393c54a298SLorenzo Stoakes limit_pages = rlimit(RLIMIT_MEMLOCK); 11403c54a298SLorenzo Stoakes limit_pages >>= PAGE_SHIFT; 11413c54a298SLorenzo Stoakes 11423c54a298SLorenzo Stoakes return locked_pages <= limit_pages; 1143363ee17fSDavidlohr Bueso } 1144363ee17fSDavidlohr Bueso 1145be83bbf8SLinus Torvalds static inline u64 file_mmap_size_max(struct file *file, struct inode *inode) 1146be83bbf8SLinus Torvalds { 1147be83bbf8SLinus Torvalds if (S_ISREG(inode->i_mode)) 1148423913adSLinus Torvalds return MAX_LFS_FILESIZE; 1149be83bbf8SLinus Torvalds 1150be83bbf8SLinus Torvalds if (S_ISBLK(inode->i_mode)) 1151be83bbf8SLinus Torvalds return MAX_LFS_FILESIZE; 1152be83bbf8SLinus Torvalds 115376f34950SIvan Khoronzhuk if (S_ISSOCK(inode->i_mode)) 115476f34950SIvan Khoronzhuk return MAX_LFS_FILESIZE; 115576f34950SIvan Khoronzhuk 1156be83bbf8SLinus Torvalds /* Special "we do even unsigned file positions" case */ 1157be83bbf8SLinus Torvalds if (file->f_mode & FMODE_UNSIGNED_OFFSET) 1158be83bbf8SLinus Torvalds return 0; 1159be83bbf8SLinus Torvalds 1160be83bbf8SLinus Torvalds /* Yes, random drivers might want more. But I'm tired of buggy drivers */ 1161be83bbf8SLinus Torvalds return ULONG_MAX; 1162be83bbf8SLinus Torvalds } 1163be83bbf8SLinus Torvalds 1164be83bbf8SLinus Torvalds static inline bool file_mmap_ok(struct file *file, struct inode *inode, 1165be83bbf8SLinus Torvalds unsigned long pgoff, unsigned long len) 1166be83bbf8SLinus Torvalds { 1167be83bbf8SLinus Torvalds u64 maxsize = file_mmap_size_max(file, inode); 1168be83bbf8SLinus Torvalds 1169be83bbf8SLinus Torvalds if (maxsize && len > maxsize) 1170be83bbf8SLinus Torvalds return false; 1171be83bbf8SLinus Torvalds maxsize -= len; 1172be83bbf8SLinus Torvalds if (pgoff > maxsize >> PAGE_SHIFT) 1173be83bbf8SLinus Torvalds return false; 1174be83bbf8SLinus Torvalds return true; 1175be83bbf8SLinus Torvalds } 1176be83bbf8SLinus Torvalds 117740401530SAl Viro /* 11783e4e28c5SMichel Lespinasse * The caller must write-lock current->mm->mmap_lock. 11791da177e4SLinus Torvalds */ 11801fcfd8dbSOleg Nesterov unsigned long do_mmap(struct file *file, unsigned long addr, 11811da177e4SLinus Torvalds unsigned long len, unsigned long prot, 118245e55300SPeter Collingbourne unsigned long flags, unsigned long pgoff, 118345e55300SPeter Collingbourne unsigned long *populate, struct list_head *uf) 11841da177e4SLinus Torvalds { 11851da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 118645e55300SPeter Collingbourne vm_flags_t vm_flags; 118762b5f7d0SDave Hansen int pkey = 0; 11881da177e4SLinus Torvalds 118941badc15SMichel Lespinasse *populate = 0; 1190bebeb3d6SMichel Lespinasse 1191e37609bbSPiotr Kwapulinski if (!len) 1192e37609bbSPiotr Kwapulinski return -EINVAL; 1193e37609bbSPiotr Kwapulinski 11941da177e4SLinus Torvalds /* 11951da177e4SLinus Torvalds * Does the application expect PROT_READ to imply PROT_EXEC? 11961da177e4SLinus Torvalds * 11971da177e4SLinus Torvalds * (the exception is when the underlying filesystem is noexec 11981da177e4SLinus Torvalds * mounted, in which case we dont add PROT_EXEC.) 11991da177e4SLinus Torvalds */ 12001da177e4SLinus Torvalds if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC)) 120190f8572bSEric W. Biederman if (!(file && path_noexec(&file->f_path))) 12021da177e4SLinus Torvalds prot |= PROT_EXEC; 12031da177e4SLinus Torvalds 1204a4ff8e86SMichal Hocko /* force arch specific MAP_FIXED handling in get_unmapped_area */ 1205a4ff8e86SMichal Hocko if (flags & MAP_FIXED_NOREPLACE) 1206a4ff8e86SMichal Hocko flags |= MAP_FIXED; 1207a4ff8e86SMichal Hocko 12087cd94146SEric Paris if (!(flags & MAP_FIXED)) 12097cd94146SEric Paris addr = round_hint_to_min(addr); 12107cd94146SEric Paris 12111da177e4SLinus Torvalds /* Careful about overflows.. */ 12121da177e4SLinus Torvalds len = PAGE_ALIGN(len); 12139206de95SAl Viro if (!len) 12141da177e4SLinus Torvalds return -ENOMEM; 12151da177e4SLinus Torvalds 12161da177e4SLinus Torvalds /* offset overflow? */ 12171da177e4SLinus Torvalds if ((pgoff + (len >> PAGE_SHIFT)) < pgoff) 12181da177e4SLinus Torvalds return -EOVERFLOW; 12191da177e4SLinus Torvalds 12201da177e4SLinus Torvalds /* Too many mappings? */ 12211da177e4SLinus Torvalds if (mm->map_count > sysctl_max_map_count) 12221da177e4SLinus Torvalds return -ENOMEM; 12231da177e4SLinus Torvalds 12241da177e4SLinus Torvalds /* Obtain the address to map to. we verify (or select) it and ensure 12251da177e4SLinus Torvalds * that it represents a valid section of the address space. 12261da177e4SLinus Torvalds */ 12271da177e4SLinus Torvalds addr = get_unmapped_area(file, addr, len, pgoff, flags); 1228ff68dac6SGaowei Pu if (IS_ERR_VALUE(addr)) 12291da177e4SLinus Torvalds return addr; 12301da177e4SLinus Torvalds 1231a4ff8e86SMichal Hocko if (flags & MAP_FIXED_NOREPLACE) { 123235e43c5fSLiam Howlett if (find_vma_intersection(mm, addr, addr + len)) 1233a4ff8e86SMichal Hocko return -EEXIST; 1234a4ff8e86SMichal Hocko } 1235a4ff8e86SMichal Hocko 123662b5f7d0SDave Hansen if (prot == PROT_EXEC) { 123762b5f7d0SDave Hansen pkey = execute_only_pkey(mm); 123862b5f7d0SDave Hansen if (pkey < 0) 123962b5f7d0SDave Hansen pkey = 0; 124062b5f7d0SDave Hansen } 124162b5f7d0SDave Hansen 12421da177e4SLinus Torvalds /* Do simple checking here so the lower-level routines won't have 12431da177e4SLinus Torvalds * to. we assume access permissions have been handled by the open 12441da177e4SLinus Torvalds * of the memory object, so we don't do any here. 12451da177e4SLinus Torvalds */ 124645e55300SPeter Collingbourne vm_flags = calc_vm_prot_bits(prot, pkey) | calc_vm_flag_bits(flags) | 12471da177e4SLinus Torvalds mm->def_flags | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC; 12481da177e4SLinus Torvalds 1249cdf7b341SHuang Shijie if (flags & MAP_LOCKED) 12501da177e4SLinus Torvalds if (!can_do_mlock()) 12511da177e4SLinus Torvalds return -EPERM; 1252ba470de4SRik van Riel 1253b0cc5e89SAndrew Morton if (!mlock_future_ok(mm, vm_flags, len)) 12541da177e4SLinus Torvalds return -EAGAIN; 12551da177e4SLinus Torvalds 12561da177e4SLinus Torvalds if (file) { 1257077bf22bSOleg Nesterov struct inode *inode = file_inode(file); 12581c972597SDan Williams unsigned long flags_mask; 12591c972597SDan Williams 1260be83bbf8SLinus Torvalds if (!file_mmap_ok(file, inode, pgoff, len)) 1261be83bbf8SLinus Torvalds return -EOVERFLOW; 1262be83bbf8SLinus Torvalds 12631c972597SDan Williams flags_mask = LEGACY_MAP_MASK | file->f_op->mmap_supported_flags; 1264077bf22bSOleg Nesterov 12651da177e4SLinus Torvalds switch (flags & MAP_TYPE) { 12661da177e4SLinus Torvalds case MAP_SHARED: 12671c972597SDan Williams /* 12681c972597SDan Williams * Force use of MAP_SHARED_VALIDATE with non-legacy 12691c972597SDan Williams * flags. E.g. MAP_SYNC is dangerous to use with 12701c972597SDan Williams * MAP_SHARED as you don't know which consistency model 12711c972597SDan Williams * you will get. We silently ignore unsupported flags 12721c972597SDan Williams * with MAP_SHARED to preserve backward compatibility. 12731c972597SDan Williams */ 12741c972597SDan Williams flags &= LEGACY_MAP_MASK; 1275e4a9bc58SJoe Perches fallthrough; 12761c972597SDan Williams case MAP_SHARED_VALIDATE: 12771c972597SDan Williams if (flags & ~flags_mask) 12781c972597SDan Williams return -EOPNOTSUPP; 1279dc617f29SDarrick J. Wong if (prot & PROT_WRITE) { 1280dc617f29SDarrick J. Wong if (!(file->f_mode & FMODE_WRITE)) 12811da177e4SLinus Torvalds return -EACCES; 1282dc617f29SDarrick J. Wong if (IS_SWAPFILE(file->f_mapping->host)) 1283dc617f29SDarrick J. Wong return -ETXTBSY; 1284dc617f29SDarrick J. Wong } 12851da177e4SLinus Torvalds 12861da177e4SLinus Torvalds /* 12871da177e4SLinus Torvalds * Make sure we don't allow writing to an append-only 12881da177e4SLinus Torvalds * file.. 12891da177e4SLinus Torvalds */ 12901da177e4SLinus Torvalds if (IS_APPEND(inode) && (file->f_mode & FMODE_WRITE)) 12911da177e4SLinus Torvalds return -EACCES; 12921da177e4SLinus Torvalds 12931da177e4SLinus Torvalds vm_flags |= VM_SHARED | VM_MAYSHARE; 12941da177e4SLinus Torvalds if (!(file->f_mode & FMODE_WRITE)) 12951da177e4SLinus Torvalds vm_flags &= ~(VM_MAYWRITE | VM_SHARED); 1296e4a9bc58SJoe Perches fallthrough; 12971da177e4SLinus Torvalds case MAP_PRIVATE: 12981da177e4SLinus Torvalds if (!(file->f_mode & FMODE_READ)) 12991da177e4SLinus Torvalds return -EACCES; 130090f8572bSEric W. Biederman if (path_noexec(&file->f_path)) { 130180c5606cSLinus Torvalds if (vm_flags & VM_EXEC) 130280c5606cSLinus Torvalds return -EPERM; 130380c5606cSLinus Torvalds vm_flags &= ~VM_MAYEXEC; 130480c5606cSLinus Torvalds } 130580c5606cSLinus Torvalds 130672c2d531SAl Viro if (!file->f_op->mmap) 130780c5606cSLinus Torvalds return -ENODEV; 1308b2c56e4fSOleg Nesterov if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP)) 1309b2c56e4fSOleg Nesterov return -EINVAL; 13101da177e4SLinus Torvalds break; 13111da177e4SLinus Torvalds 13121da177e4SLinus Torvalds default: 13131da177e4SLinus Torvalds return -EINVAL; 13141da177e4SLinus Torvalds } 13151da177e4SLinus Torvalds } else { 13161da177e4SLinus Torvalds switch (flags & MAP_TYPE) { 13171da177e4SLinus Torvalds case MAP_SHARED: 1318b2c56e4fSOleg Nesterov if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP)) 1319b2c56e4fSOleg Nesterov return -EINVAL; 1320ce363942STejun Heo /* 1321ce363942STejun Heo * Ignore pgoff. 1322ce363942STejun Heo */ 1323ce363942STejun Heo pgoff = 0; 13241da177e4SLinus Torvalds vm_flags |= VM_SHARED | VM_MAYSHARE; 13251da177e4SLinus Torvalds break; 13261da177e4SLinus Torvalds case MAP_PRIVATE: 13271da177e4SLinus Torvalds /* 13281da177e4SLinus Torvalds * Set pgoff according to addr for anon_vma. 13291da177e4SLinus Torvalds */ 13301da177e4SLinus Torvalds pgoff = addr >> PAGE_SHIFT; 13311da177e4SLinus Torvalds break; 13321da177e4SLinus Torvalds default: 13331da177e4SLinus Torvalds return -EINVAL; 13341da177e4SLinus Torvalds } 13351da177e4SLinus Torvalds } 13361da177e4SLinus Torvalds 1337c22c0d63SMichel Lespinasse /* 1338c22c0d63SMichel Lespinasse * Set 'VM_NORESERVE' if we should not account for the 1339c22c0d63SMichel Lespinasse * memory use of this mapping. 1340c22c0d63SMichel Lespinasse */ 1341c22c0d63SMichel Lespinasse if (flags & MAP_NORESERVE) { 1342c22c0d63SMichel Lespinasse /* We honor MAP_NORESERVE if allowed to overcommit */ 1343c22c0d63SMichel Lespinasse if (sysctl_overcommit_memory != OVERCOMMIT_NEVER) 1344c22c0d63SMichel Lespinasse vm_flags |= VM_NORESERVE; 1345c22c0d63SMichel Lespinasse 1346c22c0d63SMichel Lespinasse /* hugetlb applies strict overcommit unless MAP_NORESERVE */ 1347c22c0d63SMichel Lespinasse if (file && is_file_hugepages(file)) 1348c22c0d63SMichel Lespinasse vm_flags |= VM_NORESERVE; 1349c22c0d63SMichel Lespinasse } 1350c22c0d63SMichel Lespinasse 1351897ab3e0SMike Rapoport addr = mmap_region(file, addr, len, vm_flags, pgoff, uf); 135209a9f1d2SMichel Lespinasse if (!IS_ERR_VALUE(addr) && 135309a9f1d2SMichel Lespinasse ((vm_flags & VM_LOCKED) || 135409a9f1d2SMichel Lespinasse (flags & (MAP_POPULATE | MAP_NONBLOCK)) == MAP_POPULATE)) 135541badc15SMichel Lespinasse *populate = len; 1356bebeb3d6SMichel Lespinasse return addr; 13570165ab44SMiklos Szeredi } 13586be5ceb0SLinus Torvalds 1359a90f590aSDominik Brodowski unsigned long ksys_mmap_pgoff(unsigned long addr, unsigned long len, 1360a90f590aSDominik Brodowski unsigned long prot, unsigned long flags, 1361a90f590aSDominik Brodowski unsigned long fd, unsigned long pgoff) 136266f0dc48SHugh Dickins { 136366f0dc48SHugh Dickins struct file *file = NULL; 13641e3ee14bSChen Gang unsigned long retval; 136566f0dc48SHugh Dickins 136666f0dc48SHugh Dickins if (!(flags & MAP_ANONYMOUS)) { 1367120a795dSAl Viro audit_mmap_fd(fd, flags); 136866f0dc48SHugh Dickins file = fget(fd); 136966f0dc48SHugh Dickins if (!file) 13701e3ee14bSChen Gang return -EBADF; 13717bba8f0eSZhen Lei if (is_file_hugepages(file)) { 1372af73e4d9SNaoya Horiguchi len = ALIGN(len, huge_page_size(hstate_file(file))); 13737bba8f0eSZhen Lei } else if (unlikely(flags & MAP_HUGETLB)) { 1374493af578SJörn Engel retval = -EINVAL; 1375493af578SJörn Engel goto out_fput; 13767bba8f0eSZhen Lei } 137766f0dc48SHugh Dickins } else if (flags & MAP_HUGETLB) { 1378c103a4dcSAndrew Morton struct hstate *hs; 1379af73e4d9SNaoya Horiguchi 138020ac2893SAnshuman Khandual hs = hstate_sizelog((flags >> MAP_HUGE_SHIFT) & MAP_HUGE_MASK); 1381091d0d55SLi Zefan if (!hs) 1382091d0d55SLi Zefan return -EINVAL; 1383091d0d55SLi Zefan 1384091d0d55SLi Zefan len = ALIGN(len, huge_page_size(hs)); 138566f0dc48SHugh Dickins /* 138666f0dc48SHugh Dickins * VM_NORESERVE is used because the reservations will be 138766f0dc48SHugh Dickins * taken when vm_ops->mmap() is called 138866f0dc48SHugh Dickins */ 1389af73e4d9SNaoya Horiguchi file = hugetlb_file_setup(HUGETLB_ANON_FILE, len, 139042d7395fSAndi Kleen VM_NORESERVE, 139183c1fd76Szhangyiru HUGETLB_ANONHUGE_INODE, 139242d7395fSAndi Kleen (flags >> MAP_HUGE_SHIFT) & MAP_HUGE_MASK); 139366f0dc48SHugh Dickins if (IS_ERR(file)) 139466f0dc48SHugh Dickins return PTR_ERR(file); 139566f0dc48SHugh Dickins } 139666f0dc48SHugh Dickins 13979fbeb5abSMichal Hocko retval = vm_mmap_pgoff(file, addr, len, prot, flags, pgoff); 1398493af578SJörn Engel out_fput: 139966f0dc48SHugh Dickins if (file) 140066f0dc48SHugh Dickins fput(file); 140166f0dc48SHugh Dickins return retval; 140266f0dc48SHugh Dickins } 140366f0dc48SHugh Dickins 1404a90f590aSDominik Brodowski SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len, 1405a90f590aSDominik Brodowski unsigned long, prot, unsigned long, flags, 1406a90f590aSDominik Brodowski unsigned long, fd, unsigned long, pgoff) 1407a90f590aSDominik Brodowski { 1408a90f590aSDominik Brodowski return ksys_mmap_pgoff(addr, len, prot, flags, fd, pgoff); 1409a90f590aSDominik Brodowski } 1410a90f590aSDominik Brodowski 1411a4679373SChristoph Hellwig #ifdef __ARCH_WANT_SYS_OLD_MMAP 1412a4679373SChristoph Hellwig struct mmap_arg_struct { 1413a4679373SChristoph Hellwig unsigned long addr; 1414a4679373SChristoph Hellwig unsigned long len; 1415a4679373SChristoph Hellwig unsigned long prot; 1416a4679373SChristoph Hellwig unsigned long flags; 1417a4679373SChristoph Hellwig unsigned long fd; 1418a4679373SChristoph Hellwig unsigned long offset; 1419a4679373SChristoph Hellwig }; 1420a4679373SChristoph Hellwig 1421a4679373SChristoph Hellwig SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg) 1422a4679373SChristoph Hellwig { 1423a4679373SChristoph Hellwig struct mmap_arg_struct a; 1424a4679373SChristoph Hellwig 1425a4679373SChristoph Hellwig if (copy_from_user(&a, arg, sizeof(a))) 1426a4679373SChristoph Hellwig return -EFAULT; 1427de1741a1SAlexander Kuleshov if (offset_in_page(a.offset)) 1428a4679373SChristoph Hellwig return -EINVAL; 1429a4679373SChristoph Hellwig 1430a90f590aSDominik Brodowski return ksys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd, 1431a4679373SChristoph Hellwig a.offset >> PAGE_SHIFT); 1432a4679373SChristoph Hellwig } 1433a4679373SChristoph Hellwig #endif /* __ARCH_WANT_SYS_OLD_MMAP */ 1434a4679373SChristoph Hellwig 143554cbbbf3SLorenzo Stoakes static bool vm_ops_needs_writenotify(const struct vm_operations_struct *vm_ops) 143654cbbbf3SLorenzo Stoakes { 143754cbbbf3SLorenzo Stoakes return vm_ops && (vm_ops->page_mkwrite || vm_ops->pfn_mkwrite); 143854cbbbf3SLorenzo Stoakes } 143954cbbbf3SLorenzo Stoakes 144054cbbbf3SLorenzo Stoakes static bool vma_is_shared_writable(struct vm_area_struct *vma) 144154cbbbf3SLorenzo Stoakes { 144254cbbbf3SLorenzo Stoakes return (vma->vm_flags & (VM_WRITE | VM_SHARED)) == 144354cbbbf3SLorenzo Stoakes (VM_WRITE | VM_SHARED); 144454cbbbf3SLorenzo Stoakes } 144554cbbbf3SLorenzo Stoakes 144654cbbbf3SLorenzo Stoakes static bool vma_fs_can_writeback(struct vm_area_struct *vma) 144754cbbbf3SLorenzo Stoakes { 144854cbbbf3SLorenzo Stoakes /* No managed pages to writeback. */ 144954cbbbf3SLorenzo Stoakes if (vma->vm_flags & VM_PFNMAP) 145054cbbbf3SLorenzo Stoakes return false; 145154cbbbf3SLorenzo Stoakes 145254cbbbf3SLorenzo Stoakes return vma->vm_file && vma->vm_file->f_mapping && 145354cbbbf3SLorenzo Stoakes mapping_can_writeback(vma->vm_file->f_mapping); 145454cbbbf3SLorenzo Stoakes } 145554cbbbf3SLorenzo Stoakes 145654cbbbf3SLorenzo Stoakes /* 145754cbbbf3SLorenzo Stoakes * Does this VMA require the underlying folios to have their dirty state 145854cbbbf3SLorenzo Stoakes * tracked? 145954cbbbf3SLorenzo Stoakes */ 146054cbbbf3SLorenzo Stoakes bool vma_needs_dirty_tracking(struct vm_area_struct *vma) 146154cbbbf3SLorenzo Stoakes { 146254cbbbf3SLorenzo Stoakes /* Only shared, writable VMAs require dirty tracking. */ 146354cbbbf3SLorenzo Stoakes if (!vma_is_shared_writable(vma)) 146454cbbbf3SLorenzo Stoakes return false; 146554cbbbf3SLorenzo Stoakes 146654cbbbf3SLorenzo Stoakes /* Does the filesystem need to be notified? */ 146754cbbbf3SLorenzo Stoakes if (vm_ops_needs_writenotify(vma->vm_ops)) 146854cbbbf3SLorenzo Stoakes return true; 146954cbbbf3SLorenzo Stoakes 147054cbbbf3SLorenzo Stoakes /* 147154cbbbf3SLorenzo Stoakes * Even if the filesystem doesn't indicate a need for writenotify, if it 147254cbbbf3SLorenzo Stoakes * can writeback, dirty tracking is still required. 147354cbbbf3SLorenzo Stoakes */ 147454cbbbf3SLorenzo Stoakes return vma_fs_can_writeback(vma); 147554cbbbf3SLorenzo Stoakes } 147654cbbbf3SLorenzo Stoakes 14774e950f6fSAlexey Dobriyan /* 14788bb4e7a2SWei Yang * Some shared mappings will want the pages marked read-only 14794e950f6fSAlexey Dobriyan * to track write events. If so, we'll downgrade vm_page_prot 14804e950f6fSAlexey Dobriyan * to the private version (using protection_map[] without the 14814e950f6fSAlexey Dobriyan * VM_SHARED bit). 14824e950f6fSAlexey Dobriyan */ 14836d2329f8SAndrea Arcangeli int vma_wants_writenotify(struct vm_area_struct *vma, pgprot_t vm_page_prot) 14844e950f6fSAlexey Dobriyan { 14854e950f6fSAlexey Dobriyan /* If it was private or non-writable, the write bit is already clear */ 148654cbbbf3SLorenzo Stoakes if (!vma_is_shared_writable(vma)) 14874e950f6fSAlexey Dobriyan return 0; 14884e950f6fSAlexey Dobriyan 14894e950f6fSAlexey Dobriyan /* The backer wishes to know when pages are first written to? */ 149054cbbbf3SLorenzo Stoakes if (vm_ops_needs_writenotify(vma->vm_ops)) 14914e950f6fSAlexey Dobriyan return 1; 14924e950f6fSAlexey Dobriyan 149364e45507SPeter Feiner /* The open routine did something to the protections that pgprot_modify 149464e45507SPeter Feiner * won't preserve? */ 14956d2329f8SAndrea Arcangeli if (pgprot_val(vm_page_prot) != 149654cbbbf3SLorenzo Stoakes pgprot_val(vm_pgprot_modify(vm_page_prot, vma->vm_flags))) 14974e950f6fSAlexey Dobriyan return 0; 14984e950f6fSAlexey Dobriyan 1499f96f7a40SDavid Hildenbrand /* 1500f96f7a40SDavid Hildenbrand * Do we need to track softdirty? hugetlb does not support softdirty 1501f96f7a40SDavid Hildenbrand * tracking yet. 1502f96f7a40SDavid Hildenbrand */ 1503f96f7a40SDavid Hildenbrand if (vma_soft_dirty_enabled(vma) && !is_vm_hugetlb_page(vma)) 150464e45507SPeter Feiner return 1; 150564e45507SPeter Feiner 150651d3d5ebSDavid Hildenbrand /* Do we need write faults for uffd-wp tracking? */ 150751d3d5ebSDavid Hildenbrand if (userfaultfd_wp(vma)) 150851d3d5ebSDavid Hildenbrand return 1; 150951d3d5ebSDavid Hildenbrand 15104e950f6fSAlexey Dobriyan /* Can the mapping track the dirty pages? */ 151154cbbbf3SLorenzo Stoakes return vma_fs_can_writeback(vma); 15124e950f6fSAlexey Dobriyan } 15134e950f6fSAlexey Dobriyan 1514fc8744adSLinus Torvalds /* 1515fc8744adSLinus Torvalds * We account for memory if it's a private writeable mapping, 15165a6fe125SMel Gorman * not hugepages and VM_NORESERVE wasn't set. 1517fc8744adSLinus Torvalds */ 1518ca16d140SKOSAKI Motohiro static inline int accountable_mapping(struct file *file, vm_flags_t vm_flags) 1519fc8744adSLinus Torvalds { 15205a6fe125SMel Gorman /* 15215a6fe125SMel Gorman * hugetlb has its own accounting separate from the core VM 15225a6fe125SMel Gorman * VM_HUGETLB may not be set yet so we cannot check for that flag. 15235a6fe125SMel Gorman */ 15245a6fe125SMel Gorman if (file && is_file_hugepages(file)) 15255a6fe125SMel Gorman return 0; 15265a6fe125SMel Gorman 1527fc8744adSLinus Torvalds return (vm_flags & (VM_NORESERVE | VM_SHARED | VM_WRITE)) == VM_WRITE; 1528fc8744adSLinus Torvalds } 1529fc8744adSLinus Torvalds 15303499a131SLiam R. Howlett /** 15313499a131SLiam R. Howlett * unmapped_area() - Find an area between the low_limit and the high_limit with 15323499a131SLiam R. Howlett * the correct alignment and offset, all from @info. Note: current->mm is used 15333499a131SLiam R. Howlett * for the search. 15343499a131SLiam R. Howlett * 153582b24936SVernon Yang * @info: The unmapped area information including the range [low_limit - 153682b24936SVernon Yang * high_limit), the alignment offset and mask. 15373499a131SLiam R. Howlett * 15383499a131SLiam R. Howlett * Return: A memory address or -ENOMEM. 15393499a131SLiam R. Howlett */ 1540baceaf1cSJaewon Kim static unsigned long unmapped_area(struct vm_unmapped_area_info *info) 1541db4fbfb9SMichel Lespinasse { 15426b008640SLinus Torvalds unsigned long length, gap; 15436b008640SLinus Torvalds unsigned long low_limit, high_limit; 154458c5d0d6SLiam R. Howlett struct vm_area_struct *tmp; 1545db4fbfb9SMichel Lespinasse 15463499a131SLiam R. Howlett MA_STATE(mas, ¤t->mm->mm_mt, 0, 0); 1547db4fbfb9SMichel Lespinasse 1548db4fbfb9SMichel Lespinasse /* Adjust search length to account for worst case alignment overhead */ 1549db4fbfb9SMichel Lespinasse length = info->length + info->align_mask; 1550db4fbfb9SMichel Lespinasse if (length < info->length) 1551db4fbfb9SMichel Lespinasse return -ENOMEM; 1552db4fbfb9SMichel Lespinasse 155358c5d0d6SLiam R. Howlett low_limit = info->low_limit; 15546b008640SLinus Torvalds if (low_limit < mmap_min_addr) 15556b008640SLinus Torvalds low_limit = mmap_min_addr; 15566b008640SLinus Torvalds high_limit = info->high_limit; 155758c5d0d6SLiam R. Howlett retry: 15586b008640SLinus Torvalds if (mas_empty_area(&mas, low_limit, high_limit - 1, length)) 15593499a131SLiam R. Howlett return -ENOMEM; 15603499a131SLiam R. Howlett 1561d4af56c5SLiam R. Howlett gap = mas.index; 1562d4af56c5SLiam R. Howlett gap += (info->align_offset - gap) & info->align_mask; 156358c5d0d6SLiam R. Howlett tmp = mas_next(&mas, ULONG_MAX); 156458c5d0d6SLiam R. Howlett if (tmp && (tmp->vm_flags & VM_GROWSDOWN)) { /* Avoid prev check if possible */ 156558c5d0d6SLiam R. Howlett if (vm_start_gap(tmp) < gap + length - 1) { 156658c5d0d6SLiam R. Howlett low_limit = tmp->vm_end; 156758c5d0d6SLiam R. Howlett mas_reset(&mas); 156858c5d0d6SLiam R. Howlett goto retry; 156958c5d0d6SLiam R. Howlett } 157058c5d0d6SLiam R. Howlett } else { 157158c5d0d6SLiam R. Howlett tmp = mas_prev(&mas, 0); 157258c5d0d6SLiam R. Howlett if (tmp && vm_end_gap(tmp) > gap) { 157358c5d0d6SLiam R. Howlett low_limit = vm_end_gap(tmp); 157458c5d0d6SLiam R. Howlett mas_reset(&mas); 157558c5d0d6SLiam R. Howlett goto retry; 157658c5d0d6SLiam R. Howlett } 157758c5d0d6SLiam R. Howlett } 157858c5d0d6SLiam R. Howlett 15793499a131SLiam R. Howlett return gap; 1580db4fbfb9SMichel Lespinasse } 1581db4fbfb9SMichel Lespinasse 15823499a131SLiam R. Howlett /** 15833499a131SLiam R. Howlett * unmapped_area_topdown() - Find an area between the low_limit and the 158482b24936SVernon Yang * high_limit with the correct alignment and offset at the highest available 15853499a131SLiam R. Howlett * address, all from @info. Note: current->mm is used for the search. 15863499a131SLiam R. Howlett * 158782b24936SVernon Yang * @info: The unmapped area information including the range [low_limit - 158882b24936SVernon Yang * high_limit), the alignment offset and mask. 15893499a131SLiam R. Howlett * 15903499a131SLiam R. Howlett * Return: A memory address or -ENOMEM. 15913499a131SLiam R. Howlett */ 1592baceaf1cSJaewon Kim static unsigned long unmapped_area_topdown(struct vm_unmapped_area_info *info) 1593db4fbfb9SMichel Lespinasse { 15946b008640SLinus Torvalds unsigned long length, gap, gap_end; 15956b008640SLinus Torvalds unsigned long low_limit, high_limit; 159658c5d0d6SLiam R. Howlett struct vm_area_struct *tmp; 1597d4af56c5SLiam R. Howlett 15983499a131SLiam R. Howlett MA_STATE(mas, ¤t->mm->mm_mt, 0, 0); 1599db4fbfb9SMichel Lespinasse /* Adjust search length to account for worst case alignment overhead */ 1600db4fbfb9SMichel Lespinasse length = info->length + info->align_mask; 1601db4fbfb9SMichel Lespinasse if (length < info->length) 1602db4fbfb9SMichel Lespinasse return -ENOMEM; 1603db4fbfb9SMichel Lespinasse 16046b008640SLinus Torvalds low_limit = info->low_limit; 16056b008640SLinus Torvalds if (low_limit < mmap_min_addr) 16066b008640SLinus Torvalds low_limit = mmap_min_addr; 160758c5d0d6SLiam R. Howlett high_limit = info->high_limit; 160858c5d0d6SLiam R. Howlett retry: 16096b008640SLinus Torvalds if (mas_empty_area_rev(&mas, low_limit, high_limit - 1, length)) 16103499a131SLiam R. Howlett return -ENOMEM; 16113499a131SLiam R. Howlett 1612d4af56c5SLiam R. Howlett gap = mas.last + 1 - info->length; 1613d4af56c5SLiam R. Howlett gap -= (gap - info->align_offset) & info->align_mask; 161458c5d0d6SLiam R. Howlett gap_end = mas.last; 161558c5d0d6SLiam R. Howlett tmp = mas_next(&mas, ULONG_MAX); 161658c5d0d6SLiam R. Howlett if (tmp && (tmp->vm_flags & VM_GROWSDOWN)) { /* Avoid prev check if possible */ 161758c5d0d6SLiam R. Howlett if (vm_start_gap(tmp) <= gap_end) { 161858c5d0d6SLiam R. Howlett high_limit = vm_start_gap(tmp); 161958c5d0d6SLiam R. Howlett mas_reset(&mas); 162058c5d0d6SLiam R. Howlett goto retry; 162158c5d0d6SLiam R. Howlett } 162258c5d0d6SLiam R. Howlett } else { 162358c5d0d6SLiam R. Howlett tmp = mas_prev(&mas, 0); 162458c5d0d6SLiam R. Howlett if (tmp && vm_end_gap(tmp) > gap) { 162558c5d0d6SLiam R. Howlett high_limit = tmp->vm_start; 162658c5d0d6SLiam R. Howlett mas_reset(&mas); 162758c5d0d6SLiam R. Howlett goto retry; 162858c5d0d6SLiam R. Howlett } 162958c5d0d6SLiam R. Howlett } 163058c5d0d6SLiam R. Howlett 16313499a131SLiam R. Howlett return gap; 1632db4fbfb9SMichel Lespinasse } 1633db4fbfb9SMichel Lespinasse 1634baceaf1cSJaewon Kim /* 1635baceaf1cSJaewon Kim * Search for an unmapped address range. 1636baceaf1cSJaewon Kim * 1637baceaf1cSJaewon Kim * We are looking for a range that: 1638baceaf1cSJaewon Kim * - does not intersect with any VMA; 1639baceaf1cSJaewon Kim * - is contained within the [low_limit, high_limit) interval; 1640baceaf1cSJaewon Kim * - is at least the desired size. 1641baceaf1cSJaewon Kim * - satisfies (begin_addr & align_mask) == (align_offset & align_mask) 1642baceaf1cSJaewon Kim */ 1643baceaf1cSJaewon Kim unsigned long vm_unmapped_area(struct vm_unmapped_area_info *info) 1644baceaf1cSJaewon Kim { 1645df529cabSJaewon Kim unsigned long addr; 1646df529cabSJaewon Kim 1647baceaf1cSJaewon Kim if (info->flags & VM_UNMAPPED_AREA_TOPDOWN) 1648df529cabSJaewon Kim addr = unmapped_area_topdown(info); 1649baceaf1cSJaewon Kim else 1650df529cabSJaewon Kim addr = unmapped_area(info); 1651df529cabSJaewon Kim 1652df529cabSJaewon Kim trace_vm_unmapped_area(addr, info); 1653df529cabSJaewon Kim return addr; 1654baceaf1cSJaewon Kim } 1655f6795053SSteve Capper 16561da177e4SLinus Torvalds /* Get an address range which is currently unmapped. 16571da177e4SLinus Torvalds * For shmat() with addr=0. 16581da177e4SLinus Torvalds * 16591da177e4SLinus Torvalds * Ugly calling convention alert: 16601da177e4SLinus Torvalds * Return value with the low bits set means error value, 16611da177e4SLinus Torvalds * ie 16621da177e4SLinus Torvalds * if (ret & ~PAGE_MASK) 16631da177e4SLinus Torvalds * error = ret; 16641da177e4SLinus Torvalds * 16651da177e4SLinus Torvalds * This function "knows" that -ENOMEM has the bits set. 16661da177e4SLinus Torvalds */ 16671da177e4SLinus Torvalds unsigned long 16684b439e25SChristophe Leroy generic_get_unmapped_area(struct file *filp, unsigned long addr, 16694b439e25SChristophe Leroy unsigned long len, unsigned long pgoff, 16704b439e25SChristophe Leroy unsigned long flags) 16711da177e4SLinus Torvalds { 16721da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 16731be7107fSHugh Dickins struct vm_area_struct *vma, *prev; 1674db4fbfb9SMichel Lespinasse struct vm_unmapped_area_info info; 16752cb4de08SChristophe Leroy const unsigned long mmap_end = arch_get_mmap_end(addr, len, flags); 16761da177e4SLinus Torvalds 1677f6795053SSteve Capper if (len > mmap_end - mmap_min_addr) 16781da177e4SLinus Torvalds return -ENOMEM; 16791da177e4SLinus Torvalds 168006abdfb4SBenjamin Herrenschmidt if (flags & MAP_FIXED) 168106abdfb4SBenjamin Herrenschmidt return addr; 168206abdfb4SBenjamin Herrenschmidt 16831da177e4SLinus Torvalds if (addr) { 16841da177e4SLinus Torvalds addr = PAGE_ALIGN(addr); 16851be7107fSHugh Dickins vma = find_vma_prev(mm, addr, &prev); 1686f6795053SSteve Capper if (mmap_end - len >= addr && addr >= mmap_min_addr && 16871be7107fSHugh Dickins (!vma || addr + len <= vm_start_gap(vma)) && 16881be7107fSHugh Dickins (!prev || addr >= vm_end_gap(prev))) 16891da177e4SLinus Torvalds return addr; 16901da177e4SLinus Torvalds } 16911da177e4SLinus Torvalds 1692db4fbfb9SMichel Lespinasse info.flags = 0; 1693db4fbfb9SMichel Lespinasse info.length = len; 16944e99b021SHeiko Carstens info.low_limit = mm->mmap_base; 1695f6795053SSteve Capper info.high_limit = mmap_end; 1696db4fbfb9SMichel Lespinasse info.align_mask = 0; 169709ef5283SJaewon Kim info.align_offset = 0; 1698db4fbfb9SMichel Lespinasse return vm_unmapped_area(&info); 16991da177e4SLinus Torvalds } 17004b439e25SChristophe Leroy 17014b439e25SChristophe Leroy #ifndef HAVE_ARCH_UNMAPPED_AREA 17024b439e25SChristophe Leroy unsigned long 17034b439e25SChristophe Leroy arch_get_unmapped_area(struct file *filp, unsigned long addr, 17044b439e25SChristophe Leroy unsigned long len, unsigned long pgoff, 17054b439e25SChristophe Leroy unsigned long flags) 17064b439e25SChristophe Leroy { 17074b439e25SChristophe Leroy return generic_get_unmapped_area(filp, addr, len, pgoff, flags); 17084b439e25SChristophe Leroy } 17091da177e4SLinus Torvalds #endif 17101da177e4SLinus Torvalds 17111da177e4SLinus Torvalds /* 17121da177e4SLinus Torvalds * This mmap-allocator allocates new areas top-down from below the 17131da177e4SLinus Torvalds * stack's low limit (the base): 17141da177e4SLinus Torvalds */ 17151da177e4SLinus Torvalds unsigned long 17164b439e25SChristophe Leroy generic_get_unmapped_area_topdown(struct file *filp, unsigned long addr, 171743cca0b1SYang Fan unsigned long len, unsigned long pgoff, 171843cca0b1SYang Fan unsigned long flags) 17191da177e4SLinus Torvalds { 17201be7107fSHugh Dickins struct vm_area_struct *vma, *prev; 17211da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 1722db4fbfb9SMichel Lespinasse struct vm_unmapped_area_info info; 17232cb4de08SChristophe Leroy const unsigned long mmap_end = arch_get_mmap_end(addr, len, flags); 17241da177e4SLinus Torvalds 17251da177e4SLinus Torvalds /* requested length too big for entire address space */ 1726f6795053SSteve Capper if (len > mmap_end - mmap_min_addr) 17271da177e4SLinus Torvalds return -ENOMEM; 17281da177e4SLinus Torvalds 172906abdfb4SBenjamin Herrenschmidt if (flags & MAP_FIXED) 173006abdfb4SBenjamin Herrenschmidt return addr; 173106abdfb4SBenjamin Herrenschmidt 17321da177e4SLinus Torvalds /* requesting a specific address */ 17331da177e4SLinus Torvalds if (addr) { 17341da177e4SLinus Torvalds addr = PAGE_ALIGN(addr); 17351be7107fSHugh Dickins vma = find_vma_prev(mm, addr, &prev); 1736f6795053SSteve Capper if (mmap_end - len >= addr && addr >= mmap_min_addr && 17371be7107fSHugh Dickins (!vma || addr + len <= vm_start_gap(vma)) && 17381be7107fSHugh Dickins (!prev || addr >= vm_end_gap(prev))) 17391da177e4SLinus Torvalds return addr; 17401da177e4SLinus Torvalds } 17411da177e4SLinus Torvalds 1742db4fbfb9SMichel Lespinasse info.flags = VM_UNMAPPED_AREA_TOPDOWN; 1743db4fbfb9SMichel Lespinasse info.length = len; 17446b008640SLinus Torvalds info.low_limit = PAGE_SIZE; 1745f6795053SSteve Capper info.high_limit = arch_get_mmap_base(addr, mm->mmap_base); 1746db4fbfb9SMichel Lespinasse info.align_mask = 0; 174709ef5283SJaewon Kim info.align_offset = 0; 1748db4fbfb9SMichel Lespinasse addr = vm_unmapped_area(&info); 1749b716ad95SXiao Guangrong 17501da177e4SLinus Torvalds /* 17511da177e4SLinus Torvalds * A failed mmap() very likely causes application failure, 17521da177e4SLinus Torvalds * so fall back to the bottom-up function here. This scenario 17531da177e4SLinus Torvalds * can happen with large stack limits and large mmap() 17541da177e4SLinus Torvalds * allocations. 17551da177e4SLinus Torvalds */ 1756de1741a1SAlexander Kuleshov if (offset_in_page(addr)) { 1757db4fbfb9SMichel Lespinasse VM_BUG_ON(addr != -ENOMEM); 1758db4fbfb9SMichel Lespinasse info.flags = 0; 1759db4fbfb9SMichel Lespinasse info.low_limit = TASK_UNMAPPED_BASE; 1760f6795053SSteve Capper info.high_limit = mmap_end; 1761db4fbfb9SMichel Lespinasse addr = vm_unmapped_area(&info); 1762db4fbfb9SMichel Lespinasse } 17631da177e4SLinus Torvalds 17641da177e4SLinus Torvalds return addr; 17651da177e4SLinus Torvalds } 17664b439e25SChristophe Leroy 17674b439e25SChristophe Leroy #ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN 17684b439e25SChristophe Leroy unsigned long 17694b439e25SChristophe Leroy arch_get_unmapped_area_topdown(struct file *filp, unsigned long addr, 17704b439e25SChristophe Leroy unsigned long len, unsigned long pgoff, 17714b439e25SChristophe Leroy unsigned long flags) 17724b439e25SChristophe Leroy { 17734b439e25SChristophe Leroy return generic_get_unmapped_area_topdown(filp, addr, len, pgoff, flags); 17744b439e25SChristophe Leroy } 17751da177e4SLinus Torvalds #endif 17761da177e4SLinus Torvalds 17771da177e4SLinus Torvalds unsigned long 17781da177e4SLinus Torvalds get_unmapped_area(struct file *file, unsigned long addr, unsigned long len, 17791da177e4SLinus Torvalds unsigned long pgoff, unsigned long flags) 17801da177e4SLinus Torvalds { 178106abdfb4SBenjamin Herrenschmidt unsigned long (*get_area)(struct file *, unsigned long, 178206abdfb4SBenjamin Herrenschmidt unsigned long, unsigned long, unsigned long); 178307ab67c8SLinus Torvalds 17849206de95SAl Viro unsigned long error = arch_mmap_check(addr, len, flags); 17859206de95SAl Viro if (error) 17869206de95SAl Viro return error; 17879206de95SAl Viro 17889206de95SAl Viro /* Careful about overflows.. */ 17899206de95SAl Viro if (len > TASK_SIZE) 17909206de95SAl Viro return -ENOMEM; 17919206de95SAl Viro 179207ab67c8SLinus Torvalds get_area = current->mm->get_unmapped_area; 1793c01d5b30SHugh Dickins if (file) { 1794c01d5b30SHugh Dickins if (file->f_op->get_unmapped_area) 179507ab67c8SLinus Torvalds get_area = file->f_op->get_unmapped_area; 1796c01d5b30SHugh Dickins } else if (flags & MAP_SHARED) { 1797c01d5b30SHugh Dickins /* 1798c01d5b30SHugh Dickins * mmap_region() will call shmem_zero_setup() to create a file, 1799c01d5b30SHugh Dickins * so use shmem's get_unmapped_area in case it can be huge. 180045e55300SPeter Collingbourne * do_mmap() will clear pgoff, so match alignment. 1801c01d5b30SHugh Dickins */ 1802c01d5b30SHugh Dickins pgoff = 0; 1803c01d5b30SHugh Dickins get_area = shmem_get_unmapped_area; 1804c01d5b30SHugh Dickins } 1805c01d5b30SHugh Dickins 180607ab67c8SLinus Torvalds addr = get_area(file, addr, len, pgoff, flags); 180707ab67c8SLinus Torvalds if (IS_ERR_VALUE(addr)) 180807ab67c8SLinus Torvalds return addr; 180907ab67c8SLinus Torvalds 18101da177e4SLinus Torvalds if (addr > TASK_SIZE - len) 18111da177e4SLinus Torvalds return -ENOMEM; 1812de1741a1SAlexander Kuleshov if (offset_in_page(addr)) 18131da177e4SLinus Torvalds return -EINVAL; 181406abdfb4SBenjamin Herrenschmidt 18159ac4ed4bSAl Viro error = security_mmap_addr(addr); 18169ac4ed4bSAl Viro return error ? error : addr; 18171da177e4SLinus Torvalds } 18181da177e4SLinus Torvalds 18191da177e4SLinus Torvalds EXPORT_SYMBOL(get_unmapped_area); 18201da177e4SLinus Torvalds 1821be8432e7SLiam R. Howlett /** 1822abdba2ddSLiam R. Howlett * find_vma_intersection() - Look up the first VMA which intersects the interval 1823abdba2ddSLiam R. Howlett * @mm: The process address space. 1824abdba2ddSLiam R. Howlett * @start_addr: The inclusive start user address. 1825abdba2ddSLiam R. Howlett * @end_addr: The exclusive end user address. 1826abdba2ddSLiam R. Howlett * 1827abdba2ddSLiam R. Howlett * Returns: The first VMA within the provided range, %NULL otherwise. Assumes 1828abdba2ddSLiam R. Howlett * start_addr < end_addr. 1829abdba2ddSLiam R. Howlett */ 1830abdba2ddSLiam R. Howlett struct vm_area_struct *find_vma_intersection(struct mm_struct *mm, 1831abdba2ddSLiam R. Howlett unsigned long start_addr, 1832abdba2ddSLiam R. Howlett unsigned long end_addr) 1833abdba2ddSLiam R. Howlett { 1834abdba2ddSLiam R. Howlett unsigned long index = start_addr; 1835abdba2ddSLiam R. Howlett 1836abdba2ddSLiam R. Howlett mmap_assert_locked(mm); 18377964cf8cSLiam R. Howlett return mt_find(&mm->mm_mt, &index, end_addr - 1); 1838abdba2ddSLiam R. Howlett } 1839abdba2ddSLiam R. Howlett EXPORT_SYMBOL(find_vma_intersection); 1840abdba2ddSLiam R. Howlett 1841abdba2ddSLiam R. Howlett /** 1842be8432e7SLiam R. Howlett * find_vma() - Find the VMA for a given address, or the next VMA. 1843be8432e7SLiam R. Howlett * @mm: The mm_struct to check 1844be8432e7SLiam R. Howlett * @addr: The address 1845be8432e7SLiam R. Howlett * 1846be8432e7SLiam R. Howlett * Returns: The VMA associated with addr, or the next VMA. 1847be8432e7SLiam R. Howlett * May return %NULL in the case of no VMA at addr or above. 1848be8432e7SLiam R. Howlett */ 18491da177e4SLinus Torvalds struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr) 18501da177e4SLinus Torvalds { 1851be8432e7SLiam R. Howlett unsigned long index = addr; 18521da177e4SLinus Torvalds 18535b78ed24SLuigi Rizzo mmap_assert_locked(mm); 18547964cf8cSLiam R. Howlett return mt_find(&mm->mm_mt, &index, ULONG_MAX); 18551da177e4SLinus Torvalds } 18561da177e4SLinus Torvalds EXPORT_SYMBOL(find_vma); 18571da177e4SLinus Torvalds 18587fdbd37dSLiam R. Howlett /** 18597fdbd37dSLiam R. Howlett * find_vma_prev() - Find the VMA for a given address, or the next vma and 18607fdbd37dSLiam R. Howlett * set %pprev to the previous VMA, if any. 18617fdbd37dSLiam R. Howlett * @mm: The mm_struct to check 18627fdbd37dSLiam R. Howlett * @addr: The address 18637fdbd37dSLiam R. Howlett * @pprev: The pointer to set to the previous VMA 18647fdbd37dSLiam R. Howlett * 18657fdbd37dSLiam R. Howlett * Note that RCU lock is missing here since the external mmap_lock() is used 18667fdbd37dSLiam R. Howlett * instead. 18677fdbd37dSLiam R. Howlett * 18687fdbd37dSLiam R. Howlett * Returns: The VMA associated with @addr, or the next vma. 18697fdbd37dSLiam R. Howlett * May return %NULL in the case of no vma at addr or above. 18706bd4837dSKOSAKI Motohiro */ 18711da177e4SLinus Torvalds struct vm_area_struct * 18721da177e4SLinus Torvalds find_vma_prev(struct mm_struct *mm, unsigned long addr, 18731da177e4SLinus Torvalds struct vm_area_struct **pprev) 18741da177e4SLinus Torvalds { 18756bd4837dSKOSAKI Motohiro struct vm_area_struct *vma; 18767fdbd37dSLiam R. Howlett MA_STATE(mas, &mm->mm_mt, addr, addr); 18771da177e4SLinus Torvalds 18787fdbd37dSLiam R. Howlett vma = mas_walk(&mas); 18797fdbd37dSLiam R. Howlett *pprev = mas_prev(&mas, 0); 18807fdbd37dSLiam R. Howlett if (!vma) 18817fdbd37dSLiam R. Howlett vma = mas_next(&mas, ULONG_MAX); 18826bd4837dSKOSAKI Motohiro return vma; 18831da177e4SLinus Torvalds } 18841da177e4SLinus Torvalds 18851da177e4SLinus Torvalds /* 18861da177e4SLinus Torvalds * Verify that the stack growth is acceptable and 18871da177e4SLinus Torvalds * update accounting. This is shared with both the 18881da177e4SLinus Torvalds * grow-up and grow-down cases. 18891da177e4SLinus Torvalds */ 18901be7107fSHugh Dickins static int acct_stack_growth(struct vm_area_struct *vma, 18911be7107fSHugh Dickins unsigned long size, unsigned long grow) 18921da177e4SLinus Torvalds { 18931da177e4SLinus Torvalds struct mm_struct *mm = vma->vm_mm; 18941be7107fSHugh Dickins unsigned long new_start; 18951da177e4SLinus Torvalds 18961da177e4SLinus Torvalds /* address space limit tests */ 189784638335SKonstantin Khlebnikov if (!may_expand_vm(mm, vma->vm_flags, grow)) 18981da177e4SLinus Torvalds return -ENOMEM; 18991da177e4SLinus Torvalds 19001da177e4SLinus Torvalds /* Stack limit test */ 190124c79d8eSKrzysztof Opasiak if (size > rlimit(RLIMIT_STACK)) 19021da177e4SLinus Torvalds return -ENOMEM; 19031da177e4SLinus Torvalds 19041da177e4SLinus Torvalds /* mlock limit tests */ 1905b0cc5e89SAndrew Morton if (!mlock_future_ok(mm, vma->vm_flags, grow << PAGE_SHIFT)) 19061da177e4SLinus Torvalds return -ENOMEM; 19071da177e4SLinus Torvalds 19080d59a01bSAdam Litke /* Check to ensure the stack will not grow into a hugetlb-only region */ 19090d59a01bSAdam Litke new_start = (vma->vm_flags & VM_GROWSUP) ? vma->vm_start : 19100d59a01bSAdam Litke vma->vm_end - size; 19110d59a01bSAdam Litke if (is_hugepage_only_range(vma->vm_mm, new_start, size)) 19120d59a01bSAdam Litke return -EFAULT; 19130d59a01bSAdam Litke 19141da177e4SLinus Torvalds /* 19151da177e4SLinus Torvalds * Overcommit.. This must be the final test, as it will 19161da177e4SLinus Torvalds * update security statistics. 19171da177e4SLinus Torvalds */ 191805fa199dSHugh Dickins if (security_vm_enough_memory_mm(mm, grow)) 19191da177e4SLinus Torvalds return -ENOMEM; 19201da177e4SLinus Torvalds 19211da177e4SLinus Torvalds return 0; 19221da177e4SLinus Torvalds } 19231da177e4SLinus Torvalds 192446dea3d0SHugh Dickins #if defined(CONFIG_STACK_GROWSUP) || defined(CONFIG_IA64) 19251da177e4SLinus Torvalds /* 192646dea3d0SHugh Dickins * PA-RISC uses this for its stack; IA64 for its Register Backing Store. 192746dea3d0SHugh Dickins * vma is the last one with address > vma->vm_end. Have to extend vma. 19281da177e4SLinus Torvalds */ 19298d7071afSLinus Torvalds static int expand_upwards(struct vm_area_struct *vma, unsigned long address) 19301da177e4SLinus Torvalds { 193109357814SOleg Nesterov struct mm_struct *mm = vma->vm_mm; 19321be7107fSHugh Dickins struct vm_area_struct *next; 19331be7107fSHugh Dickins unsigned long gap_addr; 193412352d3cSKonstantin Khlebnikov int error = 0; 1935b5df0922SLiam R. Howlett MA_STATE(mas, &mm->mm_mt, vma->vm_start, address); 19361da177e4SLinus Torvalds 19371da177e4SLinus Torvalds if (!(vma->vm_flags & VM_GROWSUP)) 19381da177e4SLinus Torvalds return -EFAULT; 19391da177e4SLinus Torvalds 1940bd726c90SHelge Deller /* Guard against exceeding limits of the address space. */ 19411be7107fSHugh Dickins address &= PAGE_MASK; 194237511fb5SHelge Deller if (address >= (TASK_SIZE & PAGE_MASK)) 194312352d3cSKonstantin Khlebnikov return -ENOMEM; 1944bd726c90SHelge Deller address += PAGE_SIZE; 194512352d3cSKonstantin Khlebnikov 19461be7107fSHugh Dickins /* Enforce stack_guard_gap */ 19471be7107fSHugh Dickins gap_addr = address + stack_guard_gap; 1948bd726c90SHelge Deller 1949bd726c90SHelge Deller /* Guard against overflow */ 1950bd726c90SHelge Deller if (gap_addr < address || gap_addr > TASK_SIZE) 1951bd726c90SHelge Deller gap_addr = TASK_SIZE; 1952bd726c90SHelge Deller 1953763ecb03SLiam R. Howlett next = find_vma_intersection(mm, vma->vm_end, gap_addr); 1954763ecb03SLiam R. Howlett if (next && vma_is_accessible(next)) { 19551be7107fSHugh Dickins if (!(next->vm_flags & VM_GROWSUP)) 19561be7107fSHugh Dickins return -ENOMEM; 19571be7107fSHugh Dickins /* Check that both stack segments have the same anon_vma? */ 19581be7107fSHugh Dickins } 19591be7107fSHugh Dickins 1960b5df0922SLiam R. Howlett if (next) 1961b5df0922SLiam R. Howlett mas_prev_range(&mas, address); 1962b5df0922SLiam R. Howlett 1963b5df0922SLiam R. Howlett __mas_set_range(&mas, vma->vm_start, address - 1); 1964da089254SLiam R. Howlett if (mas_preallocate(&mas, vma, GFP_KERNEL)) 19651da177e4SLinus Torvalds return -ENOMEM; 19661da177e4SLinus Torvalds 1967d4af56c5SLiam R. Howlett /* We must make sure the anon_vma is allocated. */ 1968d4af56c5SLiam R. Howlett if (unlikely(anon_vma_prepare(vma))) { 1969d4af56c5SLiam R. Howlett mas_destroy(&mas); 1970d4af56c5SLiam R. Howlett return -ENOMEM; 1971d4af56c5SLiam R. Howlett } 1972d4af56c5SLiam R. Howlett 1973c137381fSSuren Baghdasaryan /* Lock the VMA before expanding to prevent concurrent page faults */ 1974c137381fSSuren Baghdasaryan vma_start_write(vma); 19751da177e4SLinus Torvalds /* 19761da177e4SLinus Torvalds * vma->vm_start/vm_end cannot change under us because the caller 1977c1e8d7c6SMichel Lespinasse * is required to hold the mmap_lock in read mode. We need the 19781da177e4SLinus Torvalds * anon_vma lock to serialize against concurrent expand_stacks. 19791da177e4SLinus Torvalds */ 198012352d3cSKonstantin Khlebnikov anon_vma_lock_write(vma->anon_vma); 19811da177e4SLinus Torvalds 19821da177e4SLinus Torvalds /* Somebody else might have raced and expanded it already */ 19831da177e4SLinus Torvalds if (address > vma->vm_end) { 19841da177e4SLinus Torvalds unsigned long size, grow; 19851da177e4SLinus Torvalds 19861da177e4SLinus Torvalds size = address - vma->vm_start; 19871da177e4SLinus Torvalds grow = (address - vma->vm_end) >> PAGE_SHIFT; 19881da177e4SLinus Torvalds 198942c36f63SHugh Dickins error = -ENOMEM; 199042c36f63SHugh Dickins if (vma->vm_pgoff + (size >> PAGE_SHIFT) >= vma->vm_pgoff) { 19911da177e4SLinus Torvalds error = acct_stack_growth(vma, size, grow); 19923af9e859SEric B Munson if (!error) { 19934128997bSMichel Lespinasse /* 1994524e00b3SLiam R. Howlett * We only hold a shared mmap_lock lock here, so 1995524e00b3SLiam R. Howlett * we need to protect against concurrent vma 1996524e00b3SLiam R. Howlett * expansions. anon_vma_lock_write() doesn't 1997524e00b3SLiam R. Howlett * help here, as we don't guarantee that all 1998524e00b3SLiam R. Howlett * growable vmas in a mm share the same root 1999524e00b3SLiam R. Howlett * anon vma. So, we reuse mm->page_table_lock 2000524e00b3SLiam R. Howlett * to guard against concurrent vma expansions. 20014128997bSMichel Lespinasse */ 200209357814SOleg Nesterov spin_lock(&mm->page_table_lock); 200387e8827bSOleg Nesterov if (vma->vm_flags & VM_LOCKED) 200409357814SOleg Nesterov mm->locked_vm += grow; 200584638335SKonstantin Khlebnikov vm_stat_account(mm, vma->vm_flags, grow); 2006bf181b9fSMichel Lespinasse anon_vma_interval_tree_pre_update_vma(vma); 20071da177e4SLinus Torvalds vma->vm_end = address; 2008d4af56c5SLiam R. Howlett /* Overwrite old entry in mtree. */ 2009fbcc3104SLiam R. Howlett mas_store_prealloc(&mas, vma); 2010bf181b9fSMichel Lespinasse anon_vma_interval_tree_post_update_vma(vma); 201109357814SOleg Nesterov spin_unlock(&mm->page_table_lock); 20124128997bSMichel Lespinasse 20133af9e859SEric B Munson perf_event_mmap(vma); 20143af9e859SEric B Munson } 20151da177e4SLinus Torvalds } 201642c36f63SHugh Dickins } 201712352d3cSKonstantin Khlebnikov anon_vma_unlock_write(vma->anon_vma); 2018c791576cSYang Shi khugepaged_enter_vma(vma, vma->vm_flags); 2019d4af56c5SLiam R. Howlett mas_destroy(&mas); 20202574d5e4SLiam R. Howlett validate_mm(mm); 20211da177e4SLinus Torvalds return error; 20221da177e4SLinus Torvalds } 202346dea3d0SHugh Dickins #endif /* CONFIG_STACK_GROWSUP || CONFIG_IA64 */ 202446dea3d0SHugh Dickins 20251da177e4SLinus Torvalds /* 20261da177e4SLinus Torvalds * vma is the first one with address < vma->vm_start. Have to extend vma. 20278d7071afSLinus Torvalds * mmap_lock held for writing. 20281da177e4SLinus Torvalds */ 2029524e00b3SLiam R. Howlett int expand_downwards(struct vm_area_struct *vma, unsigned long address) 20301da177e4SLinus Torvalds { 203109357814SOleg Nesterov struct mm_struct *mm = vma->vm_mm; 2032763ecb03SLiam R. Howlett MA_STATE(mas, &mm->mm_mt, vma->vm_start, vma->vm_start); 20331be7107fSHugh Dickins struct vm_area_struct *prev; 20340a1d5299SJann Horn int error = 0; 20351da177e4SLinus Torvalds 20368d7071afSLinus Torvalds if (!(vma->vm_flags & VM_GROWSDOWN)) 20378d7071afSLinus Torvalds return -EFAULT; 20388d7071afSLinus Torvalds 20398869477aSEric Paris address &= PAGE_MASK; 20408b35ca3eSBen Hutchings if (address < mmap_min_addr || address < FIRST_USER_ADDRESS) 20410a1d5299SJann Horn return -EPERM; 20428869477aSEric Paris 20431be7107fSHugh Dickins /* Enforce stack_guard_gap */ 2044763ecb03SLiam R. Howlett prev = mas_prev(&mas, 0); 20451be7107fSHugh Dickins /* Check that both stack segments have the same anon_vma? */ 2046f440fa1aSLiam R. Howlett if (prev) { 2047f440fa1aSLiam R. Howlett if (!(prev->vm_flags & VM_GROWSDOWN) && 2048f440fa1aSLiam R. Howlett vma_is_accessible(prev) && 2049f440fa1aSLiam R. Howlett (address - prev->vm_end < stack_guard_gap)) 205032e4e6d5SOleg Nesterov return -ENOMEM; 20511be7107fSHugh Dickins } 20521be7107fSHugh Dickins 2053b5df0922SLiam R. Howlett if (prev) 2054b5df0922SLiam R. Howlett mas_next_range(&mas, vma->vm_start); 2055b5df0922SLiam R. Howlett 2056b5df0922SLiam R. Howlett __mas_set_range(&mas, address, vma->vm_end - 1); 2057da089254SLiam R. Howlett if (mas_preallocate(&mas, vma, GFP_KERNEL)) 205812352d3cSKonstantin Khlebnikov return -ENOMEM; 20591da177e4SLinus Torvalds 2060d4af56c5SLiam R. Howlett /* We must make sure the anon_vma is allocated. */ 2061d4af56c5SLiam R. Howlett if (unlikely(anon_vma_prepare(vma))) { 2062d4af56c5SLiam R. Howlett mas_destroy(&mas); 2063d4af56c5SLiam R. Howlett return -ENOMEM; 2064d4af56c5SLiam R. Howlett } 2065d4af56c5SLiam R. Howlett 2066c137381fSSuren Baghdasaryan /* Lock the VMA before expanding to prevent concurrent page faults */ 2067c137381fSSuren Baghdasaryan vma_start_write(vma); 20681da177e4SLinus Torvalds /* 20691da177e4SLinus Torvalds * vma->vm_start/vm_end cannot change under us because the caller 2070c1e8d7c6SMichel Lespinasse * is required to hold the mmap_lock in read mode. We need the 20711da177e4SLinus Torvalds * anon_vma lock to serialize against concurrent expand_stacks. 20721da177e4SLinus Torvalds */ 207312352d3cSKonstantin Khlebnikov anon_vma_lock_write(vma->anon_vma); 20741da177e4SLinus Torvalds 20751da177e4SLinus Torvalds /* Somebody else might have raced and expanded it already */ 20761da177e4SLinus Torvalds if (address < vma->vm_start) { 20771da177e4SLinus Torvalds unsigned long size, grow; 20781da177e4SLinus Torvalds 20791da177e4SLinus Torvalds size = vma->vm_end - address; 20801da177e4SLinus Torvalds grow = (vma->vm_start - address) >> PAGE_SHIFT; 20811da177e4SLinus Torvalds 2082a626ca6aSLinus Torvalds error = -ENOMEM; 2083a626ca6aSLinus Torvalds if (grow <= vma->vm_pgoff) { 20841da177e4SLinus Torvalds error = acct_stack_growth(vma, size, grow); 20851da177e4SLinus Torvalds if (!error) { 20864128997bSMichel Lespinasse /* 2087524e00b3SLiam R. Howlett * We only hold a shared mmap_lock lock here, so 2088524e00b3SLiam R. Howlett * we need to protect against concurrent vma 2089524e00b3SLiam R. Howlett * expansions. anon_vma_lock_write() doesn't 2090524e00b3SLiam R. Howlett * help here, as we don't guarantee that all 2091524e00b3SLiam R. Howlett * growable vmas in a mm share the same root 2092524e00b3SLiam R. Howlett * anon vma. So, we reuse mm->page_table_lock 2093524e00b3SLiam R. Howlett * to guard against concurrent vma expansions. 20944128997bSMichel Lespinasse */ 209509357814SOleg Nesterov spin_lock(&mm->page_table_lock); 209687e8827bSOleg Nesterov if (vma->vm_flags & VM_LOCKED) 209709357814SOleg Nesterov mm->locked_vm += grow; 209884638335SKonstantin Khlebnikov vm_stat_account(mm, vma->vm_flags, grow); 2099bf181b9fSMichel Lespinasse anon_vma_interval_tree_pre_update_vma(vma); 21001da177e4SLinus Torvalds vma->vm_start = address; 21011da177e4SLinus Torvalds vma->vm_pgoff -= grow; 2102d4af56c5SLiam R. Howlett /* Overwrite old entry in mtree. */ 2103fbcc3104SLiam R. Howlett mas_store_prealloc(&mas, vma); 2104bf181b9fSMichel Lespinasse anon_vma_interval_tree_post_update_vma(vma); 210509357814SOleg Nesterov spin_unlock(&mm->page_table_lock); 21064128997bSMichel Lespinasse 21073af9e859SEric B Munson perf_event_mmap(vma); 21081da177e4SLinus Torvalds } 21091da177e4SLinus Torvalds } 2110a626ca6aSLinus Torvalds } 211112352d3cSKonstantin Khlebnikov anon_vma_unlock_write(vma->anon_vma); 2112c791576cSYang Shi khugepaged_enter_vma(vma, vma->vm_flags); 2113d4af56c5SLiam R. Howlett mas_destroy(&mas); 21142574d5e4SLiam R. Howlett validate_mm(mm); 21151da177e4SLinus Torvalds return error; 21161da177e4SLinus Torvalds } 21171da177e4SLinus Torvalds 21181be7107fSHugh Dickins /* enforced gap between the expanding stack and other mappings. */ 21191be7107fSHugh Dickins unsigned long stack_guard_gap = 256UL<<PAGE_SHIFT; 21201be7107fSHugh Dickins 21211be7107fSHugh Dickins static int __init cmdline_parse_stack_guard_gap(char *p) 21221be7107fSHugh Dickins { 21231be7107fSHugh Dickins unsigned long val; 21241be7107fSHugh Dickins char *endptr; 21251be7107fSHugh Dickins 21261be7107fSHugh Dickins val = simple_strtoul(p, &endptr, 10); 21271be7107fSHugh Dickins if (!*endptr) 21281be7107fSHugh Dickins stack_guard_gap = val << PAGE_SHIFT; 21291be7107fSHugh Dickins 2130e6d09493SRandy Dunlap return 1; 21311be7107fSHugh Dickins } 21321be7107fSHugh Dickins __setup("stack_guard_gap=", cmdline_parse_stack_guard_gap); 21331be7107fSHugh Dickins 2134b6a2fea3SOllie Wild #ifdef CONFIG_STACK_GROWSUP 21358d7071afSLinus Torvalds int expand_stack_locked(struct vm_area_struct *vma, unsigned long address) 2136b6a2fea3SOllie Wild { 2137b6a2fea3SOllie Wild return expand_upwards(vma, address); 2138b6a2fea3SOllie Wild } 2139b6a2fea3SOllie Wild 21408d7071afSLinus Torvalds struct vm_area_struct *find_extend_vma_locked(struct mm_struct *mm, unsigned long addr) 2141b6a2fea3SOllie Wild { 2142b6a2fea3SOllie Wild struct vm_area_struct *vma, *prev; 2143b6a2fea3SOllie Wild 2144b6a2fea3SOllie Wild addr &= PAGE_MASK; 2145b6a2fea3SOllie Wild vma = find_vma_prev(mm, addr, &prev); 2146b6a2fea3SOllie Wild if (vma && (vma->vm_start <= addr)) 2147b6a2fea3SOllie Wild return vma; 2148f440fa1aSLiam R. Howlett if (!prev) 2149f440fa1aSLiam R. Howlett return NULL; 21508d7071afSLinus Torvalds if (expand_stack_locked(prev, addr)) 2151b6a2fea3SOllie Wild return NULL; 2152cea10a19SMichel Lespinasse if (prev->vm_flags & VM_LOCKED) 2153fc05f566SKirill A. Shutemov populate_vma_page_range(prev, addr, prev->vm_end, NULL); 2154b6a2fea3SOllie Wild return prev; 2155b6a2fea3SOllie Wild } 2156b6a2fea3SOllie Wild #else 21578d7071afSLinus Torvalds int expand_stack_locked(struct vm_area_struct *vma, unsigned long address) 2158b6a2fea3SOllie Wild { 2159f440fa1aSLiam R. Howlett if (unlikely(!(vma->vm_flags & VM_GROWSDOWN))) 2160f440fa1aSLiam R. Howlett return -EINVAL; 2161b6a2fea3SOllie Wild return expand_downwards(vma, address); 2162b6a2fea3SOllie Wild } 2163b6a2fea3SOllie Wild 21648d7071afSLinus Torvalds struct vm_area_struct *find_extend_vma_locked(struct mm_struct *mm, unsigned long addr) 21651da177e4SLinus Torvalds { 21661da177e4SLinus Torvalds struct vm_area_struct *vma; 21671da177e4SLinus Torvalds unsigned long start; 21681da177e4SLinus Torvalds 21691da177e4SLinus Torvalds addr &= PAGE_MASK; 21701da177e4SLinus Torvalds vma = find_vma(mm, addr); 21711da177e4SLinus Torvalds if (!vma) 21721da177e4SLinus Torvalds return NULL; 21731da177e4SLinus Torvalds if (vma->vm_start <= addr) 21741da177e4SLinus Torvalds return vma; 21751da177e4SLinus Torvalds start = vma->vm_start; 21768d7071afSLinus Torvalds if (expand_stack_locked(vma, addr)) 21771da177e4SLinus Torvalds return NULL; 2178cea10a19SMichel Lespinasse if (vma->vm_flags & VM_LOCKED) 2179fc05f566SKirill A. Shutemov populate_vma_page_range(vma, addr, start, NULL); 21801da177e4SLinus Torvalds return vma; 21811da177e4SLinus Torvalds } 21821da177e4SLinus Torvalds #endif 21831da177e4SLinus Torvalds 21848d7071afSLinus Torvalds /* 21858d7071afSLinus Torvalds * IA64 has some horrid mapping rules: it can expand both up and down, 21868d7071afSLinus Torvalds * but with various special rules. 21878d7071afSLinus Torvalds * 21888d7071afSLinus Torvalds * We'll get rid of this architecture eventually, so the ugliness is 21898d7071afSLinus Torvalds * temporary. 21908d7071afSLinus Torvalds */ 21918d7071afSLinus Torvalds #ifdef CONFIG_IA64 21928d7071afSLinus Torvalds static inline bool vma_expand_ok(struct vm_area_struct *vma, unsigned long addr) 2193f440fa1aSLiam R. Howlett { 21948d7071afSLinus Torvalds return REGION_NUMBER(addr) == REGION_NUMBER(vma->vm_start) && 21958d7071afSLinus Torvalds REGION_OFFSET(addr) < RGN_MAP_LIMIT; 2196f440fa1aSLiam R. Howlett } 21978d7071afSLinus Torvalds 21988d7071afSLinus Torvalds /* 21998d7071afSLinus Torvalds * IA64 stacks grow down, but there's a special register backing store 22008d7071afSLinus Torvalds * that can grow up. Only sequentially, though, so the new address must 22018d7071afSLinus Torvalds * match vm_end. 22028d7071afSLinus Torvalds */ 22038d7071afSLinus Torvalds static inline int vma_expand_up(struct vm_area_struct *vma, unsigned long addr) 22048d7071afSLinus Torvalds { 22058d7071afSLinus Torvalds if (!vma_expand_ok(vma, addr)) 22068d7071afSLinus Torvalds return -EFAULT; 22078d7071afSLinus Torvalds if (vma->vm_end != (addr & PAGE_MASK)) 22088d7071afSLinus Torvalds return -EFAULT; 22098d7071afSLinus Torvalds return expand_upwards(vma, addr); 22108d7071afSLinus Torvalds } 22118d7071afSLinus Torvalds 22128d7071afSLinus Torvalds static inline bool vma_expand_down(struct vm_area_struct *vma, unsigned long addr) 22138d7071afSLinus Torvalds { 22148d7071afSLinus Torvalds if (!vma_expand_ok(vma, addr)) 22158d7071afSLinus Torvalds return -EFAULT; 22168d7071afSLinus Torvalds return expand_downwards(vma, addr); 22178d7071afSLinus Torvalds } 22188d7071afSLinus Torvalds 22198d7071afSLinus Torvalds #elif defined(CONFIG_STACK_GROWSUP) 22208d7071afSLinus Torvalds 22218d7071afSLinus Torvalds #define vma_expand_up(vma,addr) expand_upwards(vma, addr) 22228d7071afSLinus Torvalds #define vma_expand_down(vma, addr) (-EFAULT) 22238d7071afSLinus Torvalds 22248d7071afSLinus Torvalds #else 22258d7071afSLinus Torvalds 22268d7071afSLinus Torvalds #define vma_expand_up(vma,addr) (-EFAULT) 22278d7071afSLinus Torvalds #define vma_expand_down(vma, addr) expand_downwards(vma, addr) 22288d7071afSLinus Torvalds 22298d7071afSLinus Torvalds #endif 22308d7071afSLinus Torvalds 22318d7071afSLinus Torvalds /* 22328d7071afSLinus Torvalds * expand_stack(): legacy interface for page faulting. Don't use unless 22338d7071afSLinus Torvalds * you have to. 22348d7071afSLinus Torvalds * 22358d7071afSLinus Torvalds * This is called with the mm locked for reading, drops the lock, takes 22368d7071afSLinus Torvalds * the lock for writing, tries to look up a vma again, expands it if 22378d7071afSLinus Torvalds * necessary, and downgrades the lock to reading again. 22388d7071afSLinus Torvalds * 22398d7071afSLinus Torvalds * If no vma is found or it can't be expanded, it returns NULL and has 22408d7071afSLinus Torvalds * dropped the lock. 22418d7071afSLinus Torvalds */ 22428d7071afSLinus Torvalds struct vm_area_struct *expand_stack(struct mm_struct *mm, unsigned long addr) 22438d7071afSLinus Torvalds { 22448d7071afSLinus Torvalds struct vm_area_struct *vma, *prev; 22458d7071afSLinus Torvalds 22468d7071afSLinus Torvalds mmap_read_unlock(mm); 22478d7071afSLinus Torvalds if (mmap_write_lock_killable(mm)) 22488d7071afSLinus Torvalds return NULL; 22498d7071afSLinus Torvalds 22508d7071afSLinus Torvalds vma = find_vma_prev(mm, addr, &prev); 22518d7071afSLinus Torvalds if (vma && vma->vm_start <= addr) 22528d7071afSLinus Torvalds goto success; 22538d7071afSLinus Torvalds 22548d7071afSLinus Torvalds if (prev && !vma_expand_up(prev, addr)) { 22558d7071afSLinus Torvalds vma = prev; 22568d7071afSLinus Torvalds goto success; 22578d7071afSLinus Torvalds } 22588d7071afSLinus Torvalds 22598d7071afSLinus Torvalds if (vma && !vma_expand_down(vma, addr)) 22608d7071afSLinus Torvalds goto success; 22618d7071afSLinus Torvalds 22628d7071afSLinus Torvalds mmap_write_unlock(mm); 22638d7071afSLinus Torvalds return NULL; 22648d7071afSLinus Torvalds 22658d7071afSLinus Torvalds success: 22668d7071afSLinus Torvalds mmap_write_downgrade(mm); 22678d7071afSLinus Torvalds return vma; 22688d7071afSLinus Torvalds } 2269e1d6d01aSJesse Barnes 22702c0b3814SHugh Dickins /* 2271763ecb03SLiam R. Howlett * Ok - we have the memory areas we should free on a maple tree so release them, 2272763ecb03SLiam R. Howlett * and do the vma updates. 22731da177e4SLinus Torvalds * 22742c0b3814SHugh Dickins * Called with the mm semaphore held. 22751da177e4SLinus Torvalds */ 2276763ecb03SLiam R. Howlett static inline void remove_mt(struct mm_struct *mm, struct ma_state *mas) 22771da177e4SLinus Torvalds { 22784f74d2c8SLinus Torvalds unsigned long nr_accounted = 0; 2279763ecb03SLiam R. Howlett struct vm_area_struct *vma; 22804f74d2c8SLinus Torvalds 2281365e9c87SHugh Dickins /* Update high watermark before we lower total_vm */ 2282365e9c87SHugh Dickins update_hiwater_vm(mm); 2283763ecb03SLiam R. Howlett mas_for_each(mas, vma, ULONG_MAX) { 2284ab50b8edSHugh Dickins long nrpages = vma_pages(vma); 22851da177e4SLinus Torvalds 22864f74d2c8SLinus Torvalds if (vma->vm_flags & VM_ACCOUNT) 22874f74d2c8SLinus Torvalds nr_accounted += nrpages; 228884638335SKonstantin Khlebnikov vm_stat_account(mm, vma->vm_flags, -nrpages); 22890d2ebf9cSSuren Baghdasaryan remove_vma(vma, false); 2290763ecb03SLiam R. Howlett } 22914f74d2c8SLinus Torvalds vm_unacct_memory(nr_accounted); 22921da177e4SLinus Torvalds } 22931da177e4SLinus Torvalds 22941da177e4SLinus Torvalds /* 22951da177e4SLinus Torvalds * Get rid of page table information in the indicated region. 22961da177e4SLinus Torvalds * 2297f10df686SPaolo 'Blaisorblade' Giarrusso * Called with the mm semaphore held. 22981da177e4SLinus Torvalds */ 2299fd892593SLiam R. Howlett static void unmap_region(struct mm_struct *mm, struct ma_state *mas, 2300e0da382cSHugh Dickins struct vm_area_struct *vma, struct vm_area_struct *prev, 2301fd892593SLiam R. Howlett struct vm_area_struct *next, unsigned long start, 2302fd892593SLiam R. Howlett unsigned long end, unsigned long tree_end, bool mm_wr_locked) 23031da177e4SLinus Torvalds { 2304d16dfc55SPeter Zijlstra struct mmu_gather tlb; 2305fd892593SLiam R. Howlett unsigned long mt_start = mas->index; 23061da177e4SLinus Torvalds 23071da177e4SLinus Torvalds lru_add_drain(); 2308a72afd87SWill Deacon tlb_gather_mmu(&tlb, mm); 2309365e9c87SHugh Dickins update_hiwater_rss(mm); 2310fd892593SLiam R. Howlett unmap_vmas(&tlb, mas, vma, start, end, tree_end, mm_wr_locked); 2311fd892593SLiam R. Howlett mas_set(mas, mt_start); 2312fd892593SLiam R. Howlett free_pgtables(&tlb, mas, vma, prev ? prev->vm_end : FIRST_USER_ADDRESS, 231398e51a22SSuren Baghdasaryan next ? next->vm_start : USER_PGTABLES_CEILING, 231498e51a22SSuren Baghdasaryan mm_wr_locked); 2315ae8eba8bSWill Deacon tlb_finish_mmu(&tlb); 23161da177e4SLinus Torvalds } 23171da177e4SLinus Torvalds 23181da177e4SLinus Torvalds /* 2319def5efe0SDavid Rientjes * __split_vma() bypasses sysctl_max_map_count checking. We use this where it 2320def5efe0SDavid Rientjes * has already been checked or doesn't make sense to fail. 23210fd5a9e2SLiam R. Howlett * VMA Iterator will point to the end VMA. 23221da177e4SLinus Torvalds */ 23239760ebffSLiam R. Howlett int __split_vma(struct vma_iterator *vmi, struct vm_area_struct *vma, 23241da177e4SLinus Torvalds unsigned long addr, int new_below) 23251da177e4SLinus Torvalds { 2326b2b3b886SLiam R. Howlett struct vma_prepare vp; 23271da177e4SLinus Torvalds struct vm_area_struct *new; 2328e3975891SChen Gang int err; 23299760ebffSLiam R. Howlett 2330b2b3b886SLiam R. Howlett WARN_ON(vma->vm_start >= addr); 2331b2b3b886SLiam R. Howlett WARN_ON(vma->vm_end <= addr); 2332b2b3b886SLiam R. Howlett 2333dd3b614fSDmitry Safonov if (vma->vm_ops && vma->vm_ops->may_split) { 2334dd3b614fSDmitry Safonov err = vma->vm_ops->may_split(vma, addr); 233531383c68SDan Williams if (err) 233631383c68SDan Williams return err; 233731383c68SDan Williams } 23381da177e4SLinus Torvalds 23393928d4f5SLinus Torvalds new = vm_area_dup(vma); 23401da177e4SLinus Torvalds if (!new) 2341e3975891SChen Gang return -ENOMEM; 23421da177e4SLinus Torvalds 2343b2b3b886SLiam R. Howlett if (new_below) { 23441da177e4SLinus Torvalds new->vm_end = addr; 2345b2b3b886SLiam R. Howlett } else { 23461da177e4SLinus Torvalds new->vm_start = addr; 23471da177e4SLinus Torvalds new->vm_pgoff += ((addr - vma->vm_start) >> PAGE_SHIFT); 23481da177e4SLinus Torvalds } 23491da177e4SLinus Torvalds 2350b5df0922SLiam R. Howlett err = -ENOMEM; 2351b5df0922SLiam R. Howlett vma_iter_config(vmi, new->vm_start, new->vm_end); 2352b5df0922SLiam R. Howlett if (vma_iter_prealloc(vmi, new)) 2353b5df0922SLiam R. Howlett goto out_free_vma; 2354b5df0922SLiam R. Howlett 2355ef0855d3SOleg Nesterov err = vma_dup_policy(vma, new); 2356ef0855d3SOleg Nesterov if (err) 2357b2b3b886SLiam R. Howlett goto out_free_vmi; 23581da177e4SLinus Torvalds 2359c4ea95d7SDaniel Forrest err = anon_vma_clone(new, vma); 2360c4ea95d7SDaniel Forrest if (err) 23615beb4930SRik van Riel goto out_free_mpol; 23625beb4930SRik van Riel 2363e9714acfSKonstantin Khlebnikov if (new->vm_file) 23641da177e4SLinus Torvalds get_file(new->vm_file); 23651da177e4SLinus Torvalds 23661da177e4SLinus Torvalds if (new->vm_ops && new->vm_ops->open) 23671da177e4SLinus Torvalds new->vm_ops->open(new); 23681da177e4SLinus Torvalds 2369b2b3b886SLiam R. Howlett init_vma_prep(&vp, vma); 2370b2b3b886SLiam R. Howlett vp.insert = new; 2371b2b3b886SLiam R. Howlett vma_prepare(&vp); 2372ccf1d78dSSuren Baghdasaryan vma_adjust_trans_huge(vma, vma->vm_start, addr, 0); 23731da177e4SLinus Torvalds 2374b2b3b886SLiam R. Howlett if (new_below) { 2375b2b3b886SLiam R. Howlett vma->vm_start = addr; 2376b2b3b886SLiam R. Howlett vma->vm_pgoff += (addr - new->vm_start) >> PAGE_SHIFT; 2377b2b3b886SLiam R. Howlett } else { 2378b2b3b886SLiam R. Howlett vma->vm_end = addr; 23799760ebffSLiam R. Howlett } 23805beb4930SRik van Riel 2381b2b3b886SLiam R. Howlett /* vma_complete stores the new vma */ 2382b2b3b886SLiam R. Howlett vma_complete(&vp, vmi, vma->vm_mm); 2383b2b3b886SLiam R. Howlett 2384b2b3b886SLiam R. Howlett /* Success. */ 2385b2b3b886SLiam R. Howlett if (new_below) 2386b2b3b886SLiam R. Howlett vma_next(vmi); 2387b2b3b886SLiam R. Howlett return 0; 2388b2b3b886SLiam R. Howlett 23895beb4930SRik van Riel out_free_mpol: 2390ef0855d3SOleg Nesterov mpol_put(vma_policy(new)); 2391b2b3b886SLiam R. Howlett out_free_vmi: 2392b2b3b886SLiam R. Howlett vma_iter_free(vmi); 23935beb4930SRik van Riel out_free_vma: 23943928d4f5SLinus Torvalds vm_area_free(new); 23955beb4930SRik van Riel return err; 23961da177e4SLinus Torvalds } 23971da177e4SLinus Torvalds 2398659ace58SKOSAKI Motohiro /* 2399659ace58SKOSAKI Motohiro * Split a vma into two pieces at address 'addr', a new vma is allocated 2400659ace58SKOSAKI Motohiro * either for the first part or the tail. 2401659ace58SKOSAKI Motohiro */ 24029760ebffSLiam R. Howlett int split_vma(struct vma_iterator *vmi, struct vm_area_struct *vma, 2403659ace58SKOSAKI Motohiro unsigned long addr, int new_below) 2404659ace58SKOSAKI Motohiro { 24059760ebffSLiam R. Howlett if (vma->vm_mm->map_count >= sysctl_max_map_count) 2406659ace58SKOSAKI Motohiro return -ENOMEM; 2407659ace58SKOSAKI Motohiro 24089760ebffSLiam R. Howlett return __split_vma(vmi, vma, addr, new_below); 2409f2ebfe43SLiam R. Howlett } 2410f2ebfe43SLiam R. Howlett 241111f9a21aSLiam R. Howlett /* 2412183654ceSLiam R. Howlett * do_vmi_align_munmap() - munmap the aligned region from @start to @end. 2413183654ceSLiam R. Howlett * @vmi: The vma iterator 241411f9a21aSLiam R. Howlett * @vma: The starting vm_area_struct 241511f9a21aSLiam R. Howlett * @mm: The mm_struct 241611f9a21aSLiam R. Howlett * @start: The aligned start address to munmap. 241711f9a21aSLiam R. Howlett * @end: The aligned end address to munmap. 241811f9a21aSLiam R. Howlett * @uf: The userfaultfd list_head 2419408579cdSLiam R. Howlett * @unlock: Set to true to drop the mmap_lock. unlocking only happens on 2420408579cdSLiam R. Howlett * success. 242111f9a21aSLiam R. Howlett * 2422408579cdSLiam R. Howlett * Return: 0 on success and drops the lock if so directed, error and leaves the 2423408579cdSLiam R. Howlett * lock held otherwise. 242411f9a21aSLiam R. Howlett */ 242511f9a21aSLiam R. Howlett static int 2426183654ceSLiam R. Howlett do_vmi_align_munmap(struct vma_iterator *vmi, struct vm_area_struct *vma, 242711f9a21aSLiam R. Howlett struct mm_struct *mm, unsigned long start, 2428408579cdSLiam R. Howlett unsigned long end, struct list_head *uf, bool unlock) 242911f9a21aSLiam R. Howlett { 2430763ecb03SLiam R. Howlett struct vm_area_struct *prev, *next = NULL; 2431763ecb03SLiam R. Howlett struct maple_tree mt_detach; 2432763ecb03SLiam R. Howlett int count = 0; 2433d4af56c5SLiam R. Howlett int error = -ENOMEM; 2434606c812eSLiam R. Howlett unsigned long locked_vm = 0; 2435763ecb03SLiam R. Howlett MA_STATE(mas_detach, &mt_detach, 0, 0); 24363dd44325SLiam R. Howlett mt_init_flags(&mt_detach, vmi->mas.tree->ma_flags & MT_FLAGS_LOCK_MASK); 243702fdb25fSLiam R. Howlett mt_on_stack(mt_detach); 2438524e00b3SLiam R. Howlett 24391da177e4SLinus Torvalds /* 24401da177e4SLinus Torvalds * If we need to split any vma, do it now to save pain later. 24411da177e4SLinus Torvalds * 24421da177e4SLinus Torvalds * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially 24431da177e4SLinus Torvalds * unmapped vm_area_struct will remain in use: so lower split_vma 24441da177e4SLinus Torvalds * places tmp vma above, and higher split_vma places tmp vma below. 24451da177e4SLinus Torvalds */ 2446763ecb03SLiam R. Howlett 2447763ecb03SLiam R. Howlett /* Does it split the first one? */ 2448146425a3SHugh Dickins if (start > vma->vm_start) { 2449659ace58SKOSAKI Motohiro 2450659ace58SKOSAKI Motohiro /* 2451659ace58SKOSAKI Motohiro * Make sure that map_count on return from munmap() will 2452659ace58SKOSAKI Motohiro * not exceed its limit; but let map_count go just above 2453659ace58SKOSAKI Motohiro * its limit temporarily, to help free resources as expected. 2454659ace58SKOSAKI Motohiro */ 2455659ace58SKOSAKI Motohiro if (end < vma->vm_end && mm->map_count >= sysctl_max_map_count) 2456d4af56c5SLiam R. Howlett goto map_count_exceeded; 2457659ace58SKOSAKI Motohiro 24586935e052SLiam R. Howlett error = __split_vma(vmi, vma, start, 1); 24591da177e4SLinus Torvalds if (error) 2460763ecb03SLiam R. Howlett goto start_split_failed; 24611da177e4SLinus Torvalds } 24621da177e4SLinus Torvalds 2463763ecb03SLiam R. Howlett /* 2464763ecb03SLiam R. Howlett * Detach a range of VMAs from the mm. Using next as a temp variable as 2465763ecb03SLiam R. Howlett * it is always overwritten. 2466763ecb03SLiam R. Howlett */ 24676935e052SLiam R. Howlett next = vma; 24686935e052SLiam R. Howlett do { 2469763ecb03SLiam R. Howlett /* Does it split the end? */ 2470763ecb03SLiam R. Howlett if (next->vm_end > end) { 24716b73cff2SLiam R. Howlett error = __split_vma(vmi, next, end, 0); 24721da177e4SLinus Torvalds if (error) 2473763ecb03SLiam R. Howlett goto end_split_failed; 247411f9a21aSLiam R. Howlett } 2475606c812eSLiam R. Howlett vma_start_write(next); 2476fd892593SLiam R. Howlett mas_set(&mas_detach, count); 24776c26bd43SDavid Woodhouse error = mas_store_gfp(&mas_detach, next, GFP_KERNEL); 24786c26bd43SDavid Woodhouse if (error) 2479606c812eSLiam R. Howlett goto munmap_gather_failed; 2480606c812eSLiam R. Howlett vma_mark_detached(next, true); 2481606c812eSLiam R. Howlett if (next->vm_flags & VM_LOCKED) 2482606c812eSLiam R. Howlett locked_vm += vma_pages(next); 2483763ecb03SLiam R. Howlett 2484763ecb03SLiam R. Howlett count++; 24852376dd7cSAndrea Arcangeli if (unlikely(uf)) { 24862376dd7cSAndrea Arcangeli /* 24872376dd7cSAndrea Arcangeli * If userfaultfd_unmap_prep returns an error the vmas 2488f0953a1bSIngo Molnar * will remain split, but userland will get a 24892376dd7cSAndrea Arcangeli * highly unexpected error anyway. This is no 24902376dd7cSAndrea Arcangeli * different than the case where the first of the two 24912376dd7cSAndrea Arcangeli * __split_vma fails, but we don't undo the first 24922376dd7cSAndrea Arcangeli * split, despite we could. This is unlikely enough 24932376dd7cSAndrea Arcangeli * failure that it's not worth optimizing it for. 24942376dd7cSAndrea Arcangeli */ 249565ac1320SLiam R. Howlett error = userfaultfd_unmap_prep(next, start, end, uf); 249611f9a21aSLiam R. Howlett 24972376dd7cSAndrea Arcangeli if (error) 2498d4af56c5SLiam R. Howlett goto userfaultfd_error; 24992376dd7cSAndrea Arcangeli } 2500ba470de4SRik van Riel #ifdef CONFIG_DEBUG_VM_MAPLE_TREE 2501ba470de4SRik van Riel BUG_ON(next->vm_start < start); 2502ba470de4SRik van Riel BUG_ON(next->vm_start > end); 25031da177e4SLinus Torvalds #endif 25046935e052SLiam R. Howlett } for_each_vma_range(*vmi, next, end); 25052376dd7cSAndrea Arcangeli 2506763ecb03SLiam R. Howlett #if defined(CONFIG_DEBUG_VM_MAPLE_TREE) 2507763ecb03SLiam R. Howlett /* Make sure no VMAs are about to be lost. */ 2508763ecb03SLiam R. Howlett { 2509fd892593SLiam R. Howlett MA_STATE(test, &mt_detach, 0, 0); 2510763ecb03SLiam R. Howlett struct vm_area_struct *vma_mas, *vma_test; 2511763ecb03SLiam R. Howlett int test_count = 0; 2512763ecb03SLiam R. Howlett 2513183654ceSLiam R. Howlett vma_iter_set(vmi, start); 2514763ecb03SLiam R. Howlett rcu_read_lock(); 2515fd892593SLiam R. Howlett vma_test = mas_find(&test, count - 1); 2516183654ceSLiam R. Howlett for_each_vma_range(*vmi, vma_mas, end) { 2517763ecb03SLiam R. Howlett BUG_ON(vma_mas != vma_test); 2518763ecb03SLiam R. Howlett test_count++; 2519fd892593SLiam R. Howlett vma_test = mas_next(&test, count - 1); 2520763ecb03SLiam R. Howlett } 2521763ecb03SLiam R. Howlett rcu_read_unlock(); 2522763ecb03SLiam R. Howlett BUG_ON(count != test_count); 2523763ecb03SLiam R. Howlett } 2524763ecb03SLiam R. Howlett #endif 25256935e052SLiam R. Howlett 25266935e052SLiam R. Howlett while (vma_iter_addr(vmi) > start) 25276935e052SLiam R. Howlett vma_iter_prev_range(vmi); 25286935e052SLiam R. Howlett 25296c26bd43SDavid Woodhouse error = vma_iter_clear_gfp(vmi, start, end, GFP_KERNEL); 25306c26bd43SDavid Woodhouse if (error) 2531606c812eSLiam R. Howlett goto clear_tree_failed; 25320378c0a0SLiam R. Howlett 25336c26bd43SDavid Woodhouse /* Point of no return */ 2534606c812eSLiam R. Howlett mm->locked_vm -= locked_vm; 2535763ecb03SLiam R. Howlett mm->map_count -= count; 2536408579cdSLiam R. Howlett if (unlock) 2537d8ed45c5SMichel Lespinasse mmap_write_downgrade(mm); 2538dd2283f2SYang Shi 25396935e052SLiam R. Howlett prev = vma_iter_prev_range(vmi); 25406935e052SLiam R. Howlett next = vma_next(vmi); 25416935e052SLiam R. Howlett if (next) 25426935e052SLiam R. Howlett vma_iter_prev_range(vmi); 25436935e052SLiam R. Howlett 254468f48381SSuren Baghdasaryan /* 254568f48381SSuren Baghdasaryan * We can free page tables without write-locking mmap_lock because VMAs 254668f48381SSuren Baghdasaryan * were isolated before we downgraded mmap_lock. 254768f48381SSuren Baghdasaryan */ 2548fd892593SLiam R. Howlett mas_set(&mas_detach, 1); 2549fd892593SLiam R. Howlett unmap_region(mm, &mas_detach, vma, prev, next, start, end, count, 2550fd892593SLiam R. Howlett !unlock); 2551763ecb03SLiam R. Howlett /* Statistics and freeing VMAs */ 2552fd892593SLiam R. Howlett mas_set(&mas_detach, 0); 2553763ecb03SLiam R. Howlett remove_mt(mm, &mas_detach); 2554ae80b404SLinus Torvalds validate_mm(mm); 2555408579cdSLiam R. Howlett if (unlock) 2556408579cdSLiam R. Howlett mmap_read_unlock(mm); 25571da177e4SLinus Torvalds 255802fdb25fSLiam R. Howlett __mt_destroy(&mt_detach); 2559408579cdSLiam R. Howlett return 0; 2560d4af56c5SLiam R. Howlett 2561606c812eSLiam R. Howlett clear_tree_failed: 2562d4af56c5SLiam R. Howlett userfaultfd_error: 2563606c812eSLiam R. Howlett munmap_gather_failed: 2564763ecb03SLiam R. Howlett end_split_failed: 2565606c812eSLiam R. Howlett mas_set(&mas_detach, 0); 2566606c812eSLiam R. Howlett mas_for_each(&mas_detach, next, end) 2567606c812eSLiam R. Howlett vma_mark_detached(next, false); 2568606c812eSLiam R. Howlett 2569763ecb03SLiam R. Howlett __mt_destroy(&mt_detach); 2570763ecb03SLiam R. Howlett start_split_failed: 2571763ecb03SLiam R. Howlett map_count_exceeded: 2572b5641a5dSLinus Torvalds validate_mm(mm); 2573d4af56c5SLiam R. Howlett return error; 25741da177e4SLinus Torvalds } 25751da177e4SLinus Torvalds 257611f9a21aSLiam R. Howlett /* 2577183654ceSLiam R. Howlett * do_vmi_munmap() - munmap a given range. 2578183654ceSLiam R. Howlett * @vmi: The vma iterator 257911f9a21aSLiam R. Howlett * @mm: The mm_struct 258011f9a21aSLiam R. Howlett * @start: The start address to munmap 258111f9a21aSLiam R. Howlett * @len: The length of the range to munmap 258211f9a21aSLiam R. Howlett * @uf: The userfaultfd list_head 2583408579cdSLiam R. Howlett * @unlock: set to true if the user wants to drop the mmap_lock on success 258411f9a21aSLiam R. Howlett * 258511f9a21aSLiam R. Howlett * This function takes a @mas that is either pointing to the previous VMA or set 258611f9a21aSLiam R. Howlett * to MA_START and sets it up to remove the mapping(s). The @len will be 258711f9a21aSLiam R. Howlett * aligned and any arch_unmap work will be preformed. 258811f9a21aSLiam R. Howlett * 2589408579cdSLiam R. Howlett * Return: 0 on success and drops the lock if so directed, error and leaves the 2590408579cdSLiam R. Howlett * lock held otherwise. 259111f9a21aSLiam R. Howlett */ 2592183654ceSLiam R. Howlett int do_vmi_munmap(struct vma_iterator *vmi, struct mm_struct *mm, 259311f9a21aSLiam R. Howlett unsigned long start, size_t len, struct list_head *uf, 2594408579cdSLiam R. Howlett bool unlock) 259511f9a21aSLiam R. Howlett { 259611f9a21aSLiam R. Howlett unsigned long end; 259711f9a21aSLiam R. Howlett struct vm_area_struct *vma; 259811f9a21aSLiam R. Howlett 259911f9a21aSLiam R. Howlett if ((offset_in_page(start)) || start > TASK_SIZE || len > TASK_SIZE-start) 260011f9a21aSLiam R. Howlett return -EINVAL; 260111f9a21aSLiam R. Howlett 260211f9a21aSLiam R. Howlett end = start + PAGE_ALIGN(len); 260311f9a21aSLiam R. Howlett if (end == start) 260411f9a21aSLiam R. Howlett return -EINVAL; 260511f9a21aSLiam R. Howlett 260611f9a21aSLiam R. Howlett /* arch_unmap() might do unmaps itself. */ 260711f9a21aSLiam R. Howlett arch_unmap(mm, start, end); 260811f9a21aSLiam R. Howlett 260911f9a21aSLiam R. Howlett /* Find the first overlapping VMA */ 2610183654ceSLiam R. Howlett vma = vma_find(vmi, end); 2611408579cdSLiam R. Howlett if (!vma) { 2612408579cdSLiam R. Howlett if (unlock) 2613408579cdSLiam R. Howlett mmap_write_unlock(mm); 261411f9a21aSLiam R. Howlett return 0; 2615408579cdSLiam R. Howlett } 261611f9a21aSLiam R. Howlett 2617408579cdSLiam R. Howlett return do_vmi_align_munmap(vmi, vma, mm, start, end, uf, unlock); 261811f9a21aSLiam R. Howlett } 261911f9a21aSLiam R. Howlett 262011f9a21aSLiam R. Howlett /* do_munmap() - Wrapper function for non-maple tree aware do_munmap() calls. 262111f9a21aSLiam R. Howlett * @mm: The mm_struct 262211f9a21aSLiam R. Howlett * @start: The start address to munmap 262311f9a21aSLiam R. Howlett * @len: The length to be munmapped. 262411f9a21aSLiam R. Howlett * @uf: The userfaultfd list_head 2625408579cdSLiam R. Howlett * 2626408579cdSLiam R. Howlett * Return: 0 on success, error otherwise. 262711f9a21aSLiam R. Howlett */ 2628dd2283f2SYang Shi int do_munmap(struct mm_struct *mm, unsigned long start, size_t len, 2629dd2283f2SYang Shi struct list_head *uf) 2630dd2283f2SYang Shi { 2631183654ceSLiam R. Howlett VMA_ITERATOR(vmi, mm, start); 263211f9a21aSLiam R. Howlett 2633183654ceSLiam R. Howlett return do_vmi_munmap(&vmi, mm, start, len, uf, false); 2634dd2283f2SYang Shi } 2635dd2283f2SYang Shi 2636e99668a5SLiam R. Howlett unsigned long mmap_region(struct file *file, unsigned long addr, 2637e99668a5SLiam R. Howlett unsigned long len, vm_flags_t vm_flags, unsigned long pgoff, 2638e99668a5SLiam R. Howlett struct list_head *uf) 2639e99668a5SLiam R. Howlett { 2640e99668a5SLiam R. Howlett struct mm_struct *mm = current->mm; 2641e99668a5SLiam R. Howlett struct vm_area_struct *vma = NULL; 2642e99668a5SLiam R. Howlett struct vm_area_struct *next, *prev, *merge; 2643e99668a5SLiam R. Howlett pgoff_t pglen = len >> PAGE_SHIFT; 2644e99668a5SLiam R. Howlett unsigned long charged = 0; 2645e99668a5SLiam R. Howlett unsigned long end = addr + len; 2646e99668a5SLiam R. Howlett unsigned long merge_start = addr, merge_end = end; 2647e99668a5SLiam R. Howlett pgoff_t vm_pgoff; 2648e99668a5SLiam R. Howlett int error; 2649183654ceSLiam R. Howlett VMA_ITERATOR(vmi, mm, addr); 2650e99668a5SLiam R. Howlett 2651e99668a5SLiam R. Howlett /* Check against address space limit. */ 2652e99668a5SLiam R. Howlett if (!may_expand_vm(mm, vm_flags, len >> PAGE_SHIFT)) { 2653e99668a5SLiam R. Howlett unsigned long nr_pages; 2654e99668a5SLiam R. Howlett 2655e99668a5SLiam R. Howlett /* 2656e99668a5SLiam R. Howlett * MAP_FIXED may remove pages of mappings that intersects with 2657e99668a5SLiam R. Howlett * requested mapping. Account for the pages it would unmap. 2658e99668a5SLiam R. Howlett */ 2659e99668a5SLiam R. Howlett nr_pages = count_vma_pages_range(mm, addr, end); 2660e99668a5SLiam R. Howlett 2661e99668a5SLiam R. Howlett if (!may_expand_vm(mm, vm_flags, 2662e99668a5SLiam R. Howlett (len >> PAGE_SHIFT) - nr_pages)) 2663e99668a5SLiam R. Howlett return -ENOMEM; 2664e99668a5SLiam R. Howlett } 2665e99668a5SLiam R. Howlett 2666e99668a5SLiam R. Howlett /* Unmap any existing mapping in the area */ 2667183654ceSLiam R. Howlett if (do_vmi_munmap(&vmi, mm, addr, len, uf, false)) 2668e99668a5SLiam R. Howlett return -ENOMEM; 2669e99668a5SLiam R. Howlett 2670e99668a5SLiam R. Howlett /* 2671e99668a5SLiam R. Howlett * Private writable mapping: check memory availability 2672e99668a5SLiam R. Howlett */ 2673e99668a5SLiam R. Howlett if (accountable_mapping(file, vm_flags)) { 2674e99668a5SLiam R. Howlett charged = len >> PAGE_SHIFT; 2675e99668a5SLiam R. Howlett if (security_vm_enough_memory_mm(mm, charged)) 2676e99668a5SLiam R. Howlett return -ENOMEM; 2677e99668a5SLiam R. Howlett vm_flags |= VM_ACCOUNT; 2678e99668a5SLiam R. Howlett } 2679e99668a5SLiam R. Howlett 2680183654ceSLiam R. Howlett next = vma_next(&vmi); 2681183654ceSLiam R. Howlett prev = vma_prev(&vmi); 268253bee98dSLiam R. Howlett if (vm_flags & VM_SPECIAL) { 268353bee98dSLiam R. Howlett if (prev) 268453bee98dSLiam R. Howlett vma_iter_next_range(&vmi); 2685e99668a5SLiam R. Howlett goto cannot_expand; 268653bee98dSLiam R. Howlett } 2687e99668a5SLiam R. Howlett 2688e99668a5SLiam R. Howlett /* Attempt to expand an old mapping */ 2689e99668a5SLiam R. Howlett /* Check next */ 2690e99668a5SLiam R. Howlett if (next && next->vm_start == end && !vma_policy(next) && 2691e99668a5SLiam R. Howlett can_vma_merge_before(next, vm_flags, NULL, file, pgoff+pglen, 2692e99668a5SLiam R. Howlett NULL_VM_UFFD_CTX, NULL)) { 2693e99668a5SLiam R. Howlett merge_end = next->vm_end; 2694e99668a5SLiam R. Howlett vma = next; 2695e99668a5SLiam R. Howlett vm_pgoff = next->vm_pgoff - pglen; 2696e99668a5SLiam R. Howlett } 2697e99668a5SLiam R. Howlett 2698e99668a5SLiam R. Howlett /* Check prev */ 2699e99668a5SLiam R. Howlett if (prev && prev->vm_end == addr && !vma_policy(prev) && 2700e99668a5SLiam R. Howlett (vma ? can_vma_merge_after(prev, vm_flags, vma->anon_vma, file, 2701e99668a5SLiam R. Howlett pgoff, vma->vm_userfaultfd_ctx, NULL) : 2702e99668a5SLiam R. Howlett can_vma_merge_after(prev, vm_flags, NULL, file, pgoff, 2703e99668a5SLiam R. Howlett NULL_VM_UFFD_CTX, NULL))) { 2704e99668a5SLiam R. Howlett merge_start = prev->vm_start; 2705e99668a5SLiam R. Howlett vma = prev; 2706e99668a5SLiam R. Howlett vm_pgoff = prev->vm_pgoff; 270753bee98dSLiam R. Howlett } else if (prev) { 270853bee98dSLiam R. Howlett vma_iter_next_range(&vmi); 2709e99668a5SLiam R. Howlett } 2710e99668a5SLiam R. Howlett 2711e99668a5SLiam R. Howlett /* Actually expand, if possible */ 2712e99668a5SLiam R. Howlett if (vma && 27133c441ab7SLiam R. Howlett !vma_expand(&vmi, vma, merge_start, merge_end, vm_pgoff, next)) { 2714e99668a5SLiam R. Howlett khugepaged_enter_vma(vma, vm_flags); 2715e99668a5SLiam R. Howlett goto expanded; 2716e99668a5SLiam R. Howlett } 2717e99668a5SLiam R. Howlett 271853bee98dSLiam R. Howlett if (vma == prev) 271953bee98dSLiam R. Howlett vma_iter_set(&vmi, addr); 2720e99668a5SLiam R. Howlett cannot_expand: 27215c1c03deSLiam R. Howlett 2722e99668a5SLiam R. Howlett /* 2723e99668a5SLiam R. Howlett * Determine the object being mapped and call the appropriate 2724e99668a5SLiam R. Howlett * specific mapper. the address has already been validated, but 2725e99668a5SLiam R. Howlett * not unmapped, but the maps are removed from the list. 2726e99668a5SLiam R. Howlett */ 2727e99668a5SLiam R. Howlett vma = vm_area_alloc(mm); 2728e99668a5SLiam R. Howlett if (!vma) { 2729e99668a5SLiam R. Howlett error = -ENOMEM; 2730e99668a5SLiam R. Howlett goto unacct_error; 2731e99668a5SLiam R. Howlett } 2732e99668a5SLiam R. Howlett 273353bee98dSLiam R. Howlett vma_iter_config(&vmi, addr, end); 2734e99668a5SLiam R. Howlett vma->vm_start = addr; 2735e99668a5SLiam R. Howlett vma->vm_end = end; 27361c71222eSSuren Baghdasaryan vm_flags_init(vma, vm_flags); 2737e99668a5SLiam R. Howlett vma->vm_page_prot = vm_get_page_prot(vm_flags); 2738e99668a5SLiam R. Howlett vma->vm_pgoff = pgoff; 2739e99668a5SLiam R. Howlett 2740e99668a5SLiam R. Howlett if (file) { 2741e99668a5SLiam R. Howlett if (vm_flags & VM_SHARED) { 2742e99668a5SLiam R. Howlett error = mapping_map_writable(file->f_mapping); 2743e99668a5SLiam R. Howlett if (error) 2744e99668a5SLiam R. Howlett goto free_vma; 2745e99668a5SLiam R. Howlett } 2746e99668a5SLiam R. Howlett 2747e99668a5SLiam R. Howlett vma->vm_file = get_file(file); 2748e99668a5SLiam R. Howlett error = call_mmap(file, vma); 2749e99668a5SLiam R. Howlett if (error) 2750e99668a5SLiam R. Howlett goto unmap_and_free_vma; 2751e99668a5SLiam R. Howlett 2752a57b7051SLiam Howlett /* 2753a57b7051SLiam Howlett * Expansion is handled above, merging is handled below. 2754a57b7051SLiam Howlett * Drivers should not alter the address of the VMA. 2755e99668a5SLiam R. Howlett */ 2756a57b7051SLiam Howlett error = -EINVAL; 2757cc8d1b09SLiam R. Howlett if (WARN_ON((addr != vma->vm_start))) 2758a57b7051SLiam Howlett goto close_and_free_vma; 2759e99668a5SLiam R. Howlett 276053bee98dSLiam R. Howlett vma_iter_config(&vmi, addr, end); 2761e99668a5SLiam R. Howlett /* 2762e99668a5SLiam R. Howlett * If vm_flags changed after call_mmap(), we should try merge 2763e99668a5SLiam R. Howlett * vma again as we may succeed this time. 2764e99668a5SLiam R. Howlett */ 2765e99668a5SLiam R. Howlett if (unlikely(vm_flags != vma->vm_flags && prev)) { 27669760ebffSLiam R. Howlett merge = vma_merge(&vmi, mm, prev, vma->vm_start, 27679760ebffSLiam R. Howlett vma->vm_end, vma->vm_flags, NULL, 27689760ebffSLiam R. Howlett vma->vm_file, vma->vm_pgoff, NULL, 27699760ebffSLiam R. Howlett NULL_VM_UFFD_CTX, NULL); 2770e99668a5SLiam R. Howlett if (merge) { 2771e99668a5SLiam R. Howlett /* 2772e99668a5SLiam R. Howlett * ->mmap() can change vma->vm_file and fput 2773e99668a5SLiam R. Howlett * the original file. So fput the vma->vm_file 2774e99668a5SLiam R. Howlett * here or we would add an extra fput for file 2775e99668a5SLiam R. Howlett * and cause general protection fault 2776e99668a5SLiam R. Howlett * ultimately. 2777e99668a5SLiam R. Howlett */ 2778e99668a5SLiam R. Howlett fput(vma->vm_file); 2779e99668a5SLiam R. Howlett vm_area_free(vma); 2780e99668a5SLiam R. Howlett vma = merge; 2781e99668a5SLiam R. Howlett /* Update vm_flags to pick up the change. */ 2782e99668a5SLiam R. Howlett vm_flags = vma->vm_flags; 2783e99668a5SLiam R. Howlett goto unmap_writable; 2784e99668a5SLiam R. Howlett } 2785e99668a5SLiam R. Howlett } 2786e99668a5SLiam R. Howlett 2787e99668a5SLiam R. Howlett vm_flags = vma->vm_flags; 2788e99668a5SLiam R. Howlett } else if (vm_flags & VM_SHARED) { 2789e99668a5SLiam R. Howlett error = shmem_zero_setup(vma); 2790e99668a5SLiam R. Howlett if (error) 2791e99668a5SLiam R. Howlett goto free_vma; 2792e99668a5SLiam R. Howlett } else { 2793e99668a5SLiam R. Howlett vma_set_anonymous(vma); 2794e99668a5SLiam R. Howlett } 2795e99668a5SLiam R. Howlett 2796b507808eSJoey Gouly if (map_deny_write_exec(vma, vma->vm_flags)) { 2797b507808eSJoey Gouly error = -EACCES; 2798b507808eSJoey Gouly goto close_and_free_vma; 2799b507808eSJoey Gouly } 2800b507808eSJoey Gouly 2801e99668a5SLiam R. Howlett /* Allow architectures to sanity-check the vm_flags */ 2802e99668a5SLiam R. Howlett error = -EINVAL; 2803cc8d1b09SLiam R. Howlett if (!arch_validate_flags(vma->vm_flags)) 2804deb0f656SCarlos Llamas goto close_and_free_vma; 2805e99668a5SLiam R. Howlett 2806e99668a5SLiam R. Howlett error = -ENOMEM; 2807b5df0922SLiam R. Howlett if (vma_iter_prealloc(&vmi, vma)) 28085789151eSMike Kravetz goto close_and_free_vma; 2809e99668a5SLiam R. Howlett 28101c7873e3SHugh Dickins /* Lock the VMA since it is modified after insertion into VMA tree */ 28111c7873e3SHugh Dickins vma_start_write(vma); 2812183654ceSLiam R. Howlett vma_iter_store(&vmi, vma); 2813e99668a5SLiam R. Howlett mm->map_count++; 2814e99668a5SLiam R. Howlett if (vma->vm_file) { 28156852c46cSYu Ma i_mmap_lock_write(vma->vm_file->f_mapping); 2816e99668a5SLiam R. Howlett if (vma->vm_flags & VM_SHARED) 2817e99668a5SLiam R. Howlett mapping_allow_writable(vma->vm_file->f_mapping); 2818e99668a5SLiam R. Howlett 2819e99668a5SLiam R. Howlett flush_dcache_mmap_lock(vma->vm_file->f_mapping); 2820e99668a5SLiam R. Howlett vma_interval_tree_insert(vma, &vma->vm_file->f_mapping->i_mmap); 2821e99668a5SLiam R. Howlett flush_dcache_mmap_unlock(vma->vm_file->f_mapping); 2822e99668a5SLiam R. Howlett i_mmap_unlock_write(vma->vm_file->f_mapping); 2823e99668a5SLiam R. Howlett } 2824e99668a5SLiam R. Howlett 2825e99668a5SLiam R. Howlett /* 2826e99668a5SLiam R. Howlett * vma_merge() calls khugepaged_enter_vma() either, the below 2827e99668a5SLiam R. Howlett * call covers the non-merge case. 2828e99668a5SLiam R. Howlett */ 2829e99668a5SLiam R. Howlett khugepaged_enter_vma(vma, vma->vm_flags); 2830e99668a5SLiam R. Howlett 2831e99668a5SLiam R. Howlett /* Once vma denies write, undo our temporary denial count */ 2832e99668a5SLiam R. Howlett unmap_writable: 2833e99668a5SLiam R. Howlett if (file && vm_flags & VM_SHARED) 2834e99668a5SLiam R. Howlett mapping_unmap_writable(file->f_mapping); 2835e99668a5SLiam R. Howlett file = vma->vm_file; 2836d7597f59SStefan Roesch ksm_add_vma(vma); 2837e99668a5SLiam R. Howlett expanded: 2838e99668a5SLiam R. Howlett perf_event_mmap(vma); 2839e99668a5SLiam R. Howlett 2840e99668a5SLiam R. Howlett vm_stat_account(mm, vm_flags, len >> PAGE_SHIFT); 2841e99668a5SLiam R. Howlett if (vm_flags & VM_LOCKED) { 2842e99668a5SLiam R. Howlett if ((vm_flags & VM_SPECIAL) || vma_is_dax(vma) || 2843e99668a5SLiam R. Howlett is_vm_hugetlb_page(vma) || 2844e99668a5SLiam R. Howlett vma == get_gate_vma(current->mm)) 2845e430a95aSSuren Baghdasaryan vm_flags_clear(vma, VM_LOCKED_MASK); 2846e99668a5SLiam R. Howlett else 2847e99668a5SLiam R. Howlett mm->locked_vm += (len >> PAGE_SHIFT); 2848e99668a5SLiam R. Howlett } 2849e99668a5SLiam R. Howlett 2850e99668a5SLiam R. Howlett if (file) 2851e99668a5SLiam R. Howlett uprobe_mmap(vma); 2852e99668a5SLiam R. Howlett 2853e99668a5SLiam R. Howlett /* 2854e99668a5SLiam R. Howlett * New (or expanded) vma always get soft dirty status. 2855e99668a5SLiam R. Howlett * Otherwise user-space soft-dirty page tracker won't 2856e99668a5SLiam R. Howlett * be able to distinguish situation when vma area unmapped, 2857e99668a5SLiam R. Howlett * then new mapped in-place (which must be aimed as 2858e99668a5SLiam R. Howlett * a completely new data area). 2859e99668a5SLiam R. Howlett */ 28601c71222eSSuren Baghdasaryan vm_flags_set(vma, VM_SOFTDIRTY); 2861e99668a5SLiam R. Howlett 2862e99668a5SLiam R. Howlett vma_set_page_prot(vma); 2863e99668a5SLiam R. Howlett 2864e99668a5SLiam R. Howlett validate_mm(mm); 2865e99668a5SLiam R. Howlett return addr; 2866e99668a5SLiam R. Howlett 2867deb0f656SCarlos Llamas close_and_free_vma: 2868cc8d1b09SLiam R. Howlett if (file && vma->vm_ops && vma->vm_ops->close) 2869deb0f656SCarlos Llamas vma->vm_ops->close(vma); 2870cc8d1b09SLiam R. Howlett 2871cc8d1b09SLiam R. Howlett if (file || vma->vm_file) { 2872e99668a5SLiam R. Howlett unmap_and_free_vma: 2873e99668a5SLiam R. Howlett fput(vma->vm_file); 2874e99668a5SLiam R. Howlett vma->vm_file = NULL; 2875e99668a5SLiam R. Howlett 2876fd892593SLiam R. Howlett vma_iter_set(&vmi, vma->vm_end); 2877e99668a5SLiam R. Howlett /* Undo any partial mapping done by a device driver. */ 2878fd892593SLiam R. Howlett unmap_region(mm, &vmi.mas, vma, prev, next, vma->vm_start, 2879fd892593SLiam R. Howlett vma->vm_end, vma->vm_end, true); 2880cc8d1b09SLiam R. Howlett } 2881cc674ab3SLi Zetao if (file && (vm_flags & VM_SHARED)) 2882e99668a5SLiam R. Howlett mapping_unmap_writable(file->f_mapping); 2883e99668a5SLiam R. Howlett free_vma: 2884e99668a5SLiam R. Howlett vm_area_free(vma); 2885e99668a5SLiam R. Howlett unacct_error: 2886e99668a5SLiam R. Howlett if (charged) 2887e99668a5SLiam R. Howlett vm_unacct_memory(charged); 2888e99668a5SLiam R. Howlett validate_mm(mm); 2889e99668a5SLiam R. Howlett return error; 2890e99668a5SLiam R. Howlett } 2891e99668a5SLiam R. Howlett 2892408579cdSLiam R. Howlett static int __vm_munmap(unsigned long start, size_t len, bool unlock) 2893a46ef99dSLinus Torvalds { 2894a46ef99dSLinus Torvalds int ret; 2895bfce281cSAl Viro struct mm_struct *mm = current->mm; 2896897ab3e0SMike Rapoport LIST_HEAD(uf); 2897183654ceSLiam R. Howlett VMA_ITERATOR(vmi, mm, start); 2898a46ef99dSLinus Torvalds 2899d8ed45c5SMichel Lespinasse if (mmap_write_lock_killable(mm)) 2900ae798783SMichal Hocko return -EINTR; 2901ae798783SMichal Hocko 2902408579cdSLiam R. Howlett ret = do_vmi_munmap(&vmi, mm, start, len, &uf, unlock); 2903408579cdSLiam R. Howlett if (ret || !unlock) 2904d8ed45c5SMichel Lespinasse mmap_write_unlock(mm); 2905dd2283f2SYang Shi 2906897ab3e0SMike Rapoport userfaultfd_unmap_complete(mm, &uf); 2907a46ef99dSLinus Torvalds return ret; 2908a46ef99dSLinus Torvalds } 2909dd2283f2SYang Shi 2910dd2283f2SYang Shi int vm_munmap(unsigned long start, size_t len) 2911dd2283f2SYang Shi { 2912dd2283f2SYang Shi return __vm_munmap(start, len, false); 2913dd2283f2SYang Shi } 2914a46ef99dSLinus Torvalds EXPORT_SYMBOL(vm_munmap); 2915a46ef99dSLinus Torvalds 29166a6160a7SHeiko Carstens SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len) 29171da177e4SLinus Torvalds { 2918ce18d171SCatalin Marinas addr = untagged_addr(addr); 2919dd2283f2SYang Shi return __vm_munmap(addr, len, true); 29201da177e4SLinus Torvalds } 29211da177e4SLinus Torvalds 2922c8d78c18SKirill A. Shutemov 2923c8d78c18SKirill A. Shutemov /* 2924c8d78c18SKirill A. Shutemov * Emulation of deprecated remap_file_pages() syscall. 2925c8d78c18SKirill A. Shutemov */ 2926c8d78c18SKirill A. Shutemov SYSCALL_DEFINE5(remap_file_pages, unsigned long, start, unsigned long, size, 2927c8d78c18SKirill A. Shutemov unsigned long, prot, unsigned long, pgoff, unsigned long, flags) 2928c8d78c18SKirill A. Shutemov { 2929c8d78c18SKirill A. Shutemov 2930c8d78c18SKirill A. Shutemov struct mm_struct *mm = current->mm; 2931c8d78c18SKirill A. Shutemov struct vm_area_struct *vma; 2932c8d78c18SKirill A. Shutemov unsigned long populate = 0; 2933c8d78c18SKirill A. Shutemov unsigned long ret = -EINVAL; 2934c8d78c18SKirill A. Shutemov struct file *file; 2935c8d78c18SKirill A. Shutemov 2936ee65728eSMike Rapoport pr_warn_once("%s (%d) uses deprecated remap_file_pages() syscall. See Documentation/mm/remap_file_pages.rst.\n", 2937c8d78c18SKirill A. Shutemov current->comm, current->pid); 2938c8d78c18SKirill A. Shutemov 2939c8d78c18SKirill A. Shutemov if (prot) 2940c8d78c18SKirill A. Shutemov return ret; 2941c8d78c18SKirill A. Shutemov start = start & PAGE_MASK; 2942c8d78c18SKirill A. Shutemov size = size & PAGE_MASK; 2943c8d78c18SKirill A. Shutemov 2944c8d78c18SKirill A. Shutemov if (start + size <= start) 2945c8d78c18SKirill A. Shutemov return ret; 2946c8d78c18SKirill A. Shutemov 2947c8d78c18SKirill A. Shutemov /* Does pgoff wrap? */ 2948c8d78c18SKirill A. Shutemov if (pgoff + (size >> PAGE_SHIFT) < pgoff) 2949c8d78c18SKirill A. Shutemov return ret; 2950c8d78c18SKirill A. Shutemov 2951d8ed45c5SMichel Lespinasse if (mmap_write_lock_killable(mm)) 2952dc0ef0dfSMichal Hocko return -EINTR; 2953dc0ef0dfSMichal Hocko 29549b593cb2SLiam R. Howlett vma = vma_lookup(mm, start); 2955c8d78c18SKirill A. Shutemov 2956c8d78c18SKirill A. Shutemov if (!vma || !(vma->vm_flags & VM_SHARED)) 2957c8d78c18SKirill A. Shutemov goto out; 2958c8d78c18SKirill A. Shutemov 295948f7df32SKirill A. Shutemov if (start + size > vma->vm_end) { 2960763ecb03SLiam R. Howlett VMA_ITERATOR(vmi, mm, vma->vm_end); 2961763ecb03SLiam R. Howlett struct vm_area_struct *next, *prev = vma; 296248f7df32SKirill A. Shutemov 2963763ecb03SLiam R. Howlett for_each_vma_range(vmi, next, start + size) { 296448f7df32SKirill A. Shutemov /* hole between vmas ? */ 2965763ecb03SLiam R. Howlett if (next->vm_start != prev->vm_end) 296648f7df32SKirill A. Shutemov goto out; 296748f7df32SKirill A. Shutemov 296848f7df32SKirill A. Shutemov if (next->vm_file != vma->vm_file) 296948f7df32SKirill A. Shutemov goto out; 297048f7df32SKirill A. Shutemov 297148f7df32SKirill A. Shutemov if (next->vm_flags != vma->vm_flags) 297248f7df32SKirill A. Shutemov goto out; 297348f7df32SKirill A. Shutemov 29741db43d3fSLiam Howlett if (start + size <= next->vm_end) 29751db43d3fSLiam Howlett break; 29761db43d3fSLiam Howlett 2977763ecb03SLiam R. Howlett prev = next; 297848f7df32SKirill A. Shutemov } 297948f7df32SKirill A. Shutemov 298048f7df32SKirill A. Shutemov if (!next) 2981c8d78c18SKirill A. Shutemov goto out; 2982c8d78c18SKirill A. Shutemov } 2983c8d78c18SKirill A. Shutemov 2984c8d78c18SKirill A. Shutemov prot |= vma->vm_flags & VM_READ ? PROT_READ : 0; 2985c8d78c18SKirill A. Shutemov prot |= vma->vm_flags & VM_WRITE ? PROT_WRITE : 0; 2986c8d78c18SKirill A. Shutemov prot |= vma->vm_flags & VM_EXEC ? PROT_EXEC : 0; 2987c8d78c18SKirill A. Shutemov 2988c8d78c18SKirill A. Shutemov flags &= MAP_NONBLOCK; 2989c8d78c18SKirill A. Shutemov flags |= MAP_SHARED | MAP_FIXED | MAP_POPULATE; 2990fce000b1SLiam Howlett if (vma->vm_flags & VM_LOCKED) 2991c8d78c18SKirill A. Shutemov flags |= MAP_LOCKED; 299248f7df32SKirill A. Shutemov 2993c8d78c18SKirill A. Shutemov file = get_file(vma->vm_file); 299445e55300SPeter Collingbourne ret = do_mmap(vma->vm_file, start, size, 2995897ab3e0SMike Rapoport prot, flags, pgoff, &populate, NULL); 2996c8d78c18SKirill A. Shutemov fput(file); 2997c8d78c18SKirill A. Shutemov out: 2998d8ed45c5SMichel Lespinasse mmap_write_unlock(mm); 2999c8d78c18SKirill A. Shutemov if (populate) 3000c8d78c18SKirill A. Shutemov mm_populate(ret, populate); 3001c8d78c18SKirill A. Shutemov if (!IS_ERR_VALUE(ret)) 3002c8d78c18SKirill A. Shutemov ret = 0; 3003c8d78c18SKirill A. Shutemov return ret; 3004c8d78c18SKirill A. Shutemov } 3005c8d78c18SKirill A. Shutemov 30061da177e4SLinus Torvalds /* 300727b26701SLiam R. Howlett * do_vma_munmap() - Unmap a full or partial vma. 300827b26701SLiam R. Howlett * @vmi: The vma iterator pointing at the vma 300927b26701SLiam R. Howlett * @vma: The first vma to be munmapped 301027b26701SLiam R. Howlett * @start: the start of the address to unmap 301127b26701SLiam R. Howlett * @end: The end of the address to unmap 30122e7ce7d3SLiam R. Howlett * @uf: The userfaultfd list_head 3013408579cdSLiam R. Howlett * @unlock: Drop the lock on success 30142e7ce7d3SLiam R. Howlett * 301527b26701SLiam R. Howlett * unmaps a VMA mapping when the vma iterator is already in position. 301627b26701SLiam R. Howlett * Does not handle alignment. 3017408579cdSLiam R. Howlett * 3018408579cdSLiam R. Howlett * Return: 0 on success drops the lock of so directed, error on failure and will 3019408579cdSLiam R. Howlett * still hold the lock. 30201da177e4SLinus Torvalds */ 302127b26701SLiam R. Howlett int do_vma_munmap(struct vma_iterator *vmi, struct vm_area_struct *vma, 3022408579cdSLiam R. Howlett unsigned long start, unsigned long end, struct list_head *uf, 3023408579cdSLiam R. Howlett bool unlock) 30242e7ce7d3SLiam R. Howlett { 30252e7ce7d3SLiam R. Howlett struct mm_struct *mm = vma->vm_mm; 30262e7ce7d3SLiam R. Howlett 302727b26701SLiam R. Howlett arch_unmap(mm, start, end); 3028b5641a5dSLinus Torvalds return do_vmi_align_munmap(vmi, vma, mm, start, end, uf, unlock); 30292e7ce7d3SLiam R. Howlett } 30302e7ce7d3SLiam R. Howlett 30312e7ce7d3SLiam R. Howlett /* 30322e7ce7d3SLiam R. Howlett * do_brk_flags() - Increase the brk vma if the flags match. 303392fed820SLiam R. Howlett * @vmi: The vma iterator 30342e7ce7d3SLiam R. Howlett * @addr: The start address 30352e7ce7d3SLiam R. Howlett * @len: The length of the increase 30362e7ce7d3SLiam R. Howlett * @vma: The vma, 30372e7ce7d3SLiam R. Howlett * @flags: The VMA Flags 30382e7ce7d3SLiam R. Howlett * 30392e7ce7d3SLiam R. Howlett * Extend the brk VMA from addr to addr + len. If the VMA is NULL or the flags 30402e7ce7d3SLiam R. Howlett * do not match then create a new anonymous VMA. Eventually we may be able to 30412e7ce7d3SLiam R. Howlett * do some brk-specific accounting here. 30422e7ce7d3SLiam R. Howlett */ 304392fed820SLiam R. Howlett static int do_brk_flags(struct vma_iterator *vmi, struct vm_area_struct *vma, 3044763ecb03SLiam R. Howlett unsigned long addr, unsigned long len, unsigned long flags) 30451da177e4SLinus Torvalds { 30461da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 3047287051b1SLiam R. Howlett struct vma_prepare vp; 30482e7ce7d3SLiam R. Howlett 30492e7ce7d3SLiam R. Howlett /* 30502e7ce7d3SLiam R. Howlett * Check against address space limits by the changed size 30512e7ce7d3SLiam R. Howlett * Note: This happens *after* clearing old mappings in some code paths. 30522e7ce7d3SLiam R. Howlett */ 305316e72e9bSDenys Vlasenko flags |= VM_DATA_DEFAULT_FLAGS | VM_ACCOUNT | mm->def_flags; 305484638335SKonstantin Khlebnikov if (!may_expand_vm(mm, flags, len >> PAGE_SHIFT)) 30551da177e4SLinus Torvalds return -ENOMEM; 30561da177e4SLinus Torvalds 30571da177e4SLinus Torvalds if (mm->map_count > sysctl_max_map_count) 30581da177e4SLinus Torvalds return -ENOMEM; 30591da177e4SLinus Torvalds 3060191c5424SAl Viro if (security_vm_enough_memory_mm(mm, len >> PAGE_SHIFT)) 30611da177e4SLinus Torvalds return -ENOMEM; 30621da177e4SLinus Torvalds 30631da177e4SLinus Torvalds /* 30642e7ce7d3SLiam R. Howlett * Expand the existing vma if possible; Note that singular lists do not 30652e7ce7d3SLiam R. Howlett * occur after forking, so the expand will only happen on new VMAs. 30661da177e4SLinus Torvalds */ 30676c28ca64SLiam Howlett if (vma && vma->vm_end == addr && !vma_policy(vma) && 30686c28ca64SLiam Howlett can_vma_merge_after(vma, flags, NULL, NULL, 30696c28ca64SLiam Howlett addr >> PAGE_SHIFT, NULL_VM_UFFD_CTX, NULL)) { 3070b5df0922SLiam R. Howlett vma_iter_config(vmi, vma->vm_start, addr + len); 3071b5df0922SLiam R. Howlett if (vma_iter_prealloc(vmi, vma)) 3072675eaca1SAlistair Popple goto unacct_fail; 307328c5609fSLiam Howlett 3074287051b1SLiam R. Howlett init_vma_prep(&vp, vma); 3075287051b1SLiam R. Howlett vma_prepare(&vp); 3076ccf1d78dSSuren Baghdasaryan vma_adjust_trans_huge(vma, vma->vm_start, addr + len, 0); 30772e7ce7d3SLiam R. Howlett vma->vm_end = addr + len; 30781c71222eSSuren Baghdasaryan vm_flags_set(vma, VM_SOFTDIRTY); 307992fed820SLiam R. Howlett vma_iter_store(vmi, vma); 30802e7ce7d3SLiam R. Howlett 3081287051b1SLiam R. Howlett vma_complete(&vp, vmi, mm); 30822e7ce7d3SLiam R. Howlett khugepaged_enter_vma(vma, flags); 30832e7ce7d3SLiam R. Howlett goto out; 30842e7ce7d3SLiam R. Howlett } 30852e7ce7d3SLiam R. Howlett 3086b5df0922SLiam R. Howlett if (vma) 3087b5df0922SLiam R. Howlett vma_iter_next_range(vmi); 30882e7ce7d3SLiam R. Howlett /* create a vma struct for an anonymous mapping */ 30892e7ce7d3SLiam R. Howlett vma = vm_area_alloc(mm); 30902e7ce7d3SLiam R. Howlett if (!vma) 3091675eaca1SAlistair Popple goto unacct_fail; 30921da177e4SLinus Torvalds 3093bfd40eafSKirill A. Shutemov vma_set_anonymous(vma); 30941da177e4SLinus Torvalds vma->vm_start = addr; 30951da177e4SLinus Torvalds vma->vm_end = addr + len; 30962e7ce7d3SLiam R. Howlett vma->vm_pgoff = addr >> PAGE_SHIFT; 30971c71222eSSuren Baghdasaryan vm_flags_init(vma, flags); 30983ed75eb8SColy Li vma->vm_page_prot = vm_get_page_prot(flags); 3099*ad9f0063SSuren Baghdasaryan vma_start_write(vma); 310092fed820SLiam R. Howlett if (vma_iter_store_gfp(vmi, vma, GFP_KERNEL)) 31012e7ce7d3SLiam R. Howlett goto mas_store_fail; 3102d4af56c5SLiam R. Howlett 31032e7ce7d3SLiam R. Howlett mm->map_count++; 31042574d5e4SLiam R. Howlett validate_mm(mm); 3105d7597f59SStefan Roesch ksm_add_vma(vma); 31061da177e4SLinus Torvalds out: 31073af9e859SEric B Munson perf_event_mmap(vma); 31081da177e4SLinus Torvalds mm->total_vm += len >> PAGE_SHIFT; 310984638335SKonstantin Khlebnikov mm->data_vm += len >> PAGE_SHIFT; 3110128557ffSMichel Lespinasse if (flags & VM_LOCKED) 3111ba470de4SRik van Riel mm->locked_vm += (len >> PAGE_SHIFT); 31121c71222eSSuren Baghdasaryan vm_flags_set(vma, VM_SOFTDIRTY); 31135d22fc25SLinus Torvalds return 0; 3114d4af56c5SLiam R. Howlett 31152e7ce7d3SLiam R. Howlett mas_store_fail: 3116d4af56c5SLiam R. Howlett vm_area_free(vma); 3117675eaca1SAlistair Popple unacct_fail: 31182e7ce7d3SLiam R. Howlett vm_unacct_memory(len >> PAGE_SHIFT); 31192e7ce7d3SLiam R. Howlett return -ENOMEM; 31201da177e4SLinus Torvalds } 31211da177e4SLinus Torvalds 3122bb177a73SMichal Hocko int vm_brk_flags(unsigned long addr, unsigned long request, unsigned long flags) 3123e4eb1ff6SLinus Torvalds { 3124e4eb1ff6SLinus Torvalds struct mm_struct *mm = current->mm; 31252e7ce7d3SLiam R. Howlett struct vm_area_struct *vma = NULL; 3126bb177a73SMichal Hocko unsigned long len; 31275d22fc25SLinus Torvalds int ret; 3128128557ffSMichel Lespinasse bool populate; 3129897ab3e0SMike Rapoport LIST_HEAD(uf); 313092fed820SLiam R. Howlett VMA_ITERATOR(vmi, mm, addr); 3131e4eb1ff6SLinus Torvalds 3132bb177a73SMichal Hocko len = PAGE_ALIGN(request); 3133bb177a73SMichal Hocko if (len < request) 3134bb177a73SMichal Hocko return -ENOMEM; 3135bb177a73SMichal Hocko if (!len) 3136bb177a73SMichal Hocko return 0; 3137bb177a73SMichal Hocko 3138d8ed45c5SMichel Lespinasse if (mmap_write_lock_killable(mm)) 31392d6c9282SMichal Hocko return -EINTR; 31402d6c9282SMichal Hocko 31412e7ce7d3SLiam R. Howlett /* Until we need other flags, refuse anything except VM_EXEC. */ 31422e7ce7d3SLiam R. Howlett if ((flags & (~VM_EXEC)) != 0) 31432e7ce7d3SLiam R. Howlett return -EINVAL; 31442e7ce7d3SLiam R. Howlett 31452e7ce7d3SLiam R. Howlett ret = check_brk_limits(addr, len); 31462e7ce7d3SLiam R. Howlett if (ret) 31472e7ce7d3SLiam R. Howlett goto limits_failed; 31482e7ce7d3SLiam R. Howlett 3149183654ceSLiam R. Howlett ret = do_vmi_munmap(&vmi, mm, addr, len, &uf, 0); 31502e7ce7d3SLiam R. Howlett if (ret) 31512e7ce7d3SLiam R. Howlett goto munmap_failed; 31522e7ce7d3SLiam R. Howlett 315392fed820SLiam R. Howlett vma = vma_prev(&vmi); 315492fed820SLiam R. Howlett ret = do_brk_flags(&vmi, vma, addr, len, flags); 3155128557ffSMichel Lespinasse populate = ((mm->def_flags & VM_LOCKED) != 0); 3156d8ed45c5SMichel Lespinasse mmap_write_unlock(mm); 3157897ab3e0SMike Rapoport userfaultfd_unmap_complete(mm, &uf); 31585d22fc25SLinus Torvalds if (populate && !ret) 3159128557ffSMichel Lespinasse mm_populate(addr, len); 3160e4eb1ff6SLinus Torvalds return ret; 31612e7ce7d3SLiam R. Howlett 31622e7ce7d3SLiam R. Howlett munmap_failed: 31632e7ce7d3SLiam R. Howlett limits_failed: 31642e7ce7d3SLiam R. Howlett mmap_write_unlock(mm); 31652e7ce7d3SLiam R. Howlett return ret; 3166e4eb1ff6SLinus Torvalds } 316716e72e9bSDenys Vlasenko EXPORT_SYMBOL(vm_brk_flags); 316816e72e9bSDenys Vlasenko 316916e72e9bSDenys Vlasenko int vm_brk(unsigned long addr, unsigned long len) 317016e72e9bSDenys Vlasenko { 317116e72e9bSDenys Vlasenko return vm_brk_flags(addr, len, 0); 317216e72e9bSDenys Vlasenko } 3173e4eb1ff6SLinus Torvalds EXPORT_SYMBOL(vm_brk); 31741da177e4SLinus Torvalds 31751da177e4SLinus Torvalds /* Release all mmaps. */ 31761da177e4SLinus Torvalds void exit_mmap(struct mm_struct *mm) 31771da177e4SLinus Torvalds { 3178d16dfc55SPeter Zijlstra struct mmu_gather tlb; 3179ba470de4SRik van Riel struct vm_area_struct *vma; 31801da177e4SLinus Torvalds unsigned long nr_accounted = 0; 3181763ecb03SLiam R. Howlett MA_STATE(mas, &mm->mm_mt, 0, 0); 3182763ecb03SLiam R. Howlett int count = 0; 31831da177e4SLinus Torvalds 3184d6dd61c8SJeremy Fitzhardinge /* mm's last user has gone, and its about to be pulled down */ 3185cddb8a5cSAndrea Arcangeli mmu_notifier_release(mm); 3186d6dd61c8SJeremy Fitzhardinge 3187bf3980c8SSuren Baghdasaryan mmap_read_lock(mm); 31889480c53eSJeremy Fitzhardinge arch_exit_mmap(mm); 31899480c53eSJeremy Fitzhardinge 3190763ecb03SLiam R. Howlett vma = mas_find(&mas, ULONG_MAX); 319164591e86SSuren Baghdasaryan if (!vma) { 319264591e86SSuren Baghdasaryan /* Can happen if dup_mmap() received an OOM */ 3193bf3980c8SSuren Baghdasaryan mmap_read_unlock(mm); 31949480c53eSJeremy Fitzhardinge return; 319564591e86SSuren Baghdasaryan } 31969480c53eSJeremy Fitzhardinge 31971da177e4SLinus Torvalds lru_add_drain(); 31981da177e4SLinus Torvalds flush_cache_mm(mm); 3199d8b45053SWill Deacon tlb_gather_mmu_fullmm(&tlb, mm); 3200901608d9SOleg Nesterov /* update_hiwater_rss(mm) here? but nobody should be looking */ 3201763ecb03SLiam R. Howlett /* Use ULONG_MAX here to ensure all VMAs in the mm are unmapped */ 3202fd892593SLiam R. Howlett unmap_vmas(&tlb, &mas, vma, 0, ULONG_MAX, ULONG_MAX, false); 3203bf3980c8SSuren Baghdasaryan mmap_read_unlock(mm); 3204bf3980c8SSuren Baghdasaryan 3205bf3980c8SSuren Baghdasaryan /* 3206bf3980c8SSuren Baghdasaryan * Set MMF_OOM_SKIP to hide this task from the oom killer/reaper 3207b3541d91SSuren Baghdasaryan * because the memory has been already freed. 3208bf3980c8SSuren Baghdasaryan */ 3209bf3980c8SSuren Baghdasaryan set_bit(MMF_OOM_SKIP, &mm->flags); 3210bf3980c8SSuren Baghdasaryan mmap_write_lock(mm); 32113dd44325SLiam R. Howlett mt_clear_in_rcu(&mm->mm_mt); 3212fd892593SLiam R. Howlett mas_set(&mas, vma->vm_end); 3213fd892593SLiam R. Howlett free_pgtables(&tlb, &mas, vma, FIRST_USER_ADDRESS, 321498e51a22SSuren Baghdasaryan USER_PGTABLES_CEILING, true); 3215ae8eba8bSWill Deacon tlb_finish_mmu(&tlb); 32161da177e4SLinus Torvalds 3217763ecb03SLiam R. Howlett /* 3218763ecb03SLiam R. Howlett * Walk the list again, actually closing and freeing it, with preemption 3219763ecb03SLiam R. Howlett * enabled, without holding any MM locks besides the unreachable 3220763ecb03SLiam R. Howlett * mmap_write_lock. 3221763ecb03SLiam R. Howlett */ 3222fd892593SLiam R. Howlett mas_set(&mas, vma->vm_end); 3223763ecb03SLiam R. Howlett do { 32244f74d2c8SLinus Torvalds if (vma->vm_flags & VM_ACCOUNT) 32254f74d2c8SLinus Torvalds nr_accounted += vma_pages(vma); 32260d2ebf9cSSuren Baghdasaryan remove_vma(vma, true); 3227763ecb03SLiam R. Howlett count++; 32280a3b3c25SPaul E. McKenney cond_resched(); 3229763ecb03SLiam R. Howlett } while ((vma = mas_find(&mas, ULONG_MAX)) != NULL); 3230763ecb03SLiam R. Howlett 3231763ecb03SLiam R. Howlett BUG_ON(count != mm->map_count); 3232d4af56c5SLiam R. Howlett 3233d4af56c5SLiam R. Howlett trace_exit_mmap(mm); 3234d4af56c5SLiam R. Howlett __mt_destroy(&mm->mm_mt); 323564591e86SSuren Baghdasaryan mmap_write_unlock(mm); 32364f74d2c8SLinus Torvalds vm_unacct_memory(nr_accounted); 32371da177e4SLinus Torvalds } 32381da177e4SLinus Torvalds 32391da177e4SLinus Torvalds /* Insert vm structure into process list sorted by address 32401da177e4SLinus Torvalds * and into the inode's i_mmap tree. If vm_file is non-NULL 3241c8c06efaSDavidlohr Bueso * then i_mmap_rwsem is taken here. 32421da177e4SLinus Torvalds */ 32431da177e4SLinus Torvalds int insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma) 32441da177e4SLinus Torvalds { 3245d4af56c5SLiam R. Howlett unsigned long charged = vma_pages(vma); 32461da177e4SLinus Torvalds 3247d4af56c5SLiam R. Howlett 3248d0601a50SLiam R. Howlett if (find_vma_intersection(mm, vma->vm_start, vma->vm_end)) 3249c9d13f5fSChen Gang return -ENOMEM; 3250d4af56c5SLiam R. Howlett 3251c9d13f5fSChen Gang if ((vma->vm_flags & VM_ACCOUNT) && 3252d4af56c5SLiam R. Howlett security_vm_enough_memory_mm(mm, charged)) 3253c9d13f5fSChen Gang return -ENOMEM; 3254c9d13f5fSChen Gang 32551da177e4SLinus Torvalds /* 32561da177e4SLinus Torvalds * The vm_pgoff of a purely anonymous vma should be irrelevant 32571da177e4SLinus Torvalds * until its first write fault, when page's anon_vma and index 32581da177e4SLinus Torvalds * are set. But now set the vm_pgoff it will almost certainly 32591da177e4SLinus Torvalds * end up with (unless mremap moves it elsewhere before that 32601da177e4SLinus Torvalds * first wfault), so /proc/pid/maps tells a consistent story. 32611da177e4SLinus Torvalds * 32621da177e4SLinus Torvalds * By setting it to reflect the virtual start address of the 32631da177e4SLinus Torvalds * vma, merges and splits can happen in a seamless way, just 32641da177e4SLinus Torvalds * using the existing file pgoff checks and manipulations. 32658332326eSLiao Pingfang * Similarly in do_mmap and in do_brk_flags. 32661da177e4SLinus Torvalds */ 32678a9cc3b5SOleg Nesterov if (vma_is_anonymous(vma)) { 32681da177e4SLinus Torvalds BUG_ON(vma->anon_vma); 32691da177e4SLinus Torvalds vma->vm_pgoff = vma->vm_start >> PAGE_SHIFT; 32701da177e4SLinus Torvalds } 32712b144498SSrikar Dronamraju 3272763ecb03SLiam R. Howlett if (vma_link(mm, vma)) { 3273d4af56c5SLiam R. Howlett vm_unacct_memory(charged); 3274d4af56c5SLiam R. Howlett return -ENOMEM; 3275d4af56c5SLiam R. Howlett } 3276d4af56c5SLiam R. Howlett 32771da177e4SLinus Torvalds return 0; 32781da177e4SLinus Torvalds } 32791da177e4SLinus Torvalds 32801da177e4SLinus Torvalds /* 32811da177e4SLinus Torvalds * Copy the vma structure to a new location in the same mm, 32821da177e4SLinus Torvalds * prior to moving page table entries, to effect an mremap move. 32831da177e4SLinus Torvalds */ 32841da177e4SLinus Torvalds struct vm_area_struct *copy_vma(struct vm_area_struct **vmap, 328538a76013SMichel Lespinasse unsigned long addr, unsigned long len, pgoff_t pgoff, 328638a76013SMichel Lespinasse bool *need_rmap_locks) 32871da177e4SLinus Torvalds { 32881da177e4SLinus Torvalds struct vm_area_struct *vma = *vmap; 32891da177e4SLinus Torvalds unsigned long vma_start = vma->vm_start; 32901da177e4SLinus Torvalds struct mm_struct *mm = vma->vm_mm; 32911da177e4SLinus Torvalds struct vm_area_struct *new_vma, *prev; 3292948f017bSAndrea Arcangeli bool faulted_in_anon_vma = true; 3293076f16bfSLiam R. Howlett VMA_ITERATOR(vmi, mm, addr); 32941da177e4SLinus Torvalds 32951da177e4SLinus Torvalds /* 32961da177e4SLinus Torvalds * If anonymous vma has not yet been faulted, update new pgoff 32971da177e4SLinus Torvalds * to match new location, to increase its chance of merging. 32981da177e4SLinus Torvalds */ 3299ce75799bSOleg Nesterov if (unlikely(vma_is_anonymous(vma) && !vma->anon_vma)) { 33001da177e4SLinus Torvalds pgoff = addr >> PAGE_SHIFT; 3301948f017bSAndrea Arcangeli faulted_in_anon_vma = false; 3302948f017bSAndrea Arcangeli } 33031da177e4SLinus Torvalds 3304763ecb03SLiam R. Howlett new_vma = find_vma_prev(mm, addr, &prev); 3305763ecb03SLiam R. Howlett if (new_vma && new_vma->vm_start < addr + len) 33066597d783SHugh Dickins return NULL; /* should never get here */ 3307524e00b3SLiam R. Howlett 33089760ebffSLiam R. Howlett new_vma = vma_merge(&vmi, mm, prev, addr, addr + len, vma->vm_flags, 330919a809afSAndrea Arcangeli vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma), 33105c26f6acSSuren Baghdasaryan vma->vm_userfaultfd_ctx, anon_vma_name(vma)); 33111da177e4SLinus Torvalds if (new_vma) { 33121da177e4SLinus Torvalds /* 33131da177e4SLinus Torvalds * Source vma may have been merged into new_vma 33141da177e4SLinus Torvalds */ 3315948f017bSAndrea Arcangeli if (unlikely(vma_start >= new_vma->vm_start && 3316948f017bSAndrea Arcangeli vma_start < new_vma->vm_end)) { 3317948f017bSAndrea Arcangeli /* 3318948f017bSAndrea Arcangeli * The only way we can get a vma_merge with 3319948f017bSAndrea Arcangeli * self during an mremap is if the vma hasn't 3320948f017bSAndrea Arcangeli * been faulted in yet and we were allowed to 3321948f017bSAndrea Arcangeli * reset the dst vma->vm_pgoff to the 3322948f017bSAndrea Arcangeli * destination address of the mremap to allow 3323948f017bSAndrea Arcangeli * the merge to happen. mremap must change the 3324948f017bSAndrea Arcangeli * vm_pgoff linearity between src and dst vmas 3325948f017bSAndrea Arcangeli * (in turn preventing a vma_merge) to be 3326948f017bSAndrea Arcangeli * safe. It is only safe to keep the vm_pgoff 3327948f017bSAndrea Arcangeli * linear if there are no pages mapped yet. 3328948f017bSAndrea Arcangeli */ 332981d1b09cSSasha Levin VM_BUG_ON_VMA(faulted_in_anon_vma, new_vma); 333038a76013SMichel Lespinasse *vmap = vma = new_vma; 3331108d6642SMichel Lespinasse } 333238a76013SMichel Lespinasse *need_rmap_locks = (new_vma->vm_pgoff <= vma->vm_pgoff); 33331da177e4SLinus Torvalds } else { 33343928d4f5SLinus Torvalds new_vma = vm_area_dup(vma); 3335e3975891SChen Gang if (!new_vma) 3336e3975891SChen Gang goto out; 33371da177e4SLinus Torvalds new_vma->vm_start = addr; 33381da177e4SLinus Torvalds new_vma->vm_end = addr + len; 33391da177e4SLinus Torvalds new_vma->vm_pgoff = pgoff; 3340ef0855d3SOleg Nesterov if (vma_dup_policy(vma, new_vma)) 3341523d4e20SMichel Lespinasse goto out_free_vma; 3342523d4e20SMichel Lespinasse if (anon_vma_clone(new_vma, vma)) 3343523d4e20SMichel Lespinasse goto out_free_mempol; 3344e9714acfSKonstantin Khlebnikov if (new_vma->vm_file) 33451da177e4SLinus Torvalds get_file(new_vma->vm_file); 33461da177e4SLinus Torvalds if (new_vma->vm_ops && new_vma->vm_ops->open) 33471da177e4SLinus Torvalds new_vma->vm_ops->open(new_vma); 3348763ecb03SLiam R. Howlett if (vma_link(mm, new_vma)) 3349524e00b3SLiam R. Howlett goto out_vma_link; 335038a76013SMichel Lespinasse *need_rmap_locks = false; 33511da177e4SLinus Torvalds } 33521da177e4SLinus Torvalds return new_vma; 33535beb4930SRik van Riel 3354524e00b3SLiam R. Howlett out_vma_link: 3355524e00b3SLiam R. Howlett if (new_vma->vm_ops && new_vma->vm_ops->close) 3356524e00b3SLiam R. Howlett new_vma->vm_ops->close(new_vma); 335792b73996SLiam Howlett 335892b73996SLiam Howlett if (new_vma->vm_file) 335992b73996SLiam Howlett fput(new_vma->vm_file); 336092b73996SLiam Howlett 336192b73996SLiam Howlett unlink_anon_vmas(new_vma); 33625beb4930SRik van Riel out_free_mempol: 3363ef0855d3SOleg Nesterov mpol_put(vma_policy(new_vma)); 33645beb4930SRik van Riel out_free_vma: 33653928d4f5SLinus Torvalds vm_area_free(new_vma); 3366e3975891SChen Gang out: 33675beb4930SRik van Riel return NULL; 33681da177e4SLinus Torvalds } 3369119f657cSakpm@osdl.org 3370119f657cSakpm@osdl.org /* 3371119f657cSakpm@osdl.org * Return true if the calling process may expand its vm space by the passed 3372119f657cSakpm@osdl.org * number of pages 3373119f657cSakpm@osdl.org */ 337484638335SKonstantin Khlebnikov bool may_expand_vm(struct mm_struct *mm, vm_flags_t flags, unsigned long npages) 3375119f657cSakpm@osdl.org { 337684638335SKonstantin Khlebnikov if (mm->total_vm + npages > rlimit(RLIMIT_AS) >> PAGE_SHIFT) 337784638335SKonstantin Khlebnikov return false; 3378119f657cSakpm@osdl.org 3379d977d56cSKonstantin Khlebnikov if (is_data_mapping(flags) && 3380d977d56cSKonstantin Khlebnikov mm->data_vm + npages > rlimit(RLIMIT_DATA) >> PAGE_SHIFT) { 3381f4fcd558SKonstantin Khlebnikov /* Workaround for Valgrind */ 3382f4fcd558SKonstantin Khlebnikov if (rlimit(RLIMIT_DATA) == 0 && 3383f4fcd558SKonstantin Khlebnikov mm->data_vm + npages <= rlimit_max(RLIMIT_DATA) >> PAGE_SHIFT) 3384f4fcd558SKonstantin Khlebnikov return true; 338557a7702bSDavid Woodhouse 338657a7702bSDavid Woodhouse pr_warn_once("%s (%d): VmData %lu exceed data ulimit %lu. Update limits%s.\n", 3387d977d56cSKonstantin Khlebnikov current->comm, current->pid, 3388d977d56cSKonstantin Khlebnikov (mm->data_vm + npages) << PAGE_SHIFT, 338957a7702bSDavid Woodhouse rlimit(RLIMIT_DATA), 339057a7702bSDavid Woodhouse ignore_rlimit_data ? "" : " or use boot option ignore_rlimit_data"); 339157a7702bSDavid Woodhouse 339257a7702bSDavid Woodhouse if (!ignore_rlimit_data) 3393d977d56cSKonstantin Khlebnikov return false; 3394d977d56cSKonstantin Khlebnikov } 3395119f657cSakpm@osdl.org 339684638335SKonstantin Khlebnikov return true; 339784638335SKonstantin Khlebnikov } 339884638335SKonstantin Khlebnikov 339984638335SKonstantin Khlebnikov void vm_stat_account(struct mm_struct *mm, vm_flags_t flags, long npages) 340084638335SKonstantin Khlebnikov { 34017866076bSPeng Liu WRITE_ONCE(mm->total_vm, READ_ONCE(mm->total_vm)+npages); 340284638335SKonstantin Khlebnikov 3403d977d56cSKonstantin Khlebnikov if (is_exec_mapping(flags)) 340484638335SKonstantin Khlebnikov mm->exec_vm += npages; 3405d977d56cSKonstantin Khlebnikov else if (is_stack_mapping(flags)) 340684638335SKonstantin Khlebnikov mm->stack_vm += npages; 3407d977d56cSKonstantin Khlebnikov else if (is_data_mapping(flags)) 340884638335SKonstantin Khlebnikov mm->data_vm += npages; 3409119f657cSakpm@osdl.org } 3410fa5dc22fSRoland McGrath 3411b3ec9f33SSouptick Joarder static vm_fault_t special_mapping_fault(struct vm_fault *vmf); 3412a62c34bdSAndy Lutomirski 3413a62c34bdSAndy Lutomirski /* 3414a62c34bdSAndy Lutomirski * Having a close hook prevents vma merging regardless of flags. 3415a62c34bdSAndy Lutomirski */ 3416a62c34bdSAndy Lutomirski static void special_mapping_close(struct vm_area_struct *vma) 3417a62c34bdSAndy Lutomirski { 3418a62c34bdSAndy Lutomirski } 3419a62c34bdSAndy Lutomirski 3420a62c34bdSAndy Lutomirski static const char *special_mapping_name(struct vm_area_struct *vma) 3421a62c34bdSAndy Lutomirski { 3422a62c34bdSAndy Lutomirski return ((struct vm_special_mapping *)vma->vm_private_data)->name; 3423a62c34bdSAndy Lutomirski } 3424a62c34bdSAndy Lutomirski 342514d07113SBrian Geffon static int special_mapping_mremap(struct vm_area_struct *new_vma) 3426b059a453SDmitry Safonov { 3427b059a453SDmitry Safonov struct vm_special_mapping *sm = new_vma->vm_private_data; 3428b059a453SDmitry Safonov 3429280e87e9SDmitry Safonov if (WARN_ON_ONCE(current->mm != new_vma->vm_mm)) 3430280e87e9SDmitry Safonov return -EFAULT; 3431280e87e9SDmitry Safonov 3432b059a453SDmitry Safonov if (sm->mremap) 3433b059a453SDmitry Safonov return sm->mremap(sm, new_vma); 3434280e87e9SDmitry Safonov 3435b059a453SDmitry Safonov return 0; 3436b059a453SDmitry Safonov } 3437b059a453SDmitry Safonov 3438871402e0SDmitry Safonov static int special_mapping_split(struct vm_area_struct *vma, unsigned long addr) 3439871402e0SDmitry Safonov { 3440871402e0SDmitry Safonov /* 3441871402e0SDmitry Safonov * Forbid splitting special mappings - kernel has expectations over 3442871402e0SDmitry Safonov * the number of pages in mapping. Together with VM_DONTEXPAND 3443871402e0SDmitry Safonov * the size of vma should stay the same over the special mapping's 3444871402e0SDmitry Safonov * lifetime. 3445871402e0SDmitry Safonov */ 3446871402e0SDmitry Safonov return -EINVAL; 3447871402e0SDmitry Safonov } 3448871402e0SDmitry Safonov 3449a62c34bdSAndy Lutomirski static const struct vm_operations_struct special_mapping_vmops = { 3450a62c34bdSAndy Lutomirski .close = special_mapping_close, 3451a62c34bdSAndy Lutomirski .fault = special_mapping_fault, 3452b059a453SDmitry Safonov .mremap = special_mapping_mremap, 3453a62c34bdSAndy Lutomirski .name = special_mapping_name, 3454af34ebebSDmitry Safonov /* vDSO code relies that VVAR can't be accessed remotely */ 3455af34ebebSDmitry Safonov .access = NULL, 3456871402e0SDmitry Safonov .may_split = special_mapping_split, 3457a62c34bdSAndy Lutomirski }; 3458a62c34bdSAndy Lutomirski 3459a62c34bdSAndy Lutomirski static const struct vm_operations_struct legacy_special_mapping_vmops = { 3460a62c34bdSAndy Lutomirski .close = special_mapping_close, 3461a62c34bdSAndy Lutomirski .fault = special_mapping_fault, 3462a62c34bdSAndy Lutomirski }; 3463fa5dc22fSRoland McGrath 3464b3ec9f33SSouptick Joarder static vm_fault_t special_mapping_fault(struct vm_fault *vmf) 3465fa5dc22fSRoland McGrath { 346611bac800SDave Jiang struct vm_area_struct *vma = vmf->vma; 3467b1d0e4f5SNick Piggin pgoff_t pgoff; 3468fa5dc22fSRoland McGrath struct page **pages; 3469fa5dc22fSRoland McGrath 3470f872f540SAndy Lutomirski if (vma->vm_ops == &legacy_special_mapping_vmops) { 3471a62c34bdSAndy Lutomirski pages = vma->vm_private_data; 3472f872f540SAndy Lutomirski } else { 3473f872f540SAndy Lutomirski struct vm_special_mapping *sm = vma->vm_private_data; 3474f872f540SAndy Lutomirski 3475f872f540SAndy Lutomirski if (sm->fault) 347611bac800SDave Jiang return sm->fault(sm, vmf->vma, vmf); 3477f872f540SAndy Lutomirski 3478f872f540SAndy Lutomirski pages = sm->pages; 3479f872f540SAndy Lutomirski } 3480a62c34bdSAndy Lutomirski 34818a9cc3b5SOleg Nesterov for (pgoff = vmf->pgoff; pgoff && *pages; ++pages) 3482b1d0e4f5SNick Piggin pgoff--; 3483fa5dc22fSRoland McGrath 3484fa5dc22fSRoland McGrath if (*pages) { 3485fa5dc22fSRoland McGrath struct page *page = *pages; 3486fa5dc22fSRoland McGrath get_page(page); 3487b1d0e4f5SNick Piggin vmf->page = page; 3488b1d0e4f5SNick Piggin return 0; 3489fa5dc22fSRoland McGrath } 3490fa5dc22fSRoland McGrath 3491b1d0e4f5SNick Piggin return VM_FAULT_SIGBUS; 3492fa5dc22fSRoland McGrath } 3493fa5dc22fSRoland McGrath 3494a62c34bdSAndy Lutomirski static struct vm_area_struct *__install_special_mapping( 3495a62c34bdSAndy Lutomirski struct mm_struct *mm, 3496fa5dc22fSRoland McGrath unsigned long addr, unsigned long len, 349727f28b97SChen Gang unsigned long vm_flags, void *priv, 349827f28b97SChen Gang const struct vm_operations_struct *ops) 3499fa5dc22fSRoland McGrath { 3500462e635eSTavis Ormandy int ret; 3501fa5dc22fSRoland McGrath struct vm_area_struct *vma; 3502fa5dc22fSRoland McGrath 3503490fc053SLinus Torvalds vma = vm_area_alloc(mm); 3504fa5dc22fSRoland McGrath if (unlikely(vma == NULL)) 35053935ed6aSStefani Seibold return ERR_PTR(-ENOMEM); 3506fa5dc22fSRoland McGrath 3507fa5dc22fSRoland McGrath vma->vm_start = addr; 3508fa5dc22fSRoland McGrath vma->vm_end = addr + len; 3509fa5dc22fSRoland McGrath 3510e430a95aSSuren Baghdasaryan vm_flags_init(vma, (vm_flags | mm->def_flags | 3511e430a95aSSuren Baghdasaryan VM_DONTEXPAND | VM_SOFTDIRTY) & ~VM_LOCKED_MASK); 35123ed75eb8SColy Li vma->vm_page_prot = vm_get_page_prot(vma->vm_flags); 3513fa5dc22fSRoland McGrath 3514a62c34bdSAndy Lutomirski vma->vm_ops = ops; 3515a62c34bdSAndy Lutomirski vma->vm_private_data = priv; 3516fa5dc22fSRoland McGrath 3517462e635eSTavis Ormandy ret = insert_vm_struct(mm, vma); 3518462e635eSTavis Ormandy if (ret) 3519462e635eSTavis Ormandy goto out; 3520fa5dc22fSRoland McGrath 352184638335SKonstantin Khlebnikov vm_stat_account(mm, vma->vm_flags, len >> PAGE_SHIFT); 3522fa5dc22fSRoland McGrath 3523cdd6c482SIngo Molnar perf_event_mmap(vma); 3524089dd79dSPeter Zijlstra 35253935ed6aSStefani Seibold return vma; 3526462e635eSTavis Ormandy 3527462e635eSTavis Ormandy out: 35283928d4f5SLinus Torvalds vm_area_free(vma); 35293935ed6aSStefani Seibold return ERR_PTR(ret); 35303935ed6aSStefani Seibold } 35313935ed6aSStefani Seibold 35322eefd878SDmitry Safonov bool vma_is_special_mapping(const struct vm_area_struct *vma, 35332eefd878SDmitry Safonov const struct vm_special_mapping *sm) 35342eefd878SDmitry Safonov { 35352eefd878SDmitry Safonov return vma->vm_private_data == sm && 35362eefd878SDmitry Safonov (vma->vm_ops == &special_mapping_vmops || 35372eefd878SDmitry Safonov vma->vm_ops == &legacy_special_mapping_vmops); 35382eefd878SDmitry Safonov } 35392eefd878SDmitry Safonov 3540a62c34bdSAndy Lutomirski /* 3541c1e8d7c6SMichel Lespinasse * Called with mm->mmap_lock held for writing. 3542a62c34bdSAndy Lutomirski * Insert a new vma covering the given region, with the given flags. 3543a62c34bdSAndy Lutomirski * Its pages are supplied by the given array of struct page *. 3544a62c34bdSAndy Lutomirski * The array can be shorter than len >> PAGE_SHIFT if it's null-terminated. 3545a62c34bdSAndy Lutomirski * The region past the last page supplied will always produce SIGBUS. 3546a62c34bdSAndy Lutomirski * The array pointer and the pages it points to are assumed to stay alive 3547a62c34bdSAndy Lutomirski * for as long as this mapping might exist. 3548a62c34bdSAndy Lutomirski */ 3549a62c34bdSAndy Lutomirski struct vm_area_struct *_install_special_mapping( 3550a62c34bdSAndy Lutomirski struct mm_struct *mm, 3551a62c34bdSAndy Lutomirski unsigned long addr, unsigned long len, 3552a62c34bdSAndy Lutomirski unsigned long vm_flags, const struct vm_special_mapping *spec) 3553a62c34bdSAndy Lutomirski { 355427f28b97SChen Gang return __install_special_mapping(mm, addr, len, vm_flags, (void *)spec, 355527f28b97SChen Gang &special_mapping_vmops); 3556a62c34bdSAndy Lutomirski } 3557a62c34bdSAndy Lutomirski 35583935ed6aSStefani Seibold int install_special_mapping(struct mm_struct *mm, 35593935ed6aSStefani Seibold unsigned long addr, unsigned long len, 35603935ed6aSStefani Seibold unsigned long vm_flags, struct page **pages) 35613935ed6aSStefani Seibold { 3562a62c34bdSAndy Lutomirski struct vm_area_struct *vma = __install_special_mapping( 356327f28b97SChen Gang mm, addr, len, vm_flags, (void *)pages, 356427f28b97SChen Gang &legacy_special_mapping_vmops); 35653935ed6aSStefani Seibold 356614bd5b45SDuan Jiong return PTR_ERR_OR_ZERO(vma); 3567fa5dc22fSRoland McGrath } 35687906d00cSAndrea Arcangeli 35697906d00cSAndrea Arcangeli static DEFINE_MUTEX(mm_all_locks_mutex); 35707906d00cSAndrea Arcangeli 3571454ed842SPeter Zijlstra static void vm_lock_anon_vma(struct mm_struct *mm, struct anon_vma *anon_vma) 35727906d00cSAndrea Arcangeli { 3573f808c13fSDavidlohr Bueso if (!test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_root.rb_node)) { 35747906d00cSAndrea Arcangeli /* 35757906d00cSAndrea Arcangeli * The LSB of head.next can't change from under us 35767906d00cSAndrea Arcangeli * because we hold the mm_all_locks_mutex. 35777906d00cSAndrea Arcangeli */ 3578da1c55f1SMichel Lespinasse down_write_nest_lock(&anon_vma->root->rwsem, &mm->mmap_lock); 35797906d00cSAndrea Arcangeli /* 35807906d00cSAndrea Arcangeli * We can safely modify head.next after taking the 35815a505085SIngo Molnar * anon_vma->root->rwsem. If some other vma in this mm shares 35827906d00cSAndrea Arcangeli * the same anon_vma we won't take it again. 35837906d00cSAndrea Arcangeli * 35847906d00cSAndrea Arcangeli * No need of atomic instructions here, head.next 35857906d00cSAndrea Arcangeli * can't change from under us thanks to the 35865a505085SIngo Molnar * anon_vma->root->rwsem. 35877906d00cSAndrea Arcangeli */ 35887906d00cSAndrea Arcangeli if (__test_and_set_bit(0, (unsigned long *) 3589f808c13fSDavidlohr Bueso &anon_vma->root->rb_root.rb_root.rb_node)) 35907906d00cSAndrea Arcangeli BUG(); 35917906d00cSAndrea Arcangeli } 35927906d00cSAndrea Arcangeli } 35937906d00cSAndrea Arcangeli 3594454ed842SPeter Zijlstra static void vm_lock_mapping(struct mm_struct *mm, struct address_space *mapping) 35957906d00cSAndrea Arcangeli { 35967906d00cSAndrea Arcangeli if (!test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) { 35977906d00cSAndrea Arcangeli /* 35987906d00cSAndrea Arcangeli * AS_MM_ALL_LOCKS can't change from under us because 35997906d00cSAndrea Arcangeli * we hold the mm_all_locks_mutex. 36007906d00cSAndrea Arcangeli * 36017906d00cSAndrea Arcangeli * Operations on ->flags have to be atomic because 36027906d00cSAndrea Arcangeli * even if AS_MM_ALL_LOCKS is stable thanks to the 36037906d00cSAndrea Arcangeli * mm_all_locks_mutex, there may be other cpus 36047906d00cSAndrea Arcangeli * changing other bitflags in parallel to us. 36057906d00cSAndrea Arcangeli */ 36067906d00cSAndrea Arcangeli if (test_and_set_bit(AS_MM_ALL_LOCKS, &mapping->flags)) 36077906d00cSAndrea Arcangeli BUG(); 3608da1c55f1SMichel Lespinasse down_write_nest_lock(&mapping->i_mmap_rwsem, &mm->mmap_lock); 36097906d00cSAndrea Arcangeli } 36107906d00cSAndrea Arcangeli } 36117906d00cSAndrea Arcangeli 36127906d00cSAndrea Arcangeli /* 36137906d00cSAndrea Arcangeli * This operation locks against the VM for all pte/vma/mm related 36147906d00cSAndrea Arcangeli * operations that could ever happen on a certain mm. This includes 36157906d00cSAndrea Arcangeli * vmtruncate, try_to_unmap, and all page faults. 36167906d00cSAndrea Arcangeli * 3617c1e8d7c6SMichel Lespinasse * The caller must take the mmap_lock in write mode before calling 36187906d00cSAndrea Arcangeli * mm_take_all_locks(). The caller isn't allowed to release the 3619c1e8d7c6SMichel Lespinasse * mmap_lock until mm_drop_all_locks() returns. 36207906d00cSAndrea Arcangeli * 3621c1e8d7c6SMichel Lespinasse * mmap_lock in write mode is required in order to block all operations 36227906d00cSAndrea Arcangeli * that could modify pagetables and free pages without need of 362327ba0644SKirill A. Shutemov * altering the vma layout. It's also needed in write mode to avoid new 36247906d00cSAndrea Arcangeli * anon_vmas to be associated with existing vmas. 36257906d00cSAndrea Arcangeli * 36267906d00cSAndrea Arcangeli * A single task can't take more than one mm_take_all_locks() in a row 36277906d00cSAndrea Arcangeli * or it would deadlock. 36287906d00cSAndrea Arcangeli * 3629bf181b9fSMichel Lespinasse * The LSB in anon_vma->rb_root.rb_node and the AS_MM_ALL_LOCKS bitflag in 36307906d00cSAndrea Arcangeli * mapping->flags avoid to take the same lock twice, if more than one 36317906d00cSAndrea Arcangeli * vma in this mm is backed by the same anon_vma or address_space. 36327906d00cSAndrea Arcangeli * 363388f306b6SKirill A. Shutemov * We take locks in following order, accordingly to comment at beginning 363488f306b6SKirill A. Shutemov * of mm/rmap.c: 363588f306b6SKirill A. Shutemov * - all hugetlbfs_i_mmap_rwsem_key locks (aka mapping->i_mmap_rwsem for 363688f306b6SKirill A. Shutemov * hugetlb mapping); 3637eeff9a5dSSuren Baghdasaryan * - all vmas marked locked 363888f306b6SKirill A. Shutemov * - all i_mmap_rwsem locks; 363988f306b6SKirill A. Shutemov * - all anon_vma->rwseml 364088f306b6SKirill A. Shutemov * 364188f306b6SKirill A. Shutemov * We can take all locks within these types randomly because the VM code 364288f306b6SKirill A. Shutemov * doesn't nest them and we protected from parallel mm_take_all_locks() by 364388f306b6SKirill A. Shutemov * mm_all_locks_mutex. 36447906d00cSAndrea Arcangeli * 36457906d00cSAndrea Arcangeli * mm_take_all_locks() and mm_drop_all_locks are expensive operations 36467906d00cSAndrea Arcangeli * that may have to take thousand of locks. 36477906d00cSAndrea Arcangeli * 36487906d00cSAndrea Arcangeli * mm_take_all_locks() can fail if it's interrupted by signals. 36497906d00cSAndrea Arcangeli */ 36507906d00cSAndrea Arcangeli int mm_take_all_locks(struct mm_struct *mm) 36517906d00cSAndrea Arcangeli { 36527906d00cSAndrea Arcangeli struct vm_area_struct *vma; 36535beb4930SRik van Riel struct anon_vma_chain *avc; 3654763ecb03SLiam R. Howlett MA_STATE(mas, &mm->mm_mt, 0, 0); 36557906d00cSAndrea Arcangeli 3656325bca1fSRolf Eike Beer mmap_assert_write_locked(mm); 36577906d00cSAndrea Arcangeli 36587906d00cSAndrea Arcangeli mutex_lock(&mm_all_locks_mutex); 36597906d00cSAndrea Arcangeli 366090717566SJann Horn /* 366190717566SJann Horn * vma_start_write() does not have a complement in mm_drop_all_locks() 366290717566SJann Horn * because vma_start_write() is always asymmetrical; it marks a VMA as 366390717566SJann Horn * being written to until mmap_write_unlock() or mmap_write_downgrade() 366490717566SJann Horn * is reached. 366590717566SJann Horn */ 3666763ecb03SLiam R. Howlett mas_for_each(&mas, vma, ULONG_MAX) { 36677906d00cSAndrea Arcangeli if (signal_pending(current)) 36687906d00cSAndrea Arcangeli goto out_unlock; 3669eeff9a5dSSuren Baghdasaryan vma_start_write(vma); 3670eeff9a5dSSuren Baghdasaryan } 3671eeff9a5dSSuren Baghdasaryan 3672eeff9a5dSSuren Baghdasaryan mas_set(&mas, 0); 3673eeff9a5dSSuren Baghdasaryan mas_for_each(&mas, vma, ULONG_MAX) { 3674eeff9a5dSSuren Baghdasaryan if (signal_pending(current)) 3675eeff9a5dSSuren Baghdasaryan goto out_unlock; 367688f306b6SKirill A. Shutemov if (vma->vm_file && vma->vm_file->f_mapping && 367788f306b6SKirill A. Shutemov is_vm_hugetlb_page(vma)) 367888f306b6SKirill A. Shutemov vm_lock_mapping(mm, vma->vm_file->f_mapping); 367988f306b6SKirill A. Shutemov } 368088f306b6SKirill A. Shutemov 3681763ecb03SLiam R. Howlett mas_set(&mas, 0); 3682763ecb03SLiam R. Howlett mas_for_each(&mas, vma, ULONG_MAX) { 368388f306b6SKirill A. Shutemov if (signal_pending(current)) 368488f306b6SKirill A. Shutemov goto out_unlock; 368588f306b6SKirill A. Shutemov if (vma->vm_file && vma->vm_file->f_mapping && 368688f306b6SKirill A. Shutemov !is_vm_hugetlb_page(vma)) 3687454ed842SPeter Zijlstra vm_lock_mapping(mm, vma->vm_file->f_mapping); 36887906d00cSAndrea Arcangeli } 36897cd5a02fSPeter Zijlstra 3690763ecb03SLiam R. Howlett mas_set(&mas, 0); 3691763ecb03SLiam R. Howlett mas_for_each(&mas, vma, ULONG_MAX) { 36927cd5a02fSPeter Zijlstra if (signal_pending(current)) 36937cd5a02fSPeter Zijlstra goto out_unlock; 36947cd5a02fSPeter Zijlstra if (vma->anon_vma) 36955beb4930SRik van Riel list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) 36965beb4930SRik van Riel vm_lock_anon_vma(mm, avc->anon_vma); 36977cd5a02fSPeter Zijlstra } 36987cd5a02fSPeter Zijlstra 3699584cff54SKautuk Consul return 0; 37007906d00cSAndrea Arcangeli 37017906d00cSAndrea Arcangeli out_unlock: 37027906d00cSAndrea Arcangeli mm_drop_all_locks(mm); 3703584cff54SKautuk Consul return -EINTR; 37047906d00cSAndrea Arcangeli } 37057906d00cSAndrea Arcangeli 37067906d00cSAndrea Arcangeli static void vm_unlock_anon_vma(struct anon_vma *anon_vma) 37077906d00cSAndrea Arcangeli { 3708f808c13fSDavidlohr Bueso if (test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_root.rb_node)) { 37097906d00cSAndrea Arcangeli /* 37107906d00cSAndrea Arcangeli * The LSB of head.next can't change to 0 from under 37117906d00cSAndrea Arcangeli * us because we hold the mm_all_locks_mutex. 37127906d00cSAndrea Arcangeli * 37137906d00cSAndrea Arcangeli * We must however clear the bitflag before unlocking 3714bf181b9fSMichel Lespinasse * the vma so the users using the anon_vma->rb_root will 37157906d00cSAndrea Arcangeli * never see our bitflag. 37167906d00cSAndrea Arcangeli * 37177906d00cSAndrea Arcangeli * No need of atomic instructions here, head.next 37187906d00cSAndrea Arcangeli * can't change from under us until we release the 37195a505085SIngo Molnar * anon_vma->root->rwsem. 37207906d00cSAndrea Arcangeli */ 37217906d00cSAndrea Arcangeli if (!__test_and_clear_bit(0, (unsigned long *) 3722f808c13fSDavidlohr Bueso &anon_vma->root->rb_root.rb_root.rb_node)) 37237906d00cSAndrea Arcangeli BUG(); 372408b52706SKonstantin Khlebnikov anon_vma_unlock_write(anon_vma); 37257906d00cSAndrea Arcangeli } 37267906d00cSAndrea Arcangeli } 37277906d00cSAndrea Arcangeli 37287906d00cSAndrea Arcangeli static void vm_unlock_mapping(struct address_space *mapping) 37297906d00cSAndrea Arcangeli { 37307906d00cSAndrea Arcangeli if (test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) { 37317906d00cSAndrea Arcangeli /* 37327906d00cSAndrea Arcangeli * AS_MM_ALL_LOCKS can't change to 0 from under us 37337906d00cSAndrea Arcangeli * because we hold the mm_all_locks_mutex. 37347906d00cSAndrea Arcangeli */ 373583cde9e8SDavidlohr Bueso i_mmap_unlock_write(mapping); 37367906d00cSAndrea Arcangeli if (!test_and_clear_bit(AS_MM_ALL_LOCKS, 37377906d00cSAndrea Arcangeli &mapping->flags)) 37387906d00cSAndrea Arcangeli BUG(); 37397906d00cSAndrea Arcangeli } 37407906d00cSAndrea Arcangeli } 37417906d00cSAndrea Arcangeli 37427906d00cSAndrea Arcangeli /* 3743c1e8d7c6SMichel Lespinasse * The mmap_lock cannot be released by the caller until 37447906d00cSAndrea Arcangeli * mm_drop_all_locks() returns. 37457906d00cSAndrea Arcangeli */ 37467906d00cSAndrea Arcangeli void mm_drop_all_locks(struct mm_struct *mm) 37477906d00cSAndrea Arcangeli { 37487906d00cSAndrea Arcangeli struct vm_area_struct *vma; 37495beb4930SRik van Riel struct anon_vma_chain *avc; 3750763ecb03SLiam R. Howlett MA_STATE(mas, &mm->mm_mt, 0, 0); 37517906d00cSAndrea Arcangeli 3752325bca1fSRolf Eike Beer mmap_assert_write_locked(mm); 37537906d00cSAndrea Arcangeli BUG_ON(!mutex_is_locked(&mm_all_locks_mutex)); 37547906d00cSAndrea Arcangeli 3755763ecb03SLiam R. Howlett mas_for_each(&mas, vma, ULONG_MAX) { 37567906d00cSAndrea Arcangeli if (vma->anon_vma) 37575beb4930SRik van Riel list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) 37585beb4930SRik van Riel vm_unlock_anon_vma(avc->anon_vma); 37597906d00cSAndrea Arcangeli if (vma->vm_file && vma->vm_file->f_mapping) 37607906d00cSAndrea Arcangeli vm_unlock_mapping(vma->vm_file->f_mapping); 37617906d00cSAndrea Arcangeli } 37627906d00cSAndrea Arcangeli 37637906d00cSAndrea Arcangeli mutex_unlock(&mm_all_locks_mutex); 37647906d00cSAndrea Arcangeli } 37658feae131SDavid Howells 37668feae131SDavid Howells /* 37673edf41d8Sseokhoon.yoon * initialise the percpu counter for VM 37688feae131SDavid Howells */ 37698feae131SDavid Howells void __init mmap_init(void) 37708feae131SDavid Howells { 377100a62ce9SKOSAKI Motohiro int ret; 377200a62ce9SKOSAKI Motohiro 3773908c7f19STejun Heo ret = percpu_counter_init(&vm_committed_as, 0, GFP_KERNEL); 377400a62ce9SKOSAKI Motohiro VM_BUG_ON(ret); 37758feae131SDavid Howells } 3776c9b1d098SAndrew Shewmaker 3777c9b1d098SAndrew Shewmaker /* 3778c9b1d098SAndrew Shewmaker * Initialise sysctl_user_reserve_kbytes. 3779c9b1d098SAndrew Shewmaker * 3780c9b1d098SAndrew Shewmaker * This is intended to prevent a user from starting a single memory hogging 3781c9b1d098SAndrew Shewmaker * process, such that they cannot recover (kill the hog) in OVERCOMMIT_NEVER 3782c9b1d098SAndrew Shewmaker * mode. 3783c9b1d098SAndrew Shewmaker * 3784c9b1d098SAndrew Shewmaker * The default value is min(3% of free memory, 128MB) 3785c9b1d098SAndrew Shewmaker * 128MB is enough to recover with sshd/login, bash, and top/kill. 3786c9b1d098SAndrew Shewmaker */ 37871640879aSAndrew Shewmaker static int init_user_reserve(void) 3788c9b1d098SAndrew Shewmaker { 3789c9b1d098SAndrew Shewmaker unsigned long free_kbytes; 3790c9b1d098SAndrew Shewmaker 3791b1773e0eSZhangPeng free_kbytes = K(global_zone_page_state(NR_FREE_PAGES)); 3792c9b1d098SAndrew Shewmaker 3793c9b1d098SAndrew Shewmaker sysctl_user_reserve_kbytes = min(free_kbytes / 32, 1UL << 17); 3794c9b1d098SAndrew Shewmaker return 0; 3795c9b1d098SAndrew Shewmaker } 3796a64fb3cdSPaul Gortmaker subsys_initcall(init_user_reserve); 37974eeab4f5SAndrew Shewmaker 37984eeab4f5SAndrew Shewmaker /* 37994eeab4f5SAndrew Shewmaker * Initialise sysctl_admin_reserve_kbytes. 38004eeab4f5SAndrew Shewmaker * 38014eeab4f5SAndrew Shewmaker * The purpose of sysctl_admin_reserve_kbytes is to allow the sys admin 38024eeab4f5SAndrew Shewmaker * to log in and kill a memory hogging process. 38034eeab4f5SAndrew Shewmaker * 38044eeab4f5SAndrew Shewmaker * Systems with more than 256MB will reserve 8MB, enough to recover 38054eeab4f5SAndrew Shewmaker * with sshd, bash, and top in OVERCOMMIT_GUESS. Smaller systems will 38064eeab4f5SAndrew Shewmaker * only reserve 3% of free pages by default. 38074eeab4f5SAndrew Shewmaker */ 38081640879aSAndrew Shewmaker static int init_admin_reserve(void) 38094eeab4f5SAndrew Shewmaker { 38104eeab4f5SAndrew Shewmaker unsigned long free_kbytes; 38114eeab4f5SAndrew Shewmaker 3812b1773e0eSZhangPeng free_kbytes = K(global_zone_page_state(NR_FREE_PAGES)); 38134eeab4f5SAndrew Shewmaker 38144eeab4f5SAndrew Shewmaker sysctl_admin_reserve_kbytes = min(free_kbytes / 32, 1UL << 13); 38154eeab4f5SAndrew Shewmaker return 0; 38164eeab4f5SAndrew Shewmaker } 3817a64fb3cdSPaul Gortmaker subsys_initcall(init_admin_reserve); 38181640879aSAndrew Shewmaker 38191640879aSAndrew Shewmaker /* 38201640879aSAndrew Shewmaker * Reinititalise user and admin reserves if memory is added or removed. 38211640879aSAndrew Shewmaker * 38221640879aSAndrew Shewmaker * The default user reserve max is 128MB, and the default max for the 38231640879aSAndrew Shewmaker * admin reserve is 8MB. These are usually, but not always, enough to 38241640879aSAndrew Shewmaker * enable recovery from a memory hogging process using login/sshd, a shell, 38251640879aSAndrew Shewmaker * and tools like top. It may make sense to increase or even disable the 38261640879aSAndrew Shewmaker * reserve depending on the existence of swap or variations in the recovery 38271640879aSAndrew Shewmaker * tools. So, the admin may have changed them. 38281640879aSAndrew Shewmaker * 38291640879aSAndrew Shewmaker * If memory is added and the reserves have been eliminated or increased above 38301640879aSAndrew Shewmaker * the default max, then we'll trust the admin. 38311640879aSAndrew Shewmaker * 38321640879aSAndrew Shewmaker * If memory is removed and there isn't enough free memory, then we 38331640879aSAndrew Shewmaker * need to reset the reserves. 38341640879aSAndrew Shewmaker * 38351640879aSAndrew Shewmaker * Otherwise keep the reserve set by the admin. 38361640879aSAndrew Shewmaker */ 38371640879aSAndrew Shewmaker static int reserve_mem_notifier(struct notifier_block *nb, 38381640879aSAndrew Shewmaker unsigned long action, void *data) 38391640879aSAndrew Shewmaker { 38401640879aSAndrew Shewmaker unsigned long tmp, free_kbytes; 38411640879aSAndrew Shewmaker 38421640879aSAndrew Shewmaker switch (action) { 38431640879aSAndrew Shewmaker case MEM_ONLINE: 38441640879aSAndrew Shewmaker /* Default max is 128MB. Leave alone if modified by operator. */ 38451640879aSAndrew Shewmaker tmp = sysctl_user_reserve_kbytes; 38461640879aSAndrew Shewmaker if (0 < tmp && tmp < (1UL << 17)) 38471640879aSAndrew Shewmaker init_user_reserve(); 38481640879aSAndrew Shewmaker 38491640879aSAndrew Shewmaker /* Default max is 8MB. Leave alone if modified by operator. */ 38501640879aSAndrew Shewmaker tmp = sysctl_admin_reserve_kbytes; 38511640879aSAndrew Shewmaker if (0 < tmp && tmp < (1UL << 13)) 38521640879aSAndrew Shewmaker init_admin_reserve(); 38531640879aSAndrew Shewmaker 38541640879aSAndrew Shewmaker break; 38551640879aSAndrew Shewmaker case MEM_OFFLINE: 3856b1773e0eSZhangPeng free_kbytes = K(global_zone_page_state(NR_FREE_PAGES)); 38571640879aSAndrew Shewmaker 38581640879aSAndrew Shewmaker if (sysctl_user_reserve_kbytes > free_kbytes) { 38591640879aSAndrew Shewmaker init_user_reserve(); 38601640879aSAndrew Shewmaker pr_info("vm.user_reserve_kbytes reset to %lu\n", 38611640879aSAndrew Shewmaker sysctl_user_reserve_kbytes); 38621640879aSAndrew Shewmaker } 38631640879aSAndrew Shewmaker 38641640879aSAndrew Shewmaker if (sysctl_admin_reserve_kbytes > free_kbytes) { 38651640879aSAndrew Shewmaker init_admin_reserve(); 38661640879aSAndrew Shewmaker pr_info("vm.admin_reserve_kbytes reset to %lu\n", 38671640879aSAndrew Shewmaker sysctl_admin_reserve_kbytes); 38681640879aSAndrew Shewmaker } 38691640879aSAndrew Shewmaker break; 38701640879aSAndrew Shewmaker default: 38711640879aSAndrew Shewmaker break; 38721640879aSAndrew Shewmaker } 38731640879aSAndrew Shewmaker return NOTIFY_OK; 38741640879aSAndrew Shewmaker } 38751640879aSAndrew Shewmaker 38761640879aSAndrew Shewmaker static int __meminit init_reserve_notifier(void) 38771640879aSAndrew Shewmaker { 38781eeaa4fdSLiu Shixin if (hotplug_memory_notifier(reserve_mem_notifier, DEFAULT_CALLBACK_PRI)) 3879b1de0d13SMitchel Humpherys pr_err("Failed registering memory add/remove notifier for admin reserve\n"); 38801640879aSAndrew Shewmaker 38811640879aSAndrew Shewmaker return 0; 38821640879aSAndrew Shewmaker } 3883a64fb3cdSPaul Gortmaker subsys_initcall(init_reserve_notifier); 3884