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; 67*71a5849aSSami Tolvanen int mmap_rnd_bits_max __ro_after_init = 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, 10830afc8c3SYajun Deng struct address_space *mapping) 1091da177e4SLinus Torvalds { 110e8e17ee9SLorenzo Stoakes if (vma_is_shared_maywrite(vma)) 1114bb5f5d9SDavid Herrmann mapping_unmap_writable(mapping); 1121da177e4SLinus Torvalds 1131da177e4SLinus Torvalds flush_dcache_mmap_lock(mapping); 1146b2dbba8SMichel Lespinasse vma_interval_tree_remove(vma, &mapping->i_mmap); 1151da177e4SLinus Torvalds flush_dcache_mmap_unlock(mapping); 1161da177e4SLinus Torvalds } 1171da177e4SLinus Torvalds 1181da177e4SLinus Torvalds /* 1196b2dbba8SMichel Lespinasse * Unlink a file-based vm structure from its interval tree, to hide 120a8fb5618SHugh Dickins * vma from rmap and vmtruncate before freeing its page tables. 1211da177e4SLinus Torvalds */ 122a8fb5618SHugh Dickins void unlink_file_vma(struct vm_area_struct *vma) 1231da177e4SLinus Torvalds { 1241da177e4SLinus Torvalds struct file *file = vma->vm_file; 1251da177e4SLinus Torvalds 1261da177e4SLinus Torvalds if (file) { 1271da177e4SLinus Torvalds struct address_space *mapping = file->f_mapping; 12883cde9e8SDavidlohr Bueso i_mmap_lock_write(mapping); 12930afc8c3SYajun Deng __remove_shared_vm_struct(vma, mapping); 13083cde9e8SDavidlohr Bueso i_mmap_unlock_write(mapping); 1311da177e4SLinus Torvalds } 132a8fb5618SHugh Dickins } 133a8fb5618SHugh Dickins 134a8fb5618SHugh Dickins /* 135763ecb03SLiam R. Howlett * Close a vm structure and free it. 136a8fb5618SHugh Dickins */ 1370d2ebf9cSSuren Baghdasaryan static void remove_vma(struct vm_area_struct *vma, bool unreachable) 138a8fb5618SHugh Dickins { 139a8fb5618SHugh Dickins might_sleep(); 1401da177e4SLinus Torvalds if (vma->vm_ops && vma->vm_ops->close) 1411da177e4SLinus Torvalds vma->vm_ops->close(vma); 142e9714acfSKonstantin Khlebnikov if (vma->vm_file) 143a8fb5618SHugh Dickins fput(vma->vm_file); 144f0be3d32SLee Schermerhorn mpol_put(vma_policy(vma)); 1450d2ebf9cSSuren Baghdasaryan if (unreachable) 1460d2ebf9cSSuren Baghdasaryan __vm_area_free(vma); 1470d2ebf9cSSuren Baghdasaryan else 1483928d4f5SLinus Torvalds vm_area_free(vma); 1491da177e4SLinus Torvalds } 1501da177e4SLinus Torvalds 151b62b633eSLiam R. Howlett static inline struct vm_area_struct *vma_prev_limit(struct vma_iterator *vmi, 152b62b633eSLiam R. Howlett unsigned long min) 153b62b633eSLiam R. Howlett { 154b62b633eSLiam R. Howlett return mas_prev(&vmi->mas, min); 155b62b633eSLiam R. Howlett } 156b62b633eSLiam R. Howlett 1572e7ce7d3SLiam R. Howlett /* 1582e7ce7d3SLiam R. Howlett * check_brk_limits() - Use platform specific check of range & verify mlock 1592e7ce7d3SLiam R. Howlett * limits. 1602e7ce7d3SLiam R. Howlett * @addr: The address to check 1612e7ce7d3SLiam R. Howlett * @len: The size of increase. 1622e7ce7d3SLiam R. Howlett * 1632e7ce7d3SLiam R. Howlett * Return: 0 on success. 1642e7ce7d3SLiam R. Howlett */ 1652e7ce7d3SLiam R. Howlett static int check_brk_limits(unsigned long addr, unsigned long len) 1662e7ce7d3SLiam R. Howlett { 1672e7ce7d3SLiam R. Howlett unsigned long mapped_addr; 1682e7ce7d3SLiam R. Howlett 1692e7ce7d3SLiam R. Howlett mapped_addr = get_unmapped_area(NULL, addr, len, 0, MAP_FIXED); 1702e7ce7d3SLiam R. Howlett if (IS_ERR_VALUE(mapped_addr)) 1712e7ce7d3SLiam R. Howlett return mapped_addr; 1722e7ce7d3SLiam R. Howlett 173b0cc5e89SAndrew Morton return mlock_future_ok(current->mm, current->mm->def_flags, len) 1743c54a298SLorenzo Stoakes ? 0 : -EAGAIN; 1752e7ce7d3SLiam R. Howlett } 17692fed820SLiam R. Howlett static int do_brk_flags(struct vma_iterator *vmi, struct vm_area_struct *brkvma, 177763ecb03SLiam R. Howlett unsigned long addr, unsigned long request, unsigned long flags); 1786a6160a7SHeiko Carstens SYSCALL_DEFINE1(brk, unsigned long, brk) 1791da177e4SLinus Torvalds { 1809bc8039eSYang Shi unsigned long newbrk, oldbrk, origbrk; 1811da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 1822e7ce7d3SLiam R. Howlett struct vm_area_struct *brkvma, *next = NULL; 183a5b4592cSJiri Kosina unsigned long min_brk; 184408579cdSLiam R. Howlett bool populate = false; 185897ab3e0SMike Rapoport LIST_HEAD(uf); 18692fed820SLiam R. Howlett struct vma_iterator vmi; 1871da177e4SLinus Torvalds 188d8ed45c5SMichel Lespinasse if (mmap_write_lock_killable(mm)) 189dc0ef0dfSMichal Hocko return -EINTR; 1901da177e4SLinus Torvalds 1919bc8039eSYang Shi origbrk = mm->brk; 1929bc8039eSYang Shi 193a5b4592cSJiri Kosina #ifdef CONFIG_COMPAT_BRK 1945520e894SJiri Kosina /* 1955520e894SJiri Kosina * CONFIG_COMPAT_BRK can still be overridden by setting 1965520e894SJiri Kosina * randomize_va_space to 2, which will still cause mm->start_brk 1975520e894SJiri Kosina * to be arbitrarily shifted 1985520e894SJiri Kosina */ 1994471a675SJiri Kosina if (current->brk_randomized) 2005520e894SJiri Kosina min_brk = mm->start_brk; 2015520e894SJiri Kosina else 2025520e894SJiri Kosina min_brk = mm->end_data; 203a5b4592cSJiri Kosina #else 204a5b4592cSJiri Kosina min_brk = mm->start_brk; 205a5b4592cSJiri Kosina #endif 206a5b4592cSJiri Kosina if (brk < min_brk) 2071da177e4SLinus Torvalds goto out; 2081e624196SRam Gupta 2091e624196SRam Gupta /* 2101e624196SRam Gupta * Check against rlimit here. If this check is done later after the test 2111e624196SRam Gupta * of oldbrk with newbrk then it can escape the test and let the data 2121e624196SRam Gupta * segment grow beyond its set limit the in case where the limit is 2131e624196SRam Gupta * not page aligned -Ram Gupta 2141e624196SRam Gupta */ 2158764b338SCyrill Gorcunov if (check_data_rlimit(rlimit(RLIMIT_DATA), brk, mm->start_brk, 2168764b338SCyrill Gorcunov mm->end_data, mm->start_data)) 2171e624196SRam Gupta goto out; 2181e624196SRam Gupta 2191da177e4SLinus Torvalds newbrk = PAGE_ALIGN(brk); 2201da177e4SLinus Torvalds oldbrk = PAGE_ALIGN(mm->brk); 2219bc8039eSYang Shi if (oldbrk == newbrk) { 2229bc8039eSYang Shi mm->brk = brk; 2239bc8039eSYang Shi goto success; 2249bc8039eSYang Shi } 2251da177e4SLinus Torvalds 226408579cdSLiam R. Howlett /* Always allow shrinking brk. */ 2271da177e4SLinus Torvalds if (brk <= mm->brk) { 2282e7ce7d3SLiam R. Howlett /* Search one past newbrk */ 22992fed820SLiam R. Howlett vma_iter_init(&vmi, mm, newbrk); 23092fed820SLiam R. Howlett brkvma = vma_find(&vmi, oldbrk); 231f5ad5083SJason A. Donenfeld if (!brkvma || brkvma->vm_start >= oldbrk) 2322e7ce7d3SLiam R. Howlett goto out; /* mapping intersects with an existing non-brk vma. */ 2339bc8039eSYang Shi /* 2342e7ce7d3SLiam R. Howlett * mm->brk must be protected by write mmap_lock. 235408579cdSLiam R. Howlett * do_vma_munmap() will drop the lock on success, so update it 23627b26701SLiam R. Howlett * before calling do_vma_munmap(). 2379bc8039eSYang Shi */ 2389bc8039eSYang Shi mm->brk = brk; 239408579cdSLiam R. Howlett if (do_vma_munmap(&vmi, brkvma, newbrk, oldbrk, &uf, true)) 2401da177e4SLinus Torvalds goto out; 241408579cdSLiam R. Howlett 242408579cdSLiam R. Howlett goto success_unlocked; 2431da177e4SLinus Torvalds } 2441da177e4SLinus Torvalds 2452e7ce7d3SLiam R. Howlett if (check_brk_limits(oldbrk, newbrk - oldbrk)) 2462e7ce7d3SLiam R. Howlett goto out; 2472e7ce7d3SLiam R. Howlett 2482e7ce7d3SLiam R. Howlett /* 2492e7ce7d3SLiam R. Howlett * Only check if the next VMA is within the stack_guard_gap of the 2502e7ce7d3SLiam R. Howlett * expansion area 2512e7ce7d3SLiam R. Howlett */ 25292fed820SLiam R. Howlett vma_iter_init(&vmi, mm, oldbrk); 25392fed820SLiam R. Howlett next = vma_find(&vmi, newbrk + PAGE_SIZE + stack_guard_gap); 2541be7107fSHugh Dickins if (next && newbrk + PAGE_SIZE > vm_start_gap(next)) 2551da177e4SLinus Torvalds goto out; 2561da177e4SLinus Torvalds 25792fed820SLiam R. Howlett brkvma = vma_prev_limit(&vmi, mm->start_brk); 2581da177e4SLinus Torvalds /* Ok, looks good - let it rip. */ 25992fed820SLiam R. Howlett if (do_brk_flags(&vmi, brkvma, oldbrk, newbrk - oldbrk, 0) < 0) 2601da177e4SLinus Torvalds goto out; 2612e7ce7d3SLiam R. Howlett 2621da177e4SLinus Torvalds mm->brk = brk; 263408579cdSLiam R. Howlett if (mm->def_flags & VM_LOCKED) 264408579cdSLiam R. Howlett populate = true; 2659bc8039eSYang Shi 2669bc8039eSYang Shi success: 267d8ed45c5SMichel Lespinasse mmap_write_unlock(mm); 268408579cdSLiam R. Howlett success_unlocked: 269897ab3e0SMike Rapoport userfaultfd_unmap_complete(mm, &uf); 270128557ffSMichel Lespinasse if (populate) 271128557ffSMichel Lespinasse mm_populate(oldbrk, newbrk - oldbrk); 272128557ffSMichel Lespinasse return brk; 273128557ffSMichel Lespinasse 2741da177e4SLinus Torvalds out: 275408579cdSLiam R. Howlett mm->brk = origbrk; 276d8ed45c5SMichel Lespinasse mmap_write_unlock(mm); 277b7204006SAdrian Huang return origbrk; 2781da177e4SLinus Torvalds } 2791da177e4SLinus Torvalds 280d4af56c5SLiam R. Howlett #if defined(CONFIG_DEBUG_VM_MAPLE_TREE) 281eafd4dc4SRashika Kheria static void validate_mm(struct mm_struct *mm) 2821da177e4SLinus Torvalds { 2831da177e4SLinus Torvalds int bug = 0; 2841da177e4SLinus Torvalds int i = 0; 285763ecb03SLiam R. Howlett struct vm_area_struct *vma; 286b50e195fSLiam R. Howlett VMA_ITERATOR(vmi, mm, 0); 287ff26f70fSAndrew Morton 288b50e195fSLiam R. Howlett mt_validate(&mm->mm_mt); 289b50e195fSLiam R. Howlett for_each_vma(vmi, vma) { 290524e00b3SLiam R. Howlett #ifdef CONFIG_DEBUG_VM_RB 29112352d3cSKonstantin Khlebnikov struct anon_vma *anon_vma = vma->anon_vma; 292ed8ea815SMichel Lespinasse struct anon_vma_chain *avc; 293b50e195fSLiam R. Howlett #endif 294b50e195fSLiam R. Howlett unsigned long vmi_start, vmi_end; 295b50e195fSLiam R. Howlett bool warn = 0; 296ff26f70fSAndrew Morton 297b50e195fSLiam R. Howlett vmi_start = vma_iter_addr(&vmi); 298b50e195fSLiam R. Howlett vmi_end = vma_iter_end(&vmi); 299b50e195fSLiam R. Howlett if (VM_WARN_ON_ONCE_MM(vma->vm_end != vmi_end, mm)) 300b50e195fSLiam R. Howlett warn = 1; 301b50e195fSLiam R. Howlett 302b50e195fSLiam R. Howlett if (VM_WARN_ON_ONCE_MM(vma->vm_start != vmi_start, mm)) 303b50e195fSLiam R. Howlett warn = 1; 304b50e195fSLiam R. Howlett 305b50e195fSLiam R. Howlett if (warn) { 306b50e195fSLiam R. Howlett pr_emerg("issue in %s\n", current->comm); 307b50e195fSLiam R. Howlett dump_stack(); 308b50e195fSLiam R. Howlett dump_vma(vma); 309b50e195fSLiam R. Howlett pr_emerg("tree range: %px start %lx end %lx\n", vma, 310b50e195fSLiam R. Howlett vmi_start, vmi_end - 1); 311b50e195fSLiam R. Howlett vma_iter_dump_tree(&vmi); 312b50e195fSLiam R. Howlett } 313b50e195fSLiam R. Howlett 314b50e195fSLiam R. Howlett #ifdef CONFIG_DEBUG_VM_RB 31512352d3cSKonstantin Khlebnikov if (anon_vma) { 31612352d3cSKonstantin Khlebnikov anon_vma_lock_read(anon_vma); 317ed8ea815SMichel Lespinasse list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) 318ed8ea815SMichel Lespinasse anon_vma_interval_tree_verify(avc); 31912352d3cSKonstantin Khlebnikov anon_vma_unlock_read(anon_vma); 32012352d3cSKonstantin Khlebnikov } 321524e00b3SLiam R. Howlett #endif 3221da177e4SLinus Torvalds i++; 3231da177e4SLinus Torvalds } 3245a0768f6SMichel Lespinasse if (i != mm->map_count) { 325b50e195fSLiam R. Howlett pr_emerg("map_count %d vma iterator %d\n", mm->map_count, i); 3265a0768f6SMichel Lespinasse bug = 1; 3275a0768f6SMichel Lespinasse } 32896dad67fSSasha Levin VM_BUG_ON_MM(bug, mm); 3291da177e4SLinus Torvalds } 330524e00b3SLiam R. Howlett 331524e00b3SLiam R. Howlett #else /* !CONFIG_DEBUG_VM_MAPLE_TREE */ 3321da177e4SLinus Torvalds #define validate_mm(mm) do { } while (0) 333524e00b3SLiam R. Howlett #endif /* CONFIG_DEBUG_VM_MAPLE_TREE */ 334d3737187SMichel Lespinasse 335bf181b9fSMichel Lespinasse /* 336bf181b9fSMichel Lespinasse * vma has some anon_vma assigned, and is already inserted on that 337bf181b9fSMichel Lespinasse * anon_vma's interval trees. 338bf181b9fSMichel Lespinasse * 339bf181b9fSMichel Lespinasse * Before updating the vma's vm_start / vm_end / vm_pgoff fields, the 340bf181b9fSMichel Lespinasse * vma must be removed from the anon_vma's interval trees using 341bf181b9fSMichel Lespinasse * anon_vma_interval_tree_pre_update_vma(). 342bf181b9fSMichel Lespinasse * 343bf181b9fSMichel Lespinasse * After the update, the vma will be reinserted using 344bf181b9fSMichel Lespinasse * anon_vma_interval_tree_post_update_vma(). 345bf181b9fSMichel Lespinasse * 346c1e8d7c6SMichel Lespinasse * The entire update must be protected by exclusive mmap_lock and by 347bf181b9fSMichel Lespinasse * the root anon_vma's mutex. 348bf181b9fSMichel Lespinasse */ 349bf181b9fSMichel Lespinasse static inline void 350bf181b9fSMichel Lespinasse anon_vma_interval_tree_pre_update_vma(struct vm_area_struct *vma) 351bf181b9fSMichel Lespinasse { 352bf181b9fSMichel Lespinasse struct anon_vma_chain *avc; 353bf181b9fSMichel Lespinasse 354bf181b9fSMichel Lespinasse list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) 355bf181b9fSMichel Lespinasse anon_vma_interval_tree_remove(avc, &avc->anon_vma->rb_root); 356bf181b9fSMichel Lespinasse } 357bf181b9fSMichel Lespinasse 358bf181b9fSMichel Lespinasse static inline void 359bf181b9fSMichel Lespinasse anon_vma_interval_tree_post_update_vma(struct vm_area_struct *vma) 360bf181b9fSMichel Lespinasse { 361bf181b9fSMichel Lespinasse struct anon_vma_chain *avc; 362bf181b9fSMichel Lespinasse 363bf181b9fSMichel Lespinasse list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) 364bf181b9fSMichel Lespinasse anon_vma_interval_tree_insert(avc, &avc->anon_vma->rb_root); 365bf181b9fSMichel Lespinasse } 366bf181b9fSMichel Lespinasse 367e8420a8eSCyril Hrubis static unsigned long count_vma_pages_range(struct mm_struct *mm, 368e8420a8eSCyril Hrubis unsigned long addr, unsigned long end) 369e8420a8eSCyril Hrubis { 3702e3af1dbSMatthew Wilcox (Oracle) VMA_ITERATOR(vmi, mm, addr); 371e8420a8eSCyril Hrubis struct vm_area_struct *vma; 3722e3af1dbSMatthew Wilcox (Oracle) unsigned long nr_pages = 0; 373e8420a8eSCyril Hrubis 3742e3af1dbSMatthew Wilcox (Oracle) for_each_vma_range(vmi, vma, end) { 3752e3af1dbSMatthew Wilcox (Oracle) unsigned long vm_start = max(addr, vma->vm_start); 3762e3af1dbSMatthew Wilcox (Oracle) unsigned long vm_end = min(end, vma->vm_end); 377e8420a8eSCyril Hrubis 3782e3af1dbSMatthew Wilcox (Oracle) nr_pages += PHYS_PFN(vm_end - vm_start); 379e8420a8eSCyril Hrubis } 380e8420a8eSCyril Hrubis 381e8420a8eSCyril Hrubis return nr_pages; 382e8420a8eSCyril Hrubis } 383e8420a8eSCyril Hrubis 384c154124fSLiam R. Howlett static void __vma_link_file(struct vm_area_struct *vma, 385c154124fSLiam R. Howlett struct address_space *mapping) 3861da177e4SLinus Torvalds { 387e8e17ee9SLorenzo Stoakes if (vma_is_shared_maywrite(vma)) 388cf508b58SMiaohe Lin mapping_allow_writable(mapping); 3891da177e4SLinus Torvalds 3901da177e4SLinus Torvalds flush_dcache_mmap_lock(mapping); 3916b2dbba8SMichel Lespinasse vma_interval_tree_insert(vma, &mapping->i_mmap); 3921da177e4SLinus Torvalds flush_dcache_mmap_unlock(mapping); 3931da177e4SLinus Torvalds } 3941da177e4SLinus Torvalds 39530afc8c3SYajun Deng static void vma_link_file(struct vm_area_struct *vma) 39630afc8c3SYajun Deng { 39730afc8c3SYajun Deng struct file *file = vma->vm_file; 39830afc8c3SYajun Deng struct address_space *mapping; 39930afc8c3SYajun Deng 40030afc8c3SYajun Deng if (file) { 40130afc8c3SYajun Deng mapping = file->f_mapping; 40230afc8c3SYajun Deng i_mmap_lock_write(mapping); 40330afc8c3SYajun Deng __vma_link_file(vma, mapping); 40430afc8c3SYajun Deng i_mmap_unlock_write(mapping); 40530afc8c3SYajun Deng } 40630afc8c3SYajun Deng } 40730afc8c3SYajun Deng 408763ecb03SLiam R. Howlett static int vma_link(struct mm_struct *mm, struct vm_area_struct *vma) 4091da177e4SLinus Torvalds { 41079e4f2caSLiam R. Howlett VMA_ITERATOR(vmi, mm, 0); 4111da177e4SLinus Torvalds 412b5df0922SLiam R. Howlett vma_iter_config(&vmi, vma->vm_start, vma->vm_end); 413b5df0922SLiam R. Howlett if (vma_iter_prealloc(&vmi, vma)) 414d4af56c5SLiam R. Howlett return -ENOMEM; 415d4af56c5SLiam R. Howlett 416ad9f0063SSuren Baghdasaryan vma_start_write(vma); 4176852c46cSYu Ma vma_iter_store(&vmi, vma); 41830afc8c3SYajun Deng vma_link_file(vma); 4191da177e4SLinus Torvalds mm->map_count++; 4201da177e4SLinus Torvalds validate_mm(mm); 421d4af56c5SLiam R. Howlett return 0; 4221da177e4SLinus Torvalds } 4231da177e4SLinus Torvalds 4241da177e4SLinus Torvalds /* 42568cefec5SLiam R. Howlett * init_multi_vma_prep() - Initializer for struct vma_prepare 42668cefec5SLiam R. Howlett * @vp: The vma_prepare struct 42768cefec5SLiam R. Howlett * @vma: The vma that will be altered once locked 42868cefec5SLiam R. Howlett * @next: The next vma if it is to be adjusted 42968cefec5SLiam R. Howlett * @remove: The first vma to be removed 43068cefec5SLiam R. Howlett * @remove2: The second vma to be removed 43168cefec5SLiam R. Howlett */ 43268cefec5SLiam R. Howlett static inline void init_multi_vma_prep(struct vma_prepare *vp, 43368cefec5SLiam R. Howlett struct vm_area_struct *vma, struct vm_area_struct *next, 43468cefec5SLiam R. Howlett struct vm_area_struct *remove, struct vm_area_struct *remove2) 43568cefec5SLiam R. Howlett { 43668cefec5SLiam R. Howlett memset(vp, 0, sizeof(struct vma_prepare)); 43768cefec5SLiam R. Howlett vp->vma = vma; 43868cefec5SLiam R. Howlett vp->anon_vma = vma->anon_vma; 43968cefec5SLiam R. Howlett vp->remove = remove; 44068cefec5SLiam R. Howlett vp->remove2 = remove2; 44168cefec5SLiam R. Howlett vp->adj_next = next; 44268cefec5SLiam R. Howlett if (!vp->anon_vma && next) 44368cefec5SLiam R. Howlett vp->anon_vma = next->anon_vma; 44468cefec5SLiam R. Howlett 44568cefec5SLiam R. Howlett vp->file = vma->vm_file; 44668cefec5SLiam R. Howlett if (vp->file) 44768cefec5SLiam R. Howlett vp->mapping = vma->vm_file->f_mapping; 44868cefec5SLiam R. Howlett 44968cefec5SLiam R. Howlett } 45068cefec5SLiam R. Howlett 45168cefec5SLiam R. Howlett /* 45268cefec5SLiam R. Howlett * init_vma_prep() - Initializer wrapper for vma_prepare struct 45368cefec5SLiam R. Howlett * @vp: The vma_prepare struct 45468cefec5SLiam R. Howlett * @vma: The vma that will be altered once locked 45568cefec5SLiam R. Howlett */ 45668cefec5SLiam R. Howlett static inline void init_vma_prep(struct vma_prepare *vp, 45768cefec5SLiam R. Howlett struct vm_area_struct *vma) 45868cefec5SLiam R. Howlett { 45968cefec5SLiam R. Howlett init_multi_vma_prep(vp, vma, NULL, NULL, NULL); 46068cefec5SLiam R. Howlett } 46168cefec5SLiam R. Howlett 46268cefec5SLiam R. Howlett 46368cefec5SLiam R. Howlett /* 464440703e0SLiam R. Howlett * vma_prepare() - Helper function for handling locking VMAs prior to altering 465440703e0SLiam R. Howlett * @vp: The initialized vma_prepare struct 466440703e0SLiam R. Howlett */ 467440703e0SLiam R. Howlett static inline void vma_prepare(struct vma_prepare *vp) 468440703e0SLiam R. Howlett { 469440703e0SLiam R. Howlett if (vp->file) { 470440703e0SLiam R. Howlett uprobe_munmap(vp->vma, vp->vma->vm_start, vp->vma->vm_end); 471440703e0SLiam R. Howlett 472440703e0SLiam R. Howlett if (vp->adj_next) 473440703e0SLiam R. Howlett uprobe_munmap(vp->adj_next, vp->adj_next->vm_start, 474440703e0SLiam R. Howlett vp->adj_next->vm_end); 475440703e0SLiam R. Howlett 476440703e0SLiam R. Howlett i_mmap_lock_write(vp->mapping); 477440703e0SLiam R. Howlett if (vp->insert && vp->insert->vm_file) { 478440703e0SLiam R. Howlett /* 479440703e0SLiam R. Howlett * Put into interval tree now, so instantiated pages 480440703e0SLiam R. Howlett * are visible to arm/parisc __flush_dcache_page 481440703e0SLiam R. Howlett * throughout; but we cannot insert into address 482440703e0SLiam R. Howlett * space until vma start or end is updated. 483440703e0SLiam R. Howlett */ 484440703e0SLiam R. Howlett __vma_link_file(vp->insert, 485440703e0SLiam R. Howlett vp->insert->vm_file->f_mapping); 486440703e0SLiam R. Howlett } 487440703e0SLiam R. Howlett } 488440703e0SLiam R. Howlett 489440703e0SLiam R. Howlett if (vp->anon_vma) { 490440703e0SLiam R. Howlett anon_vma_lock_write(vp->anon_vma); 491440703e0SLiam R. Howlett anon_vma_interval_tree_pre_update_vma(vp->vma); 492440703e0SLiam R. Howlett if (vp->adj_next) 493440703e0SLiam R. Howlett anon_vma_interval_tree_pre_update_vma(vp->adj_next); 494440703e0SLiam R. Howlett } 495440703e0SLiam R. Howlett 496440703e0SLiam R. Howlett if (vp->file) { 497440703e0SLiam R. Howlett flush_dcache_mmap_lock(vp->mapping); 498440703e0SLiam R. Howlett vma_interval_tree_remove(vp->vma, &vp->mapping->i_mmap); 499440703e0SLiam R. Howlett if (vp->adj_next) 500440703e0SLiam R. Howlett vma_interval_tree_remove(vp->adj_next, 501440703e0SLiam R. Howlett &vp->mapping->i_mmap); 502440703e0SLiam R. Howlett } 503440703e0SLiam R. Howlett 504440703e0SLiam R. Howlett } 505440703e0SLiam R. Howlett 506440703e0SLiam R. Howlett /* 507440703e0SLiam R. Howlett * vma_complete- Helper function for handling the unlocking after altering VMAs, 508440703e0SLiam R. Howlett * or for inserting a VMA. 509440703e0SLiam R. Howlett * 510440703e0SLiam R. Howlett * @vp: The vma_prepare struct 511440703e0SLiam R. Howlett * @vmi: The vma iterator 512440703e0SLiam R. Howlett * @mm: The mm_struct 513440703e0SLiam R. Howlett */ 514440703e0SLiam R. Howlett static inline void vma_complete(struct vma_prepare *vp, 515440703e0SLiam R. Howlett struct vma_iterator *vmi, struct mm_struct *mm) 516440703e0SLiam R. Howlett { 517440703e0SLiam R. Howlett if (vp->file) { 518440703e0SLiam R. Howlett if (vp->adj_next) 519440703e0SLiam R. Howlett vma_interval_tree_insert(vp->adj_next, 520440703e0SLiam R. Howlett &vp->mapping->i_mmap); 521440703e0SLiam R. Howlett vma_interval_tree_insert(vp->vma, &vp->mapping->i_mmap); 522440703e0SLiam R. Howlett flush_dcache_mmap_unlock(vp->mapping); 523440703e0SLiam R. Howlett } 524440703e0SLiam R. Howlett 525440703e0SLiam R. Howlett if (vp->remove && vp->file) { 52630afc8c3SYajun Deng __remove_shared_vm_struct(vp->remove, vp->mapping); 527440703e0SLiam R. Howlett if (vp->remove2) 52830afc8c3SYajun Deng __remove_shared_vm_struct(vp->remove2, vp->mapping); 529440703e0SLiam R. Howlett } else if (vp->insert) { 530440703e0SLiam R. Howlett /* 531440703e0SLiam R. Howlett * split_vma has split insert from vma, and needs 532440703e0SLiam R. Howlett * us to insert it before dropping the locks 533440703e0SLiam R. Howlett * (it may either follow vma or precede it). 534440703e0SLiam R. Howlett */ 535440703e0SLiam R. Howlett vma_iter_store(vmi, vp->insert); 536440703e0SLiam R. Howlett mm->map_count++; 537440703e0SLiam R. Howlett } 538440703e0SLiam R. Howlett 539440703e0SLiam R. Howlett if (vp->anon_vma) { 540440703e0SLiam R. Howlett anon_vma_interval_tree_post_update_vma(vp->vma); 541440703e0SLiam R. Howlett if (vp->adj_next) 542440703e0SLiam R. Howlett anon_vma_interval_tree_post_update_vma(vp->adj_next); 543440703e0SLiam R. Howlett anon_vma_unlock_write(vp->anon_vma); 544440703e0SLiam R. Howlett } 545440703e0SLiam R. Howlett 546440703e0SLiam R. Howlett if (vp->file) { 547440703e0SLiam R. Howlett i_mmap_unlock_write(vp->mapping); 548440703e0SLiam R. Howlett uprobe_mmap(vp->vma); 549440703e0SLiam R. Howlett 550440703e0SLiam R. Howlett if (vp->adj_next) 551440703e0SLiam R. Howlett uprobe_mmap(vp->adj_next); 552440703e0SLiam R. Howlett } 553440703e0SLiam R. Howlett 554440703e0SLiam R. Howlett if (vp->remove) { 555440703e0SLiam R. Howlett again: 556457f67beSSuren Baghdasaryan vma_mark_detached(vp->remove, true); 557440703e0SLiam R. Howlett if (vp->file) { 558440703e0SLiam R. Howlett uprobe_munmap(vp->remove, vp->remove->vm_start, 559440703e0SLiam R. Howlett vp->remove->vm_end); 560440703e0SLiam R. Howlett fput(vp->file); 561440703e0SLiam R. Howlett } 562440703e0SLiam R. Howlett if (vp->remove->anon_vma) 563440703e0SLiam R. Howlett anon_vma_merge(vp->vma, vp->remove); 564440703e0SLiam R. Howlett mm->map_count--; 565440703e0SLiam R. Howlett mpol_put(vma_policy(vp->remove)); 566440703e0SLiam R. Howlett if (!vp->remove2) 567440703e0SLiam R. Howlett WARN_ON_ONCE(vp->vma->vm_end < vp->remove->vm_end); 568440703e0SLiam R. Howlett vm_area_free(vp->remove); 569440703e0SLiam R. Howlett 570440703e0SLiam R. Howlett /* 571440703e0SLiam R. Howlett * In mprotect's case 6 (see comments on vma_merge), 5725ff783f1SVlastimil Babka * we are removing both mid and next vmas 573440703e0SLiam R. Howlett */ 574440703e0SLiam R. Howlett if (vp->remove2) { 575440703e0SLiam R. Howlett vp->remove = vp->remove2; 576440703e0SLiam R. Howlett vp->remove2 = NULL; 577440703e0SLiam R. Howlett goto again; 578440703e0SLiam R. Howlett } 579440703e0SLiam R. Howlett } 580440703e0SLiam R. Howlett if (vp->insert && vp->file) 581440703e0SLiam R. Howlett uprobe_mmap(vp->insert); 5822574d5e4SLiam R. Howlett validate_mm(mm); 583440703e0SLiam R. Howlett } 584440703e0SLiam R. Howlett 585440703e0SLiam R. Howlett /* 58604241ffeSLiam R. Howlett * dup_anon_vma() - Helper function to duplicate anon_vma 58704241ffeSLiam R. Howlett * @dst: The destination VMA 58804241ffeSLiam R. Howlett * @src: The source VMA 589824135c4SLiam R. Howlett * @dup: Pointer to the destination VMA when successful. 59004241ffeSLiam R. Howlett * 59104241ffeSLiam R. Howlett * Returns: 0 on success. 59204241ffeSLiam R. Howlett */ 59304241ffeSLiam R. Howlett static inline int dup_anon_vma(struct vm_area_struct *dst, 594824135c4SLiam R. Howlett struct vm_area_struct *src, struct vm_area_struct **dup) 59504241ffeSLiam R. Howlett { 59604241ffeSLiam R. Howlett /* 59704241ffeSLiam R. Howlett * Easily overlooked: when mprotect shifts the boundary, make sure the 59804241ffeSLiam R. Howlett * expanding vma has anon_vma set if the shrinking vma had, to cover any 59904241ffeSLiam R. Howlett * anon pages imported. 60004241ffeSLiam R. Howlett */ 60104241ffeSLiam R. Howlett if (src->anon_vma && !dst->anon_vma) { 602824135c4SLiam R. Howlett int ret; 603824135c4SLiam R. Howlett 604c9d6e982SSuren Baghdasaryan vma_assert_write_locked(dst); 60504241ffeSLiam R. Howlett dst->anon_vma = src->anon_vma; 606824135c4SLiam R. Howlett ret = anon_vma_clone(dst, src); 607824135c4SLiam R. Howlett if (ret) 608824135c4SLiam R. Howlett return ret; 609824135c4SLiam R. Howlett 610824135c4SLiam R. Howlett *dup = dst; 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 { 637824135c4SLiam R. Howlett struct vm_area_struct *anon_dup = NULL; 63868cefec5SLiam R. Howlett bool remove_next = false; 6399303d3e1SLiam R. Howlett struct vma_prepare vp; 6409303d3e1SLiam R. Howlett 641c9d6e982SSuren Baghdasaryan vma_start_write(vma); 6429303d3e1SLiam R. Howlett if (next && (vma != next) && (end == next->vm_end)) { 64304241ffeSLiam R. Howlett int ret; 6449303d3e1SLiam R. Howlett 64504241ffeSLiam R. Howlett remove_next = true; 646c9d6e982SSuren Baghdasaryan vma_start_write(next); 647824135c4SLiam R. Howlett ret = dup_anon_vma(vma, next, &anon_dup); 64804241ffeSLiam R. Howlett if (ret) 64904241ffeSLiam R. Howlett return ret; 6509303d3e1SLiam R. Howlett } 6519303d3e1SLiam R. Howlett 65268cefec5SLiam R. Howlett init_multi_vma_prep(&vp, vma, NULL, remove_next ? next : NULL, NULL); 6539303d3e1SLiam R. Howlett /* Not merging but overwriting any part of next is not handled. */ 6549303d3e1SLiam R. Howlett VM_WARN_ON(next && !vp.remove && 6559303d3e1SLiam R. Howlett next != vma && end > next->vm_start); 6569303d3e1SLiam R. Howlett /* Only handles expanding */ 6579303d3e1SLiam R. Howlett VM_WARN_ON(vma->vm_start < start || vma->vm_end > end); 6589303d3e1SLiam R. Howlett 659b5df0922SLiam R. Howlett /* Note: vma iterator must be pointing to 'start' */ 660b5df0922SLiam R. Howlett vma_iter_config(vmi, start, end); 661b5df0922SLiam R. Howlett if (vma_iter_prealloc(vmi, vma)) 6629303d3e1SLiam R. Howlett goto nomem; 6639303d3e1SLiam R. Howlett 664ccf1d78dSSuren Baghdasaryan vma_prepare(&vp); 6659303d3e1SLiam R. Howlett vma_adjust_trans_huge(vma, start, end, 0); 666412c6ef9SYajun Deng vma_set_range(vma, start, end, pgoff); 6679303d3e1SLiam R. Howlett vma_iter_store(vmi, vma); 6689303d3e1SLiam R. Howlett 6699303d3e1SLiam R. Howlett vma_complete(&vp, vmi, vma->vm_mm); 6709303d3e1SLiam R. Howlett return 0; 6719303d3e1SLiam R. Howlett 6729303d3e1SLiam R. Howlett nomem: 673824135c4SLiam R. Howlett if (anon_dup) 674824135c4SLiam R. Howlett unlink_anon_vmas(anon_dup); 6759303d3e1SLiam R. Howlett return -ENOMEM; 6769303d3e1SLiam R. Howlett } 677cf51e86dSLiam R. Howlett 678cf51e86dSLiam R. Howlett /* 679cf51e86dSLiam R. Howlett * vma_shrink() - Reduce an existing VMAs memory area 680cf51e86dSLiam R. Howlett * @vmi: The vma iterator 681cf51e86dSLiam R. Howlett * @vma: The VMA to modify 682cf51e86dSLiam R. Howlett * @start: The new start 683cf51e86dSLiam R. Howlett * @end: The new end 684cf51e86dSLiam R. Howlett * 685cf51e86dSLiam R. Howlett * Returns: 0 on success, -ENOMEM otherwise 686cf51e86dSLiam R. Howlett */ 687cf51e86dSLiam R. Howlett int vma_shrink(struct vma_iterator *vmi, struct vm_area_struct *vma, 688cf51e86dSLiam R. Howlett unsigned long start, unsigned long end, pgoff_t pgoff) 689cf51e86dSLiam R. Howlett { 690cf51e86dSLiam R. Howlett struct vma_prepare vp; 691cf51e86dSLiam R. Howlett 692cf51e86dSLiam R. Howlett WARN_ON((vma->vm_start != start) && (vma->vm_end != end)); 693cf51e86dSLiam R. Howlett 694b5df0922SLiam R. Howlett if (vma->vm_start < start) 695b5df0922SLiam R. Howlett vma_iter_config(vmi, vma->vm_start, start); 696b5df0922SLiam R. Howlett else 697b5df0922SLiam R. Howlett vma_iter_config(vmi, end, vma->vm_end); 698b5df0922SLiam R. Howlett 699b5df0922SLiam R. Howlett if (vma_iter_prealloc(vmi, NULL)) 700cf51e86dSLiam R. Howlett return -ENOMEM; 701cf51e86dSLiam R. Howlett 702c9d6e982SSuren Baghdasaryan vma_start_write(vma); 703c9d6e982SSuren Baghdasaryan 704cf51e86dSLiam R. Howlett init_vma_prep(&vp, vma); 705cf51e86dSLiam R. Howlett vma_prepare(&vp); 706ccf1d78dSSuren Baghdasaryan vma_adjust_trans_huge(vma, start, end, 0); 707cf51e86dSLiam R. Howlett 708b5df0922SLiam R. Howlett vma_iter_clear(vmi); 709412c6ef9SYajun Deng vma_set_range(vma, start, end, pgoff); 710cf51e86dSLiam R. Howlett vma_complete(&vp, vmi, vma->vm_mm); 711cf51e86dSLiam R. Howlett return 0; 712cf51e86dSLiam R. Howlett } 713cf51e86dSLiam R. Howlett 7149303d3e1SLiam R. Howlett /* 7151da177e4SLinus Torvalds * If the vma has a ->close operation then the driver probably needs to release 716714965caSVlastimil Babka * per-vma resources, so we don't attempt to merge those if the caller indicates 717714965caSVlastimil Babka * the current vma may be removed as part of the merge. 7181da177e4SLinus Torvalds */ 7192dbf4010SVlastimil Babka static inline bool is_mergeable_vma(struct vm_area_struct *vma, 72019a809afSAndrea Arcangeli struct file *file, unsigned long vm_flags, 7219a10064fSColin Cross struct vm_userfaultfd_ctx vm_userfaultfd_ctx, 722714965caSVlastimil Babka struct anon_vma_name *anon_name, bool may_remove_vma) 7231da177e4SLinus Torvalds { 72434228d47SCyrill Gorcunov /* 72534228d47SCyrill Gorcunov * VM_SOFTDIRTY should not prevent from VMA merging, if we 72634228d47SCyrill Gorcunov * match the flags but dirty bit -- the caller should mark 72734228d47SCyrill Gorcunov * merged VMA as dirty. If dirty bit won't be excluded from 7288bb4e7a2SWei Yang * comparison, we increase pressure on the memory system forcing 72934228d47SCyrill Gorcunov * the kernel to generate new VMAs when old one could be 73034228d47SCyrill Gorcunov * extended instead. 73134228d47SCyrill Gorcunov */ 73234228d47SCyrill Gorcunov if ((vma->vm_flags ^ vm_flags) & ~VM_SOFTDIRTY) 7332dbf4010SVlastimil Babka return false; 7341da177e4SLinus Torvalds if (vma->vm_file != file) 7352dbf4010SVlastimil Babka return false; 736714965caSVlastimil Babka if (may_remove_vma && vma->vm_ops && vma->vm_ops->close) 7372dbf4010SVlastimil Babka return false; 73819a809afSAndrea Arcangeli if (!is_mergeable_vm_userfaultfd_ctx(vma, vm_userfaultfd_ctx)) 7392dbf4010SVlastimil Babka return false; 7405c26f6acSSuren Baghdasaryan if (!anon_vma_name_eq(anon_vma_name(vma), anon_name)) 7412dbf4010SVlastimil Babka return false; 7422dbf4010SVlastimil Babka return true; 7431da177e4SLinus Torvalds } 7441da177e4SLinus Torvalds 7452dbf4010SVlastimil Babka static inline bool is_mergeable_anon_vma(struct anon_vma *anon_vma1, 7462dbf4010SVlastimil Babka struct anon_vma *anon_vma2, struct vm_area_struct *vma) 7471da177e4SLinus Torvalds { 748965f55deSShaohua Li /* 749965f55deSShaohua Li * The list_is_singular() test is to avoid merging VMA cloned from 750965f55deSShaohua Li * parents. This can improve scalability caused by anon_vma lock. 751965f55deSShaohua Li */ 752965f55deSShaohua Li if ((!anon_vma1 || !anon_vma2) && (!vma || 753965f55deSShaohua Li list_is_singular(&vma->anon_vma_chain))) 7542dbf4010SVlastimil Babka return true; 755965f55deSShaohua Li return anon_vma1 == anon_vma2; 7561da177e4SLinus Torvalds } 7571da177e4SLinus Torvalds 7581da177e4SLinus Torvalds /* 7591da177e4SLinus Torvalds * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff) 7601da177e4SLinus Torvalds * in front of (at a lower virtual address and file offset than) the vma. 7611da177e4SLinus Torvalds * 7621da177e4SLinus Torvalds * We cannot merge two vmas if they have differently assigned (non-NULL) 7631da177e4SLinus Torvalds * anon_vmas, nor if same anon_vma is assigned but offsets incompatible. 7641da177e4SLinus Torvalds * 7651da177e4SLinus Torvalds * We don't check here for the merged mmap wrapping around the end of pagecache 76645e55300SPeter Collingbourne * indices (16TB on ia32) because do_mmap() does not permit mmap's which 7671da177e4SLinus Torvalds * wrap, nor mmaps which cover the final page at index -1UL. 768714965caSVlastimil Babka * 769714965caSVlastimil Babka * We assume the vma may be removed as part of the merge. 7701da177e4SLinus Torvalds */ 7712dbf4010SVlastimil Babka static bool 7721da177e4SLinus Torvalds can_vma_merge_before(struct vm_area_struct *vma, unsigned long vm_flags, 77319a809afSAndrea Arcangeli struct anon_vma *anon_vma, struct file *file, 7742dbf4010SVlastimil Babka pgoff_t vm_pgoff, struct vm_userfaultfd_ctx vm_userfaultfd_ctx, 7755c26f6acSSuren Baghdasaryan struct anon_vma_name *anon_name) 7761da177e4SLinus Torvalds { 777714965caSVlastimil Babka if (is_mergeable_vma(vma, file, vm_flags, vm_userfaultfd_ctx, anon_name, true) && 778965f55deSShaohua Li is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) { 7791da177e4SLinus Torvalds if (vma->vm_pgoff == vm_pgoff) 7802dbf4010SVlastimil Babka return true; 7811da177e4SLinus Torvalds } 7822dbf4010SVlastimil Babka return false; 7831da177e4SLinus Torvalds } 7841da177e4SLinus Torvalds 7851da177e4SLinus Torvalds /* 7861da177e4SLinus Torvalds * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff) 7871da177e4SLinus Torvalds * beyond (at a higher virtual address and file offset than) the vma. 7881da177e4SLinus Torvalds * 7891da177e4SLinus Torvalds * We cannot merge two vmas if they have differently assigned (non-NULL) 7901da177e4SLinus Torvalds * anon_vmas, nor if same anon_vma is assigned but offsets incompatible. 791714965caSVlastimil Babka * 792714965caSVlastimil Babka * We assume that vma is not removed as part of the merge. 7931da177e4SLinus Torvalds */ 7942dbf4010SVlastimil Babka static bool 7951da177e4SLinus Torvalds can_vma_merge_after(struct vm_area_struct *vma, unsigned long vm_flags, 79619a809afSAndrea Arcangeli struct anon_vma *anon_vma, struct file *file, 7972dbf4010SVlastimil Babka pgoff_t vm_pgoff, struct vm_userfaultfd_ctx vm_userfaultfd_ctx, 7985c26f6acSSuren Baghdasaryan struct anon_vma_name *anon_name) 7991da177e4SLinus Torvalds { 800714965caSVlastimil Babka if (is_mergeable_vma(vma, file, vm_flags, vm_userfaultfd_ctx, anon_name, false) && 801965f55deSShaohua Li is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) { 8021da177e4SLinus Torvalds pgoff_t vm_pglen; 803d6e93217SLibin vm_pglen = vma_pages(vma); 8041da177e4SLinus Torvalds if (vma->vm_pgoff + vm_pglen == vm_pgoff) 8052dbf4010SVlastimil Babka return true; 8061da177e4SLinus Torvalds } 8072dbf4010SVlastimil Babka return false; 8081da177e4SLinus Torvalds } 8091da177e4SLinus Torvalds 8101da177e4SLinus Torvalds /* 8119a10064fSColin Cross * Given a mapping request (addr,end,vm_flags,file,pgoff,anon_name), 8129a10064fSColin Cross * figure out whether that can be merged with its predecessor or its 8139a10064fSColin Cross * successor. Or both (it neatly fills a hole). 8141da177e4SLinus Torvalds * 8151da177e4SLinus Torvalds * In most cases - when called for mmap, brk or mremap - [addr,end) is 8161da177e4SLinus Torvalds * certain not to be mapped by the time vma_merge is called; but when 8171da177e4SLinus Torvalds * called for mprotect, it is certain to be already mapped (either at 8181da177e4SLinus Torvalds * an offset within prev, or at the start of next), and the flags of 8191da177e4SLinus Torvalds * this area are about to be changed to vm_flags - and the no-change 8201da177e4SLinus Torvalds * case has already been eliminated. 8211da177e4SLinus Torvalds * 822fcfccd91SLorenzo Stoakes * The following mprotect cases have to be considered, where **** is 8231da177e4SLinus Torvalds * the area passed down from mprotect_fixup, never extending beyond one 824fcfccd91SLorenzo Stoakes * vma, PPPP is the previous vma, CCCC is a concurrent vma that starts 825fcfccd91SLorenzo Stoakes * at the same address as **** and is of the same or larger span, and 826fcfccd91SLorenzo Stoakes * NNNN the next vma after ****: 8271da177e4SLinus Torvalds * 828fcfccd91SLorenzo Stoakes * **** **** **** 829fcfccd91SLorenzo Stoakes * PPPPPPNNNNNN PPPPPPNNNNNN PPPPPPCCCCCC 8305d42ab29SWei Yang * cannot merge might become might become 831fcfccd91SLorenzo Stoakes * PPNNNNNNNNNN PPPPPPPPPPCC 8325d42ab29SWei Yang * mmap, brk or case 4 below case 5 below 8335d42ab29SWei Yang * mremap move: 834fcfccd91SLorenzo Stoakes * **** **** 835fcfccd91SLorenzo Stoakes * PPPP NNNN PPPPCCCCNNNN 8365d42ab29SWei Yang * might become might become 8375d42ab29SWei Yang * PPPPPPPPPPPP 1 or PPPPPPPPPPPP 6 or 838fcfccd91SLorenzo Stoakes * PPPPPPPPNNNN 2 or PPPPPPPPNNNN 7 or 839fcfccd91SLorenzo Stoakes * PPPPNNNNNNNN 3 PPPPNNNNNNNN 8 8401da177e4SLinus Torvalds * 841fcfccd91SLorenzo Stoakes * It is important for case 8 that the vma CCCC overlapping the 842fcfccd91SLorenzo Stoakes * region **** is never going to extended over NNNN. Instead NNNN must 843fcfccd91SLorenzo Stoakes * be extended in region **** and CCCC must be removed. This way in 8440503ea8fSLiam R. Howlett * all cases where vma_merge succeeds, the moment vma_merge drops the 845e86f15eeSAndrea Arcangeli * rmap_locks, the properties of the merged vma will be already 846e86f15eeSAndrea Arcangeli * correct for the whole merged range. Some of those properties like 847e86f15eeSAndrea Arcangeli * vm_page_prot/vm_flags may be accessed by rmap_walks and they must 848e86f15eeSAndrea Arcangeli * be correct for the whole merged range immediately after the 849fcfccd91SLorenzo Stoakes * rmap_locks are released. Otherwise if NNNN would be removed and 850fcfccd91SLorenzo Stoakes * CCCC would be extended over the NNNN range, remove_migration_ptes 851e86f15eeSAndrea Arcangeli * or other rmap walkers (if working on addresses beyond the "end" 852fcfccd91SLorenzo Stoakes * parameter) may establish ptes with the wrong permissions of CCCC 853fcfccd91SLorenzo Stoakes * instead of the right permissions of NNNN. 8540503ea8fSLiam R. Howlett * 8550503ea8fSLiam R. Howlett * In the code below: 8560503ea8fSLiam R. Howlett * PPPP is represented by *prev 857fcfccd91SLorenzo Stoakes * CCCC is represented by *curr or not represented at all (NULL) 858fcfccd91SLorenzo Stoakes * NNNN is represented by *next or not represented at all (NULL) 859fcfccd91SLorenzo Stoakes * **** is not represented - it will be merged and the vma containing the 8609e8a39d2SVlastimil Babka * area is returned, or the function will return NULL 8611da177e4SLinus Torvalds */ 86293bf5d4aSLorenzo Stoakes static struct vm_area_struct 8632c8b9474SYajun Deng *vma_merge(struct vma_iterator *vmi, struct vm_area_struct *prev, 8642c8b9474SYajun Deng struct vm_area_struct *src, unsigned long addr, unsigned long end, 8652c8b9474SYajun Deng unsigned long vm_flags, pgoff_t pgoff, struct mempolicy *policy, 8669a10064fSColin Cross struct vm_userfaultfd_ctx vm_userfaultfd_ctx, 8675c26f6acSSuren Baghdasaryan struct anon_vma_name *anon_name) 8681da177e4SLinus Torvalds { 8692c8b9474SYajun Deng struct mm_struct *mm = src->vm_mm; 8702c8b9474SYajun Deng struct anon_vma *anon_vma = src->anon_vma; 8712c8b9474SYajun Deng struct file *file = src->vm_file; 872b0729ae0SLorenzo Stoakes struct vm_area_struct *curr, *next, *res; 8730503ea8fSLiam R. Howlett struct vm_area_struct *vma, *adjust, *remove, *remove2; 874824135c4SLiam R. Howlett struct vm_area_struct *anon_dup = NULL; 8750173db4fSLorenzo Stoakes struct vma_prepare vp; 8760173db4fSLorenzo Stoakes pgoff_t vma_pgoff; 8770173db4fSLorenzo Stoakes int err = 0; 878eef19944SJakub Matěna bool merge_prev = false; 879eef19944SJakub Matěna bool merge_next = false; 8800503ea8fSLiam R. Howlett bool vma_expanded = false; 8810503ea8fSLiam R. Howlett unsigned long vma_start = addr; 8820173db4fSLorenzo Stoakes unsigned long vma_end = end; 8830173db4fSLorenzo Stoakes pgoff_t pglen = (end - addr) >> PAGE_SHIFT; 8840173db4fSLorenzo Stoakes long adj_start = 0; 8851da177e4SLinus Torvalds 8861da177e4SLinus Torvalds /* 8871da177e4SLinus Torvalds * We later require that vma->vm_flags == vm_flags, 8881da177e4SLinus Torvalds * so this tests vma->vm_flags & VM_SPECIAL, too. 8891da177e4SLinus Torvalds */ 8901da177e4SLinus Torvalds if (vm_flags & VM_SPECIAL) 8911da177e4SLinus Torvalds return NULL; 8921da177e4SLinus Torvalds 89300cd00a6SLorenzo Stoakes /* Does the input range span an existing VMA? (cases 5 - 8) */ 89400cd00a6SLorenzo Stoakes curr = find_vma_intersection(mm, prev ? prev->vm_end : 0, end); 8951da177e4SLinus Torvalds 89600cd00a6SLorenzo Stoakes if (!curr || /* cases 1 - 4 */ 89700cd00a6SLorenzo Stoakes end == curr->vm_end) /* cases 6 - 8, adjacent VMA */ 89800cd00a6SLorenzo Stoakes next = vma_lookup(mm, end); 89900cd00a6SLorenzo Stoakes else 90000cd00a6SLorenzo Stoakes next = NULL; /* case 5 */ 901e86f15eeSAndrea Arcangeli 9020503ea8fSLiam R. Howlett if (prev) { 9030503ea8fSLiam R. Howlett vma_start = prev->vm_start; 9040503ea8fSLiam R. Howlett vma_pgoff = prev->vm_pgoff; 9050173db4fSLorenzo Stoakes 906eef19944SJakub Matěna /* Can we merge the predecessor? */ 9070173db4fSLorenzo Stoakes if (addr == prev->vm_end && mpol_equal(vma_policy(prev), policy) 9080503ea8fSLiam R. Howlett && can_vma_merge_after(prev, vm_flags, anon_vma, file, 9090503ea8fSLiam R. Howlett pgoff, vm_userfaultfd_ctx, anon_name)) { 910eef19944SJakub Matěna merge_prev = true; 91118b098afSLiam R. Howlett vma_prev(vmi); 9121da177e4SLinus Torvalds } 9130503ea8fSLiam R. Howlett } 914b0729ae0SLorenzo Stoakes 915eef19944SJakub Matěna /* Can we merge the successor? */ 91600cd00a6SLorenzo Stoakes if (next && mpol_equal(policy, vma_policy(next)) && 9170173db4fSLorenzo Stoakes can_vma_merge_before(next, vm_flags, anon_vma, file, pgoff+pglen, 9189a10064fSColin Cross vm_userfaultfd_ctx, anon_name)) { 919eef19944SJakub Matěna merge_next = true; 920eef19944SJakub Matěna } 9210503ea8fSLiam R. Howlett 92229417d29SLorenzo Stoakes /* Verify some invariant that must be enforced by the caller. */ 92329417d29SLorenzo Stoakes VM_WARN_ON(prev && addr <= prev->vm_start); 92429417d29SLorenzo Stoakes VM_WARN_ON(curr && (addr != curr->vm_start || end > curr->vm_end)); 92529417d29SLorenzo Stoakes VM_WARN_ON(addr >= end); 92629417d29SLorenzo Stoakes 9270173db4fSLorenzo Stoakes if (!merge_prev && !merge_next) 9280173db4fSLorenzo Stoakes return NULL; /* Not mergeable. */ 9290173db4fSLorenzo Stoakes 930c9d6e982SSuren Baghdasaryan if (merge_prev) 931c9d6e982SSuren Baghdasaryan vma_start_write(prev); 932c9d6e982SSuren Baghdasaryan 9330173db4fSLorenzo Stoakes res = vma = prev; 9340503ea8fSLiam R. Howlett remove = remove2 = adjust = NULL; 9350173db4fSLorenzo Stoakes 936eef19944SJakub Matěna /* Can we merge both the predecessor and the successor? */ 937eef19944SJakub Matěna if (merge_prev && merge_next && 9380503ea8fSLiam R. Howlett is_mergeable_anon_vma(prev->anon_vma, next->anon_vma, NULL)) { 939c9d6e982SSuren Baghdasaryan vma_start_write(next); 9405ff783f1SVlastimil Babka remove = next; /* case 1 */ 9410503ea8fSLiam R. Howlett vma_end = next->vm_end; 942824135c4SLiam R. Howlett err = dup_anon_vma(prev, next, &anon_dup); 943fcfccd91SLorenzo Stoakes if (curr) { /* case 6 */ 944c9d6e982SSuren Baghdasaryan vma_start_write(curr); 945fcfccd91SLorenzo Stoakes remove = curr; 9460503ea8fSLiam R. Howlett remove2 = next; 9479a12d103SLiam R. Howlett /* 9489a12d103SLiam R. Howlett * Note that the dup_anon_vma below cannot overwrite err 9499a12d103SLiam R. Howlett * since the first caller would do nothing unless next 9509a12d103SLiam R. Howlett * has an anon_vma. 9519a12d103SLiam R. Howlett */ 9525ff783f1SVlastimil Babka if (!next->anon_vma) 953824135c4SLiam R. Howlett err = dup_anon_vma(prev, curr, &anon_dup); 9540503ea8fSLiam R. Howlett } 9550173db4fSLorenzo Stoakes } else if (merge_prev) { /* case 2 */ 956fcfccd91SLorenzo Stoakes if (curr) { 957c9d6e982SSuren Baghdasaryan vma_start_write(curr); 958fcfccd91SLorenzo Stoakes if (end == curr->vm_end) { /* case 7 */ 959fc0c8f90SVlastimil Babka /* 960fc0c8f90SVlastimil Babka * can_vma_merge_after() assumed we would not be 961fc0c8f90SVlastimil Babka * removing prev vma, so it skipped the check 962fc0c8f90SVlastimil Babka * for vm_ops->close, but we are removing curr 963fc0c8f90SVlastimil Babka */ 964fc0c8f90SVlastimil Babka if (curr->vm_ops && curr->vm_ops->close) 965fc0c8f90SVlastimil Babka err = -EINVAL; 966fcfccd91SLorenzo Stoakes remove = curr; 9670503ea8fSLiam R. Howlett } else { /* case 5 */ 968fcfccd91SLorenzo Stoakes adjust = curr; 969fcfccd91SLorenzo Stoakes adj_start = (end - curr->vm_start); 9700503ea8fSLiam R. Howlett } 971fc0c8f90SVlastimil Babka if (!err) 972fc0c8f90SVlastimil Babka err = dup_anon_vma(prev, curr, &anon_dup); 9730503ea8fSLiam R. Howlett } 9740173db4fSLorenzo Stoakes } else { /* merge_next */ 975c9d6e982SSuren Baghdasaryan vma_start_write(next); 976eef19944SJakub Matěna res = next; 9770503ea8fSLiam R. Howlett if (prev && addr < prev->vm_end) { /* case 4 */ 978c9d6e982SSuren Baghdasaryan vma_start_write(prev); 9790503ea8fSLiam R. Howlett vma_end = addr; 980183b7a60SVlastimil Babka adjust = next; 9811e76454fSVlastimil Babka adj_start = -(prev->vm_end - addr); 982824135c4SLiam R. Howlett err = dup_anon_vma(next, prev, &anon_dup); 9830503ea8fSLiam R. Howlett } else { 984b0729ae0SLorenzo Stoakes /* 985b0729ae0SLorenzo Stoakes * Note that cases 3 and 8 are the ONLY ones where prev 986b0729ae0SLorenzo Stoakes * is permitted to be (but is not necessarily) NULL. 987b0729ae0SLorenzo Stoakes */ 9880503ea8fSLiam R. Howlett vma = next; /* case 3 */ 9890503ea8fSLiam R. Howlett vma_start = addr; 9900503ea8fSLiam R. Howlett vma_end = next->vm_end; 9917e775787SVlastimil Babka vma_pgoff = next->vm_pgoff - pglen; 992fcfccd91SLorenzo Stoakes if (curr) { /* case 8 */ 993fcfccd91SLorenzo Stoakes vma_pgoff = curr->vm_pgoff; 994c9d6e982SSuren Baghdasaryan vma_start_write(curr); 995fcfccd91SLorenzo Stoakes remove = curr; 996824135c4SLiam R. Howlett err = dup_anon_vma(next, curr, &anon_dup); 9970503ea8fSLiam R. Howlett } 9980503ea8fSLiam R. Howlett } 9991da177e4SLinus Torvalds } 10001da177e4SLinus Torvalds 10010173db4fSLorenzo Stoakes /* Error in anon_vma clone. */ 1002eef19944SJakub Matěna if (err) 10031419430cSLiam R. Howlett goto anon_vma_fail; 10040503ea8fSLiam R. Howlett 1005b5df0922SLiam R. Howlett if (vma_start < vma->vm_start || vma_end > vma->vm_end) 1006b5df0922SLiam R. Howlett vma_expanded = true; 1007b5df0922SLiam R. Howlett 1008b5df0922SLiam R. Howlett if (vma_expanded) { 1009b5df0922SLiam R. Howlett vma_iter_config(vmi, vma_start, vma_end); 1010b5df0922SLiam R. Howlett } else { 1011b5df0922SLiam R. Howlett vma_iter_config(vmi, adjust->vm_start + adj_start, 1012b5df0922SLiam R. Howlett adjust->vm_end); 1013b5df0922SLiam R. Howlett } 1014b5df0922SLiam R. Howlett 1015b5df0922SLiam R. Howlett if (vma_iter_prealloc(vmi, vma)) 10161419430cSLiam R. Howlett goto prealloc_fail; 10170503ea8fSLiam R. Howlett 10180503ea8fSLiam R. Howlett init_multi_vma_prep(&vp, vma, adjust, remove, remove2); 10190503ea8fSLiam R. Howlett VM_WARN_ON(vp.anon_vma && adjust && adjust->anon_vma && 10200503ea8fSLiam R. Howlett vp.anon_vma != adjust->anon_vma); 10210503ea8fSLiam R. Howlett 10220503ea8fSLiam R. Howlett vma_prepare(&vp); 1023ccf1d78dSSuren Baghdasaryan vma_adjust_trans_huge(vma, vma_start, vma_end, adj_start); 1024412c6ef9SYajun Deng vma_set_range(vma, vma_start, vma_end, vma_pgoff); 10250503ea8fSLiam R. Howlett 10260503ea8fSLiam R. Howlett if (vma_expanded) 10270503ea8fSLiam R. Howlett vma_iter_store(vmi, vma); 10280503ea8fSLiam R. Howlett 10291e76454fSVlastimil Babka if (adj_start) { 10301e76454fSVlastimil Babka adjust->vm_start += adj_start; 10311e76454fSVlastimil Babka adjust->vm_pgoff += adj_start >> PAGE_SHIFT; 10321e76454fSVlastimil Babka if (adj_start < 0) { 10330503ea8fSLiam R. Howlett WARN_ON(vma_expanded); 10340503ea8fSLiam R. Howlett vma_iter_store(vmi, next); 10350503ea8fSLiam R. Howlett } 10360503ea8fSLiam R. Howlett } 10370503ea8fSLiam R. Howlett 10380503ea8fSLiam R. Howlett vma_complete(&vp, vmi, mm); 1039eef19944SJakub Matěna khugepaged_enter_vma(res, vm_flags); 10409760ebffSLiam R. Howlett return res; 10411419430cSLiam R. Howlett 10421419430cSLiam R. Howlett prealloc_fail: 1043824135c4SLiam R. Howlett if (anon_dup) 1044824135c4SLiam R. Howlett unlink_anon_vmas(anon_dup); 1045824135c4SLiam R. Howlett 10461419430cSLiam R. Howlett anon_vma_fail: 10471419430cSLiam R. Howlett vma_iter_set(vmi, addr); 10481419430cSLiam R. Howlett vma_iter_load(vmi); 10491419430cSLiam R. Howlett return NULL; 1050f2ebfe43SLiam R. Howlett } 1051f2ebfe43SLiam R. Howlett 10521da177e4SLinus Torvalds /* 1053b4f315b4SEthon Paul * Rough compatibility check to quickly see if it's even worth looking 1054d0e9fe17SLinus Torvalds * at sharing an anon_vma. 1055d0e9fe17SLinus Torvalds * 1056d0e9fe17SLinus Torvalds * They need to have the same vm_file, and the flags can only differ 1057d0e9fe17SLinus Torvalds * in things that mprotect may change. 1058d0e9fe17SLinus Torvalds * 1059d0e9fe17SLinus Torvalds * NOTE! The fact that we share an anon_vma doesn't _have_ to mean that 1060d0e9fe17SLinus Torvalds * we can merge the two vma's. For example, we refuse to merge a vma if 1061d0e9fe17SLinus Torvalds * there is a vm_ops->close() function, because that indicates that the 1062d0e9fe17SLinus Torvalds * driver is doing some kind of reference counting. But that doesn't 1063d0e9fe17SLinus Torvalds * really matter for the anon_vma sharing case. 1064d0e9fe17SLinus Torvalds */ 1065d0e9fe17SLinus Torvalds static int anon_vma_compatible(struct vm_area_struct *a, struct vm_area_struct *b) 1066d0e9fe17SLinus Torvalds { 1067d0e9fe17SLinus Torvalds return a->vm_end == b->vm_start && 1068d0e9fe17SLinus Torvalds mpol_equal(vma_policy(a), vma_policy(b)) && 1069d0e9fe17SLinus Torvalds a->vm_file == b->vm_file && 10706cb4d9a2SAnshuman Khandual !((a->vm_flags ^ b->vm_flags) & ~(VM_ACCESS_FLAGS | VM_SOFTDIRTY)) && 1071d0e9fe17SLinus Torvalds b->vm_pgoff == a->vm_pgoff + ((b->vm_start - a->vm_start) >> PAGE_SHIFT); 1072d0e9fe17SLinus Torvalds } 1073d0e9fe17SLinus Torvalds 1074d0e9fe17SLinus Torvalds /* 1075d0e9fe17SLinus Torvalds * Do some basic sanity checking to see if we can re-use the anon_vma 1076d0e9fe17SLinus Torvalds * from 'old'. The 'a'/'b' vma's are in VM order - one of them will be 1077d0e9fe17SLinus Torvalds * the same as 'old', the other will be the new one that is trying 1078d0e9fe17SLinus Torvalds * to share the anon_vma. 1079d0e9fe17SLinus Torvalds * 10805b449489SFlorian Rommel * NOTE! This runs with mmap_lock held for reading, so it is possible that 1081d0e9fe17SLinus Torvalds * the anon_vma of 'old' is concurrently in the process of being set up 1082d0e9fe17SLinus Torvalds * by another page fault trying to merge _that_. But that's ok: if it 1083d0e9fe17SLinus Torvalds * is being set up, that automatically means that it will be a singleton 1084d0e9fe17SLinus Torvalds * acceptable for merging, so we can do all of this optimistically. But 10854db0c3c2SJason Low * we do that READ_ONCE() to make sure that we never re-load the pointer. 1086d0e9fe17SLinus Torvalds * 1087d0e9fe17SLinus Torvalds * IOW: that the "list_is_singular()" test on the anon_vma_chain only 1088d0e9fe17SLinus Torvalds * matters for the 'stable anon_vma' case (ie the thing we want to avoid 1089d0e9fe17SLinus Torvalds * is to return an anon_vma that is "complex" due to having gone through 1090d0e9fe17SLinus Torvalds * a fork). 1091d0e9fe17SLinus Torvalds * 1092d0e9fe17SLinus Torvalds * We also make sure that the two vma's are compatible (adjacent, 1093d0e9fe17SLinus Torvalds * and with the same memory policies). That's all stable, even with just 10945b449489SFlorian Rommel * a read lock on the mmap_lock. 1095d0e9fe17SLinus Torvalds */ 1096d0e9fe17SLinus Torvalds static struct anon_vma *reusable_anon_vma(struct vm_area_struct *old, struct vm_area_struct *a, struct vm_area_struct *b) 1097d0e9fe17SLinus Torvalds { 1098d0e9fe17SLinus Torvalds if (anon_vma_compatible(a, b)) { 10994db0c3c2SJason Low struct anon_vma *anon_vma = READ_ONCE(old->anon_vma); 1100d0e9fe17SLinus Torvalds 1101d0e9fe17SLinus Torvalds if (anon_vma && list_is_singular(&old->anon_vma_chain)) 1102d0e9fe17SLinus Torvalds return anon_vma; 1103d0e9fe17SLinus Torvalds } 1104d0e9fe17SLinus Torvalds return NULL; 1105d0e9fe17SLinus Torvalds } 1106d0e9fe17SLinus Torvalds 1107d0e9fe17SLinus Torvalds /* 11081da177e4SLinus Torvalds * find_mergeable_anon_vma is used by anon_vma_prepare, to check 11091da177e4SLinus Torvalds * neighbouring vmas for a suitable anon_vma, before it goes off 11101da177e4SLinus Torvalds * to allocate a new anon_vma. It checks because a repetitive 11111da177e4SLinus Torvalds * sequence of mprotects and faults may otherwise lead to distinct 11121da177e4SLinus Torvalds * anon_vmas being allocated, preventing vma merge in subsequent 11131da177e4SLinus Torvalds * mprotect. 11141da177e4SLinus Torvalds */ 11151da177e4SLinus Torvalds struct anon_vma *find_mergeable_anon_vma(struct vm_area_struct *vma) 11161da177e4SLinus Torvalds { 1117763ecb03SLiam R. Howlett MA_STATE(mas, &vma->vm_mm->mm_mt, vma->vm_end, vma->vm_end); 1118a67c8caaSMiaohe Lin struct anon_vma *anon_vma = NULL; 1119763ecb03SLiam R. Howlett struct vm_area_struct *prev, *next; 11201da177e4SLinus Torvalds 1121a67c8caaSMiaohe Lin /* Try next first. */ 1122763ecb03SLiam R. Howlett next = mas_walk(&mas); 1123763ecb03SLiam R. Howlett if (next) { 1124763ecb03SLiam R. Howlett anon_vma = reusable_anon_vma(next, vma, next); 1125d0e9fe17SLinus Torvalds if (anon_vma) 1126d0e9fe17SLinus Torvalds return anon_vma; 1127a67c8caaSMiaohe Lin } 11281da177e4SLinus Torvalds 1129763ecb03SLiam R. Howlett prev = mas_prev(&mas, 0); 1130763ecb03SLiam R. Howlett VM_BUG_ON_VMA(prev != vma, vma); 1131763ecb03SLiam R. Howlett prev = mas_prev(&mas, 0); 1132a67c8caaSMiaohe Lin /* Try prev next. */ 1133763ecb03SLiam R. Howlett if (prev) 1134763ecb03SLiam R. Howlett anon_vma = reusable_anon_vma(prev, prev, vma); 1135a67c8caaSMiaohe Lin 11361da177e4SLinus Torvalds /* 1137a67c8caaSMiaohe Lin * We might reach here with anon_vma == NULL if we can't find 1138a67c8caaSMiaohe Lin * any reusable anon_vma. 11391da177e4SLinus Torvalds * There's no absolute need to look only at touching neighbours: 11401da177e4SLinus Torvalds * we could search further afield for "compatible" anon_vmas. 11411da177e4SLinus Torvalds * But it would probably just be a waste of time searching, 11421da177e4SLinus Torvalds * or lead to too many vmas hanging off the same anon_vma. 11431da177e4SLinus Torvalds * We're trying to allow mprotect remerging later on, 11441da177e4SLinus Torvalds * not trying to minimize memory used for anon_vmas. 11451da177e4SLinus Torvalds */ 1146a67c8caaSMiaohe Lin return anon_vma; 11471da177e4SLinus Torvalds } 11481da177e4SLinus Torvalds 11491da177e4SLinus Torvalds /* 115040401530SAl Viro * If a hint addr is less than mmap_min_addr change hint to be as 115140401530SAl Viro * low as possible but still greater than mmap_min_addr 115240401530SAl Viro */ 115340401530SAl Viro static inline unsigned long round_hint_to_min(unsigned long hint) 115440401530SAl Viro { 115540401530SAl Viro hint &= PAGE_MASK; 115640401530SAl Viro if (((void *)hint != NULL) && 115740401530SAl Viro (hint < mmap_min_addr)) 115840401530SAl Viro return PAGE_ALIGN(mmap_min_addr); 115940401530SAl Viro return hint; 116040401530SAl Viro } 116140401530SAl Viro 1162b0cc5e89SAndrew Morton bool mlock_future_ok(struct mm_struct *mm, unsigned long flags, 11633c54a298SLorenzo Stoakes unsigned long bytes) 1164363ee17fSDavidlohr Bueso { 11653c54a298SLorenzo Stoakes unsigned long locked_pages, limit_pages; 1166363ee17fSDavidlohr Bueso 11673c54a298SLorenzo Stoakes if (!(flags & VM_LOCKED) || capable(CAP_IPC_LOCK)) 11683c54a298SLorenzo Stoakes return true; 11693c54a298SLorenzo Stoakes 11703c54a298SLorenzo Stoakes locked_pages = bytes >> PAGE_SHIFT; 11713c54a298SLorenzo Stoakes locked_pages += mm->locked_vm; 11723c54a298SLorenzo Stoakes 11733c54a298SLorenzo Stoakes limit_pages = rlimit(RLIMIT_MEMLOCK); 11743c54a298SLorenzo Stoakes limit_pages >>= PAGE_SHIFT; 11753c54a298SLorenzo Stoakes 11763c54a298SLorenzo Stoakes return locked_pages <= limit_pages; 1177363ee17fSDavidlohr Bueso } 1178363ee17fSDavidlohr Bueso 1179be83bbf8SLinus Torvalds static inline u64 file_mmap_size_max(struct file *file, struct inode *inode) 1180be83bbf8SLinus Torvalds { 1181be83bbf8SLinus Torvalds if (S_ISREG(inode->i_mode)) 1182423913adSLinus Torvalds return MAX_LFS_FILESIZE; 1183be83bbf8SLinus Torvalds 1184be83bbf8SLinus Torvalds if (S_ISBLK(inode->i_mode)) 1185be83bbf8SLinus Torvalds return MAX_LFS_FILESIZE; 1186be83bbf8SLinus Torvalds 118776f34950SIvan Khoronzhuk if (S_ISSOCK(inode->i_mode)) 118876f34950SIvan Khoronzhuk return MAX_LFS_FILESIZE; 118976f34950SIvan Khoronzhuk 1190be83bbf8SLinus Torvalds /* Special "we do even unsigned file positions" case */ 1191be83bbf8SLinus Torvalds if (file->f_mode & FMODE_UNSIGNED_OFFSET) 1192be83bbf8SLinus Torvalds return 0; 1193be83bbf8SLinus Torvalds 1194be83bbf8SLinus Torvalds /* Yes, random drivers might want more. But I'm tired of buggy drivers */ 1195be83bbf8SLinus Torvalds return ULONG_MAX; 1196be83bbf8SLinus Torvalds } 1197be83bbf8SLinus Torvalds 1198be83bbf8SLinus Torvalds static inline bool file_mmap_ok(struct file *file, struct inode *inode, 1199be83bbf8SLinus Torvalds unsigned long pgoff, unsigned long len) 1200be83bbf8SLinus Torvalds { 1201be83bbf8SLinus Torvalds u64 maxsize = file_mmap_size_max(file, inode); 1202be83bbf8SLinus Torvalds 1203be83bbf8SLinus Torvalds if (maxsize && len > maxsize) 1204be83bbf8SLinus Torvalds return false; 1205be83bbf8SLinus Torvalds maxsize -= len; 1206be83bbf8SLinus Torvalds if (pgoff > maxsize >> PAGE_SHIFT) 1207be83bbf8SLinus Torvalds return false; 1208be83bbf8SLinus Torvalds return true; 1209be83bbf8SLinus Torvalds } 1210be83bbf8SLinus Torvalds 121140401530SAl Viro /* 12123e4e28c5SMichel Lespinasse * The caller must write-lock current->mm->mmap_lock. 12131da177e4SLinus Torvalds */ 12141fcfd8dbSOleg Nesterov unsigned long do_mmap(struct file *file, unsigned long addr, 12151da177e4SLinus Torvalds unsigned long len, unsigned long prot, 1216592b5fadSYu-cheng Yu unsigned long flags, vm_flags_t vm_flags, 1217592b5fadSYu-cheng Yu unsigned long pgoff, unsigned long *populate, 1218592b5fadSYu-cheng Yu struct list_head *uf) 12191da177e4SLinus Torvalds { 12201da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 122162b5f7d0SDave Hansen int pkey = 0; 12221da177e4SLinus Torvalds 122341badc15SMichel Lespinasse *populate = 0; 1224bebeb3d6SMichel Lespinasse 1225e37609bbSPiotr Kwapulinski if (!len) 1226e37609bbSPiotr Kwapulinski return -EINVAL; 1227e37609bbSPiotr Kwapulinski 12281da177e4SLinus Torvalds /* 12291da177e4SLinus Torvalds * Does the application expect PROT_READ to imply PROT_EXEC? 12301da177e4SLinus Torvalds * 12311da177e4SLinus Torvalds * (the exception is when the underlying filesystem is noexec 1232be16dd76SMuhammad Muzammil * mounted, in which case we don't add PROT_EXEC.) 12331da177e4SLinus Torvalds */ 12341da177e4SLinus Torvalds if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC)) 123590f8572bSEric W. Biederman if (!(file && path_noexec(&file->f_path))) 12361da177e4SLinus Torvalds prot |= PROT_EXEC; 12371da177e4SLinus Torvalds 1238a4ff8e86SMichal Hocko /* force arch specific MAP_FIXED handling in get_unmapped_area */ 1239a4ff8e86SMichal Hocko if (flags & MAP_FIXED_NOREPLACE) 1240a4ff8e86SMichal Hocko flags |= MAP_FIXED; 1241a4ff8e86SMichal Hocko 12427cd94146SEric Paris if (!(flags & MAP_FIXED)) 12437cd94146SEric Paris addr = round_hint_to_min(addr); 12447cd94146SEric Paris 12451da177e4SLinus Torvalds /* Careful about overflows.. */ 12461da177e4SLinus Torvalds len = PAGE_ALIGN(len); 12479206de95SAl Viro if (!len) 12481da177e4SLinus Torvalds return -ENOMEM; 12491da177e4SLinus Torvalds 12501da177e4SLinus Torvalds /* offset overflow? */ 12511da177e4SLinus Torvalds if ((pgoff + (len >> PAGE_SHIFT)) < pgoff) 12521da177e4SLinus Torvalds return -EOVERFLOW; 12531da177e4SLinus Torvalds 12541da177e4SLinus Torvalds /* Too many mappings? */ 12551da177e4SLinus Torvalds if (mm->map_count > sysctl_max_map_count) 12561da177e4SLinus Torvalds return -ENOMEM; 12571da177e4SLinus Torvalds 12581da177e4SLinus Torvalds /* Obtain the address to map to. we verify (or select) it and ensure 12591da177e4SLinus Torvalds * that it represents a valid section of the address space. 12601da177e4SLinus Torvalds */ 12611da177e4SLinus Torvalds addr = get_unmapped_area(file, addr, len, pgoff, flags); 1262ff68dac6SGaowei Pu if (IS_ERR_VALUE(addr)) 12631da177e4SLinus Torvalds return addr; 12641da177e4SLinus Torvalds 1265a4ff8e86SMichal Hocko if (flags & MAP_FIXED_NOREPLACE) { 126635e43c5fSLiam Howlett if (find_vma_intersection(mm, addr, addr + len)) 1267a4ff8e86SMichal Hocko return -EEXIST; 1268a4ff8e86SMichal Hocko } 1269a4ff8e86SMichal Hocko 127062b5f7d0SDave Hansen if (prot == PROT_EXEC) { 127162b5f7d0SDave Hansen pkey = execute_only_pkey(mm); 127262b5f7d0SDave Hansen if (pkey < 0) 127362b5f7d0SDave Hansen pkey = 0; 127462b5f7d0SDave Hansen } 127562b5f7d0SDave Hansen 12761da177e4SLinus Torvalds /* Do simple checking here so the lower-level routines won't have 12771da177e4SLinus Torvalds * to. we assume access permissions have been handled by the open 12781da177e4SLinus Torvalds * of the memory object, so we don't do any here. 12791da177e4SLinus Torvalds */ 1280592b5fadSYu-cheng Yu vm_flags |= calc_vm_prot_bits(prot, pkey) | calc_vm_flag_bits(flags) | 12811da177e4SLinus Torvalds mm->def_flags | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC; 12821da177e4SLinus Torvalds 1283cdf7b341SHuang Shijie if (flags & MAP_LOCKED) 12841da177e4SLinus Torvalds if (!can_do_mlock()) 12851da177e4SLinus Torvalds return -EPERM; 1286ba470de4SRik van Riel 1287b0cc5e89SAndrew Morton if (!mlock_future_ok(mm, vm_flags, len)) 12881da177e4SLinus Torvalds return -EAGAIN; 12891da177e4SLinus Torvalds 12901da177e4SLinus Torvalds if (file) { 1291077bf22bSOleg Nesterov struct inode *inode = file_inode(file); 12921c972597SDan Williams unsigned long flags_mask; 12931c972597SDan Williams 1294be83bbf8SLinus Torvalds if (!file_mmap_ok(file, inode, pgoff, len)) 1295be83bbf8SLinus Torvalds return -EOVERFLOW; 1296be83bbf8SLinus Torvalds 12971c972597SDan Williams flags_mask = LEGACY_MAP_MASK | file->f_op->mmap_supported_flags; 1298077bf22bSOleg Nesterov 12991da177e4SLinus Torvalds switch (flags & MAP_TYPE) { 13001da177e4SLinus Torvalds case MAP_SHARED: 13011c972597SDan Williams /* 13021c972597SDan Williams * Force use of MAP_SHARED_VALIDATE with non-legacy 13031c972597SDan Williams * flags. E.g. MAP_SYNC is dangerous to use with 13041c972597SDan Williams * MAP_SHARED as you don't know which consistency model 13051c972597SDan Williams * you will get. We silently ignore unsupported flags 13061c972597SDan Williams * with MAP_SHARED to preserve backward compatibility. 13071c972597SDan Williams */ 13081c972597SDan Williams flags &= LEGACY_MAP_MASK; 1309e4a9bc58SJoe Perches fallthrough; 13101c972597SDan Williams case MAP_SHARED_VALIDATE: 13111c972597SDan Williams if (flags & ~flags_mask) 13121c972597SDan Williams return -EOPNOTSUPP; 1313dc617f29SDarrick J. Wong if (prot & PROT_WRITE) { 1314dc617f29SDarrick J. Wong if (!(file->f_mode & FMODE_WRITE)) 13151da177e4SLinus Torvalds return -EACCES; 1316dc617f29SDarrick J. Wong if (IS_SWAPFILE(file->f_mapping->host)) 1317dc617f29SDarrick J. Wong return -ETXTBSY; 1318dc617f29SDarrick J. Wong } 13191da177e4SLinus Torvalds 13201da177e4SLinus Torvalds /* 13211da177e4SLinus Torvalds * Make sure we don't allow writing to an append-only 13221da177e4SLinus Torvalds * file.. 13231da177e4SLinus Torvalds */ 13241da177e4SLinus Torvalds if (IS_APPEND(inode) && (file->f_mode & FMODE_WRITE)) 13251da177e4SLinus Torvalds return -EACCES; 13261da177e4SLinus Torvalds 13271da177e4SLinus Torvalds vm_flags |= VM_SHARED | VM_MAYSHARE; 13281da177e4SLinus Torvalds if (!(file->f_mode & FMODE_WRITE)) 13291da177e4SLinus Torvalds vm_flags &= ~(VM_MAYWRITE | VM_SHARED); 1330e4a9bc58SJoe Perches fallthrough; 13311da177e4SLinus Torvalds case MAP_PRIVATE: 13321da177e4SLinus Torvalds if (!(file->f_mode & FMODE_READ)) 13331da177e4SLinus Torvalds return -EACCES; 133490f8572bSEric W. Biederman if (path_noexec(&file->f_path)) { 133580c5606cSLinus Torvalds if (vm_flags & VM_EXEC) 133680c5606cSLinus Torvalds return -EPERM; 133780c5606cSLinus Torvalds vm_flags &= ~VM_MAYEXEC; 133880c5606cSLinus Torvalds } 133980c5606cSLinus Torvalds 134072c2d531SAl Viro if (!file->f_op->mmap) 134180c5606cSLinus Torvalds return -ENODEV; 1342b2c56e4fSOleg Nesterov if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP)) 1343b2c56e4fSOleg Nesterov return -EINVAL; 13441da177e4SLinus Torvalds break; 13451da177e4SLinus Torvalds 13461da177e4SLinus Torvalds default: 13471da177e4SLinus Torvalds return -EINVAL; 13481da177e4SLinus Torvalds } 13491da177e4SLinus Torvalds } else { 13501da177e4SLinus Torvalds switch (flags & MAP_TYPE) { 13511da177e4SLinus Torvalds case MAP_SHARED: 1352b2c56e4fSOleg Nesterov if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP)) 1353b2c56e4fSOleg Nesterov return -EINVAL; 1354ce363942STejun Heo /* 1355ce363942STejun Heo * Ignore pgoff. 1356ce363942STejun Heo */ 1357ce363942STejun Heo pgoff = 0; 13581da177e4SLinus Torvalds vm_flags |= VM_SHARED | VM_MAYSHARE; 13591da177e4SLinus Torvalds break; 13601da177e4SLinus Torvalds case MAP_PRIVATE: 13611da177e4SLinus Torvalds /* 13621da177e4SLinus Torvalds * Set pgoff according to addr for anon_vma. 13631da177e4SLinus Torvalds */ 13641da177e4SLinus Torvalds pgoff = addr >> PAGE_SHIFT; 13651da177e4SLinus Torvalds break; 13661da177e4SLinus Torvalds default: 13671da177e4SLinus Torvalds return -EINVAL; 13681da177e4SLinus Torvalds } 13691da177e4SLinus Torvalds } 13701da177e4SLinus Torvalds 1371c22c0d63SMichel Lespinasse /* 1372c22c0d63SMichel Lespinasse * Set 'VM_NORESERVE' if we should not account for the 1373c22c0d63SMichel Lespinasse * memory use of this mapping. 1374c22c0d63SMichel Lespinasse */ 1375c22c0d63SMichel Lespinasse if (flags & MAP_NORESERVE) { 1376c22c0d63SMichel Lespinasse /* We honor MAP_NORESERVE if allowed to overcommit */ 1377c22c0d63SMichel Lespinasse if (sysctl_overcommit_memory != OVERCOMMIT_NEVER) 1378c22c0d63SMichel Lespinasse vm_flags |= VM_NORESERVE; 1379c22c0d63SMichel Lespinasse 1380c22c0d63SMichel Lespinasse /* hugetlb applies strict overcommit unless MAP_NORESERVE */ 1381c22c0d63SMichel Lespinasse if (file && is_file_hugepages(file)) 1382c22c0d63SMichel Lespinasse vm_flags |= VM_NORESERVE; 1383c22c0d63SMichel Lespinasse } 1384c22c0d63SMichel Lespinasse 1385897ab3e0SMike Rapoport addr = mmap_region(file, addr, len, vm_flags, pgoff, uf); 138609a9f1d2SMichel Lespinasse if (!IS_ERR_VALUE(addr) && 138709a9f1d2SMichel Lespinasse ((vm_flags & VM_LOCKED) || 138809a9f1d2SMichel Lespinasse (flags & (MAP_POPULATE | MAP_NONBLOCK)) == MAP_POPULATE)) 138941badc15SMichel Lespinasse *populate = len; 1390bebeb3d6SMichel Lespinasse return addr; 13910165ab44SMiklos Szeredi } 13926be5ceb0SLinus Torvalds 1393a90f590aSDominik Brodowski unsigned long ksys_mmap_pgoff(unsigned long addr, unsigned long len, 1394a90f590aSDominik Brodowski unsigned long prot, unsigned long flags, 1395a90f590aSDominik Brodowski unsigned long fd, unsigned long pgoff) 139666f0dc48SHugh Dickins { 139766f0dc48SHugh Dickins struct file *file = NULL; 13981e3ee14bSChen Gang unsigned long retval; 139966f0dc48SHugh Dickins 140066f0dc48SHugh Dickins if (!(flags & MAP_ANONYMOUS)) { 1401120a795dSAl Viro audit_mmap_fd(fd, flags); 140266f0dc48SHugh Dickins file = fget(fd); 140366f0dc48SHugh Dickins if (!file) 14041e3ee14bSChen Gang return -EBADF; 14057bba8f0eSZhen Lei if (is_file_hugepages(file)) { 1406af73e4d9SNaoya Horiguchi len = ALIGN(len, huge_page_size(hstate_file(file))); 14077bba8f0eSZhen Lei } else if (unlikely(flags & MAP_HUGETLB)) { 1408493af578SJörn Engel retval = -EINVAL; 1409493af578SJörn Engel goto out_fput; 14107bba8f0eSZhen Lei } 141166f0dc48SHugh Dickins } else if (flags & MAP_HUGETLB) { 1412c103a4dcSAndrew Morton struct hstate *hs; 1413af73e4d9SNaoya Horiguchi 141420ac2893SAnshuman Khandual hs = hstate_sizelog((flags >> MAP_HUGE_SHIFT) & MAP_HUGE_MASK); 1415091d0d55SLi Zefan if (!hs) 1416091d0d55SLi Zefan return -EINVAL; 1417091d0d55SLi Zefan 1418091d0d55SLi Zefan len = ALIGN(len, huge_page_size(hs)); 141966f0dc48SHugh Dickins /* 142066f0dc48SHugh Dickins * VM_NORESERVE is used because the reservations will be 142166f0dc48SHugh Dickins * taken when vm_ops->mmap() is called 142266f0dc48SHugh Dickins */ 1423af73e4d9SNaoya Horiguchi file = hugetlb_file_setup(HUGETLB_ANON_FILE, len, 142442d7395fSAndi Kleen VM_NORESERVE, 142583c1fd76Szhangyiru HUGETLB_ANONHUGE_INODE, 142642d7395fSAndi Kleen (flags >> MAP_HUGE_SHIFT) & MAP_HUGE_MASK); 142766f0dc48SHugh Dickins if (IS_ERR(file)) 142866f0dc48SHugh Dickins return PTR_ERR(file); 142966f0dc48SHugh Dickins } 143066f0dc48SHugh Dickins 14319fbeb5abSMichal Hocko retval = vm_mmap_pgoff(file, addr, len, prot, flags, pgoff); 1432493af578SJörn Engel out_fput: 143366f0dc48SHugh Dickins if (file) 143466f0dc48SHugh Dickins fput(file); 143566f0dc48SHugh Dickins return retval; 143666f0dc48SHugh Dickins } 143766f0dc48SHugh Dickins 1438a90f590aSDominik Brodowski SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len, 1439a90f590aSDominik Brodowski unsigned long, prot, unsigned long, flags, 1440a90f590aSDominik Brodowski unsigned long, fd, unsigned long, pgoff) 1441a90f590aSDominik Brodowski { 1442a90f590aSDominik Brodowski return ksys_mmap_pgoff(addr, len, prot, flags, fd, pgoff); 1443a90f590aSDominik Brodowski } 1444a90f590aSDominik Brodowski 1445a4679373SChristoph Hellwig #ifdef __ARCH_WANT_SYS_OLD_MMAP 1446a4679373SChristoph Hellwig struct mmap_arg_struct { 1447a4679373SChristoph Hellwig unsigned long addr; 1448a4679373SChristoph Hellwig unsigned long len; 1449a4679373SChristoph Hellwig unsigned long prot; 1450a4679373SChristoph Hellwig unsigned long flags; 1451a4679373SChristoph Hellwig unsigned long fd; 1452a4679373SChristoph Hellwig unsigned long offset; 1453a4679373SChristoph Hellwig }; 1454a4679373SChristoph Hellwig 1455a4679373SChristoph Hellwig SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg) 1456a4679373SChristoph Hellwig { 1457a4679373SChristoph Hellwig struct mmap_arg_struct a; 1458a4679373SChristoph Hellwig 1459a4679373SChristoph Hellwig if (copy_from_user(&a, arg, sizeof(a))) 1460a4679373SChristoph Hellwig return -EFAULT; 1461de1741a1SAlexander Kuleshov if (offset_in_page(a.offset)) 1462a4679373SChristoph Hellwig return -EINVAL; 1463a4679373SChristoph Hellwig 1464a90f590aSDominik Brodowski return ksys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd, 1465a4679373SChristoph Hellwig a.offset >> PAGE_SHIFT); 1466a4679373SChristoph Hellwig } 1467a4679373SChristoph Hellwig #endif /* __ARCH_WANT_SYS_OLD_MMAP */ 1468a4679373SChristoph Hellwig 146954cbbbf3SLorenzo Stoakes static bool vm_ops_needs_writenotify(const struct vm_operations_struct *vm_ops) 147054cbbbf3SLorenzo Stoakes { 147154cbbbf3SLorenzo Stoakes return vm_ops && (vm_ops->page_mkwrite || vm_ops->pfn_mkwrite); 147254cbbbf3SLorenzo Stoakes } 147354cbbbf3SLorenzo Stoakes 147454cbbbf3SLorenzo Stoakes static bool vma_is_shared_writable(struct vm_area_struct *vma) 147554cbbbf3SLorenzo Stoakes { 147654cbbbf3SLorenzo Stoakes return (vma->vm_flags & (VM_WRITE | VM_SHARED)) == 147754cbbbf3SLorenzo Stoakes (VM_WRITE | VM_SHARED); 147854cbbbf3SLorenzo Stoakes } 147954cbbbf3SLorenzo Stoakes 148054cbbbf3SLorenzo Stoakes static bool vma_fs_can_writeback(struct vm_area_struct *vma) 148154cbbbf3SLorenzo Stoakes { 148254cbbbf3SLorenzo Stoakes /* No managed pages to writeback. */ 148354cbbbf3SLorenzo Stoakes if (vma->vm_flags & VM_PFNMAP) 148454cbbbf3SLorenzo Stoakes return false; 148554cbbbf3SLorenzo Stoakes 148654cbbbf3SLorenzo Stoakes return vma->vm_file && vma->vm_file->f_mapping && 148754cbbbf3SLorenzo Stoakes mapping_can_writeback(vma->vm_file->f_mapping); 148854cbbbf3SLorenzo Stoakes } 148954cbbbf3SLorenzo Stoakes 149054cbbbf3SLorenzo Stoakes /* 149154cbbbf3SLorenzo Stoakes * Does this VMA require the underlying folios to have their dirty state 149254cbbbf3SLorenzo Stoakes * tracked? 149354cbbbf3SLorenzo Stoakes */ 149454cbbbf3SLorenzo Stoakes bool vma_needs_dirty_tracking(struct vm_area_struct *vma) 149554cbbbf3SLorenzo Stoakes { 149654cbbbf3SLorenzo Stoakes /* Only shared, writable VMAs require dirty tracking. */ 149754cbbbf3SLorenzo Stoakes if (!vma_is_shared_writable(vma)) 149854cbbbf3SLorenzo Stoakes return false; 149954cbbbf3SLorenzo Stoakes 150054cbbbf3SLorenzo Stoakes /* Does the filesystem need to be notified? */ 150154cbbbf3SLorenzo Stoakes if (vm_ops_needs_writenotify(vma->vm_ops)) 150254cbbbf3SLorenzo Stoakes return true; 150354cbbbf3SLorenzo Stoakes 150454cbbbf3SLorenzo Stoakes /* 150554cbbbf3SLorenzo Stoakes * Even if the filesystem doesn't indicate a need for writenotify, if it 150654cbbbf3SLorenzo Stoakes * can writeback, dirty tracking is still required. 150754cbbbf3SLorenzo Stoakes */ 150854cbbbf3SLorenzo Stoakes return vma_fs_can_writeback(vma); 150954cbbbf3SLorenzo Stoakes } 151054cbbbf3SLorenzo Stoakes 15114e950f6fSAlexey Dobriyan /* 15128bb4e7a2SWei Yang * Some shared mappings will want the pages marked read-only 15134e950f6fSAlexey Dobriyan * to track write events. If so, we'll downgrade vm_page_prot 15144e950f6fSAlexey Dobriyan * to the private version (using protection_map[] without the 15154e950f6fSAlexey Dobriyan * VM_SHARED bit). 15164e950f6fSAlexey Dobriyan */ 15176d2329f8SAndrea Arcangeli int vma_wants_writenotify(struct vm_area_struct *vma, pgprot_t vm_page_prot) 15184e950f6fSAlexey Dobriyan { 15194e950f6fSAlexey Dobriyan /* If it was private or non-writable, the write bit is already clear */ 152054cbbbf3SLorenzo Stoakes if (!vma_is_shared_writable(vma)) 15214e950f6fSAlexey Dobriyan return 0; 15224e950f6fSAlexey Dobriyan 15234e950f6fSAlexey Dobriyan /* The backer wishes to know when pages are first written to? */ 152454cbbbf3SLorenzo Stoakes if (vm_ops_needs_writenotify(vma->vm_ops)) 15254e950f6fSAlexey Dobriyan return 1; 15264e950f6fSAlexey Dobriyan 152764e45507SPeter Feiner /* The open routine did something to the protections that pgprot_modify 152864e45507SPeter Feiner * won't preserve? */ 15296d2329f8SAndrea Arcangeli if (pgprot_val(vm_page_prot) != 153054cbbbf3SLorenzo Stoakes pgprot_val(vm_pgprot_modify(vm_page_prot, vma->vm_flags))) 15314e950f6fSAlexey Dobriyan return 0; 15324e950f6fSAlexey Dobriyan 1533f96f7a40SDavid Hildenbrand /* 1534f96f7a40SDavid Hildenbrand * Do we need to track softdirty? hugetlb does not support softdirty 1535f96f7a40SDavid Hildenbrand * tracking yet. 1536f96f7a40SDavid Hildenbrand */ 1537f96f7a40SDavid Hildenbrand if (vma_soft_dirty_enabled(vma) && !is_vm_hugetlb_page(vma)) 153864e45507SPeter Feiner return 1; 153964e45507SPeter Feiner 154051d3d5ebSDavid Hildenbrand /* Do we need write faults for uffd-wp tracking? */ 154151d3d5ebSDavid Hildenbrand if (userfaultfd_wp(vma)) 154251d3d5ebSDavid Hildenbrand return 1; 154351d3d5ebSDavid Hildenbrand 15444e950f6fSAlexey Dobriyan /* Can the mapping track the dirty pages? */ 154554cbbbf3SLorenzo Stoakes return vma_fs_can_writeback(vma); 15464e950f6fSAlexey Dobriyan } 15474e950f6fSAlexey Dobriyan 1548fc8744adSLinus Torvalds /* 1549fc8744adSLinus Torvalds * We account for memory if it's a private writeable mapping, 15505a6fe125SMel Gorman * not hugepages and VM_NORESERVE wasn't set. 1551fc8744adSLinus Torvalds */ 1552ca16d140SKOSAKI Motohiro static inline int accountable_mapping(struct file *file, vm_flags_t vm_flags) 1553fc8744adSLinus Torvalds { 15545a6fe125SMel Gorman /* 15555a6fe125SMel Gorman * hugetlb has its own accounting separate from the core VM 15565a6fe125SMel Gorman * VM_HUGETLB may not be set yet so we cannot check for that flag. 15575a6fe125SMel Gorman */ 15585a6fe125SMel Gorman if (file && is_file_hugepages(file)) 15595a6fe125SMel Gorman return 0; 15605a6fe125SMel Gorman 1561fc8744adSLinus Torvalds return (vm_flags & (VM_NORESERVE | VM_SHARED | VM_WRITE)) == VM_WRITE; 1562fc8744adSLinus Torvalds } 1563fc8744adSLinus Torvalds 15643499a131SLiam R. Howlett /** 15653499a131SLiam R. Howlett * unmapped_area() - Find an area between the low_limit and the high_limit with 15663499a131SLiam R. Howlett * the correct alignment and offset, all from @info. Note: current->mm is used 15673499a131SLiam R. Howlett * for the search. 15683499a131SLiam R. Howlett * 156982b24936SVernon Yang * @info: The unmapped area information including the range [low_limit - 157082b24936SVernon Yang * high_limit), the alignment offset and mask. 15713499a131SLiam R. Howlett * 15723499a131SLiam R. Howlett * Return: A memory address or -ENOMEM. 15733499a131SLiam R. Howlett */ 1574baceaf1cSJaewon Kim static unsigned long unmapped_area(struct vm_unmapped_area_info *info) 1575db4fbfb9SMichel Lespinasse { 15766b008640SLinus Torvalds unsigned long length, gap; 15776b008640SLinus Torvalds unsigned long low_limit, high_limit; 157858c5d0d6SLiam R. Howlett struct vm_area_struct *tmp; 1579db4fbfb9SMichel Lespinasse 15803499a131SLiam R. Howlett MA_STATE(mas, ¤t->mm->mm_mt, 0, 0); 1581db4fbfb9SMichel Lespinasse 1582db4fbfb9SMichel Lespinasse /* Adjust search length to account for worst case alignment overhead */ 1583db4fbfb9SMichel Lespinasse length = info->length + info->align_mask; 1584db4fbfb9SMichel Lespinasse if (length < info->length) 1585db4fbfb9SMichel Lespinasse return -ENOMEM; 1586db4fbfb9SMichel Lespinasse 158758c5d0d6SLiam R. Howlett low_limit = info->low_limit; 15886b008640SLinus Torvalds if (low_limit < mmap_min_addr) 15896b008640SLinus Torvalds low_limit = mmap_min_addr; 15906b008640SLinus Torvalds high_limit = info->high_limit; 159158c5d0d6SLiam R. Howlett retry: 15926b008640SLinus Torvalds if (mas_empty_area(&mas, low_limit, high_limit - 1, length)) 15933499a131SLiam R. Howlett return -ENOMEM; 15943499a131SLiam R. Howlett 1595d4af56c5SLiam R. Howlett gap = mas.index; 1596d4af56c5SLiam R. Howlett gap += (info->align_offset - gap) & info->align_mask; 159758c5d0d6SLiam R. Howlett tmp = mas_next(&mas, ULONG_MAX); 15980266e7c5SRick Edgecombe if (tmp && (tmp->vm_flags & VM_STARTGAP_FLAGS)) { /* Avoid prev check if possible */ 159958c5d0d6SLiam R. Howlett if (vm_start_gap(tmp) < gap + length - 1) { 160058c5d0d6SLiam R. Howlett low_limit = tmp->vm_end; 160158c5d0d6SLiam R. Howlett mas_reset(&mas); 160258c5d0d6SLiam R. Howlett goto retry; 160358c5d0d6SLiam R. Howlett } 160458c5d0d6SLiam R. Howlett } else { 160558c5d0d6SLiam R. Howlett tmp = mas_prev(&mas, 0); 160658c5d0d6SLiam R. Howlett if (tmp && vm_end_gap(tmp) > gap) { 160758c5d0d6SLiam R. Howlett low_limit = vm_end_gap(tmp); 160858c5d0d6SLiam R. Howlett mas_reset(&mas); 160958c5d0d6SLiam R. Howlett goto retry; 161058c5d0d6SLiam R. Howlett } 161158c5d0d6SLiam R. Howlett } 161258c5d0d6SLiam R. Howlett 16133499a131SLiam R. Howlett return gap; 1614db4fbfb9SMichel Lespinasse } 1615db4fbfb9SMichel Lespinasse 16163499a131SLiam R. Howlett /** 16173499a131SLiam R. Howlett * unmapped_area_topdown() - Find an area between the low_limit and the 161882b24936SVernon Yang * high_limit with the correct alignment and offset at the highest available 16193499a131SLiam R. Howlett * address, all from @info. Note: current->mm is used for the search. 16203499a131SLiam R. Howlett * 162182b24936SVernon Yang * @info: The unmapped area information including the range [low_limit - 162282b24936SVernon Yang * high_limit), the alignment offset and mask. 16233499a131SLiam R. Howlett * 16243499a131SLiam R. Howlett * Return: A memory address or -ENOMEM. 16253499a131SLiam R. Howlett */ 1626baceaf1cSJaewon Kim static unsigned long unmapped_area_topdown(struct vm_unmapped_area_info *info) 1627db4fbfb9SMichel Lespinasse { 16286b008640SLinus Torvalds unsigned long length, gap, gap_end; 16296b008640SLinus Torvalds unsigned long low_limit, high_limit; 163058c5d0d6SLiam R. Howlett struct vm_area_struct *tmp; 1631d4af56c5SLiam R. Howlett 16323499a131SLiam R. Howlett MA_STATE(mas, ¤t->mm->mm_mt, 0, 0); 1633db4fbfb9SMichel Lespinasse /* Adjust search length to account for worst case alignment overhead */ 1634db4fbfb9SMichel Lespinasse length = info->length + info->align_mask; 1635db4fbfb9SMichel Lespinasse if (length < info->length) 1636db4fbfb9SMichel Lespinasse return -ENOMEM; 1637db4fbfb9SMichel Lespinasse 16386b008640SLinus Torvalds low_limit = info->low_limit; 16396b008640SLinus Torvalds if (low_limit < mmap_min_addr) 16406b008640SLinus Torvalds low_limit = mmap_min_addr; 164158c5d0d6SLiam R. Howlett high_limit = info->high_limit; 164258c5d0d6SLiam R. Howlett retry: 16436b008640SLinus Torvalds if (mas_empty_area_rev(&mas, low_limit, high_limit - 1, length)) 16443499a131SLiam R. Howlett return -ENOMEM; 16453499a131SLiam R. Howlett 1646d4af56c5SLiam R. Howlett gap = mas.last + 1 - info->length; 1647d4af56c5SLiam R. Howlett gap -= (gap - info->align_offset) & info->align_mask; 164858c5d0d6SLiam R. Howlett gap_end = mas.last; 164958c5d0d6SLiam R. Howlett tmp = mas_next(&mas, ULONG_MAX); 16500266e7c5SRick Edgecombe if (tmp && (tmp->vm_flags & VM_STARTGAP_FLAGS)) { /* Avoid prev check if possible */ 165158c5d0d6SLiam R. Howlett if (vm_start_gap(tmp) <= gap_end) { 165258c5d0d6SLiam R. Howlett high_limit = vm_start_gap(tmp); 165358c5d0d6SLiam R. Howlett mas_reset(&mas); 165458c5d0d6SLiam R. Howlett goto retry; 165558c5d0d6SLiam R. Howlett } 165658c5d0d6SLiam R. Howlett } else { 165758c5d0d6SLiam R. Howlett tmp = mas_prev(&mas, 0); 165858c5d0d6SLiam R. Howlett if (tmp && vm_end_gap(tmp) > gap) { 165958c5d0d6SLiam R. Howlett high_limit = tmp->vm_start; 166058c5d0d6SLiam R. Howlett mas_reset(&mas); 166158c5d0d6SLiam R. Howlett goto retry; 166258c5d0d6SLiam R. Howlett } 166358c5d0d6SLiam R. Howlett } 166458c5d0d6SLiam R. Howlett 16653499a131SLiam R. Howlett return gap; 1666db4fbfb9SMichel Lespinasse } 1667db4fbfb9SMichel Lespinasse 1668baceaf1cSJaewon Kim /* 1669baceaf1cSJaewon Kim * Search for an unmapped address range. 1670baceaf1cSJaewon Kim * 1671baceaf1cSJaewon Kim * We are looking for a range that: 1672baceaf1cSJaewon Kim * - does not intersect with any VMA; 1673baceaf1cSJaewon Kim * - is contained within the [low_limit, high_limit) interval; 1674baceaf1cSJaewon Kim * - is at least the desired size. 1675baceaf1cSJaewon Kim * - satisfies (begin_addr & align_mask) == (align_offset & align_mask) 1676baceaf1cSJaewon Kim */ 1677baceaf1cSJaewon Kim unsigned long vm_unmapped_area(struct vm_unmapped_area_info *info) 1678baceaf1cSJaewon Kim { 1679df529cabSJaewon Kim unsigned long addr; 1680df529cabSJaewon Kim 1681baceaf1cSJaewon Kim if (info->flags & VM_UNMAPPED_AREA_TOPDOWN) 1682df529cabSJaewon Kim addr = unmapped_area_topdown(info); 1683baceaf1cSJaewon Kim else 1684df529cabSJaewon Kim addr = unmapped_area(info); 1685df529cabSJaewon Kim 1686df529cabSJaewon Kim trace_vm_unmapped_area(addr, info); 1687df529cabSJaewon Kim return addr; 1688baceaf1cSJaewon Kim } 1689f6795053SSteve Capper 16901da177e4SLinus Torvalds /* Get an address range which is currently unmapped. 16911da177e4SLinus Torvalds * For shmat() with addr=0. 16921da177e4SLinus Torvalds * 16931da177e4SLinus Torvalds * Ugly calling convention alert: 16941da177e4SLinus Torvalds * Return value with the low bits set means error value, 16951da177e4SLinus Torvalds * ie 16961da177e4SLinus Torvalds * if (ret & ~PAGE_MASK) 16971da177e4SLinus Torvalds * error = ret; 16981da177e4SLinus Torvalds * 16991da177e4SLinus Torvalds * This function "knows" that -ENOMEM has the bits set. 17001da177e4SLinus Torvalds */ 17011da177e4SLinus Torvalds unsigned long 17024b439e25SChristophe Leroy generic_get_unmapped_area(struct file *filp, unsigned long addr, 17034b439e25SChristophe Leroy unsigned long len, unsigned long pgoff, 17044b439e25SChristophe Leroy unsigned long flags) 17051da177e4SLinus Torvalds { 17061da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 17071be7107fSHugh Dickins struct vm_area_struct *vma, *prev; 1708db4fbfb9SMichel Lespinasse struct vm_unmapped_area_info info; 17092cb4de08SChristophe Leroy const unsigned long mmap_end = arch_get_mmap_end(addr, len, flags); 17101da177e4SLinus Torvalds 1711f6795053SSteve Capper if (len > mmap_end - mmap_min_addr) 17121da177e4SLinus Torvalds return -ENOMEM; 17131da177e4SLinus Torvalds 171406abdfb4SBenjamin Herrenschmidt if (flags & MAP_FIXED) 171506abdfb4SBenjamin Herrenschmidt return addr; 171606abdfb4SBenjamin Herrenschmidt 17171da177e4SLinus Torvalds if (addr) { 17181da177e4SLinus Torvalds addr = PAGE_ALIGN(addr); 17191be7107fSHugh Dickins vma = find_vma_prev(mm, addr, &prev); 1720f6795053SSteve Capper if (mmap_end - len >= addr && addr >= mmap_min_addr && 17211be7107fSHugh Dickins (!vma || addr + len <= vm_start_gap(vma)) && 17221be7107fSHugh Dickins (!prev || addr >= vm_end_gap(prev))) 17231da177e4SLinus Torvalds return addr; 17241da177e4SLinus Torvalds } 17251da177e4SLinus Torvalds 1726db4fbfb9SMichel Lespinasse info.flags = 0; 1727db4fbfb9SMichel Lespinasse info.length = len; 17284e99b021SHeiko Carstens info.low_limit = mm->mmap_base; 1729f6795053SSteve Capper info.high_limit = mmap_end; 1730db4fbfb9SMichel Lespinasse info.align_mask = 0; 173109ef5283SJaewon Kim info.align_offset = 0; 1732db4fbfb9SMichel Lespinasse return vm_unmapped_area(&info); 17331da177e4SLinus Torvalds } 17344b439e25SChristophe Leroy 17354b439e25SChristophe Leroy #ifndef HAVE_ARCH_UNMAPPED_AREA 17364b439e25SChristophe Leroy unsigned long 17374b439e25SChristophe Leroy arch_get_unmapped_area(struct file *filp, unsigned long addr, 17384b439e25SChristophe Leroy unsigned long len, unsigned long pgoff, 17394b439e25SChristophe Leroy unsigned long flags) 17404b439e25SChristophe Leroy { 17414b439e25SChristophe Leroy return generic_get_unmapped_area(filp, addr, len, pgoff, flags); 17424b439e25SChristophe Leroy } 17431da177e4SLinus Torvalds #endif 17441da177e4SLinus Torvalds 17451da177e4SLinus Torvalds /* 17461da177e4SLinus Torvalds * This mmap-allocator allocates new areas top-down from below the 17471da177e4SLinus Torvalds * stack's low limit (the base): 17481da177e4SLinus Torvalds */ 17491da177e4SLinus Torvalds unsigned long 17504b439e25SChristophe Leroy generic_get_unmapped_area_topdown(struct file *filp, unsigned long addr, 175143cca0b1SYang Fan unsigned long len, unsigned long pgoff, 175243cca0b1SYang Fan unsigned long flags) 17531da177e4SLinus Torvalds { 17541be7107fSHugh Dickins struct vm_area_struct *vma, *prev; 17551da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 1756db4fbfb9SMichel Lespinasse struct vm_unmapped_area_info info; 17572cb4de08SChristophe Leroy const unsigned long mmap_end = arch_get_mmap_end(addr, len, flags); 17581da177e4SLinus Torvalds 17591da177e4SLinus Torvalds /* requested length too big for entire address space */ 1760f6795053SSteve Capper if (len > mmap_end - mmap_min_addr) 17611da177e4SLinus Torvalds return -ENOMEM; 17621da177e4SLinus Torvalds 176306abdfb4SBenjamin Herrenschmidt if (flags & MAP_FIXED) 176406abdfb4SBenjamin Herrenschmidt return addr; 176506abdfb4SBenjamin Herrenschmidt 17661da177e4SLinus Torvalds /* requesting a specific address */ 17671da177e4SLinus Torvalds if (addr) { 17681da177e4SLinus Torvalds addr = PAGE_ALIGN(addr); 17691be7107fSHugh Dickins vma = find_vma_prev(mm, addr, &prev); 1770f6795053SSteve Capper if (mmap_end - len >= addr && addr >= mmap_min_addr && 17711be7107fSHugh Dickins (!vma || addr + len <= vm_start_gap(vma)) && 17721be7107fSHugh Dickins (!prev || addr >= vm_end_gap(prev))) 17731da177e4SLinus Torvalds return addr; 17741da177e4SLinus Torvalds } 17751da177e4SLinus Torvalds 1776db4fbfb9SMichel Lespinasse info.flags = VM_UNMAPPED_AREA_TOPDOWN; 1777db4fbfb9SMichel Lespinasse info.length = len; 17786b008640SLinus Torvalds info.low_limit = PAGE_SIZE; 1779f6795053SSteve Capper info.high_limit = arch_get_mmap_base(addr, mm->mmap_base); 1780db4fbfb9SMichel Lespinasse info.align_mask = 0; 178109ef5283SJaewon Kim info.align_offset = 0; 1782db4fbfb9SMichel Lespinasse addr = vm_unmapped_area(&info); 1783b716ad95SXiao Guangrong 17841da177e4SLinus Torvalds /* 17851da177e4SLinus Torvalds * A failed mmap() very likely causes application failure, 17861da177e4SLinus Torvalds * so fall back to the bottom-up function here. This scenario 17871da177e4SLinus Torvalds * can happen with large stack limits and large mmap() 17881da177e4SLinus Torvalds * allocations. 17891da177e4SLinus Torvalds */ 1790de1741a1SAlexander Kuleshov if (offset_in_page(addr)) { 1791db4fbfb9SMichel Lespinasse VM_BUG_ON(addr != -ENOMEM); 1792db4fbfb9SMichel Lespinasse info.flags = 0; 1793db4fbfb9SMichel Lespinasse info.low_limit = TASK_UNMAPPED_BASE; 1794f6795053SSteve Capper info.high_limit = mmap_end; 1795db4fbfb9SMichel Lespinasse addr = vm_unmapped_area(&info); 1796db4fbfb9SMichel Lespinasse } 17971da177e4SLinus Torvalds 17981da177e4SLinus Torvalds return addr; 17991da177e4SLinus Torvalds } 18004b439e25SChristophe Leroy 18014b439e25SChristophe Leroy #ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN 18024b439e25SChristophe Leroy unsigned long 18034b439e25SChristophe Leroy arch_get_unmapped_area_topdown(struct file *filp, unsigned long addr, 18044b439e25SChristophe Leroy unsigned long len, unsigned long pgoff, 18054b439e25SChristophe Leroy unsigned long flags) 18064b439e25SChristophe Leroy { 18074b439e25SChristophe Leroy return generic_get_unmapped_area_topdown(filp, addr, len, pgoff, flags); 18084b439e25SChristophe Leroy } 18091da177e4SLinus Torvalds #endif 18101da177e4SLinus Torvalds 18111da177e4SLinus Torvalds unsigned long 18121da177e4SLinus Torvalds get_unmapped_area(struct file *file, unsigned long addr, unsigned long len, 18131da177e4SLinus Torvalds unsigned long pgoff, unsigned long flags) 18141da177e4SLinus Torvalds { 181506abdfb4SBenjamin Herrenschmidt unsigned long (*get_area)(struct file *, unsigned long, 181606abdfb4SBenjamin Herrenschmidt unsigned long, unsigned long, unsigned long); 181707ab67c8SLinus Torvalds 18189206de95SAl Viro unsigned long error = arch_mmap_check(addr, len, flags); 18199206de95SAl Viro if (error) 18209206de95SAl Viro return error; 18219206de95SAl Viro 18229206de95SAl Viro /* Careful about overflows.. */ 18239206de95SAl Viro if (len > TASK_SIZE) 18249206de95SAl Viro return -ENOMEM; 18259206de95SAl Viro 182607ab67c8SLinus Torvalds get_area = current->mm->get_unmapped_area; 1827c01d5b30SHugh Dickins if (file) { 1828c01d5b30SHugh Dickins if (file->f_op->get_unmapped_area) 182907ab67c8SLinus Torvalds get_area = file->f_op->get_unmapped_area; 1830c01d5b30SHugh Dickins } else if (flags & MAP_SHARED) { 1831c01d5b30SHugh Dickins /* 1832c01d5b30SHugh Dickins * mmap_region() will call shmem_zero_setup() to create a file, 1833c01d5b30SHugh Dickins * so use shmem's get_unmapped_area in case it can be huge. 1834c01d5b30SHugh Dickins */ 1835c01d5b30SHugh Dickins get_area = shmem_get_unmapped_area; 1836efa7df3eSRik van Riel } else if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE)) { 1837efa7df3eSRik van Riel /* Ensures that larger anonymous mappings are THP aligned. */ 1838efa7df3eSRik van Riel get_area = thp_get_unmapped_area; 1839c01d5b30SHugh Dickins } 1840c01d5b30SHugh Dickins 184196204e15SRyan Roberts /* Always treat pgoff as zero for anonymous memory. */ 184296204e15SRyan Roberts if (!file) 184396204e15SRyan Roberts pgoff = 0; 184496204e15SRyan Roberts 184507ab67c8SLinus Torvalds addr = get_area(file, addr, len, pgoff, flags); 184607ab67c8SLinus Torvalds if (IS_ERR_VALUE(addr)) 184707ab67c8SLinus Torvalds return addr; 184807ab67c8SLinus Torvalds 18491da177e4SLinus Torvalds if (addr > TASK_SIZE - len) 18501da177e4SLinus Torvalds return -ENOMEM; 1851de1741a1SAlexander Kuleshov if (offset_in_page(addr)) 18521da177e4SLinus Torvalds return -EINVAL; 185306abdfb4SBenjamin Herrenschmidt 18549ac4ed4bSAl Viro error = security_mmap_addr(addr); 18559ac4ed4bSAl Viro return error ? error : addr; 18561da177e4SLinus Torvalds } 18571da177e4SLinus Torvalds 18581da177e4SLinus Torvalds EXPORT_SYMBOL(get_unmapped_area); 18591da177e4SLinus Torvalds 1860be8432e7SLiam R. Howlett /** 1861abdba2ddSLiam R. Howlett * find_vma_intersection() - Look up the first VMA which intersects the interval 1862abdba2ddSLiam R. Howlett * @mm: The process address space. 1863abdba2ddSLiam R. Howlett * @start_addr: The inclusive start user address. 1864abdba2ddSLiam R. Howlett * @end_addr: The exclusive end user address. 1865abdba2ddSLiam R. Howlett * 1866abdba2ddSLiam R. Howlett * Returns: The first VMA within the provided range, %NULL otherwise. Assumes 1867abdba2ddSLiam R. Howlett * start_addr < end_addr. 1868abdba2ddSLiam R. Howlett */ 1869abdba2ddSLiam R. Howlett struct vm_area_struct *find_vma_intersection(struct mm_struct *mm, 1870abdba2ddSLiam R. Howlett unsigned long start_addr, 1871abdba2ddSLiam R. Howlett unsigned long end_addr) 1872abdba2ddSLiam R. Howlett { 1873abdba2ddSLiam R. Howlett unsigned long index = start_addr; 1874abdba2ddSLiam R. Howlett 1875abdba2ddSLiam R. Howlett mmap_assert_locked(mm); 18767964cf8cSLiam R. Howlett return mt_find(&mm->mm_mt, &index, end_addr - 1); 1877abdba2ddSLiam R. Howlett } 1878abdba2ddSLiam R. Howlett EXPORT_SYMBOL(find_vma_intersection); 1879abdba2ddSLiam R. Howlett 1880abdba2ddSLiam R. Howlett /** 1881be8432e7SLiam R. Howlett * find_vma() - Find the VMA for a given address, or the next VMA. 1882be8432e7SLiam R. Howlett * @mm: The mm_struct to check 1883be8432e7SLiam R. Howlett * @addr: The address 1884be8432e7SLiam R. Howlett * 1885be8432e7SLiam R. Howlett * Returns: The VMA associated with addr, or the next VMA. 1886be8432e7SLiam R. Howlett * May return %NULL in the case of no VMA at addr or above. 1887be8432e7SLiam R. Howlett */ 18881da177e4SLinus Torvalds struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr) 18891da177e4SLinus Torvalds { 1890be8432e7SLiam R. Howlett unsigned long index = addr; 18911da177e4SLinus Torvalds 18925b78ed24SLuigi Rizzo mmap_assert_locked(mm); 18937964cf8cSLiam R. Howlett return mt_find(&mm->mm_mt, &index, ULONG_MAX); 18941da177e4SLinus Torvalds } 18951da177e4SLinus Torvalds EXPORT_SYMBOL(find_vma); 18961da177e4SLinus Torvalds 18977fdbd37dSLiam R. Howlett /** 18987fdbd37dSLiam R. Howlett * find_vma_prev() - Find the VMA for a given address, or the next vma and 18997fdbd37dSLiam R. Howlett * set %pprev to the previous VMA, if any. 19007fdbd37dSLiam R. Howlett * @mm: The mm_struct to check 19017fdbd37dSLiam R. Howlett * @addr: The address 19027fdbd37dSLiam R. Howlett * @pprev: The pointer to set to the previous VMA 19037fdbd37dSLiam R. Howlett * 19047fdbd37dSLiam R. Howlett * Note that RCU lock is missing here since the external mmap_lock() is used 19057fdbd37dSLiam R. Howlett * instead. 19067fdbd37dSLiam R. Howlett * 19077fdbd37dSLiam R. Howlett * Returns: The VMA associated with @addr, or the next vma. 19087fdbd37dSLiam R. Howlett * May return %NULL in the case of no vma at addr or above. 19096bd4837dSKOSAKI Motohiro */ 19101da177e4SLinus Torvalds struct vm_area_struct * 19111da177e4SLinus Torvalds find_vma_prev(struct mm_struct *mm, unsigned long addr, 19121da177e4SLinus Torvalds struct vm_area_struct **pprev) 19131da177e4SLinus Torvalds { 19146bd4837dSKOSAKI Motohiro struct vm_area_struct *vma; 19157fdbd37dSLiam R. Howlett MA_STATE(mas, &mm->mm_mt, addr, addr); 19161da177e4SLinus Torvalds 19177fdbd37dSLiam R. Howlett vma = mas_walk(&mas); 19187fdbd37dSLiam R. Howlett *pprev = mas_prev(&mas, 0); 19197fdbd37dSLiam R. Howlett if (!vma) 19207fdbd37dSLiam R. Howlett vma = mas_next(&mas, ULONG_MAX); 19216bd4837dSKOSAKI Motohiro return vma; 19221da177e4SLinus Torvalds } 19231da177e4SLinus Torvalds 19241da177e4SLinus Torvalds /* 19251da177e4SLinus Torvalds * Verify that the stack growth is acceptable and 19261da177e4SLinus Torvalds * update accounting. This is shared with both the 19271da177e4SLinus Torvalds * grow-up and grow-down cases. 19281da177e4SLinus Torvalds */ 19291be7107fSHugh Dickins static int acct_stack_growth(struct vm_area_struct *vma, 19301be7107fSHugh Dickins unsigned long size, unsigned long grow) 19311da177e4SLinus Torvalds { 19321da177e4SLinus Torvalds struct mm_struct *mm = vma->vm_mm; 19331be7107fSHugh Dickins unsigned long new_start; 19341da177e4SLinus Torvalds 19351da177e4SLinus Torvalds /* address space limit tests */ 193684638335SKonstantin Khlebnikov if (!may_expand_vm(mm, vma->vm_flags, grow)) 19371da177e4SLinus Torvalds return -ENOMEM; 19381da177e4SLinus Torvalds 19391da177e4SLinus Torvalds /* Stack limit test */ 194024c79d8eSKrzysztof Opasiak if (size > rlimit(RLIMIT_STACK)) 19411da177e4SLinus Torvalds return -ENOMEM; 19421da177e4SLinus Torvalds 19431da177e4SLinus Torvalds /* mlock limit tests */ 1944b0cc5e89SAndrew Morton if (!mlock_future_ok(mm, vma->vm_flags, grow << PAGE_SHIFT)) 19451da177e4SLinus Torvalds return -ENOMEM; 19461da177e4SLinus Torvalds 19470d59a01bSAdam Litke /* Check to ensure the stack will not grow into a hugetlb-only region */ 19480d59a01bSAdam Litke new_start = (vma->vm_flags & VM_GROWSUP) ? vma->vm_start : 19490d59a01bSAdam Litke vma->vm_end - size; 19500d59a01bSAdam Litke if (is_hugepage_only_range(vma->vm_mm, new_start, size)) 19510d59a01bSAdam Litke return -EFAULT; 19520d59a01bSAdam Litke 19531da177e4SLinus Torvalds /* 19541da177e4SLinus Torvalds * Overcommit.. This must be the final test, as it will 19551da177e4SLinus Torvalds * update security statistics. 19561da177e4SLinus Torvalds */ 195705fa199dSHugh Dickins if (security_vm_enough_memory_mm(mm, grow)) 19581da177e4SLinus Torvalds return -ENOMEM; 19591da177e4SLinus Torvalds 19601da177e4SLinus Torvalds return 0; 19611da177e4SLinus Torvalds } 19621da177e4SLinus Torvalds 1963cf8e8658SArd Biesheuvel #if defined(CONFIG_STACK_GROWSUP) 19641da177e4SLinus Torvalds /* 1965cf8e8658SArd Biesheuvel * PA-RISC uses this for its stack. 196646dea3d0SHugh Dickins * vma is the last one with address > vma->vm_end. Have to extend vma. 19671da177e4SLinus Torvalds */ 19688d7071afSLinus Torvalds static int expand_upwards(struct vm_area_struct *vma, unsigned long address) 19691da177e4SLinus Torvalds { 197009357814SOleg Nesterov struct mm_struct *mm = vma->vm_mm; 19711be7107fSHugh Dickins struct vm_area_struct *next; 19721be7107fSHugh Dickins unsigned long gap_addr; 197312352d3cSKonstantin Khlebnikov int error = 0; 1974b5df0922SLiam R. Howlett MA_STATE(mas, &mm->mm_mt, vma->vm_start, address); 19751da177e4SLinus Torvalds 19761da177e4SLinus Torvalds if (!(vma->vm_flags & VM_GROWSUP)) 19771da177e4SLinus Torvalds return -EFAULT; 19781da177e4SLinus Torvalds 1979bd726c90SHelge Deller /* Guard against exceeding limits of the address space. */ 19801be7107fSHugh Dickins address &= PAGE_MASK; 198137511fb5SHelge Deller if (address >= (TASK_SIZE & PAGE_MASK)) 198212352d3cSKonstantin Khlebnikov return -ENOMEM; 1983bd726c90SHelge Deller address += PAGE_SIZE; 198412352d3cSKonstantin Khlebnikov 19851be7107fSHugh Dickins /* Enforce stack_guard_gap */ 19861be7107fSHugh Dickins gap_addr = address + stack_guard_gap; 1987bd726c90SHelge Deller 1988bd726c90SHelge Deller /* Guard against overflow */ 1989bd726c90SHelge Deller if (gap_addr < address || gap_addr > TASK_SIZE) 1990bd726c90SHelge Deller gap_addr = TASK_SIZE; 1991bd726c90SHelge Deller 1992763ecb03SLiam R. Howlett next = find_vma_intersection(mm, vma->vm_end, gap_addr); 1993763ecb03SLiam R. Howlett if (next && vma_is_accessible(next)) { 19941be7107fSHugh Dickins if (!(next->vm_flags & VM_GROWSUP)) 19951be7107fSHugh Dickins return -ENOMEM; 19961be7107fSHugh Dickins /* Check that both stack segments have the same anon_vma? */ 19971be7107fSHugh Dickins } 19981be7107fSHugh Dickins 1999b5df0922SLiam R. Howlett if (next) 2000b5df0922SLiam R. Howlett mas_prev_range(&mas, address); 2001b5df0922SLiam R. Howlett 2002b5df0922SLiam R. Howlett __mas_set_range(&mas, vma->vm_start, address - 1); 2003da089254SLiam R. Howlett if (mas_preallocate(&mas, vma, GFP_KERNEL)) 20041da177e4SLinus Torvalds return -ENOMEM; 20051da177e4SLinus Torvalds 2006d4af56c5SLiam R. Howlett /* We must make sure the anon_vma is allocated. */ 2007d4af56c5SLiam R. Howlett if (unlikely(anon_vma_prepare(vma))) { 2008d4af56c5SLiam R. Howlett mas_destroy(&mas); 2009d4af56c5SLiam R. Howlett return -ENOMEM; 2010d4af56c5SLiam R. Howlett } 2011d4af56c5SLiam R. Howlett 2012c137381fSSuren Baghdasaryan /* Lock the VMA before expanding to prevent concurrent page faults */ 2013c137381fSSuren Baghdasaryan vma_start_write(vma); 20141da177e4SLinus Torvalds /* 20151da177e4SLinus Torvalds * vma->vm_start/vm_end cannot change under us because the caller 2016c1e8d7c6SMichel Lespinasse * is required to hold the mmap_lock in read mode. We need the 20171da177e4SLinus Torvalds * anon_vma lock to serialize against concurrent expand_stacks. 20181da177e4SLinus Torvalds */ 201912352d3cSKonstantin Khlebnikov anon_vma_lock_write(vma->anon_vma); 20201da177e4SLinus Torvalds 20211da177e4SLinus Torvalds /* Somebody else might have raced and expanded it already */ 20221da177e4SLinus Torvalds if (address > vma->vm_end) { 20231da177e4SLinus Torvalds unsigned long size, grow; 20241da177e4SLinus Torvalds 20251da177e4SLinus Torvalds size = address - vma->vm_start; 20261da177e4SLinus Torvalds grow = (address - vma->vm_end) >> PAGE_SHIFT; 20271da177e4SLinus Torvalds 202842c36f63SHugh Dickins error = -ENOMEM; 202942c36f63SHugh Dickins if (vma->vm_pgoff + (size >> PAGE_SHIFT) >= vma->vm_pgoff) { 20301da177e4SLinus Torvalds error = acct_stack_growth(vma, size, grow); 20313af9e859SEric B Munson if (!error) { 20324128997bSMichel Lespinasse /* 2033524e00b3SLiam R. Howlett * We only hold a shared mmap_lock lock here, so 2034524e00b3SLiam R. Howlett * we need to protect against concurrent vma 2035524e00b3SLiam R. Howlett * expansions. anon_vma_lock_write() doesn't 2036524e00b3SLiam R. Howlett * help here, as we don't guarantee that all 2037524e00b3SLiam R. Howlett * growable vmas in a mm share the same root 2038524e00b3SLiam R. Howlett * anon vma. So, we reuse mm->page_table_lock 2039524e00b3SLiam R. Howlett * to guard against concurrent vma expansions. 20404128997bSMichel Lespinasse */ 204109357814SOleg Nesterov spin_lock(&mm->page_table_lock); 204287e8827bSOleg Nesterov if (vma->vm_flags & VM_LOCKED) 204309357814SOleg Nesterov mm->locked_vm += grow; 204484638335SKonstantin Khlebnikov vm_stat_account(mm, vma->vm_flags, grow); 2045bf181b9fSMichel Lespinasse anon_vma_interval_tree_pre_update_vma(vma); 20461da177e4SLinus Torvalds vma->vm_end = address; 2047d4af56c5SLiam R. Howlett /* Overwrite old entry in mtree. */ 2048fbcc3104SLiam R. Howlett mas_store_prealloc(&mas, vma); 2049bf181b9fSMichel Lespinasse anon_vma_interval_tree_post_update_vma(vma); 205009357814SOleg Nesterov spin_unlock(&mm->page_table_lock); 20514128997bSMichel Lespinasse 20523af9e859SEric B Munson perf_event_mmap(vma); 20533af9e859SEric B Munson } 20541da177e4SLinus Torvalds } 205542c36f63SHugh Dickins } 205612352d3cSKonstantin Khlebnikov anon_vma_unlock_write(vma->anon_vma); 2057d4af56c5SLiam R. Howlett mas_destroy(&mas); 20582574d5e4SLiam R. Howlett validate_mm(mm); 20591da177e4SLinus Torvalds return error; 20601da177e4SLinus Torvalds } 2061cf8e8658SArd Biesheuvel #endif /* CONFIG_STACK_GROWSUP */ 206246dea3d0SHugh Dickins 20631da177e4SLinus Torvalds /* 20641da177e4SLinus Torvalds * vma is the first one with address < vma->vm_start. Have to extend vma. 20658d7071afSLinus Torvalds * mmap_lock held for writing. 20661da177e4SLinus Torvalds */ 2067524e00b3SLiam R. Howlett int expand_downwards(struct vm_area_struct *vma, unsigned long address) 20681da177e4SLinus Torvalds { 206909357814SOleg Nesterov struct mm_struct *mm = vma->vm_mm; 2070763ecb03SLiam R. Howlett MA_STATE(mas, &mm->mm_mt, vma->vm_start, vma->vm_start); 20711be7107fSHugh Dickins struct vm_area_struct *prev; 20720a1d5299SJann Horn int error = 0; 20731da177e4SLinus Torvalds 20748d7071afSLinus Torvalds if (!(vma->vm_flags & VM_GROWSDOWN)) 20758d7071afSLinus Torvalds return -EFAULT; 20768d7071afSLinus Torvalds 20778869477aSEric Paris address &= PAGE_MASK; 20788b35ca3eSBen Hutchings if (address < mmap_min_addr || address < FIRST_USER_ADDRESS) 20790a1d5299SJann Horn return -EPERM; 20808869477aSEric Paris 20811be7107fSHugh Dickins /* Enforce stack_guard_gap */ 2082763ecb03SLiam R. Howlett prev = mas_prev(&mas, 0); 20831be7107fSHugh Dickins /* Check that both stack segments have the same anon_vma? */ 2084f440fa1aSLiam R. Howlett if (prev) { 2085f440fa1aSLiam R. Howlett if (!(prev->vm_flags & VM_GROWSDOWN) && 2086f440fa1aSLiam R. Howlett vma_is_accessible(prev) && 2087f440fa1aSLiam R. Howlett (address - prev->vm_end < stack_guard_gap)) 208832e4e6d5SOleg Nesterov return -ENOMEM; 20891be7107fSHugh Dickins } 20901be7107fSHugh Dickins 2091b5df0922SLiam R. Howlett if (prev) 2092b5df0922SLiam R. Howlett mas_next_range(&mas, vma->vm_start); 2093b5df0922SLiam R. Howlett 2094b5df0922SLiam R. Howlett __mas_set_range(&mas, address, vma->vm_end - 1); 2095da089254SLiam R. Howlett if (mas_preallocate(&mas, vma, GFP_KERNEL)) 209612352d3cSKonstantin Khlebnikov return -ENOMEM; 20971da177e4SLinus Torvalds 2098d4af56c5SLiam R. Howlett /* We must make sure the anon_vma is allocated. */ 2099d4af56c5SLiam R. Howlett if (unlikely(anon_vma_prepare(vma))) { 2100d4af56c5SLiam R. Howlett mas_destroy(&mas); 2101d4af56c5SLiam R. Howlett return -ENOMEM; 2102d4af56c5SLiam R. Howlett } 2103d4af56c5SLiam R. Howlett 2104c137381fSSuren Baghdasaryan /* Lock the VMA before expanding to prevent concurrent page faults */ 2105c137381fSSuren Baghdasaryan vma_start_write(vma); 21061da177e4SLinus Torvalds /* 21071da177e4SLinus Torvalds * vma->vm_start/vm_end cannot change under us because the caller 2108c1e8d7c6SMichel Lespinasse * is required to hold the mmap_lock in read mode. We need the 21091da177e4SLinus Torvalds * anon_vma lock to serialize against concurrent expand_stacks. 21101da177e4SLinus Torvalds */ 211112352d3cSKonstantin Khlebnikov anon_vma_lock_write(vma->anon_vma); 21121da177e4SLinus Torvalds 21131da177e4SLinus Torvalds /* Somebody else might have raced and expanded it already */ 21141da177e4SLinus Torvalds if (address < vma->vm_start) { 21151da177e4SLinus Torvalds unsigned long size, grow; 21161da177e4SLinus Torvalds 21171da177e4SLinus Torvalds size = vma->vm_end - address; 21181da177e4SLinus Torvalds grow = (vma->vm_start - address) >> PAGE_SHIFT; 21191da177e4SLinus Torvalds 2120a626ca6aSLinus Torvalds error = -ENOMEM; 2121a626ca6aSLinus Torvalds if (grow <= vma->vm_pgoff) { 21221da177e4SLinus Torvalds error = acct_stack_growth(vma, size, grow); 21231da177e4SLinus Torvalds if (!error) { 21244128997bSMichel Lespinasse /* 2125524e00b3SLiam R. Howlett * We only hold a shared mmap_lock lock here, so 2126524e00b3SLiam R. Howlett * we need to protect against concurrent vma 2127524e00b3SLiam R. Howlett * expansions. anon_vma_lock_write() doesn't 2128524e00b3SLiam R. Howlett * help here, as we don't guarantee that all 2129524e00b3SLiam R. Howlett * growable vmas in a mm share the same root 2130524e00b3SLiam R. Howlett * anon vma. So, we reuse mm->page_table_lock 2131524e00b3SLiam R. Howlett * to guard against concurrent vma expansions. 21324128997bSMichel Lespinasse */ 213309357814SOleg Nesterov spin_lock(&mm->page_table_lock); 213487e8827bSOleg Nesterov if (vma->vm_flags & VM_LOCKED) 213509357814SOleg Nesterov mm->locked_vm += grow; 213684638335SKonstantin Khlebnikov vm_stat_account(mm, vma->vm_flags, grow); 2137bf181b9fSMichel Lespinasse anon_vma_interval_tree_pre_update_vma(vma); 21381da177e4SLinus Torvalds vma->vm_start = address; 21391da177e4SLinus Torvalds vma->vm_pgoff -= grow; 2140d4af56c5SLiam R. Howlett /* Overwrite old entry in mtree. */ 2141fbcc3104SLiam R. Howlett mas_store_prealloc(&mas, vma); 2142bf181b9fSMichel Lespinasse anon_vma_interval_tree_post_update_vma(vma); 214309357814SOleg Nesterov spin_unlock(&mm->page_table_lock); 21444128997bSMichel Lespinasse 21453af9e859SEric B Munson perf_event_mmap(vma); 21461da177e4SLinus Torvalds } 21471da177e4SLinus Torvalds } 2148a626ca6aSLinus Torvalds } 214912352d3cSKonstantin Khlebnikov anon_vma_unlock_write(vma->anon_vma); 2150d4af56c5SLiam R. Howlett mas_destroy(&mas); 21512574d5e4SLiam R. Howlett validate_mm(mm); 21521da177e4SLinus Torvalds return error; 21531da177e4SLinus Torvalds } 21541da177e4SLinus Torvalds 21551be7107fSHugh Dickins /* enforced gap between the expanding stack and other mappings. */ 21561be7107fSHugh Dickins unsigned long stack_guard_gap = 256UL<<PAGE_SHIFT; 21571be7107fSHugh Dickins 21581be7107fSHugh Dickins static int __init cmdline_parse_stack_guard_gap(char *p) 21591be7107fSHugh Dickins { 21601be7107fSHugh Dickins unsigned long val; 21611be7107fSHugh Dickins char *endptr; 21621be7107fSHugh Dickins 21631be7107fSHugh Dickins val = simple_strtoul(p, &endptr, 10); 21641be7107fSHugh Dickins if (!*endptr) 21651be7107fSHugh Dickins stack_guard_gap = val << PAGE_SHIFT; 21661be7107fSHugh Dickins 2167e6d09493SRandy Dunlap return 1; 21681be7107fSHugh Dickins } 21691be7107fSHugh Dickins __setup("stack_guard_gap=", cmdline_parse_stack_guard_gap); 21701be7107fSHugh Dickins 2171b6a2fea3SOllie Wild #ifdef CONFIG_STACK_GROWSUP 21728d7071afSLinus Torvalds int expand_stack_locked(struct vm_area_struct *vma, unsigned long address) 2173b6a2fea3SOllie Wild { 2174b6a2fea3SOllie Wild return expand_upwards(vma, address); 2175b6a2fea3SOllie Wild } 2176b6a2fea3SOllie Wild 21778d7071afSLinus Torvalds struct vm_area_struct *find_extend_vma_locked(struct mm_struct *mm, unsigned long addr) 2178b6a2fea3SOllie Wild { 2179b6a2fea3SOllie Wild struct vm_area_struct *vma, *prev; 2180b6a2fea3SOllie Wild 2181b6a2fea3SOllie Wild addr &= PAGE_MASK; 2182b6a2fea3SOllie Wild vma = find_vma_prev(mm, addr, &prev); 2183b6a2fea3SOllie Wild if (vma && (vma->vm_start <= addr)) 2184b6a2fea3SOllie Wild return vma; 2185f440fa1aSLiam R. Howlett if (!prev) 2186f440fa1aSLiam R. Howlett return NULL; 21878d7071afSLinus Torvalds if (expand_stack_locked(prev, addr)) 2188b6a2fea3SOllie Wild return NULL; 2189cea10a19SMichel Lespinasse if (prev->vm_flags & VM_LOCKED) 2190fc05f566SKirill A. Shutemov populate_vma_page_range(prev, addr, prev->vm_end, NULL); 2191b6a2fea3SOllie Wild return prev; 2192b6a2fea3SOllie Wild } 2193b6a2fea3SOllie Wild #else 21948d7071afSLinus Torvalds int expand_stack_locked(struct vm_area_struct *vma, unsigned long address) 2195b6a2fea3SOllie Wild { 2196b6a2fea3SOllie Wild return expand_downwards(vma, address); 2197b6a2fea3SOllie Wild } 2198b6a2fea3SOllie Wild 21998d7071afSLinus Torvalds struct vm_area_struct *find_extend_vma_locked(struct mm_struct *mm, unsigned long addr) 22001da177e4SLinus Torvalds { 22011da177e4SLinus Torvalds struct vm_area_struct *vma; 22021da177e4SLinus Torvalds unsigned long start; 22031da177e4SLinus Torvalds 22041da177e4SLinus Torvalds addr &= PAGE_MASK; 22051da177e4SLinus Torvalds vma = find_vma(mm, addr); 22061da177e4SLinus Torvalds if (!vma) 22071da177e4SLinus Torvalds return NULL; 22081da177e4SLinus Torvalds if (vma->vm_start <= addr) 22091da177e4SLinus Torvalds return vma; 22101da177e4SLinus Torvalds start = vma->vm_start; 22118d7071afSLinus Torvalds if (expand_stack_locked(vma, addr)) 22121da177e4SLinus Torvalds return NULL; 2213cea10a19SMichel Lespinasse if (vma->vm_flags & VM_LOCKED) 2214fc05f566SKirill A. Shutemov populate_vma_page_range(vma, addr, start, NULL); 22151da177e4SLinus Torvalds return vma; 22161da177e4SLinus Torvalds } 22171da177e4SLinus Torvalds #endif 22181da177e4SLinus Torvalds 221969e583eaSLukas Bulwahn #if 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 */ 2323adb20b0cSLorenzo Stoakes static 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 2369c9d6e982SSuren Baghdasaryan vma_start_write(vma); 2370c9d6e982SSuren Baghdasaryan vma_start_write(new); 2371c9d6e982SSuren Baghdasaryan 2372b2b3b886SLiam R. Howlett init_vma_prep(&vp, vma); 2373b2b3b886SLiam R. Howlett vp.insert = new; 2374b2b3b886SLiam R. Howlett vma_prepare(&vp); 2375ccf1d78dSSuren Baghdasaryan vma_adjust_trans_huge(vma, vma->vm_start, addr, 0); 23761da177e4SLinus Torvalds 2377b2b3b886SLiam R. Howlett if (new_below) { 2378b2b3b886SLiam R. Howlett vma->vm_start = addr; 2379b2b3b886SLiam R. Howlett vma->vm_pgoff += (addr - new->vm_start) >> PAGE_SHIFT; 2380b2b3b886SLiam R. Howlett } else { 2381b2b3b886SLiam R. Howlett vma->vm_end = addr; 23829760ebffSLiam R. Howlett } 23835beb4930SRik van Riel 2384b2b3b886SLiam R. Howlett /* vma_complete stores the new vma */ 2385b2b3b886SLiam R. Howlett vma_complete(&vp, vmi, vma->vm_mm); 2386b2b3b886SLiam R. Howlett 2387b2b3b886SLiam R. Howlett /* Success. */ 2388b2b3b886SLiam R. Howlett if (new_below) 2389b2b3b886SLiam R. Howlett vma_next(vmi); 2390b2b3b886SLiam R. Howlett return 0; 2391b2b3b886SLiam R. Howlett 23925beb4930SRik van Riel out_free_mpol: 2393ef0855d3SOleg Nesterov mpol_put(vma_policy(new)); 2394b2b3b886SLiam R. Howlett out_free_vmi: 2395b2b3b886SLiam R. Howlett vma_iter_free(vmi); 23965beb4930SRik van Riel out_free_vma: 23973928d4f5SLinus Torvalds vm_area_free(new); 23985beb4930SRik van Riel return err; 23991da177e4SLinus Torvalds } 24001da177e4SLinus Torvalds 2401659ace58SKOSAKI Motohiro /* 2402659ace58SKOSAKI Motohiro * Split a vma into two pieces at address 'addr', a new vma is allocated 2403659ace58SKOSAKI Motohiro * either for the first part or the tail. 2404659ace58SKOSAKI Motohiro */ 2405adb20b0cSLorenzo Stoakes static int split_vma(struct vma_iterator *vmi, struct vm_area_struct *vma, 2406659ace58SKOSAKI Motohiro unsigned long addr, int new_below) 2407659ace58SKOSAKI Motohiro { 24089760ebffSLiam R. Howlett if (vma->vm_mm->map_count >= sysctl_max_map_count) 2409659ace58SKOSAKI Motohiro return -ENOMEM; 2410659ace58SKOSAKI Motohiro 24119760ebffSLiam R. Howlett return __split_vma(vmi, vma, addr, new_below); 2412f2ebfe43SLiam R. Howlett } 2413f2ebfe43SLiam R. Howlett 241411f9a21aSLiam R. Howlett /* 241594d7d923SLorenzo Stoakes * We are about to modify one or multiple of a VMA's flags, policy, userfaultfd 241694d7d923SLorenzo Stoakes * context and anonymous VMA name within the range [start, end). 241794d7d923SLorenzo Stoakes * 241894d7d923SLorenzo Stoakes * As a result, we might be able to merge the newly modified VMA range with an 241994d7d923SLorenzo Stoakes * adjacent VMA with identical properties. 242094d7d923SLorenzo Stoakes * 242194d7d923SLorenzo Stoakes * If no merge is possible and the range does not span the entirety of the VMA, 242294d7d923SLorenzo Stoakes * we then need to split the VMA to accommodate the change. 242394d7d923SLorenzo Stoakes * 242494d7d923SLorenzo Stoakes * The function returns either the merged VMA, the original VMA if a split was 242594d7d923SLorenzo Stoakes * required instead, or an error if the split failed. 242694d7d923SLorenzo Stoakes */ 242794d7d923SLorenzo Stoakes struct vm_area_struct *vma_modify(struct vma_iterator *vmi, 242894d7d923SLorenzo Stoakes struct vm_area_struct *prev, 242994d7d923SLorenzo Stoakes struct vm_area_struct *vma, 243094d7d923SLorenzo Stoakes unsigned long start, unsigned long end, 243194d7d923SLorenzo Stoakes unsigned long vm_flags, 243294d7d923SLorenzo Stoakes struct mempolicy *policy, 243394d7d923SLorenzo Stoakes struct vm_userfaultfd_ctx uffd_ctx, 243494d7d923SLorenzo Stoakes struct anon_vma_name *anon_name) 243594d7d923SLorenzo Stoakes { 243694d7d923SLorenzo Stoakes pgoff_t pgoff = vma->vm_pgoff + ((start - vma->vm_start) >> PAGE_SHIFT); 243794d7d923SLorenzo Stoakes struct vm_area_struct *merged; 243894d7d923SLorenzo Stoakes 24392c8b9474SYajun Deng merged = vma_merge(vmi, prev, vma, start, end, vm_flags, 24402c8b9474SYajun Deng pgoff, policy, uffd_ctx, anon_name); 244194d7d923SLorenzo Stoakes if (merged) 244294d7d923SLorenzo Stoakes return merged; 244394d7d923SLorenzo Stoakes 244494d7d923SLorenzo Stoakes if (vma->vm_start < start) { 244594d7d923SLorenzo Stoakes int err = split_vma(vmi, vma, start, 1); 244694d7d923SLorenzo Stoakes 244794d7d923SLorenzo Stoakes if (err) 244894d7d923SLorenzo Stoakes return ERR_PTR(err); 244994d7d923SLorenzo Stoakes } 245094d7d923SLorenzo Stoakes 245194d7d923SLorenzo Stoakes if (vma->vm_end > end) { 245294d7d923SLorenzo Stoakes int err = split_vma(vmi, vma, end, 0); 245394d7d923SLorenzo Stoakes 245494d7d923SLorenzo Stoakes if (err) 245594d7d923SLorenzo Stoakes return ERR_PTR(err); 245694d7d923SLorenzo Stoakes } 245794d7d923SLorenzo Stoakes 245894d7d923SLorenzo Stoakes return vma; 245994d7d923SLorenzo Stoakes } 246094d7d923SLorenzo Stoakes 246194d7d923SLorenzo Stoakes /* 24624b5f2d20SLorenzo Stoakes * Attempt to merge a newly mapped VMA with those adjacent to it. The caller 24634b5f2d20SLorenzo Stoakes * must ensure that [start, end) does not overlap any existing VMA. 24644b5f2d20SLorenzo Stoakes */ 24654b5f2d20SLorenzo Stoakes static struct vm_area_struct 24664b5f2d20SLorenzo Stoakes *vma_merge_new_vma(struct vma_iterator *vmi, struct vm_area_struct *prev, 24674b5f2d20SLorenzo Stoakes struct vm_area_struct *vma, unsigned long start, 24684b5f2d20SLorenzo Stoakes unsigned long end, pgoff_t pgoff) 24694b5f2d20SLorenzo Stoakes { 24702c8b9474SYajun Deng return vma_merge(vmi, prev, vma, start, end, vma->vm_flags, pgoff, 24712c8b9474SYajun Deng vma_policy(vma), vma->vm_userfaultfd_ctx, anon_vma_name(vma)); 24724b5f2d20SLorenzo Stoakes } 24734b5f2d20SLorenzo Stoakes 24744b5f2d20SLorenzo Stoakes /* 247593bf5d4aSLorenzo Stoakes * Expand vma by delta bytes, potentially merging with an immediately adjacent 247693bf5d4aSLorenzo Stoakes * VMA with identical properties. 247793bf5d4aSLorenzo Stoakes */ 247893bf5d4aSLorenzo Stoakes struct vm_area_struct *vma_merge_extend(struct vma_iterator *vmi, 247993bf5d4aSLorenzo Stoakes struct vm_area_struct *vma, 248093bf5d4aSLorenzo Stoakes unsigned long delta) 248193bf5d4aSLorenzo Stoakes { 248293bf5d4aSLorenzo Stoakes pgoff_t pgoff = vma->vm_pgoff + vma_pages(vma); 248393bf5d4aSLorenzo Stoakes 248493bf5d4aSLorenzo Stoakes /* vma is specified as prev, so case 1 or 2 will apply. */ 24852c8b9474SYajun Deng return vma_merge(vmi, vma, vma, vma->vm_end, vma->vm_end + delta, 24862c8b9474SYajun Deng vma->vm_flags, pgoff, vma_policy(vma), 24872c8b9474SYajun Deng vma->vm_userfaultfd_ctx, anon_vma_name(vma)); 248893bf5d4aSLorenzo Stoakes } 248993bf5d4aSLorenzo Stoakes 249093bf5d4aSLorenzo Stoakes /* 2491183654ceSLiam R. Howlett * do_vmi_align_munmap() - munmap the aligned region from @start to @end. 2492183654ceSLiam R. Howlett * @vmi: The vma iterator 249311f9a21aSLiam R. Howlett * @vma: The starting vm_area_struct 249411f9a21aSLiam R. Howlett * @mm: The mm_struct 249511f9a21aSLiam R. Howlett * @start: The aligned start address to munmap. 249611f9a21aSLiam R. Howlett * @end: The aligned end address to munmap. 249711f9a21aSLiam R. Howlett * @uf: The userfaultfd list_head 2498408579cdSLiam R. Howlett * @unlock: Set to true to drop the mmap_lock. unlocking only happens on 2499408579cdSLiam R. Howlett * success. 250011f9a21aSLiam R. Howlett * 2501408579cdSLiam R. Howlett * Return: 0 on success and drops the lock if so directed, error and leaves the 2502408579cdSLiam R. Howlett * lock held otherwise. 250311f9a21aSLiam R. Howlett */ 250411f9a21aSLiam R. Howlett static int 2505183654ceSLiam R. Howlett do_vmi_align_munmap(struct vma_iterator *vmi, struct vm_area_struct *vma, 250611f9a21aSLiam R. Howlett struct mm_struct *mm, unsigned long start, 2507408579cdSLiam R. Howlett unsigned long end, struct list_head *uf, bool unlock) 250811f9a21aSLiam R. Howlett { 2509763ecb03SLiam R. Howlett struct vm_area_struct *prev, *next = NULL; 2510763ecb03SLiam R. Howlett struct maple_tree mt_detach; 2511763ecb03SLiam R. Howlett int count = 0; 2512d4af56c5SLiam R. Howlett int error = -ENOMEM; 2513606c812eSLiam R. Howlett unsigned long locked_vm = 0; 2514763ecb03SLiam R. Howlett MA_STATE(mas_detach, &mt_detach, 0, 0); 25153dd44325SLiam R. Howlett mt_init_flags(&mt_detach, vmi->mas.tree->ma_flags & MT_FLAGS_LOCK_MASK); 251602fdb25fSLiam R. Howlett mt_on_stack(mt_detach); 2517524e00b3SLiam R. Howlett 25181da177e4SLinus Torvalds /* 25191da177e4SLinus Torvalds * If we need to split any vma, do it now to save pain later. 25201da177e4SLinus Torvalds * 25211da177e4SLinus Torvalds * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially 25221da177e4SLinus Torvalds * unmapped vm_area_struct will remain in use: so lower split_vma 25231da177e4SLinus Torvalds * places tmp vma above, and higher split_vma places tmp vma below. 25241da177e4SLinus Torvalds */ 2525763ecb03SLiam R. Howlett 2526763ecb03SLiam R. Howlett /* Does it split the first one? */ 2527146425a3SHugh Dickins if (start > vma->vm_start) { 2528659ace58SKOSAKI Motohiro 2529659ace58SKOSAKI Motohiro /* 2530659ace58SKOSAKI Motohiro * Make sure that map_count on return from munmap() will 2531659ace58SKOSAKI Motohiro * not exceed its limit; but let map_count go just above 2532659ace58SKOSAKI Motohiro * its limit temporarily, to help free resources as expected. 2533659ace58SKOSAKI Motohiro */ 2534659ace58SKOSAKI Motohiro if (end < vma->vm_end && mm->map_count >= sysctl_max_map_count) 2535d4af56c5SLiam R. Howlett goto map_count_exceeded; 2536659ace58SKOSAKI Motohiro 25376935e052SLiam R. Howlett error = __split_vma(vmi, vma, start, 1); 25381da177e4SLinus Torvalds if (error) 2539763ecb03SLiam R. Howlett goto start_split_failed; 25401da177e4SLinus Torvalds } 25411da177e4SLinus Torvalds 2542763ecb03SLiam R. Howlett /* 2543763ecb03SLiam R. Howlett * Detach a range of VMAs from the mm. Using next as a temp variable as 2544763ecb03SLiam R. Howlett * it is always overwritten. 2545763ecb03SLiam R. Howlett */ 25466935e052SLiam R. Howlett next = vma; 25476935e052SLiam R. Howlett do { 2548763ecb03SLiam R. Howlett /* Does it split the end? */ 2549763ecb03SLiam R. Howlett if (next->vm_end > end) { 25506b73cff2SLiam R. Howlett error = __split_vma(vmi, next, end, 0); 25511da177e4SLinus Torvalds if (error) 2552763ecb03SLiam R. Howlett goto end_split_failed; 255311f9a21aSLiam R. Howlett } 2554606c812eSLiam R. Howlett vma_start_write(next); 2555fd892593SLiam R. Howlett mas_set(&mas_detach, count); 25566c26bd43SDavid Woodhouse error = mas_store_gfp(&mas_detach, next, GFP_KERNEL); 25576c26bd43SDavid Woodhouse if (error) 2558606c812eSLiam R. Howlett goto munmap_gather_failed; 2559606c812eSLiam R. Howlett vma_mark_detached(next, true); 2560606c812eSLiam R. Howlett if (next->vm_flags & VM_LOCKED) 2561606c812eSLiam R. Howlett locked_vm += vma_pages(next); 2562763ecb03SLiam R. Howlett 2563763ecb03SLiam R. Howlett count++; 25642376dd7cSAndrea Arcangeli if (unlikely(uf)) { 25652376dd7cSAndrea Arcangeli /* 25662376dd7cSAndrea Arcangeli * If userfaultfd_unmap_prep returns an error the vmas 2567f0953a1bSIngo Molnar * will remain split, but userland will get a 25682376dd7cSAndrea Arcangeli * highly unexpected error anyway. This is no 25692376dd7cSAndrea Arcangeli * different than the case where the first of the two 25702376dd7cSAndrea Arcangeli * __split_vma fails, but we don't undo the first 25712376dd7cSAndrea Arcangeli * split, despite we could. This is unlikely enough 25722376dd7cSAndrea Arcangeli * failure that it's not worth optimizing it for. 25732376dd7cSAndrea Arcangeli */ 257465ac1320SLiam R. Howlett error = userfaultfd_unmap_prep(next, start, end, uf); 257511f9a21aSLiam R. Howlett 25762376dd7cSAndrea Arcangeli if (error) 2577d4af56c5SLiam R. Howlett goto userfaultfd_error; 25782376dd7cSAndrea Arcangeli } 2579ba470de4SRik van Riel #ifdef CONFIG_DEBUG_VM_MAPLE_TREE 2580ba470de4SRik van Riel BUG_ON(next->vm_start < start); 2581ba470de4SRik van Riel BUG_ON(next->vm_start > end); 25821da177e4SLinus Torvalds #endif 25836935e052SLiam R. Howlett } for_each_vma_range(*vmi, next, end); 25842376dd7cSAndrea Arcangeli 2585763ecb03SLiam R. Howlett #if defined(CONFIG_DEBUG_VM_MAPLE_TREE) 2586763ecb03SLiam R. Howlett /* Make sure no VMAs are about to be lost. */ 2587763ecb03SLiam R. Howlett { 2588fd892593SLiam R. Howlett MA_STATE(test, &mt_detach, 0, 0); 2589763ecb03SLiam R. Howlett struct vm_area_struct *vma_mas, *vma_test; 2590763ecb03SLiam R. Howlett int test_count = 0; 2591763ecb03SLiam R. Howlett 2592183654ceSLiam R. Howlett vma_iter_set(vmi, start); 2593763ecb03SLiam R. Howlett rcu_read_lock(); 2594fd892593SLiam R. Howlett vma_test = mas_find(&test, count - 1); 2595183654ceSLiam R. Howlett for_each_vma_range(*vmi, vma_mas, end) { 2596763ecb03SLiam R. Howlett BUG_ON(vma_mas != vma_test); 2597763ecb03SLiam R. Howlett test_count++; 2598fd892593SLiam R. Howlett vma_test = mas_next(&test, count - 1); 2599763ecb03SLiam R. Howlett } 2600763ecb03SLiam R. Howlett rcu_read_unlock(); 2601763ecb03SLiam R. Howlett BUG_ON(count != test_count); 2602763ecb03SLiam R. Howlett } 2603763ecb03SLiam R. Howlett #endif 26046935e052SLiam R. Howlett 26056935e052SLiam R. Howlett while (vma_iter_addr(vmi) > start) 26066935e052SLiam R. Howlett vma_iter_prev_range(vmi); 26076935e052SLiam R. Howlett 26086c26bd43SDavid Woodhouse error = vma_iter_clear_gfp(vmi, start, end, GFP_KERNEL); 26096c26bd43SDavid Woodhouse if (error) 2610606c812eSLiam R. Howlett goto clear_tree_failed; 26110378c0a0SLiam R. Howlett 26126c26bd43SDavid Woodhouse /* Point of no return */ 2613606c812eSLiam R. Howlett mm->locked_vm -= locked_vm; 2614763ecb03SLiam R. Howlett mm->map_count -= count; 2615408579cdSLiam R. Howlett if (unlock) 2616d8ed45c5SMichel Lespinasse mmap_write_downgrade(mm); 2617dd2283f2SYang Shi 26186935e052SLiam R. Howlett prev = vma_iter_prev_range(vmi); 26196935e052SLiam R. Howlett next = vma_next(vmi); 26206935e052SLiam R. Howlett if (next) 26216935e052SLiam R. Howlett vma_iter_prev_range(vmi); 26226935e052SLiam R. Howlett 262368f48381SSuren Baghdasaryan /* 262468f48381SSuren Baghdasaryan * We can free page tables without write-locking mmap_lock because VMAs 262568f48381SSuren Baghdasaryan * were isolated before we downgraded mmap_lock. 262668f48381SSuren Baghdasaryan */ 2627fd892593SLiam R. Howlett mas_set(&mas_detach, 1); 2628fd892593SLiam R. Howlett unmap_region(mm, &mas_detach, vma, prev, next, start, end, count, 2629fd892593SLiam R. Howlett !unlock); 2630763ecb03SLiam R. Howlett /* Statistics and freeing VMAs */ 2631fd892593SLiam R. Howlett mas_set(&mas_detach, 0); 2632763ecb03SLiam R. Howlett remove_mt(mm, &mas_detach); 2633ae80b404SLinus Torvalds validate_mm(mm); 2634408579cdSLiam R. Howlett if (unlock) 2635408579cdSLiam R. Howlett mmap_read_unlock(mm); 26361da177e4SLinus Torvalds 263702fdb25fSLiam R. Howlett __mt_destroy(&mt_detach); 2638408579cdSLiam R. Howlett return 0; 2639d4af56c5SLiam R. Howlett 2640606c812eSLiam R. Howlett clear_tree_failed: 2641d4af56c5SLiam R. Howlett userfaultfd_error: 2642606c812eSLiam R. Howlett munmap_gather_failed: 2643763ecb03SLiam R. Howlett end_split_failed: 2644606c812eSLiam R. Howlett mas_set(&mas_detach, 0); 2645606c812eSLiam R. Howlett mas_for_each(&mas_detach, next, end) 2646606c812eSLiam R. Howlett vma_mark_detached(next, false); 2647606c812eSLiam R. Howlett 2648763ecb03SLiam R. Howlett __mt_destroy(&mt_detach); 2649763ecb03SLiam R. Howlett start_split_failed: 2650763ecb03SLiam R. Howlett map_count_exceeded: 2651b5641a5dSLinus Torvalds validate_mm(mm); 2652d4af56c5SLiam R. Howlett return error; 26531da177e4SLinus Torvalds } 26541da177e4SLinus Torvalds 265511f9a21aSLiam R. Howlett /* 2656183654ceSLiam R. Howlett * do_vmi_munmap() - munmap a given range. 2657183654ceSLiam R. Howlett * @vmi: The vma iterator 265811f9a21aSLiam R. Howlett * @mm: The mm_struct 265911f9a21aSLiam R. Howlett * @start: The start address to munmap 266011f9a21aSLiam R. Howlett * @len: The length of the range to munmap 266111f9a21aSLiam R. Howlett * @uf: The userfaultfd list_head 2662408579cdSLiam R. Howlett * @unlock: set to true if the user wants to drop the mmap_lock on success 266311f9a21aSLiam R. Howlett * 266411f9a21aSLiam R. Howlett * This function takes a @mas that is either pointing to the previous VMA or set 266511f9a21aSLiam R. Howlett * to MA_START and sets it up to remove the mapping(s). The @len will be 266611f9a21aSLiam R. Howlett * aligned and any arch_unmap work will be preformed. 266711f9a21aSLiam R. Howlett * 2668408579cdSLiam R. Howlett * Return: 0 on success and drops the lock if so directed, error and leaves the 2669408579cdSLiam R. Howlett * lock held otherwise. 267011f9a21aSLiam R. Howlett */ 2671183654ceSLiam R. Howlett int do_vmi_munmap(struct vma_iterator *vmi, struct mm_struct *mm, 267211f9a21aSLiam R. Howlett unsigned long start, size_t len, struct list_head *uf, 2673408579cdSLiam R. Howlett bool unlock) 267411f9a21aSLiam R. Howlett { 267511f9a21aSLiam R. Howlett unsigned long end; 267611f9a21aSLiam R. Howlett struct vm_area_struct *vma; 267711f9a21aSLiam R. Howlett 267811f9a21aSLiam R. Howlett if ((offset_in_page(start)) || start > TASK_SIZE || len > TASK_SIZE-start) 267911f9a21aSLiam R. Howlett return -EINVAL; 268011f9a21aSLiam R. Howlett 268111f9a21aSLiam R. Howlett end = start + PAGE_ALIGN(len); 268211f9a21aSLiam R. Howlett if (end == start) 268311f9a21aSLiam R. Howlett return -EINVAL; 268411f9a21aSLiam R. Howlett 268511f9a21aSLiam R. Howlett /* arch_unmap() might do unmaps itself. */ 268611f9a21aSLiam R. Howlett arch_unmap(mm, start, end); 268711f9a21aSLiam R. Howlett 268811f9a21aSLiam R. Howlett /* Find the first overlapping VMA */ 2689183654ceSLiam R. Howlett vma = vma_find(vmi, end); 2690408579cdSLiam R. Howlett if (!vma) { 2691408579cdSLiam R. Howlett if (unlock) 2692408579cdSLiam R. Howlett mmap_write_unlock(mm); 269311f9a21aSLiam R. Howlett return 0; 2694408579cdSLiam R. Howlett } 269511f9a21aSLiam R. Howlett 2696408579cdSLiam R. Howlett return do_vmi_align_munmap(vmi, vma, mm, start, end, uf, unlock); 269711f9a21aSLiam R. Howlett } 269811f9a21aSLiam R. Howlett 269911f9a21aSLiam R. Howlett /* do_munmap() - Wrapper function for non-maple tree aware do_munmap() calls. 270011f9a21aSLiam R. Howlett * @mm: The mm_struct 270111f9a21aSLiam R. Howlett * @start: The start address to munmap 270211f9a21aSLiam R. Howlett * @len: The length to be munmapped. 270311f9a21aSLiam R. Howlett * @uf: The userfaultfd list_head 2704408579cdSLiam R. Howlett * 2705408579cdSLiam R. Howlett * Return: 0 on success, error otherwise. 270611f9a21aSLiam R. Howlett */ 2707dd2283f2SYang Shi int do_munmap(struct mm_struct *mm, unsigned long start, size_t len, 2708dd2283f2SYang Shi struct list_head *uf) 2709dd2283f2SYang Shi { 2710183654ceSLiam R. Howlett VMA_ITERATOR(vmi, mm, start); 271111f9a21aSLiam R. Howlett 2712183654ceSLiam R. Howlett return do_vmi_munmap(&vmi, mm, start, len, uf, false); 2713dd2283f2SYang Shi } 2714dd2283f2SYang Shi 2715e99668a5SLiam R. Howlett unsigned long mmap_region(struct file *file, unsigned long addr, 2716e99668a5SLiam R. Howlett unsigned long len, vm_flags_t vm_flags, unsigned long pgoff, 2717e99668a5SLiam R. Howlett struct list_head *uf) 2718e99668a5SLiam R. Howlett { 2719e99668a5SLiam R. Howlett struct mm_struct *mm = current->mm; 2720e99668a5SLiam R. Howlett struct vm_area_struct *vma = NULL; 2721e99668a5SLiam R. Howlett struct vm_area_struct *next, *prev, *merge; 2722e99668a5SLiam R. Howlett pgoff_t pglen = len >> PAGE_SHIFT; 2723e99668a5SLiam R. Howlett unsigned long charged = 0; 2724e99668a5SLiam R. Howlett unsigned long end = addr + len; 2725e99668a5SLiam R. Howlett unsigned long merge_start = addr, merge_end = end; 272615897894SLorenzo Stoakes bool writable_file_mapping = false; 2727e99668a5SLiam R. Howlett pgoff_t vm_pgoff; 2728e99668a5SLiam R. Howlett int error; 2729183654ceSLiam R. Howlett VMA_ITERATOR(vmi, mm, addr); 2730e99668a5SLiam R. Howlett 2731e99668a5SLiam R. Howlett /* Check against address space limit. */ 2732e99668a5SLiam R. Howlett if (!may_expand_vm(mm, vm_flags, len >> PAGE_SHIFT)) { 2733e99668a5SLiam R. Howlett unsigned long nr_pages; 2734e99668a5SLiam R. Howlett 2735e99668a5SLiam R. Howlett /* 2736e99668a5SLiam R. Howlett * MAP_FIXED may remove pages of mappings that intersects with 2737e99668a5SLiam R. Howlett * requested mapping. Account for the pages it would unmap. 2738e99668a5SLiam R. Howlett */ 2739e99668a5SLiam R. Howlett nr_pages = count_vma_pages_range(mm, addr, end); 2740e99668a5SLiam R. Howlett 2741e99668a5SLiam R. Howlett if (!may_expand_vm(mm, vm_flags, 2742e99668a5SLiam R. Howlett (len >> PAGE_SHIFT) - nr_pages)) 2743e99668a5SLiam R. Howlett return -ENOMEM; 2744e99668a5SLiam R. Howlett } 2745e99668a5SLiam R. Howlett 2746e99668a5SLiam R. Howlett /* Unmap any existing mapping in the area */ 2747183654ceSLiam R. Howlett if (do_vmi_munmap(&vmi, mm, addr, len, uf, false)) 2748e99668a5SLiam R. Howlett return -ENOMEM; 2749e99668a5SLiam R. Howlett 2750e99668a5SLiam R. Howlett /* 2751e99668a5SLiam R. Howlett * Private writable mapping: check memory availability 2752e99668a5SLiam R. Howlett */ 2753e99668a5SLiam R. Howlett if (accountable_mapping(file, vm_flags)) { 2754e99668a5SLiam R. Howlett charged = len >> PAGE_SHIFT; 2755e99668a5SLiam R. Howlett if (security_vm_enough_memory_mm(mm, charged)) 2756e99668a5SLiam R. Howlett return -ENOMEM; 2757e99668a5SLiam R. Howlett vm_flags |= VM_ACCOUNT; 2758e99668a5SLiam R. Howlett } 2759e99668a5SLiam R. Howlett 2760183654ceSLiam R. Howlett next = vma_next(&vmi); 2761183654ceSLiam R. Howlett prev = vma_prev(&vmi); 276253bee98dSLiam R. Howlett if (vm_flags & VM_SPECIAL) { 276353bee98dSLiam R. Howlett if (prev) 276453bee98dSLiam R. Howlett vma_iter_next_range(&vmi); 2765e99668a5SLiam R. Howlett goto cannot_expand; 276653bee98dSLiam R. Howlett } 2767e99668a5SLiam R. Howlett 2768e99668a5SLiam R. Howlett /* Attempt to expand an old mapping */ 2769e99668a5SLiam R. Howlett /* Check next */ 2770e99668a5SLiam R. Howlett if (next && next->vm_start == end && !vma_policy(next) && 2771e99668a5SLiam R. Howlett can_vma_merge_before(next, vm_flags, NULL, file, pgoff+pglen, 2772e99668a5SLiam R. Howlett NULL_VM_UFFD_CTX, NULL)) { 2773e99668a5SLiam R. Howlett merge_end = next->vm_end; 2774e99668a5SLiam R. Howlett vma = next; 2775e99668a5SLiam R. Howlett vm_pgoff = next->vm_pgoff - pglen; 2776e99668a5SLiam R. Howlett } 2777e99668a5SLiam R. Howlett 2778e99668a5SLiam R. Howlett /* Check prev */ 2779e99668a5SLiam R. Howlett if (prev && prev->vm_end == addr && !vma_policy(prev) && 2780e99668a5SLiam R. Howlett (vma ? can_vma_merge_after(prev, vm_flags, vma->anon_vma, file, 2781e99668a5SLiam R. Howlett pgoff, vma->vm_userfaultfd_ctx, NULL) : 2782e99668a5SLiam R. Howlett can_vma_merge_after(prev, vm_flags, NULL, file, pgoff, 2783e99668a5SLiam R. Howlett NULL_VM_UFFD_CTX, NULL))) { 2784e99668a5SLiam R. Howlett merge_start = prev->vm_start; 2785e99668a5SLiam R. Howlett vma = prev; 2786e99668a5SLiam R. Howlett vm_pgoff = prev->vm_pgoff; 278753bee98dSLiam R. Howlett } else if (prev) { 278853bee98dSLiam R. Howlett vma_iter_next_range(&vmi); 2789e99668a5SLiam R. Howlett } 2790e99668a5SLiam R. Howlett 2791e99668a5SLiam R. Howlett /* Actually expand, if possible */ 2792e99668a5SLiam R. Howlett if (vma && 27933c441ab7SLiam R. Howlett !vma_expand(&vmi, vma, merge_start, merge_end, vm_pgoff, next)) { 2794e99668a5SLiam R. Howlett khugepaged_enter_vma(vma, vm_flags); 2795e99668a5SLiam R. Howlett goto expanded; 2796e99668a5SLiam R. Howlett } 2797e99668a5SLiam R. Howlett 279853bee98dSLiam R. Howlett if (vma == prev) 279953bee98dSLiam R. Howlett vma_iter_set(&vmi, addr); 2800e99668a5SLiam R. Howlett cannot_expand: 28015c1c03deSLiam R. Howlett 2802e99668a5SLiam R. Howlett /* 2803e99668a5SLiam R. Howlett * Determine the object being mapped and call the appropriate 2804e99668a5SLiam R. Howlett * specific mapper. the address has already been validated, but 2805e99668a5SLiam R. Howlett * not unmapped, but the maps are removed from the list. 2806e99668a5SLiam R. Howlett */ 2807e99668a5SLiam R. Howlett vma = vm_area_alloc(mm); 2808e99668a5SLiam R. Howlett if (!vma) { 2809e99668a5SLiam R. Howlett error = -ENOMEM; 2810e99668a5SLiam R. Howlett goto unacct_error; 2811e99668a5SLiam R. Howlett } 2812e99668a5SLiam R. Howlett 281353bee98dSLiam R. Howlett vma_iter_config(&vmi, addr, end); 2814412c6ef9SYajun Deng vma_set_range(vma, addr, end, pgoff); 28151c71222eSSuren Baghdasaryan vm_flags_init(vma, vm_flags); 2816e99668a5SLiam R. Howlett vma->vm_page_prot = vm_get_page_prot(vm_flags); 2817e99668a5SLiam R. Howlett 2818e99668a5SLiam R. Howlett if (file) { 2819e99668a5SLiam R. Howlett vma->vm_file = get_file(file); 2820e99668a5SLiam R. Howlett error = call_mmap(file, vma); 2821e99668a5SLiam R. Howlett if (error) 2822e99668a5SLiam R. Howlett goto unmap_and_free_vma; 2823e99668a5SLiam R. Howlett 282415897894SLorenzo Stoakes if (vma_is_shared_maywrite(vma)) { 282515897894SLorenzo Stoakes error = mapping_map_writable(file->f_mapping); 282615897894SLorenzo Stoakes if (error) 282715897894SLorenzo Stoakes goto close_and_free_vma; 282815897894SLorenzo Stoakes 282915897894SLorenzo Stoakes writable_file_mapping = true; 283015897894SLorenzo Stoakes } 283115897894SLorenzo Stoakes 2832a57b7051SLiam Howlett /* 2833a57b7051SLiam Howlett * Expansion is handled above, merging is handled below. 2834a57b7051SLiam Howlett * Drivers should not alter the address of the VMA. 2835e99668a5SLiam R. Howlett */ 2836a57b7051SLiam Howlett error = -EINVAL; 2837cc8d1b09SLiam R. Howlett if (WARN_ON((addr != vma->vm_start))) 2838a57b7051SLiam Howlett goto close_and_free_vma; 2839e99668a5SLiam R. Howlett 284053bee98dSLiam R. Howlett vma_iter_config(&vmi, addr, end); 2841e99668a5SLiam R. Howlett /* 2842e99668a5SLiam R. Howlett * If vm_flags changed after call_mmap(), we should try merge 2843e99668a5SLiam R. Howlett * vma again as we may succeed this time. 2844e99668a5SLiam R. Howlett */ 2845e99668a5SLiam R. Howlett if (unlikely(vm_flags != vma->vm_flags && prev)) { 28464b5f2d20SLorenzo Stoakes merge = vma_merge_new_vma(&vmi, prev, vma, 28474b5f2d20SLorenzo Stoakes vma->vm_start, vma->vm_end, 28484b5f2d20SLorenzo Stoakes vma->vm_pgoff); 2849e99668a5SLiam R. Howlett if (merge) { 2850e99668a5SLiam R. Howlett /* 2851e99668a5SLiam R. Howlett * ->mmap() can change vma->vm_file and fput 2852e99668a5SLiam R. Howlett * the original file. So fput the vma->vm_file 2853e99668a5SLiam R. Howlett * here or we would add an extra fput for file 2854e99668a5SLiam R. Howlett * and cause general protection fault 2855e99668a5SLiam R. Howlett * ultimately. 2856e99668a5SLiam R. Howlett */ 2857e99668a5SLiam R. Howlett fput(vma->vm_file); 2858e99668a5SLiam R. Howlett vm_area_free(vma); 2859e99668a5SLiam R. Howlett vma = merge; 2860e99668a5SLiam R. Howlett /* Update vm_flags to pick up the change. */ 2861e99668a5SLiam R. Howlett vm_flags = vma->vm_flags; 2862e99668a5SLiam R. Howlett goto unmap_writable; 2863e99668a5SLiam R. Howlett } 2864e99668a5SLiam R. Howlett } 2865e99668a5SLiam R. Howlett 2866e99668a5SLiam R. Howlett vm_flags = vma->vm_flags; 2867e99668a5SLiam R. Howlett } else if (vm_flags & VM_SHARED) { 2868e99668a5SLiam R. Howlett error = shmem_zero_setup(vma); 2869e99668a5SLiam R. Howlett if (error) 2870e99668a5SLiam R. Howlett goto free_vma; 2871e99668a5SLiam R. Howlett } else { 2872e99668a5SLiam R. Howlett vma_set_anonymous(vma); 2873e99668a5SLiam R. Howlett } 2874e99668a5SLiam R. Howlett 2875b507808eSJoey Gouly if (map_deny_write_exec(vma, vma->vm_flags)) { 2876b507808eSJoey Gouly error = -EACCES; 2877b507808eSJoey Gouly goto close_and_free_vma; 2878b507808eSJoey Gouly } 2879b507808eSJoey Gouly 2880e99668a5SLiam R. Howlett /* Allow architectures to sanity-check the vm_flags */ 2881e99668a5SLiam R. Howlett error = -EINVAL; 2882cc8d1b09SLiam R. Howlett if (!arch_validate_flags(vma->vm_flags)) 2883deb0f656SCarlos Llamas goto close_and_free_vma; 2884e99668a5SLiam R. Howlett 2885e99668a5SLiam R. Howlett error = -ENOMEM; 2886b5df0922SLiam R. Howlett if (vma_iter_prealloc(&vmi, vma)) 28875789151eSMike Kravetz goto close_and_free_vma; 2888e99668a5SLiam R. Howlett 28891c7873e3SHugh Dickins /* Lock the VMA since it is modified after insertion into VMA tree */ 28901c7873e3SHugh Dickins vma_start_write(vma); 2891183654ceSLiam R. Howlett vma_iter_store(&vmi, vma); 2892e99668a5SLiam R. Howlett mm->map_count++; 289330afc8c3SYajun Deng vma_link_file(vma); 2894e99668a5SLiam R. Howlett 2895e99668a5SLiam R. Howlett /* 2896e99668a5SLiam R. Howlett * vma_merge() calls khugepaged_enter_vma() either, the below 2897e99668a5SLiam R. Howlett * call covers the non-merge case. 2898e99668a5SLiam R. Howlett */ 2899e99668a5SLiam R. Howlett khugepaged_enter_vma(vma, vma->vm_flags); 2900e99668a5SLiam R. Howlett 2901e99668a5SLiam R. Howlett /* Once vma denies write, undo our temporary denial count */ 2902e99668a5SLiam R. Howlett unmap_writable: 290315897894SLorenzo Stoakes if (writable_file_mapping) 2904e99668a5SLiam R. Howlett mapping_unmap_writable(file->f_mapping); 2905e99668a5SLiam R. Howlett file = vma->vm_file; 2906d7597f59SStefan Roesch ksm_add_vma(vma); 2907e99668a5SLiam R. Howlett expanded: 2908e99668a5SLiam R. Howlett perf_event_mmap(vma); 2909e99668a5SLiam R. Howlett 2910e99668a5SLiam R. Howlett vm_stat_account(mm, vm_flags, len >> PAGE_SHIFT); 2911e99668a5SLiam R. Howlett if (vm_flags & VM_LOCKED) { 2912e99668a5SLiam R. Howlett if ((vm_flags & VM_SPECIAL) || vma_is_dax(vma) || 2913e99668a5SLiam R. Howlett is_vm_hugetlb_page(vma) || 2914e99668a5SLiam R. Howlett vma == get_gate_vma(current->mm)) 2915e430a95aSSuren Baghdasaryan vm_flags_clear(vma, VM_LOCKED_MASK); 2916e99668a5SLiam R. Howlett else 2917e99668a5SLiam R. Howlett mm->locked_vm += (len >> PAGE_SHIFT); 2918e99668a5SLiam R. Howlett } 2919e99668a5SLiam R. Howlett 2920e99668a5SLiam R. Howlett if (file) 2921e99668a5SLiam R. Howlett uprobe_mmap(vma); 2922e99668a5SLiam R. Howlett 2923e99668a5SLiam R. Howlett /* 2924e99668a5SLiam R. Howlett * New (or expanded) vma always get soft dirty status. 2925e99668a5SLiam R. Howlett * Otherwise user-space soft-dirty page tracker won't 2926e99668a5SLiam R. Howlett * be able to distinguish situation when vma area unmapped, 2927e99668a5SLiam R. Howlett * then new mapped in-place (which must be aimed as 2928e99668a5SLiam R. Howlett * a completely new data area). 2929e99668a5SLiam R. Howlett */ 29301c71222eSSuren Baghdasaryan vm_flags_set(vma, VM_SOFTDIRTY); 2931e99668a5SLiam R. Howlett 2932e99668a5SLiam R. Howlett vma_set_page_prot(vma); 2933e99668a5SLiam R. Howlett 2934e99668a5SLiam R. Howlett validate_mm(mm); 2935e99668a5SLiam R. Howlett return addr; 2936e99668a5SLiam R. Howlett 2937deb0f656SCarlos Llamas close_and_free_vma: 2938cc8d1b09SLiam R. Howlett if (file && vma->vm_ops && vma->vm_ops->close) 2939deb0f656SCarlos Llamas vma->vm_ops->close(vma); 2940cc8d1b09SLiam R. Howlett 2941cc8d1b09SLiam R. Howlett if (file || vma->vm_file) { 2942e99668a5SLiam R. Howlett unmap_and_free_vma: 2943e99668a5SLiam R. Howlett fput(vma->vm_file); 2944e99668a5SLiam R. Howlett vma->vm_file = NULL; 2945e99668a5SLiam R. Howlett 2946fd892593SLiam R. Howlett vma_iter_set(&vmi, vma->vm_end); 2947e99668a5SLiam R. Howlett /* Undo any partial mapping done by a device driver. */ 2948fd892593SLiam R. Howlett unmap_region(mm, &vmi.mas, vma, prev, next, vma->vm_start, 2949fd892593SLiam R. Howlett vma->vm_end, vma->vm_end, true); 2950cc8d1b09SLiam R. Howlett } 295115897894SLorenzo Stoakes if (writable_file_mapping) 2952e99668a5SLiam R. Howlett mapping_unmap_writable(file->f_mapping); 2953e99668a5SLiam R. Howlett free_vma: 2954e99668a5SLiam R. Howlett vm_area_free(vma); 2955e99668a5SLiam R. Howlett unacct_error: 2956e99668a5SLiam R. Howlett if (charged) 2957e99668a5SLiam R. Howlett vm_unacct_memory(charged); 2958e99668a5SLiam R. Howlett validate_mm(mm); 2959e99668a5SLiam R. Howlett return error; 2960e99668a5SLiam R. Howlett } 2961e99668a5SLiam R. Howlett 2962408579cdSLiam R. Howlett static int __vm_munmap(unsigned long start, size_t len, bool unlock) 2963a46ef99dSLinus Torvalds { 2964a46ef99dSLinus Torvalds int ret; 2965bfce281cSAl Viro struct mm_struct *mm = current->mm; 2966897ab3e0SMike Rapoport LIST_HEAD(uf); 2967183654ceSLiam R. Howlett VMA_ITERATOR(vmi, mm, start); 2968a46ef99dSLinus Torvalds 2969d8ed45c5SMichel Lespinasse if (mmap_write_lock_killable(mm)) 2970ae798783SMichal Hocko return -EINTR; 2971ae798783SMichal Hocko 2972408579cdSLiam R. Howlett ret = do_vmi_munmap(&vmi, mm, start, len, &uf, unlock); 2973408579cdSLiam R. Howlett if (ret || !unlock) 2974d8ed45c5SMichel Lespinasse mmap_write_unlock(mm); 2975dd2283f2SYang Shi 2976897ab3e0SMike Rapoport userfaultfd_unmap_complete(mm, &uf); 2977a46ef99dSLinus Torvalds return ret; 2978a46ef99dSLinus Torvalds } 2979dd2283f2SYang Shi 2980dd2283f2SYang Shi int vm_munmap(unsigned long start, size_t len) 2981dd2283f2SYang Shi { 2982dd2283f2SYang Shi return __vm_munmap(start, len, false); 2983dd2283f2SYang Shi } 2984a46ef99dSLinus Torvalds EXPORT_SYMBOL(vm_munmap); 2985a46ef99dSLinus Torvalds 29866a6160a7SHeiko Carstens SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len) 29871da177e4SLinus Torvalds { 2988ce18d171SCatalin Marinas addr = untagged_addr(addr); 2989dd2283f2SYang Shi return __vm_munmap(addr, len, true); 29901da177e4SLinus Torvalds } 29911da177e4SLinus Torvalds 2992c8d78c18SKirill A. Shutemov 2993c8d78c18SKirill A. Shutemov /* 2994c8d78c18SKirill A. Shutemov * Emulation of deprecated remap_file_pages() syscall. 2995c8d78c18SKirill A. Shutemov */ 2996c8d78c18SKirill A. Shutemov SYSCALL_DEFINE5(remap_file_pages, unsigned long, start, unsigned long, size, 2997c8d78c18SKirill A. Shutemov unsigned long, prot, unsigned long, pgoff, unsigned long, flags) 2998c8d78c18SKirill A. Shutemov { 2999c8d78c18SKirill A. Shutemov 3000c8d78c18SKirill A. Shutemov struct mm_struct *mm = current->mm; 3001c8d78c18SKirill A. Shutemov struct vm_area_struct *vma; 3002c8d78c18SKirill A. Shutemov unsigned long populate = 0; 3003c8d78c18SKirill A. Shutemov unsigned long ret = -EINVAL; 3004c8d78c18SKirill A. Shutemov struct file *file; 3005c8d78c18SKirill A. Shutemov 3006ee65728eSMike Rapoport pr_warn_once("%s (%d) uses deprecated remap_file_pages() syscall. See Documentation/mm/remap_file_pages.rst.\n", 3007c8d78c18SKirill A. Shutemov current->comm, current->pid); 3008c8d78c18SKirill A. Shutemov 3009c8d78c18SKirill A. Shutemov if (prot) 3010c8d78c18SKirill A. Shutemov return ret; 3011c8d78c18SKirill A. Shutemov start = start & PAGE_MASK; 3012c8d78c18SKirill A. Shutemov size = size & PAGE_MASK; 3013c8d78c18SKirill A. Shutemov 3014c8d78c18SKirill A. Shutemov if (start + size <= start) 3015c8d78c18SKirill A. Shutemov return ret; 3016c8d78c18SKirill A. Shutemov 3017c8d78c18SKirill A. Shutemov /* Does pgoff wrap? */ 3018c8d78c18SKirill A. Shutemov if (pgoff + (size >> PAGE_SHIFT) < pgoff) 3019c8d78c18SKirill A. Shutemov return ret; 3020c8d78c18SKirill A. Shutemov 3021d8ed45c5SMichel Lespinasse if (mmap_write_lock_killable(mm)) 3022dc0ef0dfSMichal Hocko return -EINTR; 3023dc0ef0dfSMichal Hocko 30249b593cb2SLiam R. Howlett vma = vma_lookup(mm, start); 3025c8d78c18SKirill A. Shutemov 3026c8d78c18SKirill A. Shutemov if (!vma || !(vma->vm_flags & VM_SHARED)) 3027c8d78c18SKirill A. Shutemov goto out; 3028c8d78c18SKirill A. Shutemov 302948f7df32SKirill A. Shutemov if (start + size > vma->vm_end) { 3030763ecb03SLiam R. Howlett VMA_ITERATOR(vmi, mm, vma->vm_end); 3031763ecb03SLiam R. Howlett struct vm_area_struct *next, *prev = vma; 303248f7df32SKirill A. Shutemov 3033763ecb03SLiam R. Howlett for_each_vma_range(vmi, next, start + size) { 303448f7df32SKirill A. Shutemov /* hole between vmas ? */ 3035763ecb03SLiam R. Howlett if (next->vm_start != prev->vm_end) 303648f7df32SKirill A. Shutemov goto out; 303748f7df32SKirill A. Shutemov 303848f7df32SKirill A. Shutemov if (next->vm_file != vma->vm_file) 303948f7df32SKirill A. Shutemov goto out; 304048f7df32SKirill A. Shutemov 304148f7df32SKirill A. Shutemov if (next->vm_flags != vma->vm_flags) 304248f7df32SKirill A. Shutemov goto out; 304348f7df32SKirill A. Shutemov 30441db43d3fSLiam Howlett if (start + size <= next->vm_end) 30451db43d3fSLiam Howlett break; 30461db43d3fSLiam Howlett 3047763ecb03SLiam R. Howlett prev = next; 304848f7df32SKirill A. Shutemov } 304948f7df32SKirill A. Shutemov 305048f7df32SKirill A. Shutemov if (!next) 3051c8d78c18SKirill A. Shutemov goto out; 3052c8d78c18SKirill A. Shutemov } 3053c8d78c18SKirill A. Shutemov 3054c8d78c18SKirill A. Shutemov prot |= vma->vm_flags & VM_READ ? PROT_READ : 0; 3055c8d78c18SKirill A. Shutemov prot |= vma->vm_flags & VM_WRITE ? PROT_WRITE : 0; 3056c8d78c18SKirill A. Shutemov prot |= vma->vm_flags & VM_EXEC ? PROT_EXEC : 0; 3057c8d78c18SKirill A. Shutemov 3058c8d78c18SKirill A. Shutemov flags &= MAP_NONBLOCK; 3059c8d78c18SKirill A. Shutemov flags |= MAP_SHARED | MAP_FIXED | MAP_POPULATE; 3060fce000b1SLiam Howlett if (vma->vm_flags & VM_LOCKED) 3061c8d78c18SKirill A. Shutemov flags |= MAP_LOCKED; 306248f7df32SKirill A. Shutemov 3063c8d78c18SKirill A. Shutemov file = get_file(vma->vm_file); 306445e55300SPeter Collingbourne ret = do_mmap(vma->vm_file, start, size, 3065592b5fadSYu-cheng Yu prot, flags, 0, pgoff, &populate, NULL); 3066c8d78c18SKirill A. Shutemov fput(file); 3067c8d78c18SKirill A. Shutemov out: 3068d8ed45c5SMichel Lespinasse mmap_write_unlock(mm); 3069c8d78c18SKirill A. Shutemov if (populate) 3070c8d78c18SKirill A. Shutemov mm_populate(ret, populate); 3071c8d78c18SKirill A. Shutemov if (!IS_ERR_VALUE(ret)) 3072c8d78c18SKirill A. Shutemov ret = 0; 3073c8d78c18SKirill A. Shutemov return ret; 3074c8d78c18SKirill A. Shutemov } 3075c8d78c18SKirill A. Shutemov 30761da177e4SLinus Torvalds /* 307727b26701SLiam R. Howlett * do_vma_munmap() - Unmap a full or partial vma. 307827b26701SLiam R. Howlett * @vmi: The vma iterator pointing at the vma 307927b26701SLiam R. Howlett * @vma: The first vma to be munmapped 308027b26701SLiam R. Howlett * @start: the start of the address to unmap 308127b26701SLiam R. Howlett * @end: The end of the address to unmap 30822e7ce7d3SLiam R. Howlett * @uf: The userfaultfd list_head 3083408579cdSLiam R. Howlett * @unlock: Drop the lock on success 30842e7ce7d3SLiam R. Howlett * 308527b26701SLiam R. Howlett * unmaps a VMA mapping when the vma iterator is already in position. 308627b26701SLiam R. Howlett * Does not handle alignment. 3087408579cdSLiam R. Howlett * 3088408579cdSLiam R. Howlett * Return: 0 on success drops the lock of so directed, error on failure and will 3089408579cdSLiam R. Howlett * still hold the lock. 30901da177e4SLinus Torvalds */ 309127b26701SLiam R. Howlett int do_vma_munmap(struct vma_iterator *vmi, struct vm_area_struct *vma, 3092408579cdSLiam R. Howlett unsigned long start, unsigned long end, struct list_head *uf, 3093408579cdSLiam R. Howlett bool unlock) 30942e7ce7d3SLiam R. Howlett { 30952e7ce7d3SLiam R. Howlett struct mm_struct *mm = vma->vm_mm; 30962e7ce7d3SLiam R. Howlett 309727b26701SLiam R. Howlett arch_unmap(mm, start, end); 3098b5641a5dSLinus Torvalds return do_vmi_align_munmap(vmi, vma, mm, start, end, uf, unlock); 30992e7ce7d3SLiam R. Howlett } 31002e7ce7d3SLiam R. Howlett 31012e7ce7d3SLiam R. Howlett /* 31022e7ce7d3SLiam R. Howlett * do_brk_flags() - Increase the brk vma if the flags match. 310392fed820SLiam R. Howlett * @vmi: The vma iterator 31042e7ce7d3SLiam R. Howlett * @addr: The start address 31052e7ce7d3SLiam R. Howlett * @len: The length of the increase 31062e7ce7d3SLiam R. Howlett * @vma: The vma, 31072e7ce7d3SLiam R. Howlett * @flags: The VMA Flags 31082e7ce7d3SLiam R. Howlett * 31092e7ce7d3SLiam R. Howlett * Extend the brk VMA from addr to addr + len. If the VMA is NULL or the flags 31102e7ce7d3SLiam R. Howlett * do not match then create a new anonymous VMA. Eventually we may be able to 31112e7ce7d3SLiam R. Howlett * do some brk-specific accounting here. 31122e7ce7d3SLiam R. Howlett */ 311392fed820SLiam R. Howlett static int do_brk_flags(struct vma_iterator *vmi, struct vm_area_struct *vma, 3114763ecb03SLiam R. Howlett unsigned long addr, unsigned long len, unsigned long flags) 31151da177e4SLinus Torvalds { 31161da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 3117287051b1SLiam R. Howlett struct vma_prepare vp; 31182e7ce7d3SLiam R. Howlett 31192e7ce7d3SLiam R. Howlett /* 31202e7ce7d3SLiam R. Howlett * Check against address space limits by the changed size 31212e7ce7d3SLiam R. Howlett * Note: This happens *after* clearing old mappings in some code paths. 31222e7ce7d3SLiam R. Howlett */ 312316e72e9bSDenys Vlasenko flags |= VM_DATA_DEFAULT_FLAGS | VM_ACCOUNT | mm->def_flags; 312484638335SKonstantin Khlebnikov if (!may_expand_vm(mm, flags, len >> PAGE_SHIFT)) 31251da177e4SLinus Torvalds return -ENOMEM; 31261da177e4SLinus Torvalds 31271da177e4SLinus Torvalds if (mm->map_count > sysctl_max_map_count) 31281da177e4SLinus Torvalds return -ENOMEM; 31291da177e4SLinus Torvalds 3130191c5424SAl Viro if (security_vm_enough_memory_mm(mm, len >> PAGE_SHIFT)) 31311da177e4SLinus Torvalds return -ENOMEM; 31321da177e4SLinus Torvalds 31331da177e4SLinus Torvalds /* 31342e7ce7d3SLiam R. Howlett * Expand the existing vma if possible; Note that singular lists do not 31352e7ce7d3SLiam R. Howlett * occur after forking, so the expand will only happen on new VMAs. 31361da177e4SLinus Torvalds */ 31376c28ca64SLiam Howlett if (vma && vma->vm_end == addr && !vma_policy(vma) && 31386c28ca64SLiam Howlett can_vma_merge_after(vma, flags, NULL, NULL, 31396c28ca64SLiam Howlett addr >> PAGE_SHIFT, NULL_VM_UFFD_CTX, NULL)) { 3140b5df0922SLiam R. Howlett vma_iter_config(vmi, vma->vm_start, addr + len); 3141b5df0922SLiam R. Howlett if (vma_iter_prealloc(vmi, vma)) 3142675eaca1SAlistair Popple goto unacct_fail; 314328c5609fSLiam Howlett 3144c9d6e982SSuren Baghdasaryan vma_start_write(vma); 3145c9d6e982SSuren Baghdasaryan 3146287051b1SLiam R. Howlett init_vma_prep(&vp, vma); 3147287051b1SLiam R. Howlett vma_prepare(&vp); 3148ccf1d78dSSuren Baghdasaryan vma_adjust_trans_huge(vma, vma->vm_start, addr + len, 0); 31492e7ce7d3SLiam R. Howlett vma->vm_end = addr + len; 31501c71222eSSuren Baghdasaryan vm_flags_set(vma, VM_SOFTDIRTY); 315192fed820SLiam R. Howlett vma_iter_store(vmi, vma); 31522e7ce7d3SLiam R. Howlett 3153287051b1SLiam R. Howlett vma_complete(&vp, vmi, mm); 31542e7ce7d3SLiam R. Howlett khugepaged_enter_vma(vma, flags); 31552e7ce7d3SLiam R. Howlett goto out; 31562e7ce7d3SLiam R. Howlett } 31572e7ce7d3SLiam R. Howlett 3158b5df0922SLiam R. Howlett if (vma) 3159b5df0922SLiam R. Howlett vma_iter_next_range(vmi); 31602e7ce7d3SLiam R. Howlett /* create a vma struct for an anonymous mapping */ 31612e7ce7d3SLiam R. Howlett vma = vm_area_alloc(mm); 31622e7ce7d3SLiam R. Howlett if (!vma) 3163675eaca1SAlistair Popple goto unacct_fail; 31641da177e4SLinus Torvalds 3165bfd40eafSKirill A. Shutemov vma_set_anonymous(vma); 3166412c6ef9SYajun Deng vma_set_range(vma, addr, addr + len, addr >> PAGE_SHIFT); 31671c71222eSSuren Baghdasaryan vm_flags_init(vma, flags); 31683ed75eb8SColy Li vma->vm_page_prot = vm_get_page_prot(flags); 3169ad9f0063SSuren Baghdasaryan vma_start_write(vma); 317092fed820SLiam R. Howlett if (vma_iter_store_gfp(vmi, vma, GFP_KERNEL)) 31712e7ce7d3SLiam R. Howlett goto mas_store_fail; 3172d4af56c5SLiam R. Howlett 31732e7ce7d3SLiam R. Howlett mm->map_count++; 31742574d5e4SLiam R. Howlett validate_mm(mm); 3175d7597f59SStefan Roesch ksm_add_vma(vma); 31761da177e4SLinus Torvalds out: 31773af9e859SEric B Munson perf_event_mmap(vma); 31781da177e4SLinus Torvalds mm->total_vm += len >> PAGE_SHIFT; 317984638335SKonstantin Khlebnikov mm->data_vm += len >> PAGE_SHIFT; 3180128557ffSMichel Lespinasse if (flags & VM_LOCKED) 3181ba470de4SRik van Riel mm->locked_vm += (len >> PAGE_SHIFT); 31821c71222eSSuren Baghdasaryan vm_flags_set(vma, VM_SOFTDIRTY); 31835d22fc25SLinus Torvalds return 0; 3184d4af56c5SLiam R. Howlett 31852e7ce7d3SLiam R. Howlett mas_store_fail: 3186d4af56c5SLiam R. Howlett vm_area_free(vma); 3187675eaca1SAlistair Popple unacct_fail: 31882e7ce7d3SLiam R. Howlett vm_unacct_memory(len >> PAGE_SHIFT); 31892e7ce7d3SLiam R. Howlett return -ENOMEM; 31901da177e4SLinus Torvalds } 31911da177e4SLinus Torvalds 3192bb177a73SMichal Hocko int vm_brk_flags(unsigned long addr, unsigned long request, unsigned long flags) 3193e4eb1ff6SLinus Torvalds { 3194e4eb1ff6SLinus Torvalds struct mm_struct *mm = current->mm; 31952e7ce7d3SLiam R. Howlett struct vm_area_struct *vma = NULL; 3196bb177a73SMichal Hocko unsigned long len; 31975d22fc25SLinus Torvalds int ret; 3198128557ffSMichel Lespinasse bool populate; 3199897ab3e0SMike Rapoport LIST_HEAD(uf); 320092fed820SLiam R. Howlett VMA_ITERATOR(vmi, mm, addr); 3201e4eb1ff6SLinus Torvalds 3202bb177a73SMichal Hocko len = PAGE_ALIGN(request); 3203bb177a73SMichal Hocko if (len < request) 3204bb177a73SMichal Hocko return -ENOMEM; 3205bb177a73SMichal Hocko if (!len) 3206bb177a73SMichal Hocko return 0; 3207bb177a73SMichal Hocko 32082e7ce7d3SLiam R. Howlett /* Until we need other flags, refuse anything except VM_EXEC. */ 32092e7ce7d3SLiam R. Howlett if ((flags & (~VM_EXEC)) != 0) 32102e7ce7d3SLiam R. Howlett return -EINVAL; 32112e7ce7d3SLiam R. Howlett 3212e0f81ab1SSebastian Ott if (mmap_write_lock_killable(mm)) 3213e0f81ab1SSebastian Ott return -EINTR; 3214e0f81ab1SSebastian Ott 32152e7ce7d3SLiam R. Howlett ret = check_brk_limits(addr, len); 32162e7ce7d3SLiam R. Howlett if (ret) 32172e7ce7d3SLiam R. Howlett goto limits_failed; 32182e7ce7d3SLiam R. Howlett 3219183654ceSLiam R. Howlett ret = do_vmi_munmap(&vmi, mm, addr, len, &uf, 0); 32202e7ce7d3SLiam R. Howlett if (ret) 32212e7ce7d3SLiam R. Howlett goto munmap_failed; 32222e7ce7d3SLiam R. Howlett 322392fed820SLiam R. Howlett vma = vma_prev(&vmi); 322492fed820SLiam R. Howlett ret = do_brk_flags(&vmi, vma, addr, len, flags); 3225128557ffSMichel Lespinasse populate = ((mm->def_flags & VM_LOCKED) != 0); 3226d8ed45c5SMichel Lespinasse mmap_write_unlock(mm); 3227897ab3e0SMike Rapoport userfaultfd_unmap_complete(mm, &uf); 32285d22fc25SLinus Torvalds if (populate && !ret) 3229128557ffSMichel Lespinasse mm_populate(addr, len); 3230e4eb1ff6SLinus Torvalds return ret; 32312e7ce7d3SLiam R. Howlett 32322e7ce7d3SLiam R. Howlett munmap_failed: 32332e7ce7d3SLiam R. Howlett limits_failed: 32342e7ce7d3SLiam R. Howlett mmap_write_unlock(mm); 32352e7ce7d3SLiam R. Howlett return ret; 3236e4eb1ff6SLinus Torvalds } 323716e72e9bSDenys Vlasenko EXPORT_SYMBOL(vm_brk_flags); 323816e72e9bSDenys Vlasenko 32391da177e4SLinus Torvalds /* Release all mmaps. */ 32401da177e4SLinus Torvalds void exit_mmap(struct mm_struct *mm) 32411da177e4SLinus Torvalds { 3242d16dfc55SPeter Zijlstra struct mmu_gather tlb; 3243ba470de4SRik van Riel struct vm_area_struct *vma; 32441da177e4SLinus Torvalds unsigned long nr_accounted = 0; 3245763ecb03SLiam R. Howlett MA_STATE(mas, &mm->mm_mt, 0, 0); 3246763ecb03SLiam R. Howlett int count = 0; 32471da177e4SLinus Torvalds 3248d6dd61c8SJeremy Fitzhardinge /* mm's last user has gone, and its about to be pulled down */ 3249cddb8a5cSAndrea Arcangeli mmu_notifier_release(mm); 3250d6dd61c8SJeremy Fitzhardinge 3251bf3980c8SSuren Baghdasaryan mmap_read_lock(mm); 32529480c53eSJeremy Fitzhardinge arch_exit_mmap(mm); 32539480c53eSJeremy Fitzhardinge 3254763ecb03SLiam R. Howlett vma = mas_find(&mas, ULONG_MAX); 3255d2406291SPeng Zhang if (!vma || unlikely(xa_is_zero(vma))) { 325664591e86SSuren Baghdasaryan /* Can happen if dup_mmap() received an OOM */ 3257bf3980c8SSuren Baghdasaryan mmap_read_unlock(mm); 3258d2406291SPeng Zhang mmap_write_lock(mm); 3259d2406291SPeng Zhang goto destroy; 326064591e86SSuren Baghdasaryan } 32619480c53eSJeremy Fitzhardinge 32621da177e4SLinus Torvalds lru_add_drain(); 32631da177e4SLinus Torvalds flush_cache_mm(mm); 3264d8b45053SWill Deacon tlb_gather_mmu_fullmm(&tlb, mm); 3265901608d9SOleg Nesterov /* update_hiwater_rss(mm) here? but nobody should be looking */ 3266763ecb03SLiam R. Howlett /* Use ULONG_MAX here to ensure all VMAs in the mm are unmapped */ 3267fd892593SLiam R. Howlett unmap_vmas(&tlb, &mas, vma, 0, ULONG_MAX, ULONG_MAX, false); 3268bf3980c8SSuren Baghdasaryan mmap_read_unlock(mm); 3269bf3980c8SSuren Baghdasaryan 3270bf3980c8SSuren Baghdasaryan /* 3271bf3980c8SSuren Baghdasaryan * Set MMF_OOM_SKIP to hide this task from the oom killer/reaper 3272b3541d91SSuren Baghdasaryan * because the memory has been already freed. 3273bf3980c8SSuren Baghdasaryan */ 3274bf3980c8SSuren Baghdasaryan set_bit(MMF_OOM_SKIP, &mm->flags); 3275bf3980c8SSuren Baghdasaryan mmap_write_lock(mm); 32763dd44325SLiam R. Howlett mt_clear_in_rcu(&mm->mm_mt); 3277fd892593SLiam R. Howlett mas_set(&mas, vma->vm_end); 3278fd892593SLiam R. Howlett free_pgtables(&tlb, &mas, vma, FIRST_USER_ADDRESS, 327998e51a22SSuren Baghdasaryan USER_PGTABLES_CEILING, true); 3280ae8eba8bSWill Deacon tlb_finish_mmu(&tlb); 32811da177e4SLinus Torvalds 3282763ecb03SLiam R. Howlett /* 3283763ecb03SLiam R. Howlett * Walk the list again, actually closing and freeing it, with preemption 3284763ecb03SLiam R. Howlett * enabled, without holding any MM locks besides the unreachable 3285763ecb03SLiam R. Howlett * mmap_write_lock. 3286763ecb03SLiam R. Howlett */ 3287fd892593SLiam R. Howlett mas_set(&mas, vma->vm_end); 3288763ecb03SLiam R. Howlett do { 32894f74d2c8SLinus Torvalds if (vma->vm_flags & VM_ACCOUNT) 32904f74d2c8SLinus Torvalds nr_accounted += vma_pages(vma); 32910d2ebf9cSSuren Baghdasaryan remove_vma(vma, true); 3292763ecb03SLiam R. Howlett count++; 32930a3b3c25SPaul E. McKenney cond_resched(); 3294d2406291SPeng Zhang vma = mas_find(&mas, ULONG_MAX); 3295d2406291SPeng Zhang } while (vma && likely(!xa_is_zero(vma))); 3296763ecb03SLiam R. Howlett 3297763ecb03SLiam R. Howlett BUG_ON(count != mm->map_count); 3298d4af56c5SLiam R. Howlett 3299d4af56c5SLiam R. Howlett trace_exit_mmap(mm); 3300d2406291SPeng Zhang destroy: 3301d4af56c5SLiam R. Howlett __mt_destroy(&mm->mm_mt); 330264591e86SSuren Baghdasaryan mmap_write_unlock(mm); 33034f74d2c8SLinus Torvalds vm_unacct_memory(nr_accounted); 33041da177e4SLinus Torvalds } 33051da177e4SLinus Torvalds 33061da177e4SLinus Torvalds /* Insert vm structure into process list sorted by address 33071da177e4SLinus Torvalds * and into the inode's i_mmap tree. If vm_file is non-NULL 3308c8c06efaSDavidlohr Bueso * then i_mmap_rwsem is taken here. 33091da177e4SLinus Torvalds */ 33101da177e4SLinus Torvalds int insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma) 33111da177e4SLinus Torvalds { 3312d4af56c5SLiam R. Howlett unsigned long charged = vma_pages(vma); 33131da177e4SLinus Torvalds 3314d4af56c5SLiam R. Howlett 3315d0601a50SLiam R. Howlett if (find_vma_intersection(mm, vma->vm_start, vma->vm_end)) 3316c9d13f5fSChen Gang return -ENOMEM; 3317d4af56c5SLiam R. Howlett 3318c9d13f5fSChen Gang if ((vma->vm_flags & VM_ACCOUNT) && 3319d4af56c5SLiam R. Howlett security_vm_enough_memory_mm(mm, charged)) 3320c9d13f5fSChen Gang return -ENOMEM; 3321c9d13f5fSChen Gang 33221da177e4SLinus Torvalds /* 33231da177e4SLinus Torvalds * The vm_pgoff of a purely anonymous vma should be irrelevant 33241da177e4SLinus Torvalds * until its first write fault, when page's anon_vma and index 33251da177e4SLinus Torvalds * are set. But now set the vm_pgoff it will almost certainly 33261da177e4SLinus Torvalds * end up with (unless mremap moves it elsewhere before that 33271da177e4SLinus Torvalds * first wfault), so /proc/pid/maps tells a consistent story. 33281da177e4SLinus Torvalds * 33291da177e4SLinus Torvalds * By setting it to reflect the virtual start address of the 33301da177e4SLinus Torvalds * vma, merges and splits can happen in a seamless way, just 33311da177e4SLinus Torvalds * using the existing file pgoff checks and manipulations. 33328332326eSLiao Pingfang * Similarly in do_mmap and in do_brk_flags. 33331da177e4SLinus Torvalds */ 33348a9cc3b5SOleg Nesterov if (vma_is_anonymous(vma)) { 33351da177e4SLinus Torvalds BUG_ON(vma->anon_vma); 33361da177e4SLinus Torvalds vma->vm_pgoff = vma->vm_start >> PAGE_SHIFT; 33371da177e4SLinus Torvalds } 33382b144498SSrikar Dronamraju 3339763ecb03SLiam R. Howlett if (vma_link(mm, vma)) { 3340dd34d9feSAnthony Yznaga if (vma->vm_flags & VM_ACCOUNT) 3341d4af56c5SLiam R. Howlett vm_unacct_memory(charged); 3342d4af56c5SLiam R. Howlett return -ENOMEM; 3343d4af56c5SLiam R. Howlett } 3344d4af56c5SLiam R. Howlett 33451da177e4SLinus Torvalds return 0; 33461da177e4SLinus Torvalds } 33471da177e4SLinus Torvalds 33481da177e4SLinus Torvalds /* 33491da177e4SLinus Torvalds * Copy the vma structure to a new location in the same mm, 33501da177e4SLinus Torvalds * prior to moving page table entries, to effect an mremap move. 33511da177e4SLinus Torvalds */ 33521da177e4SLinus Torvalds struct vm_area_struct *copy_vma(struct vm_area_struct **vmap, 335338a76013SMichel Lespinasse unsigned long addr, unsigned long len, pgoff_t pgoff, 335438a76013SMichel Lespinasse bool *need_rmap_locks) 33551da177e4SLinus Torvalds { 33561da177e4SLinus Torvalds struct vm_area_struct *vma = *vmap; 33571da177e4SLinus Torvalds unsigned long vma_start = vma->vm_start; 33581da177e4SLinus Torvalds struct mm_struct *mm = vma->vm_mm; 33591da177e4SLinus Torvalds struct vm_area_struct *new_vma, *prev; 3360948f017bSAndrea Arcangeli bool faulted_in_anon_vma = true; 3361076f16bfSLiam R. Howlett VMA_ITERATOR(vmi, mm, addr); 33621da177e4SLinus Torvalds 33631da177e4SLinus Torvalds /* 33641da177e4SLinus Torvalds * If anonymous vma has not yet been faulted, update new pgoff 33651da177e4SLinus Torvalds * to match new location, to increase its chance of merging. 33661da177e4SLinus Torvalds */ 3367ce75799bSOleg Nesterov if (unlikely(vma_is_anonymous(vma) && !vma->anon_vma)) { 33681da177e4SLinus Torvalds pgoff = addr >> PAGE_SHIFT; 3369948f017bSAndrea Arcangeli faulted_in_anon_vma = false; 3370948f017bSAndrea Arcangeli } 33711da177e4SLinus Torvalds 3372763ecb03SLiam R. Howlett new_vma = find_vma_prev(mm, addr, &prev); 3373763ecb03SLiam R. Howlett if (new_vma && new_vma->vm_start < addr + len) 33746597d783SHugh Dickins return NULL; /* should never get here */ 3375524e00b3SLiam R. Howlett 33764b5f2d20SLorenzo Stoakes new_vma = vma_merge_new_vma(&vmi, prev, vma, addr, addr + len, pgoff); 33771da177e4SLinus Torvalds if (new_vma) { 33781da177e4SLinus Torvalds /* 33791da177e4SLinus Torvalds * Source vma may have been merged into new_vma 33801da177e4SLinus Torvalds */ 3381948f017bSAndrea Arcangeli if (unlikely(vma_start >= new_vma->vm_start && 3382948f017bSAndrea Arcangeli vma_start < new_vma->vm_end)) { 3383948f017bSAndrea Arcangeli /* 3384948f017bSAndrea Arcangeli * The only way we can get a vma_merge with 3385948f017bSAndrea Arcangeli * self during an mremap is if the vma hasn't 3386948f017bSAndrea Arcangeli * been faulted in yet and we were allowed to 3387948f017bSAndrea Arcangeli * reset the dst vma->vm_pgoff to the 3388948f017bSAndrea Arcangeli * destination address of the mremap to allow 3389948f017bSAndrea Arcangeli * the merge to happen. mremap must change the 3390948f017bSAndrea Arcangeli * vm_pgoff linearity between src and dst vmas 3391948f017bSAndrea Arcangeli * (in turn preventing a vma_merge) to be 3392948f017bSAndrea Arcangeli * safe. It is only safe to keep the vm_pgoff 3393948f017bSAndrea Arcangeli * linear if there are no pages mapped yet. 3394948f017bSAndrea Arcangeli */ 339581d1b09cSSasha Levin VM_BUG_ON_VMA(faulted_in_anon_vma, new_vma); 339638a76013SMichel Lespinasse *vmap = vma = new_vma; 3397108d6642SMichel Lespinasse } 339838a76013SMichel Lespinasse *need_rmap_locks = (new_vma->vm_pgoff <= vma->vm_pgoff); 33991da177e4SLinus Torvalds } else { 34003928d4f5SLinus Torvalds new_vma = vm_area_dup(vma); 3401e3975891SChen Gang if (!new_vma) 3402e3975891SChen Gang goto out; 3403412c6ef9SYajun Deng vma_set_range(new_vma, addr, addr + len, pgoff); 3404ef0855d3SOleg Nesterov if (vma_dup_policy(vma, new_vma)) 3405523d4e20SMichel Lespinasse goto out_free_vma; 3406523d4e20SMichel Lespinasse if (anon_vma_clone(new_vma, vma)) 3407523d4e20SMichel Lespinasse goto out_free_mempol; 3408e9714acfSKonstantin Khlebnikov if (new_vma->vm_file) 34091da177e4SLinus Torvalds get_file(new_vma->vm_file); 34101da177e4SLinus Torvalds if (new_vma->vm_ops && new_vma->vm_ops->open) 34111da177e4SLinus Torvalds new_vma->vm_ops->open(new_vma); 3412763ecb03SLiam R. Howlett if (vma_link(mm, new_vma)) 3413524e00b3SLiam R. Howlett goto out_vma_link; 341438a76013SMichel Lespinasse *need_rmap_locks = false; 34151da177e4SLinus Torvalds } 34161da177e4SLinus Torvalds return new_vma; 34175beb4930SRik van Riel 3418524e00b3SLiam R. Howlett out_vma_link: 3419524e00b3SLiam R. Howlett if (new_vma->vm_ops && new_vma->vm_ops->close) 3420524e00b3SLiam R. Howlett new_vma->vm_ops->close(new_vma); 342192b73996SLiam Howlett 342292b73996SLiam Howlett if (new_vma->vm_file) 342392b73996SLiam Howlett fput(new_vma->vm_file); 342492b73996SLiam Howlett 342592b73996SLiam Howlett unlink_anon_vmas(new_vma); 34265beb4930SRik van Riel out_free_mempol: 3427ef0855d3SOleg Nesterov mpol_put(vma_policy(new_vma)); 34285beb4930SRik van Riel out_free_vma: 34293928d4f5SLinus Torvalds vm_area_free(new_vma); 3430e3975891SChen Gang out: 34315beb4930SRik van Riel return NULL; 34321da177e4SLinus Torvalds } 3433119f657cSakpm@osdl.org 3434119f657cSakpm@osdl.org /* 3435119f657cSakpm@osdl.org * Return true if the calling process may expand its vm space by the passed 3436119f657cSakpm@osdl.org * number of pages 3437119f657cSakpm@osdl.org */ 343884638335SKonstantin Khlebnikov bool may_expand_vm(struct mm_struct *mm, vm_flags_t flags, unsigned long npages) 3439119f657cSakpm@osdl.org { 344084638335SKonstantin Khlebnikov if (mm->total_vm + npages > rlimit(RLIMIT_AS) >> PAGE_SHIFT) 344184638335SKonstantin Khlebnikov return false; 3442119f657cSakpm@osdl.org 3443d977d56cSKonstantin Khlebnikov if (is_data_mapping(flags) && 3444d977d56cSKonstantin Khlebnikov mm->data_vm + npages > rlimit(RLIMIT_DATA) >> PAGE_SHIFT) { 3445f4fcd558SKonstantin Khlebnikov /* Workaround for Valgrind */ 3446f4fcd558SKonstantin Khlebnikov if (rlimit(RLIMIT_DATA) == 0 && 3447f4fcd558SKonstantin Khlebnikov mm->data_vm + npages <= rlimit_max(RLIMIT_DATA) >> PAGE_SHIFT) 3448f4fcd558SKonstantin Khlebnikov return true; 344957a7702bSDavid Woodhouse 345057a7702bSDavid Woodhouse pr_warn_once("%s (%d): VmData %lu exceed data ulimit %lu. Update limits%s.\n", 3451d977d56cSKonstantin Khlebnikov current->comm, current->pid, 3452d977d56cSKonstantin Khlebnikov (mm->data_vm + npages) << PAGE_SHIFT, 345357a7702bSDavid Woodhouse rlimit(RLIMIT_DATA), 345457a7702bSDavid Woodhouse ignore_rlimit_data ? "" : " or use boot option ignore_rlimit_data"); 345557a7702bSDavid Woodhouse 345657a7702bSDavid Woodhouse if (!ignore_rlimit_data) 3457d977d56cSKonstantin Khlebnikov return false; 3458d977d56cSKonstantin Khlebnikov } 3459119f657cSakpm@osdl.org 346084638335SKonstantin Khlebnikov return true; 346184638335SKonstantin Khlebnikov } 346284638335SKonstantin Khlebnikov 346384638335SKonstantin Khlebnikov void vm_stat_account(struct mm_struct *mm, vm_flags_t flags, long npages) 346484638335SKonstantin Khlebnikov { 34657866076bSPeng Liu WRITE_ONCE(mm->total_vm, READ_ONCE(mm->total_vm)+npages); 346684638335SKonstantin Khlebnikov 3467d977d56cSKonstantin Khlebnikov if (is_exec_mapping(flags)) 346884638335SKonstantin Khlebnikov mm->exec_vm += npages; 3469d977d56cSKonstantin Khlebnikov else if (is_stack_mapping(flags)) 347084638335SKonstantin Khlebnikov mm->stack_vm += npages; 3471d977d56cSKonstantin Khlebnikov else if (is_data_mapping(flags)) 347284638335SKonstantin Khlebnikov mm->data_vm += npages; 3473119f657cSakpm@osdl.org } 3474fa5dc22fSRoland McGrath 3475b3ec9f33SSouptick Joarder static vm_fault_t special_mapping_fault(struct vm_fault *vmf); 3476a62c34bdSAndy Lutomirski 3477a62c34bdSAndy Lutomirski /* 3478a62c34bdSAndy Lutomirski * Having a close hook prevents vma merging regardless of flags. 3479a62c34bdSAndy Lutomirski */ 3480a62c34bdSAndy Lutomirski static void special_mapping_close(struct vm_area_struct *vma) 3481a62c34bdSAndy Lutomirski { 3482a62c34bdSAndy Lutomirski } 3483a62c34bdSAndy Lutomirski 3484a62c34bdSAndy Lutomirski static const char *special_mapping_name(struct vm_area_struct *vma) 3485a62c34bdSAndy Lutomirski { 3486a62c34bdSAndy Lutomirski return ((struct vm_special_mapping *)vma->vm_private_data)->name; 3487a62c34bdSAndy Lutomirski } 3488a62c34bdSAndy Lutomirski 348914d07113SBrian Geffon static int special_mapping_mremap(struct vm_area_struct *new_vma) 3490b059a453SDmitry Safonov { 3491b059a453SDmitry Safonov struct vm_special_mapping *sm = new_vma->vm_private_data; 3492b059a453SDmitry Safonov 3493280e87e9SDmitry Safonov if (WARN_ON_ONCE(current->mm != new_vma->vm_mm)) 3494280e87e9SDmitry Safonov return -EFAULT; 3495280e87e9SDmitry Safonov 3496b059a453SDmitry Safonov if (sm->mremap) 3497b059a453SDmitry Safonov return sm->mremap(sm, new_vma); 3498280e87e9SDmitry Safonov 3499b059a453SDmitry Safonov return 0; 3500b059a453SDmitry Safonov } 3501b059a453SDmitry Safonov 3502871402e0SDmitry Safonov static int special_mapping_split(struct vm_area_struct *vma, unsigned long addr) 3503871402e0SDmitry Safonov { 3504871402e0SDmitry Safonov /* 3505871402e0SDmitry Safonov * Forbid splitting special mappings - kernel has expectations over 3506871402e0SDmitry Safonov * the number of pages in mapping. Together with VM_DONTEXPAND 3507871402e0SDmitry Safonov * the size of vma should stay the same over the special mapping's 3508871402e0SDmitry Safonov * lifetime. 3509871402e0SDmitry Safonov */ 3510871402e0SDmitry Safonov return -EINVAL; 3511871402e0SDmitry Safonov } 3512871402e0SDmitry Safonov 3513a62c34bdSAndy Lutomirski static const struct vm_operations_struct special_mapping_vmops = { 3514a62c34bdSAndy Lutomirski .close = special_mapping_close, 3515a62c34bdSAndy Lutomirski .fault = special_mapping_fault, 3516b059a453SDmitry Safonov .mremap = special_mapping_mremap, 3517a62c34bdSAndy Lutomirski .name = special_mapping_name, 3518af34ebebSDmitry Safonov /* vDSO code relies that VVAR can't be accessed remotely */ 3519af34ebebSDmitry Safonov .access = NULL, 3520871402e0SDmitry Safonov .may_split = special_mapping_split, 3521a62c34bdSAndy Lutomirski }; 3522a62c34bdSAndy Lutomirski 3523a62c34bdSAndy Lutomirski static const struct vm_operations_struct legacy_special_mapping_vmops = { 3524a62c34bdSAndy Lutomirski .close = special_mapping_close, 3525a62c34bdSAndy Lutomirski .fault = special_mapping_fault, 3526a62c34bdSAndy Lutomirski }; 3527fa5dc22fSRoland McGrath 3528b3ec9f33SSouptick Joarder static vm_fault_t special_mapping_fault(struct vm_fault *vmf) 3529fa5dc22fSRoland McGrath { 353011bac800SDave Jiang struct vm_area_struct *vma = vmf->vma; 3531b1d0e4f5SNick Piggin pgoff_t pgoff; 3532fa5dc22fSRoland McGrath struct page **pages; 3533fa5dc22fSRoland McGrath 3534f872f540SAndy Lutomirski if (vma->vm_ops == &legacy_special_mapping_vmops) { 3535a62c34bdSAndy Lutomirski pages = vma->vm_private_data; 3536f872f540SAndy Lutomirski } else { 3537f872f540SAndy Lutomirski struct vm_special_mapping *sm = vma->vm_private_data; 3538f872f540SAndy Lutomirski 3539f872f540SAndy Lutomirski if (sm->fault) 354011bac800SDave Jiang return sm->fault(sm, vmf->vma, vmf); 3541f872f540SAndy Lutomirski 3542f872f540SAndy Lutomirski pages = sm->pages; 3543f872f540SAndy Lutomirski } 3544a62c34bdSAndy Lutomirski 35458a9cc3b5SOleg Nesterov for (pgoff = vmf->pgoff; pgoff && *pages; ++pages) 3546b1d0e4f5SNick Piggin pgoff--; 3547fa5dc22fSRoland McGrath 3548fa5dc22fSRoland McGrath if (*pages) { 3549fa5dc22fSRoland McGrath struct page *page = *pages; 3550fa5dc22fSRoland McGrath get_page(page); 3551b1d0e4f5SNick Piggin vmf->page = page; 3552b1d0e4f5SNick Piggin return 0; 3553fa5dc22fSRoland McGrath } 3554fa5dc22fSRoland McGrath 3555b1d0e4f5SNick Piggin return VM_FAULT_SIGBUS; 3556fa5dc22fSRoland McGrath } 3557fa5dc22fSRoland McGrath 3558a62c34bdSAndy Lutomirski static struct vm_area_struct *__install_special_mapping( 3559a62c34bdSAndy Lutomirski struct mm_struct *mm, 3560fa5dc22fSRoland McGrath unsigned long addr, unsigned long len, 356127f28b97SChen Gang unsigned long vm_flags, void *priv, 356227f28b97SChen Gang const struct vm_operations_struct *ops) 3563fa5dc22fSRoland McGrath { 3564462e635eSTavis Ormandy int ret; 3565fa5dc22fSRoland McGrath struct vm_area_struct *vma; 3566fa5dc22fSRoland McGrath 3567490fc053SLinus Torvalds vma = vm_area_alloc(mm); 3568fa5dc22fSRoland McGrath if (unlikely(vma == NULL)) 35693935ed6aSStefani Seibold return ERR_PTR(-ENOMEM); 3570fa5dc22fSRoland McGrath 3571412c6ef9SYajun Deng vma_set_range(vma, addr, addr + len, 0); 3572e430a95aSSuren Baghdasaryan vm_flags_init(vma, (vm_flags | mm->def_flags | 3573e430a95aSSuren Baghdasaryan VM_DONTEXPAND | VM_SOFTDIRTY) & ~VM_LOCKED_MASK); 35743ed75eb8SColy Li vma->vm_page_prot = vm_get_page_prot(vma->vm_flags); 3575fa5dc22fSRoland McGrath 3576a62c34bdSAndy Lutomirski vma->vm_ops = ops; 3577a62c34bdSAndy Lutomirski vma->vm_private_data = priv; 3578fa5dc22fSRoland McGrath 3579462e635eSTavis Ormandy ret = insert_vm_struct(mm, vma); 3580462e635eSTavis Ormandy if (ret) 3581462e635eSTavis Ormandy goto out; 3582fa5dc22fSRoland McGrath 358384638335SKonstantin Khlebnikov vm_stat_account(mm, vma->vm_flags, len >> PAGE_SHIFT); 3584fa5dc22fSRoland McGrath 3585cdd6c482SIngo Molnar perf_event_mmap(vma); 3586089dd79dSPeter Zijlstra 35873935ed6aSStefani Seibold return vma; 3588462e635eSTavis Ormandy 3589462e635eSTavis Ormandy out: 35903928d4f5SLinus Torvalds vm_area_free(vma); 35913935ed6aSStefani Seibold return ERR_PTR(ret); 35923935ed6aSStefani Seibold } 35933935ed6aSStefani Seibold 35942eefd878SDmitry Safonov bool vma_is_special_mapping(const struct vm_area_struct *vma, 35952eefd878SDmitry Safonov const struct vm_special_mapping *sm) 35962eefd878SDmitry Safonov { 35972eefd878SDmitry Safonov return vma->vm_private_data == sm && 35982eefd878SDmitry Safonov (vma->vm_ops == &special_mapping_vmops || 35992eefd878SDmitry Safonov vma->vm_ops == &legacy_special_mapping_vmops); 36002eefd878SDmitry Safonov } 36012eefd878SDmitry Safonov 3602a62c34bdSAndy Lutomirski /* 3603c1e8d7c6SMichel Lespinasse * Called with mm->mmap_lock held for writing. 3604a62c34bdSAndy Lutomirski * Insert a new vma covering the given region, with the given flags. 3605a62c34bdSAndy Lutomirski * Its pages are supplied by the given array of struct page *. 3606a62c34bdSAndy Lutomirski * The array can be shorter than len >> PAGE_SHIFT if it's null-terminated. 3607a62c34bdSAndy Lutomirski * The region past the last page supplied will always produce SIGBUS. 3608a62c34bdSAndy Lutomirski * The array pointer and the pages it points to are assumed to stay alive 3609a62c34bdSAndy Lutomirski * for as long as this mapping might exist. 3610a62c34bdSAndy Lutomirski */ 3611a62c34bdSAndy Lutomirski struct vm_area_struct *_install_special_mapping( 3612a62c34bdSAndy Lutomirski struct mm_struct *mm, 3613a62c34bdSAndy Lutomirski unsigned long addr, unsigned long len, 3614a62c34bdSAndy Lutomirski unsigned long vm_flags, const struct vm_special_mapping *spec) 3615a62c34bdSAndy Lutomirski { 361627f28b97SChen Gang return __install_special_mapping(mm, addr, len, vm_flags, (void *)spec, 361727f28b97SChen Gang &special_mapping_vmops); 3618a62c34bdSAndy Lutomirski } 3619a62c34bdSAndy Lutomirski 36203935ed6aSStefani Seibold int install_special_mapping(struct mm_struct *mm, 36213935ed6aSStefani Seibold unsigned long addr, unsigned long len, 36223935ed6aSStefani Seibold unsigned long vm_flags, struct page **pages) 36233935ed6aSStefani Seibold { 3624a62c34bdSAndy Lutomirski struct vm_area_struct *vma = __install_special_mapping( 362527f28b97SChen Gang mm, addr, len, vm_flags, (void *)pages, 362627f28b97SChen Gang &legacy_special_mapping_vmops); 36273935ed6aSStefani Seibold 362814bd5b45SDuan Jiong return PTR_ERR_OR_ZERO(vma); 3629fa5dc22fSRoland McGrath } 36307906d00cSAndrea Arcangeli 36317906d00cSAndrea Arcangeli static DEFINE_MUTEX(mm_all_locks_mutex); 36327906d00cSAndrea Arcangeli 3633454ed842SPeter Zijlstra static void vm_lock_anon_vma(struct mm_struct *mm, struct anon_vma *anon_vma) 36347906d00cSAndrea Arcangeli { 3635f808c13fSDavidlohr Bueso if (!test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_root.rb_node)) { 36367906d00cSAndrea Arcangeli /* 36377906d00cSAndrea Arcangeli * The LSB of head.next can't change from under us 36387906d00cSAndrea Arcangeli * because we hold the mm_all_locks_mutex. 36397906d00cSAndrea Arcangeli */ 3640da1c55f1SMichel Lespinasse down_write_nest_lock(&anon_vma->root->rwsem, &mm->mmap_lock); 36417906d00cSAndrea Arcangeli /* 36427906d00cSAndrea Arcangeli * We can safely modify head.next after taking the 36435a505085SIngo Molnar * anon_vma->root->rwsem. If some other vma in this mm shares 36447906d00cSAndrea Arcangeli * the same anon_vma we won't take it again. 36457906d00cSAndrea Arcangeli * 36467906d00cSAndrea Arcangeli * No need of atomic instructions here, head.next 36477906d00cSAndrea Arcangeli * can't change from under us thanks to the 36485a505085SIngo Molnar * anon_vma->root->rwsem. 36497906d00cSAndrea Arcangeli */ 36507906d00cSAndrea Arcangeli if (__test_and_set_bit(0, (unsigned long *) 3651f808c13fSDavidlohr Bueso &anon_vma->root->rb_root.rb_root.rb_node)) 36527906d00cSAndrea Arcangeli BUG(); 36537906d00cSAndrea Arcangeli } 36547906d00cSAndrea Arcangeli } 36557906d00cSAndrea Arcangeli 3656454ed842SPeter Zijlstra static void vm_lock_mapping(struct mm_struct *mm, struct address_space *mapping) 36577906d00cSAndrea Arcangeli { 36587906d00cSAndrea Arcangeli if (!test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) { 36597906d00cSAndrea Arcangeli /* 36607906d00cSAndrea Arcangeli * AS_MM_ALL_LOCKS can't change from under us because 36617906d00cSAndrea Arcangeli * we hold the mm_all_locks_mutex. 36627906d00cSAndrea Arcangeli * 36637906d00cSAndrea Arcangeli * Operations on ->flags have to be atomic because 36647906d00cSAndrea Arcangeli * even if AS_MM_ALL_LOCKS is stable thanks to the 36657906d00cSAndrea Arcangeli * mm_all_locks_mutex, there may be other cpus 36667906d00cSAndrea Arcangeli * changing other bitflags in parallel to us. 36677906d00cSAndrea Arcangeli */ 36687906d00cSAndrea Arcangeli if (test_and_set_bit(AS_MM_ALL_LOCKS, &mapping->flags)) 36697906d00cSAndrea Arcangeli BUG(); 3670da1c55f1SMichel Lespinasse down_write_nest_lock(&mapping->i_mmap_rwsem, &mm->mmap_lock); 36717906d00cSAndrea Arcangeli } 36727906d00cSAndrea Arcangeli } 36737906d00cSAndrea Arcangeli 36747906d00cSAndrea Arcangeli /* 36757906d00cSAndrea Arcangeli * This operation locks against the VM for all pte/vma/mm related 36767906d00cSAndrea Arcangeli * operations that could ever happen on a certain mm. This includes 36777906d00cSAndrea Arcangeli * vmtruncate, try_to_unmap, and all page faults. 36787906d00cSAndrea Arcangeli * 3679c1e8d7c6SMichel Lespinasse * The caller must take the mmap_lock in write mode before calling 36807906d00cSAndrea Arcangeli * mm_take_all_locks(). The caller isn't allowed to release the 3681c1e8d7c6SMichel Lespinasse * mmap_lock until mm_drop_all_locks() returns. 36827906d00cSAndrea Arcangeli * 3683c1e8d7c6SMichel Lespinasse * mmap_lock in write mode is required in order to block all operations 36847906d00cSAndrea Arcangeli * that could modify pagetables and free pages without need of 368527ba0644SKirill A. Shutemov * altering the vma layout. It's also needed in write mode to avoid new 36867906d00cSAndrea Arcangeli * anon_vmas to be associated with existing vmas. 36877906d00cSAndrea Arcangeli * 36887906d00cSAndrea Arcangeli * A single task can't take more than one mm_take_all_locks() in a row 36897906d00cSAndrea Arcangeli * or it would deadlock. 36907906d00cSAndrea Arcangeli * 3691bf181b9fSMichel Lespinasse * The LSB in anon_vma->rb_root.rb_node and the AS_MM_ALL_LOCKS bitflag in 36927906d00cSAndrea Arcangeli * mapping->flags avoid to take the same lock twice, if more than one 36937906d00cSAndrea Arcangeli * vma in this mm is backed by the same anon_vma or address_space. 36947906d00cSAndrea Arcangeli * 369588f306b6SKirill A. Shutemov * We take locks in following order, accordingly to comment at beginning 369688f306b6SKirill A. Shutemov * of mm/rmap.c: 369788f306b6SKirill A. Shutemov * - all hugetlbfs_i_mmap_rwsem_key locks (aka mapping->i_mmap_rwsem for 369888f306b6SKirill A. Shutemov * hugetlb mapping); 3699eeff9a5dSSuren Baghdasaryan * - all vmas marked locked 370088f306b6SKirill A. Shutemov * - all i_mmap_rwsem locks; 370188f306b6SKirill A. Shutemov * - all anon_vma->rwseml 370288f306b6SKirill A. Shutemov * 370388f306b6SKirill A. Shutemov * We can take all locks within these types randomly because the VM code 370488f306b6SKirill A. Shutemov * doesn't nest them and we protected from parallel mm_take_all_locks() by 370588f306b6SKirill A. Shutemov * mm_all_locks_mutex. 37067906d00cSAndrea Arcangeli * 37077906d00cSAndrea Arcangeli * mm_take_all_locks() and mm_drop_all_locks are expensive operations 37087906d00cSAndrea Arcangeli * that may have to take thousand of locks. 37097906d00cSAndrea Arcangeli * 37107906d00cSAndrea Arcangeli * mm_take_all_locks() can fail if it's interrupted by signals. 37117906d00cSAndrea Arcangeli */ 37127906d00cSAndrea Arcangeli int mm_take_all_locks(struct mm_struct *mm) 37137906d00cSAndrea Arcangeli { 37147906d00cSAndrea Arcangeli struct vm_area_struct *vma; 37155beb4930SRik van Riel struct anon_vma_chain *avc; 3716763ecb03SLiam R. Howlett MA_STATE(mas, &mm->mm_mt, 0, 0); 37177906d00cSAndrea Arcangeli 3718325bca1fSRolf Eike Beer mmap_assert_write_locked(mm); 37197906d00cSAndrea Arcangeli 37207906d00cSAndrea Arcangeli mutex_lock(&mm_all_locks_mutex); 37217906d00cSAndrea Arcangeli 372290717566SJann Horn /* 372390717566SJann Horn * vma_start_write() does not have a complement in mm_drop_all_locks() 372490717566SJann Horn * because vma_start_write() is always asymmetrical; it marks a VMA as 372590717566SJann Horn * being written to until mmap_write_unlock() or mmap_write_downgrade() 372690717566SJann Horn * is reached. 372790717566SJann Horn */ 3728763ecb03SLiam R. Howlett mas_for_each(&mas, vma, ULONG_MAX) { 37297906d00cSAndrea Arcangeli if (signal_pending(current)) 37307906d00cSAndrea Arcangeli goto out_unlock; 3731eeff9a5dSSuren Baghdasaryan vma_start_write(vma); 3732eeff9a5dSSuren Baghdasaryan } 3733eeff9a5dSSuren Baghdasaryan 3734eeff9a5dSSuren Baghdasaryan mas_set(&mas, 0); 3735eeff9a5dSSuren Baghdasaryan mas_for_each(&mas, vma, ULONG_MAX) { 3736eeff9a5dSSuren Baghdasaryan if (signal_pending(current)) 3737eeff9a5dSSuren Baghdasaryan goto out_unlock; 373888f306b6SKirill A. Shutemov if (vma->vm_file && vma->vm_file->f_mapping && 373988f306b6SKirill A. Shutemov is_vm_hugetlb_page(vma)) 374088f306b6SKirill A. Shutemov vm_lock_mapping(mm, vma->vm_file->f_mapping); 374188f306b6SKirill A. Shutemov } 374288f306b6SKirill A. Shutemov 3743763ecb03SLiam R. Howlett mas_set(&mas, 0); 3744763ecb03SLiam R. Howlett mas_for_each(&mas, vma, ULONG_MAX) { 374588f306b6SKirill A. Shutemov if (signal_pending(current)) 374688f306b6SKirill A. Shutemov goto out_unlock; 374788f306b6SKirill A. Shutemov if (vma->vm_file && vma->vm_file->f_mapping && 374888f306b6SKirill A. Shutemov !is_vm_hugetlb_page(vma)) 3749454ed842SPeter Zijlstra vm_lock_mapping(mm, vma->vm_file->f_mapping); 37507906d00cSAndrea Arcangeli } 37517cd5a02fSPeter Zijlstra 3752763ecb03SLiam R. Howlett mas_set(&mas, 0); 3753763ecb03SLiam R. Howlett mas_for_each(&mas, vma, ULONG_MAX) { 37547cd5a02fSPeter Zijlstra if (signal_pending(current)) 37557cd5a02fSPeter Zijlstra goto out_unlock; 37567cd5a02fSPeter Zijlstra if (vma->anon_vma) 37575beb4930SRik van Riel list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) 37585beb4930SRik van Riel vm_lock_anon_vma(mm, avc->anon_vma); 37597cd5a02fSPeter Zijlstra } 37607cd5a02fSPeter Zijlstra 3761584cff54SKautuk Consul return 0; 37627906d00cSAndrea Arcangeli 37637906d00cSAndrea Arcangeli out_unlock: 37647906d00cSAndrea Arcangeli mm_drop_all_locks(mm); 3765584cff54SKautuk Consul return -EINTR; 37667906d00cSAndrea Arcangeli } 37677906d00cSAndrea Arcangeli 37687906d00cSAndrea Arcangeli static void vm_unlock_anon_vma(struct anon_vma *anon_vma) 37697906d00cSAndrea Arcangeli { 3770f808c13fSDavidlohr Bueso if (test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_root.rb_node)) { 37717906d00cSAndrea Arcangeli /* 37727906d00cSAndrea Arcangeli * The LSB of head.next can't change to 0 from under 37737906d00cSAndrea Arcangeli * us because we hold the mm_all_locks_mutex. 37747906d00cSAndrea Arcangeli * 37757906d00cSAndrea Arcangeli * We must however clear the bitflag before unlocking 3776bf181b9fSMichel Lespinasse * the vma so the users using the anon_vma->rb_root will 37777906d00cSAndrea Arcangeli * never see our bitflag. 37787906d00cSAndrea Arcangeli * 37797906d00cSAndrea Arcangeli * No need of atomic instructions here, head.next 37807906d00cSAndrea Arcangeli * can't change from under us until we release the 37815a505085SIngo Molnar * anon_vma->root->rwsem. 37827906d00cSAndrea Arcangeli */ 37837906d00cSAndrea Arcangeli if (!__test_and_clear_bit(0, (unsigned long *) 3784f808c13fSDavidlohr Bueso &anon_vma->root->rb_root.rb_root.rb_node)) 37857906d00cSAndrea Arcangeli BUG(); 378608b52706SKonstantin Khlebnikov anon_vma_unlock_write(anon_vma); 37877906d00cSAndrea Arcangeli } 37887906d00cSAndrea Arcangeli } 37897906d00cSAndrea Arcangeli 37907906d00cSAndrea Arcangeli static void vm_unlock_mapping(struct address_space *mapping) 37917906d00cSAndrea Arcangeli { 37927906d00cSAndrea Arcangeli if (test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) { 37937906d00cSAndrea Arcangeli /* 37947906d00cSAndrea Arcangeli * AS_MM_ALL_LOCKS can't change to 0 from under us 37957906d00cSAndrea Arcangeli * because we hold the mm_all_locks_mutex. 37967906d00cSAndrea Arcangeli */ 379783cde9e8SDavidlohr Bueso i_mmap_unlock_write(mapping); 37987906d00cSAndrea Arcangeli if (!test_and_clear_bit(AS_MM_ALL_LOCKS, 37997906d00cSAndrea Arcangeli &mapping->flags)) 38007906d00cSAndrea Arcangeli BUG(); 38017906d00cSAndrea Arcangeli } 38027906d00cSAndrea Arcangeli } 38037906d00cSAndrea Arcangeli 38047906d00cSAndrea Arcangeli /* 3805c1e8d7c6SMichel Lespinasse * The mmap_lock cannot be released by the caller until 38067906d00cSAndrea Arcangeli * mm_drop_all_locks() returns. 38077906d00cSAndrea Arcangeli */ 38087906d00cSAndrea Arcangeli void mm_drop_all_locks(struct mm_struct *mm) 38097906d00cSAndrea Arcangeli { 38107906d00cSAndrea Arcangeli struct vm_area_struct *vma; 38115beb4930SRik van Riel struct anon_vma_chain *avc; 3812763ecb03SLiam R. Howlett MA_STATE(mas, &mm->mm_mt, 0, 0); 38137906d00cSAndrea Arcangeli 3814325bca1fSRolf Eike Beer mmap_assert_write_locked(mm); 38157906d00cSAndrea Arcangeli BUG_ON(!mutex_is_locked(&mm_all_locks_mutex)); 38167906d00cSAndrea Arcangeli 3817763ecb03SLiam R. Howlett mas_for_each(&mas, vma, ULONG_MAX) { 38187906d00cSAndrea Arcangeli if (vma->anon_vma) 38195beb4930SRik van Riel list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) 38205beb4930SRik van Riel vm_unlock_anon_vma(avc->anon_vma); 38217906d00cSAndrea Arcangeli if (vma->vm_file && vma->vm_file->f_mapping) 38227906d00cSAndrea Arcangeli vm_unlock_mapping(vma->vm_file->f_mapping); 38237906d00cSAndrea Arcangeli } 38247906d00cSAndrea Arcangeli 38257906d00cSAndrea Arcangeli mutex_unlock(&mm_all_locks_mutex); 38267906d00cSAndrea Arcangeli } 38278feae131SDavid Howells 38288feae131SDavid Howells /* 38293edf41d8Sseokhoon.yoon * initialise the percpu counter for VM 38308feae131SDavid Howells */ 38318feae131SDavid Howells void __init mmap_init(void) 38328feae131SDavid Howells { 383300a62ce9SKOSAKI Motohiro int ret; 383400a62ce9SKOSAKI Motohiro 3835908c7f19STejun Heo ret = percpu_counter_init(&vm_committed_as, 0, GFP_KERNEL); 383600a62ce9SKOSAKI Motohiro VM_BUG_ON(ret); 38378feae131SDavid Howells } 3838c9b1d098SAndrew Shewmaker 3839c9b1d098SAndrew Shewmaker /* 3840c9b1d098SAndrew Shewmaker * Initialise sysctl_user_reserve_kbytes. 3841c9b1d098SAndrew Shewmaker * 3842c9b1d098SAndrew Shewmaker * This is intended to prevent a user from starting a single memory hogging 3843c9b1d098SAndrew Shewmaker * process, such that they cannot recover (kill the hog) in OVERCOMMIT_NEVER 3844c9b1d098SAndrew Shewmaker * mode. 3845c9b1d098SAndrew Shewmaker * 3846c9b1d098SAndrew Shewmaker * The default value is min(3% of free memory, 128MB) 3847c9b1d098SAndrew Shewmaker * 128MB is enough to recover with sshd/login, bash, and top/kill. 3848c9b1d098SAndrew Shewmaker */ 38491640879aSAndrew Shewmaker static int init_user_reserve(void) 3850c9b1d098SAndrew Shewmaker { 3851c9b1d098SAndrew Shewmaker unsigned long free_kbytes; 3852c9b1d098SAndrew Shewmaker 3853b1773e0eSZhangPeng free_kbytes = K(global_zone_page_state(NR_FREE_PAGES)); 3854c9b1d098SAndrew Shewmaker 38559c793854SYajun Deng sysctl_user_reserve_kbytes = min(free_kbytes / 32, SZ_128K); 3856c9b1d098SAndrew Shewmaker return 0; 3857c9b1d098SAndrew Shewmaker } 3858a64fb3cdSPaul Gortmaker subsys_initcall(init_user_reserve); 38594eeab4f5SAndrew Shewmaker 38604eeab4f5SAndrew Shewmaker /* 38614eeab4f5SAndrew Shewmaker * Initialise sysctl_admin_reserve_kbytes. 38624eeab4f5SAndrew Shewmaker * 38634eeab4f5SAndrew Shewmaker * The purpose of sysctl_admin_reserve_kbytes is to allow the sys admin 38644eeab4f5SAndrew Shewmaker * to log in and kill a memory hogging process. 38654eeab4f5SAndrew Shewmaker * 38664eeab4f5SAndrew Shewmaker * Systems with more than 256MB will reserve 8MB, enough to recover 38674eeab4f5SAndrew Shewmaker * with sshd, bash, and top in OVERCOMMIT_GUESS. Smaller systems will 38684eeab4f5SAndrew Shewmaker * only reserve 3% of free pages by default. 38694eeab4f5SAndrew Shewmaker */ 38701640879aSAndrew Shewmaker static int init_admin_reserve(void) 38714eeab4f5SAndrew Shewmaker { 38724eeab4f5SAndrew Shewmaker unsigned long free_kbytes; 38734eeab4f5SAndrew Shewmaker 3874b1773e0eSZhangPeng free_kbytes = K(global_zone_page_state(NR_FREE_PAGES)); 38754eeab4f5SAndrew Shewmaker 38769c793854SYajun Deng sysctl_admin_reserve_kbytes = min(free_kbytes / 32, SZ_8K); 38774eeab4f5SAndrew Shewmaker return 0; 38784eeab4f5SAndrew Shewmaker } 3879a64fb3cdSPaul Gortmaker subsys_initcall(init_admin_reserve); 38801640879aSAndrew Shewmaker 38811640879aSAndrew Shewmaker /* 38821640879aSAndrew Shewmaker * Reinititalise user and admin reserves if memory is added or removed. 38831640879aSAndrew Shewmaker * 38841640879aSAndrew Shewmaker * The default user reserve max is 128MB, and the default max for the 38851640879aSAndrew Shewmaker * admin reserve is 8MB. These are usually, but not always, enough to 38861640879aSAndrew Shewmaker * enable recovery from a memory hogging process using login/sshd, a shell, 38871640879aSAndrew Shewmaker * and tools like top. It may make sense to increase or even disable the 38881640879aSAndrew Shewmaker * reserve depending on the existence of swap or variations in the recovery 38891640879aSAndrew Shewmaker * tools. So, the admin may have changed them. 38901640879aSAndrew Shewmaker * 38911640879aSAndrew Shewmaker * If memory is added and the reserves have been eliminated or increased above 38921640879aSAndrew Shewmaker * the default max, then we'll trust the admin. 38931640879aSAndrew Shewmaker * 38941640879aSAndrew Shewmaker * If memory is removed and there isn't enough free memory, then we 38951640879aSAndrew Shewmaker * need to reset the reserves. 38961640879aSAndrew Shewmaker * 38971640879aSAndrew Shewmaker * Otherwise keep the reserve set by the admin. 38981640879aSAndrew Shewmaker */ 38991640879aSAndrew Shewmaker static int reserve_mem_notifier(struct notifier_block *nb, 39001640879aSAndrew Shewmaker unsigned long action, void *data) 39011640879aSAndrew Shewmaker { 39021640879aSAndrew Shewmaker unsigned long tmp, free_kbytes; 39031640879aSAndrew Shewmaker 39041640879aSAndrew Shewmaker switch (action) { 39051640879aSAndrew Shewmaker case MEM_ONLINE: 39061640879aSAndrew Shewmaker /* Default max is 128MB. Leave alone if modified by operator. */ 39071640879aSAndrew Shewmaker tmp = sysctl_user_reserve_kbytes; 39089c793854SYajun Deng if (tmp > 0 && tmp < SZ_128K) 39091640879aSAndrew Shewmaker init_user_reserve(); 39101640879aSAndrew Shewmaker 39111640879aSAndrew Shewmaker /* Default max is 8MB. Leave alone if modified by operator. */ 39121640879aSAndrew Shewmaker tmp = sysctl_admin_reserve_kbytes; 39139c793854SYajun Deng if (tmp > 0 && tmp < SZ_8K) 39141640879aSAndrew Shewmaker init_admin_reserve(); 39151640879aSAndrew Shewmaker 39161640879aSAndrew Shewmaker break; 39171640879aSAndrew Shewmaker case MEM_OFFLINE: 3918b1773e0eSZhangPeng free_kbytes = K(global_zone_page_state(NR_FREE_PAGES)); 39191640879aSAndrew Shewmaker 39201640879aSAndrew Shewmaker if (sysctl_user_reserve_kbytes > free_kbytes) { 39211640879aSAndrew Shewmaker init_user_reserve(); 39221640879aSAndrew Shewmaker pr_info("vm.user_reserve_kbytes reset to %lu\n", 39231640879aSAndrew Shewmaker sysctl_user_reserve_kbytes); 39241640879aSAndrew Shewmaker } 39251640879aSAndrew Shewmaker 39261640879aSAndrew Shewmaker if (sysctl_admin_reserve_kbytes > free_kbytes) { 39271640879aSAndrew Shewmaker init_admin_reserve(); 39281640879aSAndrew Shewmaker pr_info("vm.admin_reserve_kbytes reset to %lu\n", 39291640879aSAndrew Shewmaker sysctl_admin_reserve_kbytes); 39301640879aSAndrew Shewmaker } 39311640879aSAndrew Shewmaker break; 39321640879aSAndrew Shewmaker default: 39331640879aSAndrew Shewmaker break; 39341640879aSAndrew Shewmaker } 39351640879aSAndrew Shewmaker return NOTIFY_OK; 39361640879aSAndrew Shewmaker } 39371640879aSAndrew Shewmaker 39381640879aSAndrew Shewmaker static int __meminit init_reserve_notifier(void) 39391640879aSAndrew Shewmaker { 39401eeaa4fdSLiu Shixin if (hotplug_memory_notifier(reserve_mem_notifier, DEFAULT_CALLBACK_PRI)) 3941b1de0d13SMitchel Humpherys pr_err("Failed registering memory add/remove notifier for admin reserve\n"); 39421640879aSAndrew Shewmaker 39431640879aSAndrew Shewmaker return 0; 39441640879aSAndrew Shewmaker } 3945a64fb3cdSPaul Gortmaker subsys_initcall(init_reserve_notifier); 3946