11da177e4SLinus Torvalds /* 21da177e4SLinus Torvalds * mm/mmap.c 31da177e4SLinus Torvalds * 41da177e4SLinus Torvalds * Written by obz. 51da177e4SLinus Torvalds * 6046c6884SAlan Cox * Address space accounting code <alan@lxorguk.ukuu.org.uk> 71da177e4SLinus Torvalds */ 81da177e4SLinus Torvalds 9b1de0d13SMitchel Humpherys #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt 10b1de0d13SMitchel Humpherys 11e8420a8eSCyril Hrubis #include <linux/kernel.h> 121da177e4SLinus Torvalds #include <linux/slab.h> 134af3c9ccSAlexey Dobriyan #include <linux/backing-dev.h> 141da177e4SLinus Torvalds #include <linux/mm.h> 15615d6e87SDavidlohr Bueso #include <linux/vmacache.h> 161da177e4SLinus Torvalds #include <linux/shm.h> 171da177e4SLinus Torvalds #include <linux/mman.h> 181da177e4SLinus Torvalds #include <linux/pagemap.h> 191da177e4SLinus Torvalds #include <linux/swap.h> 201da177e4SLinus Torvalds #include <linux/syscalls.h> 21c59ede7bSRandy.Dunlap #include <linux/capability.h> 221da177e4SLinus Torvalds #include <linux/init.h> 231da177e4SLinus Torvalds #include <linux/file.h> 241da177e4SLinus Torvalds #include <linux/fs.h> 251da177e4SLinus Torvalds #include <linux/personality.h> 261da177e4SLinus Torvalds #include <linux/security.h> 271da177e4SLinus Torvalds #include <linux/hugetlb.h> 28*c01d5b30SHugh Dickins #include <linux/shmem_fs.h> 291da177e4SLinus Torvalds #include <linux/profile.h> 30b95f1b31SPaul Gortmaker #include <linux/export.h> 311da177e4SLinus Torvalds #include <linux/mount.h> 321da177e4SLinus Torvalds #include <linux/mempolicy.h> 331da177e4SLinus Torvalds #include <linux/rmap.h> 34cddb8a5cSAndrea Arcangeli #include <linux/mmu_notifier.h> 3582f71ae4SKonstantin Khlebnikov #include <linux/mmdebug.h> 36cdd6c482SIngo Molnar #include <linux/perf_event.h> 37120a795dSAl Viro #include <linux/audit.h> 38b15d00b6SAndrea Arcangeli #include <linux/khugepaged.h> 392b144498SSrikar Dronamraju #include <linux/uprobes.h> 40d3737187SMichel Lespinasse #include <linux/rbtree_augmented.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> 471da177e4SLinus Torvalds 481da177e4SLinus Torvalds #include <asm/uaccess.h> 491da177e4SLinus Torvalds #include <asm/cacheflush.h> 501da177e4SLinus Torvalds #include <asm/tlb.h> 51d6dd61c8SJeremy Fitzhardinge #include <asm/mmu_context.h> 521da177e4SLinus Torvalds 5342b77728SJan Beulich #include "internal.h" 5442b77728SJan Beulich 553a459756SKirill Korotaev #ifndef arch_mmap_check 563a459756SKirill Korotaev #define arch_mmap_check(addr, len, flags) (0) 573a459756SKirill Korotaev #endif 583a459756SKirill Korotaev 59d07e2259SDaniel Cashman #ifdef CONFIG_HAVE_ARCH_MMAP_RND_BITS 60d07e2259SDaniel Cashman const int mmap_rnd_bits_min = CONFIG_ARCH_MMAP_RND_BITS_MIN; 61d07e2259SDaniel Cashman const int mmap_rnd_bits_max = CONFIG_ARCH_MMAP_RND_BITS_MAX; 62d07e2259SDaniel Cashman int mmap_rnd_bits __read_mostly = CONFIG_ARCH_MMAP_RND_BITS; 63d07e2259SDaniel Cashman #endif 64d07e2259SDaniel Cashman #ifdef CONFIG_HAVE_ARCH_MMAP_RND_COMPAT_BITS 65d07e2259SDaniel Cashman const int mmap_rnd_compat_bits_min = CONFIG_ARCH_MMAP_RND_COMPAT_BITS_MIN; 66d07e2259SDaniel Cashman const int mmap_rnd_compat_bits_max = CONFIG_ARCH_MMAP_RND_COMPAT_BITS_MAX; 67d07e2259SDaniel Cashman int mmap_rnd_compat_bits __read_mostly = CONFIG_ARCH_MMAP_RND_COMPAT_BITS; 68d07e2259SDaniel Cashman #endif 69d07e2259SDaniel Cashman 70f4fcd558SKonstantin Khlebnikov static bool ignore_rlimit_data; 71d977d56cSKonstantin Khlebnikov core_param(ignore_rlimit_data, ignore_rlimit_data, bool, 0644); 72d07e2259SDaniel Cashman 73e0da382cSHugh Dickins static void unmap_region(struct mm_struct *mm, 74e0da382cSHugh Dickins struct vm_area_struct *vma, struct vm_area_struct *prev, 75e0da382cSHugh Dickins unsigned long start, unsigned long end); 76e0da382cSHugh Dickins 771da177e4SLinus Torvalds /* description of effects of mapping type and prot in current implementation. 781da177e4SLinus Torvalds * this is due to the limited x86 page protection hardware. The expected 791da177e4SLinus Torvalds * behavior is in parens: 801da177e4SLinus Torvalds * 811da177e4SLinus Torvalds * map_type prot 821da177e4SLinus Torvalds * PROT_NONE PROT_READ PROT_WRITE PROT_EXEC 831da177e4SLinus Torvalds * MAP_SHARED r: (no) no r: (yes) yes r: (no) yes r: (no) yes 841da177e4SLinus Torvalds * w: (no) no w: (no) no w: (yes) yes w: (no) no 851da177e4SLinus Torvalds * x: (no) no x: (no) yes x: (no) yes x: (yes) yes 861da177e4SLinus Torvalds * 871da177e4SLinus Torvalds * MAP_PRIVATE r: (no) no r: (yes) yes r: (no) yes r: (no) yes 881da177e4SLinus Torvalds * w: (no) no w: (no) no w: (copy) copy w: (no) no 891da177e4SLinus Torvalds * x: (no) no x: (no) yes x: (no) yes x: (yes) yes 901da177e4SLinus Torvalds * 911da177e4SLinus Torvalds */ 921da177e4SLinus Torvalds pgprot_t protection_map[16] = { 931da177e4SLinus Torvalds __P000, __P001, __P010, __P011, __P100, __P101, __P110, __P111, 941da177e4SLinus Torvalds __S000, __S001, __S010, __S011, __S100, __S101, __S110, __S111 951da177e4SLinus Torvalds }; 961da177e4SLinus Torvalds 97804af2cfSHugh Dickins pgprot_t vm_get_page_prot(unsigned long vm_flags) 98804af2cfSHugh Dickins { 99b845f313SDave Kleikamp return __pgprot(pgprot_val(protection_map[vm_flags & 100b845f313SDave Kleikamp (VM_READ|VM_WRITE|VM_EXEC|VM_SHARED)]) | 101b845f313SDave Kleikamp pgprot_val(arch_vm_get_page_prot(vm_flags))); 102804af2cfSHugh Dickins } 103804af2cfSHugh Dickins EXPORT_SYMBOL(vm_get_page_prot); 104804af2cfSHugh Dickins 10564e45507SPeter Feiner static pgprot_t vm_pgprot_modify(pgprot_t oldprot, unsigned long vm_flags) 10664e45507SPeter Feiner { 10764e45507SPeter Feiner return pgprot_modify(oldprot, vm_get_page_prot(vm_flags)); 10864e45507SPeter Feiner } 10964e45507SPeter Feiner 11064e45507SPeter Feiner /* Update vma->vm_page_prot to reflect vma->vm_flags. */ 11164e45507SPeter Feiner void vma_set_page_prot(struct vm_area_struct *vma) 11264e45507SPeter Feiner { 11364e45507SPeter Feiner unsigned long vm_flags = vma->vm_flags; 11464e45507SPeter Feiner 11564e45507SPeter Feiner vma->vm_page_prot = vm_pgprot_modify(vma->vm_page_prot, vm_flags); 11664e45507SPeter Feiner if (vma_wants_writenotify(vma)) { 11764e45507SPeter Feiner vm_flags &= ~VM_SHARED; 11864e45507SPeter Feiner vma->vm_page_prot = vm_pgprot_modify(vma->vm_page_prot, 11964e45507SPeter Feiner vm_flags); 12064e45507SPeter Feiner } 12164e45507SPeter Feiner } 12264e45507SPeter Feiner 1231da177e4SLinus Torvalds /* 124c8c06efaSDavidlohr Bueso * Requires inode->i_mapping->i_mmap_rwsem 1251da177e4SLinus Torvalds */ 1261da177e4SLinus Torvalds static void __remove_shared_vm_struct(struct vm_area_struct *vma, 1271da177e4SLinus Torvalds struct file *file, struct address_space *mapping) 1281da177e4SLinus Torvalds { 1291da177e4SLinus Torvalds if (vma->vm_flags & VM_DENYWRITE) 130496ad9aaSAl Viro atomic_inc(&file_inode(file)->i_writecount); 1311da177e4SLinus Torvalds if (vma->vm_flags & VM_SHARED) 1324bb5f5d9SDavid Herrmann mapping_unmap_writable(mapping); 1331da177e4SLinus Torvalds 1341da177e4SLinus Torvalds flush_dcache_mmap_lock(mapping); 1356b2dbba8SMichel Lespinasse vma_interval_tree_remove(vma, &mapping->i_mmap); 1361da177e4SLinus Torvalds flush_dcache_mmap_unlock(mapping); 1371da177e4SLinus Torvalds } 1381da177e4SLinus Torvalds 1391da177e4SLinus Torvalds /* 1406b2dbba8SMichel Lespinasse * Unlink a file-based vm structure from its interval tree, to hide 141a8fb5618SHugh Dickins * vma from rmap and vmtruncate before freeing its page tables. 1421da177e4SLinus Torvalds */ 143a8fb5618SHugh Dickins void unlink_file_vma(struct vm_area_struct *vma) 1441da177e4SLinus Torvalds { 1451da177e4SLinus Torvalds struct file *file = vma->vm_file; 1461da177e4SLinus Torvalds 1471da177e4SLinus Torvalds if (file) { 1481da177e4SLinus Torvalds struct address_space *mapping = file->f_mapping; 14983cde9e8SDavidlohr Bueso i_mmap_lock_write(mapping); 1501da177e4SLinus Torvalds __remove_shared_vm_struct(vma, file, mapping); 15183cde9e8SDavidlohr Bueso i_mmap_unlock_write(mapping); 1521da177e4SLinus Torvalds } 153a8fb5618SHugh Dickins } 154a8fb5618SHugh Dickins 155a8fb5618SHugh Dickins /* 156a8fb5618SHugh Dickins * Close a vm structure and free it, returning the next. 157a8fb5618SHugh Dickins */ 158a8fb5618SHugh Dickins static struct vm_area_struct *remove_vma(struct vm_area_struct *vma) 159a8fb5618SHugh Dickins { 160a8fb5618SHugh Dickins struct vm_area_struct *next = vma->vm_next; 161a8fb5618SHugh Dickins 162a8fb5618SHugh Dickins might_sleep(); 1631da177e4SLinus Torvalds if (vma->vm_ops && vma->vm_ops->close) 1641da177e4SLinus Torvalds vma->vm_ops->close(vma); 165e9714acfSKonstantin Khlebnikov if (vma->vm_file) 166a8fb5618SHugh Dickins fput(vma->vm_file); 167f0be3d32SLee Schermerhorn mpol_put(vma_policy(vma)); 1681da177e4SLinus Torvalds kmem_cache_free(vm_area_cachep, vma); 169a8fb5618SHugh Dickins return next; 1701da177e4SLinus Torvalds } 1711da177e4SLinus Torvalds 1725d22fc25SLinus Torvalds static int do_brk(unsigned long addr, unsigned long len); 173e4eb1ff6SLinus Torvalds 1746a6160a7SHeiko Carstens SYSCALL_DEFINE1(brk, unsigned long, brk) 1751da177e4SLinus Torvalds { 1768764b338SCyrill Gorcunov unsigned long retval; 1771da177e4SLinus Torvalds unsigned long newbrk, oldbrk; 1781da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 179a5b4592cSJiri Kosina unsigned long min_brk; 180128557ffSMichel Lespinasse bool populate; 1811da177e4SLinus Torvalds 182dc0ef0dfSMichal Hocko if (down_write_killable(&mm->mmap_sem)) 183dc0ef0dfSMichal Hocko return -EINTR; 1841da177e4SLinus Torvalds 185a5b4592cSJiri Kosina #ifdef CONFIG_COMPAT_BRK 1865520e894SJiri Kosina /* 1875520e894SJiri Kosina * CONFIG_COMPAT_BRK can still be overridden by setting 1885520e894SJiri Kosina * randomize_va_space to 2, which will still cause mm->start_brk 1895520e894SJiri Kosina * to be arbitrarily shifted 1905520e894SJiri Kosina */ 1914471a675SJiri Kosina if (current->brk_randomized) 1925520e894SJiri Kosina min_brk = mm->start_brk; 1935520e894SJiri Kosina else 1945520e894SJiri Kosina min_brk = mm->end_data; 195a5b4592cSJiri Kosina #else 196a5b4592cSJiri Kosina min_brk = mm->start_brk; 197a5b4592cSJiri Kosina #endif 198a5b4592cSJiri Kosina if (brk < min_brk) 1991da177e4SLinus Torvalds goto out; 2001e624196SRam Gupta 2011e624196SRam Gupta /* 2021e624196SRam Gupta * Check against rlimit here. If this check is done later after the test 2031e624196SRam Gupta * of oldbrk with newbrk then it can escape the test and let the data 2041e624196SRam Gupta * segment grow beyond its set limit the in case where the limit is 2051e624196SRam Gupta * not page aligned -Ram Gupta 2061e624196SRam Gupta */ 2078764b338SCyrill Gorcunov if (check_data_rlimit(rlimit(RLIMIT_DATA), brk, mm->start_brk, 2088764b338SCyrill Gorcunov mm->end_data, mm->start_data)) 2091e624196SRam Gupta goto out; 2101e624196SRam Gupta 2111da177e4SLinus Torvalds newbrk = PAGE_ALIGN(brk); 2121da177e4SLinus Torvalds oldbrk = PAGE_ALIGN(mm->brk); 2131da177e4SLinus Torvalds if (oldbrk == newbrk) 2141da177e4SLinus Torvalds goto set_brk; 2151da177e4SLinus Torvalds 2161da177e4SLinus Torvalds /* Always allow shrinking brk. */ 2171da177e4SLinus Torvalds if (brk <= mm->brk) { 2181da177e4SLinus Torvalds if (!do_munmap(mm, newbrk, oldbrk-newbrk)) 2191da177e4SLinus Torvalds goto set_brk; 2201da177e4SLinus Torvalds goto out; 2211da177e4SLinus Torvalds } 2221da177e4SLinus Torvalds 2231da177e4SLinus Torvalds /* Check against existing mmap mappings. */ 2241da177e4SLinus Torvalds if (find_vma_intersection(mm, oldbrk, newbrk+PAGE_SIZE)) 2251da177e4SLinus Torvalds goto out; 2261da177e4SLinus Torvalds 2271da177e4SLinus Torvalds /* Ok, looks good - let it rip. */ 2285d22fc25SLinus Torvalds if (do_brk(oldbrk, newbrk-oldbrk) < 0) 2291da177e4SLinus Torvalds goto out; 230128557ffSMichel Lespinasse 2311da177e4SLinus Torvalds set_brk: 2321da177e4SLinus Torvalds mm->brk = brk; 233128557ffSMichel Lespinasse populate = newbrk > oldbrk && (mm->def_flags & VM_LOCKED) != 0; 234128557ffSMichel Lespinasse up_write(&mm->mmap_sem); 235128557ffSMichel Lespinasse if (populate) 236128557ffSMichel Lespinasse mm_populate(oldbrk, newbrk - oldbrk); 237128557ffSMichel Lespinasse return brk; 238128557ffSMichel Lespinasse 2391da177e4SLinus Torvalds out: 2401da177e4SLinus Torvalds retval = mm->brk; 2411da177e4SLinus Torvalds up_write(&mm->mmap_sem); 2421da177e4SLinus Torvalds return retval; 2431da177e4SLinus Torvalds } 2441da177e4SLinus Torvalds 245d3737187SMichel Lespinasse static long vma_compute_subtree_gap(struct vm_area_struct *vma) 246d3737187SMichel Lespinasse { 247d3737187SMichel Lespinasse unsigned long max, subtree_gap; 248d3737187SMichel Lespinasse max = vma->vm_start; 249d3737187SMichel Lespinasse if (vma->vm_prev) 250d3737187SMichel Lespinasse max -= vma->vm_prev->vm_end; 251d3737187SMichel Lespinasse if (vma->vm_rb.rb_left) { 252d3737187SMichel Lespinasse subtree_gap = rb_entry(vma->vm_rb.rb_left, 253d3737187SMichel Lespinasse struct vm_area_struct, vm_rb)->rb_subtree_gap; 254d3737187SMichel Lespinasse if (subtree_gap > max) 255d3737187SMichel Lespinasse max = subtree_gap; 256d3737187SMichel Lespinasse } 257d3737187SMichel Lespinasse if (vma->vm_rb.rb_right) { 258d3737187SMichel Lespinasse subtree_gap = rb_entry(vma->vm_rb.rb_right, 259d3737187SMichel Lespinasse struct vm_area_struct, vm_rb)->rb_subtree_gap; 260d3737187SMichel Lespinasse if (subtree_gap > max) 261d3737187SMichel Lespinasse max = subtree_gap; 262d3737187SMichel Lespinasse } 263d3737187SMichel Lespinasse return max; 264d3737187SMichel Lespinasse } 265d3737187SMichel Lespinasse 266ed8ea815SMichel Lespinasse #ifdef CONFIG_DEBUG_VM_RB 267acf128d0SAndrea Arcangeli static int browse_rb(struct mm_struct *mm) 2681da177e4SLinus Torvalds { 269acf128d0SAndrea Arcangeli struct rb_root *root = &mm->mm_rb; 2705a0768f6SMichel Lespinasse int i = 0, j, bug = 0; 2711da177e4SLinus Torvalds struct rb_node *nd, *pn = NULL; 2721da177e4SLinus Torvalds unsigned long prev = 0, pend = 0; 2731da177e4SLinus Torvalds 2741da177e4SLinus Torvalds for (nd = rb_first(root); nd; nd = rb_next(nd)) { 2751da177e4SLinus Torvalds struct vm_area_struct *vma; 2761da177e4SLinus Torvalds vma = rb_entry(nd, struct vm_area_struct, vm_rb); 2775a0768f6SMichel Lespinasse if (vma->vm_start < prev) { 278ff26f70fSAndrew Morton pr_emerg("vm_start %lx < prev %lx\n", 279ff26f70fSAndrew Morton vma->vm_start, prev); 2805a0768f6SMichel Lespinasse bug = 1; 2815a0768f6SMichel Lespinasse } 2825a0768f6SMichel Lespinasse if (vma->vm_start < pend) { 283ff26f70fSAndrew Morton pr_emerg("vm_start %lx < pend %lx\n", 284ff26f70fSAndrew Morton vma->vm_start, pend); 2855a0768f6SMichel Lespinasse bug = 1; 2865a0768f6SMichel Lespinasse } 2875a0768f6SMichel Lespinasse if (vma->vm_start > vma->vm_end) { 288ff26f70fSAndrew Morton pr_emerg("vm_start %lx > vm_end %lx\n", 289ff26f70fSAndrew Morton vma->vm_start, vma->vm_end); 2905a0768f6SMichel Lespinasse bug = 1; 2915a0768f6SMichel Lespinasse } 292acf128d0SAndrea Arcangeli spin_lock(&mm->page_table_lock); 2935a0768f6SMichel Lespinasse if (vma->rb_subtree_gap != vma_compute_subtree_gap(vma)) { 2948542bdfcSSasha Levin pr_emerg("free gap %lx, correct %lx\n", 2955a0768f6SMichel Lespinasse vma->rb_subtree_gap, 2965a0768f6SMichel Lespinasse vma_compute_subtree_gap(vma)); 2975a0768f6SMichel Lespinasse bug = 1; 2985a0768f6SMichel Lespinasse } 299acf128d0SAndrea Arcangeli spin_unlock(&mm->page_table_lock); 3001da177e4SLinus Torvalds i++; 3011da177e4SLinus Torvalds pn = nd; 302d1af65d1SDavid Miller prev = vma->vm_start; 303d1af65d1SDavid Miller pend = vma->vm_end; 3041da177e4SLinus Torvalds } 3051da177e4SLinus Torvalds j = 0; 3065a0768f6SMichel Lespinasse for (nd = pn; nd; nd = rb_prev(nd)) 3071da177e4SLinus Torvalds j++; 3085a0768f6SMichel Lespinasse if (i != j) { 3098542bdfcSSasha Levin pr_emerg("backwards %d, forwards %d\n", j, i); 3105a0768f6SMichel Lespinasse bug = 1; 3111da177e4SLinus Torvalds } 3125a0768f6SMichel Lespinasse return bug ? -1 : i; 3131da177e4SLinus Torvalds } 3141da177e4SLinus Torvalds 315d3737187SMichel Lespinasse static void validate_mm_rb(struct rb_root *root, struct vm_area_struct *ignore) 316d3737187SMichel Lespinasse { 317d3737187SMichel Lespinasse struct rb_node *nd; 318d3737187SMichel Lespinasse 319d3737187SMichel Lespinasse for (nd = rb_first(root); nd; nd = rb_next(nd)) { 320d3737187SMichel Lespinasse struct vm_area_struct *vma; 321d3737187SMichel Lespinasse vma = rb_entry(nd, struct vm_area_struct, vm_rb); 32296dad67fSSasha Levin VM_BUG_ON_VMA(vma != ignore && 32396dad67fSSasha Levin vma->rb_subtree_gap != vma_compute_subtree_gap(vma), 32496dad67fSSasha Levin vma); 325d3737187SMichel Lespinasse } 3261da177e4SLinus Torvalds } 3271da177e4SLinus Torvalds 328eafd4dc4SRashika Kheria static void validate_mm(struct mm_struct *mm) 3291da177e4SLinus Torvalds { 3301da177e4SLinus Torvalds int bug = 0; 3311da177e4SLinus Torvalds int i = 0; 3325a0768f6SMichel Lespinasse unsigned long highest_address = 0; 333ed8ea815SMichel Lespinasse struct vm_area_struct *vma = mm->mmap; 334ff26f70fSAndrew Morton 335ed8ea815SMichel Lespinasse while (vma) { 33612352d3cSKonstantin Khlebnikov struct anon_vma *anon_vma = vma->anon_vma; 337ed8ea815SMichel Lespinasse struct anon_vma_chain *avc; 338ff26f70fSAndrew Morton 33912352d3cSKonstantin Khlebnikov if (anon_vma) { 34012352d3cSKonstantin Khlebnikov anon_vma_lock_read(anon_vma); 341ed8ea815SMichel Lespinasse list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) 342ed8ea815SMichel Lespinasse anon_vma_interval_tree_verify(avc); 34312352d3cSKonstantin Khlebnikov anon_vma_unlock_read(anon_vma); 34412352d3cSKonstantin Khlebnikov } 34512352d3cSKonstantin Khlebnikov 3465a0768f6SMichel Lespinasse highest_address = vma->vm_end; 347ed8ea815SMichel Lespinasse vma = vma->vm_next; 3481da177e4SLinus Torvalds i++; 3491da177e4SLinus Torvalds } 3505a0768f6SMichel Lespinasse if (i != mm->map_count) { 3518542bdfcSSasha Levin pr_emerg("map_count %d vm_next %d\n", mm->map_count, i); 3525a0768f6SMichel Lespinasse bug = 1; 3535a0768f6SMichel Lespinasse } 3545a0768f6SMichel Lespinasse if (highest_address != mm->highest_vm_end) { 3558542bdfcSSasha Levin pr_emerg("mm->highest_vm_end %lx, found %lx\n", 3565a0768f6SMichel Lespinasse mm->highest_vm_end, highest_address); 3575a0768f6SMichel Lespinasse bug = 1; 3585a0768f6SMichel Lespinasse } 359acf128d0SAndrea Arcangeli i = browse_rb(mm); 3605a0768f6SMichel Lespinasse if (i != mm->map_count) { 361ff26f70fSAndrew Morton if (i != -1) 3628542bdfcSSasha Levin pr_emerg("map_count %d rb %d\n", mm->map_count, i); 3635a0768f6SMichel Lespinasse bug = 1; 3645a0768f6SMichel Lespinasse } 36596dad67fSSasha Levin VM_BUG_ON_MM(bug, mm); 3661da177e4SLinus Torvalds } 3671da177e4SLinus Torvalds #else 368d3737187SMichel Lespinasse #define validate_mm_rb(root, ignore) do { } while (0) 3691da177e4SLinus Torvalds #define validate_mm(mm) do { } while (0) 3701da177e4SLinus Torvalds #endif 3711da177e4SLinus Torvalds 372d3737187SMichel Lespinasse RB_DECLARE_CALLBACKS(static, vma_gap_callbacks, struct vm_area_struct, vm_rb, 373d3737187SMichel Lespinasse unsigned long, rb_subtree_gap, vma_compute_subtree_gap) 374d3737187SMichel Lespinasse 375d3737187SMichel Lespinasse /* 376d3737187SMichel Lespinasse * Update augmented rbtree rb_subtree_gap values after vma->vm_start or 377d3737187SMichel Lespinasse * vma->vm_prev->vm_end values changed, without modifying the vma's position 378d3737187SMichel Lespinasse * in the rbtree. 379d3737187SMichel Lespinasse */ 380d3737187SMichel Lespinasse static void vma_gap_update(struct vm_area_struct *vma) 381d3737187SMichel Lespinasse { 382d3737187SMichel Lespinasse /* 383d3737187SMichel Lespinasse * As it turns out, RB_DECLARE_CALLBACKS() already created a callback 384d3737187SMichel Lespinasse * function that does exacltly what we want. 385d3737187SMichel Lespinasse */ 386d3737187SMichel Lespinasse vma_gap_callbacks_propagate(&vma->vm_rb, NULL); 387d3737187SMichel Lespinasse } 388d3737187SMichel Lespinasse 389d3737187SMichel Lespinasse static inline void vma_rb_insert(struct vm_area_struct *vma, 390d3737187SMichel Lespinasse struct rb_root *root) 391d3737187SMichel Lespinasse { 392d3737187SMichel Lespinasse /* All rb_subtree_gap values must be consistent prior to insertion */ 393d3737187SMichel Lespinasse validate_mm_rb(root, NULL); 394d3737187SMichel Lespinasse 395d3737187SMichel Lespinasse rb_insert_augmented(&vma->vm_rb, root, &vma_gap_callbacks); 396d3737187SMichel Lespinasse } 397d3737187SMichel Lespinasse 398d3737187SMichel Lespinasse static void vma_rb_erase(struct vm_area_struct *vma, struct rb_root *root) 399d3737187SMichel Lespinasse { 400d3737187SMichel Lespinasse /* 401d3737187SMichel Lespinasse * All rb_subtree_gap values must be consistent prior to erase, 402d3737187SMichel Lespinasse * with the possible exception of the vma being erased. 403d3737187SMichel Lespinasse */ 404d3737187SMichel Lespinasse validate_mm_rb(root, vma); 405d3737187SMichel Lespinasse 406d3737187SMichel Lespinasse /* 407d3737187SMichel Lespinasse * Note rb_erase_augmented is a fairly large inline function, 408d3737187SMichel Lespinasse * so make sure we instantiate it only once with our desired 409d3737187SMichel Lespinasse * augmented rbtree callbacks. 410d3737187SMichel Lespinasse */ 411d3737187SMichel Lespinasse rb_erase_augmented(&vma->vm_rb, root, &vma_gap_callbacks); 412d3737187SMichel Lespinasse } 413d3737187SMichel Lespinasse 414bf181b9fSMichel Lespinasse /* 415bf181b9fSMichel Lespinasse * vma has some anon_vma assigned, and is already inserted on that 416bf181b9fSMichel Lespinasse * anon_vma's interval trees. 417bf181b9fSMichel Lespinasse * 418bf181b9fSMichel Lespinasse * Before updating the vma's vm_start / vm_end / vm_pgoff fields, the 419bf181b9fSMichel Lespinasse * vma must be removed from the anon_vma's interval trees using 420bf181b9fSMichel Lespinasse * anon_vma_interval_tree_pre_update_vma(). 421bf181b9fSMichel Lespinasse * 422bf181b9fSMichel Lespinasse * After the update, the vma will be reinserted using 423bf181b9fSMichel Lespinasse * anon_vma_interval_tree_post_update_vma(). 424bf181b9fSMichel Lespinasse * 425bf181b9fSMichel Lespinasse * The entire update must be protected by exclusive mmap_sem and by 426bf181b9fSMichel Lespinasse * the root anon_vma's mutex. 427bf181b9fSMichel Lespinasse */ 428bf181b9fSMichel Lespinasse static inline void 429bf181b9fSMichel Lespinasse anon_vma_interval_tree_pre_update_vma(struct vm_area_struct *vma) 430bf181b9fSMichel Lespinasse { 431bf181b9fSMichel Lespinasse struct anon_vma_chain *avc; 432bf181b9fSMichel Lespinasse 433bf181b9fSMichel Lespinasse list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) 434bf181b9fSMichel Lespinasse anon_vma_interval_tree_remove(avc, &avc->anon_vma->rb_root); 435bf181b9fSMichel Lespinasse } 436bf181b9fSMichel Lespinasse 437bf181b9fSMichel Lespinasse static inline void 438bf181b9fSMichel Lespinasse anon_vma_interval_tree_post_update_vma(struct vm_area_struct *vma) 439bf181b9fSMichel Lespinasse { 440bf181b9fSMichel Lespinasse struct anon_vma_chain *avc; 441bf181b9fSMichel Lespinasse 442bf181b9fSMichel Lespinasse list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) 443bf181b9fSMichel Lespinasse anon_vma_interval_tree_insert(avc, &avc->anon_vma->rb_root); 444bf181b9fSMichel Lespinasse } 445bf181b9fSMichel Lespinasse 4466597d783SHugh Dickins static int find_vma_links(struct mm_struct *mm, unsigned long addr, 4476597d783SHugh Dickins unsigned long end, struct vm_area_struct **pprev, 4486597d783SHugh Dickins struct rb_node ***rb_link, struct rb_node **rb_parent) 4491da177e4SLinus Torvalds { 4501da177e4SLinus Torvalds struct rb_node **__rb_link, *__rb_parent, *rb_prev; 4511da177e4SLinus Torvalds 4521da177e4SLinus Torvalds __rb_link = &mm->mm_rb.rb_node; 4531da177e4SLinus Torvalds rb_prev = __rb_parent = NULL; 4541da177e4SLinus Torvalds 4551da177e4SLinus Torvalds while (*__rb_link) { 4561da177e4SLinus Torvalds struct vm_area_struct *vma_tmp; 4571da177e4SLinus Torvalds 4581da177e4SLinus Torvalds __rb_parent = *__rb_link; 4591da177e4SLinus Torvalds vma_tmp = rb_entry(__rb_parent, struct vm_area_struct, vm_rb); 4601da177e4SLinus Torvalds 4611da177e4SLinus Torvalds if (vma_tmp->vm_end > addr) { 4626597d783SHugh Dickins /* Fail if an existing vma overlaps the area */ 4636597d783SHugh Dickins if (vma_tmp->vm_start < end) 4646597d783SHugh Dickins return -ENOMEM; 4651da177e4SLinus Torvalds __rb_link = &__rb_parent->rb_left; 4661da177e4SLinus Torvalds } else { 4671da177e4SLinus Torvalds rb_prev = __rb_parent; 4681da177e4SLinus Torvalds __rb_link = &__rb_parent->rb_right; 4691da177e4SLinus Torvalds } 4701da177e4SLinus Torvalds } 4711da177e4SLinus Torvalds 4721da177e4SLinus Torvalds *pprev = NULL; 4731da177e4SLinus Torvalds if (rb_prev) 4741da177e4SLinus Torvalds *pprev = rb_entry(rb_prev, struct vm_area_struct, vm_rb); 4751da177e4SLinus Torvalds *rb_link = __rb_link; 4761da177e4SLinus Torvalds *rb_parent = __rb_parent; 4776597d783SHugh Dickins return 0; 4781da177e4SLinus Torvalds } 4791da177e4SLinus Torvalds 480e8420a8eSCyril Hrubis static unsigned long count_vma_pages_range(struct mm_struct *mm, 481e8420a8eSCyril Hrubis unsigned long addr, unsigned long end) 482e8420a8eSCyril Hrubis { 483e8420a8eSCyril Hrubis unsigned long nr_pages = 0; 484e8420a8eSCyril Hrubis struct vm_area_struct *vma; 485e8420a8eSCyril Hrubis 486e8420a8eSCyril Hrubis /* Find first overlaping mapping */ 487e8420a8eSCyril Hrubis vma = find_vma_intersection(mm, addr, end); 488e8420a8eSCyril Hrubis if (!vma) 489e8420a8eSCyril Hrubis return 0; 490e8420a8eSCyril Hrubis 491e8420a8eSCyril Hrubis nr_pages = (min(end, vma->vm_end) - 492e8420a8eSCyril Hrubis max(addr, vma->vm_start)) >> PAGE_SHIFT; 493e8420a8eSCyril Hrubis 494e8420a8eSCyril Hrubis /* Iterate over the rest of the overlaps */ 495e8420a8eSCyril Hrubis for (vma = vma->vm_next; vma; vma = vma->vm_next) { 496e8420a8eSCyril Hrubis unsigned long overlap_len; 497e8420a8eSCyril Hrubis 498e8420a8eSCyril Hrubis if (vma->vm_start > end) 499e8420a8eSCyril Hrubis break; 500e8420a8eSCyril Hrubis 501e8420a8eSCyril Hrubis overlap_len = min(end, vma->vm_end) - vma->vm_start; 502e8420a8eSCyril Hrubis nr_pages += overlap_len >> PAGE_SHIFT; 503e8420a8eSCyril Hrubis } 504e8420a8eSCyril Hrubis 505e8420a8eSCyril Hrubis return nr_pages; 506e8420a8eSCyril Hrubis } 507e8420a8eSCyril Hrubis 5081da177e4SLinus Torvalds void __vma_link_rb(struct mm_struct *mm, struct vm_area_struct *vma, 5091da177e4SLinus Torvalds struct rb_node **rb_link, struct rb_node *rb_parent) 5101da177e4SLinus Torvalds { 511d3737187SMichel Lespinasse /* Update tracking information for the gap following the new vma. */ 512d3737187SMichel Lespinasse if (vma->vm_next) 513d3737187SMichel Lespinasse vma_gap_update(vma->vm_next); 514d3737187SMichel Lespinasse else 515d3737187SMichel Lespinasse mm->highest_vm_end = vma->vm_end; 516d3737187SMichel Lespinasse 517d3737187SMichel Lespinasse /* 518d3737187SMichel Lespinasse * vma->vm_prev wasn't known when we followed the rbtree to find the 519d3737187SMichel Lespinasse * correct insertion point for that vma. As a result, we could not 520d3737187SMichel Lespinasse * update the vma vm_rb parents rb_subtree_gap values on the way down. 521d3737187SMichel Lespinasse * So, we first insert the vma with a zero rb_subtree_gap value 522d3737187SMichel Lespinasse * (to be consistent with what we did on the way down), and then 523d3737187SMichel Lespinasse * immediately update the gap to the correct value. Finally we 524d3737187SMichel Lespinasse * rebalance the rbtree after all augmented values have been set. 525d3737187SMichel Lespinasse */ 5261da177e4SLinus Torvalds rb_link_node(&vma->vm_rb, rb_parent, rb_link); 527d3737187SMichel Lespinasse vma->rb_subtree_gap = 0; 528d3737187SMichel Lespinasse vma_gap_update(vma); 529d3737187SMichel Lespinasse vma_rb_insert(vma, &mm->mm_rb); 5301da177e4SLinus Torvalds } 5311da177e4SLinus Torvalds 532cb8f488cSDenys Vlasenko static void __vma_link_file(struct vm_area_struct *vma) 5331da177e4SLinus Torvalds { 5341da177e4SLinus Torvalds struct file *file; 5351da177e4SLinus Torvalds 5361da177e4SLinus Torvalds file = vma->vm_file; 5371da177e4SLinus Torvalds if (file) { 5381da177e4SLinus Torvalds struct address_space *mapping = file->f_mapping; 5391da177e4SLinus Torvalds 5401da177e4SLinus Torvalds if (vma->vm_flags & VM_DENYWRITE) 541496ad9aaSAl Viro atomic_dec(&file_inode(file)->i_writecount); 5421da177e4SLinus Torvalds if (vma->vm_flags & VM_SHARED) 5434bb5f5d9SDavid Herrmann atomic_inc(&mapping->i_mmap_writable); 5441da177e4SLinus Torvalds 5451da177e4SLinus Torvalds flush_dcache_mmap_lock(mapping); 5466b2dbba8SMichel Lespinasse vma_interval_tree_insert(vma, &mapping->i_mmap); 5471da177e4SLinus Torvalds flush_dcache_mmap_unlock(mapping); 5481da177e4SLinus Torvalds } 5491da177e4SLinus Torvalds } 5501da177e4SLinus Torvalds 5511da177e4SLinus Torvalds static void 5521da177e4SLinus Torvalds __vma_link(struct mm_struct *mm, struct vm_area_struct *vma, 5531da177e4SLinus Torvalds struct vm_area_struct *prev, struct rb_node **rb_link, 5541da177e4SLinus Torvalds struct rb_node *rb_parent) 5551da177e4SLinus Torvalds { 5561da177e4SLinus Torvalds __vma_link_list(mm, vma, prev, rb_parent); 5571da177e4SLinus Torvalds __vma_link_rb(mm, vma, rb_link, rb_parent); 5581da177e4SLinus Torvalds } 5591da177e4SLinus Torvalds 5601da177e4SLinus Torvalds static void vma_link(struct mm_struct *mm, struct vm_area_struct *vma, 5611da177e4SLinus Torvalds struct vm_area_struct *prev, struct rb_node **rb_link, 5621da177e4SLinus Torvalds struct rb_node *rb_parent) 5631da177e4SLinus Torvalds { 5641da177e4SLinus Torvalds struct address_space *mapping = NULL; 5651da177e4SLinus Torvalds 56664ac4940SHuang Shijie if (vma->vm_file) { 5671da177e4SLinus Torvalds mapping = vma->vm_file->f_mapping; 56883cde9e8SDavidlohr Bueso i_mmap_lock_write(mapping); 56964ac4940SHuang Shijie } 5701da177e4SLinus Torvalds 5711da177e4SLinus Torvalds __vma_link(mm, vma, prev, rb_link, rb_parent); 5721da177e4SLinus Torvalds __vma_link_file(vma); 5731da177e4SLinus Torvalds 5741da177e4SLinus Torvalds if (mapping) 57583cde9e8SDavidlohr Bueso i_mmap_unlock_write(mapping); 5761da177e4SLinus Torvalds 5771da177e4SLinus Torvalds mm->map_count++; 5781da177e4SLinus Torvalds validate_mm(mm); 5791da177e4SLinus Torvalds } 5801da177e4SLinus Torvalds 5811da177e4SLinus Torvalds /* 58288f6b4c3SKautuk Consul * Helper for vma_adjust() in the split_vma insert case: insert a vma into the 5836b2dbba8SMichel Lespinasse * mm's list and rbtree. It has already been inserted into the interval tree. 5841da177e4SLinus Torvalds */ 58548aae425SZhenwenXu static void __insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma) 5861da177e4SLinus Torvalds { 5876597d783SHugh Dickins struct vm_area_struct *prev; 5881da177e4SLinus Torvalds struct rb_node **rb_link, *rb_parent; 5891da177e4SLinus Torvalds 5906597d783SHugh Dickins if (find_vma_links(mm, vma->vm_start, vma->vm_end, 5916597d783SHugh Dickins &prev, &rb_link, &rb_parent)) 5926597d783SHugh Dickins BUG(); 5931da177e4SLinus Torvalds __vma_link(mm, vma, prev, rb_link, rb_parent); 5941da177e4SLinus Torvalds mm->map_count++; 5951da177e4SLinus Torvalds } 5961da177e4SLinus Torvalds 5971da177e4SLinus Torvalds static inline void 5981da177e4SLinus Torvalds __vma_unlink(struct mm_struct *mm, struct vm_area_struct *vma, 5991da177e4SLinus Torvalds struct vm_area_struct *prev) 6001da177e4SLinus Torvalds { 601d3737187SMichel Lespinasse struct vm_area_struct *next; 602297c5eeeSLinus Torvalds 603d3737187SMichel Lespinasse vma_rb_erase(vma, &mm->mm_rb); 604d3737187SMichel Lespinasse prev->vm_next = next = vma->vm_next; 605297c5eeeSLinus Torvalds if (next) 606297c5eeeSLinus Torvalds next->vm_prev = prev; 607615d6e87SDavidlohr Bueso 608615d6e87SDavidlohr Bueso /* Kill the cache */ 609615d6e87SDavidlohr Bueso vmacache_invalidate(mm); 6101da177e4SLinus Torvalds } 6111da177e4SLinus Torvalds 6121da177e4SLinus Torvalds /* 6131da177e4SLinus Torvalds * We cannot adjust vm_start, vm_end, vm_pgoff fields of a vma that 6141da177e4SLinus Torvalds * is already present in an i_mmap tree without adjusting the tree. 6151da177e4SLinus Torvalds * The following helper function should be used when such adjustments 6161da177e4SLinus Torvalds * are necessary. The "insert" vma (if any) is to be inserted 6171da177e4SLinus Torvalds * before we drop the necessary locks. 6181da177e4SLinus Torvalds */ 6195beb4930SRik van Riel int vma_adjust(struct vm_area_struct *vma, unsigned long start, 6201da177e4SLinus Torvalds unsigned long end, pgoff_t pgoff, struct vm_area_struct *insert) 6211da177e4SLinus Torvalds { 6221da177e4SLinus Torvalds struct mm_struct *mm = vma->vm_mm; 6231da177e4SLinus Torvalds struct vm_area_struct *next = vma->vm_next; 6241da177e4SLinus Torvalds struct vm_area_struct *importer = NULL; 6251da177e4SLinus Torvalds struct address_space *mapping = NULL; 6266b2dbba8SMichel Lespinasse struct rb_root *root = NULL; 627012f1800SRik van Riel struct anon_vma *anon_vma = NULL; 6281da177e4SLinus Torvalds struct file *file = vma->vm_file; 629d3737187SMichel Lespinasse bool start_changed = false, end_changed = false; 6301da177e4SLinus Torvalds long adjust_next = 0; 6311da177e4SLinus Torvalds int remove_next = 0; 6321da177e4SLinus Torvalds 6331da177e4SLinus Torvalds if (next && !insert) { 634287d97acSLinus Torvalds struct vm_area_struct *exporter = NULL; 635287d97acSLinus Torvalds 6361da177e4SLinus Torvalds if (end >= next->vm_end) { 6371da177e4SLinus Torvalds /* 6381da177e4SLinus Torvalds * vma expands, overlapping all the next, and 6391da177e4SLinus Torvalds * perhaps the one after too (mprotect case 6). 6401da177e4SLinus Torvalds */ 6411da177e4SLinus Torvalds again: remove_next = 1 + (end > next->vm_end); 6421da177e4SLinus Torvalds end = next->vm_end; 643287d97acSLinus Torvalds exporter = next; 6441da177e4SLinus Torvalds importer = vma; 6451da177e4SLinus Torvalds } else if (end > next->vm_start) { 6461da177e4SLinus Torvalds /* 6471da177e4SLinus Torvalds * vma expands, overlapping part of the next: 6481da177e4SLinus Torvalds * mprotect case 5 shifting the boundary up. 6491da177e4SLinus Torvalds */ 6501da177e4SLinus Torvalds adjust_next = (end - next->vm_start) >> PAGE_SHIFT; 651287d97acSLinus Torvalds exporter = next; 6521da177e4SLinus Torvalds importer = vma; 6531da177e4SLinus Torvalds } else if (end < vma->vm_end) { 6541da177e4SLinus Torvalds /* 6551da177e4SLinus Torvalds * vma shrinks, and !insert tells it's not 6561da177e4SLinus Torvalds * split_vma inserting another: so it must be 6571da177e4SLinus Torvalds * mprotect case 4 shifting the boundary down. 6581da177e4SLinus Torvalds */ 6591da177e4SLinus Torvalds adjust_next = -((vma->vm_end - end) >> PAGE_SHIFT); 660287d97acSLinus Torvalds exporter = vma; 6611da177e4SLinus Torvalds importer = next; 6621da177e4SLinus Torvalds } 6631da177e4SLinus Torvalds 6645beb4930SRik van Riel /* 6655beb4930SRik van Riel * Easily overlooked: when mprotect shifts the boundary, 6665beb4930SRik van Riel * make sure the expanding vma has anon_vma set if the 6675beb4930SRik van Riel * shrinking vma had, to cover any anon pages imported. 6685beb4930SRik van Riel */ 669287d97acSLinus Torvalds if (exporter && exporter->anon_vma && !importer->anon_vma) { 670c4ea95d7SDaniel Forrest int error; 671c4ea95d7SDaniel Forrest 672287d97acSLinus Torvalds importer->anon_vma = exporter->anon_vma; 673b800c91aSKonstantin Khlebnikov error = anon_vma_clone(importer, exporter); 6743fe89b3eSLeon Yu if (error) 675b800c91aSKonstantin Khlebnikov return error; 676b800c91aSKonstantin Khlebnikov } 6775beb4930SRik van Riel } 6785beb4930SRik van Riel 67937f9f559SKirill A. Shutemov vma_adjust_trans_huge(vma, start, end, adjust_next); 68037f9f559SKirill A. Shutemov 6811da177e4SLinus Torvalds if (file) { 6821da177e4SLinus Torvalds mapping = file->f_mapping; 6831da177e4SLinus Torvalds root = &mapping->i_mmap; 684cbc91f71SSrikar Dronamraju uprobe_munmap(vma, vma->vm_start, vma->vm_end); 685682968e0SSrikar Dronamraju 686682968e0SSrikar Dronamraju if (adjust_next) 68727ba0644SKirill A. Shutemov uprobe_munmap(next, next->vm_start, next->vm_end); 688682968e0SSrikar Dronamraju 68983cde9e8SDavidlohr Bueso i_mmap_lock_write(mapping); 6901da177e4SLinus Torvalds if (insert) { 6911da177e4SLinus Torvalds /* 6926b2dbba8SMichel Lespinasse * Put into interval tree now, so instantiated pages 6931da177e4SLinus Torvalds * are visible to arm/parisc __flush_dcache_page 6941da177e4SLinus Torvalds * throughout; but we cannot insert into address 6951da177e4SLinus Torvalds * space until vma start or end is updated. 6961da177e4SLinus Torvalds */ 6971da177e4SLinus Torvalds __vma_link_file(insert); 6981da177e4SLinus Torvalds } 6991da177e4SLinus Torvalds } 7001da177e4SLinus Torvalds 701012f1800SRik van Riel anon_vma = vma->anon_vma; 702bf181b9fSMichel Lespinasse if (!anon_vma && adjust_next) 703bf181b9fSMichel Lespinasse anon_vma = next->anon_vma; 704bf181b9fSMichel Lespinasse if (anon_vma) { 70581d1b09cSSasha Levin VM_BUG_ON_VMA(adjust_next && next->anon_vma && 70681d1b09cSSasha Levin anon_vma != next->anon_vma, next); 7074fc3f1d6SIngo Molnar anon_vma_lock_write(anon_vma); 708bf181b9fSMichel Lespinasse anon_vma_interval_tree_pre_update_vma(vma); 709bf181b9fSMichel Lespinasse if (adjust_next) 710bf181b9fSMichel Lespinasse anon_vma_interval_tree_pre_update_vma(next); 711bf181b9fSMichel Lespinasse } 712012f1800SRik van Riel 7131da177e4SLinus Torvalds if (root) { 7141da177e4SLinus Torvalds flush_dcache_mmap_lock(mapping); 7156b2dbba8SMichel Lespinasse vma_interval_tree_remove(vma, root); 7161da177e4SLinus Torvalds if (adjust_next) 7176b2dbba8SMichel Lespinasse vma_interval_tree_remove(next, root); 7181da177e4SLinus Torvalds } 7191da177e4SLinus Torvalds 720d3737187SMichel Lespinasse if (start != vma->vm_start) { 7211da177e4SLinus Torvalds vma->vm_start = start; 722d3737187SMichel Lespinasse start_changed = true; 723d3737187SMichel Lespinasse } 724d3737187SMichel Lespinasse if (end != vma->vm_end) { 7251da177e4SLinus Torvalds vma->vm_end = end; 726d3737187SMichel Lespinasse end_changed = true; 727d3737187SMichel Lespinasse } 7281da177e4SLinus Torvalds vma->vm_pgoff = pgoff; 7291da177e4SLinus Torvalds if (adjust_next) { 7301da177e4SLinus Torvalds next->vm_start += adjust_next << PAGE_SHIFT; 7311da177e4SLinus Torvalds next->vm_pgoff += adjust_next; 7321da177e4SLinus Torvalds } 7331da177e4SLinus Torvalds 7341da177e4SLinus Torvalds if (root) { 7351da177e4SLinus Torvalds if (adjust_next) 7366b2dbba8SMichel Lespinasse vma_interval_tree_insert(next, root); 7376b2dbba8SMichel Lespinasse vma_interval_tree_insert(vma, root); 7381da177e4SLinus Torvalds flush_dcache_mmap_unlock(mapping); 7391da177e4SLinus Torvalds } 7401da177e4SLinus Torvalds 7411da177e4SLinus Torvalds if (remove_next) { 7421da177e4SLinus Torvalds /* 7431da177e4SLinus Torvalds * vma_merge has merged next into vma, and needs 7441da177e4SLinus Torvalds * us to remove next before dropping the locks. 7451da177e4SLinus Torvalds */ 7461da177e4SLinus Torvalds __vma_unlink(mm, next, vma); 7471da177e4SLinus Torvalds if (file) 7481da177e4SLinus Torvalds __remove_shared_vm_struct(next, file, mapping); 7491da177e4SLinus Torvalds } else if (insert) { 7501da177e4SLinus Torvalds /* 7511da177e4SLinus Torvalds * split_vma has split insert from vma, and needs 7521da177e4SLinus Torvalds * us to insert it before dropping the locks 7531da177e4SLinus Torvalds * (it may either follow vma or precede it). 7541da177e4SLinus Torvalds */ 7551da177e4SLinus Torvalds __insert_vm_struct(mm, insert); 756d3737187SMichel Lespinasse } else { 757d3737187SMichel Lespinasse if (start_changed) 758d3737187SMichel Lespinasse vma_gap_update(vma); 759d3737187SMichel Lespinasse if (end_changed) { 760d3737187SMichel Lespinasse if (!next) 761d3737187SMichel Lespinasse mm->highest_vm_end = end; 762d3737187SMichel Lespinasse else if (!adjust_next) 763d3737187SMichel Lespinasse vma_gap_update(next); 764d3737187SMichel Lespinasse } 7651da177e4SLinus Torvalds } 7661da177e4SLinus Torvalds 767bf181b9fSMichel Lespinasse if (anon_vma) { 768bf181b9fSMichel Lespinasse anon_vma_interval_tree_post_update_vma(vma); 769bf181b9fSMichel Lespinasse if (adjust_next) 770bf181b9fSMichel Lespinasse anon_vma_interval_tree_post_update_vma(next); 77108b52706SKonstantin Khlebnikov anon_vma_unlock_write(anon_vma); 772bf181b9fSMichel Lespinasse } 7731da177e4SLinus Torvalds if (mapping) 77483cde9e8SDavidlohr Bueso i_mmap_unlock_write(mapping); 7751da177e4SLinus Torvalds 7762b144498SSrikar Dronamraju if (root) { 7777b2d81d4SIngo Molnar uprobe_mmap(vma); 7782b144498SSrikar Dronamraju 7792b144498SSrikar Dronamraju if (adjust_next) 7807b2d81d4SIngo Molnar uprobe_mmap(next); 7812b144498SSrikar Dronamraju } 7822b144498SSrikar Dronamraju 7831da177e4SLinus Torvalds if (remove_next) { 784925d1c40SMatt Helsley if (file) { 785cbc91f71SSrikar Dronamraju uprobe_munmap(next, next->vm_start, next->vm_end); 7861da177e4SLinus Torvalds fput(file); 787925d1c40SMatt Helsley } 7885beb4930SRik van Riel if (next->anon_vma) 7895beb4930SRik van Riel anon_vma_merge(vma, next); 7901da177e4SLinus Torvalds mm->map_count--; 7913964acd0SOleg Nesterov mpol_put(vma_policy(next)); 7921da177e4SLinus Torvalds kmem_cache_free(vm_area_cachep, next); 7931da177e4SLinus Torvalds /* 7941da177e4SLinus Torvalds * In mprotect's case 6 (see comments on vma_merge), 7951da177e4SLinus Torvalds * we must remove another next too. It would clutter 7961da177e4SLinus Torvalds * up the code too much to do both in one go. 7971da177e4SLinus Torvalds */ 7981da177e4SLinus Torvalds next = vma->vm_next; 799d3737187SMichel Lespinasse if (remove_next == 2) 8001da177e4SLinus Torvalds goto again; 801d3737187SMichel Lespinasse else if (next) 802d3737187SMichel Lespinasse vma_gap_update(next); 803d3737187SMichel Lespinasse else 804d3737187SMichel Lespinasse mm->highest_vm_end = end; 8051da177e4SLinus Torvalds } 8062b144498SSrikar Dronamraju if (insert && file) 8077b2d81d4SIngo Molnar uprobe_mmap(insert); 8081da177e4SLinus Torvalds 8091da177e4SLinus Torvalds validate_mm(mm); 8105beb4930SRik van Riel 8115beb4930SRik van Riel return 0; 8121da177e4SLinus Torvalds } 8131da177e4SLinus Torvalds 8141da177e4SLinus Torvalds /* 8151da177e4SLinus Torvalds * If the vma has a ->close operation then the driver probably needs to release 8161da177e4SLinus Torvalds * per-vma resources, so we don't attempt to merge those. 8171da177e4SLinus Torvalds */ 8181da177e4SLinus Torvalds static inline int is_mergeable_vma(struct vm_area_struct *vma, 81919a809afSAndrea Arcangeli struct file *file, unsigned long vm_flags, 82019a809afSAndrea Arcangeli struct vm_userfaultfd_ctx vm_userfaultfd_ctx) 8211da177e4SLinus Torvalds { 82234228d47SCyrill Gorcunov /* 82334228d47SCyrill Gorcunov * VM_SOFTDIRTY should not prevent from VMA merging, if we 82434228d47SCyrill Gorcunov * match the flags but dirty bit -- the caller should mark 82534228d47SCyrill Gorcunov * merged VMA as dirty. If dirty bit won't be excluded from 82634228d47SCyrill Gorcunov * comparison, we increase pressue on the memory system forcing 82734228d47SCyrill Gorcunov * the kernel to generate new VMAs when old one could be 82834228d47SCyrill Gorcunov * extended instead. 82934228d47SCyrill Gorcunov */ 83034228d47SCyrill Gorcunov if ((vma->vm_flags ^ vm_flags) & ~VM_SOFTDIRTY) 8311da177e4SLinus Torvalds return 0; 8321da177e4SLinus Torvalds if (vma->vm_file != file) 8331da177e4SLinus Torvalds return 0; 8341da177e4SLinus Torvalds if (vma->vm_ops && vma->vm_ops->close) 8351da177e4SLinus Torvalds return 0; 83619a809afSAndrea Arcangeli if (!is_mergeable_vm_userfaultfd_ctx(vma, vm_userfaultfd_ctx)) 83719a809afSAndrea Arcangeli return 0; 8381da177e4SLinus Torvalds return 1; 8391da177e4SLinus Torvalds } 8401da177e4SLinus Torvalds 8411da177e4SLinus Torvalds static inline int is_mergeable_anon_vma(struct anon_vma *anon_vma1, 842965f55deSShaohua Li struct anon_vma *anon_vma2, 843965f55deSShaohua Li struct vm_area_struct *vma) 8441da177e4SLinus Torvalds { 845965f55deSShaohua Li /* 846965f55deSShaohua Li * The list_is_singular() test is to avoid merging VMA cloned from 847965f55deSShaohua Li * parents. This can improve scalability caused by anon_vma lock. 848965f55deSShaohua Li */ 849965f55deSShaohua Li if ((!anon_vma1 || !anon_vma2) && (!vma || 850965f55deSShaohua Li list_is_singular(&vma->anon_vma_chain))) 851965f55deSShaohua Li return 1; 852965f55deSShaohua Li return anon_vma1 == anon_vma2; 8531da177e4SLinus Torvalds } 8541da177e4SLinus Torvalds 8551da177e4SLinus Torvalds /* 8561da177e4SLinus Torvalds * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff) 8571da177e4SLinus Torvalds * in front of (at a lower virtual address and file offset than) the vma. 8581da177e4SLinus Torvalds * 8591da177e4SLinus Torvalds * We cannot merge two vmas if they have differently assigned (non-NULL) 8601da177e4SLinus Torvalds * anon_vmas, nor if same anon_vma is assigned but offsets incompatible. 8611da177e4SLinus Torvalds * 8621da177e4SLinus Torvalds * We don't check here for the merged mmap wrapping around the end of pagecache 8631da177e4SLinus Torvalds * indices (16TB on ia32) because do_mmap_pgoff() does not permit mmap's which 8641da177e4SLinus Torvalds * wrap, nor mmaps which cover the final page at index -1UL. 8651da177e4SLinus Torvalds */ 8661da177e4SLinus Torvalds static int 8671da177e4SLinus Torvalds can_vma_merge_before(struct vm_area_struct *vma, unsigned long vm_flags, 86819a809afSAndrea Arcangeli struct anon_vma *anon_vma, struct file *file, 86919a809afSAndrea Arcangeli pgoff_t vm_pgoff, 87019a809afSAndrea Arcangeli struct vm_userfaultfd_ctx vm_userfaultfd_ctx) 8711da177e4SLinus Torvalds { 87219a809afSAndrea Arcangeli if (is_mergeable_vma(vma, file, vm_flags, vm_userfaultfd_ctx) && 873965f55deSShaohua Li is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) { 8741da177e4SLinus Torvalds if (vma->vm_pgoff == vm_pgoff) 8751da177e4SLinus Torvalds return 1; 8761da177e4SLinus Torvalds } 8771da177e4SLinus Torvalds return 0; 8781da177e4SLinus Torvalds } 8791da177e4SLinus Torvalds 8801da177e4SLinus Torvalds /* 8811da177e4SLinus Torvalds * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff) 8821da177e4SLinus Torvalds * beyond (at a higher virtual address and file offset than) the vma. 8831da177e4SLinus Torvalds * 8841da177e4SLinus Torvalds * We cannot merge two vmas if they have differently assigned (non-NULL) 8851da177e4SLinus Torvalds * anon_vmas, nor if same anon_vma is assigned but offsets incompatible. 8861da177e4SLinus Torvalds */ 8871da177e4SLinus Torvalds static int 8881da177e4SLinus Torvalds can_vma_merge_after(struct vm_area_struct *vma, unsigned long vm_flags, 88919a809afSAndrea Arcangeli struct anon_vma *anon_vma, struct file *file, 89019a809afSAndrea Arcangeli pgoff_t vm_pgoff, 89119a809afSAndrea Arcangeli struct vm_userfaultfd_ctx vm_userfaultfd_ctx) 8921da177e4SLinus Torvalds { 89319a809afSAndrea Arcangeli if (is_mergeable_vma(vma, file, vm_flags, vm_userfaultfd_ctx) && 894965f55deSShaohua Li is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) { 8951da177e4SLinus Torvalds pgoff_t vm_pglen; 896d6e93217SLibin vm_pglen = vma_pages(vma); 8971da177e4SLinus Torvalds if (vma->vm_pgoff + vm_pglen == vm_pgoff) 8981da177e4SLinus Torvalds return 1; 8991da177e4SLinus Torvalds } 9001da177e4SLinus Torvalds return 0; 9011da177e4SLinus Torvalds } 9021da177e4SLinus Torvalds 9031da177e4SLinus Torvalds /* 9041da177e4SLinus Torvalds * Given a mapping request (addr,end,vm_flags,file,pgoff), figure out 9051da177e4SLinus Torvalds * whether that can be merged with its predecessor or its successor. 9061da177e4SLinus Torvalds * Or both (it neatly fills a hole). 9071da177e4SLinus Torvalds * 9081da177e4SLinus Torvalds * In most cases - when called for mmap, brk or mremap - [addr,end) is 9091da177e4SLinus Torvalds * certain not to be mapped by the time vma_merge is called; but when 9101da177e4SLinus Torvalds * called for mprotect, it is certain to be already mapped (either at 9111da177e4SLinus Torvalds * an offset within prev, or at the start of next), and the flags of 9121da177e4SLinus Torvalds * this area are about to be changed to vm_flags - and the no-change 9131da177e4SLinus Torvalds * case has already been eliminated. 9141da177e4SLinus Torvalds * 9151da177e4SLinus Torvalds * The following mprotect cases have to be considered, where AAAA is 9161da177e4SLinus Torvalds * the area passed down from mprotect_fixup, never extending beyond one 9171da177e4SLinus Torvalds * vma, PPPPPP is the prev vma specified, and NNNNNN the next vma after: 9181da177e4SLinus Torvalds * 9191da177e4SLinus Torvalds * AAAA AAAA AAAA AAAA 9201da177e4SLinus Torvalds * PPPPPPNNNNNN PPPPPPNNNNNN PPPPPPNNNNNN PPPPNNNNXXXX 9211da177e4SLinus Torvalds * cannot merge might become might become might become 9221da177e4SLinus Torvalds * PPNNNNNNNNNN PPPPPPPPPPNN PPPPPPPPPPPP 6 or 9231da177e4SLinus Torvalds * mmap, brk or case 4 below case 5 below PPPPPPPPXXXX 7 or 9241da177e4SLinus Torvalds * mremap move: PPPPNNNNNNNN 8 9251da177e4SLinus Torvalds * AAAA 9261da177e4SLinus Torvalds * PPPP NNNN PPPPPPPPPPPP PPPPPPPPNNNN PPPPNNNNNNNN 9271da177e4SLinus Torvalds * might become case 1 below case 2 below case 3 below 9281da177e4SLinus Torvalds * 9291da177e4SLinus Torvalds * Odd one out? Case 8, because it extends NNNN but needs flags of XXXX: 9301da177e4SLinus Torvalds * mprotect_fixup updates vm_flags & vm_page_prot on successful return. 9311da177e4SLinus Torvalds */ 9321da177e4SLinus Torvalds struct vm_area_struct *vma_merge(struct mm_struct *mm, 9331da177e4SLinus Torvalds struct vm_area_struct *prev, unsigned long addr, 9341da177e4SLinus Torvalds unsigned long end, unsigned long vm_flags, 9351da177e4SLinus Torvalds struct anon_vma *anon_vma, struct file *file, 93619a809afSAndrea Arcangeli pgoff_t pgoff, struct mempolicy *policy, 93719a809afSAndrea Arcangeli struct vm_userfaultfd_ctx vm_userfaultfd_ctx) 9381da177e4SLinus Torvalds { 9391da177e4SLinus Torvalds pgoff_t pglen = (end - addr) >> PAGE_SHIFT; 9401da177e4SLinus Torvalds struct vm_area_struct *area, *next; 9415beb4930SRik van Riel int err; 9421da177e4SLinus Torvalds 9431da177e4SLinus Torvalds /* 9441da177e4SLinus Torvalds * We later require that vma->vm_flags == vm_flags, 9451da177e4SLinus Torvalds * so this tests vma->vm_flags & VM_SPECIAL, too. 9461da177e4SLinus Torvalds */ 9471da177e4SLinus Torvalds if (vm_flags & VM_SPECIAL) 9481da177e4SLinus Torvalds return NULL; 9491da177e4SLinus Torvalds 9501da177e4SLinus Torvalds if (prev) 9511da177e4SLinus Torvalds next = prev->vm_next; 9521da177e4SLinus Torvalds else 9531da177e4SLinus Torvalds next = mm->mmap; 9541da177e4SLinus Torvalds area = next; 9551da177e4SLinus Torvalds if (next && next->vm_end == end) /* cases 6, 7, 8 */ 9561da177e4SLinus Torvalds next = next->vm_next; 9571da177e4SLinus Torvalds 9581da177e4SLinus Torvalds /* 9591da177e4SLinus Torvalds * Can it merge with the predecessor? 9601da177e4SLinus Torvalds */ 9611da177e4SLinus Torvalds if (prev && prev->vm_end == addr && 9621da177e4SLinus Torvalds mpol_equal(vma_policy(prev), policy) && 9631da177e4SLinus Torvalds can_vma_merge_after(prev, vm_flags, 96419a809afSAndrea Arcangeli anon_vma, file, pgoff, 96519a809afSAndrea Arcangeli vm_userfaultfd_ctx)) { 9661da177e4SLinus Torvalds /* 9671da177e4SLinus Torvalds * OK, it can. Can we now merge in the successor as well? 9681da177e4SLinus Torvalds */ 9691da177e4SLinus Torvalds if (next && end == next->vm_start && 9701da177e4SLinus Torvalds mpol_equal(policy, vma_policy(next)) && 9711da177e4SLinus Torvalds can_vma_merge_before(next, vm_flags, 97219a809afSAndrea Arcangeli anon_vma, file, 97319a809afSAndrea Arcangeli pgoff+pglen, 97419a809afSAndrea Arcangeli vm_userfaultfd_ctx) && 9751da177e4SLinus Torvalds is_mergeable_anon_vma(prev->anon_vma, 976965f55deSShaohua Li next->anon_vma, NULL)) { 9771da177e4SLinus Torvalds /* cases 1, 6 */ 9785beb4930SRik van Riel err = vma_adjust(prev, prev->vm_start, 9791da177e4SLinus Torvalds next->vm_end, prev->vm_pgoff, NULL); 9801da177e4SLinus Torvalds } else /* cases 2, 5, 7 */ 9815beb4930SRik van Riel err = vma_adjust(prev, prev->vm_start, 9821da177e4SLinus Torvalds end, prev->vm_pgoff, NULL); 9835beb4930SRik van Riel if (err) 9845beb4930SRik van Riel return NULL; 9856d50e60cSDavid Rientjes khugepaged_enter_vma_merge(prev, vm_flags); 9861da177e4SLinus Torvalds return prev; 9871da177e4SLinus Torvalds } 9881da177e4SLinus Torvalds 9891da177e4SLinus Torvalds /* 9901da177e4SLinus Torvalds * Can this new request be merged in front of next? 9911da177e4SLinus Torvalds */ 9921da177e4SLinus Torvalds if (next && end == next->vm_start && 9931da177e4SLinus Torvalds mpol_equal(policy, vma_policy(next)) && 9941da177e4SLinus Torvalds can_vma_merge_before(next, vm_flags, 99519a809afSAndrea Arcangeli anon_vma, file, pgoff+pglen, 99619a809afSAndrea Arcangeli vm_userfaultfd_ctx)) { 9971da177e4SLinus Torvalds if (prev && addr < prev->vm_end) /* case 4 */ 9985beb4930SRik van Riel err = vma_adjust(prev, prev->vm_start, 9991da177e4SLinus Torvalds addr, prev->vm_pgoff, NULL); 10001da177e4SLinus Torvalds else /* cases 3, 8 */ 10015beb4930SRik van Riel err = vma_adjust(area, addr, next->vm_end, 10021da177e4SLinus Torvalds next->vm_pgoff - pglen, NULL); 10035beb4930SRik van Riel if (err) 10045beb4930SRik van Riel return NULL; 10056d50e60cSDavid Rientjes khugepaged_enter_vma_merge(area, vm_flags); 10061da177e4SLinus Torvalds return area; 10071da177e4SLinus Torvalds } 10081da177e4SLinus Torvalds 10091da177e4SLinus Torvalds return NULL; 10101da177e4SLinus Torvalds } 10111da177e4SLinus Torvalds 10121da177e4SLinus Torvalds /* 1013d0e9fe17SLinus Torvalds * Rough compatbility check to quickly see if it's even worth looking 1014d0e9fe17SLinus Torvalds * at sharing an anon_vma. 1015d0e9fe17SLinus Torvalds * 1016d0e9fe17SLinus Torvalds * They need to have the same vm_file, and the flags can only differ 1017d0e9fe17SLinus Torvalds * in things that mprotect may change. 1018d0e9fe17SLinus Torvalds * 1019d0e9fe17SLinus Torvalds * NOTE! The fact that we share an anon_vma doesn't _have_ to mean that 1020d0e9fe17SLinus Torvalds * we can merge the two vma's. For example, we refuse to merge a vma if 1021d0e9fe17SLinus Torvalds * there is a vm_ops->close() function, because that indicates that the 1022d0e9fe17SLinus Torvalds * driver is doing some kind of reference counting. But that doesn't 1023d0e9fe17SLinus Torvalds * really matter for the anon_vma sharing case. 1024d0e9fe17SLinus Torvalds */ 1025d0e9fe17SLinus Torvalds static int anon_vma_compatible(struct vm_area_struct *a, struct vm_area_struct *b) 1026d0e9fe17SLinus Torvalds { 1027d0e9fe17SLinus Torvalds return a->vm_end == b->vm_start && 1028d0e9fe17SLinus Torvalds mpol_equal(vma_policy(a), vma_policy(b)) && 1029d0e9fe17SLinus Torvalds a->vm_file == b->vm_file && 103034228d47SCyrill Gorcunov !((a->vm_flags ^ b->vm_flags) & ~(VM_READ|VM_WRITE|VM_EXEC|VM_SOFTDIRTY)) && 1031d0e9fe17SLinus Torvalds b->vm_pgoff == a->vm_pgoff + ((b->vm_start - a->vm_start) >> PAGE_SHIFT); 1032d0e9fe17SLinus Torvalds } 1033d0e9fe17SLinus Torvalds 1034d0e9fe17SLinus Torvalds /* 1035d0e9fe17SLinus Torvalds * Do some basic sanity checking to see if we can re-use the anon_vma 1036d0e9fe17SLinus Torvalds * from 'old'. The 'a'/'b' vma's are in VM order - one of them will be 1037d0e9fe17SLinus Torvalds * the same as 'old', the other will be the new one that is trying 1038d0e9fe17SLinus Torvalds * to share the anon_vma. 1039d0e9fe17SLinus Torvalds * 1040d0e9fe17SLinus Torvalds * NOTE! This runs with mm_sem held for reading, so it is possible that 1041d0e9fe17SLinus Torvalds * the anon_vma of 'old' is concurrently in the process of being set up 1042d0e9fe17SLinus Torvalds * by another page fault trying to merge _that_. But that's ok: if it 1043d0e9fe17SLinus Torvalds * is being set up, that automatically means that it will be a singleton 1044d0e9fe17SLinus Torvalds * acceptable for merging, so we can do all of this optimistically. But 10454db0c3c2SJason Low * we do that READ_ONCE() to make sure that we never re-load the pointer. 1046d0e9fe17SLinus Torvalds * 1047d0e9fe17SLinus Torvalds * IOW: that the "list_is_singular()" test on the anon_vma_chain only 1048d0e9fe17SLinus Torvalds * matters for the 'stable anon_vma' case (ie the thing we want to avoid 1049d0e9fe17SLinus Torvalds * is to return an anon_vma that is "complex" due to having gone through 1050d0e9fe17SLinus Torvalds * a fork). 1051d0e9fe17SLinus Torvalds * 1052d0e9fe17SLinus Torvalds * We also make sure that the two vma's are compatible (adjacent, 1053d0e9fe17SLinus Torvalds * and with the same memory policies). That's all stable, even with just 1054d0e9fe17SLinus Torvalds * a read lock on the mm_sem. 1055d0e9fe17SLinus Torvalds */ 1056d0e9fe17SLinus Torvalds static struct anon_vma *reusable_anon_vma(struct vm_area_struct *old, struct vm_area_struct *a, struct vm_area_struct *b) 1057d0e9fe17SLinus Torvalds { 1058d0e9fe17SLinus Torvalds if (anon_vma_compatible(a, b)) { 10594db0c3c2SJason Low struct anon_vma *anon_vma = READ_ONCE(old->anon_vma); 1060d0e9fe17SLinus Torvalds 1061d0e9fe17SLinus Torvalds if (anon_vma && list_is_singular(&old->anon_vma_chain)) 1062d0e9fe17SLinus Torvalds return anon_vma; 1063d0e9fe17SLinus Torvalds } 1064d0e9fe17SLinus Torvalds return NULL; 1065d0e9fe17SLinus Torvalds } 1066d0e9fe17SLinus Torvalds 1067d0e9fe17SLinus Torvalds /* 10681da177e4SLinus Torvalds * find_mergeable_anon_vma is used by anon_vma_prepare, to check 10691da177e4SLinus Torvalds * neighbouring vmas for a suitable anon_vma, before it goes off 10701da177e4SLinus Torvalds * to allocate a new anon_vma. It checks because a repetitive 10711da177e4SLinus Torvalds * sequence of mprotects and faults may otherwise lead to distinct 10721da177e4SLinus Torvalds * anon_vmas being allocated, preventing vma merge in subsequent 10731da177e4SLinus Torvalds * mprotect. 10741da177e4SLinus Torvalds */ 10751da177e4SLinus Torvalds struct anon_vma *find_mergeable_anon_vma(struct vm_area_struct *vma) 10761da177e4SLinus Torvalds { 1077d0e9fe17SLinus Torvalds struct anon_vma *anon_vma; 10781da177e4SLinus Torvalds struct vm_area_struct *near; 10791da177e4SLinus Torvalds 10801da177e4SLinus Torvalds near = vma->vm_next; 10811da177e4SLinus Torvalds if (!near) 10821da177e4SLinus Torvalds goto try_prev; 10831da177e4SLinus Torvalds 1084d0e9fe17SLinus Torvalds anon_vma = reusable_anon_vma(near, vma, near); 1085d0e9fe17SLinus Torvalds if (anon_vma) 1086d0e9fe17SLinus Torvalds return anon_vma; 10871da177e4SLinus Torvalds try_prev: 10889be34c9dSLinus Torvalds near = vma->vm_prev; 10891da177e4SLinus Torvalds if (!near) 10901da177e4SLinus Torvalds goto none; 10911da177e4SLinus Torvalds 1092d0e9fe17SLinus Torvalds anon_vma = reusable_anon_vma(near, near, vma); 1093d0e9fe17SLinus Torvalds if (anon_vma) 1094d0e9fe17SLinus Torvalds return anon_vma; 10951da177e4SLinus Torvalds none: 10961da177e4SLinus Torvalds /* 10971da177e4SLinus Torvalds * There's no absolute need to look only at touching neighbours: 10981da177e4SLinus Torvalds * we could search further afield for "compatible" anon_vmas. 10991da177e4SLinus Torvalds * But it would probably just be a waste of time searching, 11001da177e4SLinus Torvalds * or lead to too many vmas hanging off the same anon_vma. 11011da177e4SLinus Torvalds * We're trying to allow mprotect remerging later on, 11021da177e4SLinus Torvalds * not trying to minimize memory used for anon_vmas. 11031da177e4SLinus Torvalds */ 11041da177e4SLinus Torvalds return NULL; 11051da177e4SLinus Torvalds } 11061da177e4SLinus Torvalds 11071da177e4SLinus Torvalds /* 110840401530SAl Viro * If a hint addr is less than mmap_min_addr change hint to be as 110940401530SAl Viro * low as possible but still greater than mmap_min_addr 111040401530SAl Viro */ 111140401530SAl Viro static inline unsigned long round_hint_to_min(unsigned long hint) 111240401530SAl Viro { 111340401530SAl Viro hint &= PAGE_MASK; 111440401530SAl Viro if (((void *)hint != NULL) && 111540401530SAl Viro (hint < mmap_min_addr)) 111640401530SAl Viro return PAGE_ALIGN(mmap_min_addr); 111740401530SAl Viro return hint; 111840401530SAl Viro } 111940401530SAl Viro 1120363ee17fSDavidlohr Bueso static inline int mlock_future_check(struct mm_struct *mm, 1121363ee17fSDavidlohr Bueso unsigned long flags, 1122363ee17fSDavidlohr Bueso unsigned long len) 1123363ee17fSDavidlohr Bueso { 1124363ee17fSDavidlohr Bueso unsigned long locked, lock_limit; 1125363ee17fSDavidlohr Bueso 1126363ee17fSDavidlohr Bueso /* mlock MCL_FUTURE? */ 1127363ee17fSDavidlohr Bueso if (flags & VM_LOCKED) { 1128363ee17fSDavidlohr Bueso locked = len >> PAGE_SHIFT; 1129363ee17fSDavidlohr Bueso locked += mm->locked_vm; 1130363ee17fSDavidlohr Bueso lock_limit = rlimit(RLIMIT_MEMLOCK); 1131363ee17fSDavidlohr Bueso lock_limit >>= PAGE_SHIFT; 1132363ee17fSDavidlohr Bueso if (locked > lock_limit && !capable(CAP_IPC_LOCK)) 1133363ee17fSDavidlohr Bueso return -EAGAIN; 1134363ee17fSDavidlohr Bueso } 1135363ee17fSDavidlohr Bueso return 0; 1136363ee17fSDavidlohr Bueso } 1137363ee17fSDavidlohr Bueso 113840401530SAl Viro /* 113927f5de79SJianjun Kong * The caller must hold down_write(¤t->mm->mmap_sem). 11401da177e4SLinus Torvalds */ 11411fcfd8dbSOleg Nesterov unsigned long do_mmap(struct file *file, unsigned long addr, 11421da177e4SLinus Torvalds unsigned long len, unsigned long prot, 11431fcfd8dbSOleg Nesterov unsigned long flags, vm_flags_t vm_flags, 11441fcfd8dbSOleg Nesterov unsigned long pgoff, unsigned long *populate) 11451da177e4SLinus Torvalds { 11461da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 114762b5f7d0SDave Hansen int pkey = 0; 11481da177e4SLinus Torvalds 114941badc15SMichel Lespinasse *populate = 0; 1150bebeb3d6SMichel Lespinasse 1151e37609bbSPiotr Kwapulinski if (!len) 1152e37609bbSPiotr Kwapulinski return -EINVAL; 1153e37609bbSPiotr Kwapulinski 11541da177e4SLinus Torvalds /* 11551da177e4SLinus Torvalds * Does the application expect PROT_READ to imply PROT_EXEC? 11561da177e4SLinus Torvalds * 11571da177e4SLinus Torvalds * (the exception is when the underlying filesystem is noexec 11581da177e4SLinus Torvalds * mounted, in which case we dont add PROT_EXEC.) 11591da177e4SLinus Torvalds */ 11601da177e4SLinus Torvalds if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC)) 116190f8572bSEric W. Biederman if (!(file && path_noexec(&file->f_path))) 11621da177e4SLinus Torvalds prot |= PROT_EXEC; 11631da177e4SLinus Torvalds 11647cd94146SEric Paris if (!(flags & MAP_FIXED)) 11657cd94146SEric Paris addr = round_hint_to_min(addr); 11667cd94146SEric Paris 11671da177e4SLinus Torvalds /* Careful about overflows.. */ 11681da177e4SLinus Torvalds len = PAGE_ALIGN(len); 11699206de95SAl Viro if (!len) 11701da177e4SLinus Torvalds return -ENOMEM; 11711da177e4SLinus Torvalds 11721da177e4SLinus Torvalds /* offset overflow? */ 11731da177e4SLinus Torvalds if ((pgoff + (len >> PAGE_SHIFT)) < pgoff) 11741da177e4SLinus Torvalds return -EOVERFLOW; 11751da177e4SLinus Torvalds 11761da177e4SLinus Torvalds /* Too many mappings? */ 11771da177e4SLinus Torvalds if (mm->map_count > sysctl_max_map_count) 11781da177e4SLinus Torvalds return -ENOMEM; 11791da177e4SLinus Torvalds 11801da177e4SLinus Torvalds /* Obtain the address to map to. we verify (or select) it and ensure 11811da177e4SLinus Torvalds * that it represents a valid section of the address space. 11821da177e4SLinus Torvalds */ 11831da177e4SLinus Torvalds addr = get_unmapped_area(file, addr, len, pgoff, flags); 1184de1741a1SAlexander Kuleshov if (offset_in_page(addr)) 11851da177e4SLinus Torvalds return addr; 11861da177e4SLinus Torvalds 118762b5f7d0SDave Hansen if (prot == PROT_EXEC) { 118862b5f7d0SDave Hansen pkey = execute_only_pkey(mm); 118962b5f7d0SDave Hansen if (pkey < 0) 119062b5f7d0SDave Hansen pkey = 0; 119162b5f7d0SDave Hansen } 119262b5f7d0SDave Hansen 11931da177e4SLinus Torvalds /* Do simple checking here so the lower-level routines won't have 11941da177e4SLinus Torvalds * to. we assume access permissions have been handled by the open 11951da177e4SLinus Torvalds * of the memory object, so we don't do any here. 11961da177e4SLinus Torvalds */ 119762b5f7d0SDave Hansen vm_flags |= calc_vm_prot_bits(prot, pkey) | calc_vm_flag_bits(flags) | 11981da177e4SLinus Torvalds mm->def_flags | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC; 11991da177e4SLinus Torvalds 1200cdf7b341SHuang Shijie if (flags & MAP_LOCKED) 12011da177e4SLinus Torvalds if (!can_do_mlock()) 12021da177e4SLinus Torvalds return -EPERM; 1203ba470de4SRik van Riel 1204363ee17fSDavidlohr Bueso if (mlock_future_check(mm, vm_flags, len)) 12051da177e4SLinus Torvalds return -EAGAIN; 12061da177e4SLinus Torvalds 12071da177e4SLinus Torvalds if (file) { 1208077bf22bSOleg Nesterov struct inode *inode = file_inode(file); 1209077bf22bSOleg Nesterov 12101da177e4SLinus Torvalds switch (flags & MAP_TYPE) { 12111da177e4SLinus Torvalds case MAP_SHARED: 12121da177e4SLinus Torvalds if ((prot&PROT_WRITE) && !(file->f_mode&FMODE_WRITE)) 12131da177e4SLinus Torvalds return -EACCES; 12141da177e4SLinus Torvalds 12151da177e4SLinus Torvalds /* 12161da177e4SLinus Torvalds * Make sure we don't allow writing to an append-only 12171da177e4SLinus Torvalds * file.. 12181da177e4SLinus Torvalds */ 12191da177e4SLinus Torvalds if (IS_APPEND(inode) && (file->f_mode & FMODE_WRITE)) 12201da177e4SLinus Torvalds return -EACCES; 12211da177e4SLinus Torvalds 12221da177e4SLinus Torvalds /* 12231da177e4SLinus Torvalds * Make sure there are no mandatory locks on the file. 12241da177e4SLinus Torvalds */ 1225d7a06983SJeff Layton if (locks_verify_locked(file)) 12261da177e4SLinus Torvalds return -EAGAIN; 12271da177e4SLinus Torvalds 12281da177e4SLinus Torvalds vm_flags |= VM_SHARED | VM_MAYSHARE; 12291da177e4SLinus Torvalds if (!(file->f_mode & FMODE_WRITE)) 12301da177e4SLinus Torvalds vm_flags &= ~(VM_MAYWRITE | VM_SHARED); 12311da177e4SLinus Torvalds 12321da177e4SLinus Torvalds /* fall through */ 12331da177e4SLinus Torvalds case MAP_PRIVATE: 12341da177e4SLinus Torvalds if (!(file->f_mode & FMODE_READ)) 12351da177e4SLinus Torvalds return -EACCES; 123690f8572bSEric W. Biederman if (path_noexec(&file->f_path)) { 123780c5606cSLinus Torvalds if (vm_flags & VM_EXEC) 123880c5606cSLinus Torvalds return -EPERM; 123980c5606cSLinus Torvalds vm_flags &= ~VM_MAYEXEC; 124080c5606cSLinus Torvalds } 124180c5606cSLinus Torvalds 124272c2d531SAl Viro if (!file->f_op->mmap) 124380c5606cSLinus Torvalds return -ENODEV; 1244b2c56e4fSOleg Nesterov if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP)) 1245b2c56e4fSOleg Nesterov return -EINVAL; 12461da177e4SLinus Torvalds break; 12471da177e4SLinus Torvalds 12481da177e4SLinus Torvalds default: 12491da177e4SLinus Torvalds return -EINVAL; 12501da177e4SLinus Torvalds } 12511da177e4SLinus Torvalds } else { 12521da177e4SLinus Torvalds switch (flags & MAP_TYPE) { 12531da177e4SLinus Torvalds case MAP_SHARED: 1254b2c56e4fSOleg Nesterov if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP)) 1255b2c56e4fSOleg Nesterov return -EINVAL; 1256ce363942STejun Heo /* 1257ce363942STejun Heo * Ignore pgoff. 1258ce363942STejun Heo */ 1259ce363942STejun Heo pgoff = 0; 12601da177e4SLinus Torvalds vm_flags |= VM_SHARED | VM_MAYSHARE; 12611da177e4SLinus Torvalds break; 12621da177e4SLinus Torvalds case MAP_PRIVATE: 12631da177e4SLinus Torvalds /* 12641da177e4SLinus Torvalds * Set pgoff according to addr for anon_vma. 12651da177e4SLinus Torvalds */ 12661da177e4SLinus Torvalds pgoff = addr >> PAGE_SHIFT; 12671da177e4SLinus Torvalds break; 12681da177e4SLinus Torvalds default: 12691da177e4SLinus Torvalds return -EINVAL; 12701da177e4SLinus Torvalds } 12711da177e4SLinus Torvalds } 12721da177e4SLinus Torvalds 1273c22c0d63SMichel Lespinasse /* 1274c22c0d63SMichel Lespinasse * Set 'VM_NORESERVE' if we should not account for the 1275c22c0d63SMichel Lespinasse * memory use of this mapping. 1276c22c0d63SMichel Lespinasse */ 1277c22c0d63SMichel Lespinasse if (flags & MAP_NORESERVE) { 1278c22c0d63SMichel Lespinasse /* We honor MAP_NORESERVE if allowed to overcommit */ 1279c22c0d63SMichel Lespinasse if (sysctl_overcommit_memory != OVERCOMMIT_NEVER) 1280c22c0d63SMichel Lespinasse vm_flags |= VM_NORESERVE; 1281c22c0d63SMichel Lespinasse 1282c22c0d63SMichel Lespinasse /* hugetlb applies strict overcommit unless MAP_NORESERVE */ 1283c22c0d63SMichel Lespinasse if (file && is_file_hugepages(file)) 1284c22c0d63SMichel Lespinasse vm_flags |= VM_NORESERVE; 1285c22c0d63SMichel Lespinasse } 1286c22c0d63SMichel Lespinasse 1287c22c0d63SMichel Lespinasse addr = mmap_region(file, addr, len, vm_flags, pgoff); 128809a9f1d2SMichel Lespinasse if (!IS_ERR_VALUE(addr) && 128909a9f1d2SMichel Lespinasse ((vm_flags & VM_LOCKED) || 129009a9f1d2SMichel Lespinasse (flags & (MAP_POPULATE | MAP_NONBLOCK)) == MAP_POPULATE)) 129141badc15SMichel Lespinasse *populate = len; 1292bebeb3d6SMichel Lespinasse return addr; 12930165ab44SMiklos Szeredi } 12946be5ceb0SLinus Torvalds 129566f0dc48SHugh Dickins SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len, 129666f0dc48SHugh Dickins unsigned long, prot, unsigned long, flags, 129766f0dc48SHugh Dickins unsigned long, fd, unsigned long, pgoff) 129866f0dc48SHugh Dickins { 129966f0dc48SHugh Dickins struct file *file = NULL; 13001e3ee14bSChen Gang unsigned long retval; 130166f0dc48SHugh Dickins 130266f0dc48SHugh Dickins if (!(flags & MAP_ANONYMOUS)) { 1303120a795dSAl Viro audit_mmap_fd(fd, flags); 130466f0dc48SHugh Dickins file = fget(fd); 130566f0dc48SHugh Dickins if (!file) 13061e3ee14bSChen Gang return -EBADF; 1307af73e4d9SNaoya Horiguchi if (is_file_hugepages(file)) 1308af73e4d9SNaoya Horiguchi len = ALIGN(len, huge_page_size(hstate_file(file))); 1309493af578SJörn Engel retval = -EINVAL; 1310493af578SJörn Engel if (unlikely(flags & MAP_HUGETLB && !is_file_hugepages(file))) 1311493af578SJörn Engel goto out_fput; 131266f0dc48SHugh Dickins } else if (flags & MAP_HUGETLB) { 131366f0dc48SHugh Dickins struct user_struct *user = NULL; 1314c103a4dcSAndrew Morton struct hstate *hs; 1315af73e4d9SNaoya Horiguchi 1316c103a4dcSAndrew Morton hs = hstate_sizelog((flags >> MAP_HUGE_SHIFT) & SHM_HUGE_MASK); 1317091d0d55SLi Zefan if (!hs) 1318091d0d55SLi Zefan return -EINVAL; 1319091d0d55SLi Zefan 1320091d0d55SLi Zefan len = ALIGN(len, huge_page_size(hs)); 132166f0dc48SHugh Dickins /* 132266f0dc48SHugh Dickins * VM_NORESERVE is used because the reservations will be 132366f0dc48SHugh Dickins * taken when vm_ops->mmap() is called 132466f0dc48SHugh Dickins * A dummy user value is used because we are not locking 132566f0dc48SHugh Dickins * memory so no accounting is necessary 132666f0dc48SHugh Dickins */ 1327af73e4d9SNaoya Horiguchi file = hugetlb_file_setup(HUGETLB_ANON_FILE, len, 132842d7395fSAndi Kleen VM_NORESERVE, 132942d7395fSAndi Kleen &user, HUGETLB_ANONHUGE_INODE, 133042d7395fSAndi Kleen (flags >> MAP_HUGE_SHIFT) & MAP_HUGE_MASK); 133166f0dc48SHugh Dickins if (IS_ERR(file)) 133266f0dc48SHugh Dickins return PTR_ERR(file); 133366f0dc48SHugh Dickins } 133466f0dc48SHugh Dickins 133566f0dc48SHugh Dickins flags &= ~(MAP_EXECUTABLE | MAP_DENYWRITE); 133666f0dc48SHugh Dickins 13379fbeb5abSMichal Hocko retval = vm_mmap_pgoff(file, addr, len, prot, flags, pgoff); 1338493af578SJörn Engel out_fput: 133966f0dc48SHugh Dickins if (file) 134066f0dc48SHugh Dickins fput(file); 134166f0dc48SHugh Dickins return retval; 134266f0dc48SHugh Dickins } 134366f0dc48SHugh Dickins 1344a4679373SChristoph Hellwig #ifdef __ARCH_WANT_SYS_OLD_MMAP 1345a4679373SChristoph Hellwig struct mmap_arg_struct { 1346a4679373SChristoph Hellwig unsigned long addr; 1347a4679373SChristoph Hellwig unsigned long len; 1348a4679373SChristoph Hellwig unsigned long prot; 1349a4679373SChristoph Hellwig unsigned long flags; 1350a4679373SChristoph Hellwig unsigned long fd; 1351a4679373SChristoph Hellwig unsigned long offset; 1352a4679373SChristoph Hellwig }; 1353a4679373SChristoph Hellwig 1354a4679373SChristoph Hellwig SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg) 1355a4679373SChristoph Hellwig { 1356a4679373SChristoph Hellwig struct mmap_arg_struct a; 1357a4679373SChristoph Hellwig 1358a4679373SChristoph Hellwig if (copy_from_user(&a, arg, sizeof(a))) 1359a4679373SChristoph Hellwig return -EFAULT; 1360de1741a1SAlexander Kuleshov if (offset_in_page(a.offset)) 1361a4679373SChristoph Hellwig return -EINVAL; 1362a4679373SChristoph Hellwig 1363a4679373SChristoph Hellwig return sys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd, 1364a4679373SChristoph Hellwig a.offset >> PAGE_SHIFT); 1365a4679373SChristoph Hellwig } 1366a4679373SChristoph Hellwig #endif /* __ARCH_WANT_SYS_OLD_MMAP */ 1367a4679373SChristoph Hellwig 13684e950f6fSAlexey Dobriyan /* 13694e950f6fSAlexey Dobriyan * Some shared mappigns will want the pages marked read-only 13704e950f6fSAlexey Dobriyan * to track write events. If so, we'll downgrade vm_page_prot 13714e950f6fSAlexey Dobriyan * to the private version (using protection_map[] without the 13724e950f6fSAlexey Dobriyan * VM_SHARED bit). 13734e950f6fSAlexey Dobriyan */ 13744e950f6fSAlexey Dobriyan int vma_wants_writenotify(struct vm_area_struct *vma) 13754e950f6fSAlexey Dobriyan { 1376ca16d140SKOSAKI Motohiro vm_flags_t vm_flags = vma->vm_flags; 13778a04446aSKirill A. Shutemov const struct vm_operations_struct *vm_ops = vma->vm_ops; 13784e950f6fSAlexey Dobriyan 13794e950f6fSAlexey Dobriyan /* If it was private or non-writable, the write bit is already clear */ 13804e950f6fSAlexey Dobriyan if ((vm_flags & (VM_WRITE|VM_SHARED)) != ((VM_WRITE|VM_SHARED))) 13814e950f6fSAlexey Dobriyan return 0; 13824e950f6fSAlexey Dobriyan 13834e950f6fSAlexey Dobriyan /* The backer wishes to know when pages are first written to? */ 13848a04446aSKirill A. Shutemov if (vm_ops && (vm_ops->page_mkwrite || vm_ops->pfn_mkwrite)) 13854e950f6fSAlexey Dobriyan return 1; 13864e950f6fSAlexey Dobriyan 138764e45507SPeter Feiner /* The open routine did something to the protections that pgprot_modify 138864e45507SPeter Feiner * won't preserve? */ 13894e950f6fSAlexey Dobriyan if (pgprot_val(vma->vm_page_prot) != 139064e45507SPeter Feiner pgprot_val(vm_pgprot_modify(vma->vm_page_prot, vm_flags))) 13914e950f6fSAlexey Dobriyan return 0; 13924e950f6fSAlexey Dobriyan 139364e45507SPeter Feiner /* Do we need to track softdirty? */ 139464e45507SPeter Feiner if (IS_ENABLED(CONFIG_MEM_SOFT_DIRTY) && !(vm_flags & VM_SOFTDIRTY)) 139564e45507SPeter Feiner return 1; 139664e45507SPeter Feiner 13974e950f6fSAlexey Dobriyan /* Specialty mapping? */ 13984b6e1e37SKonstantin Khlebnikov if (vm_flags & VM_PFNMAP) 13994e950f6fSAlexey Dobriyan return 0; 14004e950f6fSAlexey Dobriyan 14014e950f6fSAlexey Dobriyan /* Can the mapping track the dirty pages? */ 14024e950f6fSAlexey Dobriyan return vma->vm_file && vma->vm_file->f_mapping && 14034e950f6fSAlexey Dobriyan mapping_cap_account_dirty(vma->vm_file->f_mapping); 14044e950f6fSAlexey Dobriyan } 14054e950f6fSAlexey Dobriyan 1406fc8744adSLinus Torvalds /* 1407fc8744adSLinus Torvalds * We account for memory if it's a private writeable mapping, 14085a6fe125SMel Gorman * not hugepages and VM_NORESERVE wasn't set. 1409fc8744adSLinus Torvalds */ 1410ca16d140SKOSAKI Motohiro static inline int accountable_mapping(struct file *file, vm_flags_t vm_flags) 1411fc8744adSLinus Torvalds { 14125a6fe125SMel Gorman /* 14135a6fe125SMel Gorman * hugetlb has its own accounting separate from the core VM 14145a6fe125SMel Gorman * VM_HUGETLB may not be set yet so we cannot check for that flag. 14155a6fe125SMel Gorman */ 14165a6fe125SMel Gorman if (file && is_file_hugepages(file)) 14175a6fe125SMel Gorman return 0; 14185a6fe125SMel Gorman 1419fc8744adSLinus Torvalds return (vm_flags & (VM_NORESERVE | VM_SHARED | VM_WRITE)) == VM_WRITE; 1420fc8744adSLinus Torvalds } 1421fc8744adSLinus Torvalds 14220165ab44SMiklos Szeredi unsigned long mmap_region(struct file *file, unsigned long addr, 1423c22c0d63SMichel Lespinasse unsigned long len, vm_flags_t vm_flags, unsigned long pgoff) 14240165ab44SMiklos Szeredi { 14250165ab44SMiklos Szeredi struct mm_struct *mm = current->mm; 14260165ab44SMiklos Szeredi struct vm_area_struct *vma, *prev; 14270165ab44SMiklos Szeredi int error; 14280165ab44SMiklos Szeredi struct rb_node **rb_link, *rb_parent; 14290165ab44SMiklos Szeredi unsigned long charged = 0; 14300165ab44SMiklos Szeredi 1431e8420a8eSCyril Hrubis /* Check against address space limit. */ 143284638335SKonstantin Khlebnikov if (!may_expand_vm(mm, vm_flags, len >> PAGE_SHIFT)) { 1433e8420a8eSCyril Hrubis unsigned long nr_pages; 1434e8420a8eSCyril Hrubis 1435e8420a8eSCyril Hrubis /* 1436e8420a8eSCyril Hrubis * MAP_FIXED may remove pages of mappings that intersects with 1437e8420a8eSCyril Hrubis * requested mapping. Account for the pages it would unmap. 1438e8420a8eSCyril Hrubis */ 1439e8420a8eSCyril Hrubis nr_pages = count_vma_pages_range(mm, addr, addr + len); 1440e8420a8eSCyril Hrubis 144184638335SKonstantin Khlebnikov if (!may_expand_vm(mm, vm_flags, 144284638335SKonstantin Khlebnikov (len >> PAGE_SHIFT) - nr_pages)) 1443e8420a8eSCyril Hrubis return -ENOMEM; 1444e8420a8eSCyril Hrubis } 1445e8420a8eSCyril Hrubis 14461da177e4SLinus Torvalds /* Clear old maps */ 14479fcd1457SRasmus Villemoes while (find_vma_links(mm, addr, addr + len, &prev, &rb_link, 14489fcd1457SRasmus Villemoes &rb_parent)) { 14491da177e4SLinus Torvalds if (do_munmap(mm, addr, len)) 14501da177e4SLinus Torvalds return -ENOMEM; 14511da177e4SLinus Torvalds } 14521da177e4SLinus Torvalds 1453fc8744adSLinus Torvalds /* 14541da177e4SLinus Torvalds * Private writable mapping: check memory availability 14551da177e4SLinus Torvalds */ 14565a6fe125SMel Gorman if (accountable_mapping(file, vm_flags)) { 14571da177e4SLinus Torvalds charged = len >> PAGE_SHIFT; 1458191c5424SAl Viro if (security_vm_enough_memory_mm(mm, charged)) 14591da177e4SLinus Torvalds return -ENOMEM; 14601da177e4SLinus Torvalds vm_flags |= VM_ACCOUNT; 14611da177e4SLinus Torvalds } 14621da177e4SLinus Torvalds 14631da177e4SLinus Torvalds /* 1464de33c8dbSLinus Torvalds * Can we just expand an old mapping? 14651da177e4SLinus Torvalds */ 146619a809afSAndrea Arcangeli vma = vma_merge(mm, prev, addr, addr + len, vm_flags, 146719a809afSAndrea Arcangeli NULL, file, pgoff, NULL, NULL_VM_UFFD_CTX); 1468ba470de4SRik van Riel if (vma) 14691da177e4SLinus Torvalds goto out; 14701da177e4SLinus Torvalds 14711da177e4SLinus Torvalds /* 14721da177e4SLinus Torvalds * Determine the object being mapped and call the appropriate 14731da177e4SLinus Torvalds * specific mapper. the address has already been validated, but 14741da177e4SLinus Torvalds * not unmapped, but the maps are removed from the list. 14751da177e4SLinus Torvalds */ 1476c5e3b83eSPekka Enberg vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL); 14771da177e4SLinus Torvalds if (!vma) { 14781da177e4SLinus Torvalds error = -ENOMEM; 14791da177e4SLinus Torvalds goto unacct_error; 14801da177e4SLinus Torvalds } 14811da177e4SLinus Torvalds 14821da177e4SLinus Torvalds vma->vm_mm = mm; 14831da177e4SLinus Torvalds vma->vm_start = addr; 14841da177e4SLinus Torvalds vma->vm_end = addr + len; 14851da177e4SLinus Torvalds vma->vm_flags = vm_flags; 14863ed75eb8SColy Li vma->vm_page_prot = vm_get_page_prot(vm_flags); 14871da177e4SLinus Torvalds vma->vm_pgoff = pgoff; 14885beb4930SRik van Riel INIT_LIST_HEAD(&vma->anon_vma_chain); 14891da177e4SLinus Torvalds 14901da177e4SLinus Torvalds if (file) { 14911da177e4SLinus Torvalds if (vm_flags & VM_DENYWRITE) { 14921da177e4SLinus Torvalds error = deny_write_access(file); 14931da177e4SLinus Torvalds if (error) 14941da177e4SLinus Torvalds goto free_vma; 14951da177e4SLinus Torvalds } 14964bb5f5d9SDavid Herrmann if (vm_flags & VM_SHARED) { 14974bb5f5d9SDavid Herrmann error = mapping_map_writable(file->f_mapping); 14984bb5f5d9SDavid Herrmann if (error) 14994bb5f5d9SDavid Herrmann goto allow_write_and_free_vma; 15004bb5f5d9SDavid Herrmann } 15014bb5f5d9SDavid Herrmann 15024bb5f5d9SDavid Herrmann /* ->mmap() can change vma->vm_file, but must guarantee that 15034bb5f5d9SDavid Herrmann * vma_link() below can deny write-access if VM_DENYWRITE is set 15044bb5f5d9SDavid Herrmann * and map writably if VM_SHARED is set. This usually means the 15054bb5f5d9SDavid Herrmann * new file must not have been exposed to user-space, yet. 15064bb5f5d9SDavid Herrmann */ 1507cb0942b8SAl Viro vma->vm_file = get_file(file); 15081da177e4SLinus Torvalds error = file->f_op->mmap(file, vma); 15091da177e4SLinus Torvalds if (error) 15101da177e4SLinus Torvalds goto unmap_and_free_vma; 15111da177e4SLinus Torvalds 15121da177e4SLinus Torvalds /* Can addr have changed?? 15131da177e4SLinus Torvalds * 15141da177e4SLinus Torvalds * Answer: Yes, several device drivers can do it in their 15151da177e4SLinus Torvalds * f_op->mmap method. -DaveM 15162897b4d2SJoonsoo Kim * Bug: If addr is changed, prev, rb_link, rb_parent should 15172897b4d2SJoonsoo Kim * be updated for vma_link() 15181da177e4SLinus Torvalds */ 15192897b4d2SJoonsoo Kim WARN_ON_ONCE(addr != vma->vm_start); 15202897b4d2SJoonsoo Kim 15211da177e4SLinus Torvalds addr = vma->vm_start; 15221da177e4SLinus Torvalds vm_flags = vma->vm_flags; 1523f8dbf0a7SHuang Shijie } else if (vm_flags & VM_SHARED) { 1524f8dbf0a7SHuang Shijie error = shmem_zero_setup(vma); 1525f8dbf0a7SHuang Shijie if (error) 1526f8dbf0a7SHuang Shijie goto free_vma; 1527f8dbf0a7SHuang Shijie } 15281da177e4SLinus Torvalds 15294d3d5b41SOleg Nesterov vma_link(mm, vma, prev, rb_link, rb_parent); 15304d3d5b41SOleg Nesterov /* Once vma denies write, undo our temporary denial count */ 15314bb5f5d9SDavid Herrmann if (file) { 15324bb5f5d9SDavid Herrmann if (vm_flags & VM_SHARED) 15334bb5f5d9SDavid Herrmann mapping_unmap_writable(file->f_mapping); 1534e8686772SOleg Nesterov if (vm_flags & VM_DENYWRITE) 1535e8686772SOleg Nesterov allow_write_access(file); 15364bb5f5d9SDavid Herrmann } 1537e8686772SOleg Nesterov file = vma->vm_file; 15381da177e4SLinus Torvalds out: 1539cdd6c482SIngo Molnar perf_event_mmap(vma); 15400a4a9391SPeter Zijlstra 154184638335SKonstantin Khlebnikov vm_stat_account(mm, vm_flags, len >> PAGE_SHIFT); 15421da177e4SLinus Torvalds if (vm_flags & VM_LOCKED) { 1543bebeb3d6SMichel Lespinasse if (!((vm_flags & VM_SPECIAL) || is_vm_hugetlb_page(vma) || 1544bebeb3d6SMichel Lespinasse vma == get_gate_vma(current->mm))) 154506f9d8c2SKOSAKI Motohiro mm->locked_vm += (len >> PAGE_SHIFT); 1546bebeb3d6SMichel Lespinasse else 1547de60f5f1SEric B Munson vma->vm_flags &= VM_LOCKED_CLEAR_MASK; 1548bebeb3d6SMichel Lespinasse } 15492b144498SSrikar Dronamraju 1550c7a3a88cSOleg Nesterov if (file) 1551c7a3a88cSOleg Nesterov uprobe_mmap(vma); 15522b144498SSrikar Dronamraju 1553d9104d1cSCyrill Gorcunov /* 1554d9104d1cSCyrill Gorcunov * New (or expanded) vma always get soft dirty status. 1555d9104d1cSCyrill Gorcunov * Otherwise user-space soft-dirty page tracker won't 1556d9104d1cSCyrill Gorcunov * be able to distinguish situation when vma area unmapped, 1557d9104d1cSCyrill Gorcunov * then new mapped in-place (which must be aimed as 1558d9104d1cSCyrill Gorcunov * a completely new data area). 1559d9104d1cSCyrill Gorcunov */ 1560d9104d1cSCyrill Gorcunov vma->vm_flags |= VM_SOFTDIRTY; 1561d9104d1cSCyrill Gorcunov 156264e45507SPeter Feiner vma_set_page_prot(vma); 156364e45507SPeter Feiner 15641da177e4SLinus Torvalds return addr; 15651da177e4SLinus Torvalds 15661da177e4SLinus Torvalds unmap_and_free_vma: 15671da177e4SLinus Torvalds vma->vm_file = NULL; 15681da177e4SLinus Torvalds fput(file); 15691da177e4SLinus Torvalds 15701da177e4SLinus Torvalds /* Undo any partial mapping done by a device driver. */ 1571e0da382cSHugh Dickins unmap_region(mm, vma, prev, vma->vm_start, vma->vm_end); 1572e0da382cSHugh Dickins charged = 0; 15734bb5f5d9SDavid Herrmann if (vm_flags & VM_SHARED) 15744bb5f5d9SDavid Herrmann mapping_unmap_writable(file->f_mapping); 15754bb5f5d9SDavid Herrmann allow_write_and_free_vma: 15764bb5f5d9SDavid Herrmann if (vm_flags & VM_DENYWRITE) 15774bb5f5d9SDavid Herrmann allow_write_access(file); 15781da177e4SLinus Torvalds free_vma: 15791da177e4SLinus Torvalds kmem_cache_free(vm_area_cachep, vma); 15801da177e4SLinus Torvalds unacct_error: 15811da177e4SLinus Torvalds if (charged) 15821da177e4SLinus Torvalds vm_unacct_memory(charged); 15831da177e4SLinus Torvalds return error; 15841da177e4SLinus Torvalds } 15851da177e4SLinus Torvalds 1586db4fbfb9SMichel Lespinasse unsigned long unmapped_area(struct vm_unmapped_area_info *info) 1587db4fbfb9SMichel Lespinasse { 1588db4fbfb9SMichel Lespinasse /* 1589db4fbfb9SMichel Lespinasse * We implement the search by looking for an rbtree node that 1590db4fbfb9SMichel Lespinasse * immediately follows a suitable gap. That is, 1591db4fbfb9SMichel Lespinasse * - gap_start = vma->vm_prev->vm_end <= info->high_limit - length; 1592db4fbfb9SMichel Lespinasse * - gap_end = vma->vm_start >= info->low_limit + length; 1593db4fbfb9SMichel Lespinasse * - gap_end - gap_start >= length 1594db4fbfb9SMichel Lespinasse */ 1595db4fbfb9SMichel Lespinasse 1596db4fbfb9SMichel Lespinasse struct mm_struct *mm = current->mm; 1597db4fbfb9SMichel Lespinasse struct vm_area_struct *vma; 1598db4fbfb9SMichel Lespinasse unsigned long length, low_limit, high_limit, gap_start, gap_end; 1599db4fbfb9SMichel Lespinasse 1600db4fbfb9SMichel Lespinasse /* Adjust search length to account for worst case alignment overhead */ 1601db4fbfb9SMichel Lespinasse length = info->length + info->align_mask; 1602db4fbfb9SMichel Lespinasse if (length < info->length) 1603db4fbfb9SMichel Lespinasse return -ENOMEM; 1604db4fbfb9SMichel Lespinasse 1605db4fbfb9SMichel Lespinasse /* Adjust search limits by the desired length */ 1606db4fbfb9SMichel Lespinasse if (info->high_limit < length) 1607db4fbfb9SMichel Lespinasse return -ENOMEM; 1608db4fbfb9SMichel Lespinasse high_limit = info->high_limit - length; 1609db4fbfb9SMichel Lespinasse 1610db4fbfb9SMichel Lespinasse if (info->low_limit > high_limit) 1611db4fbfb9SMichel Lespinasse return -ENOMEM; 1612db4fbfb9SMichel Lespinasse low_limit = info->low_limit + length; 1613db4fbfb9SMichel Lespinasse 1614db4fbfb9SMichel Lespinasse /* Check if rbtree root looks promising */ 1615db4fbfb9SMichel Lespinasse if (RB_EMPTY_ROOT(&mm->mm_rb)) 1616db4fbfb9SMichel Lespinasse goto check_highest; 1617db4fbfb9SMichel Lespinasse vma = rb_entry(mm->mm_rb.rb_node, struct vm_area_struct, vm_rb); 1618db4fbfb9SMichel Lespinasse if (vma->rb_subtree_gap < length) 1619db4fbfb9SMichel Lespinasse goto check_highest; 1620db4fbfb9SMichel Lespinasse 1621db4fbfb9SMichel Lespinasse while (true) { 1622db4fbfb9SMichel Lespinasse /* Visit left subtree if it looks promising */ 1623db4fbfb9SMichel Lespinasse gap_end = vma->vm_start; 1624db4fbfb9SMichel Lespinasse if (gap_end >= low_limit && vma->vm_rb.rb_left) { 1625db4fbfb9SMichel Lespinasse struct vm_area_struct *left = 1626db4fbfb9SMichel Lespinasse rb_entry(vma->vm_rb.rb_left, 1627db4fbfb9SMichel Lespinasse struct vm_area_struct, vm_rb); 1628db4fbfb9SMichel Lespinasse if (left->rb_subtree_gap >= length) { 1629db4fbfb9SMichel Lespinasse vma = left; 1630db4fbfb9SMichel Lespinasse continue; 1631db4fbfb9SMichel Lespinasse } 1632db4fbfb9SMichel Lespinasse } 1633db4fbfb9SMichel Lespinasse 1634db4fbfb9SMichel Lespinasse gap_start = vma->vm_prev ? vma->vm_prev->vm_end : 0; 1635db4fbfb9SMichel Lespinasse check_current: 1636db4fbfb9SMichel Lespinasse /* Check if current node has a suitable gap */ 1637db4fbfb9SMichel Lespinasse if (gap_start > high_limit) 1638db4fbfb9SMichel Lespinasse return -ENOMEM; 1639db4fbfb9SMichel Lespinasse if (gap_end >= low_limit && gap_end - gap_start >= length) 1640db4fbfb9SMichel Lespinasse goto found; 1641db4fbfb9SMichel Lespinasse 1642db4fbfb9SMichel Lespinasse /* Visit right subtree if it looks promising */ 1643db4fbfb9SMichel Lespinasse if (vma->vm_rb.rb_right) { 1644db4fbfb9SMichel Lespinasse struct vm_area_struct *right = 1645db4fbfb9SMichel Lespinasse rb_entry(vma->vm_rb.rb_right, 1646db4fbfb9SMichel Lespinasse struct vm_area_struct, vm_rb); 1647db4fbfb9SMichel Lespinasse if (right->rb_subtree_gap >= length) { 1648db4fbfb9SMichel Lespinasse vma = right; 1649db4fbfb9SMichel Lespinasse continue; 1650db4fbfb9SMichel Lespinasse } 1651db4fbfb9SMichel Lespinasse } 1652db4fbfb9SMichel Lespinasse 1653db4fbfb9SMichel Lespinasse /* Go back up the rbtree to find next candidate node */ 1654db4fbfb9SMichel Lespinasse while (true) { 1655db4fbfb9SMichel Lespinasse struct rb_node *prev = &vma->vm_rb; 1656db4fbfb9SMichel Lespinasse if (!rb_parent(prev)) 1657db4fbfb9SMichel Lespinasse goto check_highest; 1658db4fbfb9SMichel Lespinasse vma = rb_entry(rb_parent(prev), 1659db4fbfb9SMichel Lespinasse struct vm_area_struct, vm_rb); 1660db4fbfb9SMichel Lespinasse if (prev == vma->vm_rb.rb_left) { 1661db4fbfb9SMichel Lespinasse gap_start = vma->vm_prev->vm_end; 1662db4fbfb9SMichel Lespinasse gap_end = vma->vm_start; 1663db4fbfb9SMichel Lespinasse goto check_current; 1664db4fbfb9SMichel Lespinasse } 1665db4fbfb9SMichel Lespinasse } 1666db4fbfb9SMichel Lespinasse } 1667db4fbfb9SMichel Lespinasse 1668db4fbfb9SMichel Lespinasse check_highest: 1669db4fbfb9SMichel Lespinasse /* Check highest gap, which does not precede any rbtree node */ 1670db4fbfb9SMichel Lespinasse gap_start = mm->highest_vm_end; 1671db4fbfb9SMichel Lespinasse gap_end = ULONG_MAX; /* Only for VM_BUG_ON below */ 1672db4fbfb9SMichel Lespinasse if (gap_start > high_limit) 1673db4fbfb9SMichel Lespinasse return -ENOMEM; 1674db4fbfb9SMichel Lespinasse 1675db4fbfb9SMichel Lespinasse found: 1676db4fbfb9SMichel Lespinasse /* We found a suitable gap. Clip it with the original low_limit. */ 1677db4fbfb9SMichel Lespinasse if (gap_start < info->low_limit) 1678db4fbfb9SMichel Lespinasse gap_start = info->low_limit; 1679db4fbfb9SMichel Lespinasse 1680db4fbfb9SMichel Lespinasse /* Adjust gap address to the desired alignment */ 1681db4fbfb9SMichel Lespinasse gap_start += (info->align_offset - gap_start) & info->align_mask; 1682db4fbfb9SMichel Lespinasse 1683db4fbfb9SMichel Lespinasse VM_BUG_ON(gap_start + info->length > info->high_limit); 1684db4fbfb9SMichel Lespinasse VM_BUG_ON(gap_start + info->length > gap_end); 1685db4fbfb9SMichel Lespinasse return gap_start; 1686db4fbfb9SMichel Lespinasse } 1687db4fbfb9SMichel Lespinasse 1688db4fbfb9SMichel Lespinasse unsigned long unmapped_area_topdown(struct vm_unmapped_area_info *info) 1689db4fbfb9SMichel Lespinasse { 1690db4fbfb9SMichel Lespinasse struct mm_struct *mm = current->mm; 1691db4fbfb9SMichel Lespinasse struct vm_area_struct *vma; 1692db4fbfb9SMichel Lespinasse unsigned long length, low_limit, high_limit, gap_start, gap_end; 1693db4fbfb9SMichel Lespinasse 1694db4fbfb9SMichel Lespinasse /* Adjust search length to account for worst case alignment overhead */ 1695db4fbfb9SMichel Lespinasse length = info->length + info->align_mask; 1696db4fbfb9SMichel Lespinasse if (length < info->length) 1697db4fbfb9SMichel Lespinasse return -ENOMEM; 1698db4fbfb9SMichel Lespinasse 1699db4fbfb9SMichel Lespinasse /* 1700db4fbfb9SMichel Lespinasse * Adjust search limits by the desired length. 1701db4fbfb9SMichel Lespinasse * See implementation comment at top of unmapped_area(). 1702db4fbfb9SMichel Lespinasse */ 1703db4fbfb9SMichel Lespinasse gap_end = info->high_limit; 1704db4fbfb9SMichel Lespinasse if (gap_end < length) 1705db4fbfb9SMichel Lespinasse return -ENOMEM; 1706db4fbfb9SMichel Lespinasse high_limit = gap_end - length; 1707db4fbfb9SMichel Lespinasse 1708db4fbfb9SMichel Lespinasse if (info->low_limit > high_limit) 1709db4fbfb9SMichel Lespinasse return -ENOMEM; 1710db4fbfb9SMichel Lespinasse low_limit = info->low_limit + length; 1711db4fbfb9SMichel Lespinasse 1712db4fbfb9SMichel Lespinasse /* Check highest gap, which does not precede any rbtree node */ 1713db4fbfb9SMichel Lespinasse gap_start = mm->highest_vm_end; 1714db4fbfb9SMichel Lespinasse if (gap_start <= high_limit) 1715db4fbfb9SMichel Lespinasse goto found_highest; 1716db4fbfb9SMichel Lespinasse 1717db4fbfb9SMichel Lespinasse /* Check if rbtree root looks promising */ 1718db4fbfb9SMichel Lespinasse if (RB_EMPTY_ROOT(&mm->mm_rb)) 1719db4fbfb9SMichel Lespinasse return -ENOMEM; 1720db4fbfb9SMichel Lespinasse vma = rb_entry(mm->mm_rb.rb_node, struct vm_area_struct, vm_rb); 1721db4fbfb9SMichel Lespinasse if (vma->rb_subtree_gap < length) 1722db4fbfb9SMichel Lespinasse return -ENOMEM; 1723db4fbfb9SMichel Lespinasse 1724db4fbfb9SMichel Lespinasse while (true) { 1725db4fbfb9SMichel Lespinasse /* Visit right subtree if it looks promising */ 1726db4fbfb9SMichel Lespinasse gap_start = vma->vm_prev ? vma->vm_prev->vm_end : 0; 1727db4fbfb9SMichel Lespinasse if (gap_start <= high_limit && vma->vm_rb.rb_right) { 1728db4fbfb9SMichel Lespinasse struct vm_area_struct *right = 1729db4fbfb9SMichel Lespinasse rb_entry(vma->vm_rb.rb_right, 1730db4fbfb9SMichel Lespinasse struct vm_area_struct, vm_rb); 1731db4fbfb9SMichel Lespinasse if (right->rb_subtree_gap >= length) { 1732db4fbfb9SMichel Lespinasse vma = right; 1733db4fbfb9SMichel Lespinasse continue; 1734db4fbfb9SMichel Lespinasse } 1735db4fbfb9SMichel Lespinasse } 1736db4fbfb9SMichel Lespinasse 1737db4fbfb9SMichel Lespinasse check_current: 1738db4fbfb9SMichel Lespinasse /* Check if current node has a suitable gap */ 1739db4fbfb9SMichel Lespinasse gap_end = vma->vm_start; 1740db4fbfb9SMichel Lespinasse if (gap_end < low_limit) 1741db4fbfb9SMichel Lespinasse return -ENOMEM; 1742db4fbfb9SMichel Lespinasse if (gap_start <= high_limit && gap_end - gap_start >= length) 1743db4fbfb9SMichel Lespinasse goto found; 1744db4fbfb9SMichel Lespinasse 1745db4fbfb9SMichel Lespinasse /* Visit left subtree if it looks promising */ 1746db4fbfb9SMichel Lespinasse if (vma->vm_rb.rb_left) { 1747db4fbfb9SMichel Lespinasse struct vm_area_struct *left = 1748db4fbfb9SMichel Lespinasse rb_entry(vma->vm_rb.rb_left, 1749db4fbfb9SMichel Lespinasse struct vm_area_struct, vm_rb); 1750db4fbfb9SMichel Lespinasse if (left->rb_subtree_gap >= length) { 1751db4fbfb9SMichel Lespinasse vma = left; 1752db4fbfb9SMichel Lespinasse continue; 1753db4fbfb9SMichel Lespinasse } 1754db4fbfb9SMichel Lespinasse } 1755db4fbfb9SMichel Lespinasse 1756db4fbfb9SMichel Lespinasse /* Go back up the rbtree to find next candidate node */ 1757db4fbfb9SMichel Lespinasse while (true) { 1758db4fbfb9SMichel Lespinasse struct rb_node *prev = &vma->vm_rb; 1759db4fbfb9SMichel Lespinasse if (!rb_parent(prev)) 1760db4fbfb9SMichel Lespinasse return -ENOMEM; 1761db4fbfb9SMichel Lespinasse vma = rb_entry(rb_parent(prev), 1762db4fbfb9SMichel Lespinasse struct vm_area_struct, vm_rb); 1763db4fbfb9SMichel Lespinasse if (prev == vma->vm_rb.rb_right) { 1764db4fbfb9SMichel Lespinasse gap_start = vma->vm_prev ? 1765db4fbfb9SMichel Lespinasse vma->vm_prev->vm_end : 0; 1766db4fbfb9SMichel Lespinasse goto check_current; 1767db4fbfb9SMichel Lespinasse } 1768db4fbfb9SMichel Lespinasse } 1769db4fbfb9SMichel Lespinasse } 1770db4fbfb9SMichel Lespinasse 1771db4fbfb9SMichel Lespinasse found: 1772db4fbfb9SMichel Lespinasse /* We found a suitable gap. Clip it with the original high_limit. */ 1773db4fbfb9SMichel Lespinasse if (gap_end > info->high_limit) 1774db4fbfb9SMichel Lespinasse gap_end = info->high_limit; 1775db4fbfb9SMichel Lespinasse 1776db4fbfb9SMichel Lespinasse found_highest: 1777db4fbfb9SMichel Lespinasse /* Compute highest gap address at the desired alignment */ 1778db4fbfb9SMichel Lespinasse gap_end -= info->length; 1779db4fbfb9SMichel Lespinasse gap_end -= (gap_end - info->align_offset) & info->align_mask; 1780db4fbfb9SMichel Lespinasse 1781db4fbfb9SMichel Lespinasse VM_BUG_ON(gap_end < info->low_limit); 1782db4fbfb9SMichel Lespinasse VM_BUG_ON(gap_end < gap_start); 1783db4fbfb9SMichel Lespinasse return gap_end; 1784db4fbfb9SMichel Lespinasse } 1785db4fbfb9SMichel Lespinasse 17861da177e4SLinus Torvalds /* Get an address range which is currently unmapped. 17871da177e4SLinus Torvalds * For shmat() with addr=0. 17881da177e4SLinus Torvalds * 17891da177e4SLinus Torvalds * Ugly calling convention alert: 17901da177e4SLinus Torvalds * Return value with the low bits set means error value, 17911da177e4SLinus Torvalds * ie 17921da177e4SLinus Torvalds * if (ret & ~PAGE_MASK) 17931da177e4SLinus Torvalds * error = ret; 17941da177e4SLinus Torvalds * 17951da177e4SLinus Torvalds * This function "knows" that -ENOMEM has the bits set. 17961da177e4SLinus Torvalds */ 17971da177e4SLinus Torvalds #ifndef HAVE_ARCH_UNMAPPED_AREA 17981da177e4SLinus Torvalds unsigned long 17991da177e4SLinus Torvalds arch_get_unmapped_area(struct file *filp, unsigned long addr, 18001da177e4SLinus Torvalds unsigned long len, unsigned long pgoff, unsigned long flags) 18011da177e4SLinus Torvalds { 18021da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 18031da177e4SLinus Torvalds struct vm_area_struct *vma; 1804db4fbfb9SMichel Lespinasse struct vm_unmapped_area_info info; 18051da177e4SLinus Torvalds 18062afc745fSAkira Takeuchi if (len > TASK_SIZE - mmap_min_addr) 18071da177e4SLinus Torvalds return -ENOMEM; 18081da177e4SLinus Torvalds 180906abdfb4SBenjamin Herrenschmidt if (flags & MAP_FIXED) 181006abdfb4SBenjamin Herrenschmidt return addr; 181106abdfb4SBenjamin Herrenschmidt 18121da177e4SLinus Torvalds if (addr) { 18131da177e4SLinus Torvalds addr = PAGE_ALIGN(addr); 18141da177e4SLinus Torvalds vma = find_vma(mm, addr); 18152afc745fSAkira Takeuchi if (TASK_SIZE - len >= addr && addr >= mmap_min_addr && 18161da177e4SLinus Torvalds (!vma || addr + len <= vma->vm_start)) 18171da177e4SLinus Torvalds return addr; 18181da177e4SLinus Torvalds } 18191da177e4SLinus Torvalds 1820db4fbfb9SMichel Lespinasse info.flags = 0; 1821db4fbfb9SMichel Lespinasse info.length = len; 18224e99b021SHeiko Carstens info.low_limit = mm->mmap_base; 1823db4fbfb9SMichel Lespinasse info.high_limit = TASK_SIZE; 1824db4fbfb9SMichel Lespinasse info.align_mask = 0; 1825db4fbfb9SMichel Lespinasse return vm_unmapped_area(&info); 18261da177e4SLinus Torvalds } 18271da177e4SLinus Torvalds #endif 18281da177e4SLinus Torvalds 18291da177e4SLinus Torvalds /* 18301da177e4SLinus Torvalds * This mmap-allocator allocates new areas top-down from below the 18311da177e4SLinus Torvalds * stack's low limit (the base): 18321da177e4SLinus Torvalds */ 18331da177e4SLinus Torvalds #ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN 18341da177e4SLinus Torvalds unsigned long 18351da177e4SLinus Torvalds arch_get_unmapped_area_topdown(struct file *filp, const unsigned long addr0, 18361da177e4SLinus Torvalds const unsigned long len, const unsigned long pgoff, 18371da177e4SLinus Torvalds const unsigned long flags) 18381da177e4SLinus Torvalds { 18391da177e4SLinus Torvalds struct vm_area_struct *vma; 18401da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 1841db4fbfb9SMichel Lespinasse unsigned long addr = addr0; 1842db4fbfb9SMichel Lespinasse struct vm_unmapped_area_info info; 18431da177e4SLinus Torvalds 18441da177e4SLinus Torvalds /* requested length too big for entire address space */ 18452afc745fSAkira Takeuchi if (len > TASK_SIZE - mmap_min_addr) 18461da177e4SLinus Torvalds return -ENOMEM; 18471da177e4SLinus Torvalds 184806abdfb4SBenjamin Herrenschmidt if (flags & MAP_FIXED) 184906abdfb4SBenjamin Herrenschmidt return addr; 185006abdfb4SBenjamin Herrenschmidt 18511da177e4SLinus Torvalds /* requesting a specific address */ 18521da177e4SLinus Torvalds if (addr) { 18531da177e4SLinus Torvalds addr = PAGE_ALIGN(addr); 18541da177e4SLinus Torvalds vma = find_vma(mm, addr); 18552afc745fSAkira Takeuchi if (TASK_SIZE - len >= addr && addr >= mmap_min_addr && 18561da177e4SLinus Torvalds (!vma || addr + len <= vma->vm_start)) 18571da177e4SLinus Torvalds return addr; 18581da177e4SLinus Torvalds } 18591da177e4SLinus Torvalds 1860db4fbfb9SMichel Lespinasse info.flags = VM_UNMAPPED_AREA_TOPDOWN; 1861db4fbfb9SMichel Lespinasse info.length = len; 18622afc745fSAkira Takeuchi info.low_limit = max(PAGE_SIZE, mmap_min_addr); 1863db4fbfb9SMichel Lespinasse info.high_limit = mm->mmap_base; 1864db4fbfb9SMichel Lespinasse info.align_mask = 0; 1865db4fbfb9SMichel Lespinasse addr = vm_unmapped_area(&info); 1866b716ad95SXiao Guangrong 18671da177e4SLinus Torvalds /* 18681da177e4SLinus Torvalds * A failed mmap() very likely causes application failure, 18691da177e4SLinus Torvalds * so fall back to the bottom-up function here. This scenario 18701da177e4SLinus Torvalds * can happen with large stack limits and large mmap() 18711da177e4SLinus Torvalds * allocations. 18721da177e4SLinus Torvalds */ 1873de1741a1SAlexander Kuleshov if (offset_in_page(addr)) { 1874db4fbfb9SMichel Lespinasse VM_BUG_ON(addr != -ENOMEM); 1875db4fbfb9SMichel Lespinasse info.flags = 0; 1876db4fbfb9SMichel Lespinasse info.low_limit = TASK_UNMAPPED_BASE; 1877db4fbfb9SMichel Lespinasse info.high_limit = TASK_SIZE; 1878db4fbfb9SMichel Lespinasse addr = vm_unmapped_area(&info); 1879db4fbfb9SMichel Lespinasse } 18801da177e4SLinus Torvalds 18811da177e4SLinus Torvalds return addr; 18821da177e4SLinus Torvalds } 18831da177e4SLinus Torvalds #endif 18841da177e4SLinus Torvalds 18851da177e4SLinus Torvalds unsigned long 18861da177e4SLinus Torvalds get_unmapped_area(struct file *file, unsigned long addr, unsigned long len, 18871da177e4SLinus Torvalds unsigned long pgoff, unsigned long flags) 18881da177e4SLinus Torvalds { 188906abdfb4SBenjamin Herrenschmidt unsigned long (*get_area)(struct file *, unsigned long, 189006abdfb4SBenjamin Herrenschmidt unsigned long, unsigned long, unsigned long); 189107ab67c8SLinus Torvalds 18929206de95SAl Viro unsigned long error = arch_mmap_check(addr, len, flags); 18939206de95SAl Viro if (error) 18949206de95SAl Viro return error; 18959206de95SAl Viro 18969206de95SAl Viro /* Careful about overflows.. */ 18979206de95SAl Viro if (len > TASK_SIZE) 18989206de95SAl Viro return -ENOMEM; 18999206de95SAl Viro 190007ab67c8SLinus Torvalds get_area = current->mm->get_unmapped_area; 1901*c01d5b30SHugh Dickins if (file) { 1902*c01d5b30SHugh Dickins if (file->f_op->get_unmapped_area) 190307ab67c8SLinus Torvalds get_area = file->f_op->get_unmapped_area; 1904*c01d5b30SHugh Dickins } else if (flags & MAP_SHARED) { 1905*c01d5b30SHugh Dickins /* 1906*c01d5b30SHugh Dickins * mmap_region() will call shmem_zero_setup() to create a file, 1907*c01d5b30SHugh Dickins * so use shmem's get_unmapped_area in case it can be huge. 1908*c01d5b30SHugh Dickins * do_mmap_pgoff() will clear pgoff, so match alignment. 1909*c01d5b30SHugh Dickins */ 1910*c01d5b30SHugh Dickins pgoff = 0; 1911*c01d5b30SHugh Dickins get_area = shmem_get_unmapped_area; 1912*c01d5b30SHugh Dickins } 1913*c01d5b30SHugh Dickins 191407ab67c8SLinus Torvalds addr = get_area(file, addr, len, pgoff, flags); 191507ab67c8SLinus Torvalds if (IS_ERR_VALUE(addr)) 191607ab67c8SLinus Torvalds return addr; 191707ab67c8SLinus Torvalds 19181da177e4SLinus Torvalds if (addr > TASK_SIZE - len) 19191da177e4SLinus Torvalds return -ENOMEM; 1920de1741a1SAlexander Kuleshov if (offset_in_page(addr)) 19211da177e4SLinus Torvalds return -EINVAL; 192206abdfb4SBenjamin Herrenschmidt 19239ac4ed4bSAl Viro error = security_mmap_addr(addr); 19249ac4ed4bSAl Viro return error ? error : addr; 19251da177e4SLinus Torvalds } 19261da177e4SLinus Torvalds 19271da177e4SLinus Torvalds EXPORT_SYMBOL(get_unmapped_area); 19281da177e4SLinus Torvalds 19291da177e4SLinus Torvalds /* Look up the first VMA which satisfies addr < vm_end, NULL if none. */ 19301da177e4SLinus Torvalds struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr) 19311da177e4SLinus Torvalds { 1932615d6e87SDavidlohr Bueso struct rb_node *rb_node; 1933615d6e87SDavidlohr Bueso struct vm_area_struct *vma; 19341da177e4SLinus Torvalds 19351da177e4SLinus Torvalds /* Check the cache first. */ 1936615d6e87SDavidlohr Bueso vma = vmacache_find(mm, addr); 1937615d6e87SDavidlohr Bueso if (likely(vma)) 1938615d6e87SDavidlohr Bueso return vma; 19391da177e4SLinus Torvalds 19401da177e4SLinus Torvalds rb_node = mm->mm_rb.rb_node; 19411da177e4SLinus Torvalds 19421da177e4SLinus Torvalds while (rb_node) { 1943615d6e87SDavidlohr Bueso struct vm_area_struct *tmp; 19441da177e4SLinus Torvalds 1945615d6e87SDavidlohr Bueso tmp = rb_entry(rb_node, struct vm_area_struct, vm_rb); 19461da177e4SLinus Torvalds 1947615d6e87SDavidlohr Bueso if (tmp->vm_end > addr) { 1948615d6e87SDavidlohr Bueso vma = tmp; 1949615d6e87SDavidlohr Bueso if (tmp->vm_start <= addr) 19501da177e4SLinus Torvalds break; 19511da177e4SLinus Torvalds rb_node = rb_node->rb_left; 19521da177e4SLinus Torvalds } else 19531da177e4SLinus Torvalds rb_node = rb_node->rb_right; 19541da177e4SLinus Torvalds } 1955615d6e87SDavidlohr Bueso 19561da177e4SLinus Torvalds if (vma) 1957615d6e87SDavidlohr Bueso vmacache_update(addr, vma); 19581da177e4SLinus Torvalds return vma; 19591da177e4SLinus Torvalds } 19601da177e4SLinus Torvalds 19611da177e4SLinus Torvalds EXPORT_SYMBOL(find_vma); 19621da177e4SLinus Torvalds 19636bd4837dSKOSAKI Motohiro /* 19646bd4837dSKOSAKI Motohiro * Same as find_vma, but also return a pointer to the previous VMA in *pprev. 19656bd4837dSKOSAKI Motohiro */ 19661da177e4SLinus Torvalds struct vm_area_struct * 19671da177e4SLinus Torvalds find_vma_prev(struct mm_struct *mm, unsigned long addr, 19681da177e4SLinus Torvalds struct vm_area_struct **pprev) 19691da177e4SLinus Torvalds { 19706bd4837dSKOSAKI Motohiro struct vm_area_struct *vma; 19711da177e4SLinus Torvalds 19726bd4837dSKOSAKI Motohiro vma = find_vma(mm, addr); 197383cd904dSMikulas Patocka if (vma) { 197483cd904dSMikulas Patocka *pprev = vma->vm_prev; 197583cd904dSMikulas Patocka } else { 197683cd904dSMikulas Patocka struct rb_node *rb_node = mm->mm_rb.rb_node; 197783cd904dSMikulas Patocka *pprev = NULL; 197883cd904dSMikulas Patocka while (rb_node) { 197983cd904dSMikulas Patocka *pprev = rb_entry(rb_node, struct vm_area_struct, vm_rb); 198083cd904dSMikulas Patocka rb_node = rb_node->rb_right; 198183cd904dSMikulas Patocka } 198283cd904dSMikulas Patocka } 19836bd4837dSKOSAKI Motohiro return vma; 19841da177e4SLinus Torvalds } 19851da177e4SLinus Torvalds 19861da177e4SLinus Torvalds /* 19871da177e4SLinus Torvalds * Verify that the stack growth is acceptable and 19881da177e4SLinus Torvalds * update accounting. This is shared with both the 19891da177e4SLinus Torvalds * grow-up and grow-down cases. 19901da177e4SLinus Torvalds */ 19911da177e4SLinus Torvalds static int acct_stack_growth(struct vm_area_struct *vma, unsigned long size, unsigned long grow) 19921da177e4SLinus Torvalds { 19931da177e4SLinus Torvalds struct mm_struct *mm = vma->vm_mm; 19941da177e4SLinus Torvalds struct rlimit *rlim = current->signal->rlim; 1995690eac53SLinus Torvalds unsigned long new_start, actual_size; 19961da177e4SLinus Torvalds 19971da177e4SLinus Torvalds /* address space limit tests */ 199884638335SKonstantin Khlebnikov if (!may_expand_vm(mm, vma->vm_flags, grow)) 19991da177e4SLinus Torvalds return -ENOMEM; 20001da177e4SLinus Torvalds 20011da177e4SLinus Torvalds /* Stack limit test */ 2002690eac53SLinus Torvalds actual_size = size; 2003690eac53SLinus Torvalds if (size && (vma->vm_flags & (VM_GROWSUP | VM_GROWSDOWN))) 2004690eac53SLinus Torvalds actual_size -= PAGE_SIZE; 20054db0c3c2SJason Low if (actual_size > READ_ONCE(rlim[RLIMIT_STACK].rlim_cur)) 20061da177e4SLinus Torvalds return -ENOMEM; 20071da177e4SLinus Torvalds 20081da177e4SLinus Torvalds /* mlock limit tests */ 20091da177e4SLinus Torvalds if (vma->vm_flags & VM_LOCKED) { 20101da177e4SLinus Torvalds unsigned long locked; 20111da177e4SLinus Torvalds unsigned long limit; 20121da177e4SLinus Torvalds locked = mm->locked_vm + grow; 20134db0c3c2SJason Low limit = READ_ONCE(rlim[RLIMIT_MEMLOCK].rlim_cur); 201459e99e5bSJiri Slaby limit >>= PAGE_SHIFT; 20151da177e4SLinus Torvalds if (locked > limit && !capable(CAP_IPC_LOCK)) 20161da177e4SLinus Torvalds return -ENOMEM; 20171da177e4SLinus Torvalds } 20181da177e4SLinus Torvalds 20190d59a01bSAdam Litke /* Check to ensure the stack will not grow into a hugetlb-only region */ 20200d59a01bSAdam Litke new_start = (vma->vm_flags & VM_GROWSUP) ? vma->vm_start : 20210d59a01bSAdam Litke vma->vm_end - size; 20220d59a01bSAdam Litke if (is_hugepage_only_range(vma->vm_mm, new_start, size)) 20230d59a01bSAdam Litke return -EFAULT; 20240d59a01bSAdam Litke 20251da177e4SLinus Torvalds /* 20261da177e4SLinus Torvalds * Overcommit.. This must be the final test, as it will 20271da177e4SLinus Torvalds * update security statistics. 20281da177e4SLinus Torvalds */ 202905fa199dSHugh Dickins if (security_vm_enough_memory_mm(mm, grow)) 20301da177e4SLinus Torvalds return -ENOMEM; 20311da177e4SLinus Torvalds 20321da177e4SLinus Torvalds return 0; 20331da177e4SLinus Torvalds } 20341da177e4SLinus Torvalds 203546dea3d0SHugh Dickins #if defined(CONFIG_STACK_GROWSUP) || defined(CONFIG_IA64) 20361da177e4SLinus Torvalds /* 203746dea3d0SHugh Dickins * PA-RISC uses this for its stack; IA64 for its Register Backing Store. 203846dea3d0SHugh Dickins * vma is the last one with address > vma->vm_end. Have to extend vma. 20391da177e4SLinus Torvalds */ 204046dea3d0SHugh Dickins int expand_upwards(struct vm_area_struct *vma, unsigned long address) 20411da177e4SLinus Torvalds { 204209357814SOleg Nesterov struct mm_struct *mm = vma->vm_mm; 204312352d3cSKonstantin Khlebnikov int error = 0; 20441da177e4SLinus Torvalds 20451da177e4SLinus Torvalds if (!(vma->vm_flags & VM_GROWSUP)) 20461da177e4SLinus Torvalds return -EFAULT; 20471da177e4SLinus Torvalds 204812352d3cSKonstantin Khlebnikov /* Guard against wrapping around to address 0. */ 204912352d3cSKonstantin Khlebnikov if (address < PAGE_ALIGN(address+4)) 205012352d3cSKonstantin Khlebnikov address = PAGE_ALIGN(address+4); 205112352d3cSKonstantin Khlebnikov else 205212352d3cSKonstantin Khlebnikov return -ENOMEM; 205312352d3cSKonstantin Khlebnikov 205412352d3cSKonstantin Khlebnikov /* We must make sure the anon_vma is allocated. */ 20551da177e4SLinus Torvalds if (unlikely(anon_vma_prepare(vma))) 20561da177e4SLinus Torvalds return -ENOMEM; 20571da177e4SLinus Torvalds 20581da177e4SLinus Torvalds /* 20591da177e4SLinus Torvalds * vma->vm_start/vm_end cannot change under us because the caller 20601da177e4SLinus Torvalds * is required to hold the mmap_sem in read mode. We need the 20611da177e4SLinus Torvalds * anon_vma lock to serialize against concurrent expand_stacks. 20621da177e4SLinus Torvalds */ 206312352d3cSKonstantin Khlebnikov anon_vma_lock_write(vma->anon_vma); 20641da177e4SLinus Torvalds 20651da177e4SLinus Torvalds /* Somebody else might have raced and expanded it already */ 20661da177e4SLinus Torvalds if (address > vma->vm_end) { 20671da177e4SLinus Torvalds unsigned long size, grow; 20681da177e4SLinus Torvalds 20691da177e4SLinus Torvalds size = address - vma->vm_start; 20701da177e4SLinus Torvalds grow = (address - vma->vm_end) >> PAGE_SHIFT; 20711da177e4SLinus Torvalds 207242c36f63SHugh Dickins error = -ENOMEM; 207342c36f63SHugh Dickins if (vma->vm_pgoff + (size >> PAGE_SHIFT) >= vma->vm_pgoff) { 20741da177e4SLinus Torvalds error = acct_stack_growth(vma, size, grow); 20753af9e859SEric B Munson if (!error) { 20764128997bSMichel Lespinasse /* 20774128997bSMichel Lespinasse * vma_gap_update() doesn't support concurrent 20784128997bSMichel Lespinasse * updates, but we only hold a shared mmap_sem 20794128997bSMichel Lespinasse * lock here, so we need to protect against 20804128997bSMichel Lespinasse * concurrent vma expansions. 208112352d3cSKonstantin Khlebnikov * anon_vma_lock_write() doesn't help here, as 20824128997bSMichel Lespinasse * we don't guarantee that all growable vmas 20834128997bSMichel Lespinasse * in a mm share the same root anon vma. 20844128997bSMichel Lespinasse * So, we reuse mm->page_table_lock to guard 20854128997bSMichel Lespinasse * against concurrent vma expansions. 20864128997bSMichel Lespinasse */ 208709357814SOleg Nesterov spin_lock(&mm->page_table_lock); 208887e8827bSOleg Nesterov if (vma->vm_flags & VM_LOCKED) 208909357814SOleg Nesterov mm->locked_vm += grow; 209084638335SKonstantin Khlebnikov vm_stat_account(mm, vma->vm_flags, grow); 2091bf181b9fSMichel Lespinasse anon_vma_interval_tree_pre_update_vma(vma); 20921da177e4SLinus Torvalds vma->vm_end = address; 2093bf181b9fSMichel Lespinasse anon_vma_interval_tree_post_update_vma(vma); 2094d3737187SMichel Lespinasse if (vma->vm_next) 2095d3737187SMichel Lespinasse vma_gap_update(vma->vm_next); 2096d3737187SMichel Lespinasse else 209709357814SOleg Nesterov mm->highest_vm_end = address; 209809357814SOleg Nesterov spin_unlock(&mm->page_table_lock); 20994128997bSMichel Lespinasse 21003af9e859SEric B Munson perf_event_mmap(vma); 21013af9e859SEric B Munson } 21021da177e4SLinus Torvalds } 210342c36f63SHugh Dickins } 210412352d3cSKonstantin Khlebnikov anon_vma_unlock_write(vma->anon_vma); 21056d50e60cSDavid Rientjes khugepaged_enter_vma_merge(vma, vma->vm_flags); 210609357814SOleg Nesterov validate_mm(mm); 21071da177e4SLinus Torvalds return error; 21081da177e4SLinus Torvalds } 210946dea3d0SHugh Dickins #endif /* CONFIG_STACK_GROWSUP || CONFIG_IA64 */ 211046dea3d0SHugh Dickins 21111da177e4SLinus Torvalds /* 21121da177e4SLinus Torvalds * vma is the first one with address < vma->vm_start. Have to extend vma. 21131da177e4SLinus Torvalds */ 2114d05f3169SMichal Hocko int expand_downwards(struct vm_area_struct *vma, 2115b6a2fea3SOllie Wild unsigned long address) 21161da177e4SLinus Torvalds { 211709357814SOleg Nesterov struct mm_struct *mm = vma->vm_mm; 21181da177e4SLinus Torvalds int error; 21191da177e4SLinus Torvalds 21208869477aSEric Paris address &= PAGE_MASK; 2121e5467859SAl Viro error = security_mmap_addr(address); 21228869477aSEric Paris if (error) 21238869477aSEric Paris return error; 21248869477aSEric Paris 212512352d3cSKonstantin Khlebnikov /* We must make sure the anon_vma is allocated. */ 212612352d3cSKonstantin Khlebnikov if (unlikely(anon_vma_prepare(vma))) 212712352d3cSKonstantin Khlebnikov return -ENOMEM; 21281da177e4SLinus Torvalds 21291da177e4SLinus Torvalds /* 21301da177e4SLinus Torvalds * vma->vm_start/vm_end cannot change under us because the caller 21311da177e4SLinus Torvalds * is required to hold the mmap_sem in read mode. We need the 21321da177e4SLinus Torvalds * anon_vma lock to serialize against concurrent expand_stacks. 21331da177e4SLinus Torvalds */ 213412352d3cSKonstantin Khlebnikov anon_vma_lock_write(vma->anon_vma); 21351da177e4SLinus Torvalds 21361da177e4SLinus Torvalds /* Somebody else might have raced and expanded it already */ 21371da177e4SLinus Torvalds if (address < vma->vm_start) { 21381da177e4SLinus Torvalds unsigned long size, grow; 21391da177e4SLinus Torvalds 21401da177e4SLinus Torvalds size = vma->vm_end - address; 21411da177e4SLinus Torvalds grow = (vma->vm_start - address) >> PAGE_SHIFT; 21421da177e4SLinus Torvalds 2143a626ca6aSLinus Torvalds error = -ENOMEM; 2144a626ca6aSLinus Torvalds if (grow <= vma->vm_pgoff) { 21451da177e4SLinus Torvalds error = acct_stack_growth(vma, size, grow); 21461da177e4SLinus Torvalds if (!error) { 21474128997bSMichel Lespinasse /* 21484128997bSMichel Lespinasse * vma_gap_update() doesn't support concurrent 21494128997bSMichel Lespinasse * updates, but we only hold a shared mmap_sem 21504128997bSMichel Lespinasse * lock here, so we need to protect against 21514128997bSMichel Lespinasse * concurrent vma expansions. 215212352d3cSKonstantin Khlebnikov * anon_vma_lock_write() doesn't help here, as 21534128997bSMichel Lespinasse * we don't guarantee that all growable vmas 21544128997bSMichel Lespinasse * in a mm share the same root anon vma. 21554128997bSMichel Lespinasse * So, we reuse mm->page_table_lock to guard 21564128997bSMichel Lespinasse * against concurrent vma expansions. 21574128997bSMichel Lespinasse */ 215809357814SOleg Nesterov spin_lock(&mm->page_table_lock); 215987e8827bSOleg Nesterov if (vma->vm_flags & VM_LOCKED) 216009357814SOleg Nesterov mm->locked_vm += grow; 216184638335SKonstantin Khlebnikov vm_stat_account(mm, vma->vm_flags, grow); 2162bf181b9fSMichel Lespinasse anon_vma_interval_tree_pre_update_vma(vma); 21631da177e4SLinus Torvalds vma->vm_start = address; 21641da177e4SLinus Torvalds vma->vm_pgoff -= grow; 2165bf181b9fSMichel Lespinasse anon_vma_interval_tree_post_update_vma(vma); 2166d3737187SMichel Lespinasse vma_gap_update(vma); 216709357814SOleg Nesterov spin_unlock(&mm->page_table_lock); 21684128997bSMichel Lespinasse 21693af9e859SEric B Munson perf_event_mmap(vma); 21701da177e4SLinus Torvalds } 21711da177e4SLinus Torvalds } 2172a626ca6aSLinus Torvalds } 217312352d3cSKonstantin Khlebnikov anon_vma_unlock_write(vma->anon_vma); 21746d50e60cSDavid Rientjes khugepaged_enter_vma_merge(vma, vma->vm_flags); 217509357814SOleg Nesterov validate_mm(mm); 21761da177e4SLinus Torvalds return error; 21771da177e4SLinus Torvalds } 21781da177e4SLinus Torvalds 217909884964SLinus Torvalds /* 218009884964SLinus Torvalds * Note how expand_stack() refuses to expand the stack all the way to 218109884964SLinus Torvalds * abut the next virtual mapping, *unless* that mapping itself is also 218209884964SLinus Torvalds * a stack mapping. We want to leave room for a guard page, after all 218309884964SLinus Torvalds * (the guard page itself is not added here, that is done by the 218409884964SLinus Torvalds * actual page faulting logic) 218509884964SLinus Torvalds * 218609884964SLinus Torvalds * This matches the behavior of the guard page logic (see mm/memory.c: 218709884964SLinus Torvalds * check_stack_guard_page()), which only allows the guard page to be 218809884964SLinus Torvalds * removed under these circumstances. 218909884964SLinus Torvalds */ 2190b6a2fea3SOllie Wild #ifdef CONFIG_STACK_GROWSUP 2191b6a2fea3SOllie Wild int expand_stack(struct vm_area_struct *vma, unsigned long address) 2192b6a2fea3SOllie Wild { 219309884964SLinus Torvalds struct vm_area_struct *next; 219409884964SLinus Torvalds 219509884964SLinus Torvalds address &= PAGE_MASK; 219609884964SLinus Torvalds next = vma->vm_next; 219709884964SLinus Torvalds if (next && next->vm_start == address + PAGE_SIZE) { 219809884964SLinus Torvalds if (!(next->vm_flags & VM_GROWSUP)) 219909884964SLinus Torvalds return -ENOMEM; 220009884964SLinus Torvalds } 2201b6a2fea3SOllie Wild return expand_upwards(vma, address); 2202b6a2fea3SOllie Wild } 2203b6a2fea3SOllie Wild 2204b6a2fea3SOllie Wild struct vm_area_struct * 2205b6a2fea3SOllie Wild find_extend_vma(struct mm_struct *mm, unsigned long addr) 2206b6a2fea3SOllie Wild { 2207b6a2fea3SOllie Wild struct vm_area_struct *vma, *prev; 2208b6a2fea3SOllie Wild 2209b6a2fea3SOllie Wild addr &= PAGE_MASK; 2210b6a2fea3SOllie Wild vma = find_vma_prev(mm, addr, &prev); 2211b6a2fea3SOllie Wild if (vma && (vma->vm_start <= addr)) 2212b6a2fea3SOllie Wild return vma; 22131c127185SDenys Vlasenko if (!prev || expand_stack(prev, addr)) 2214b6a2fea3SOllie Wild return NULL; 2215cea10a19SMichel Lespinasse if (prev->vm_flags & VM_LOCKED) 2216fc05f566SKirill A. Shutemov populate_vma_page_range(prev, addr, prev->vm_end, NULL); 2217b6a2fea3SOllie Wild return prev; 2218b6a2fea3SOllie Wild } 2219b6a2fea3SOllie Wild #else 2220b6a2fea3SOllie Wild int expand_stack(struct vm_area_struct *vma, unsigned long address) 2221b6a2fea3SOllie Wild { 222209884964SLinus Torvalds struct vm_area_struct *prev; 222309884964SLinus Torvalds 222409884964SLinus Torvalds address &= PAGE_MASK; 222509884964SLinus Torvalds prev = vma->vm_prev; 222609884964SLinus Torvalds if (prev && prev->vm_end == address) { 222709884964SLinus Torvalds if (!(prev->vm_flags & VM_GROWSDOWN)) 222809884964SLinus Torvalds return -ENOMEM; 222909884964SLinus Torvalds } 2230b6a2fea3SOllie Wild return expand_downwards(vma, address); 2231b6a2fea3SOllie Wild } 2232b6a2fea3SOllie Wild 22331da177e4SLinus Torvalds struct vm_area_struct * 22341da177e4SLinus Torvalds find_extend_vma(struct mm_struct *mm, unsigned long addr) 22351da177e4SLinus Torvalds { 22361da177e4SLinus Torvalds struct vm_area_struct *vma; 22371da177e4SLinus Torvalds unsigned long start; 22381da177e4SLinus Torvalds 22391da177e4SLinus Torvalds addr &= PAGE_MASK; 22401da177e4SLinus Torvalds vma = find_vma(mm, addr); 22411da177e4SLinus Torvalds if (!vma) 22421da177e4SLinus Torvalds return NULL; 22431da177e4SLinus Torvalds if (vma->vm_start <= addr) 22441da177e4SLinus Torvalds return vma; 22451da177e4SLinus Torvalds if (!(vma->vm_flags & VM_GROWSDOWN)) 22461da177e4SLinus Torvalds return NULL; 22471da177e4SLinus Torvalds start = vma->vm_start; 22481da177e4SLinus Torvalds if (expand_stack(vma, addr)) 22491da177e4SLinus Torvalds return NULL; 2250cea10a19SMichel Lespinasse if (vma->vm_flags & VM_LOCKED) 2251fc05f566SKirill A. Shutemov populate_vma_page_range(vma, addr, start, NULL); 22521da177e4SLinus Torvalds return vma; 22531da177e4SLinus Torvalds } 22541da177e4SLinus Torvalds #endif 22551da177e4SLinus Torvalds 2256e1d6d01aSJesse Barnes EXPORT_SYMBOL_GPL(find_extend_vma); 2257e1d6d01aSJesse Barnes 22582c0b3814SHugh Dickins /* 22592c0b3814SHugh Dickins * Ok - we have the memory areas we should free on the vma list, 22602c0b3814SHugh Dickins * so release them, and do the vma updates. 22611da177e4SLinus Torvalds * 22622c0b3814SHugh Dickins * Called with the mm semaphore held. 22631da177e4SLinus Torvalds */ 22642c0b3814SHugh Dickins static void remove_vma_list(struct mm_struct *mm, struct vm_area_struct *vma) 22651da177e4SLinus Torvalds { 22664f74d2c8SLinus Torvalds unsigned long nr_accounted = 0; 22674f74d2c8SLinus Torvalds 2268365e9c87SHugh Dickins /* Update high watermark before we lower total_vm */ 2269365e9c87SHugh Dickins update_hiwater_vm(mm); 22702c0b3814SHugh Dickins do { 2271ab50b8edSHugh Dickins long nrpages = vma_pages(vma); 22721da177e4SLinus Torvalds 22734f74d2c8SLinus Torvalds if (vma->vm_flags & VM_ACCOUNT) 22744f74d2c8SLinus Torvalds nr_accounted += nrpages; 227584638335SKonstantin Khlebnikov vm_stat_account(mm, vma->vm_flags, -nrpages); 2276a8fb5618SHugh Dickins vma = remove_vma(vma); 2277146425a3SHugh Dickins } while (vma); 22784f74d2c8SLinus Torvalds vm_unacct_memory(nr_accounted); 22791da177e4SLinus Torvalds validate_mm(mm); 22801da177e4SLinus Torvalds } 22811da177e4SLinus Torvalds 22821da177e4SLinus Torvalds /* 22831da177e4SLinus Torvalds * Get rid of page table information in the indicated region. 22841da177e4SLinus Torvalds * 2285f10df686SPaolo 'Blaisorblade' Giarrusso * Called with the mm semaphore held. 22861da177e4SLinus Torvalds */ 22871da177e4SLinus Torvalds static void unmap_region(struct mm_struct *mm, 2288e0da382cSHugh Dickins struct vm_area_struct *vma, struct vm_area_struct *prev, 2289e0da382cSHugh Dickins unsigned long start, unsigned long end) 22901da177e4SLinus Torvalds { 2291e0da382cSHugh Dickins struct vm_area_struct *next = prev ? prev->vm_next : mm->mmap; 2292d16dfc55SPeter Zijlstra struct mmu_gather tlb; 22931da177e4SLinus Torvalds 22941da177e4SLinus Torvalds lru_add_drain(); 22952b047252SLinus Torvalds tlb_gather_mmu(&tlb, mm, start, end); 2296365e9c87SHugh Dickins update_hiwater_rss(mm); 22974f74d2c8SLinus Torvalds unmap_vmas(&tlb, vma, start, end); 2298d16dfc55SPeter Zijlstra free_pgtables(&tlb, vma, prev ? prev->vm_end : FIRST_USER_ADDRESS, 22996ee8630eSHugh Dickins next ? next->vm_start : USER_PGTABLES_CEILING); 2300d16dfc55SPeter Zijlstra tlb_finish_mmu(&tlb, start, end); 23011da177e4SLinus Torvalds } 23021da177e4SLinus Torvalds 23031da177e4SLinus Torvalds /* 23041da177e4SLinus Torvalds * Create a list of vma's touched by the unmap, removing them from the mm's 23051da177e4SLinus Torvalds * vma list as we go.. 23061da177e4SLinus Torvalds */ 23071da177e4SLinus Torvalds static void 23081da177e4SLinus Torvalds detach_vmas_to_be_unmapped(struct mm_struct *mm, struct vm_area_struct *vma, 23091da177e4SLinus Torvalds struct vm_area_struct *prev, unsigned long end) 23101da177e4SLinus Torvalds { 23111da177e4SLinus Torvalds struct vm_area_struct **insertion_point; 23121da177e4SLinus Torvalds struct vm_area_struct *tail_vma = NULL; 23131da177e4SLinus Torvalds 23141da177e4SLinus Torvalds insertion_point = (prev ? &prev->vm_next : &mm->mmap); 2315297c5eeeSLinus Torvalds vma->vm_prev = NULL; 23161da177e4SLinus Torvalds do { 2317d3737187SMichel Lespinasse vma_rb_erase(vma, &mm->mm_rb); 23181da177e4SLinus Torvalds mm->map_count--; 23191da177e4SLinus Torvalds tail_vma = vma; 23201da177e4SLinus Torvalds vma = vma->vm_next; 23211da177e4SLinus Torvalds } while (vma && vma->vm_start < end); 23221da177e4SLinus Torvalds *insertion_point = vma; 2323d3737187SMichel Lespinasse if (vma) { 2324297c5eeeSLinus Torvalds vma->vm_prev = prev; 2325d3737187SMichel Lespinasse vma_gap_update(vma); 2326d3737187SMichel Lespinasse } else 2327d3737187SMichel Lespinasse mm->highest_vm_end = prev ? prev->vm_end : 0; 23281da177e4SLinus Torvalds tail_vma->vm_next = NULL; 2329615d6e87SDavidlohr Bueso 2330615d6e87SDavidlohr Bueso /* Kill the cache */ 2331615d6e87SDavidlohr Bueso vmacache_invalidate(mm); 23321da177e4SLinus Torvalds } 23331da177e4SLinus Torvalds 23341da177e4SLinus Torvalds /* 2335659ace58SKOSAKI Motohiro * __split_vma() bypasses sysctl_max_map_count checking. We use this on the 2336659ace58SKOSAKI Motohiro * munmap path where it doesn't make sense to fail. 23371da177e4SLinus Torvalds */ 2338659ace58SKOSAKI Motohiro static int __split_vma(struct mm_struct *mm, struct vm_area_struct *vma, 23391da177e4SLinus Torvalds unsigned long addr, int new_below) 23401da177e4SLinus Torvalds { 23411da177e4SLinus Torvalds struct vm_area_struct *new; 2342e3975891SChen Gang int err; 23431da177e4SLinus Torvalds 2344a5516438SAndi Kleen if (is_vm_hugetlb_page(vma) && (addr & 2345a5516438SAndi Kleen ~(huge_page_mask(hstate_vma(vma))))) 23461da177e4SLinus Torvalds return -EINVAL; 23471da177e4SLinus Torvalds 2348e94b1766SChristoph Lameter new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL); 23491da177e4SLinus Torvalds if (!new) 2350e3975891SChen Gang return -ENOMEM; 23511da177e4SLinus Torvalds 23521da177e4SLinus Torvalds /* most fields are the same, copy all, and then fixup */ 23531da177e4SLinus Torvalds *new = *vma; 23541da177e4SLinus Torvalds 23555beb4930SRik van Riel INIT_LIST_HEAD(&new->anon_vma_chain); 23565beb4930SRik van Riel 23571da177e4SLinus Torvalds if (new_below) 23581da177e4SLinus Torvalds new->vm_end = addr; 23591da177e4SLinus Torvalds else { 23601da177e4SLinus Torvalds new->vm_start = addr; 23611da177e4SLinus Torvalds new->vm_pgoff += ((addr - vma->vm_start) >> PAGE_SHIFT); 23621da177e4SLinus Torvalds } 23631da177e4SLinus Torvalds 2364ef0855d3SOleg Nesterov err = vma_dup_policy(vma, new); 2365ef0855d3SOleg Nesterov if (err) 23665beb4930SRik van Riel goto out_free_vma; 23671da177e4SLinus Torvalds 2368c4ea95d7SDaniel Forrest err = anon_vma_clone(new, vma); 2369c4ea95d7SDaniel Forrest if (err) 23705beb4930SRik van Riel goto out_free_mpol; 23715beb4930SRik van Riel 2372e9714acfSKonstantin Khlebnikov if (new->vm_file) 23731da177e4SLinus Torvalds get_file(new->vm_file); 23741da177e4SLinus Torvalds 23751da177e4SLinus Torvalds if (new->vm_ops && new->vm_ops->open) 23761da177e4SLinus Torvalds new->vm_ops->open(new); 23771da177e4SLinus Torvalds 23781da177e4SLinus Torvalds if (new_below) 23795beb4930SRik van Riel err = vma_adjust(vma, addr, vma->vm_end, vma->vm_pgoff + 23801da177e4SLinus Torvalds ((addr - new->vm_start) >> PAGE_SHIFT), new); 23811da177e4SLinus Torvalds else 23825beb4930SRik van Riel err = vma_adjust(vma, vma->vm_start, addr, vma->vm_pgoff, new); 23831da177e4SLinus Torvalds 23845beb4930SRik van Riel /* Success. */ 23855beb4930SRik van Riel if (!err) 23861da177e4SLinus Torvalds return 0; 23875beb4930SRik van Riel 23885beb4930SRik van Riel /* Clean everything up if vma_adjust failed. */ 238958927533SRik van Riel if (new->vm_ops && new->vm_ops->close) 23905beb4930SRik van Riel new->vm_ops->close(new); 2391e9714acfSKonstantin Khlebnikov if (new->vm_file) 23925beb4930SRik van Riel fput(new->vm_file); 23932aeadc30SAndrea Arcangeli unlink_anon_vmas(new); 23945beb4930SRik van Riel out_free_mpol: 2395ef0855d3SOleg Nesterov mpol_put(vma_policy(new)); 23965beb4930SRik van Riel out_free_vma: 23975beb4930SRik van Riel kmem_cache_free(vm_area_cachep, 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 */ 2405659ace58SKOSAKI Motohiro int split_vma(struct mm_struct *mm, struct vm_area_struct *vma, 2406659ace58SKOSAKI Motohiro unsigned long addr, int new_below) 2407659ace58SKOSAKI Motohiro { 2408659ace58SKOSAKI Motohiro if (mm->map_count >= sysctl_max_map_count) 2409659ace58SKOSAKI Motohiro return -ENOMEM; 2410659ace58SKOSAKI Motohiro 2411659ace58SKOSAKI Motohiro return __split_vma(mm, vma, addr, new_below); 2412659ace58SKOSAKI Motohiro } 2413659ace58SKOSAKI Motohiro 24141da177e4SLinus Torvalds /* Munmap is split into 2 main parts -- this part which finds 24151da177e4SLinus Torvalds * what needs doing, and the areas themselves, which do the 24161da177e4SLinus Torvalds * work. This now handles partial unmappings. 24171da177e4SLinus Torvalds * Jeremy Fitzhardinge <jeremy@goop.org> 24181da177e4SLinus Torvalds */ 24191da177e4SLinus Torvalds int do_munmap(struct mm_struct *mm, unsigned long start, size_t len) 24201da177e4SLinus Torvalds { 24211da177e4SLinus Torvalds unsigned long end; 2422146425a3SHugh Dickins struct vm_area_struct *vma, *prev, *last; 24231da177e4SLinus Torvalds 2424de1741a1SAlexander Kuleshov if ((offset_in_page(start)) || start > TASK_SIZE || len > TASK_SIZE-start) 24251da177e4SLinus Torvalds return -EINVAL; 24261da177e4SLinus Torvalds 2427cc71aba3Svishnu.ps len = PAGE_ALIGN(len); 2428cc71aba3Svishnu.ps if (len == 0) 24291da177e4SLinus Torvalds return -EINVAL; 24301da177e4SLinus Torvalds 24311da177e4SLinus Torvalds /* Find the first overlapping VMA */ 24329be34c9dSLinus Torvalds vma = find_vma(mm, start); 2433146425a3SHugh Dickins if (!vma) 24341da177e4SLinus Torvalds return 0; 24359be34c9dSLinus Torvalds prev = vma->vm_prev; 2436146425a3SHugh Dickins /* we have start < vma->vm_end */ 24371da177e4SLinus Torvalds 24381da177e4SLinus Torvalds /* if it doesn't overlap, we have nothing.. */ 24391da177e4SLinus Torvalds end = start + len; 2440146425a3SHugh Dickins if (vma->vm_start >= end) 24411da177e4SLinus Torvalds return 0; 24421da177e4SLinus Torvalds 24431da177e4SLinus Torvalds /* 24441da177e4SLinus Torvalds * If we need to split any vma, do it now to save pain later. 24451da177e4SLinus Torvalds * 24461da177e4SLinus Torvalds * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially 24471da177e4SLinus Torvalds * unmapped vm_area_struct will remain in use: so lower split_vma 24481da177e4SLinus Torvalds * places tmp vma above, and higher split_vma places tmp vma below. 24491da177e4SLinus Torvalds */ 2450146425a3SHugh Dickins if (start > vma->vm_start) { 2451659ace58SKOSAKI Motohiro int error; 2452659ace58SKOSAKI Motohiro 2453659ace58SKOSAKI Motohiro /* 2454659ace58SKOSAKI Motohiro * Make sure that map_count on return from munmap() will 2455659ace58SKOSAKI Motohiro * not exceed its limit; but let map_count go just above 2456659ace58SKOSAKI Motohiro * its limit temporarily, to help free resources as expected. 2457659ace58SKOSAKI Motohiro */ 2458659ace58SKOSAKI Motohiro if (end < vma->vm_end && mm->map_count >= sysctl_max_map_count) 2459659ace58SKOSAKI Motohiro return -ENOMEM; 2460659ace58SKOSAKI Motohiro 2461659ace58SKOSAKI Motohiro error = __split_vma(mm, vma, start, 0); 24621da177e4SLinus Torvalds if (error) 24631da177e4SLinus Torvalds return error; 2464146425a3SHugh Dickins prev = vma; 24651da177e4SLinus Torvalds } 24661da177e4SLinus Torvalds 24671da177e4SLinus Torvalds /* Does it split the last one? */ 24681da177e4SLinus Torvalds last = find_vma(mm, end); 24691da177e4SLinus Torvalds if (last && end > last->vm_start) { 2470659ace58SKOSAKI Motohiro int error = __split_vma(mm, last, end, 1); 24711da177e4SLinus Torvalds if (error) 24721da177e4SLinus Torvalds return error; 24731da177e4SLinus Torvalds } 2474146425a3SHugh Dickins vma = prev ? prev->vm_next : mm->mmap; 24751da177e4SLinus Torvalds 24761da177e4SLinus Torvalds /* 2477ba470de4SRik van Riel * unlock any mlock()ed ranges before detaching vmas 2478ba470de4SRik van Riel */ 2479ba470de4SRik van Riel if (mm->locked_vm) { 2480ba470de4SRik van Riel struct vm_area_struct *tmp = vma; 2481ba470de4SRik van Riel while (tmp && tmp->vm_start < end) { 2482ba470de4SRik van Riel if (tmp->vm_flags & VM_LOCKED) { 2483ba470de4SRik van Riel mm->locked_vm -= vma_pages(tmp); 2484ba470de4SRik van Riel munlock_vma_pages_all(tmp); 2485ba470de4SRik van Riel } 2486ba470de4SRik van Riel tmp = tmp->vm_next; 2487ba470de4SRik van Riel } 2488ba470de4SRik van Riel } 2489ba470de4SRik van Riel 2490ba470de4SRik van Riel /* 24911da177e4SLinus Torvalds * Remove the vma's, and unmap the actual pages 24921da177e4SLinus Torvalds */ 2493146425a3SHugh Dickins detach_vmas_to_be_unmapped(mm, vma, prev, end); 2494146425a3SHugh Dickins unmap_region(mm, vma, prev, start, end); 24951da177e4SLinus Torvalds 24961de4fa14SDave Hansen arch_unmap(mm, vma, start, end); 24971de4fa14SDave Hansen 24981da177e4SLinus Torvalds /* Fix up all other VM information */ 24992c0b3814SHugh Dickins remove_vma_list(mm, vma); 25001da177e4SLinus Torvalds 25011da177e4SLinus Torvalds return 0; 25021da177e4SLinus Torvalds } 25031da177e4SLinus Torvalds 2504bfce281cSAl Viro int vm_munmap(unsigned long start, size_t len) 2505a46ef99dSLinus Torvalds { 2506a46ef99dSLinus Torvalds int ret; 2507bfce281cSAl Viro struct mm_struct *mm = current->mm; 2508a46ef99dSLinus Torvalds 2509ae798783SMichal Hocko if (down_write_killable(&mm->mmap_sem)) 2510ae798783SMichal Hocko return -EINTR; 2511ae798783SMichal Hocko 2512a46ef99dSLinus Torvalds ret = do_munmap(mm, start, len); 2513a46ef99dSLinus Torvalds up_write(&mm->mmap_sem); 2514a46ef99dSLinus Torvalds return ret; 2515a46ef99dSLinus Torvalds } 2516a46ef99dSLinus Torvalds EXPORT_SYMBOL(vm_munmap); 2517a46ef99dSLinus Torvalds 25186a6160a7SHeiko Carstens SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len) 25191da177e4SLinus Torvalds { 2520dc0ef0dfSMichal Hocko int ret; 2521dc0ef0dfSMichal Hocko struct mm_struct *mm = current->mm; 2522dc0ef0dfSMichal Hocko 25231da177e4SLinus Torvalds profile_munmap(addr); 2524dc0ef0dfSMichal Hocko if (down_write_killable(&mm->mmap_sem)) 2525dc0ef0dfSMichal Hocko return -EINTR; 2526dc0ef0dfSMichal Hocko ret = do_munmap(mm, addr, len); 2527dc0ef0dfSMichal Hocko up_write(&mm->mmap_sem); 2528dc0ef0dfSMichal Hocko return ret; 25291da177e4SLinus Torvalds } 25301da177e4SLinus Torvalds 2531c8d78c18SKirill A. Shutemov 2532c8d78c18SKirill A. Shutemov /* 2533c8d78c18SKirill A. Shutemov * Emulation of deprecated remap_file_pages() syscall. 2534c8d78c18SKirill A. Shutemov */ 2535c8d78c18SKirill A. Shutemov SYSCALL_DEFINE5(remap_file_pages, unsigned long, start, unsigned long, size, 2536c8d78c18SKirill A. Shutemov unsigned long, prot, unsigned long, pgoff, unsigned long, flags) 2537c8d78c18SKirill A. Shutemov { 2538c8d78c18SKirill A. Shutemov 2539c8d78c18SKirill A. Shutemov struct mm_struct *mm = current->mm; 2540c8d78c18SKirill A. Shutemov struct vm_area_struct *vma; 2541c8d78c18SKirill A. Shutemov unsigned long populate = 0; 2542c8d78c18SKirill A. Shutemov unsigned long ret = -EINVAL; 2543c8d78c18SKirill A. Shutemov struct file *file; 2544c8d78c18SKirill A. Shutemov 2545756a025fSJoe Perches pr_warn_once("%s (%d) uses deprecated remap_file_pages() syscall. See Documentation/vm/remap_file_pages.txt.\n", 2546c8d78c18SKirill A. Shutemov current->comm, current->pid); 2547c8d78c18SKirill A. Shutemov 2548c8d78c18SKirill A. Shutemov if (prot) 2549c8d78c18SKirill A. Shutemov return ret; 2550c8d78c18SKirill A. Shutemov start = start & PAGE_MASK; 2551c8d78c18SKirill A. Shutemov size = size & PAGE_MASK; 2552c8d78c18SKirill A. Shutemov 2553c8d78c18SKirill A. Shutemov if (start + size <= start) 2554c8d78c18SKirill A. Shutemov return ret; 2555c8d78c18SKirill A. Shutemov 2556c8d78c18SKirill A. Shutemov /* Does pgoff wrap? */ 2557c8d78c18SKirill A. Shutemov if (pgoff + (size >> PAGE_SHIFT) < pgoff) 2558c8d78c18SKirill A. Shutemov return ret; 2559c8d78c18SKirill A. Shutemov 2560dc0ef0dfSMichal Hocko if (down_write_killable(&mm->mmap_sem)) 2561dc0ef0dfSMichal Hocko return -EINTR; 2562dc0ef0dfSMichal Hocko 2563c8d78c18SKirill A. Shutemov vma = find_vma(mm, start); 2564c8d78c18SKirill A. Shutemov 2565c8d78c18SKirill A. Shutemov if (!vma || !(vma->vm_flags & VM_SHARED)) 2566c8d78c18SKirill A. Shutemov goto out; 2567c8d78c18SKirill A. Shutemov 256848f7df32SKirill A. Shutemov if (start < vma->vm_start) 2569c8d78c18SKirill A. Shutemov goto out; 2570c8d78c18SKirill A. Shutemov 257148f7df32SKirill A. Shutemov if (start + size > vma->vm_end) { 257248f7df32SKirill A. Shutemov struct vm_area_struct *next; 257348f7df32SKirill A. Shutemov 257448f7df32SKirill A. Shutemov for (next = vma->vm_next; next; next = next->vm_next) { 257548f7df32SKirill A. Shutemov /* hole between vmas ? */ 257648f7df32SKirill A. Shutemov if (next->vm_start != next->vm_prev->vm_end) 257748f7df32SKirill A. Shutemov goto out; 257848f7df32SKirill A. Shutemov 257948f7df32SKirill A. Shutemov if (next->vm_file != vma->vm_file) 258048f7df32SKirill A. Shutemov goto out; 258148f7df32SKirill A. Shutemov 258248f7df32SKirill A. Shutemov if (next->vm_flags != vma->vm_flags) 258348f7df32SKirill A. Shutemov goto out; 258448f7df32SKirill A. Shutemov 258548f7df32SKirill A. Shutemov if (start + size <= next->vm_end) 258648f7df32SKirill A. Shutemov break; 258748f7df32SKirill A. Shutemov } 258848f7df32SKirill A. Shutemov 258948f7df32SKirill A. Shutemov if (!next) 2590c8d78c18SKirill A. Shutemov goto out; 2591c8d78c18SKirill A. Shutemov } 2592c8d78c18SKirill A. Shutemov 2593c8d78c18SKirill A. Shutemov prot |= vma->vm_flags & VM_READ ? PROT_READ : 0; 2594c8d78c18SKirill A. Shutemov prot |= vma->vm_flags & VM_WRITE ? PROT_WRITE : 0; 2595c8d78c18SKirill A. Shutemov prot |= vma->vm_flags & VM_EXEC ? PROT_EXEC : 0; 2596c8d78c18SKirill A. Shutemov 2597c8d78c18SKirill A. Shutemov flags &= MAP_NONBLOCK; 2598c8d78c18SKirill A. Shutemov flags |= MAP_SHARED | MAP_FIXED | MAP_POPULATE; 2599c8d78c18SKirill A. Shutemov if (vma->vm_flags & VM_LOCKED) { 260048f7df32SKirill A. Shutemov struct vm_area_struct *tmp; 2601c8d78c18SKirill A. Shutemov flags |= MAP_LOCKED; 260248f7df32SKirill A. Shutemov 2603c8d78c18SKirill A. Shutemov /* drop PG_Mlocked flag for over-mapped range */ 260448f7df32SKirill A. Shutemov for (tmp = vma; tmp->vm_start >= start + size; 260548f7df32SKirill A. Shutemov tmp = tmp->vm_next) { 26069a73f61bSKirill A. Shutemov /* 26079a73f61bSKirill A. Shutemov * Split pmd and munlock page on the border 26089a73f61bSKirill A. Shutemov * of the range. 26099a73f61bSKirill A. Shutemov */ 26109a73f61bSKirill A. Shutemov vma_adjust_trans_huge(tmp, start, start + size, 0); 26119a73f61bSKirill A. Shutemov 261248f7df32SKirill A. Shutemov munlock_vma_pages_range(tmp, 261348f7df32SKirill A. Shutemov max(tmp->vm_start, start), 261448f7df32SKirill A. Shutemov min(tmp->vm_end, start + size)); 261548f7df32SKirill A. Shutemov } 2616c8d78c18SKirill A. Shutemov } 2617c8d78c18SKirill A. Shutemov 2618c8d78c18SKirill A. Shutemov file = get_file(vma->vm_file); 2619c8d78c18SKirill A. Shutemov ret = do_mmap_pgoff(vma->vm_file, start, size, 2620c8d78c18SKirill A. Shutemov prot, flags, pgoff, &populate); 2621c8d78c18SKirill A. Shutemov fput(file); 2622c8d78c18SKirill A. Shutemov out: 2623c8d78c18SKirill A. Shutemov up_write(&mm->mmap_sem); 2624c8d78c18SKirill A. Shutemov if (populate) 2625c8d78c18SKirill A. Shutemov mm_populate(ret, populate); 2626c8d78c18SKirill A. Shutemov if (!IS_ERR_VALUE(ret)) 2627c8d78c18SKirill A. Shutemov ret = 0; 2628c8d78c18SKirill A. Shutemov return ret; 2629c8d78c18SKirill A. Shutemov } 2630c8d78c18SKirill A. Shutemov 26311da177e4SLinus Torvalds static inline void verify_mm_writelocked(struct mm_struct *mm) 26321da177e4SLinus Torvalds { 2633a241ec65SPaul E. McKenney #ifdef CONFIG_DEBUG_VM 26341da177e4SLinus Torvalds if (unlikely(down_read_trylock(&mm->mmap_sem))) { 26351da177e4SLinus Torvalds WARN_ON(1); 26361da177e4SLinus Torvalds up_read(&mm->mmap_sem); 26371da177e4SLinus Torvalds } 26381da177e4SLinus Torvalds #endif 26391da177e4SLinus Torvalds } 26401da177e4SLinus Torvalds 26411da177e4SLinus Torvalds /* 26421da177e4SLinus Torvalds * this is really a simplified "do_mmap". it only handles 26431da177e4SLinus Torvalds * anonymous maps. eventually we may be able to do some 26441da177e4SLinus Torvalds * brk-specific accounting here. 26451da177e4SLinus Torvalds */ 26465d22fc25SLinus Torvalds static int do_brk(unsigned long addr, unsigned long len) 26471da177e4SLinus Torvalds { 26481da177e4SLinus Torvalds struct mm_struct *mm = current->mm; 26491da177e4SLinus Torvalds struct vm_area_struct *vma, *prev; 26501da177e4SLinus Torvalds unsigned long flags; 26511da177e4SLinus Torvalds struct rb_node **rb_link, *rb_parent; 26521da177e4SLinus Torvalds pgoff_t pgoff = addr >> PAGE_SHIFT; 26533a459756SKirill Korotaev int error; 26541da177e4SLinus Torvalds 26551da177e4SLinus Torvalds len = PAGE_ALIGN(len); 26561da177e4SLinus Torvalds if (!len) 26575d22fc25SLinus Torvalds return 0; 26581da177e4SLinus Torvalds 26593a459756SKirill Korotaev flags = VM_DATA_DEFAULT_FLAGS | VM_ACCOUNT | mm->def_flags; 26603a459756SKirill Korotaev 26612c6a1016SAl Viro error = get_unmapped_area(NULL, addr, len, 0, MAP_FIXED); 2662de1741a1SAlexander Kuleshov if (offset_in_page(error)) 26633a459756SKirill Korotaev return error; 26643a459756SKirill Korotaev 2665363ee17fSDavidlohr Bueso error = mlock_future_check(mm, mm->def_flags, len); 2666363ee17fSDavidlohr Bueso if (error) 2667363ee17fSDavidlohr Bueso return error; 26681da177e4SLinus Torvalds 26691da177e4SLinus Torvalds /* 26701da177e4SLinus Torvalds * mm->mmap_sem is required to protect against another thread 26711da177e4SLinus Torvalds * changing the mappings in case we sleep. 26721da177e4SLinus Torvalds */ 26731da177e4SLinus Torvalds verify_mm_writelocked(mm); 26741da177e4SLinus Torvalds 26751da177e4SLinus Torvalds /* 26761da177e4SLinus Torvalds * Clear old maps. this also does some error checking for us 26771da177e4SLinus Torvalds */ 26789fcd1457SRasmus Villemoes while (find_vma_links(mm, addr, addr + len, &prev, &rb_link, 26799fcd1457SRasmus Villemoes &rb_parent)) { 26801da177e4SLinus Torvalds if (do_munmap(mm, addr, len)) 26811da177e4SLinus Torvalds return -ENOMEM; 26821da177e4SLinus Torvalds } 26831da177e4SLinus Torvalds 26841da177e4SLinus Torvalds /* Check against address space limits *after* clearing old maps... */ 268584638335SKonstantin Khlebnikov if (!may_expand_vm(mm, flags, len >> PAGE_SHIFT)) 26861da177e4SLinus Torvalds return -ENOMEM; 26871da177e4SLinus Torvalds 26881da177e4SLinus Torvalds if (mm->map_count > sysctl_max_map_count) 26891da177e4SLinus Torvalds return -ENOMEM; 26901da177e4SLinus Torvalds 2691191c5424SAl Viro if (security_vm_enough_memory_mm(mm, len >> PAGE_SHIFT)) 26921da177e4SLinus Torvalds return -ENOMEM; 26931da177e4SLinus Torvalds 26941da177e4SLinus Torvalds /* Can we just expand an old private anonymous mapping? */ 2695ba470de4SRik van Riel vma = vma_merge(mm, prev, addr, addr + len, flags, 269619a809afSAndrea Arcangeli NULL, NULL, pgoff, NULL, NULL_VM_UFFD_CTX); 2697ba470de4SRik van Riel if (vma) 26981da177e4SLinus Torvalds goto out; 26991da177e4SLinus Torvalds 27001da177e4SLinus Torvalds /* 27011da177e4SLinus Torvalds * create a vma struct for an anonymous mapping 27021da177e4SLinus Torvalds */ 2703c5e3b83eSPekka Enberg vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL); 27041da177e4SLinus Torvalds if (!vma) { 27051da177e4SLinus Torvalds vm_unacct_memory(len >> PAGE_SHIFT); 27061da177e4SLinus Torvalds return -ENOMEM; 27071da177e4SLinus Torvalds } 27081da177e4SLinus Torvalds 27095beb4930SRik van Riel INIT_LIST_HEAD(&vma->anon_vma_chain); 27101da177e4SLinus Torvalds vma->vm_mm = mm; 27111da177e4SLinus Torvalds vma->vm_start = addr; 27121da177e4SLinus Torvalds vma->vm_end = addr + len; 27131da177e4SLinus Torvalds vma->vm_pgoff = pgoff; 27141da177e4SLinus Torvalds vma->vm_flags = flags; 27153ed75eb8SColy Li vma->vm_page_prot = vm_get_page_prot(flags); 27161da177e4SLinus Torvalds vma_link(mm, vma, prev, rb_link, rb_parent); 27171da177e4SLinus Torvalds out: 27183af9e859SEric B Munson perf_event_mmap(vma); 27191da177e4SLinus Torvalds mm->total_vm += len >> PAGE_SHIFT; 272084638335SKonstantin Khlebnikov mm->data_vm += len >> PAGE_SHIFT; 2721128557ffSMichel Lespinasse if (flags & VM_LOCKED) 2722ba470de4SRik van Riel mm->locked_vm += (len >> PAGE_SHIFT); 2723d9104d1cSCyrill Gorcunov vma->vm_flags |= VM_SOFTDIRTY; 27245d22fc25SLinus Torvalds return 0; 27251da177e4SLinus Torvalds } 27261da177e4SLinus Torvalds 27275d22fc25SLinus Torvalds int vm_brk(unsigned long addr, unsigned long len) 2728e4eb1ff6SLinus Torvalds { 2729e4eb1ff6SLinus Torvalds struct mm_struct *mm = current->mm; 27305d22fc25SLinus Torvalds int ret; 2731128557ffSMichel Lespinasse bool populate; 2732e4eb1ff6SLinus Torvalds 27332d6c9282SMichal Hocko if (down_write_killable(&mm->mmap_sem)) 27342d6c9282SMichal Hocko return -EINTR; 27352d6c9282SMichal Hocko 2736e4eb1ff6SLinus Torvalds ret = do_brk(addr, len); 2737128557ffSMichel Lespinasse populate = ((mm->def_flags & VM_LOCKED) != 0); 2738e4eb1ff6SLinus Torvalds up_write(&mm->mmap_sem); 27395d22fc25SLinus Torvalds if (populate && !ret) 2740128557ffSMichel Lespinasse mm_populate(addr, len); 2741e4eb1ff6SLinus Torvalds return ret; 2742e4eb1ff6SLinus Torvalds } 2743e4eb1ff6SLinus Torvalds EXPORT_SYMBOL(vm_brk); 27441da177e4SLinus Torvalds 27451da177e4SLinus Torvalds /* Release all mmaps. */ 27461da177e4SLinus Torvalds void exit_mmap(struct mm_struct *mm) 27471da177e4SLinus Torvalds { 2748d16dfc55SPeter Zijlstra struct mmu_gather tlb; 2749ba470de4SRik van Riel struct vm_area_struct *vma; 27501da177e4SLinus Torvalds unsigned long nr_accounted = 0; 27511da177e4SLinus Torvalds 2752d6dd61c8SJeremy Fitzhardinge /* mm's last user has gone, and its about to be pulled down */ 2753cddb8a5cSAndrea Arcangeli mmu_notifier_release(mm); 2754d6dd61c8SJeremy Fitzhardinge 2755ba470de4SRik van Riel if (mm->locked_vm) { 2756ba470de4SRik van Riel vma = mm->mmap; 2757ba470de4SRik van Riel while (vma) { 2758ba470de4SRik van Riel if (vma->vm_flags & VM_LOCKED) 2759ba470de4SRik van Riel munlock_vma_pages_all(vma); 2760ba470de4SRik van Riel vma = vma->vm_next; 2761ba470de4SRik van Riel } 2762ba470de4SRik van Riel } 27639480c53eSJeremy Fitzhardinge 27649480c53eSJeremy Fitzhardinge arch_exit_mmap(mm); 27659480c53eSJeremy Fitzhardinge 2766ba470de4SRik van Riel vma = mm->mmap; 27679480c53eSJeremy Fitzhardinge if (!vma) /* Can happen if dup_mmap() received an OOM */ 27689480c53eSJeremy Fitzhardinge return; 27699480c53eSJeremy Fitzhardinge 27701da177e4SLinus Torvalds lru_add_drain(); 27711da177e4SLinus Torvalds flush_cache_mm(mm); 27722b047252SLinus Torvalds tlb_gather_mmu(&tlb, mm, 0, -1); 2773901608d9SOleg Nesterov /* update_hiwater_rss(mm) here? but nobody should be looking */ 2774e0da382cSHugh Dickins /* Use -1 here to ensure all VMAs in the mm are unmapped */ 27754f74d2c8SLinus Torvalds unmap_vmas(&tlb, vma, 0, -1); 27769ba69294SHugh Dickins 27776ee8630eSHugh Dickins free_pgtables(&tlb, vma, FIRST_USER_ADDRESS, USER_PGTABLES_CEILING); 2778853f5e26SAl Viro tlb_finish_mmu(&tlb, 0, -1); 27791da177e4SLinus Torvalds 27801da177e4SLinus Torvalds /* 27818f4f8c16SHugh Dickins * Walk the list again, actually closing and freeing it, 27828f4f8c16SHugh Dickins * with preemption enabled, without holding any MM locks. 27831da177e4SLinus Torvalds */ 27844f74d2c8SLinus Torvalds while (vma) { 27854f74d2c8SLinus Torvalds if (vma->vm_flags & VM_ACCOUNT) 27864f74d2c8SLinus Torvalds nr_accounted += vma_pages(vma); 2787a8fb5618SHugh Dickins vma = remove_vma(vma); 27884f74d2c8SLinus Torvalds } 27894f74d2c8SLinus Torvalds vm_unacct_memory(nr_accounted); 27901da177e4SLinus Torvalds } 27911da177e4SLinus Torvalds 27921da177e4SLinus Torvalds /* Insert vm structure into process list sorted by address 27931da177e4SLinus Torvalds * and into the inode's i_mmap tree. If vm_file is non-NULL 2794c8c06efaSDavidlohr Bueso * then i_mmap_rwsem is taken here. 27951da177e4SLinus Torvalds */ 27961da177e4SLinus Torvalds int insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma) 27971da177e4SLinus Torvalds { 27986597d783SHugh Dickins struct vm_area_struct *prev; 27991da177e4SLinus Torvalds struct rb_node **rb_link, *rb_parent; 28001da177e4SLinus Torvalds 2801c9d13f5fSChen Gang if (find_vma_links(mm, vma->vm_start, vma->vm_end, 2802c9d13f5fSChen Gang &prev, &rb_link, &rb_parent)) 2803c9d13f5fSChen Gang return -ENOMEM; 2804c9d13f5fSChen Gang if ((vma->vm_flags & VM_ACCOUNT) && 2805c9d13f5fSChen Gang security_vm_enough_memory_mm(mm, vma_pages(vma))) 2806c9d13f5fSChen Gang return -ENOMEM; 2807c9d13f5fSChen Gang 28081da177e4SLinus Torvalds /* 28091da177e4SLinus Torvalds * The vm_pgoff of a purely anonymous vma should be irrelevant 28101da177e4SLinus Torvalds * until its first write fault, when page's anon_vma and index 28111da177e4SLinus Torvalds * are set. But now set the vm_pgoff it will almost certainly 28121da177e4SLinus Torvalds * end up with (unless mremap moves it elsewhere before that 28131da177e4SLinus Torvalds * first wfault), so /proc/pid/maps tells a consistent story. 28141da177e4SLinus Torvalds * 28151da177e4SLinus Torvalds * By setting it to reflect the virtual start address of the 28161da177e4SLinus Torvalds * vma, merges and splits can happen in a seamless way, just 28171da177e4SLinus Torvalds * using the existing file pgoff checks and manipulations. 28181da177e4SLinus Torvalds * Similarly in do_mmap_pgoff and in do_brk. 28191da177e4SLinus Torvalds */ 28208a9cc3b5SOleg Nesterov if (vma_is_anonymous(vma)) { 28211da177e4SLinus Torvalds BUG_ON(vma->anon_vma); 28221da177e4SLinus Torvalds vma->vm_pgoff = vma->vm_start >> PAGE_SHIFT; 28231da177e4SLinus Torvalds } 28242b144498SSrikar Dronamraju 28251da177e4SLinus Torvalds vma_link(mm, vma, prev, rb_link, rb_parent); 28261da177e4SLinus Torvalds return 0; 28271da177e4SLinus Torvalds } 28281da177e4SLinus Torvalds 28291da177e4SLinus Torvalds /* 28301da177e4SLinus Torvalds * Copy the vma structure to a new location in the same mm, 28311da177e4SLinus Torvalds * prior to moving page table entries, to effect an mremap move. 28321da177e4SLinus Torvalds */ 28331da177e4SLinus Torvalds struct vm_area_struct *copy_vma(struct vm_area_struct **vmap, 283438a76013SMichel Lespinasse unsigned long addr, unsigned long len, pgoff_t pgoff, 283538a76013SMichel Lespinasse bool *need_rmap_locks) 28361da177e4SLinus Torvalds { 28371da177e4SLinus Torvalds struct vm_area_struct *vma = *vmap; 28381da177e4SLinus Torvalds unsigned long vma_start = vma->vm_start; 28391da177e4SLinus Torvalds struct mm_struct *mm = vma->vm_mm; 28401da177e4SLinus Torvalds struct vm_area_struct *new_vma, *prev; 28411da177e4SLinus Torvalds struct rb_node **rb_link, *rb_parent; 2842948f017bSAndrea Arcangeli bool faulted_in_anon_vma = true; 28431da177e4SLinus Torvalds 28441da177e4SLinus Torvalds /* 28451da177e4SLinus Torvalds * If anonymous vma has not yet been faulted, update new pgoff 28461da177e4SLinus Torvalds * to match new location, to increase its chance of merging. 28471da177e4SLinus Torvalds */ 2848ce75799bSOleg Nesterov if (unlikely(vma_is_anonymous(vma) && !vma->anon_vma)) { 28491da177e4SLinus Torvalds pgoff = addr >> PAGE_SHIFT; 2850948f017bSAndrea Arcangeli faulted_in_anon_vma = false; 2851948f017bSAndrea Arcangeli } 28521da177e4SLinus Torvalds 28536597d783SHugh Dickins if (find_vma_links(mm, addr, addr + len, &prev, &rb_link, &rb_parent)) 28546597d783SHugh Dickins return NULL; /* should never get here */ 28551da177e4SLinus Torvalds new_vma = vma_merge(mm, prev, addr, addr + len, vma->vm_flags, 285619a809afSAndrea Arcangeli vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma), 285719a809afSAndrea Arcangeli vma->vm_userfaultfd_ctx); 28581da177e4SLinus Torvalds if (new_vma) { 28591da177e4SLinus Torvalds /* 28601da177e4SLinus Torvalds * Source vma may have been merged into new_vma 28611da177e4SLinus Torvalds */ 2862948f017bSAndrea Arcangeli if (unlikely(vma_start >= new_vma->vm_start && 2863948f017bSAndrea Arcangeli vma_start < new_vma->vm_end)) { 2864948f017bSAndrea Arcangeli /* 2865948f017bSAndrea Arcangeli * The only way we can get a vma_merge with 2866948f017bSAndrea Arcangeli * self during an mremap is if the vma hasn't 2867948f017bSAndrea Arcangeli * been faulted in yet and we were allowed to 2868948f017bSAndrea Arcangeli * reset the dst vma->vm_pgoff to the 2869948f017bSAndrea Arcangeli * destination address of the mremap to allow 2870948f017bSAndrea Arcangeli * the merge to happen. mremap must change the 2871948f017bSAndrea Arcangeli * vm_pgoff linearity between src and dst vmas 2872948f017bSAndrea Arcangeli * (in turn preventing a vma_merge) to be 2873948f017bSAndrea Arcangeli * safe. It is only safe to keep the vm_pgoff 2874948f017bSAndrea Arcangeli * linear if there are no pages mapped yet. 2875948f017bSAndrea Arcangeli */ 287681d1b09cSSasha Levin VM_BUG_ON_VMA(faulted_in_anon_vma, new_vma); 287738a76013SMichel Lespinasse *vmap = vma = new_vma; 2878108d6642SMichel Lespinasse } 287938a76013SMichel Lespinasse *need_rmap_locks = (new_vma->vm_pgoff <= vma->vm_pgoff); 28801da177e4SLinus Torvalds } else { 2881e94b1766SChristoph Lameter new_vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL); 2882e3975891SChen Gang if (!new_vma) 2883e3975891SChen Gang goto out; 28841da177e4SLinus Torvalds *new_vma = *vma; 28851da177e4SLinus Torvalds new_vma->vm_start = addr; 28861da177e4SLinus Torvalds new_vma->vm_end = addr + len; 28871da177e4SLinus Torvalds new_vma->vm_pgoff = pgoff; 2888ef0855d3SOleg Nesterov if (vma_dup_policy(vma, new_vma)) 2889523d4e20SMichel Lespinasse goto out_free_vma; 2890523d4e20SMichel Lespinasse INIT_LIST_HEAD(&new_vma->anon_vma_chain); 2891523d4e20SMichel Lespinasse if (anon_vma_clone(new_vma, vma)) 2892523d4e20SMichel Lespinasse goto out_free_mempol; 2893e9714acfSKonstantin Khlebnikov if (new_vma->vm_file) 28941da177e4SLinus Torvalds get_file(new_vma->vm_file); 28951da177e4SLinus Torvalds if (new_vma->vm_ops && new_vma->vm_ops->open) 28961da177e4SLinus Torvalds new_vma->vm_ops->open(new_vma); 28971da177e4SLinus Torvalds vma_link(mm, new_vma, prev, rb_link, rb_parent); 289838a76013SMichel Lespinasse *need_rmap_locks = false; 28991da177e4SLinus Torvalds } 29001da177e4SLinus Torvalds return new_vma; 29015beb4930SRik van Riel 29025beb4930SRik van Riel out_free_mempol: 2903ef0855d3SOleg Nesterov mpol_put(vma_policy(new_vma)); 29045beb4930SRik van Riel out_free_vma: 29055beb4930SRik van Riel kmem_cache_free(vm_area_cachep, new_vma); 2906e3975891SChen Gang out: 29075beb4930SRik van Riel return NULL; 29081da177e4SLinus Torvalds } 2909119f657cSakpm@osdl.org 2910119f657cSakpm@osdl.org /* 2911119f657cSakpm@osdl.org * Return true if the calling process may expand its vm space by the passed 2912119f657cSakpm@osdl.org * number of pages 2913119f657cSakpm@osdl.org */ 291484638335SKonstantin Khlebnikov bool may_expand_vm(struct mm_struct *mm, vm_flags_t flags, unsigned long npages) 2915119f657cSakpm@osdl.org { 291684638335SKonstantin Khlebnikov if (mm->total_vm + npages > rlimit(RLIMIT_AS) >> PAGE_SHIFT) 291784638335SKonstantin Khlebnikov return false; 2918119f657cSakpm@osdl.org 2919d977d56cSKonstantin Khlebnikov if (is_data_mapping(flags) && 2920d977d56cSKonstantin Khlebnikov mm->data_vm + npages > rlimit(RLIMIT_DATA) >> PAGE_SHIFT) { 2921f4fcd558SKonstantin Khlebnikov /* Workaround for Valgrind */ 2922f4fcd558SKonstantin Khlebnikov if (rlimit(RLIMIT_DATA) == 0 && 2923f4fcd558SKonstantin Khlebnikov mm->data_vm + npages <= rlimit_max(RLIMIT_DATA) >> PAGE_SHIFT) 2924f4fcd558SKonstantin Khlebnikov return true; 2925f4fcd558SKonstantin Khlebnikov if (!ignore_rlimit_data) { 2926f4fcd558SKonstantin Khlebnikov pr_warn_once("%s (%d): VmData %lu exceed data ulimit %lu. Update limits or use boot option ignore_rlimit_data.\n", 2927d977d56cSKonstantin Khlebnikov current->comm, current->pid, 2928d977d56cSKonstantin Khlebnikov (mm->data_vm + npages) << PAGE_SHIFT, 2929d977d56cSKonstantin Khlebnikov rlimit(RLIMIT_DATA)); 2930d977d56cSKonstantin Khlebnikov return false; 2931d977d56cSKonstantin Khlebnikov } 2932f4fcd558SKonstantin Khlebnikov } 2933119f657cSakpm@osdl.org 293484638335SKonstantin Khlebnikov return true; 293584638335SKonstantin Khlebnikov } 293684638335SKonstantin Khlebnikov 293784638335SKonstantin Khlebnikov void vm_stat_account(struct mm_struct *mm, vm_flags_t flags, long npages) 293884638335SKonstantin Khlebnikov { 293984638335SKonstantin Khlebnikov mm->total_vm += npages; 294084638335SKonstantin Khlebnikov 2941d977d56cSKonstantin Khlebnikov if (is_exec_mapping(flags)) 294284638335SKonstantin Khlebnikov mm->exec_vm += npages; 2943d977d56cSKonstantin Khlebnikov else if (is_stack_mapping(flags)) 294484638335SKonstantin Khlebnikov mm->stack_vm += npages; 2945d977d56cSKonstantin Khlebnikov else if (is_data_mapping(flags)) 294684638335SKonstantin Khlebnikov mm->data_vm += npages; 2947119f657cSakpm@osdl.org } 2948fa5dc22fSRoland McGrath 2949a62c34bdSAndy Lutomirski static int special_mapping_fault(struct vm_area_struct *vma, 2950a62c34bdSAndy Lutomirski struct vm_fault *vmf); 2951a62c34bdSAndy Lutomirski 2952a62c34bdSAndy Lutomirski /* 2953a62c34bdSAndy Lutomirski * Having a close hook prevents vma merging regardless of flags. 2954a62c34bdSAndy Lutomirski */ 2955a62c34bdSAndy Lutomirski static void special_mapping_close(struct vm_area_struct *vma) 2956a62c34bdSAndy Lutomirski { 2957a62c34bdSAndy Lutomirski } 2958a62c34bdSAndy Lutomirski 2959a62c34bdSAndy Lutomirski static const char *special_mapping_name(struct vm_area_struct *vma) 2960a62c34bdSAndy Lutomirski { 2961a62c34bdSAndy Lutomirski return ((struct vm_special_mapping *)vma->vm_private_data)->name; 2962a62c34bdSAndy Lutomirski } 2963a62c34bdSAndy Lutomirski 2964b059a453SDmitry Safonov static int special_mapping_mremap(struct vm_area_struct *new_vma) 2965b059a453SDmitry Safonov { 2966b059a453SDmitry Safonov struct vm_special_mapping *sm = new_vma->vm_private_data; 2967b059a453SDmitry Safonov 2968b059a453SDmitry Safonov if (sm->mremap) 2969b059a453SDmitry Safonov return sm->mremap(sm, new_vma); 2970b059a453SDmitry Safonov return 0; 2971b059a453SDmitry Safonov } 2972b059a453SDmitry Safonov 2973a62c34bdSAndy Lutomirski static const struct vm_operations_struct special_mapping_vmops = { 2974a62c34bdSAndy Lutomirski .close = special_mapping_close, 2975a62c34bdSAndy Lutomirski .fault = special_mapping_fault, 2976b059a453SDmitry Safonov .mremap = special_mapping_mremap, 2977a62c34bdSAndy Lutomirski .name = special_mapping_name, 2978a62c34bdSAndy Lutomirski }; 2979a62c34bdSAndy Lutomirski 2980a62c34bdSAndy Lutomirski static const struct vm_operations_struct legacy_special_mapping_vmops = { 2981a62c34bdSAndy Lutomirski .close = special_mapping_close, 2982a62c34bdSAndy Lutomirski .fault = special_mapping_fault, 2983a62c34bdSAndy Lutomirski }; 2984fa5dc22fSRoland McGrath 2985b1d0e4f5SNick Piggin static int special_mapping_fault(struct vm_area_struct *vma, 2986b1d0e4f5SNick Piggin struct vm_fault *vmf) 2987fa5dc22fSRoland McGrath { 2988b1d0e4f5SNick Piggin pgoff_t pgoff; 2989fa5dc22fSRoland McGrath struct page **pages; 2990fa5dc22fSRoland McGrath 2991f872f540SAndy Lutomirski if (vma->vm_ops == &legacy_special_mapping_vmops) { 2992a62c34bdSAndy Lutomirski pages = vma->vm_private_data; 2993f872f540SAndy Lutomirski } else { 2994f872f540SAndy Lutomirski struct vm_special_mapping *sm = vma->vm_private_data; 2995f872f540SAndy Lutomirski 2996f872f540SAndy Lutomirski if (sm->fault) 2997f872f540SAndy Lutomirski return sm->fault(sm, vma, vmf); 2998f872f540SAndy Lutomirski 2999f872f540SAndy Lutomirski pages = sm->pages; 3000f872f540SAndy Lutomirski } 3001a62c34bdSAndy Lutomirski 30028a9cc3b5SOleg Nesterov for (pgoff = vmf->pgoff; pgoff && *pages; ++pages) 3003b1d0e4f5SNick Piggin pgoff--; 3004fa5dc22fSRoland McGrath 3005fa5dc22fSRoland McGrath if (*pages) { 3006fa5dc22fSRoland McGrath struct page *page = *pages; 3007fa5dc22fSRoland McGrath get_page(page); 3008b1d0e4f5SNick Piggin vmf->page = page; 3009b1d0e4f5SNick Piggin return 0; 3010fa5dc22fSRoland McGrath } 3011fa5dc22fSRoland McGrath 3012b1d0e4f5SNick Piggin return VM_FAULT_SIGBUS; 3013fa5dc22fSRoland McGrath } 3014fa5dc22fSRoland McGrath 3015a62c34bdSAndy Lutomirski static struct vm_area_struct *__install_special_mapping( 3016a62c34bdSAndy Lutomirski struct mm_struct *mm, 3017fa5dc22fSRoland McGrath unsigned long addr, unsigned long len, 301827f28b97SChen Gang unsigned long vm_flags, void *priv, 301927f28b97SChen Gang const struct vm_operations_struct *ops) 3020fa5dc22fSRoland McGrath { 3021462e635eSTavis Ormandy int ret; 3022fa5dc22fSRoland McGrath struct vm_area_struct *vma; 3023fa5dc22fSRoland McGrath 3024fa5dc22fSRoland McGrath vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL); 3025fa5dc22fSRoland McGrath if (unlikely(vma == NULL)) 30263935ed6aSStefani Seibold return ERR_PTR(-ENOMEM); 3027fa5dc22fSRoland McGrath 30285beb4930SRik van Riel INIT_LIST_HEAD(&vma->anon_vma_chain); 3029fa5dc22fSRoland McGrath vma->vm_mm = mm; 3030fa5dc22fSRoland McGrath vma->vm_start = addr; 3031fa5dc22fSRoland McGrath vma->vm_end = addr + len; 3032fa5dc22fSRoland McGrath 3033d9104d1cSCyrill Gorcunov vma->vm_flags = vm_flags | mm->def_flags | VM_DONTEXPAND | VM_SOFTDIRTY; 30343ed75eb8SColy Li vma->vm_page_prot = vm_get_page_prot(vma->vm_flags); 3035fa5dc22fSRoland McGrath 3036a62c34bdSAndy Lutomirski vma->vm_ops = ops; 3037a62c34bdSAndy Lutomirski vma->vm_private_data = priv; 3038fa5dc22fSRoland McGrath 3039462e635eSTavis Ormandy ret = insert_vm_struct(mm, vma); 3040462e635eSTavis Ormandy if (ret) 3041462e635eSTavis Ormandy goto out; 3042fa5dc22fSRoland McGrath 304384638335SKonstantin Khlebnikov vm_stat_account(mm, vma->vm_flags, len >> PAGE_SHIFT); 3044fa5dc22fSRoland McGrath 3045cdd6c482SIngo Molnar perf_event_mmap(vma); 3046089dd79dSPeter Zijlstra 30473935ed6aSStefani Seibold return vma; 3048462e635eSTavis Ormandy 3049462e635eSTavis Ormandy out: 3050462e635eSTavis Ormandy kmem_cache_free(vm_area_cachep, vma); 30513935ed6aSStefani Seibold return ERR_PTR(ret); 30523935ed6aSStefani Seibold } 30533935ed6aSStefani Seibold 3054a62c34bdSAndy Lutomirski /* 3055a62c34bdSAndy Lutomirski * Called with mm->mmap_sem held for writing. 3056a62c34bdSAndy Lutomirski * Insert a new vma covering the given region, with the given flags. 3057a62c34bdSAndy Lutomirski * Its pages are supplied by the given array of struct page *. 3058a62c34bdSAndy Lutomirski * The array can be shorter than len >> PAGE_SHIFT if it's null-terminated. 3059a62c34bdSAndy Lutomirski * The region past the last page supplied will always produce SIGBUS. 3060a62c34bdSAndy Lutomirski * The array pointer and the pages it points to are assumed to stay alive 3061a62c34bdSAndy Lutomirski * for as long as this mapping might exist. 3062a62c34bdSAndy Lutomirski */ 3063a62c34bdSAndy Lutomirski struct vm_area_struct *_install_special_mapping( 3064a62c34bdSAndy Lutomirski struct mm_struct *mm, 3065a62c34bdSAndy Lutomirski unsigned long addr, unsigned long len, 3066a62c34bdSAndy Lutomirski unsigned long vm_flags, const struct vm_special_mapping *spec) 3067a62c34bdSAndy Lutomirski { 306827f28b97SChen Gang return __install_special_mapping(mm, addr, len, vm_flags, (void *)spec, 306927f28b97SChen Gang &special_mapping_vmops); 3070a62c34bdSAndy Lutomirski } 3071a62c34bdSAndy Lutomirski 30723935ed6aSStefani Seibold int install_special_mapping(struct mm_struct *mm, 30733935ed6aSStefani Seibold unsigned long addr, unsigned long len, 30743935ed6aSStefani Seibold unsigned long vm_flags, struct page **pages) 30753935ed6aSStefani Seibold { 3076a62c34bdSAndy Lutomirski struct vm_area_struct *vma = __install_special_mapping( 307727f28b97SChen Gang mm, addr, len, vm_flags, (void *)pages, 307827f28b97SChen Gang &legacy_special_mapping_vmops); 30793935ed6aSStefani Seibold 308014bd5b45SDuan Jiong return PTR_ERR_OR_ZERO(vma); 3081fa5dc22fSRoland McGrath } 30827906d00cSAndrea Arcangeli 30837906d00cSAndrea Arcangeli static DEFINE_MUTEX(mm_all_locks_mutex); 30847906d00cSAndrea Arcangeli 3085454ed842SPeter Zijlstra static void vm_lock_anon_vma(struct mm_struct *mm, struct anon_vma *anon_vma) 30867906d00cSAndrea Arcangeli { 3087bf181b9fSMichel Lespinasse if (!test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_node)) { 30887906d00cSAndrea Arcangeli /* 30897906d00cSAndrea Arcangeli * The LSB of head.next can't change from under us 30907906d00cSAndrea Arcangeli * because we hold the mm_all_locks_mutex. 30917906d00cSAndrea Arcangeli */ 3092572043c9SJiri Kosina down_write_nest_lock(&anon_vma->root->rwsem, &mm->mmap_sem); 30937906d00cSAndrea Arcangeli /* 30947906d00cSAndrea Arcangeli * We can safely modify head.next after taking the 30955a505085SIngo Molnar * anon_vma->root->rwsem. If some other vma in this mm shares 30967906d00cSAndrea Arcangeli * the same anon_vma we won't take it again. 30977906d00cSAndrea Arcangeli * 30987906d00cSAndrea Arcangeli * No need of atomic instructions here, head.next 30997906d00cSAndrea Arcangeli * can't change from under us thanks to the 31005a505085SIngo Molnar * anon_vma->root->rwsem. 31017906d00cSAndrea Arcangeli */ 31027906d00cSAndrea Arcangeli if (__test_and_set_bit(0, (unsigned long *) 3103bf181b9fSMichel Lespinasse &anon_vma->root->rb_root.rb_node)) 31047906d00cSAndrea Arcangeli BUG(); 31057906d00cSAndrea Arcangeli } 31067906d00cSAndrea Arcangeli } 31077906d00cSAndrea Arcangeli 3108454ed842SPeter Zijlstra static void vm_lock_mapping(struct mm_struct *mm, struct address_space *mapping) 31097906d00cSAndrea Arcangeli { 31107906d00cSAndrea Arcangeli if (!test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) { 31117906d00cSAndrea Arcangeli /* 31127906d00cSAndrea Arcangeli * AS_MM_ALL_LOCKS can't change from under us because 31137906d00cSAndrea Arcangeli * we hold the mm_all_locks_mutex. 31147906d00cSAndrea Arcangeli * 31157906d00cSAndrea Arcangeli * Operations on ->flags have to be atomic because 31167906d00cSAndrea Arcangeli * even if AS_MM_ALL_LOCKS is stable thanks to the 31177906d00cSAndrea Arcangeli * mm_all_locks_mutex, there may be other cpus 31187906d00cSAndrea Arcangeli * changing other bitflags in parallel to us. 31197906d00cSAndrea Arcangeli */ 31207906d00cSAndrea Arcangeli if (test_and_set_bit(AS_MM_ALL_LOCKS, &mapping->flags)) 31217906d00cSAndrea Arcangeli BUG(); 3122c8c06efaSDavidlohr Bueso down_write_nest_lock(&mapping->i_mmap_rwsem, &mm->mmap_sem); 31237906d00cSAndrea Arcangeli } 31247906d00cSAndrea Arcangeli } 31257906d00cSAndrea Arcangeli 31267906d00cSAndrea Arcangeli /* 31277906d00cSAndrea Arcangeli * This operation locks against the VM for all pte/vma/mm related 31287906d00cSAndrea Arcangeli * operations that could ever happen on a certain mm. This includes 31297906d00cSAndrea Arcangeli * vmtruncate, try_to_unmap, and all page faults. 31307906d00cSAndrea Arcangeli * 31317906d00cSAndrea Arcangeli * The caller must take the mmap_sem in write mode before calling 31327906d00cSAndrea Arcangeli * mm_take_all_locks(). The caller isn't allowed to release the 31337906d00cSAndrea Arcangeli * mmap_sem until mm_drop_all_locks() returns. 31347906d00cSAndrea Arcangeli * 31357906d00cSAndrea Arcangeli * mmap_sem in write mode is required in order to block all operations 31367906d00cSAndrea Arcangeli * that could modify pagetables and free pages without need of 313727ba0644SKirill A. Shutemov * altering the vma layout. It's also needed in write mode to avoid new 31387906d00cSAndrea Arcangeli * anon_vmas to be associated with existing vmas. 31397906d00cSAndrea Arcangeli * 31407906d00cSAndrea Arcangeli * A single task can't take more than one mm_take_all_locks() in a row 31417906d00cSAndrea Arcangeli * or it would deadlock. 31427906d00cSAndrea Arcangeli * 3143bf181b9fSMichel Lespinasse * The LSB in anon_vma->rb_root.rb_node and the AS_MM_ALL_LOCKS bitflag in 31447906d00cSAndrea Arcangeli * mapping->flags avoid to take the same lock twice, if more than one 31457906d00cSAndrea Arcangeli * vma in this mm is backed by the same anon_vma or address_space. 31467906d00cSAndrea Arcangeli * 314788f306b6SKirill A. Shutemov * We take locks in following order, accordingly to comment at beginning 314888f306b6SKirill A. Shutemov * of mm/rmap.c: 314988f306b6SKirill A. Shutemov * - all hugetlbfs_i_mmap_rwsem_key locks (aka mapping->i_mmap_rwsem for 315088f306b6SKirill A. Shutemov * hugetlb mapping); 315188f306b6SKirill A. Shutemov * - all i_mmap_rwsem locks; 315288f306b6SKirill A. Shutemov * - all anon_vma->rwseml 315388f306b6SKirill A. Shutemov * 315488f306b6SKirill A. Shutemov * We can take all locks within these types randomly because the VM code 315588f306b6SKirill A. Shutemov * doesn't nest them and we protected from parallel mm_take_all_locks() by 315688f306b6SKirill A. Shutemov * mm_all_locks_mutex. 31577906d00cSAndrea Arcangeli * 31587906d00cSAndrea Arcangeli * mm_take_all_locks() and mm_drop_all_locks are expensive operations 31597906d00cSAndrea Arcangeli * that may have to take thousand of locks. 31607906d00cSAndrea Arcangeli * 31617906d00cSAndrea Arcangeli * mm_take_all_locks() can fail if it's interrupted by signals. 31627906d00cSAndrea Arcangeli */ 31637906d00cSAndrea Arcangeli int mm_take_all_locks(struct mm_struct *mm) 31647906d00cSAndrea Arcangeli { 31657906d00cSAndrea Arcangeli struct vm_area_struct *vma; 31665beb4930SRik van Riel struct anon_vma_chain *avc; 31677906d00cSAndrea Arcangeli 31687906d00cSAndrea Arcangeli BUG_ON(down_read_trylock(&mm->mmap_sem)); 31697906d00cSAndrea Arcangeli 31707906d00cSAndrea Arcangeli mutex_lock(&mm_all_locks_mutex); 31717906d00cSAndrea Arcangeli 31727906d00cSAndrea Arcangeli for (vma = mm->mmap; vma; vma = vma->vm_next) { 31737906d00cSAndrea Arcangeli if (signal_pending(current)) 31747906d00cSAndrea Arcangeli goto out_unlock; 317588f306b6SKirill A. Shutemov if (vma->vm_file && vma->vm_file->f_mapping && 317688f306b6SKirill A. Shutemov is_vm_hugetlb_page(vma)) 317788f306b6SKirill A. Shutemov vm_lock_mapping(mm, vma->vm_file->f_mapping); 317888f306b6SKirill A. Shutemov } 317988f306b6SKirill A. Shutemov 318088f306b6SKirill A. Shutemov for (vma = mm->mmap; vma; vma = vma->vm_next) { 318188f306b6SKirill A. Shutemov if (signal_pending(current)) 318288f306b6SKirill A. Shutemov goto out_unlock; 318388f306b6SKirill A. Shutemov if (vma->vm_file && vma->vm_file->f_mapping && 318488f306b6SKirill A. Shutemov !is_vm_hugetlb_page(vma)) 3185454ed842SPeter Zijlstra vm_lock_mapping(mm, vma->vm_file->f_mapping); 31867906d00cSAndrea Arcangeli } 31877cd5a02fSPeter Zijlstra 31887cd5a02fSPeter Zijlstra for (vma = mm->mmap; vma; vma = vma->vm_next) { 31897cd5a02fSPeter Zijlstra if (signal_pending(current)) 31907cd5a02fSPeter Zijlstra goto out_unlock; 31917cd5a02fSPeter Zijlstra if (vma->anon_vma) 31925beb4930SRik van Riel list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) 31935beb4930SRik van Riel vm_lock_anon_vma(mm, avc->anon_vma); 31947cd5a02fSPeter Zijlstra } 31957cd5a02fSPeter Zijlstra 3196584cff54SKautuk Consul return 0; 31977906d00cSAndrea Arcangeli 31987906d00cSAndrea Arcangeli out_unlock: 31997906d00cSAndrea Arcangeli mm_drop_all_locks(mm); 3200584cff54SKautuk Consul return -EINTR; 32017906d00cSAndrea Arcangeli } 32027906d00cSAndrea Arcangeli 32037906d00cSAndrea Arcangeli static void vm_unlock_anon_vma(struct anon_vma *anon_vma) 32047906d00cSAndrea Arcangeli { 3205bf181b9fSMichel Lespinasse if (test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_node)) { 32067906d00cSAndrea Arcangeli /* 32077906d00cSAndrea Arcangeli * The LSB of head.next can't change to 0 from under 32087906d00cSAndrea Arcangeli * us because we hold the mm_all_locks_mutex. 32097906d00cSAndrea Arcangeli * 32107906d00cSAndrea Arcangeli * We must however clear the bitflag before unlocking 3211bf181b9fSMichel Lespinasse * the vma so the users using the anon_vma->rb_root will 32127906d00cSAndrea Arcangeli * never see our bitflag. 32137906d00cSAndrea Arcangeli * 32147906d00cSAndrea Arcangeli * No need of atomic instructions here, head.next 32157906d00cSAndrea Arcangeli * can't change from under us until we release the 32165a505085SIngo Molnar * anon_vma->root->rwsem. 32177906d00cSAndrea Arcangeli */ 32187906d00cSAndrea Arcangeli if (!__test_and_clear_bit(0, (unsigned long *) 3219bf181b9fSMichel Lespinasse &anon_vma->root->rb_root.rb_node)) 32207906d00cSAndrea Arcangeli BUG(); 322108b52706SKonstantin Khlebnikov anon_vma_unlock_write(anon_vma); 32227906d00cSAndrea Arcangeli } 32237906d00cSAndrea Arcangeli } 32247906d00cSAndrea Arcangeli 32257906d00cSAndrea Arcangeli static void vm_unlock_mapping(struct address_space *mapping) 32267906d00cSAndrea Arcangeli { 32277906d00cSAndrea Arcangeli if (test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) { 32287906d00cSAndrea Arcangeli /* 32297906d00cSAndrea Arcangeli * AS_MM_ALL_LOCKS can't change to 0 from under us 32307906d00cSAndrea Arcangeli * because we hold the mm_all_locks_mutex. 32317906d00cSAndrea Arcangeli */ 323283cde9e8SDavidlohr Bueso i_mmap_unlock_write(mapping); 32337906d00cSAndrea Arcangeli if (!test_and_clear_bit(AS_MM_ALL_LOCKS, 32347906d00cSAndrea Arcangeli &mapping->flags)) 32357906d00cSAndrea Arcangeli BUG(); 32367906d00cSAndrea Arcangeli } 32377906d00cSAndrea Arcangeli } 32387906d00cSAndrea Arcangeli 32397906d00cSAndrea Arcangeli /* 32407906d00cSAndrea Arcangeli * The mmap_sem cannot be released by the caller until 32417906d00cSAndrea Arcangeli * mm_drop_all_locks() returns. 32427906d00cSAndrea Arcangeli */ 32437906d00cSAndrea Arcangeli void mm_drop_all_locks(struct mm_struct *mm) 32447906d00cSAndrea Arcangeli { 32457906d00cSAndrea Arcangeli struct vm_area_struct *vma; 32465beb4930SRik van Riel struct anon_vma_chain *avc; 32477906d00cSAndrea Arcangeli 32487906d00cSAndrea Arcangeli BUG_ON(down_read_trylock(&mm->mmap_sem)); 32497906d00cSAndrea Arcangeli BUG_ON(!mutex_is_locked(&mm_all_locks_mutex)); 32507906d00cSAndrea Arcangeli 32517906d00cSAndrea Arcangeli for (vma = mm->mmap; vma; vma = vma->vm_next) { 32527906d00cSAndrea Arcangeli if (vma->anon_vma) 32535beb4930SRik van Riel list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) 32545beb4930SRik van Riel vm_unlock_anon_vma(avc->anon_vma); 32557906d00cSAndrea Arcangeli if (vma->vm_file && vma->vm_file->f_mapping) 32567906d00cSAndrea Arcangeli vm_unlock_mapping(vma->vm_file->f_mapping); 32577906d00cSAndrea Arcangeli } 32587906d00cSAndrea Arcangeli 32597906d00cSAndrea Arcangeli mutex_unlock(&mm_all_locks_mutex); 32607906d00cSAndrea Arcangeli } 32618feae131SDavid Howells 32628feae131SDavid Howells /* 32638feae131SDavid Howells * initialise the VMA slab 32648feae131SDavid Howells */ 32658feae131SDavid Howells void __init mmap_init(void) 32668feae131SDavid Howells { 326700a62ce9SKOSAKI Motohiro int ret; 326800a62ce9SKOSAKI Motohiro 3269908c7f19STejun Heo ret = percpu_counter_init(&vm_committed_as, 0, GFP_KERNEL); 327000a62ce9SKOSAKI Motohiro VM_BUG_ON(ret); 32718feae131SDavid Howells } 3272c9b1d098SAndrew Shewmaker 3273c9b1d098SAndrew Shewmaker /* 3274c9b1d098SAndrew Shewmaker * Initialise sysctl_user_reserve_kbytes. 3275c9b1d098SAndrew Shewmaker * 3276c9b1d098SAndrew Shewmaker * This is intended to prevent a user from starting a single memory hogging 3277c9b1d098SAndrew Shewmaker * process, such that they cannot recover (kill the hog) in OVERCOMMIT_NEVER 3278c9b1d098SAndrew Shewmaker * mode. 3279c9b1d098SAndrew Shewmaker * 3280c9b1d098SAndrew Shewmaker * The default value is min(3% of free memory, 128MB) 3281c9b1d098SAndrew Shewmaker * 128MB is enough to recover with sshd/login, bash, and top/kill. 3282c9b1d098SAndrew Shewmaker */ 32831640879aSAndrew Shewmaker static int init_user_reserve(void) 3284c9b1d098SAndrew Shewmaker { 3285c9b1d098SAndrew Shewmaker unsigned long free_kbytes; 3286c9b1d098SAndrew Shewmaker 3287c9b1d098SAndrew Shewmaker free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10); 3288c9b1d098SAndrew Shewmaker 3289c9b1d098SAndrew Shewmaker sysctl_user_reserve_kbytes = min(free_kbytes / 32, 1UL << 17); 3290c9b1d098SAndrew Shewmaker return 0; 3291c9b1d098SAndrew Shewmaker } 3292a64fb3cdSPaul Gortmaker subsys_initcall(init_user_reserve); 32934eeab4f5SAndrew Shewmaker 32944eeab4f5SAndrew Shewmaker /* 32954eeab4f5SAndrew Shewmaker * Initialise sysctl_admin_reserve_kbytes. 32964eeab4f5SAndrew Shewmaker * 32974eeab4f5SAndrew Shewmaker * The purpose of sysctl_admin_reserve_kbytes is to allow the sys admin 32984eeab4f5SAndrew Shewmaker * to log in and kill a memory hogging process. 32994eeab4f5SAndrew Shewmaker * 33004eeab4f5SAndrew Shewmaker * Systems with more than 256MB will reserve 8MB, enough to recover 33014eeab4f5SAndrew Shewmaker * with sshd, bash, and top in OVERCOMMIT_GUESS. Smaller systems will 33024eeab4f5SAndrew Shewmaker * only reserve 3% of free pages by default. 33034eeab4f5SAndrew Shewmaker */ 33041640879aSAndrew Shewmaker static int init_admin_reserve(void) 33054eeab4f5SAndrew Shewmaker { 33064eeab4f5SAndrew Shewmaker unsigned long free_kbytes; 33074eeab4f5SAndrew Shewmaker 33084eeab4f5SAndrew Shewmaker free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10); 33094eeab4f5SAndrew Shewmaker 33104eeab4f5SAndrew Shewmaker sysctl_admin_reserve_kbytes = min(free_kbytes / 32, 1UL << 13); 33114eeab4f5SAndrew Shewmaker return 0; 33124eeab4f5SAndrew Shewmaker } 3313a64fb3cdSPaul Gortmaker subsys_initcall(init_admin_reserve); 33141640879aSAndrew Shewmaker 33151640879aSAndrew Shewmaker /* 33161640879aSAndrew Shewmaker * Reinititalise user and admin reserves if memory is added or removed. 33171640879aSAndrew Shewmaker * 33181640879aSAndrew Shewmaker * The default user reserve max is 128MB, and the default max for the 33191640879aSAndrew Shewmaker * admin reserve is 8MB. These are usually, but not always, enough to 33201640879aSAndrew Shewmaker * enable recovery from a memory hogging process using login/sshd, a shell, 33211640879aSAndrew Shewmaker * and tools like top. It may make sense to increase or even disable the 33221640879aSAndrew Shewmaker * reserve depending on the existence of swap or variations in the recovery 33231640879aSAndrew Shewmaker * tools. So, the admin may have changed them. 33241640879aSAndrew Shewmaker * 33251640879aSAndrew Shewmaker * If memory is added and the reserves have been eliminated or increased above 33261640879aSAndrew Shewmaker * the default max, then we'll trust the admin. 33271640879aSAndrew Shewmaker * 33281640879aSAndrew Shewmaker * If memory is removed and there isn't enough free memory, then we 33291640879aSAndrew Shewmaker * need to reset the reserves. 33301640879aSAndrew Shewmaker * 33311640879aSAndrew Shewmaker * Otherwise keep the reserve set by the admin. 33321640879aSAndrew Shewmaker */ 33331640879aSAndrew Shewmaker static int reserve_mem_notifier(struct notifier_block *nb, 33341640879aSAndrew Shewmaker unsigned long action, void *data) 33351640879aSAndrew Shewmaker { 33361640879aSAndrew Shewmaker unsigned long tmp, free_kbytes; 33371640879aSAndrew Shewmaker 33381640879aSAndrew Shewmaker switch (action) { 33391640879aSAndrew Shewmaker case MEM_ONLINE: 33401640879aSAndrew Shewmaker /* Default max is 128MB. Leave alone if modified by operator. */ 33411640879aSAndrew Shewmaker tmp = sysctl_user_reserve_kbytes; 33421640879aSAndrew Shewmaker if (0 < tmp && tmp < (1UL << 17)) 33431640879aSAndrew Shewmaker init_user_reserve(); 33441640879aSAndrew Shewmaker 33451640879aSAndrew Shewmaker /* Default max is 8MB. Leave alone if modified by operator. */ 33461640879aSAndrew Shewmaker tmp = sysctl_admin_reserve_kbytes; 33471640879aSAndrew Shewmaker if (0 < tmp && tmp < (1UL << 13)) 33481640879aSAndrew Shewmaker init_admin_reserve(); 33491640879aSAndrew Shewmaker 33501640879aSAndrew Shewmaker break; 33511640879aSAndrew Shewmaker case MEM_OFFLINE: 33521640879aSAndrew Shewmaker free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10); 33531640879aSAndrew Shewmaker 33541640879aSAndrew Shewmaker if (sysctl_user_reserve_kbytes > free_kbytes) { 33551640879aSAndrew Shewmaker init_user_reserve(); 33561640879aSAndrew Shewmaker pr_info("vm.user_reserve_kbytes reset to %lu\n", 33571640879aSAndrew Shewmaker sysctl_user_reserve_kbytes); 33581640879aSAndrew Shewmaker } 33591640879aSAndrew Shewmaker 33601640879aSAndrew Shewmaker if (sysctl_admin_reserve_kbytes > free_kbytes) { 33611640879aSAndrew Shewmaker init_admin_reserve(); 33621640879aSAndrew Shewmaker pr_info("vm.admin_reserve_kbytes reset to %lu\n", 33631640879aSAndrew Shewmaker sysctl_admin_reserve_kbytes); 33641640879aSAndrew Shewmaker } 33651640879aSAndrew Shewmaker break; 33661640879aSAndrew Shewmaker default: 33671640879aSAndrew Shewmaker break; 33681640879aSAndrew Shewmaker } 33691640879aSAndrew Shewmaker return NOTIFY_OK; 33701640879aSAndrew Shewmaker } 33711640879aSAndrew Shewmaker 33721640879aSAndrew Shewmaker static struct notifier_block reserve_mem_nb = { 33731640879aSAndrew Shewmaker .notifier_call = reserve_mem_notifier, 33741640879aSAndrew Shewmaker }; 33751640879aSAndrew Shewmaker 33761640879aSAndrew Shewmaker static int __meminit init_reserve_notifier(void) 33771640879aSAndrew Shewmaker { 33781640879aSAndrew Shewmaker if (register_hotmemory_notifier(&reserve_mem_nb)) 3379b1de0d13SMitchel Humpherys pr_err("Failed registering memory add/remove notifier for admin reserve\n"); 33801640879aSAndrew Shewmaker 33811640879aSAndrew Shewmaker return 0; 33821640879aSAndrew Shewmaker } 3383a64fb3cdSPaul Gortmaker subsys_initcall(init_reserve_notifier); 3384