Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1 | /* |
| 2 | * mm/mmap.c |
| 3 | * |
| 4 | * Written by obz. |
| 5 | * |
Alan Cox | 046c688 | 2009-01-05 14:06:29 +0000 | [diff] [blame] | 6 | * Address space accounting code <alan@lxorguk.ukuu.org.uk> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 7 | */ |
| 8 | |
Mitchel Humpherys | b1de0d1 | 2014-06-06 14:38:30 -0700 | [diff] [blame] | 9 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt |
| 10 | |
Cyril Hrubis | e8420a8 | 2013-04-29 15:08:33 -0700 | [diff] [blame] | 11 | #include <linux/kernel.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 12 | #include <linux/slab.h> |
Alexey Dobriyan | 4af3c9c | 2007-10-16 23:29:23 -0700 | [diff] [blame] | 13 | #include <linux/backing-dev.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 14 | #include <linux/mm.h> |
Davidlohr Bueso | 615d6e8 | 2014-04-07 15:37:25 -0700 | [diff] [blame] | 15 | #include <linux/vmacache.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 16 | #include <linux/shm.h> |
| 17 | #include <linux/mman.h> |
| 18 | #include <linux/pagemap.h> |
| 19 | #include <linux/swap.h> |
| 20 | #include <linux/syscalls.h> |
Randy.Dunlap | c59ede7 | 2006-01-11 12:17:46 -0800 | [diff] [blame] | 21 | #include <linux/capability.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 22 | #include <linux/init.h> |
| 23 | #include <linux/file.h> |
| 24 | #include <linux/fs.h> |
| 25 | #include <linux/personality.h> |
| 26 | #include <linux/security.h> |
| 27 | #include <linux/hugetlb.h> |
Hugh Dickins | c01d5b3 | 2016-07-26 15:26:15 -0700 | [diff] [blame] | 28 | #include <linux/shmem_fs.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 29 | #include <linux/profile.h> |
Paul Gortmaker | b95f1b31 | 2011-10-16 02:01:52 -0400 | [diff] [blame] | 30 | #include <linux/export.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 31 | #include <linux/mount.h> |
| 32 | #include <linux/mempolicy.h> |
| 33 | #include <linux/rmap.h> |
Andrea Arcangeli | cddb8a5 | 2008-07-28 15:46:29 -0700 | [diff] [blame] | 34 | #include <linux/mmu_notifier.h> |
Konstantin Khlebnikov | 82f71ae | 2014-08-06 16:06:36 -0700 | [diff] [blame] | 35 | #include <linux/mmdebug.h> |
Ingo Molnar | cdd6c48 | 2009-09-21 12:02:48 +0200 | [diff] [blame] | 36 | #include <linux/perf_event.h> |
Al Viro | 120a795 | 2010-10-30 02:54:44 -0400 | [diff] [blame] | 37 | #include <linux/audit.h> |
Andrea Arcangeli | b15d00b | 2011-01-13 15:46:59 -0800 | [diff] [blame] | 38 | #include <linux/khugepaged.h> |
Srikar Dronamraju | 2b14449 | 2012-02-09 14:56:42 +0530 | [diff] [blame] | 39 | #include <linux/uprobes.h> |
Michel Lespinasse | d373718 | 2012-12-11 16:01:38 -0800 | [diff] [blame] | 40 | #include <linux/rbtree_augmented.h> |
Andrew Shewmaker | 1640879 | 2013-04-29 15:08:12 -0700 | [diff] [blame] | 41 | #include <linux/notifier.h> |
| 42 | #include <linux/memory.h> |
Mitchel Humpherys | b1de0d1 | 2014-06-06 14:38:30 -0700 | [diff] [blame] | 43 | #include <linux/printk.h> |
Andrea Arcangeli | 19a809a | 2015-09-04 15:46:24 -0700 | [diff] [blame] | 44 | #include <linux/userfaultfd_k.h> |
Konstantin Khlebnikov | d977d56 | 2016-02-02 16:57:43 -0800 | [diff] [blame] | 45 | #include <linux/moduleparam.h> |
Dave Hansen | 62b5f7d | 2016-02-12 13:02:40 -0800 | [diff] [blame] | 46 | #include <linux/pkeys.h> |
Andrea Arcangeli | 07b039e9 | 2017-09-06 16:25:00 -0700 | [diff] [blame] | 47 | #include <linux/oom.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 48 | |
| 49 | #include <asm/uaccess.h> |
| 50 | #include <asm/cacheflush.h> |
| 51 | #include <asm/tlb.h> |
Jeremy Fitzhardinge | d6dd61c | 2007-05-02 19:27:14 +0200 | [diff] [blame] | 52 | #include <asm/mmu_context.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 53 | |
Jan Beulich | 42b7772 | 2008-07-23 21:27:10 -0700 | [diff] [blame] | 54 | #include "internal.h" |
| 55 | |
Kirill Korotaev | 3a45975 | 2006-09-07 14:17:04 +0400 | [diff] [blame] | 56 | #ifndef arch_mmap_check |
| 57 | #define arch_mmap_check(addr, len, flags) (0) |
| 58 | #endif |
| 59 | |
Daniel Cashman | d07e225 | 2016-01-14 15:19:53 -0800 | [diff] [blame] | 60 | #ifdef CONFIG_HAVE_ARCH_MMAP_RND_BITS |
| 61 | const int mmap_rnd_bits_min = CONFIG_ARCH_MMAP_RND_BITS_MIN; |
| 62 | const int mmap_rnd_bits_max = CONFIG_ARCH_MMAP_RND_BITS_MAX; |
| 63 | int mmap_rnd_bits __read_mostly = CONFIG_ARCH_MMAP_RND_BITS; |
| 64 | #endif |
| 65 | #ifdef CONFIG_HAVE_ARCH_MMAP_RND_COMPAT_BITS |
| 66 | const int mmap_rnd_compat_bits_min = CONFIG_ARCH_MMAP_RND_COMPAT_BITS_MIN; |
| 67 | const int mmap_rnd_compat_bits_max = CONFIG_ARCH_MMAP_RND_COMPAT_BITS_MAX; |
| 68 | int mmap_rnd_compat_bits __read_mostly = CONFIG_ARCH_MMAP_RND_COMPAT_BITS; |
| 69 | #endif |
| 70 | |
Konstantin Khlebnikov | f4fcd55 | 2016-05-20 16:57:45 -0700 | [diff] [blame] | 71 | static bool ignore_rlimit_data; |
Konstantin Khlebnikov | d977d56 | 2016-02-02 16:57:43 -0800 | [diff] [blame] | 72 | core_param(ignore_rlimit_data, ignore_rlimit_data, bool, 0644); |
Daniel Cashman | d07e225 | 2016-01-14 15:19:53 -0800 | [diff] [blame] | 73 | |
Hugh Dickins | e0da382 | 2005-04-19 13:29:15 -0700 | [diff] [blame] | 74 | static void unmap_region(struct mm_struct *mm, |
| 75 | struct vm_area_struct *vma, struct vm_area_struct *prev, |
| 76 | unsigned long start, unsigned long end); |
| 77 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 78 | /* description of effects of mapping type and prot in current implementation. |
| 79 | * this is due to the limited x86 page protection hardware. The expected |
| 80 | * behavior is in parens: |
| 81 | * |
| 82 | * map_type prot |
| 83 | * PROT_NONE PROT_READ PROT_WRITE PROT_EXEC |
| 84 | * MAP_SHARED r: (no) no r: (yes) yes r: (no) yes r: (no) yes |
| 85 | * w: (no) no w: (no) no w: (yes) yes w: (no) no |
| 86 | * x: (no) no x: (no) yes x: (no) yes x: (yes) yes |
vishnu.ps | cc71aba | 2014-10-09 15:26:29 -0700 | [diff] [blame] | 87 | * |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 88 | * MAP_PRIVATE r: (no) no r: (yes) yes r: (no) yes r: (no) yes |
| 89 | * w: (no) no w: (no) no w: (copy) copy w: (no) no |
| 90 | * x: (no) no x: (no) yes x: (no) yes x: (yes) yes |
| 91 | * |
Catalin Marinas | cab15ce | 2016-08-11 18:44:50 +0100 | [diff] [blame] | 92 | * On arm64, PROT_EXEC has the following behaviour for both MAP_SHARED and |
| 93 | * MAP_PRIVATE: |
| 94 | * r: (no) no |
| 95 | * w: (no) no |
| 96 | * x: (yes) yes |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 97 | */ |
| 98 | pgprot_t protection_map[16] = { |
| 99 | __P000, __P001, __P010, __P011, __P100, __P101, __P110, __P111, |
| 100 | __S000, __S001, __S010, __S011, __S100, __S101, __S110, __S111 |
| 101 | }; |
| 102 | |
Hugh Dickins | 804af2c | 2006-07-26 21:39:49 +0100 | [diff] [blame] | 103 | pgprot_t vm_get_page_prot(unsigned long vm_flags) |
| 104 | { |
Dave Kleikamp | b845f31 | 2008-07-08 00:28:51 +1000 | [diff] [blame] | 105 | return __pgprot(pgprot_val(protection_map[vm_flags & |
| 106 | (VM_READ|VM_WRITE|VM_EXEC|VM_SHARED)]) | |
| 107 | pgprot_val(arch_vm_get_page_prot(vm_flags))); |
Hugh Dickins | 804af2c | 2006-07-26 21:39:49 +0100 | [diff] [blame] | 108 | } |
| 109 | EXPORT_SYMBOL(vm_get_page_prot); |
| 110 | |
Peter Feiner | 64e4550 | 2014-10-13 15:55:46 -0700 | [diff] [blame] | 111 | static pgprot_t vm_pgprot_modify(pgprot_t oldprot, unsigned long vm_flags) |
| 112 | { |
| 113 | return pgprot_modify(oldprot, vm_get_page_prot(vm_flags)); |
| 114 | } |
| 115 | |
| 116 | /* Update vma->vm_page_prot to reflect vma->vm_flags. */ |
| 117 | void vma_set_page_prot(struct vm_area_struct *vma) |
| 118 | { |
| 119 | unsigned long vm_flags = vma->vm_flags; |
Andrea Arcangeli | 6d2329f | 2016-10-07 17:01:22 -0700 | [diff] [blame] | 120 | pgprot_t vm_page_prot; |
Peter Feiner | 64e4550 | 2014-10-13 15:55:46 -0700 | [diff] [blame] | 121 | |
Andrea Arcangeli | 6d2329f | 2016-10-07 17:01:22 -0700 | [diff] [blame] | 122 | vm_page_prot = vm_pgprot_modify(vma->vm_page_prot, vm_flags); |
| 123 | if (vma_wants_writenotify(vma, vm_page_prot)) { |
Peter Feiner | 64e4550 | 2014-10-13 15:55:46 -0700 | [diff] [blame] | 124 | vm_flags &= ~VM_SHARED; |
Andrea Arcangeli | 6d2329f | 2016-10-07 17:01:22 -0700 | [diff] [blame] | 125 | vm_page_prot = vm_pgprot_modify(vm_page_prot, vm_flags); |
Peter Feiner | 64e4550 | 2014-10-13 15:55:46 -0700 | [diff] [blame] | 126 | } |
Andrea Arcangeli | 6d2329f | 2016-10-07 17:01:22 -0700 | [diff] [blame] | 127 | /* remove_protection_ptes reads vma->vm_page_prot without mmap_sem */ |
| 128 | WRITE_ONCE(vma->vm_page_prot, vm_page_prot); |
Peter Feiner | 64e4550 | 2014-10-13 15:55:46 -0700 | [diff] [blame] | 129 | } |
| 130 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 131 | /* |
Davidlohr Bueso | c8c06ef | 2014-12-12 16:54:24 -0800 | [diff] [blame] | 132 | * Requires inode->i_mapping->i_mmap_rwsem |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 133 | */ |
| 134 | static void __remove_shared_vm_struct(struct vm_area_struct *vma, |
| 135 | struct file *file, struct address_space *mapping) |
| 136 | { |
| 137 | if (vma->vm_flags & VM_DENYWRITE) |
Al Viro | 496ad9a | 2013-01-23 17:07:38 -0500 | [diff] [blame] | 138 | atomic_inc(&file_inode(file)->i_writecount); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 139 | if (vma->vm_flags & VM_SHARED) |
David Herrmann | 4bb5f5d | 2014-08-08 14:25:25 -0700 | [diff] [blame] | 140 | mapping_unmap_writable(mapping); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 141 | |
| 142 | flush_dcache_mmap_lock(mapping); |
Kirill A. Shutemov | 27ba064 | 2015-02-10 14:09:59 -0800 | [diff] [blame] | 143 | vma_interval_tree_remove(vma, &mapping->i_mmap); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 144 | flush_dcache_mmap_unlock(mapping); |
| 145 | } |
| 146 | |
| 147 | /* |
Michel Lespinasse | 6b2dbba | 2012-10-08 16:31:25 -0700 | [diff] [blame] | 148 | * Unlink a file-based vm structure from its interval tree, to hide |
Hugh Dickins | a8fb561 | 2005-10-29 18:15:57 -0700 | [diff] [blame] | 149 | * vma from rmap and vmtruncate before freeing its page tables. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 150 | */ |
Hugh Dickins | a8fb561 | 2005-10-29 18:15:57 -0700 | [diff] [blame] | 151 | void unlink_file_vma(struct vm_area_struct *vma) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 152 | { |
| 153 | struct file *file = vma->vm_file; |
| 154 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 155 | if (file) { |
| 156 | struct address_space *mapping = file->f_mapping; |
Davidlohr Bueso | 83cde9e | 2014-12-12 16:54:21 -0800 | [diff] [blame] | 157 | i_mmap_lock_write(mapping); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 158 | __remove_shared_vm_struct(vma, file, mapping); |
Davidlohr Bueso | 83cde9e | 2014-12-12 16:54:21 -0800 | [diff] [blame] | 159 | i_mmap_unlock_write(mapping); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 160 | } |
Hugh Dickins | a8fb561 | 2005-10-29 18:15:57 -0700 | [diff] [blame] | 161 | } |
| 162 | |
| 163 | /* |
| 164 | * Close a vm structure and free it, returning the next. |
| 165 | */ |
| 166 | static struct vm_area_struct *remove_vma(struct vm_area_struct *vma) |
| 167 | { |
| 168 | struct vm_area_struct *next = vma->vm_next; |
| 169 | |
Hugh Dickins | a8fb561 | 2005-10-29 18:15:57 -0700 | [diff] [blame] | 170 | might_sleep(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 171 | if (vma->vm_ops && vma->vm_ops->close) |
| 172 | vma->vm_ops->close(vma); |
Konstantin Khlebnikov | e9714ac | 2012-10-08 16:28:54 -0700 | [diff] [blame] | 173 | if (vma->vm_file) |
Hugh Dickins | a8fb561 | 2005-10-29 18:15:57 -0700 | [diff] [blame] | 174 | fput(vma->vm_file); |
Lee Schermerhorn | f0be3d3 | 2008-04-28 02:13:08 -0700 | [diff] [blame] | 175 | mpol_put(vma_policy(vma)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 176 | kmem_cache_free(vm_area_cachep, vma); |
Hugh Dickins | a8fb561 | 2005-10-29 18:15:57 -0700 | [diff] [blame] | 177 | return next; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 178 | } |
| 179 | |
Linus Torvalds | 5d22fc2 | 2016-05-27 15:57:31 -0700 | [diff] [blame] | 180 | static int do_brk(unsigned long addr, unsigned long len); |
Linus Torvalds | e4eb1ff | 2012-04-20 15:35:40 -0700 | [diff] [blame] | 181 | |
Heiko Carstens | 6a6160a | 2009-01-14 14:14:15 +0100 | [diff] [blame] | 182 | SYSCALL_DEFINE1(brk, unsigned long, brk) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 183 | { |
Cyrill Gorcunov | 8764b33 | 2014-10-09 15:27:32 -0700 | [diff] [blame] | 184 | unsigned long retval; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 185 | unsigned long newbrk, oldbrk; |
| 186 | struct mm_struct *mm = current->mm; |
Hugh Dickins | cfc0eb40 | 2017-06-19 04:03:24 -0700 | [diff] [blame] | 187 | struct vm_area_struct *next; |
Jiri Kosina | a5b4592 | 2008-06-05 22:46:05 -0700 | [diff] [blame] | 188 | unsigned long min_brk; |
Michel Lespinasse | 128557f | 2013-02-22 16:32:40 -0800 | [diff] [blame] | 189 | bool populate; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 190 | |
Michal Hocko | dc0ef0d | 2016-05-23 16:25:27 -0700 | [diff] [blame] | 191 | if (down_write_killable(&mm->mmap_sem)) |
| 192 | return -EINTR; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 193 | |
Jiri Kosina | a5b4592 | 2008-06-05 22:46:05 -0700 | [diff] [blame] | 194 | #ifdef CONFIG_COMPAT_BRK |
Jiri Kosina | 5520e89 | 2011-01-13 15:47:23 -0800 | [diff] [blame] | 195 | /* |
| 196 | * CONFIG_COMPAT_BRK can still be overridden by setting |
| 197 | * randomize_va_space to 2, which will still cause mm->start_brk |
| 198 | * to be arbitrarily shifted |
| 199 | */ |
Jiri Kosina | 4471a67 | 2011-04-14 15:22:09 -0700 | [diff] [blame] | 200 | if (current->brk_randomized) |
Jiri Kosina | 5520e89 | 2011-01-13 15:47:23 -0800 | [diff] [blame] | 201 | min_brk = mm->start_brk; |
| 202 | else |
| 203 | min_brk = mm->end_data; |
Jiri Kosina | a5b4592 | 2008-06-05 22:46:05 -0700 | [diff] [blame] | 204 | #else |
| 205 | min_brk = mm->start_brk; |
| 206 | #endif |
| 207 | if (brk < min_brk) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 208 | goto out; |
Ram Gupta | 1e62419 | 2006-04-10 22:52:57 -0700 | [diff] [blame] | 209 | |
| 210 | /* |
| 211 | * Check against rlimit here. If this check is done later after the test |
| 212 | * of oldbrk with newbrk then it can escape the test and let the data |
| 213 | * segment grow beyond its set limit the in case where the limit is |
| 214 | * not page aligned -Ram Gupta |
| 215 | */ |
Cyrill Gorcunov | 8764b33 | 2014-10-09 15:27:32 -0700 | [diff] [blame] | 216 | if (check_data_rlimit(rlimit(RLIMIT_DATA), brk, mm->start_brk, |
| 217 | mm->end_data, mm->start_data)) |
Ram Gupta | 1e62419 | 2006-04-10 22:52:57 -0700 | [diff] [blame] | 218 | goto out; |
| 219 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 220 | newbrk = PAGE_ALIGN(brk); |
| 221 | oldbrk = PAGE_ALIGN(mm->brk); |
| 222 | if (oldbrk == newbrk) |
| 223 | goto set_brk; |
| 224 | |
| 225 | /* Always allow shrinking brk. */ |
| 226 | if (brk <= mm->brk) { |
| 227 | if (!do_munmap(mm, newbrk, oldbrk-newbrk)) |
| 228 | goto set_brk; |
| 229 | goto out; |
| 230 | } |
| 231 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 232 | /* Check against existing mmap mappings. */ |
Hugh Dickins | cfc0eb40 | 2017-06-19 04:03:24 -0700 | [diff] [blame] | 233 | next = find_vma(mm, oldbrk); |
| 234 | if (next && newbrk + PAGE_SIZE > vm_start_gap(next)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 235 | goto out; |
| 236 | |
| 237 | /* Ok, looks good - let it rip. */ |
Linus Torvalds | 5d22fc2 | 2016-05-27 15:57:31 -0700 | [diff] [blame] | 238 | if (do_brk(oldbrk, newbrk-oldbrk) < 0) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 239 | goto out; |
Michel Lespinasse | 128557f | 2013-02-22 16:32:40 -0800 | [diff] [blame] | 240 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 241 | set_brk: |
| 242 | mm->brk = brk; |
Michel Lespinasse | 128557f | 2013-02-22 16:32:40 -0800 | [diff] [blame] | 243 | populate = newbrk > oldbrk && (mm->def_flags & VM_LOCKED) != 0; |
| 244 | up_write(&mm->mmap_sem); |
| 245 | if (populate) |
| 246 | mm_populate(oldbrk, newbrk - oldbrk); |
| 247 | return brk; |
| 248 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 249 | out: |
| 250 | retval = mm->brk; |
| 251 | up_write(&mm->mmap_sem); |
| 252 | return retval; |
| 253 | } |
| 254 | |
Michel Lespinasse | d373718 | 2012-12-11 16:01:38 -0800 | [diff] [blame] | 255 | static long vma_compute_subtree_gap(struct vm_area_struct *vma) |
| 256 | { |
Hugh Dickins | cfc0eb40 | 2017-06-19 04:03:24 -0700 | [diff] [blame] | 257 | unsigned long max, prev_end, subtree_gap; |
| 258 | |
| 259 | /* |
| 260 | * Note: in the rare case of a VM_GROWSDOWN above a VM_GROWSUP, we |
| 261 | * allow two stack_guard_gaps between them here, and when choosing |
| 262 | * an unmapped area; whereas when expanding we only require one. |
| 263 | * That's a little inconsistent, but keeps the code here simpler. |
| 264 | */ |
| 265 | max = vm_start_gap(vma); |
| 266 | if (vma->vm_prev) { |
| 267 | prev_end = vm_end_gap(vma->vm_prev); |
| 268 | if (max > prev_end) |
| 269 | max -= prev_end; |
| 270 | else |
| 271 | max = 0; |
| 272 | } |
Michel Lespinasse | d373718 | 2012-12-11 16:01:38 -0800 | [diff] [blame] | 273 | if (vma->vm_rb.rb_left) { |
| 274 | subtree_gap = rb_entry(vma->vm_rb.rb_left, |
| 275 | struct vm_area_struct, vm_rb)->rb_subtree_gap; |
| 276 | if (subtree_gap > max) |
| 277 | max = subtree_gap; |
| 278 | } |
| 279 | if (vma->vm_rb.rb_right) { |
| 280 | subtree_gap = rb_entry(vma->vm_rb.rb_right, |
| 281 | struct vm_area_struct, vm_rb)->rb_subtree_gap; |
| 282 | if (subtree_gap > max) |
| 283 | max = subtree_gap; |
| 284 | } |
| 285 | return max; |
| 286 | } |
| 287 | |
Michel Lespinasse | ed8ea81 | 2012-10-08 16:31:45 -0700 | [diff] [blame] | 288 | #ifdef CONFIG_DEBUG_VM_RB |
Andrea Arcangeli | acf128d | 2016-02-05 15:36:13 -0800 | [diff] [blame] | 289 | static int browse_rb(struct mm_struct *mm) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 290 | { |
Andrea Arcangeli | acf128d | 2016-02-05 15:36:13 -0800 | [diff] [blame] | 291 | struct rb_root *root = &mm->mm_rb; |
Michel Lespinasse | 5a0768f | 2012-12-11 16:01:42 -0800 | [diff] [blame] | 292 | int i = 0, j, bug = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 293 | struct rb_node *nd, *pn = NULL; |
| 294 | unsigned long prev = 0, pend = 0; |
| 295 | |
| 296 | for (nd = rb_first(root); nd; nd = rb_next(nd)) { |
| 297 | struct vm_area_struct *vma; |
| 298 | vma = rb_entry(nd, struct vm_area_struct, vm_rb); |
Michel Lespinasse | 5a0768f | 2012-12-11 16:01:42 -0800 | [diff] [blame] | 299 | if (vma->vm_start < prev) { |
Andrew Morton | ff26f70 | 2014-10-09 15:28:19 -0700 | [diff] [blame] | 300 | pr_emerg("vm_start %lx < prev %lx\n", |
| 301 | vma->vm_start, prev); |
Michel Lespinasse | 5a0768f | 2012-12-11 16:01:42 -0800 | [diff] [blame] | 302 | bug = 1; |
| 303 | } |
| 304 | if (vma->vm_start < pend) { |
Andrew Morton | ff26f70 | 2014-10-09 15:28:19 -0700 | [diff] [blame] | 305 | pr_emerg("vm_start %lx < pend %lx\n", |
| 306 | vma->vm_start, pend); |
Michel Lespinasse | 5a0768f | 2012-12-11 16:01:42 -0800 | [diff] [blame] | 307 | bug = 1; |
| 308 | } |
| 309 | if (vma->vm_start > vma->vm_end) { |
Andrew Morton | ff26f70 | 2014-10-09 15:28:19 -0700 | [diff] [blame] | 310 | pr_emerg("vm_start %lx > vm_end %lx\n", |
| 311 | vma->vm_start, vma->vm_end); |
Michel Lespinasse | 5a0768f | 2012-12-11 16:01:42 -0800 | [diff] [blame] | 312 | bug = 1; |
| 313 | } |
Andrea Arcangeli | acf128d | 2016-02-05 15:36:13 -0800 | [diff] [blame] | 314 | spin_lock(&mm->page_table_lock); |
Michel Lespinasse | 5a0768f | 2012-12-11 16:01:42 -0800 | [diff] [blame] | 315 | if (vma->rb_subtree_gap != vma_compute_subtree_gap(vma)) { |
Sasha Levin | 8542bdf | 2014-09-09 14:50:59 -0700 | [diff] [blame] | 316 | pr_emerg("free gap %lx, correct %lx\n", |
Michel Lespinasse | 5a0768f | 2012-12-11 16:01:42 -0800 | [diff] [blame] | 317 | vma->rb_subtree_gap, |
| 318 | vma_compute_subtree_gap(vma)); |
| 319 | bug = 1; |
| 320 | } |
Andrea Arcangeli | acf128d | 2016-02-05 15:36:13 -0800 | [diff] [blame] | 321 | spin_unlock(&mm->page_table_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 322 | i++; |
| 323 | pn = nd; |
David Miller | d1af65d | 2007-02-28 20:13:13 -0800 | [diff] [blame] | 324 | prev = vma->vm_start; |
| 325 | pend = vma->vm_end; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 326 | } |
| 327 | j = 0; |
Michel Lespinasse | 5a0768f | 2012-12-11 16:01:42 -0800 | [diff] [blame] | 328 | for (nd = pn; nd; nd = rb_prev(nd)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 329 | j++; |
Michel Lespinasse | 5a0768f | 2012-12-11 16:01:42 -0800 | [diff] [blame] | 330 | if (i != j) { |
Sasha Levin | 8542bdf | 2014-09-09 14:50:59 -0700 | [diff] [blame] | 331 | pr_emerg("backwards %d, forwards %d\n", j, i); |
Michel Lespinasse | 5a0768f | 2012-12-11 16:01:42 -0800 | [diff] [blame] | 332 | bug = 1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 333 | } |
Michel Lespinasse | 5a0768f | 2012-12-11 16:01:42 -0800 | [diff] [blame] | 334 | return bug ? -1 : i; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 335 | } |
| 336 | |
Michel Lespinasse | d373718 | 2012-12-11 16:01:38 -0800 | [diff] [blame] | 337 | static void validate_mm_rb(struct rb_root *root, struct vm_area_struct *ignore) |
| 338 | { |
| 339 | struct rb_node *nd; |
| 340 | |
| 341 | for (nd = rb_first(root); nd; nd = rb_next(nd)) { |
| 342 | struct vm_area_struct *vma; |
| 343 | vma = rb_entry(nd, struct vm_area_struct, vm_rb); |
Sasha Levin | 96dad67 | 2014-10-09 15:28:39 -0700 | [diff] [blame] | 344 | VM_BUG_ON_VMA(vma != ignore && |
| 345 | vma->rb_subtree_gap != vma_compute_subtree_gap(vma), |
| 346 | vma); |
Michel Lespinasse | d373718 | 2012-12-11 16:01:38 -0800 | [diff] [blame] | 347 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 348 | } |
| 349 | |
Rashika Kheria | eafd4dc | 2014-04-03 14:48:03 -0700 | [diff] [blame] | 350 | static void validate_mm(struct mm_struct *mm) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 351 | { |
| 352 | int bug = 0; |
| 353 | int i = 0; |
Michel Lespinasse | 5a0768f | 2012-12-11 16:01:42 -0800 | [diff] [blame] | 354 | unsigned long highest_address = 0; |
Michel Lespinasse | ed8ea81 | 2012-10-08 16:31:45 -0700 | [diff] [blame] | 355 | struct vm_area_struct *vma = mm->mmap; |
Andrew Morton | ff26f70 | 2014-10-09 15:28:19 -0700 | [diff] [blame] | 356 | |
Michel Lespinasse | ed8ea81 | 2012-10-08 16:31:45 -0700 | [diff] [blame] | 357 | while (vma) { |
Konstantin Khlebnikov | 12352d3 | 2016-02-05 15:36:50 -0800 | [diff] [blame] | 358 | struct anon_vma *anon_vma = vma->anon_vma; |
Michel Lespinasse | ed8ea81 | 2012-10-08 16:31:45 -0700 | [diff] [blame] | 359 | struct anon_vma_chain *avc; |
Andrew Morton | ff26f70 | 2014-10-09 15:28:19 -0700 | [diff] [blame] | 360 | |
Konstantin Khlebnikov | 12352d3 | 2016-02-05 15:36:50 -0800 | [diff] [blame] | 361 | if (anon_vma) { |
| 362 | anon_vma_lock_read(anon_vma); |
| 363 | list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) |
| 364 | anon_vma_interval_tree_verify(avc); |
| 365 | anon_vma_unlock_read(anon_vma); |
| 366 | } |
| 367 | |
Hugh Dickins | cfc0eb40 | 2017-06-19 04:03:24 -0700 | [diff] [blame] | 368 | highest_address = vm_end_gap(vma); |
Michel Lespinasse | ed8ea81 | 2012-10-08 16:31:45 -0700 | [diff] [blame] | 369 | vma = vma->vm_next; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 370 | i++; |
| 371 | } |
Michel Lespinasse | 5a0768f | 2012-12-11 16:01:42 -0800 | [diff] [blame] | 372 | if (i != mm->map_count) { |
Sasha Levin | 8542bdf | 2014-09-09 14:50:59 -0700 | [diff] [blame] | 373 | pr_emerg("map_count %d vm_next %d\n", mm->map_count, i); |
Michel Lespinasse | 5a0768f | 2012-12-11 16:01:42 -0800 | [diff] [blame] | 374 | bug = 1; |
| 375 | } |
| 376 | if (highest_address != mm->highest_vm_end) { |
Sasha Levin | 8542bdf | 2014-09-09 14:50:59 -0700 | [diff] [blame] | 377 | pr_emerg("mm->highest_vm_end %lx, found %lx\n", |
Andrew Morton | ff26f70 | 2014-10-09 15:28:19 -0700 | [diff] [blame] | 378 | mm->highest_vm_end, highest_address); |
Michel Lespinasse | 5a0768f | 2012-12-11 16:01:42 -0800 | [diff] [blame] | 379 | bug = 1; |
| 380 | } |
Andrea Arcangeli | acf128d | 2016-02-05 15:36:13 -0800 | [diff] [blame] | 381 | i = browse_rb(mm); |
Michel Lespinasse | 5a0768f | 2012-12-11 16:01:42 -0800 | [diff] [blame] | 382 | if (i != mm->map_count) { |
Andrew Morton | ff26f70 | 2014-10-09 15:28:19 -0700 | [diff] [blame] | 383 | if (i != -1) |
| 384 | pr_emerg("map_count %d rb %d\n", mm->map_count, i); |
Michel Lespinasse | 5a0768f | 2012-12-11 16:01:42 -0800 | [diff] [blame] | 385 | bug = 1; |
| 386 | } |
Sasha Levin | 96dad67 | 2014-10-09 15:28:39 -0700 | [diff] [blame] | 387 | VM_BUG_ON_MM(bug, mm); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 388 | } |
| 389 | #else |
Michel Lespinasse | d373718 | 2012-12-11 16:01:38 -0800 | [diff] [blame] | 390 | #define validate_mm_rb(root, ignore) do { } while (0) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 391 | #define validate_mm(mm) do { } while (0) |
| 392 | #endif |
| 393 | |
Michel Lespinasse | d373718 | 2012-12-11 16:01:38 -0800 | [diff] [blame] | 394 | RB_DECLARE_CALLBACKS(static, vma_gap_callbacks, struct vm_area_struct, vm_rb, |
| 395 | unsigned long, rb_subtree_gap, vma_compute_subtree_gap) |
| 396 | |
| 397 | /* |
| 398 | * Update augmented rbtree rb_subtree_gap values after vma->vm_start or |
| 399 | * vma->vm_prev->vm_end values changed, without modifying the vma's position |
| 400 | * in the rbtree. |
| 401 | */ |
| 402 | static void vma_gap_update(struct vm_area_struct *vma) |
| 403 | { |
| 404 | /* |
| 405 | * As it turns out, RB_DECLARE_CALLBACKS() already created a callback |
| 406 | * function that does exacltly what we want. |
| 407 | */ |
| 408 | vma_gap_callbacks_propagate(&vma->vm_rb, NULL); |
| 409 | } |
| 410 | |
| 411 | static inline void vma_rb_insert(struct vm_area_struct *vma, |
| 412 | struct rb_root *root) |
| 413 | { |
| 414 | /* All rb_subtree_gap values must be consistent prior to insertion */ |
| 415 | validate_mm_rb(root, NULL); |
| 416 | |
| 417 | rb_insert_augmented(&vma->vm_rb, root, &vma_gap_callbacks); |
| 418 | } |
| 419 | |
Andrea Arcangeli | 8f26e0b | 2016-10-07 17:01:37 -0700 | [diff] [blame] | 420 | static void __vma_rb_erase(struct vm_area_struct *vma, struct rb_root *root) |
| 421 | { |
| 422 | /* |
| 423 | * Note rb_erase_augmented is a fairly large inline function, |
| 424 | * so make sure we instantiate it only once with our desired |
| 425 | * augmented rbtree callbacks. |
| 426 | */ |
| 427 | rb_erase_augmented(&vma->vm_rb, root, &vma_gap_callbacks); |
| 428 | } |
| 429 | |
| 430 | static __always_inline void vma_rb_erase_ignore(struct vm_area_struct *vma, |
| 431 | struct rb_root *root, |
| 432 | struct vm_area_struct *ignore) |
| 433 | { |
| 434 | /* |
| 435 | * All rb_subtree_gap values must be consistent prior to erase, |
| 436 | * with the possible exception of the "next" vma being erased if |
| 437 | * next->vm_start was reduced. |
| 438 | */ |
| 439 | validate_mm_rb(root, ignore); |
| 440 | |
| 441 | __vma_rb_erase(vma, root); |
| 442 | } |
| 443 | |
| 444 | static __always_inline void vma_rb_erase(struct vm_area_struct *vma, |
| 445 | struct rb_root *root) |
Michel Lespinasse | d373718 | 2012-12-11 16:01:38 -0800 | [diff] [blame] | 446 | { |
| 447 | /* |
| 448 | * All rb_subtree_gap values must be consistent prior to erase, |
| 449 | * with the possible exception of the vma being erased. |
| 450 | */ |
| 451 | validate_mm_rb(root, vma); |
| 452 | |
Andrea Arcangeli | 8f26e0b | 2016-10-07 17:01:37 -0700 | [diff] [blame] | 453 | __vma_rb_erase(vma, root); |
Michel Lespinasse | d373718 | 2012-12-11 16:01:38 -0800 | [diff] [blame] | 454 | } |
| 455 | |
Michel Lespinasse | bf181b9 | 2012-10-08 16:31:39 -0700 | [diff] [blame] | 456 | /* |
| 457 | * vma has some anon_vma assigned, and is already inserted on that |
| 458 | * anon_vma's interval trees. |
| 459 | * |
| 460 | * Before updating the vma's vm_start / vm_end / vm_pgoff fields, the |
| 461 | * vma must be removed from the anon_vma's interval trees using |
| 462 | * anon_vma_interval_tree_pre_update_vma(). |
| 463 | * |
| 464 | * After the update, the vma will be reinserted using |
| 465 | * anon_vma_interval_tree_post_update_vma(). |
| 466 | * |
| 467 | * The entire update must be protected by exclusive mmap_sem and by |
| 468 | * the root anon_vma's mutex. |
| 469 | */ |
| 470 | static inline void |
| 471 | anon_vma_interval_tree_pre_update_vma(struct vm_area_struct *vma) |
| 472 | { |
| 473 | struct anon_vma_chain *avc; |
| 474 | |
| 475 | list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) |
| 476 | anon_vma_interval_tree_remove(avc, &avc->anon_vma->rb_root); |
| 477 | } |
| 478 | |
| 479 | static inline void |
| 480 | anon_vma_interval_tree_post_update_vma(struct vm_area_struct *vma) |
| 481 | { |
| 482 | struct anon_vma_chain *avc; |
| 483 | |
| 484 | list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) |
| 485 | anon_vma_interval_tree_insert(avc, &avc->anon_vma->rb_root); |
| 486 | } |
| 487 | |
Hugh Dickins | 6597d78 | 2012-10-08 16:29:07 -0700 | [diff] [blame] | 488 | static int find_vma_links(struct mm_struct *mm, unsigned long addr, |
| 489 | unsigned long end, struct vm_area_struct **pprev, |
| 490 | struct rb_node ***rb_link, struct rb_node **rb_parent) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 491 | { |
Hugh Dickins | 6597d78 | 2012-10-08 16:29:07 -0700 | [diff] [blame] | 492 | struct rb_node **__rb_link, *__rb_parent, *rb_prev; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 493 | |
| 494 | __rb_link = &mm->mm_rb.rb_node; |
| 495 | rb_prev = __rb_parent = NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 496 | |
| 497 | while (*__rb_link) { |
| 498 | struct vm_area_struct *vma_tmp; |
| 499 | |
| 500 | __rb_parent = *__rb_link; |
| 501 | vma_tmp = rb_entry(__rb_parent, struct vm_area_struct, vm_rb); |
| 502 | |
| 503 | if (vma_tmp->vm_end > addr) { |
Hugh Dickins | 6597d78 | 2012-10-08 16:29:07 -0700 | [diff] [blame] | 504 | /* Fail if an existing vma overlaps the area */ |
| 505 | if (vma_tmp->vm_start < end) |
| 506 | return -ENOMEM; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 507 | __rb_link = &__rb_parent->rb_left; |
| 508 | } else { |
| 509 | rb_prev = __rb_parent; |
| 510 | __rb_link = &__rb_parent->rb_right; |
| 511 | } |
| 512 | } |
| 513 | |
| 514 | *pprev = NULL; |
| 515 | if (rb_prev) |
| 516 | *pprev = rb_entry(rb_prev, struct vm_area_struct, vm_rb); |
| 517 | *rb_link = __rb_link; |
| 518 | *rb_parent = __rb_parent; |
Hugh Dickins | 6597d78 | 2012-10-08 16:29:07 -0700 | [diff] [blame] | 519 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 520 | } |
| 521 | |
Cyril Hrubis | e8420a8 | 2013-04-29 15:08:33 -0700 | [diff] [blame] | 522 | static unsigned long count_vma_pages_range(struct mm_struct *mm, |
| 523 | unsigned long addr, unsigned long end) |
| 524 | { |
| 525 | unsigned long nr_pages = 0; |
| 526 | struct vm_area_struct *vma; |
| 527 | |
| 528 | /* Find first overlaping mapping */ |
| 529 | vma = find_vma_intersection(mm, addr, end); |
| 530 | if (!vma) |
| 531 | return 0; |
| 532 | |
| 533 | nr_pages = (min(end, vma->vm_end) - |
| 534 | max(addr, vma->vm_start)) >> PAGE_SHIFT; |
| 535 | |
| 536 | /* Iterate over the rest of the overlaps */ |
| 537 | for (vma = vma->vm_next; vma; vma = vma->vm_next) { |
| 538 | unsigned long overlap_len; |
| 539 | |
| 540 | if (vma->vm_start > end) |
| 541 | break; |
| 542 | |
| 543 | overlap_len = min(end, vma->vm_end) - vma->vm_start; |
| 544 | nr_pages += overlap_len >> PAGE_SHIFT; |
| 545 | } |
| 546 | |
| 547 | return nr_pages; |
| 548 | } |
| 549 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 550 | void __vma_link_rb(struct mm_struct *mm, struct vm_area_struct *vma, |
| 551 | struct rb_node **rb_link, struct rb_node *rb_parent) |
| 552 | { |
Michel Lespinasse | d373718 | 2012-12-11 16:01:38 -0800 | [diff] [blame] | 553 | /* Update tracking information for the gap following the new vma. */ |
| 554 | if (vma->vm_next) |
| 555 | vma_gap_update(vma->vm_next); |
| 556 | else |
Hugh Dickins | cfc0eb40 | 2017-06-19 04:03:24 -0700 | [diff] [blame] | 557 | mm->highest_vm_end = vm_end_gap(vma); |
Michel Lespinasse | d373718 | 2012-12-11 16:01:38 -0800 | [diff] [blame] | 558 | |
| 559 | /* |
| 560 | * vma->vm_prev wasn't known when we followed the rbtree to find the |
| 561 | * correct insertion point for that vma. As a result, we could not |
| 562 | * update the vma vm_rb parents rb_subtree_gap values on the way down. |
| 563 | * So, we first insert the vma with a zero rb_subtree_gap value |
| 564 | * (to be consistent with what we did on the way down), and then |
| 565 | * immediately update the gap to the correct value. Finally we |
| 566 | * rebalance the rbtree after all augmented values have been set. |
| 567 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 568 | rb_link_node(&vma->vm_rb, rb_parent, rb_link); |
Michel Lespinasse | d373718 | 2012-12-11 16:01:38 -0800 | [diff] [blame] | 569 | vma->rb_subtree_gap = 0; |
| 570 | vma_gap_update(vma); |
| 571 | vma_rb_insert(vma, &mm->mm_rb); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 572 | } |
| 573 | |
Denys Vlasenko | cb8f488 | 2008-10-18 20:27:01 -0700 | [diff] [blame] | 574 | static void __vma_link_file(struct vm_area_struct *vma) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 575 | { |
ZhenwenXu | 48aae42 | 2009-01-06 14:40:21 -0800 | [diff] [blame] | 576 | struct file *file; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 577 | |
| 578 | file = vma->vm_file; |
| 579 | if (file) { |
| 580 | struct address_space *mapping = file->f_mapping; |
| 581 | |
| 582 | if (vma->vm_flags & VM_DENYWRITE) |
Al Viro | 496ad9a | 2013-01-23 17:07:38 -0500 | [diff] [blame] | 583 | atomic_dec(&file_inode(file)->i_writecount); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 584 | if (vma->vm_flags & VM_SHARED) |
David Herrmann | 4bb5f5d | 2014-08-08 14:25:25 -0700 | [diff] [blame] | 585 | atomic_inc(&mapping->i_mmap_writable); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 586 | |
| 587 | flush_dcache_mmap_lock(mapping); |
Kirill A. Shutemov | 27ba064 | 2015-02-10 14:09:59 -0800 | [diff] [blame] | 588 | vma_interval_tree_insert(vma, &mapping->i_mmap); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 589 | flush_dcache_mmap_unlock(mapping); |
| 590 | } |
| 591 | } |
| 592 | |
| 593 | static void |
| 594 | __vma_link(struct mm_struct *mm, struct vm_area_struct *vma, |
| 595 | struct vm_area_struct *prev, struct rb_node **rb_link, |
| 596 | struct rb_node *rb_parent) |
| 597 | { |
| 598 | __vma_link_list(mm, vma, prev, rb_parent); |
| 599 | __vma_link_rb(mm, vma, rb_link, rb_parent); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 600 | } |
| 601 | |
| 602 | static void vma_link(struct mm_struct *mm, struct vm_area_struct *vma, |
| 603 | struct vm_area_struct *prev, struct rb_node **rb_link, |
| 604 | struct rb_node *rb_parent) |
| 605 | { |
| 606 | struct address_space *mapping = NULL; |
| 607 | |
Huang Shijie | 64ac494 | 2014-06-04 16:07:33 -0700 | [diff] [blame] | 608 | if (vma->vm_file) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 609 | mapping = vma->vm_file->f_mapping; |
Davidlohr Bueso | 83cde9e | 2014-12-12 16:54:21 -0800 | [diff] [blame] | 610 | i_mmap_lock_write(mapping); |
Huang Shijie | 64ac494 | 2014-06-04 16:07:33 -0700 | [diff] [blame] | 611 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 612 | |
| 613 | __vma_link(mm, vma, prev, rb_link, rb_parent); |
| 614 | __vma_link_file(vma); |
| 615 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 616 | if (mapping) |
Davidlohr Bueso | 83cde9e | 2014-12-12 16:54:21 -0800 | [diff] [blame] | 617 | i_mmap_unlock_write(mapping); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 618 | |
| 619 | mm->map_count++; |
| 620 | validate_mm(mm); |
| 621 | } |
| 622 | |
| 623 | /* |
Kautuk Consul | 88f6b4c | 2012-03-21 16:34:16 -0700 | [diff] [blame] | 624 | * Helper for vma_adjust() in the split_vma insert case: insert a vma into the |
Michel Lespinasse | 6b2dbba | 2012-10-08 16:31:25 -0700 | [diff] [blame] | 625 | * mm's list and rbtree. It has already been inserted into the interval tree. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 626 | */ |
ZhenwenXu | 48aae42 | 2009-01-06 14:40:21 -0800 | [diff] [blame] | 627 | static void __insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 628 | { |
Hugh Dickins | 6597d78 | 2012-10-08 16:29:07 -0700 | [diff] [blame] | 629 | struct vm_area_struct *prev; |
ZhenwenXu | 48aae42 | 2009-01-06 14:40:21 -0800 | [diff] [blame] | 630 | struct rb_node **rb_link, *rb_parent; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 631 | |
Hugh Dickins | 6597d78 | 2012-10-08 16:29:07 -0700 | [diff] [blame] | 632 | if (find_vma_links(mm, vma->vm_start, vma->vm_end, |
| 633 | &prev, &rb_link, &rb_parent)) |
| 634 | BUG(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 635 | __vma_link(mm, vma, prev, rb_link, rb_parent); |
| 636 | mm->map_count++; |
| 637 | } |
| 638 | |
Andrea Arcangeli | e86f15e | 2016-10-07 17:01:28 -0700 | [diff] [blame] | 639 | static __always_inline void __vma_unlink_common(struct mm_struct *mm, |
| 640 | struct vm_area_struct *vma, |
| 641 | struct vm_area_struct *prev, |
Andrea Arcangeli | 8f26e0b | 2016-10-07 17:01:37 -0700 | [diff] [blame] | 642 | bool has_prev, |
| 643 | struct vm_area_struct *ignore) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 644 | { |
Michel Lespinasse | d373718 | 2012-12-11 16:01:38 -0800 | [diff] [blame] | 645 | struct vm_area_struct *next; |
Linus Torvalds | 297c5ee | 2010-08-20 16:24:55 -0700 | [diff] [blame] | 646 | |
Andrea Arcangeli | 8f26e0b | 2016-10-07 17:01:37 -0700 | [diff] [blame] | 647 | vma_rb_erase_ignore(vma, &mm->mm_rb, ignore); |
Andrea Arcangeli | e86f15e | 2016-10-07 17:01:28 -0700 | [diff] [blame] | 648 | next = vma->vm_next; |
| 649 | if (has_prev) |
| 650 | prev->vm_next = next; |
| 651 | else { |
| 652 | prev = vma->vm_prev; |
| 653 | if (prev) |
| 654 | prev->vm_next = next; |
| 655 | else |
| 656 | mm->mmap = next; |
| 657 | } |
Linus Torvalds | 297c5ee | 2010-08-20 16:24:55 -0700 | [diff] [blame] | 658 | if (next) |
| 659 | next->vm_prev = prev; |
Davidlohr Bueso | 615d6e8 | 2014-04-07 15:37:25 -0700 | [diff] [blame] | 660 | |
| 661 | /* Kill the cache */ |
| 662 | vmacache_invalidate(mm); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 663 | } |
| 664 | |
Andrea Arcangeli | e86f15e | 2016-10-07 17:01:28 -0700 | [diff] [blame] | 665 | static inline void __vma_unlink_prev(struct mm_struct *mm, |
| 666 | struct vm_area_struct *vma, |
| 667 | struct vm_area_struct *prev) |
| 668 | { |
Andrea Arcangeli | 8f26e0b | 2016-10-07 17:01:37 -0700 | [diff] [blame] | 669 | __vma_unlink_common(mm, vma, prev, true, vma); |
Andrea Arcangeli | e86f15e | 2016-10-07 17:01:28 -0700 | [diff] [blame] | 670 | } |
| 671 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 672 | /* |
| 673 | * We cannot adjust vm_start, vm_end, vm_pgoff fields of a vma that |
| 674 | * is already present in an i_mmap tree without adjusting the tree. |
| 675 | * The following helper function should be used when such adjustments |
| 676 | * are necessary. The "insert" vma (if any) is to be inserted |
| 677 | * before we drop the necessary locks. |
| 678 | */ |
Andrea Arcangeli | e86f15e | 2016-10-07 17:01:28 -0700 | [diff] [blame] | 679 | int __vma_adjust(struct vm_area_struct *vma, unsigned long start, |
| 680 | unsigned long end, pgoff_t pgoff, struct vm_area_struct *insert, |
| 681 | struct vm_area_struct *expand) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 682 | { |
| 683 | struct mm_struct *mm = vma->vm_mm; |
Andrea Arcangeli | e86f15e | 2016-10-07 17:01:28 -0700 | [diff] [blame] | 684 | struct vm_area_struct *next = vma->vm_next, *orig_vma = vma; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 685 | struct address_space *mapping = NULL; |
Michel Lespinasse | 6b2dbba | 2012-10-08 16:31:25 -0700 | [diff] [blame] | 686 | struct rb_root *root = NULL; |
Rik van Riel | 012f1800 | 2010-08-09 17:18:40 -0700 | [diff] [blame] | 687 | struct anon_vma *anon_vma = NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 688 | struct file *file = vma->vm_file; |
Michel Lespinasse | d373718 | 2012-12-11 16:01:38 -0800 | [diff] [blame] | 689 | bool start_changed = false, end_changed = false; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 690 | long adjust_next = 0; |
| 691 | int remove_next = 0; |
| 692 | |
| 693 | if (next && !insert) { |
Kirill A. Shutemov | 734537c | 2016-07-28 15:49:01 -0700 | [diff] [blame] | 694 | struct vm_area_struct *exporter = NULL, *importer = NULL; |
Linus Torvalds | 287d97a | 2010-04-10 15:22:30 -0700 | [diff] [blame] | 695 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 696 | if (end >= next->vm_end) { |
| 697 | /* |
| 698 | * vma expands, overlapping all the next, and |
| 699 | * perhaps the one after too (mprotect case 6). |
Andrea Arcangeli | 86d12e4 | 2016-10-07 17:01:34 -0700 | [diff] [blame] | 700 | * The only other cases that gets here are |
Andrea Arcangeli | e86f15e | 2016-10-07 17:01:28 -0700 | [diff] [blame] | 701 | * case 1, case 7 and case 8. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 702 | */ |
Andrea Arcangeli | e86f15e | 2016-10-07 17:01:28 -0700 | [diff] [blame] | 703 | if (next == expand) { |
| 704 | /* |
| 705 | * The only case where we don't expand "vma" |
| 706 | * and we expand "next" instead is case 8. |
| 707 | */ |
| 708 | VM_WARN_ON(end != next->vm_end); |
| 709 | /* |
| 710 | * remove_next == 3 means we're |
| 711 | * removing "vma" and that to do so we |
| 712 | * swapped "vma" and "next". |
| 713 | */ |
| 714 | remove_next = 3; |
| 715 | VM_WARN_ON(file != next->vm_file); |
| 716 | swap(vma, next); |
| 717 | } else { |
| 718 | VM_WARN_ON(expand != vma); |
| 719 | /* |
| 720 | * case 1, 6, 7, remove_next == 2 is case 6, |
| 721 | * remove_next == 1 is case 1 or 7. |
| 722 | */ |
| 723 | remove_next = 1 + (end > next->vm_end); |
| 724 | VM_WARN_ON(remove_next == 2 && |
| 725 | end != next->vm_next->vm_end); |
| 726 | VM_WARN_ON(remove_next == 1 && |
| 727 | end != next->vm_end); |
| 728 | /* trim end to next, for case 6 first pass */ |
| 729 | end = next->vm_end; |
| 730 | } |
| 731 | |
Linus Torvalds | 287d97a | 2010-04-10 15:22:30 -0700 | [diff] [blame] | 732 | exporter = next; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 733 | importer = vma; |
Kirill A. Shutemov | 734537c | 2016-07-28 15:49:01 -0700 | [diff] [blame] | 734 | |
| 735 | /* |
| 736 | * If next doesn't have anon_vma, import from vma after |
| 737 | * next, if the vma overlaps with it. |
| 738 | */ |
Andrea Arcangeli | 97a42cd | 2016-10-07 17:01:31 -0700 | [diff] [blame] | 739 | if (remove_next == 2 && !next->anon_vma) |
Kirill A. Shutemov | 734537c | 2016-07-28 15:49:01 -0700 | [diff] [blame] | 740 | exporter = next->vm_next; |
| 741 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 742 | } else if (end > next->vm_start) { |
| 743 | /* |
| 744 | * vma expands, overlapping part of the next: |
| 745 | * mprotect case 5 shifting the boundary up. |
| 746 | */ |
| 747 | adjust_next = (end - next->vm_start) >> PAGE_SHIFT; |
Linus Torvalds | 287d97a | 2010-04-10 15:22:30 -0700 | [diff] [blame] | 748 | exporter = next; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 749 | importer = vma; |
Andrea Arcangeli | e86f15e | 2016-10-07 17:01:28 -0700 | [diff] [blame] | 750 | VM_WARN_ON(expand != importer); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 751 | } else if (end < vma->vm_end) { |
| 752 | /* |
| 753 | * vma shrinks, and !insert tells it's not |
| 754 | * split_vma inserting another: so it must be |
| 755 | * mprotect case 4 shifting the boundary down. |
| 756 | */ |
vishnu.ps | cc71aba | 2014-10-09 15:26:29 -0700 | [diff] [blame] | 757 | adjust_next = -((vma->vm_end - end) >> PAGE_SHIFT); |
Linus Torvalds | 287d97a | 2010-04-10 15:22:30 -0700 | [diff] [blame] | 758 | exporter = vma; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 759 | importer = next; |
Andrea Arcangeli | e86f15e | 2016-10-07 17:01:28 -0700 | [diff] [blame] | 760 | VM_WARN_ON(expand != importer); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 761 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 762 | |
Rik van Riel | 5beb493 | 2010-03-05 13:42:07 -0800 | [diff] [blame] | 763 | /* |
| 764 | * Easily overlooked: when mprotect shifts the boundary, |
| 765 | * make sure the expanding vma has anon_vma set if the |
| 766 | * shrinking vma had, to cover any anon pages imported. |
| 767 | */ |
Linus Torvalds | 287d97a | 2010-04-10 15:22:30 -0700 | [diff] [blame] | 768 | if (exporter && exporter->anon_vma && !importer->anon_vma) { |
Daniel Forrest | c4ea95d | 2014-12-02 15:59:42 -0800 | [diff] [blame] | 769 | int error; |
| 770 | |
Linus Torvalds | 287d97a | 2010-04-10 15:22:30 -0700 | [diff] [blame] | 771 | importer->anon_vma = exporter->anon_vma; |
Konstantin Khlebnikov | b800c91 | 2015-01-11 16:54:06 +0300 | [diff] [blame] | 772 | error = anon_vma_clone(importer, exporter); |
Leon Yu | 3fe89b3 | 2015-03-25 15:55:11 -0700 | [diff] [blame] | 773 | if (error) |
Konstantin Khlebnikov | b800c91 | 2015-01-11 16:54:06 +0300 | [diff] [blame] | 774 | return error; |
Rik van Riel | 5beb493 | 2010-03-05 13:42:07 -0800 | [diff] [blame] | 775 | } |
| 776 | } |
Kirill A. Shutemov | 734537c | 2016-07-28 15:49:01 -0700 | [diff] [blame] | 777 | again: |
Andrea Arcangeli | e86f15e | 2016-10-07 17:01:28 -0700 | [diff] [blame] | 778 | vma_adjust_trans_huge(orig_vma, start, end, adjust_next); |
Kirill A. Shutemov | 37f9f55 | 2016-07-26 15:25:48 -0700 | [diff] [blame] | 779 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 780 | if (file) { |
| 781 | mapping = file->f_mapping; |
Kirill A. Shutemov | 27ba064 | 2015-02-10 14:09:59 -0800 | [diff] [blame] | 782 | root = &mapping->i_mmap; |
| 783 | uprobe_munmap(vma, vma->vm_start, vma->vm_end); |
Srikar Dronamraju | 682968e | 2012-03-30 23:56:46 +0530 | [diff] [blame] | 784 | |
Kirill A. Shutemov | 27ba064 | 2015-02-10 14:09:59 -0800 | [diff] [blame] | 785 | if (adjust_next) |
| 786 | uprobe_munmap(next, next->vm_start, next->vm_end); |
Srikar Dronamraju | 682968e | 2012-03-30 23:56:46 +0530 | [diff] [blame] | 787 | |
Davidlohr Bueso | 83cde9e | 2014-12-12 16:54:21 -0800 | [diff] [blame] | 788 | i_mmap_lock_write(mapping); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 789 | if (insert) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 790 | /* |
Michel Lespinasse | 6b2dbba | 2012-10-08 16:31:25 -0700 | [diff] [blame] | 791 | * Put into interval tree now, so instantiated pages |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 792 | * are visible to arm/parisc __flush_dcache_page |
| 793 | * throughout; but we cannot insert into address |
| 794 | * space until vma start or end is updated. |
| 795 | */ |
| 796 | __vma_link_file(insert); |
| 797 | } |
| 798 | } |
| 799 | |
Michel Lespinasse | bf181b9 | 2012-10-08 16:31:39 -0700 | [diff] [blame] | 800 | anon_vma = vma->anon_vma; |
| 801 | if (!anon_vma && adjust_next) |
| 802 | anon_vma = next->anon_vma; |
| 803 | if (anon_vma) { |
Andrea Arcangeli | e86f15e | 2016-10-07 17:01:28 -0700 | [diff] [blame] | 804 | VM_WARN_ON(adjust_next && next->anon_vma && |
| 805 | anon_vma != next->anon_vma); |
Ingo Molnar | 4fc3f1d | 2012-12-02 19:56:50 +0000 | [diff] [blame] | 806 | anon_vma_lock_write(anon_vma); |
Michel Lespinasse | bf181b9 | 2012-10-08 16:31:39 -0700 | [diff] [blame] | 807 | anon_vma_interval_tree_pre_update_vma(vma); |
| 808 | if (adjust_next) |
| 809 | anon_vma_interval_tree_pre_update_vma(next); |
| 810 | } |
Rik van Riel | 012f1800 | 2010-08-09 17:18:40 -0700 | [diff] [blame] | 811 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 812 | if (root) { |
| 813 | flush_dcache_mmap_lock(mapping); |
Michel Lespinasse | 6b2dbba | 2012-10-08 16:31:25 -0700 | [diff] [blame] | 814 | vma_interval_tree_remove(vma, root); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 815 | if (adjust_next) |
Michel Lespinasse | 6b2dbba | 2012-10-08 16:31:25 -0700 | [diff] [blame] | 816 | vma_interval_tree_remove(next, root); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 817 | } |
| 818 | |
Michel Lespinasse | d373718 | 2012-12-11 16:01:38 -0800 | [diff] [blame] | 819 | if (start != vma->vm_start) { |
| 820 | vma->vm_start = start; |
| 821 | start_changed = true; |
| 822 | } |
| 823 | if (end != vma->vm_end) { |
| 824 | vma->vm_end = end; |
| 825 | end_changed = true; |
| 826 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 827 | vma->vm_pgoff = pgoff; |
| 828 | if (adjust_next) { |
| 829 | next->vm_start += adjust_next << PAGE_SHIFT; |
| 830 | next->vm_pgoff += adjust_next; |
| 831 | } |
| 832 | |
| 833 | if (root) { |
| 834 | if (adjust_next) |
Michel Lespinasse | 6b2dbba | 2012-10-08 16:31:25 -0700 | [diff] [blame] | 835 | vma_interval_tree_insert(next, root); |
| 836 | vma_interval_tree_insert(vma, root); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 837 | flush_dcache_mmap_unlock(mapping); |
| 838 | } |
| 839 | |
| 840 | if (remove_next) { |
| 841 | /* |
| 842 | * vma_merge has merged next into vma, and needs |
| 843 | * us to remove next before dropping the locks. |
| 844 | */ |
Andrea Arcangeli | e86f15e | 2016-10-07 17:01:28 -0700 | [diff] [blame] | 845 | if (remove_next != 3) |
| 846 | __vma_unlink_prev(mm, next, vma); |
| 847 | else |
Andrea Arcangeli | 8f26e0b | 2016-10-07 17:01:37 -0700 | [diff] [blame] | 848 | /* |
| 849 | * vma is not before next if they've been |
| 850 | * swapped. |
| 851 | * |
| 852 | * pre-swap() next->vm_start was reduced so |
| 853 | * tell validate_mm_rb to ignore pre-swap() |
| 854 | * "next" (which is stored in post-swap() |
| 855 | * "vma"). |
| 856 | */ |
| 857 | __vma_unlink_common(mm, next, NULL, false, vma); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 858 | if (file) |
| 859 | __remove_shared_vm_struct(next, file, mapping); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 860 | } else if (insert) { |
| 861 | /* |
| 862 | * split_vma has split insert from vma, and needs |
| 863 | * us to insert it before dropping the locks |
| 864 | * (it may either follow vma or precede it). |
| 865 | */ |
| 866 | __insert_vm_struct(mm, insert); |
Michel Lespinasse | d373718 | 2012-12-11 16:01:38 -0800 | [diff] [blame] | 867 | } else { |
| 868 | if (start_changed) |
| 869 | vma_gap_update(vma); |
| 870 | if (end_changed) { |
| 871 | if (!next) |
Hugh Dickins | cfc0eb40 | 2017-06-19 04:03:24 -0700 | [diff] [blame] | 872 | mm->highest_vm_end = vm_end_gap(vma); |
Michel Lespinasse | d373718 | 2012-12-11 16:01:38 -0800 | [diff] [blame] | 873 | else if (!adjust_next) |
| 874 | vma_gap_update(next); |
| 875 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 876 | } |
| 877 | |
Michel Lespinasse | bf181b9 | 2012-10-08 16:31:39 -0700 | [diff] [blame] | 878 | if (anon_vma) { |
| 879 | anon_vma_interval_tree_post_update_vma(vma); |
| 880 | if (adjust_next) |
| 881 | anon_vma_interval_tree_post_update_vma(next); |
Konstantin Khlebnikov | 08b5270 | 2013-02-22 16:34:40 -0800 | [diff] [blame] | 882 | anon_vma_unlock_write(anon_vma); |
Michel Lespinasse | bf181b9 | 2012-10-08 16:31:39 -0700 | [diff] [blame] | 883 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 884 | if (mapping) |
Davidlohr Bueso | 83cde9e | 2014-12-12 16:54:21 -0800 | [diff] [blame] | 885 | i_mmap_unlock_write(mapping); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 886 | |
Srikar Dronamraju | 2b14449 | 2012-02-09 14:56:42 +0530 | [diff] [blame] | 887 | if (root) { |
Ingo Molnar | 7b2d81d | 2012-02-17 09:27:41 +0100 | [diff] [blame] | 888 | uprobe_mmap(vma); |
Srikar Dronamraju | 2b14449 | 2012-02-09 14:56:42 +0530 | [diff] [blame] | 889 | |
| 890 | if (adjust_next) |
Ingo Molnar | 7b2d81d | 2012-02-17 09:27:41 +0100 | [diff] [blame] | 891 | uprobe_mmap(next); |
Srikar Dronamraju | 2b14449 | 2012-02-09 14:56:42 +0530 | [diff] [blame] | 892 | } |
| 893 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 894 | if (remove_next) { |
Matt Helsley | 925d1c4 | 2008-04-29 01:01:36 -0700 | [diff] [blame] | 895 | if (file) { |
Srikar Dronamraju | cbc91f7 | 2012-04-11 16:05:27 +0530 | [diff] [blame] | 896 | uprobe_munmap(next, next->vm_start, next->vm_end); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 897 | fput(file); |
Matt Helsley | 925d1c4 | 2008-04-29 01:01:36 -0700 | [diff] [blame] | 898 | } |
Rik van Riel | 5beb493 | 2010-03-05 13:42:07 -0800 | [diff] [blame] | 899 | if (next->anon_vma) |
| 900 | anon_vma_merge(vma, next); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 901 | mm->map_count--; |
Oleg Nesterov | 3964acd | 2013-07-31 13:53:28 -0700 | [diff] [blame] | 902 | mpol_put(vma_policy(next)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 903 | kmem_cache_free(vm_area_cachep, next); |
| 904 | /* |
| 905 | * In mprotect's case 6 (see comments on vma_merge), |
| 906 | * we must remove another next too. It would clutter |
| 907 | * up the code too much to do both in one go. |
| 908 | */ |
Andrea Arcangeli | e86f15e | 2016-10-07 17:01:28 -0700 | [diff] [blame] | 909 | if (remove_next != 3) { |
| 910 | /* |
| 911 | * If "next" was removed and vma->vm_end was |
| 912 | * expanded (up) over it, in turn |
| 913 | * "next->vm_prev->vm_end" changed and the |
| 914 | * "vma->vm_next" gap must be updated. |
| 915 | */ |
| 916 | next = vma->vm_next; |
| 917 | } else { |
| 918 | /* |
| 919 | * For the scope of the comment "next" and |
| 920 | * "vma" considered pre-swap(): if "vma" was |
| 921 | * removed, next->vm_start was expanded (down) |
| 922 | * over it and the "next" gap must be updated. |
| 923 | * Because of the swap() the post-swap() "vma" |
| 924 | * actually points to pre-swap() "next" |
| 925 | * (post-swap() "next" as opposed is now a |
| 926 | * dangling pointer). |
| 927 | */ |
| 928 | next = vma; |
| 929 | } |
Kirill A. Shutemov | 734537c | 2016-07-28 15:49:01 -0700 | [diff] [blame] | 930 | if (remove_next == 2) { |
| 931 | remove_next = 1; |
| 932 | end = next->vm_end; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 933 | goto again; |
Kirill A. Shutemov | 734537c | 2016-07-28 15:49:01 -0700 | [diff] [blame] | 934 | } |
Michel Lespinasse | d373718 | 2012-12-11 16:01:38 -0800 | [diff] [blame] | 935 | else if (next) |
| 936 | vma_gap_update(next); |
Andrea Arcangeli | fb8c41e | 2016-10-07 17:01:25 -0700 | [diff] [blame] | 937 | else { |
| 938 | /* |
| 939 | * If remove_next == 2 we obviously can't |
| 940 | * reach this path. |
| 941 | * |
| 942 | * If remove_next == 3 we can't reach this |
| 943 | * path because pre-swap() next is always not |
| 944 | * NULL. pre-swap() "next" is not being |
| 945 | * removed and its next->vm_end is not altered |
| 946 | * (and furthermore "end" already matches |
| 947 | * next->vm_end in remove_next == 3). |
| 948 | * |
| 949 | * We reach this only in the remove_next == 1 |
| 950 | * case if the "next" vma that was removed was |
| 951 | * the highest vma of the mm. However in such |
| 952 | * case next->vm_end == "end" and the extended |
| 953 | * "vma" has vma->vm_end == next->vm_end so |
| 954 | * mm->highest_vm_end doesn't need any update |
| 955 | * in remove_next == 1 case. |
| 956 | */ |
Hugh Dickins | cfc0eb40 | 2017-06-19 04:03:24 -0700 | [diff] [blame] | 957 | VM_WARN_ON(mm->highest_vm_end != vm_end_gap(vma)); |
Andrea Arcangeli | fb8c41e | 2016-10-07 17:01:25 -0700 | [diff] [blame] | 958 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 959 | } |
Srikar Dronamraju | 2b14449 | 2012-02-09 14:56:42 +0530 | [diff] [blame] | 960 | if (insert && file) |
Ingo Molnar | 7b2d81d | 2012-02-17 09:27:41 +0100 | [diff] [blame] | 961 | uprobe_mmap(insert); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 962 | |
| 963 | validate_mm(mm); |
Rik van Riel | 5beb493 | 2010-03-05 13:42:07 -0800 | [diff] [blame] | 964 | |
| 965 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 966 | } |
| 967 | |
| 968 | /* |
| 969 | * If the vma has a ->close operation then the driver probably needs to release |
| 970 | * per-vma resources, so we don't attempt to merge those. |
| 971 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 972 | static inline int is_mergeable_vma(struct vm_area_struct *vma, |
Andrea Arcangeli | 19a809a | 2015-09-04 15:46:24 -0700 | [diff] [blame] | 973 | struct file *file, unsigned long vm_flags, |
Colin Cross | 3e4578f | 2015-10-27 16:42:08 -0700 | [diff] [blame] | 974 | struct vm_userfaultfd_ctx vm_userfaultfd_ctx, |
| 975 | const char __user *anon_name) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 976 | { |
Cyrill Gorcunov | 34228d4 | 2014-01-23 15:53:42 -0800 | [diff] [blame] | 977 | /* |
| 978 | * VM_SOFTDIRTY should not prevent from VMA merging, if we |
| 979 | * match the flags but dirty bit -- the caller should mark |
| 980 | * merged VMA as dirty. If dirty bit won't be excluded from |
| 981 | * comparison, we increase pressue on the memory system forcing |
| 982 | * the kernel to generate new VMAs when old one could be |
| 983 | * extended instead. |
| 984 | */ |
| 985 | if ((vma->vm_flags ^ vm_flags) & ~VM_SOFTDIRTY) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 986 | return 0; |
| 987 | if (vma->vm_file != file) |
| 988 | return 0; |
| 989 | if (vma->vm_ops && vma->vm_ops->close) |
| 990 | return 0; |
Andrea Arcangeli | 19a809a | 2015-09-04 15:46:24 -0700 | [diff] [blame] | 991 | if (!is_mergeable_vm_userfaultfd_ctx(vma, vm_userfaultfd_ctx)) |
| 992 | return 0; |
Colin Cross | 3e4578f | 2015-10-27 16:42:08 -0700 | [diff] [blame] | 993 | if (vma_get_anon_name(vma) != anon_name) |
| 994 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 995 | return 1; |
| 996 | } |
| 997 | |
| 998 | static inline int is_mergeable_anon_vma(struct anon_vma *anon_vma1, |
Shaohua Li | 965f55d | 2011-05-24 17:11:20 -0700 | [diff] [blame] | 999 | struct anon_vma *anon_vma2, |
| 1000 | struct vm_area_struct *vma) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1001 | { |
Shaohua Li | 965f55d | 2011-05-24 17:11:20 -0700 | [diff] [blame] | 1002 | /* |
| 1003 | * The list_is_singular() test is to avoid merging VMA cloned from |
| 1004 | * parents. This can improve scalability caused by anon_vma lock. |
| 1005 | */ |
| 1006 | if ((!anon_vma1 || !anon_vma2) && (!vma || |
| 1007 | list_is_singular(&vma->anon_vma_chain))) |
| 1008 | return 1; |
| 1009 | return anon_vma1 == anon_vma2; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1010 | } |
| 1011 | |
| 1012 | /* |
| 1013 | * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff) |
| 1014 | * in front of (at a lower virtual address and file offset than) the vma. |
| 1015 | * |
| 1016 | * We cannot merge two vmas if they have differently assigned (non-NULL) |
| 1017 | * anon_vmas, nor if same anon_vma is assigned but offsets incompatible. |
| 1018 | * |
| 1019 | * We don't check here for the merged mmap wrapping around the end of pagecache |
| 1020 | * indices (16TB on ia32) because do_mmap_pgoff() does not permit mmap's which |
| 1021 | * wrap, nor mmaps which cover the final page at index -1UL. |
| 1022 | */ |
| 1023 | static int |
| 1024 | can_vma_merge_before(struct vm_area_struct *vma, unsigned long vm_flags, |
Andrea Arcangeli | 19a809a | 2015-09-04 15:46:24 -0700 | [diff] [blame] | 1025 | struct anon_vma *anon_vma, struct file *file, |
| 1026 | pgoff_t vm_pgoff, |
Colin Cross | 3e4578f | 2015-10-27 16:42:08 -0700 | [diff] [blame] | 1027 | struct vm_userfaultfd_ctx vm_userfaultfd_ctx, |
| 1028 | const char __user *anon_name) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1029 | { |
Colin Cross | 3e4578f | 2015-10-27 16:42:08 -0700 | [diff] [blame] | 1030 | if (is_mergeable_vma(vma, file, vm_flags, vm_userfaultfd_ctx, anon_name) && |
Shaohua Li | 965f55d | 2011-05-24 17:11:20 -0700 | [diff] [blame] | 1031 | is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1032 | if (vma->vm_pgoff == vm_pgoff) |
| 1033 | return 1; |
| 1034 | } |
| 1035 | return 0; |
| 1036 | } |
| 1037 | |
| 1038 | /* |
| 1039 | * Return true if we can merge this (vm_flags,anon_vma,file,vm_pgoff) |
| 1040 | * beyond (at a higher virtual address and file offset than) the vma. |
| 1041 | * |
| 1042 | * We cannot merge two vmas if they have differently assigned (non-NULL) |
| 1043 | * anon_vmas, nor if same anon_vma is assigned but offsets incompatible. |
| 1044 | */ |
| 1045 | static int |
| 1046 | can_vma_merge_after(struct vm_area_struct *vma, unsigned long vm_flags, |
Andrea Arcangeli | 19a809a | 2015-09-04 15:46:24 -0700 | [diff] [blame] | 1047 | struct anon_vma *anon_vma, struct file *file, |
| 1048 | pgoff_t vm_pgoff, |
Colin Cross | 3e4578f | 2015-10-27 16:42:08 -0700 | [diff] [blame] | 1049 | struct vm_userfaultfd_ctx vm_userfaultfd_ctx, |
| 1050 | const char __user *anon_name) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1051 | { |
Colin Cross | 3e4578f | 2015-10-27 16:42:08 -0700 | [diff] [blame] | 1052 | if (is_mergeable_vma(vma, file, vm_flags, vm_userfaultfd_ctx, anon_name) && |
Shaohua Li | 965f55d | 2011-05-24 17:11:20 -0700 | [diff] [blame] | 1053 | is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1054 | pgoff_t vm_pglen; |
Libin | d6e9321 | 2013-07-03 15:01:26 -0700 | [diff] [blame] | 1055 | vm_pglen = vma_pages(vma); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1056 | if (vma->vm_pgoff + vm_pglen == vm_pgoff) |
| 1057 | return 1; |
| 1058 | } |
| 1059 | return 0; |
| 1060 | } |
| 1061 | |
| 1062 | /* |
Colin Cross | 3e4578f | 2015-10-27 16:42:08 -0700 | [diff] [blame] | 1063 | * Given a mapping request (addr,end,vm_flags,file,pgoff,anon_name), |
| 1064 | * figure out whether that can be merged with its predecessor or its |
| 1065 | * successor. Or both (it neatly fills a hole). |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1066 | * |
| 1067 | * In most cases - when called for mmap, brk or mremap - [addr,end) is |
| 1068 | * certain not to be mapped by the time vma_merge is called; but when |
| 1069 | * called for mprotect, it is certain to be already mapped (either at |
| 1070 | * an offset within prev, or at the start of next), and the flags of |
| 1071 | * this area are about to be changed to vm_flags - and the no-change |
| 1072 | * case has already been eliminated. |
| 1073 | * |
| 1074 | * The following mprotect cases have to be considered, where AAAA is |
| 1075 | * the area passed down from mprotect_fixup, never extending beyond one |
| 1076 | * vma, PPPPPP is the prev vma specified, and NNNNNN the next vma after: |
| 1077 | * |
| 1078 | * AAAA AAAA AAAA AAAA |
| 1079 | * PPPPPPNNNNNN PPPPPPNNNNNN PPPPPPNNNNNN PPPPNNNNXXXX |
| 1080 | * cannot merge might become might become might become |
| 1081 | * PPNNNNNNNNNN PPPPPPPPPPNN PPPPPPPPPPPP 6 or |
| 1082 | * mmap, brk or case 4 below case 5 below PPPPPPPPXXXX 7 or |
Andrea Arcangeli | e86f15e | 2016-10-07 17:01:28 -0700 | [diff] [blame] | 1083 | * mremap move: PPPPXXXXXXXX 8 |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1084 | * AAAA |
| 1085 | * PPPP NNNN PPPPPPPPPPPP PPPPPPPPNNNN PPPPNNNNNNNN |
| 1086 | * might become case 1 below case 2 below case 3 below |
| 1087 | * |
Andrea Arcangeli | e86f15e | 2016-10-07 17:01:28 -0700 | [diff] [blame] | 1088 | * It is important for case 8 that the the vma NNNN overlapping the |
| 1089 | * region AAAA is never going to extended over XXXX. Instead XXXX must |
| 1090 | * be extended in region AAAA and NNNN must be removed. This way in |
| 1091 | * all cases where vma_merge succeeds, the moment vma_adjust drops the |
| 1092 | * rmap_locks, the properties of the merged vma will be already |
| 1093 | * correct for the whole merged range. Some of those properties like |
| 1094 | * vm_page_prot/vm_flags may be accessed by rmap_walks and they must |
| 1095 | * be correct for the whole merged range immediately after the |
| 1096 | * rmap_locks are released. Otherwise if XXXX would be removed and |
| 1097 | * NNNN would be extended over the XXXX range, remove_migration_ptes |
| 1098 | * or other rmap walkers (if working on addresses beyond the "end" |
| 1099 | * parameter) may establish ptes with the wrong permissions of NNNN |
| 1100 | * instead of the right permissions of XXXX. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1101 | */ |
| 1102 | struct vm_area_struct *vma_merge(struct mm_struct *mm, |
| 1103 | struct vm_area_struct *prev, unsigned long addr, |
| 1104 | unsigned long end, unsigned long vm_flags, |
vishnu.ps | cc71aba | 2014-10-09 15:26:29 -0700 | [diff] [blame] | 1105 | struct anon_vma *anon_vma, struct file *file, |
Andrea Arcangeli | 19a809a | 2015-09-04 15:46:24 -0700 | [diff] [blame] | 1106 | pgoff_t pgoff, struct mempolicy *policy, |
Colin Cross | 3e4578f | 2015-10-27 16:42:08 -0700 | [diff] [blame] | 1107 | struct vm_userfaultfd_ctx vm_userfaultfd_ctx, |
| 1108 | const char __user *anon_name) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1109 | { |
| 1110 | pgoff_t pglen = (end - addr) >> PAGE_SHIFT; |
| 1111 | struct vm_area_struct *area, *next; |
Rik van Riel | 5beb493 | 2010-03-05 13:42:07 -0800 | [diff] [blame] | 1112 | int err; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1113 | |
| 1114 | /* |
| 1115 | * We later require that vma->vm_flags == vm_flags, |
| 1116 | * so this tests vma->vm_flags & VM_SPECIAL, too. |
| 1117 | */ |
| 1118 | if (vm_flags & VM_SPECIAL) |
| 1119 | return NULL; |
| 1120 | |
| 1121 | if (prev) |
| 1122 | next = prev->vm_next; |
| 1123 | else |
| 1124 | next = mm->mmap; |
| 1125 | area = next; |
Andrea Arcangeli | e86f15e | 2016-10-07 17:01:28 -0700 | [diff] [blame] | 1126 | if (area && area->vm_end == end) /* cases 6, 7, 8 */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1127 | next = next->vm_next; |
| 1128 | |
Andrea Arcangeli | e86f15e | 2016-10-07 17:01:28 -0700 | [diff] [blame] | 1129 | /* verify some invariant that must be enforced by the caller */ |
| 1130 | VM_WARN_ON(prev && addr <= prev->vm_start); |
| 1131 | VM_WARN_ON(area && end > area->vm_end); |
| 1132 | VM_WARN_ON(addr >= end); |
| 1133 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1134 | /* |
| 1135 | * Can it merge with the predecessor? |
| 1136 | */ |
| 1137 | if (prev && prev->vm_end == addr && |
vishnu.ps | cc71aba | 2014-10-09 15:26:29 -0700 | [diff] [blame] | 1138 | mpol_equal(vma_policy(prev), policy) && |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1139 | can_vma_merge_after(prev, vm_flags, |
Andrea Arcangeli | 19a809a | 2015-09-04 15:46:24 -0700 | [diff] [blame] | 1140 | anon_vma, file, pgoff, |
Colin Cross | 3e4578f | 2015-10-27 16:42:08 -0700 | [diff] [blame] | 1141 | vm_userfaultfd_ctx, |
| 1142 | anon_name)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1143 | /* |
| 1144 | * OK, it can. Can we now merge in the successor as well? |
| 1145 | */ |
| 1146 | if (next && end == next->vm_start && |
| 1147 | mpol_equal(policy, vma_policy(next)) && |
| 1148 | can_vma_merge_before(next, vm_flags, |
Andrea Arcangeli | 19a809a | 2015-09-04 15:46:24 -0700 | [diff] [blame] | 1149 | anon_vma, file, |
| 1150 | pgoff+pglen, |
Colin Cross | 3e4578f | 2015-10-27 16:42:08 -0700 | [diff] [blame] | 1151 | vm_userfaultfd_ctx, |
| 1152 | anon_name) && |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1153 | is_mergeable_anon_vma(prev->anon_vma, |
Shaohua Li | 965f55d | 2011-05-24 17:11:20 -0700 | [diff] [blame] | 1154 | next->anon_vma, NULL)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1155 | /* cases 1, 6 */ |
Andrea Arcangeli | e86f15e | 2016-10-07 17:01:28 -0700 | [diff] [blame] | 1156 | err = __vma_adjust(prev, prev->vm_start, |
| 1157 | next->vm_end, prev->vm_pgoff, NULL, |
| 1158 | prev); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1159 | } else /* cases 2, 5, 7 */ |
Andrea Arcangeli | e86f15e | 2016-10-07 17:01:28 -0700 | [diff] [blame] | 1160 | err = __vma_adjust(prev, prev->vm_start, |
| 1161 | end, prev->vm_pgoff, NULL, prev); |
Rik van Riel | 5beb493 | 2010-03-05 13:42:07 -0800 | [diff] [blame] | 1162 | if (err) |
| 1163 | return NULL; |
David Rientjes | 6d50e60 | 2014-10-29 14:50:31 -0700 | [diff] [blame] | 1164 | khugepaged_enter_vma_merge(prev, vm_flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1165 | return prev; |
| 1166 | } |
| 1167 | |
| 1168 | /* |
| 1169 | * Can this new request be merged in front of next? |
| 1170 | */ |
| 1171 | if (next && end == next->vm_start && |
vishnu.ps | cc71aba | 2014-10-09 15:26:29 -0700 | [diff] [blame] | 1172 | mpol_equal(policy, vma_policy(next)) && |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1173 | can_vma_merge_before(next, vm_flags, |
Andrea Arcangeli | 19a809a | 2015-09-04 15:46:24 -0700 | [diff] [blame] | 1174 | anon_vma, file, pgoff+pglen, |
Colin Cross | 3e4578f | 2015-10-27 16:42:08 -0700 | [diff] [blame] | 1175 | vm_userfaultfd_ctx, |
| 1176 | anon_name)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1177 | if (prev && addr < prev->vm_end) /* case 4 */ |
Andrea Arcangeli | e86f15e | 2016-10-07 17:01:28 -0700 | [diff] [blame] | 1178 | err = __vma_adjust(prev, prev->vm_start, |
| 1179 | addr, prev->vm_pgoff, NULL, next); |
| 1180 | else { /* cases 3, 8 */ |
| 1181 | err = __vma_adjust(area, addr, next->vm_end, |
| 1182 | next->vm_pgoff - pglen, NULL, next); |
| 1183 | /* |
| 1184 | * In case 3 area is already equal to next and |
| 1185 | * this is a noop, but in case 8 "area" has |
| 1186 | * been removed and next was expanded over it. |
| 1187 | */ |
| 1188 | area = next; |
| 1189 | } |
Rik van Riel | 5beb493 | 2010-03-05 13:42:07 -0800 | [diff] [blame] | 1190 | if (err) |
| 1191 | return NULL; |
David Rientjes | 6d50e60 | 2014-10-29 14:50:31 -0700 | [diff] [blame] | 1192 | khugepaged_enter_vma_merge(area, vm_flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1193 | return area; |
| 1194 | } |
| 1195 | |
| 1196 | return NULL; |
| 1197 | } |
| 1198 | |
| 1199 | /* |
Linus Torvalds | d0e9fe1 | 2010-04-10 10:36:19 -0700 | [diff] [blame] | 1200 | * Rough compatbility check to quickly see if it's even worth looking |
| 1201 | * at sharing an anon_vma. |
| 1202 | * |
| 1203 | * They need to have the same vm_file, and the flags can only differ |
| 1204 | * in things that mprotect may change. |
| 1205 | * |
| 1206 | * NOTE! The fact that we share an anon_vma doesn't _have_ to mean that |
| 1207 | * we can merge the two vma's. For example, we refuse to merge a vma if |
| 1208 | * there is a vm_ops->close() function, because that indicates that the |
| 1209 | * driver is doing some kind of reference counting. But that doesn't |
| 1210 | * really matter for the anon_vma sharing case. |
| 1211 | */ |
| 1212 | static int anon_vma_compatible(struct vm_area_struct *a, struct vm_area_struct *b) |
| 1213 | { |
| 1214 | return a->vm_end == b->vm_start && |
| 1215 | mpol_equal(vma_policy(a), vma_policy(b)) && |
| 1216 | a->vm_file == b->vm_file && |
Cyrill Gorcunov | 34228d4 | 2014-01-23 15:53:42 -0800 | [diff] [blame] | 1217 | !((a->vm_flags ^ b->vm_flags) & ~(VM_READ|VM_WRITE|VM_EXEC|VM_SOFTDIRTY)) && |
Linus Torvalds | d0e9fe1 | 2010-04-10 10:36:19 -0700 | [diff] [blame] | 1218 | b->vm_pgoff == a->vm_pgoff + ((b->vm_start - a->vm_start) >> PAGE_SHIFT); |
| 1219 | } |
| 1220 | |
| 1221 | /* |
| 1222 | * Do some basic sanity checking to see if we can re-use the anon_vma |
| 1223 | * from 'old'. The 'a'/'b' vma's are in VM order - one of them will be |
| 1224 | * the same as 'old', the other will be the new one that is trying |
| 1225 | * to share the anon_vma. |
| 1226 | * |
| 1227 | * NOTE! This runs with mm_sem held for reading, so it is possible that |
| 1228 | * the anon_vma of 'old' is concurrently in the process of being set up |
| 1229 | * by another page fault trying to merge _that_. But that's ok: if it |
| 1230 | * is being set up, that automatically means that it will be a singleton |
| 1231 | * acceptable for merging, so we can do all of this optimistically. But |
Jason Low | 4db0c3c | 2015-04-15 16:14:08 -0700 | [diff] [blame] | 1232 | * we do that READ_ONCE() to make sure that we never re-load the pointer. |
Linus Torvalds | d0e9fe1 | 2010-04-10 10:36:19 -0700 | [diff] [blame] | 1233 | * |
| 1234 | * IOW: that the "list_is_singular()" test on the anon_vma_chain only |
| 1235 | * matters for the 'stable anon_vma' case (ie the thing we want to avoid |
| 1236 | * is to return an anon_vma that is "complex" due to having gone through |
| 1237 | * a fork). |
| 1238 | * |
| 1239 | * We also make sure that the two vma's are compatible (adjacent, |
| 1240 | * and with the same memory policies). That's all stable, even with just |
| 1241 | * a read lock on the mm_sem. |
| 1242 | */ |
| 1243 | static struct anon_vma *reusable_anon_vma(struct vm_area_struct *old, struct vm_area_struct *a, struct vm_area_struct *b) |
| 1244 | { |
| 1245 | if (anon_vma_compatible(a, b)) { |
Jason Low | 4db0c3c | 2015-04-15 16:14:08 -0700 | [diff] [blame] | 1246 | struct anon_vma *anon_vma = READ_ONCE(old->anon_vma); |
Linus Torvalds | d0e9fe1 | 2010-04-10 10:36:19 -0700 | [diff] [blame] | 1247 | |
| 1248 | if (anon_vma && list_is_singular(&old->anon_vma_chain)) |
| 1249 | return anon_vma; |
| 1250 | } |
| 1251 | return NULL; |
| 1252 | } |
| 1253 | |
| 1254 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1255 | * find_mergeable_anon_vma is used by anon_vma_prepare, to check |
| 1256 | * neighbouring vmas for a suitable anon_vma, before it goes off |
| 1257 | * to allocate a new anon_vma. It checks because a repetitive |
| 1258 | * sequence of mprotects and faults may otherwise lead to distinct |
| 1259 | * anon_vmas being allocated, preventing vma merge in subsequent |
| 1260 | * mprotect. |
| 1261 | */ |
| 1262 | struct anon_vma *find_mergeable_anon_vma(struct vm_area_struct *vma) |
| 1263 | { |
Linus Torvalds | d0e9fe1 | 2010-04-10 10:36:19 -0700 | [diff] [blame] | 1264 | struct anon_vma *anon_vma; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1265 | struct vm_area_struct *near; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1266 | |
| 1267 | near = vma->vm_next; |
| 1268 | if (!near) |
| 1269 | goto try_prev; |
| 1270 | |
Linus Torvalds | d0e9fe1 | 2010-04-10 10:36:19 -0700 | [diff] [blame] | 1271 | anon_vma = reusable_anon_vma(near, vma, near); |
| 1272 | if (anon_vma) |
| 1273 | return anon_vma; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1274 | try_prev: |
Linus Torvalds | 9be34c9 | 2011-06-16 00:35:09 -0700 | [diff] [blame] | 1275 | near = vma->vm_prev; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1276 | if (!near) |
| 1277 | goto none; |
| 1278 | |
Linus Torvalds | d0e9fe1 | 2010-04-10 10:36:19 -0700 | [diff] [blame] | 1279 | anon_vma = reusable_anon_vma(near, near, vma); |
| 1280 | if (anon_vma) |
| 1281 | return anon_vma; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1282 | none: |
| 1283 | /* |
| 1284 | * There's no absolute need to look only at touching neighbours: |
| 1285 | * we could search further afield for "compatible" anon_vmas. |
| 1286 | * But it would probably just be a waste of time searching, |
| 1287 | * or lead to too many vmas hanging off the same anon_vma. |
| 1288 | * We're trying to allow mprotect remerging later on, |
| 1289 | * not trying to minimize memory used for anon_vmas. |
| 1290 | */ |
| 1291 | return NULL; |
| 1292 | } |
| 1293 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1294 | /* |
Al Viro | 4040153 | 2012-02-13 03:58:52 +0000 | [diff] [blame] | 1295 | * If a hint addr is less than mmap_min_addr change hint to be as |
| 1296 | * low as possible but still greater than mmap_min_addr |
| 1297 | */ |
| 1298 | static inline unsigned long round_hint_to_min(unsigned long hint) |
| 1299 | { |
| 1300 | hint &= PAGE_MASK; |
| 1301 | if (((void *)hint != NULL) && |
| 1302 | (hint < mmap_min_addr)) |
| 1303 | return PAGE_ALIGN(mmap_min_addr); |
| 1304 | return hint; |
| 1305 | } |
| 1306 | |
Davidlohr Bueso | 363ee17 | 2014-01-21 15:49:15 -0800 | [diff] [blame] | 1307 | static inline int mlock_future_check(struct mm_struct *mm, |
| 1308 | unsigned long flags, |
| 1309 | unsigned long len) |
| 1310 | { |
| 1311 | unsigned long locked, lock_limit; |
| 1312 | |
| 1313 | /* mlock MCL_FUTURE? */ |
| 1314 | if (flags & VM_LOCKED) { |
| 1315 | locked = len >> PAGE_SHIFT; |
| 1316 | locked += mm->locked_vm; |
| 1317 | lock_limit = rlimit(RLIMIT_MEMLOCK); |
| 1318 | lock_limit >>= PAGE_SHIFT; |
| 1319 | if (locked > lock_limit && !capable(CAP_IPC_LOCK)) |
| 1320 | return -EAGAIN; |
| 1321 | } |
| 1322 | return 0; |
| 1323 | } |
| 1324 | |
Al Viro | 4040153 | 2012-02-13 03:58:52 +0000 | [diff] [blame] | 1325 | /* |
Jianjun Kong | 27f5de7 | 2009-09-17 19:26:26 -0700 | [diff] [blame] | 1326 | * The caller must hold down_write(¤t->mm->mmap_sem). |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1327 | */ |
Oleg Nesterov | 1fcfd8d | 2015-09-09 15:39:29 -0700 | [diff] [blame] | 1328 | unsigned long do_mmap(struct file *file, unsigned long addr, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1329 | unsigned long len, unsigned long prot, |
Oleg Nesterov | 1fcfd8d | 2015-09-09 15:39:29 -0700 | [diff] [blame] | 1330 | unsigned long flags, vm_flags_t vm_flags, |
| 1331 | unsigned long pgoff, unsigned long *populate) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1332 | { |
vishnu.ps | cc71aba | 2014-10-09 15:26:29 -0700 | [diff] [blame] | 1333 | struct mm_struct *mm = current->mm; |
Dave Hansen | 62b5f7d | 2016-02-12 13:02:40 -0800 | [diff] [blame] | 1334 | int pkey = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1335 | |
Michel Lespinasse | 41badc1 | 2013-02-22 16:32:47 -0800 | [diff] [blame] | 1336 | *populate = 0; |
Michel Lespinasse | bebeb3d | 2013-02-22 16:32:37 -0800 | [diff] [blame] | 1337 | |
Piotr Kwapulinski | e37609b | 2015-06-24 16:58:39 -0700 | [diff] [blame] | 1338 | if (!len) |
| 1339 | return -EINVAL; |
| 1340 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1341 | /* |
| 1342 | * Does the application expect PROT_READ to imply PROT_EXEC? |
| 1343 | * |
| 1344 | * (the exception is when the underlying filesystem is noexec |
| 1345 | * mounted, in which case we dont add PROT_EXEC.) |
| 1346 | */ |
| 1347 | if ((prot & PROT_READ) && (current->personality & READ_IMPLIES_EXEC)) |
Eric W. Biederman | 90f8572 | 2015-06-29 14:42:03 -0500 | [diff] [blame] | 1348 | if (!(file && path_noexec(&file->f_path))) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1349 | prot |= PROT_EXEC; |
| 1350 | |
Eric Paris | 7cd9414 | 2007-11-26 18:47:40 -0500 | [diff] [blame] | 1351 | if (!(flags & MAP_FIXED)) |
| 1352 | addr = round_hint_to_min(addr); |
| 1353 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1354 | /* Careful about overflows.. */ |
| 1355 | len = PAGE_ALIGN(len); |
Al Viro | 9206de9 | 2009-12-03 15:23:11 -0500 | [diff] [blame] | 1356 | if (!len) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1357 | return -ENOMEM; |
| 1358 | |
| 1359 | /* offset overflow? */ |
| 1360 | if ((pgoff + (len >> PAGE_SHIFT)) < pgoff) |
vishnu.ps | cc71aba | 2014-10-09 15:26:29 -0700 | [diff] [blame] | 1361 | return -EOVERFLOW; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1362 | |
| 1363 | /* Too many mappings? */ |
| 1364 | if (mm->map_count > sysctl_max_map_count) |
| 1365 | return -ENOMEM; |
| 1366 | |
| 1367 | /* Obtain the address to map to. we verify (or select) it and ensure |
| 1368 | * that it represents a valid section of the address space. |
| 1369 | */ |
| 1370 | addr = get_unmapped_area(file, addr, len, pgoff, flags); |
Alexander Kuleshov | de1741a | 2015-11-05 18:46:54 -0800 | [diff] [blame] | 1371 | if (offset_in_page(addr)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1372 | return addr; |
| 1373 | |
Dave Hansen | 62b5f7d | 2016-02-12 13:02:40 -0800 | [diff] [blame] | 1374 | if (prot == PROT_EXEC) { |
| 1375 | pkey = execute_only_pkey(mm); |
| 1376 | if (pkey < 0) |
| 1377 | pkey = 0; |
| 1378 | } |
| 1379 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1380 | /* Do simple checking here so the lower-level routines won't have |
| 1381 | * to. we assume access permissions have been handled by the open |
| 1382 | * of the memory object, so we don't do any here. |
| 1383 | */ |
Dave Hansen | 62b5f7d | 2016-02-12 13:02:40 -0800 | [diff] [blame] | 1384 | vm_flags |= calc_vm_prot_bits(prot, pkey) | calc_vm_flag_bits(flags) | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1385 | mm->def_flags | VM_MAYREAD | VM_MAYWRITE | VM_MAYEXEC; |
| 1386 | |
Huang Shijie | cdf7b34 | 2009-09-21 17:03:36 -0700 | [diff] [blame] | 1387 | if (flags & MAP_LOCKED) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1388 | if (!can_do_mlock()) |
| 1389 | return -EPERM; |
Rik van Riel | ba470de | 2008-10-18 20:26:50 -0700 | [diff] [blame] | 1390 | |
Davidlohr Bueso | 363ee17 | 2014-01-21 15:49:15 -0800 | [diff] [blame] | 1391 | if (mlock_future_check(mm, vm_flags, len)) |
| 1392 | return -EAGAIN; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1393 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1394 | if (file) { |
Oleg Nesterov | 077bf22 | 2013-09-11 14:20:19 -0700 | [diff] [blame] | 1395 | struct inode *inode = file_inode(file); |
| 1396 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1397 | switch (flags & MAP_TYPE) { |
| 1398 | case MAP_SHARED: |
| 1399 | if ((prot&PROT_WRITE) && !(file->f_mode&FMODE_WRITE)) |
| 1400 | return -EACCES; |
| 1401 | |
| 1402 | /* |
| 1403 | * Make sure we don't allow writing to an append-only |
| 1404 | * file.. |
| 1405 | */ |
| 1406 | if (IS_APPEND(inode) && (file->f_mode & FMODE_WRITE)) |
| 1407 | return -EACCES; |
| 1408 | |
| 1409 | /* |
| 1410 | * Make sure there are no mandatory locks on the file. |
| 1411 | */ |
Jeff Layton | d7a0698 | 2014-03-10 09:54:15 -0400 | [diff] [blame] | 1412 | if (locks_verify_locked(file)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1413 | return -EAGAIN; |
| 1414 | |
| 1415 | vm_flags |= VM_SHARED | VM_MAYSHARE; |
| 1416 | if (!(file->f_mode & FMODE_WRITE)) |
| 1417 | vm_flags &= ~(VM_MAYWRITE | VM_SHARED); |
| 1418 | |
| 1419 | /* fall through */ |
| 1420 | case MAP_PRIVATE: |
| 1421 | if (!(file->f_mode & FMODE_READ)) |
| 1422 | return -EACCES; |
Eric W. Biederman | 90f8572 | 2015-06-29 14:42:03 -0500 | [diff] [blame] | 1423 | if (path_noexec(&file->f_path)) { |
Linus Torvalds | 80c5606 | 2006-10-15 14:09:55 -0700 | [diff] [blame] | 1424 | if (vm_flags & VM_EXEC) |
| 1425 | return -EPERM; |
| 1426 | vm_flags &= ~VM_MAYEXEC; |
| 1427 | } |
Linus Torvalds | 80c5606 | 2006-10-15 14:09:55 -0700 | [diff] [blame] | 1428 | |
Al Viro | 72c2d53 | 2013-09-22 16:27:52 -0400 | [diff] [blame] | 1429 | if (!file->f_op->mmap) |
Linus Torvalds | 80c5606 | 2006-10-15 14:09:55 -0700 | [diff] [blame] | 1430 | return -ENODEV; |
Oleg Nesterov | b2c56e4 | 2013-09-11 14:20:18 -0700 | [diff] [blame] | 1431 | if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP)) |
| 1432 | return -EINVAL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1433 | break; |
| 1434 | |
| 1435 | default: |
| 1436 | return -EINVAL; |
| 1437 | } |
| 1438 | } else { |
| 1439 | switch (flags & MAP_TYPE) { |
| 1440 | case MAP_SHARED: |
Oleg Nesterov | b2c56e4 | 2013-09-11 14:20:18 -0700 | [diff] [blame] | 1441 | if (vm_flags & (VM_GROWSDOWN|VM_GROWSUP)) |
| 1442 | return -EINVAL; |
Tejun Heo | ce36394 | 2008-09-03 16:09:47 +0200 | [diff] [blame] | 1443 | /* |
| 1444 | * Ignore pgoff. |
| 1445 | */ |
| 1446 | pgoff = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1447 | vm_flags |= VM_SHARED | VM_MAYSHARE; |
| 1448 | break; |
| 1449 | case MAP_PRIVATE: |
| 1450 | /* |
| 1451 | * Set pgoff according to addr for anon_vma. |
| 1452 | */ |
| 1453 | pgoff = addr >> PAGE_SHIFT; |
| 1454 | break; |
| 1455 | default: |
| 1456 | return -EINVAL; |
| 1457 | } |
| 1458 | } |
| 1459 | |
Michel Lespinasse | c22c0d6 | 2013-02-22 16:32:43 -0800 | [diff] [blame] | 1460 | /* |
| 1461 | * Set 'VM_NORESERVE' if we should not account for the |
| 1462 | * memory use of this mapping. |
| 1463 | */ |
| 1464 | if (flags & MAP_NORESERVE) { |
| 1465 | /* We honor MAP_NORESERVE if allowed to overcommit */ |
| 1466 | if (sysctl_overcommit_memory != OVERCOMMIT_NEVER) |
| 1467 | vm_flags |= VM_NORESERVE; |
| 1468 | |
| 1469 | /* hugetlb applies strict overcommit unless MAP_NORESERVE */ |
| 1470 | if (file && is_file_hugepages(file)) |
| 1471 | vm_flags |= VM_NORESERVE; |
| 1472 | } |
| 1473 | |
| 1474 | addr = mmap_region(file, addr, len, vm_flags, pgoff); |
Michel Lespinasse | 09a9f1d | 2013-03-28 16:26:23 -0700 | [diff] [blame] | 1475 | if (!IS_ERR_VALUE(addr) && |
| 1476 | ((vm_flags & VM_LOCKED) || |
| 1477 | (flags & (MAP_POPULATE | MAP_NONBLOCK)) == MAP_POPULATE)) |
Michel Lespinasse | 41badc1 | 2013-02-22 16:32:47 -0800 | [diff] [blame] | 1478 | *populate = len; |
Michel Lespinasse | bebeb3d | 2013-02-22 16:32:37 -0800 | [diff] [blame] | 1479 | return addr; |
Miklos Szeredi | 0165ab4 | 2007-07-15 23:38:26 -0700 | [diff] [blame] | 1480 | } |
Linus Torvalds | 6be5ceb | 2012-04-20 17:13:58 -0700 | [diff] [blame] | 1481 | |
Hugh Dickins | 66f0dc4 | 2009-12-30 20:17:34 +0000 | [diff] [blame] | 1482 | SYSCALL_DEFINE6(mmap_pgoff, unsigned long, addr, unsigned long, len, |
| 1483 | unsigned long, prot, unsigned long, flags, |
| 1484 | unsigned long, fd, unsigned long, pgoff) |
| 1485 | { |
| 1486 | struct file *file = NULL; |
Chen Gang | 1e3ee14 | 2015-11-05 18:48:35 -0800 | [diff] [blame] | 1487 | unsigned long retval; |
Hugh Dickins | 66f0dc4 | 2009-12-30 20:17:34 +0000 | [diff] [blame] | 1488 | |
| 1489 | if (!(flags & MAP_ANONYMOUS)) { |
Al Viro | 120a795 | 2010-10-30 02:54:44 -0400 | [diff] [blame] | 1490 | audit_mmap_fd(fd, flags); |
Hugh Dickins | 66f0dc4 | 2009-12-30 20:17:34 +0000 | [diff] [blame] | 1491 | file = fget(fd); |
| 1492 | if (!file) |
Chen Gang | 1e3ee14 | 2015-11-05 18:48:35 -0800 | [diff] [blame] | 1493 | return -EBADF; |
Naoya Horiguchi | af73e4d | 2013-05-07 16:18:13 -0700 | [diff] [blame] | 1494 | if (is_file_hugepages(file)) |
| 1495 | len = ALIGN(len, huge_page_size(hstate_file(file))); |
Jörn Engel | 493af57 | 2013-07-08 16:00:26 -0700 | [diff] [blame] | 1496 | retval = -EINVAL; |
| 1497 | if (unlikely(flags & MAP_HUGETLB && !is_file_hugepages(file))) |
| 1498 | goto out_fput; |
Hugh Dickins | 66f0dc4 | 2009-12-30 20:17:34 +0000 | [diff] [blame] | 1499 | } else if (flags & MAP_HUGETLB) { |
| 1500 | struct user_struct *user = NULL; |
Andrew Morton | c103a4d | 2013-07-08 16:01:08 -0700 | [diff] [blame] | 1501 | struct hstate *hs; |
Naoya Horiguchi | af73e4d | 2013-05-07 16:18:13 -0700 | [diff] [blame] | 1502 | |
Andrew Morton | c103a4d | 2013-07-08 16:01:08 -0700 | [diff] [blame] | 1503 | hs = hstate_sizelog((flags >> MAP_HUGE_SHIFT) & SHM_HUGE_MASK); |
Li Zefan | 091d0d5 | 2013-05-09 15:08:15 +0800 | [diff] [blame] | 1504 | if (!hs) |
| 1505 | return -EINVAL; |
| 1506 | |
| 1507 | len = ALIGN(len, huge_page_size(hs)); |
Hugh Dickins | 66f0dc4 | 2009-12-30 20:17:34 +0000 | [diff] [blame] | 1508 | /* |
| 1509 | * VM_NORESERVE is used because the reservations will be |
| 1510 | * taken when vm_ops->mmap() is called |
| 1511 | * A dummy user value is used because we are not locking |
| 1512 | * memory so no accounting is necessary |
| 1513 | */ |
Naoya Horiguchi | af73e4d | 2013-05-07 16:18:13 -0700 | [diff] [blame] | 1514 | file = hugetlb_file_setup(HUGETLB_ANON_FILE, len, |
Andi Kleen | 42d7395 | 2012-12-11 16:01:34 -0800 | [diff] [blame] | 1515 | VM_NORESERVE, |
| 1516 | &user, HUGETLB_ANONHUGE_INODE, |
| 1517 | (flags >> MAP_HUGE_SHIFT) & MAP_HUGE_MASK); |
Hugh Dickins | 66f0dc4 | 2009-12-30 20:17:34 +0000 | [diff] [blame] | 1518 | if (IS_ERR(file)) |
| 1519 | return PTR_ERR(file); |
| 1520 | } |
| 1521 | |
| 1522 | flags &= ~(MAP_EXECUTABLE | MAP_DENYWRITE); |
| 1523 | |
Michal Hocko | 9fbeb5a | 2016-05-23 16:25:30 -0700 | [diff] [blame] | 1524 | retval = vm_mmap_pgoff(file, addr, len, prot, flags, pgoff); |
Jörn Engel | 493af57 | 2013-07-08 16:00:26 -0700 | [diff] [blame] | 1525 | out_fput: |
Hugh Dickins | 66f0dc4 | 2009-12-30 20:17:34 +0000 | [diff] [blame] | 1526 | if (file) |
| 1527 | fput(file); |
Hugh Dickins | 66f0dc4 | 2009-12-30 20:17:34 +0000 | [diff] [blame] | 1528 | return retval; |
| 1529 | } |
| 1530 | |
Christoph Hellwig | a467937 | 2010-03-10 15:21:15 -0800 | [diff] [blame] | 1531 | #ifdef __ARCH_WANT_SYS_OLD_MMAP |
| 1532 | struct mmap_arg_struct { |
| 1533 | unsigned long addr; |
| 1534 | unsigned long len; |
| 1535 | unsigned long prot; |
| 1536 | unsigned long flags; |
| 1537 | unsigned long fd; |
| 1538 | unsigned long offset; |
| 1539 | }; |
| 1540 | |
| 1541 | SYSCALL_DEFINE1(old_mmap, struct mmap_arg_struct __user *, arg) |
| 1542 | { |
| 1543 | struct mmap_arg_struct a; |
| 1544 | |
| 1545 | if (copy_from_user(&a, arg, sizeof(a))) |
| 1546 | return -EFAULT; |
Alexander Kuleshov | de1741a | 2015-11-05 18:46:54 -0800 | [diff] [blame] | 1547 | if (offset_in_page(a.offset)) |
Christoph Hellwig | a467937 | 2010-03-10 15:21:15 -0800 | [diff] [blame] | 1548 | return -EINVAL; |
| 1549 | |
| 1550 | return sys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd, |
| 1551 | a.offset >> PAGE_SHIFT); |
| 1552 | } |
| 1553 | #endif /* __ARCH_WANT_SYS_OLD_MMAP */ |
| 1554 | |
Alexey Dobriyan | 4e950f6 | 2007-07-30 02:36:13 +0400 | [diff] [blame] | 1555 | /* |
| 1556 | * Some shared mappigns will want the pages marked read-only |
| 1557 | * to track write events. If so, we'll downgrade vm_page_prot |
| 1558 | * to the private version (using protection_map[] without the |
| 1559 | * VM_SHARED bit). |
| 1560 | */ |
Andrea Arcangeli | 6d2329f | 2016-10-07 17:01:22 -0700 | [diff] [blame] | 1561 | int vma_wants_writenotify(struct vm_area_struct *vma, pgprot_t vm_page_prot) |
Alexey Dobriyan | 4e950f6 | 2007-07-30 02:36:13 +0400 | [diff] [blame] | 1562 | { |
KOSAKI Motohiro | ca16d14 | 2011-05-26 19:16:19 +0900 | [diff] [blame] | 1563 | vm_flags_t vm_flags = vma->vm_flags; |
Kirill A. Shutemov | 8a04446 | 2015-09-22 14:59:12 -0700 | [diff] [blame] | 1564 | const struct vm_operations_struct *vm_ops = vma->vm_ops; |
Alexey Dobriyan | 4e950f6 | 2007-07-30 02:36:13 +0400 | [diff] [blame] | 1565 | |
| 1566 | /* If it was private or non-writable, the write bit is already clear */ |
| 1567 | if ((vm_flags & (VM_WRITE|VM_SHARED)) != ((VM_WRITE|VM_SHARED))) |
| 1568 | return 0; |
| 1569 | |
| 1570 | /* The backer wishes to know when pages are first written to? */ |
Kirill A. Shutemov | 8a04446 | 2015-09-22 14:59:12 -0700 | [diff] [blame] | 1571 | if (vm_ops && (vm_ops->page_mkwrite || vm_ops->pfn_mkwrite)) |
Alexey Dobriyan | 4e950f6 | 2007-07-30 02:36:13 +0400 | [diff] [blame] | 1572 | return 1; |
| 1573 | |
Peter Feiner | 64e4550 | 2014-10-13 15:55:46 -0700 | [diff] [blame] | 1574 | /* The open routine did something to the protections that pgprot_modify |
| 1575 | * won't preserve? */ |
Andrea Arcangeli | 6d2329f | 2016-10-07 17:01:22 -0700 | [diff] [blame] | 1576 | if (pgprot_val(vm_page_prot) != |
| 1577 | pgprot_val(vm_pgprot_modify(vm_page_prot, vm_flags))) |
Alexey Dobriyan | 4e950f6 | 2007-07-30 02:36:13 +0400 | [diff] [blame] | 1578 | return 0; |
| 1579 | |
Peter Feiner | 64e4550 | 2014-10-13 15:55:46 -0700 | [diff] [blame] | 1580 | /* Do we need to track softdirty? */ |
| 1581 | if (IS_ENABLED(CONFIG_MEM_SOFT_DIRTY) && !(vm_flags & VM_SOFTDIRTY)) |
| 1582 | return 1; |
| 1583 | |
Alexey Dobriyan | 4e950f6 | 2007-07-30 02:36:13 +0400 | [diff] [blame] | 1584 | /* Specialty mapping? */ |
Konstantin Khlebnikov | 4b6e1e3 | 2012-10-08 16:28:40 -0700 | [diff] [blame] | 1585 | if (vm_flags & VM_PFNMAP) |
Alexey Dobriyan | 4e950f6 | 2007-07-30 02:36:13 +0400 | [diff] [blame] | 1586 | return 0; |
| 1587 | |
| 1588 | /* Can the mapping track the dirty pages? */ |
| 1589 | return vma->vm_file && vma->vm_file->f_mapping && |
| 1590 | mapping_cap_account_dirty(vma->vm_file->f_mapping); |
| 1591 | } |
| 1592 | |
Linus Torvalds | fc8744a | 2009-01-31 15:08:56 -0800 | [diff] [blame] | 1593 | /* |
| 1594 | * We account for memory if it's a private writeable mapping, |
Mel Gorman | 5a6fe12 | 2009-02-10 14:02:27 +0000 | [diff] [blame] | 1595 | * not hugepages and VM_NORESERVE wasn't set. |
Linus Torvalds | fc8744a | 2009-01-31 15:08:56 -0800 | [diff] [blame] | 1596 | */ |
KOSAKI Motohiro | ca16d14 | 2011-05-26 19:16:19 +0900 | [diff] [blame] | 1597 | static inline int accountable_mapping(struct file *file, vm_flags_t vm_flags) |
Linus Torvalds | fc8744a | 2009-01-31 15:08:56 -0800 | [diff] [blame] | 1598 | { |
Mel Gorman | 5a6fe12 | 2009-02-10 14:02:27 +0000 | [diff] [blame] | 1599 | /* |
| 1600 | * hugetlb has its own accounting separate from the core VM |
| 1601 | * VM_HUGETLB may not be set yet so we cannot check for that flag. |
| 1602 | */ |
| 1603 | if (file && is_file_hugepages(file)) |
| 1604 | return 0; |
| 1605 | |
Linus Torvalds | fc8744a | 2009-01-31 15:08:56 -0800 | [diff] [blame] | 1606 | return (vm_flags & (VM_NORESERVE | VM_SHARED | VM_WRITE)) == VM_WRITE; |
| 1607 | } |
| 1608 | |
Miklos Szeredi | 0165ab4 | 2007-07-15 23:38:26 -0700 | [diff] [blame] | 1609 | unsigned long mmap_region(struct file *file, unsigned long addr, |
Michel Lespinasse | c22c0d6 | 2013-02-22 16:32:43 -0800 | [diff] [blame] | 1610 | unsigned long len, vm_flags_t vm_flags, unsigned long pgoff) |
Miklos Szeredi | 0165ab4 | 2007-07-15 23:38:26 -0700 | [diff] [blame] | 1611 | { |
| 1612 | struct mm_struct *mm = current->mm; |
| 1613 | struct vm_area_struct *vma, *prev; |
Miklos Szeredi | 0165ab4 | 2007-07-15 23:38:26 -0700 | [diff] [blame] | 1614 | int error; |
| 1615 | struct rb_node **rb_link, *rb_parent; |
| 1616 | unsigned long charged = 0; |
Miklos Szeredi | 0165ab4 | 2007-07-15 23:38:26 -0700 | [diff] [blame] | 1617 | |
Cyril Hrubis | e8420a8 | 2013-04-29 15:08:33 -0700 | [diff] [blame] | 1618 | /* Check against address space limit. */ |
Konstantin Khlebnikov | 8463833 | 2016-01-14 15:22:07 -0800 | [diff] [blame] | 1619 | if (!may_expand_vm(mm, vm_flags, len >> PAGE_SHIFT)) { |
Cyril Hrubis | e8420a8 | 2013-04-29 15:08:33 -0700 | [diff] [blame] | 1620 | unsigned long nr_pages; |
| 1621 | |
| 1622 | /* |
| 1623 | * MAP_FIXED may remove pages of mappings that intersects with |
| 1624 | * requested mapping. Account for the pages it would unmap. |
| 1625 | */ |
Cyril Hrubis | e8420a8 | 2013-04-29 15:08:33 -0700 | [diff] [blame] | 1626 | nr_pages = count_vma_pages_range(mm, addr, addr + len); |
| 1627 | |
Konstantin Khlebnikov | 8463833 | 2016-01-14 15:22:07 -0800 | [diff] [blame] | 1628 | if (!may_expand_vm(mm, vm_flags, |
| 1629 | (len >> PAGE_SHIFT) - nr_pages)) |
Cyril Hrubis | e8420a8 | 2013-04-29 15:08:33 -0700 | [diff] [blame] | 1630 | return -ENOMEM; |
| 1631 | } |
| 1632 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1633 | /* Clear old maps */ |
Rasmus Villemoes | 9fcd145 | 2015-04-15 16:14:32 -0700 | [diff] [blame] | 1634 | while (find_vma_links(mm, addr, addr + len, &prev, &rb_link, |
| 1635 | &rb_parent)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1636 | if (do_munmap(mm, addr, len)) |
| 1637 | return -ENOMEM; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1638 | } |
| 1639 | |
Linus Torvalds | fc8744a | 2009-01-31 15:08:56 -0800 | [diff] [blame] | 1640 | /* |
Linus Torvalds | fc8744a | 2009-01-31 15:08:56 -0800 | [diff] [blame] | 1641 | * Private writable mapping: check memory availability |
| 1642 | */ |
Mel Gorman | 5a6fe12 | 2009-02-10 14:02:27 +0000 | [diff] [blame] | 1643 | if (accountable_mapping(file, vm_flags)) { |
Linus Torvalds | fc8744a | 2009-01-31 15:08:56 -0800 | [diff] [blame] | 1644 | charged = len >> PAGE_SHIFT; |
Al Viro | 191c542 | 2012-02-13 03:58:52 +0000 | [diff] [blame] | 1645 | if (security_vm_enough_memory_mm(mm, charged)) |
Linus Torvalds | fc8744a | 2009-01-31 15:08:56 -0800 | [diff] [blame] | 1646 | return -ENOMEM; |
| 1647 | vm_flags |= VM_ACCOUNT; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1648 | } |
| 1649 | |
| 1650 | /* |
Linus Torvalds | de33c8d | 2009-01-29 17:46:42 -0800 | [diff] [blame] | 1651 | * Can we just expand an old mapping? |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1652 | */ |
Andrea Arcangeli | 19a809a | 2015-09-04 15:46:24 -0700 | [diff] [blame] | 1653 | vma = vma_merge(mm, prev, addr, addr + len, vm_flags, |
Colin Cross | 3e4578f | 2015-10-27 16:42:08 -0700 | [diff] [blame] | 1654 | NULL, file, pgoff, NULL, NULL_VM_UFFD_CTX, NULL); |
Linus Torvalds | de33c8d | 2009-01-29 17:46:42 -0800 | [diff] [blame] | 1655 | if (vma) |
| 1656 | goto out; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1657 | |
| 1658 | /* |
| 1659 | * Determine the object being mapped and call the appropriate |
| 1660 | * specific mapper. the address has already been validated, but |
| 1661 | * not unmapped, but the maps are removed from the list. |
| 1662 | */ |
Pekka Enberg | c5e3b83 | 2006-03-25 03:06:43 -0800 | [diff] [blame] | 1663 | vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1664 | if (!vma) { |
| 1665 | error = -ENOMEM; |
| 1666 | goto unacct_error; |
| 1667 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1668 | |
| 1669 | vma->vm_mm = mm; |
| 1670 | vma->vm_start = addr; |
| 1671 | vma->vm_end = addr + len; |
| 1672 | vma->vm_flags = vm_flags; |
Coly Li | 3ed75eb | 2007-10-18 23:39:15 -0700 | [diff] [blame] | 1673 | vma->vm_page_prot = vm_get_page_prot(vm_flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1674 | vma->vm_pgoff = pgoff; |
Rik van Riel | 5beb493 | 2010-03-05 13:42:07 -0800 | [diff] [blame] | 1675 | INIT_LIST_HEAD(&vma->anon_vma_chain); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1676 | |
| 1677 | if (file) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1678 | if (vm_flags & VM_DENYWRITE) { |
| 1679 | error = deny_write_access(file); |
| 1680 | if (error) |
| 1681 | goto free_vma; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1682 | } |
David Herrmann | 4bb5f5d | 2014-08-08 14:25:25 -0700 | [diff] [blame] | 1683 | if (vm_flags & VM_SHARED) { |
| 1684 | error = mapping_map_writable(file->f_mapping); |
| 1685 | if (error) |
| 1686 | goto allow_write_and_free_vma; |
| 1687 | } |
| 1688 | |
| 1689 | /* ->mmap() can change vma->vm_file, but must guarantee that |
| 1690 | * vma_link() below can deny write-access if VM_DENYWRITE is set |
| 1691 | * and map writably if VM_SHARED is set. This usually means the |
| 1692 | * new file must not have been exposed to user-space, yet. |
| 1693 | */ |
Al Viro | cb0942b | 2012-08-27 14:48:26 -0400 | [diff] [blame] | 1694 | vma->vm_file = get_file(file); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1695 | error = file->f_op->mmap(file, vma); |
| 1696 | if (error) |
| 1697 | goto unmap_and_free_vma; |
Huang Shijie | f8dbf0a7 | 2009-09-21 17:03:41 -0700 | [diff] [blame] | 1698 | |
| 1699 | /* Can addr have changed?? |
| 1700 | * |
| 1701 | * Answer: Yes, several device drivers can do it in their |
| 1702 | * f_op->mmap method. -DaveM |
Joonsoo Kim | 2897b4d | 2012-12-12 13:51:53 -0800 | [diff] [blame] | 1703 | * Bug: If addr is changed, prev, rb_link, rb_parent should |
| 1704 | * be updated for vma_link() |
Huang Shijie | f8dbf0a7 | 2009-09-21 17:03:41 -0700 | [diff] [blame] | 1705 | */ |
Joonsoo Kim | 2897b4d | 2012-12-12 13:51:53 -0800 | [diff] [blame] | 1706 | WARN_ON_ONCE(addr != vma->vm_start); |
| 1707 | |
Huang Shijie | f8dbf0a7 | 2009-09-21 17:03:41 -0700 | [diff] [blame] | 1708 | addr = vma->vm_start; |
Huang Shijie | f8dbf0a7 | 2009-09-21 17:03:41 -0700 | [diff] [blame] | 1709 | vm_flags = vma->vm_flags; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1710 | } else if (vm_flags & VM_SHARED) { |
| 1711 | error = shmem_zero_setup(vma); |
| 1712 | if (error) |
| 1713 | goto free_vma; |
| 1714 | } |
| 1715 | |
Linus Torvalds | de33c8d | 2009-01-29 17:46:42 -0800 | [diff] [blame] | 1716 | vma_link(mm, vma, prev, rb_link, rb_parent); |
Oleg Nesterov | 4d3d5b4 | 2008-04-28 02:12:10 -0700 | [diff] [blame] | 1717 | /* Once vma denies write, undo our temporary denial count */ |
David Herrmann | 4bb5f5d | 2014-08-08 14:25:25 -0700 | [diff] [blame] | 1718 | if (file) { |
| 1719 | if (vm_flags & VM_SHARED) |
| 1720 | mapping_unmap_writable(file->f_mapping); |
| 1721 | if (vm_flags & VM_DENYWRITE) |
| 1722 | allow_write_access(file); |
| 1723 | } |
Oleg Nesterov | e868677 | 2013-09-11 14:20:20 -0700 | [diff] [blame] | 1724 | file = vma->vm_file; |
Oleg Nesterov | 4d3d5b4 | 2008-04-28 02:12:10 -0700 | [diff] [blame] | 1725 | out: |
Ingo Molnar | cdd6c48 | 2009-09-21 12:02:48 +0200 | [diff] [blame] | 1726 | perf_event_mmap(vma); |
Peter Zijlstra | 0a4a939 | 2009-03-30 19:07:05 +0200 | [diff] [blame] | 1727 | |
Konstantin Khlebnikov | 8463833 | 2016-01-14 15:22:07 -0800 | [diff] [blame] | 1728 | vm_stat_account(mm, vm_flags, len >> PAGE_SHIFT); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1729 | if (vm_flags & VM_LOCKED) { |
Michel Lespinasse | bebeb3d | 2013-02-22 16:32:37 -0800 | [diff] [blame] | 1730 | if (!((vm_flags & VM_SPECIAL) || is_vm_hugetlb_page(vma) || |
| 1731 | vma == get_gate_vma(current->mm))) |
KOSAKI Motohiro | 06f9d8c | 2010-03-05 13:41:43 -0800 | [diff] [blame] | 1732 | mm->locked_vm += (len >> PAGE_SHIFT); |
Michel Lespinasse | bebeb3d | 2013-02-22 16:32:37 -0800 | [diff] [blame] | 1733 | else |
Eric B Munson | de60f5f | 2015-11-05 18:51:36 -0800 | [diff] [blame] | 1734 | vma->vm_flags &= VM_LOCKED_CLEAR_MASK; |
Michel Lespinasse | bebeb3d | 2013-02-22 16:32:37 -0800 | [diff] [blame] | 1735 | } |
Srikar Dronamraju | 2b14449 | 2012-02-09 14:56:42 +0530 | [diff] [blame] | 1736 | |
Oleg Nesterov | c7a3a88 | 2012-08-19 19:10:42 +0200 | [diff] [blame] | 1737 | if (file) |
| 1738 | uprobe_mmap(vma); |
Srikar Dronamraju | 2b14449 | 2012-02-09 14:56:42 +0530 | [diff] [blame] | 1739 | |
Cyrill Gorcunov | d9104d1 | 2013-09-11 14:22:24 -0700 | [diff] [blame] | 1740 | /* |
| 1741 | * New (or expanded) vma always get soft dirty status. |
| 1742 | * Otherwise user-space soft-dirty page tracker won't |
| 1743 | * be able to distinguish situation when vma area unmapped, |
| 1744 | * then new mapped in-place (which must be aimed as |
| 1745 | * a completely new data area). |
| 1746 | */ |
| 1747 | vma->vm_flags |= VM_SOFTDIRTY; |
| 1748 | |
Peter Feiner | 64e4550 | 2014-10-13 15:55:46 -0700 | [diff] [blame] | 1749 | vma_set_page_prot(vma); |
| 1750 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1751 | return addr; |
| 1752 | |
| 1753 | unmap_and_free_vma: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1754 | vma->vm_file = NULL; |
| 1755 | fput(file); |
| 1756 | |
| 1757 | /* Undo any partial mapping done by a device driver. */ |
Hugh Dickins | e0da382 | 2005-04-19 13:29:15 -0700 | [diff] [blame] | 1758 | unmap_region(mm, vma, prev, vma->vm_start, vma->vm_end); |
| 1759 | charged = 0; |
David Herrmann | 4bb5f5d | 2014-08-08 14:25:25 -0700 | [diff] [blame] | 1760 | if (vm_flags & VM_SHARED) |
| 1761 | mapping_unmap_writable(file->f_mapping); |
| 1762 | allow_write_and_free_vma: |
| 1763 | if (vm_flags & VM_DENYWRITE) |
| 1764 | allow_write_access(file); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1765 | free_vma: |
| 1766 | kmem_cache_free(vm_area_cachep, vma); |
| 1767 | unacct_error: |
| 1768 | if (charged) |
| 1769 | vm_unacct_memory(charged); |
| 1770 | return error; |
| 1771 | } |
| 1772 | |
Michel Lespinasse | db4fbfb | 2012-12-11 16:01:49 -0800 | [diff] [blame] | 1773 | unsigned long unmapped_area(struct vm_unmapped_area_info *info) |
| 1774 | { |
| 1775 | /* |
| 1776 | * We implement the search by looking for an rbtree node that |
| 1777 | * immediately follows a suitable gap. That is, |
| 1778 | * - gap_start = vma->vm_prev->vm_end <= info->high_limit - length; |
| 1779 | * - gap_end = vma->vm_start >= info->low_limit + length; |
| 1780 | * - gap_end - gap_start >= length |
| 1781 | */ |
| 1782 | |
| 1783 | struct mm_struct *mm = current->mm; |
| 1784 | struct vm_area_struct *vma; |
| 1785 | unsigned long length, low_limit, high_limit, gap_start, gap_end; |
| 1786 | |
| 1787 | /* Adjust search length to account for worst case alignment overhead */ |
| 1788 | length = info->length + info->align_mask; |
| 1789 | if (length < info->length) |
| 1790 | return -ENOMEM; |
| 1791 | |
| 1792 | /* Adjust search limits by the desired length */ |
| 1793 | if (info->high_limit < length) |
| 1794 | return -ENOMEM; |
| 1795 | high_limit = info->high_limit - length; |
| 1796 | |
| 1797 | if (info->low_limit > high_limit) |
| 1798 | return -ENOMEM; |
| 1799 | low_limit = info->low_limit + length; |
| 1800 | |
| 1801 | /* Check if rbtree root looks promising */ |
| 1802 | if (RB_EMPTY_ROOT(&mm->mm_rb)) |
| 1803 | goto check_highest; |
| 1804 | vma = rb_entry(mm->mm_rb.rb_node, struct vm_area_struct, vm_rb); |
| 1805 | if (vma->rb_subtree_gap < length) |
| 1806 | goto check_highest; |
| 1807 | |
| 1808 | while (true) { |
| 1809 | /* Visit left subtree if it looks promising */ |
Hugh Dickins | cfc0eb40 | 2017-06-19 04:03:24 -0700 | [diff] [blame] | 1810 | gap_end = vm_start_gap(vma); |
Michel Lespinasse | db4fbfb | 2012-12-11 16:01:49 -0800 | [diff] [blame] | 1811 | if (gap_end >= low_limit && vma->vm_rb.rb_left) { |
| 1812 | struct vm_area_struct *left = |
| 1813 | rb_entry(vma->vm_rb.rb_left, |
| 1814 | struct vm_area_struct, vm_rb); |
| 1815 | if (left->rb_subtree_gap >= length) { |
| 1816 | vma = left; |
| 1817 | continue; |
| 1818 | } |
| 1819 | } |
| 1820 | |
Hugh Dickins | cfc0eb40 | 2017-06-19 04:03:24 -0700 | [diff] [blame] | 1821 | gap_start = vma->vm_prev ? vm_end_gap(vma->vm_prev) : 0; |
Michel Lespinasse | db4fbfb | 2012-12-11 16:01:49 -0800 | [diff] [blame] | 1822 | check_current: |
| 1823 | /* Check if current node has a suitable gap */ |
| 1824 | if (gap_start > high_limit) |
| 1825 | return -ENOMEM; |
Hugh Dickins | ce7fe85 | 2017-06-20 02:10:44 -0700 | [diff] [blame] | 1826 | if (gap_end >= low_limit && |
| 1827 | gap_end > gap_start && gap_end - gap_start >= length) |
Michel Lespinasse | db4fbfb | 2012-12-11 16:01:49 -0800 | [diff] [blame] | 1828 | goto found; |
| 1829 | |
| 1830 | /* Visit right subtree if it looks promising */ |
| 1831 | if (vma->vm_rb.rb_right) { |
| 1832 | struct vm_area_struct *right = |
| 1833 | rb_entry(vma->vm_rb.rb_right, |
| 1834 | struct vm_area_struct, vm_rb); |
| 1835 | if (right->rb_subtree_gap >= length) { |
| 1836 | vma = right; |
| 1837 | continue; |
| 1838 | } |
| 1839 | } |
| 1840 | |
| 1841 | /* Go back up the rbtree to find next candidate node */ |
| 1842 | while (true) { |
| 1843 | struct rb_node *prev = &vma->vm_rb; |
| 1844 | if (!rb_parent(prev)) |
| 1845 | goto check_highest; |
| 1846 | vma = rb_entry(rb_parent(prev), |
| 1847 | struct vm_area_struct, vm_rb); |
| 1848 | if (prev == vma->vm_rb.rb_left) { |
Hugh Dickins | cfc0eb40 | 2017-06-19 04:03:24 -0700 | [diff] [blame] | 1849 | gap_start = vm_end_gap(vma->vm_prev); |
| 1850 | gap_end = vm_start_gap(vma); |
Michel Lespinasse | db4fbfb | 2012-12-11 16:01:49 -0800 | [diff] [blame] | 1851 | goto check_current; |
| 1852 | } |
| 1853 | } |
| 1854 | } |
| 1855 | |
| 1856 | check_highest: |
| 1857 | /* Check highest gap, which does not precede any rbtree node */ |
| 1858 | gap_start = mm->highest_vm_end; |
| 1859 | gap_end = ULONG_MAX; /* Only for VM_BUG_ON below */ |
| 1860 | if (gap_start > high_limit) |
| 1861 | return -ENOMEM; |
| 1862 | |
| 1863 | found: |
| 1864 | /* We found a suitable gap. Clip it with the original low_limit. */ |
| 1865 | if (gap_start < info->low_limit) |
| 1866 | gap_start = info->low_limit; |
| 1867 | |
| 1868 | /* Adjust gap address to the desired alignment */ |
| 1869 | gap_start += (info->align_offset - gap_start) & info->align_mask; |
| 1870 | |
| 1871 | VM_BUG_ON(gap_start + info->length > info->high_limit); |
| 1872 | VM_BUG_ON(gap_start + info->length > gap_end); |
| 1873 | return gap_start; |
| 1874 | } |
| 1875 | |
| 1876 | unsigned long unmapped_area_topdown(struct vm_unmapped_area_info *info) |
| 1877 | { |
| 1878 | struct mm_struct *mm = current->mm; |
| 1879 | struct vm_area_struct *vma; |
| 1880 | unsigned long length, low_limit, high_limit, gap_start, gap_end; |
| 1881 | |
| 1882 | /* Adjust search length to account for worst case alignment overhead */ |
| 1883 | length = info->length + info->align_mask; |
| 1884 | if (length < info->length) |
| 1885 | return -ENOMEM; |
| 1886 | |
| 1887 | /* |
| 1888 | * Adjust search limits by the desired length. |
| 1889 | * See implementation comment at top of unmapped_area(). |
| 1890 | */ |
| 1891 | gap_end = info->high_limit; |
| 1892 | if (gap_end < length) |
| 1893 | return -ENOMEM; |
| 1894 | high_limit = gap_end - length; |
| 1895 | |
| 1896 | if (info->low_limit > high_limit) |
| 1897 | return -ENOMEM; |
| 1898 | low_limit = info->low_limit + length; |
| 1899 | |
| 1900 | /* Check highest gap, which does not precede any rbtree node */ |
| 1901 | gap_start = mm->highest_vm_end; |
| 1902 | if (gap_start <= high_limit) |
| 1903 | goto found_highest; |
| 1904 | |
| 1905 | /* Check if rbtree root looks promising */ |
| 1906 | if (RB_EMPTY_ROOT(&mm->mm_rb)) |
| 1907 | return -ENOMEM; |
| 1908 | vma = rb_entry(mm->mm_rb.rb_node, struct vm_area_struct, vm_rb); |
| 1909 | if (vma->rb_subtree_gap < length) |
| 1910 | return -ENOMEM; |
| 1911 | |
| 1912 | while (true) { |
| 1913 | /* Visit right subtree if it looks promising */ |
Hugh Dickins | cfc0eb40 | 2017-06-19 04:03:24 -0700 | [diff] [blame] | 1914 | gap_start = vma->vm_prev ? vm_end_gap(vma->vm_prev) : 0; |
Michel Lespinasse | db4fbfb | 2012-12-11 16:01:49 -0800 | [diff] [blame] | 1915 | if (gap_start <= high_limit && vma->vm_rb.rb_right) { |
| 1916 | struct vm_area_struct *right = |
| 1917 | rb_entry(vma->vm_rb.rb_right, |
| 1918 | struct vm_area_struct, vm_rb); |
| 1919 | if (right->rb_subtree_gap >= length) { |
| 1920 | vma = right; |
| 1921 | continue; |
| 1922 | } |
| 1923 | } |
| 1924 | |
| 1925 | check_current: |
| 1926 | /* Check if current node has a suitable gap */ |
Hugh Dickins | cfc0eb40 | 2017-06-19 04:03:24 -0700 | [diff] [blame] | 1927 | gap_end = vm_start_gap(vma); |
Michel Lespinasse | db4fbfb | 2012-12-11 16:01:49 -0800 | [diff] [blame] | 1928 | if (gap_end < low_limit) |
| 1929 | return -ENOMEM; |
Hugh Dickins | ce7fe85 | 2017-06-20 02:10:44 -0700 | [diff] [blame] | 1930 | if (gap_start <= high_limit && |
| 1931 | gap_end > gap_start && gap_end - gap_start >= length) |
Michel Lespinasse | db4fbfb | 2012-12-11 16:01:49 -0800 | [diff] [blame] | 1932 | goto found; |
| 1933 | |
| 1934 | /* Visit left subtree if it looks promising */ |
| 1935 | if (vma->vm_rb.rb_left) { |
| 1936 | struct vm_area_struct *left = |
| 1937 | rb_entry(vma->vm_rb.rb_left, |
| 1938 | struct vm_area_struct, vm_rb); |
| 1939 | if (left->rb_subtree_gap >= length) { |
| 1940 | vma = left; |
| 1941 | continue; |
| 1942 | } |
| 1943 | } |
| 1944 | |
| 1945 | /* Go back up the rbtree to find next candidate node */ |
| 1946 | while (true) { |
| 1947 | struct rb_node *prev = &vma->vm_rb; |
| 1948 | if (!rb_parent(prev)) |
| 1949 | return -ENOMEM; |
| 1950 | vma = rb_entry(rb_parent(prev), |
| 1951 | struct vm_area_struct, vm_rb); |
| 1952 | if (prev == vma->vm_rb.rb_right) { |
| 1953 | gap_start = vma->vm_prev ? |
Hugh Dickins | cfc0eb40 | 2017-06-19 04:03:24 -0700 | [diff] [blame] | 1954 | vm_end_gap(vma->vm_prev) : 0; |
Michel Lespinasse | db4fbfb | 2012-12-11 16:01:49 -0800 | [diff] [blame] | 1955 | goto check_current; |
| 1956 | } |
| 1957 | } |
| 1958 | } |
| 1959 | |
| 1960 | found: |
| 1961 | /* We found a suitable gap. Clip it with the original high_limit. */ |
| 1962 | if (gap_end > info->high_limit) |
| 1963 | gap_end = info->high_limit; |
| 1964 | |
| 1965 | found_highest: |
| 1966 | /* Compute highest gap address at the desired alignment */ |
| 1967 | gap_end -= info->length; |
| 1968 | gap_end -= (gap_end - info->align_offset) & info->align_mask; |
| 1969 | |
| 1970 | VM_BUG_ON(gap_end < info->low_limit); |
| 1971 | VM_BUG_ON(gap_end < gap_start); |
| 1972 | return gap_end; |
| 1973 | } |
| 1974 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1975 | /* Get an address range which is currently unmapped. |
| 1976 | * For shmat() with addr=0. |
| 1977 | * |
| 1978 | * Ugly calling convention alert: |
| 1979 | * Return value with the low bits set means error value, |
| 1980 | * ie |
| 1981 | * if (ret & ~PAGE_MASK) |
| 1982 | * error = ret; |
| 1983 | * |
| 1984 | * This function "knows" that -ENOMEM has the bits set. |
| 1985 | */ |
| 1986 | #ifndef HAVE_ARCH_UNMAPPED_AREA |
| 1987 | unsigned long |
| 1988 | arch_get_unmapped_area(struct file *filp, unsigned long addr, |
| 1989 | unsigned long len, unsigned long pgoff, unsigned long flags) |
| 1990 | { |
| 1991 | struct mm_struct *mm = current->mm; |
Hugh Dickins | cfc0eb40 | 2017-06-19 04:03:24 -0700 | [diff] [blame] | 1992 | struct vm_area_struct *vma, *prev; |
Michel Lespinasse | db4fbfb | 2012-12-11 16:01:49 -0800 | [diff] [blame] | 1993 | struct vm_unmapped_area_info info; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1994 | |
Akira Takeuchi | 2afc745 | 2013-11-12 15:08:21 -0800 | [diff] [blame] | 1995 | if (len > TASK_SIZE - mmap_min_addr) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1996 | return -ENOMEM; |
| 1997 | |
Benjamin Herrenschmidt | 06abdfb | 2007-05-06 14:50:13 -0700 | [diff] [blame] | 1998 | if (flags & MAP_FIXED) |
| 1999 | return addr; |
| 2000 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2001 | if (addr) { |
| 2002 | addr = PAGE_ALIGN(addr); |
Hugh Dickins | cfc0eb40 | 2017-06-19 04:03:24 -0700 | [diff] [blame] | 2003 | vma = find_vma_prev(mm, addr, &prev); |
Akira Takeuchi | 2afc745 | 2013-11-12 15:08:21 -0800 | [diff] [blame] | 2004 | if (TASK_SIZE - len >= addr && addr >= mmap_min_addr && |
Hugh Dickins | cfc0eb40 | 2017-06-19 04:03:24 -0700 | [diff] [blame] | 2005 | (!vma || addr + len <= vm_start_gap(vma)) && |
| 2006 | (!prev || addr >= vm_end_gap(prev))) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2007 | return addr; |
| 2008 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2009 | |
Michel Lespinasse | db4fbfb | 2012-12-11 16:01:49 -0800 | [diff] [blame] | 2010 | info.flags = 0; |
| 2011 | info.length = len; |
Heiko Carstens | 4e99b02 | 2013-11-12 15:07:54 -0800 | [diff] [blame] | 2012 | info.low_limit = mm->mmap_base; |
Michel Lespinasse | db4fbfb | 2012-12-11 16:01:49 -0800 | [diff] [blame] | 2013 | info.high_limit = TASK_SIZE; |
| 2014 | info.align_mask = 0; |
| 2015 | return vm_unmapped_area(&info); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2016 | } |
vishnu.ps | cc71aba | 2014-10-09 15:26:29 -0700 | [diff] [blame] | 2017 | #endif |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2018 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2019 | /* |
| 2020 | * This mmap-allocator allocates new areas top-down from below the |
| 2021 | * stack's low limit (the base): |
| 2022 | */ |
| 2023 | #ifndef HAVE_ARCH_UNMAPPED_AREA_TOPDOWN |
| 2024 | unsigned long |
| 2025 | arch_get_unmapped_area_topdown(struct file *filp, const unsigned long addr0, |
| 2026 | const unsigned long len, const unsigned long pgoff, |
| 2027 | const unsigned long flags) |
| 2028 | { |
Hugh Dickins | cfc0eb40 | 2017-06-19 04:03:24 -0700 | [diff] [blame] | 2029 | struct vm_area_struct *vma, *prev; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2030 | struct mm_struct *mm = current->mm; |
Michel Lespinasse | db4fbfb | 2012-12-11 16:01:49 -0800 | [diff] [blame] | 2031 | unsigned long addr = addr0; |
| 2032 | struct vm_unmapped_area_info info; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2033 | |
| 2034 | /* requested length too big for entire address space */ |
Akira Takeuchi | 2afc745 | 2013-11-12 15:08:21 -0800 | [diff] [blame] | 2035 | if (len > TASK_SIZE - mmap_min_addr) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2036 | return -ENOMEM; |
| 2037 | |
Benjamin Herrenschmidt | 06abdfb | 2007-05-06 14:50:13 -0700 | [diff] [blame] | 2038 | if (flags & MAP_FIXED) |
| 2039 | return addr; |
| 2040 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2041 | /* requesting a specific address */ |
| 2042 | if (addr) { |
| 2043 | addr = PAGE_ALIGN(addr); |
Hugh Dickins | cfc0eb40 | 2017-06-19 04:03:24 -0700 | [diff] [blame] | 2044 | vma = find_vma_prev(mm, addr, &prev); |
Akira Takeuchi | 2afc745 | 2013-11-12 15:08:21 -0800 | [diff] [blame] | 2045 | if (TASK_SIZE - len >= addr && addr >= mmap_min_addr && |
Hugh Dickins | cfc0eb40 | 2017-06-19 04:03:24 -0700 | [diff] [blame] | 2046 | (!vma || addr + len <= vm_start_gap(vma)) && |
| 2047 | (!prev || addr >= vm_end_gap(prev))) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2048 | return addr; |
| 2049 | } |
| 2050 | |
Michel Lespinasse | db4fbfb | 2012-12-11 16:01:49 -0800 | [diff] [blame] | 2051 | info.flags = VM_UNMAPPED_AREA_TOPDOWN; |
| 2052 | info.length = len; |
Akira Takeuchi | 2afc745 | 2013-11-12 15:08:21 -0800 | [diff] [blame] | 2053 | info.low_limit = max(PAGE_SIZE, mmap_min_addr); |
Michel Lespinasse | db4fbfb | 2012-12-11 16:01:49 -0800 | [diff] [blame] | 2054 | info.high_limit = mm->mmap_base; |
| 2055 | info.align_mask = 0; |
| 2056 | addr = vm_unmapped_area(&info); |
Xiao Guangrong | b716ad9 | 2012-03-21 16:33:56 -0700 | [diff] [blame] | 2057 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2058 | /* |
| 2059 | * A failed mmap() very likely causes application failure, |
| 2060 | * so fall back to the bottom-up function here. This scenario |
| 2061 | * can happen with large stack limits and large mmap() |
| 2062 | * allocations. |
| 2063 | */ |
Alexander Kuleshov | de1741a | 2015-11-05 18:46:54 -0800 | [diff] [blame] | 2064 | if (offset_in_page(addr)) { |
Michel Lespinasse | db4fbfb | 2012-12-11 16:01:49 -0800 | [diff] [blame] | 2065 | VM_BUG_ON(addr != -ENOMEM); |
| 2066 | info.flags = 0; |
| 2067 | info.low_limit = TASK_UNMAPPED_BASE; |
| 2068 | info.high_limit = TASK_SIZE; |
| 2069 | addr = vm_unmapped_area(&info); |
| 2070 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2071 | |
| 2072 | return addr; |
| 2073 | } |
| 2074 | #endif |
| 2075 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2076 | unsigned long |
| 2077 | get_unmapped_area(struct file *file, unsigned long addr, unsigned long len, |
| 2078 | unsigned long pgoff, unsigned long flags) |
| 2079 | { |
Benjamin Herrenschmidt | 06abdfb | 2007-05-06 14:50:13 -0700 | [diff] [blame] | 2080 | unsigned long (*get_area)(struct file *, unsigned long, |
| 2081 | unsigned long, unsigned long, unsigned long); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2082 | |
Al Viro | 9206de9 | 2009-12-03 15:23:11 -0500 | [diff] [blame] | 2083 | unsigned long error = arch_mmap_check(addr, len, flags); |
| 2084 | if (error) |
| 2085 | return error; |
| 2086 | |
| 2087 | /* Careful about overflows.. */ |
| 2088 | if (len > TASK_SIZE) |
| 2089 | return -ENOMEM; |
| 2090 | |
Benjamin Herrenschmidt | 06abdfb | 2007-05-06 14:50:13 -0700 | [diff] [blame] | 2091 | get_area = current->mm->get_unmapped_area; |
Hugh Dickins | c01d5b3 | 2016-07-26 15:26:15 -0700 | [diff] [blame] | 2092 | if (file) { |
| 2093 | if (file->f_op->get_unmapped_area) |
| 2094 | get_area = file->f_op->get_unmapped_area; |
| 2095 | } else if (flags & MAP_SHARED) { |
| 2096 | /* |
| 2097 | * mmap_region() will call shmem_zero_setup() to create a file, |
| 2098 | * so use shmem's get_unmapped_area in case it can be huge. |
| 2099 | * do_mmap_pgoff() will clear pgoff, so match alignment. |
| 2100 | */ |
| 2101 | pgoff = 0; |
| 2102 | get_area = shmem_get_unmapped_area; |
| 2103 | } |
| 2104 | |
Benjamin Herrenschmidt | 06abdfb | 2007-05-06 14:50:13 -0700 | [diff] [blame] | 2105 | addr = get_area(file, addr, len, pgoff, flags); |
| 2106 | if (IS_ERR_VALUE(addr)) |
| 2107 | return addr; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2108 | |
Linus Torvalds | 07ab67c | 2005-05-19 22:43:37 -0700 | [diff] [blame] | 2109 | if (addr > TASK_SIZE - len) |
| 2110 | return -ENOMEM; |
Alexander Kuleshov | de1741a | 2015-11-05 18:46:54 -0800 | [diff] [blame] | 2111 | if (offset_in_page(addr)) |
Linus Torvalds | 07ab67c | 2005-05-19 22:43:37 -0700 | [diff] [blame] | 2112 | return -EINVAL; |
Benjamin Herrenschmidt | 06abdfb | 2007-05-06 14:50:13 -0700 | [diff] [blame] | 2113 | |
Al Viro | 9ac4ed4 | 2012-05-30 17:13:15 -0400 | [diff] [blame] | 2114 | error = security_mmap_addr(addr); |
| 2115 | return error ? error : addr; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2116 | } |
| 2117 | |
| 2118 | EXPORT_SYMBOL(get_unmapped_area); |
| 2119 | |
| 2120 | /* Look up the first VMA which satisfies addr < vm_end, NULL if none. */ |
ZhenwenXu | 48aae42 | 2009-01-06 14:40:21 -0800 | [diff] [blame] | 2121 | struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2122 | { |
Davidlohr Bueso | 615d6e8 | 2014-04-07 15:37:25 -0700 | [diff] [blame] | 2123 | struct rb_node *rb_node; |
| 2124 | struct vm_area_struct *vma; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2125 | |
Rajman Mekaco | 841e31e | 2012-05-29 15:06:21 -0700 | [diff] [blame] | 2126 | /* Check the cache first. */ |
Davidlohr Bueso | 615d6e8 | 2014-04-07 15:37:25 -0700 | [diff] [blame] | 2127 | vma = vmacache_find(mm, addr); |
| 2128 | if (likely(vma)) |
| 2129 | return vma; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2130 | |
Davidlohr Bueso | 615d6e8 | 2014-04-07 15:37:25 -0700 | [diff] [blame] | 2131 | rb_node = mm->mm_rb.rb_node; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2132 | |
Davidlohr Bueso | 615d6e8 | 2014-04-07 15:37:25 -0700 | [diff] [blame] | 2133 | while (rb_node) { |
| 2134 | struct vm_area_struct *tmp; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2135 | |
Davidlohr Bueso | 615d6e8 | 2014-04-07 15:37:25 -0700 | [diff] [blame] | 2136 | tmp = rb_entry(rb_node, struct vm_area_struct, vm_rb); |
Rajman Mekaco | 841e31e | 2012-05-29 15:06:21 -0700 | [diff] [blame] | 2137 | |
Davidlohr Bueso | 615d6e8 | 2014-04-07 15:37:25 -0700 | [diff] [blame] | 2138 | if (tmp->vm_end > addr) { |
| 2139 | vma = tmp; |
| 2140 | if (tmp->vm_start <= addr) |
| 2141 | break; |
| 2142 | rb_node = rb_node->rb_left; |
| 2143 | } else |
| 2144 | rb_node = rb_node->rb_right; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2145 | } |
Davidlohr Bueso | 615d6e8 | 2014-04-07 15:37:25 -0700 | [diff] [blame] | 2146 | |
| 2147 | if (vma) |
| 2148 | vmacache_update(addr, vma); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2149 | return vma; |
| 2150 | } |
| 2151 | |
| 2152 | EXPORT_SYMBOL(find_vma); |
| 2153 | |
KOSAKI Motohiro | 6bd4837 | 2012-01-10 15:08:07 -0800 | [diff] [blame] | 2154 | /* |
| 2155 | * Same as find_vma, but also return a pointer to the previous VMA in *pprev. |
KOSAKI Motohiro | 6bd4837 | 2012-01-10 15:08:07 -0800 | [diff] [blame] | 2156 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2157 | struct vm_area_struct * |
| 2158 | find_vma_prev(struct mm_struct *mm, unsigned long addr, |
| 2159 | struct vm_area_struct **pprev) |
| 2160 | { |
KOSAKI Motohiro | 6bd4837 | 2012-01-10 15:08:07 -0800 | [diff] [blame] | 2161 | struct vm_area_struct *vma; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2162 | |
KOSAKI Motohiro | 6bd4837 | 2012-01-10 15:08:07 -0800 | [diff] [blame] | 2163 | vma = find_vma(mm, addr); |
Mikulas Patocka | 83cd904 | 2012-03-04 19:52:03 -0500 | [diff] [blame] | 2164 | if (vma) { |
| 2165 | *pprev = vma->vm_prev; |
| 2166 | } else { |
| 2167 | struct rb_node *rb_node = mm->mm_rb.rb_node; |
| 2168 | *pprev = NULL; |
| 2169 | while (rb_node) { |
| 2170 | *pprev = rb_entry(rb_node, struct vm_area_struct, vm_rb); |
| 2171 | rb_node = rb_node->rb_right; |
| 2172 | } |
| 2173 | } |
KOSAKI Motohiro | 6bd4837 | 2012-01-10 15:08:07 -0800 | [diff] [blame] | 2174 | return vma; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2175 | } |
| 2176 | |
| 2177 | /* |
| 2178 | * Verify that the stack growth is acceptable and |
| 2179 | * update accounting. This is shared with both the |
| 2180 | * grow-up and grow-down cases. |
| 2181 | */ |
Hugh Dickins | cfc0eb40 | 2017-06-19 04:03:24 -0700 | [diff] [blame] | 2182 | static int acct_stack_growth(struct vm_area_struct *vma, |
| 2183 | unsigned long size, unsigned long grow) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2184 | { |
| 2185 | struct mm_struct *mm = vma->vm_mm; |
| 2186 | struct rlimit *rlim = current->signal->rlim; |
Hugh Dickins | cfc0eb40 | 2017-06-19 04:03:24 -0700 | [diff] [blame] | 2187 | unsigned long new_start; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2188 | |
| 2189 | /* address space limit tests */ |
Konstantin Khlebnikov | 8463833 | 2016-01-14 15:22:07 -0800 | [diff] [blame] | 2190 | if (!may_expand_vm(mm, vma->vm_flags, grow)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2191 | return -ENOMEM; |
| 2192 | |
| 2193 | /* Stack limit test */ |
Hugh Dickins | cfc0eb40 | 2017-06-19 04:03:24 -0700 | [diff] [blame] | 2194 | if (size > READ_ONCE(rlim[RLIMIT_STACK].rlim_cur)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2195 | return -ENOMEM; |
| 2196 | |
| 2197 | /* mlock limit tests */ |
| 2198 | if (vma->vm_flags & VM_LOCKED) { |
| 2199 | unsigned long locked; |
| 2200 | unsigned long limit; |
| 2201 | locked = mm->locked_vm + grow; |
Jason Low | 4db0c3c | 2015-04-15 16:14:08 -0700 | [diff] [blame] | 2202 | limit = READ_ONCE(rlim[RLIMIT_MEMLOCK].rlim_cur); |
Jiri Slaby | 59e99e5 | 2010-03-05 13:41:44 -0800 | [diff] [blame] | 2203 | limit >>= PAGE_SHIFT; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2204 | if (locked > limit && !capable(CAP_IPC_LOCK)) |
| 2205 | return -ENOMEM; |
| 2206 | } |
| 2207 | |
Adam Litke | 0d59a01 | 2007-01-30 14:35:39 -0800 | [diff] [blame] | 2208 | /* Check to ensure the stack will not grow into a hugetlb-only region */ |
| 2209 | new_start = (vma->vm_flags & VM_GROWSUP) ? vma->vm_start : |
| 2210 | vma->vm_end - size; |
| 2211 | if (is_hugepage_only_range(vma->vm_mm, new_start, size)) |
| 2212 | return -EFAULT; |
| 2213 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2214 | /* |
| 2215 | * Overcommit.. This must be the final test, as it will |
| 2216 | * update security statistics. |
| 2217 | */ |
Hugh Dickins | 05fa199 | 2009-04-16 21:58:12 +0100 | [diff] [blame] | 2218 | if (security_vm_enough_memory_mm(mm, grow)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2219 | return -ENOMEM; |
| 2220 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2221 | return 0; |
| 2222 | } |
| 2223 | |
Hugh Dickins | 46dea3d | 2005-10-29 18:16:20 -0700 | [diff] [blame] | 2224 | #if defined(CONFIG_STACK_GROWSUP) || defined(CONFIG_IA64) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2225 | /* |
Hugh Dickins | 46dea3d | 2005-10-29 18:16:20 -0700 | [diff] [blame] | 2226 | * PA-RISC uses this for its stack; IA64 for its Register Backing Store. |
| 2227 | * vma is the last one with address > vma->vm_end. Have to extend vma. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2228 | */ |
Hugh Dickins | 46dea3d | 2005-10-29 18:16:20 -0700 | [diff] [blame] | 2229 | int expand_upwards(struct vm_area_struct *vma, unsigned long address) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2230 | { |
Oleg Nesterov | 0935781 | 2015-11-05 18:48:17 -0800 | [diff] [blame] | 2231 | struct mm_struct *mm = vma->vm_mm; |
Hugh Dickins | cfc0eb40 | 2017-06-19 04:03:24 -0700 | [diff] [blame] | 2232 | struct vm_area_struct *next; |
| 2233 | unsigned long gap_addr; |
Konstantin Khlebnikov | 12352d3 | 2016-02-05 15:36:50 -0800 | [diff] [blame] | 2234 | int error = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2235 | |
| 2236 | if (!(vma->vm_flags & VM_GROWSUP)) |
| 2237 | return -EFAULT; |
| 2238 | |
Helge Deller | 5d10ad62 | 2017-06-19 17:34:05 +0200 | [diff] [blame] | 2239 | /* Guard against exceeding limits of the address space. */ |
Hugh Dickins | cfc0eb40 | 2017-06-19 04:03:24 -0700 | [diff] [blame] | 2240 | address &= PAGE_MASK; |
Helge Deller | 38dfd2e | 2017-07-14 14:49:38 -0700 | [diff] [blame] | 2241 | if (address >= (TASK_SIZE & PAGE_MASK)) |
Konstantin Khlebnikov | 12352d3 | 2016-02-05 15:36:50 -0800 | [diff] [blame] | 2242 | return -ENOMEM; |
Helge Deller | 5d10ad62 | 2017-06-19 17:34:05 +0200 | [diff] [blame] | 2243 | address += PAGE_SIZE; |
Konstantin Khlebnikov | 12352d3 | 2016-02-05 15:36:50 -0800 | [diff] [blame] | 2244 | |
Hugh Dickins | cfc0eb40 | 2017-06-19 04:03:24 -0700 | [diff] [blame] | 2245 | /* Enforce stack_guard_gap */ |
| 2246 | gap_addr = address + stack_guard_gap; |
Helge Deller | 5d10ad62 | 2017-06-19 17:34:05 +0200 | [diff] [blame] | 2247 | |
| 2248 | /* Guard against overflow */ |
| 2249 | if (gap_addr < address || gap_addr > TASK_SIZE) |
| 2250 | gap_addr = TASK_SIZE; |
| 2251 | |
Hugh Dickins | cfc0eb40 | 2017-06-19 04:03:24 -0700 | [diff] [blame] | 2252 | next = vma->vm_next; |
| 2253 | if (next && next->vm_start < gap_addr) { |
| 2254 | if (!(next->vm_flags & VM_GROWSUP)) |
| 2255 | return -ENOMEM; |
| 2256 | /* Check that both stack segments have the same anon_vma? */ |
| 2257 | } |
Konstantin Khlebnikov | 12352d3 | 2016-02-05 15:36:50 -0800 | [diff] [blame] | 2258 | |
| 2259 | /* We must make sure the anon_vma is allocated. */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2260 | if (unlikely(anon_vma_prepare(vma))) |
| 2261 | return -ENOMEM; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2262 | |
| 2263 | /* |
| 2264 | * vma->vm_start/vm_end cannot change under us because the caller |
| 2265 | * is required to hold the mmap_sem in read mode. We need the |
| 2266 | * anon_vma lock to serialize against concurrent expand_stacks. |
| 2267 | */ |
Konstantin Khlebnikov | 12352d3 | 2016-02-05 15:36:50 -0800 | [diff] [blame] | 2268 | anon_vma_lock_write(vma->anon_vma); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2269 | |
| 2270 | /* Somebody else might have raced and expanded it already */ |
| 2271 | if (address > vma->vm_end) { |
| 2272 | unsigned long size, grow; |
| 2273 | |
| 2274 | size = address - vma->vm_start; |
| 2275 | grow = (address - vma->vm_end) >> PAGE_SHIFT; |
| 2276 | |
Hugh Dickins | 42c36f6 | 2011-05-09 17:44:42 -0700 | [diff] [blame] | 2277 | error = -ENOMEM; |
| 2278 | if (vma->vm_pgoff + (size >> PAGE_SHIFT) >= vma->vm_pgoff) { |
| 2279 | error = acct_stack_growth(vma, size, grow); |
| 2280 | if (!error) { |
Michel Lespinasse | 4128997 | 2012-12-12 13:52:25 -0800 | [diff] [blame] | 2281 | /* |
| 2282 | * vma_gap_update() doesn't support concurrent |
| 2283 | * updates, but we only hold a shared mmap_sem |
| 2284 | * lock here, so we need to protect against |
| 2285 | * concurrent vma expansions. |
Konstantin Khlebnikov | 12352d3 | 2016-02-05 15:36:50 -0800 | [diff] [blame] | 2286 | * anon_vma_lock_write() doesn't help here, as |
Michel Lespinasse | 4128997 | 2012-12-12 13:52:25 -0800 | [diff] [blame] | 2287 | * we don't guarantee that all growable vmas |
| 2288 | * in a mm share the same root anon vma. |
| 2289 | * So, we reuse mm->page_table_lock to guard |
| 2290 | * against concurrent vma expansions. |
| 2291 | */ |
Oleg Nesterov | 0935781 | 2015-11-05 18:48:17 -0800 | [diff] [blame] | 2292 | spin_lock(&mm->page_table_lock); |
Oleg Nesterov | 87e8827 | 2015-11-05 18:48:14 -0800 | [diff] [blame] | 2293 | if (vma->vm_flags & VM_LOCKED) |
Oleg Nesterov | 0935781 | 2015-11-05 18:48:17 -0800 | [diff] [blame] | 2294 | mm->locked_vm += grow; |
Konstantin Khlebnikov | 8463833 | 2016-01-14 15:22:07 -0800 | [diff] [blame] | 2295 | vm_stat_account(mm, vma->vm_flags, grow); |
Michel Lespinasse | bf181b9 | 2012-10-08 16:31:39 -0700 | [diff] [blame] | 2296 | anon_vma_interval_tree_pre_update_vma(vma); |
Hugh Dickins | 42c36f6 | 2011-05-09 17:44:42 -0700 | [diff] [blame] | 2297 | vma->vm_end = address; |
Michel Lespinasse | bf181b9 | 2012-10-08 16:31:39 -0700 | [diff] [blame] | 2298 | anon_vma_interval_tree_post_update_vma(vma); |
Michel Lespinasse | d373718 | 2012-12-11 16:01:38 -0800 | [diff] [blame] | 2299 | if (vma->vm_next) |
| 2300 | vma_gap_update(vma->vm_next); |
| 2301 | else |
Hugh Dickins | cfc0eb40 | 2017-06-19 04:03:24 -0700 | [diff] [blame] | 2302 | mm->highest_vm_end = vm_end_gap(vma); |
Oleg Nesterov | 0935781 | 2015-11-05 18:48:17 -0800 | [diff] [blame] | 2303 | spin_unlock(&mm->page_table_lock); |
Michel Lespinasse | 4128997 | 2012-12-12 13:52:25 -0800 | [diff] [blame] | 2304 | |
Hugh Dickins | 42c36f6 | 2011-05-09 17:44:42 -0700 | [diff] [blame] | 2305 | perf_event_mmap(vma); |
| 2306 | } |
Eric B Munson | 3af9e85 | 2010-05-18 15:30:49 +0100 | [diff] [blame] | 2307 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2308 | } |
Konstantin Khlebnikov | 12352d3 | 2016-02-05 15:36:50 -0800 | [diff] [blame] | 2309 | anon_vma_unlock_write(vma->anon_vma); |
David Rientjes | 6d50e60 | 2014-10-29 14:50:31 -0700 | [diff] [blame] | 2310 | khugepaged_enter_vma_merge(vma, vma->vm_flags); |
Oleg Nesterov | 0935781 | 2015-11-05 18:48:17 -0800 | [diff] [blame] | 2311 | validate_mm(mm); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2312 | return error; |
| 2313 | } |
Hugh Dickins | 46dea3d | 2005-10-29 18:16:20 -0700 | [diff] [blame] | 2314 | #endif /* CONFIG_STACK_GROWSUP || CONFIG_IA64 */ |
| 2315 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2316 | /* |
| 2317 | * vma is the first one with address < vma->vm_start. Have to extend vma. |
| 2318 | */ |
Michal Hocko | d05f316 | 2011-05-24 17:11:44 -0700 | [diff] [blame] | 2319 | int expand_downwards(struct vm_area_struct *vma, |
Ollie Wild | b6a2fea | 2007-07-19 01:48:16 -0700 | [diff] [blame] | 2320 | unsigned long address) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2321 | { |
Oleg Nesterov | 0935781 | 2015-11-05 18:48:17 -0800 | [diff] [blame] | 2322 | struct mm_struct *mm = vma->vm_mm; |
Hugh Dickins | cfc0eb40 | 2017-06-19 04:03:24 -0700 | [diff] [blame] | 2323 | struct vm_area_struct *prev; |
| 2324 | unsigned long gap_addr; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2325 | int error; |
| 2326 | |
Eric Paris | 8869477 | 2007-11-26 18:47:26 -0500 | [diff] [blame] | 2327 | address &= PAGE_MASK; |
Al Viro | e546785 | 2012-05-30 13:30:51 -0400 | [diff] [blame] | 2328 | error = security_mmap_addr(address); |
Eric Paris | 8869477 | 2007-11-26 18:47:26 -0500 | [diff] [blame] | 2329 | if (error) |
| 2330 | return error; |
| 2331 | |
Hugh Dickins | cfc0eb40 | 2017-06-19 04:03:24 -0700 | [diff] [blame] | 2332 | /* Enforce stack_guard_gap */ |
| 2333 | gap_addr = address - stack_guard_gap; |
| 2334 | if (gap_addr > address) |
| 2335 | return -ENOMEM; |
| 2336 | prev = vma->vm_prev; |
| 2337 | if (prev && prev->vm_end > gap_addr) { |
| 2338 | if (!(prev->vm_flags & VM_GROWSDOWN)) |
| 2339 | return -ENOMEM; |
| 2340 | /* Check that both stack segments have the same anon_vma? */ |
| 2341 | } |
| 2342 | |
Konstantin Khlebnikov | 12352d3 | 2016-02-05 15:36:50 -0800 | [diff] [blame] | 2343 | /* We must make sure the anon_vma is allocated. */ |
| 2344 | if (unlikely(anon_vma_prepare(vma))) |
| 2345 | return -ENOMEM; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2346 | |
| 2347 | /* |
| 2348 | * vma->vm_start/vm_end cannot change under us because the caller |
| 2349 | * is required to hold the mmap_sem in read mode. We need the |
| 2350 | * anon_vma lock to serialize against concurrent expand_stacks. |
| 2351 | */ |
Konstantin Khlebnikov | 12352d3 | 2016-02-05 15:36:50 -0800 | [diff] [blame] | 2352 | anon_vma_lock_write(vma->anon_vma); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2353 | |
| 2354 | /* Somebody else might have raced and expanded it already */ |
| 2355 | if (address < vma->vm_start) { |
| 2356 | unsigned long size, grow; |
| 2357 | |
| 2358 | size = vma->vm_end - address; |
| 2359 | grow = (vma->vm_start - address) >> PAGE_SHIFT; |
| 2360 | |
Linus Torvalds | a626ca6 | 2011-04-13 08:07:28 -0700 | [diff] [blame] | 2361 | error = -ENOMEM; |
| 2362 | if (grow <= vma->vm_pgoff) { |
| 2363 | error = acct_stack_growth(vma, size, grow); |
| 2364 | if (!error) { |
Michel Lespinasse | 4128997 | 2012-12-12 13:52:25 -0800 | [diff] [blame] | 2365 | /* |
| 2366 | * vma_gap_update() doesn't support concurrent |
| 2367 | * updates, but we only hold a shared mmap_sem |
| 2368 | * lock here, so we need to protect against |
| 2369 | * concurrent vma expansions. |
Konstantin Khlebnikov | 12352d3 | 2016-02-05 15:36:50 -0800 | [diff] [blame] | 2370 | * anon_vma_lock_write() doesn't help here, as |
Michel Lespinasse | 4128997 | 2012-12-12 13:52:25 -0800 | [diff] [blame] | 2371 | * we don't guarantee that all growable vmas |
| 2372 | * in a mm share the same root anon vma. |
| 2373 | * So, we reuse mm->page_table_lock to guard |
| 2374 | * against concurrent vma expansions. |
| 2375 | */ |
Oleg Nesterov | 0935781 | 2015-11-05 18:48:17 -0800 | [diff] [blame] | 2376 | spin_lock(&mm->page_table_lock); |
Oleg Nesterov | 87e8827 | 2015-11-05 18:48:14 -0800 | [diff] [blame] | 2377 | if (vma->vm_flags & VM_LOCKED) |
Oleg Nesterov | 0935781 | 2015-11-05 18:48:17 -0800 | [diff] [blame] | 2378 | mm->locked_vm += grow; |
Konstantin Khlebnikov | 8463833 | 2016-01-14 15:22:07 -0800 | [diff] [blame] | 2379 | vm_stat_account(mm, vma->vm_flags, grow); |
Michel Lespinasse | bf181b9 | 2012-10-08 16:31:39 -0700 | [diff] [blame] | 2380 | anon_vma_interval_tree_pre_update_vma(vma); |
Linus Torvalds | a626ca6 | 2011-04-13 08:07:28 -0700 | [diff] [blame] | 2381 | vma->vm_start = address; |
| 2382 | vma->vm_pgoff -= grow; |
Michel Lespinasse | bf181b9 | 2012-10-08 16:31:39 -0700 | [diff] [blame] | 2383 | anon_vma_interval_tree_post_update_vma(vma); |
Michel Lespinasse | d373718 | 2012-12-11 16:01:38 -0800 | [diff] [blame] | 2384 | vma_gap_update(vma); |
Oleg Nesterov | 0935781 | 2015-11-05 18:48:17 -0800 | [diff] [blame] | 2385 | spin_unlock(&mm->page_table_lock); |
Michel Lespinasse | 4128997 | 2012-12-12 13:52:25 -0800 | [diff] [blame] | 2386 | |
Linus Torvalds | a626ca6 | 2011-04-13 08:07:28 -0700 | [diff] [blame] | 2387 | perf_event_mmap(vma); |
| 2388 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2389 | } |
| 2390 | } |
Konstantin Khlebnikov | 12352d3 | 2016-02-05 15:36:50 -0800 | [diff] [blame] | 2391 | anon_vma_unlock_write(vma->anon_vma); |
David Rientjes | 6d50e60 | 2014-10-29 14:50:31 -0700 | [diff] [blame] | 2392 | khugepaged_enter_vma_merge(vma, vma->vm_flags); |
Oleg Nesterov | 0935781 | 2015-11-05 18:48:17 -0800 | [diff] [blame] | 2393 | validate_mm(mm); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2394 | return error; |
| 2395 | } |
| 2396 | |
Hugh Dickins | cfc0eb40 | 2017-06-19 04:03:24 -0700 | [diff] [blame] | 2397 | /* enforced gap between the expanding stack and other mappings. */ |
| 2398 | unsigned long stack_guard_gap = 256UL<<PAGE_SHIFT; |
| 2399 | |
| 2400 | static int __init cmdline_parse_stack_guard_gap(char *p) |
| 2401 | { |
| 2402 | unsigned long val; |
| 2403 | char *endptr; |
| 2404 | |
| 2405 | val = simple_strtoul(p, &endptr, 10); |
| 2406 | if (!*endptr) |
| 2407 | stack_guard_gap = val << PAGE_SHIFT; |
| 2408 | |
| 2409 | return 0; |
| 2410 | } |
| 2411 | __setup("stack_guard_gap=", cmdline_parse_stack_guard_gap); |
| 2412 | |
Ollie Wild | b6a2fea | 2007-07-19 01:48:16 -0700 | [diff] [blame] | 2413 | #ifdef CONFIG_STACK_GROWSUP |
| 2414 | int expand_stack(struct vm_area_struct *vma, unsigned long address) |
| 2415 | { |
| 2416 | return expand_upwards(vma, address); |
| 2417 | } |
| 2418 | |
| 2419 | struct vm_area_struct * |
| 2420 | find_extend_vma(struct mm_struct *mm, unsigned long addr) |
| 2421 | { |
| 2422 | struct vm_area_struct *vma, *prev; |
| 2423 | |
| 2424 | addr &= PAGE_MASK; |
| 2425 | vma = find_vma_prev(mm, addr, &prev); |
| 2426 | if (vma && (vma->vm_start <= addr)) |
| 2427 | return vma; |
Denys Vlasenko | 1c12718 | 2008-11-12 01:24:41 +0100 | [diff] [blame] | 2428 | if (!prev || expand_stack(prev, addr)) |
Ollie Wild | b6a2fea | 2007-07-19 01:48:16 -0700 | [diff] [blame] | 2429 | return NULL; |
Michel Lespinasse | cea10a1 | 2013-02-22 16:32:44 -0800 | [diff] [blame] | 2430 | if (prev->vm_flags & VM_LOCKED) |
Kirill A. Shutemov | fc05f56 | 2015-04-14 15:44:39 -0700 | [diff] [blame] | 2431 | populate_vma_page_range(prev, addr, prev->vm_end, NULL); |
Ollie Wild | b6a2fea | 2007-07-19 01:48:16 -0700 | [diff] [blame] | 2432 | return prev; |
| 2433 | } |
| 2434 | #else |
| 2435 | int expand_stack(struct vm_area_struct *vma, unsigned long address) |
| 2436 | { |
| 2437 | return expand_downwards(vma, address); |
| 2438 | } |
| 2439 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2440 | struct vm_area_struct * |
vishnu.ps | cc71aba | 2014-10-09 15:26:29 -0700 | [diff] [blame] | 2441 | find_extend_vma(struct mm_struct *mm, unsigned long addr) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2442 | { |
vishnu.ps | cc71aba | 2014-10-09 15:26:29 -0700 | [diff] [blame] | 2443 | struct vm_area_struct *vma; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2444 | unsigned long start; |
| 2445 | |
| 2446 | addr &= PAGE_MASK; |
vishnu.ps | cc71aba | 2014-10-09 15:26:29 -0700 | [diff] [blame] | 2447 | vma = find_vma(mm, addr); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2448 | if (!vma) |
| 2449 | return NULL; |
| 2450 | if (vma->vm_start <= addr) |
| 2451 | return vma; |
| 2452 | if (!(vma->vm_flags & VM_GROWSDOWN)) |
| 2453 | return NULL; |
| 2454 | start = vma->vm_start; |
| 2455 | if (expand_stack(vma, addr)) |
| 2456 | return NULL; |
Michel Lespinasse | cea10a1 | 2013-02-22 16:32:44 -0800 | [diff] [blame] | 2457 | if (vma->vm_flags & VM_LOCKED) |
Kirill A. Shutemov | fc05f56 | 2015-04-14 15:44:39 -0700 | [diff] [blame] | 2458 | populate_vma_page_range(vma, addr, start, NULL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2459 | return vma; |
| 2460 | } |
| 2461 | #endif |
| 2462 | |
Jesse Barnes | e1d6d01 | 2014-12-12 16:55:27 -0800 | [diff] [blame] | 2463 | EXPORT_SYMBOL_GPL(find_extend_vma); |
| 2464 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2465 | /* |
Hugh Dickins | 2c0b381 | 2005-10-29 18:15:56 -0700 | [diff] [blame] | 2466 | * Ok - we have the memory areas we should free on the vma list, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2467 | * so release them, and do the vma updates. |
Hugh Dickins | 2c0b381 | 2005-10-29 18:15:56 -0700 | [diff] [blame] | 2468 | * |
| 2469 | * Called with the mm semaphore held. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2470 | */ |
Hugh Dickins | 2c0b381 | 2005-10-29 18:15:56 -0700 | [diff] [blame] | 2471 | static void remove_vma_list(struct mm_struct *mm, struct vm_area_struct *vma) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2472 | { |
Linus Torvalds | 4f74d2c | 2012-05-06 13:54:06 -0700 | [diff] [blame] | 2473 | unsigned long nr_accounted = 0; |
| 2474 | |
Hugh Dickins | 365e9c87 | 2005-10-29 18:16:18 -0700 | [diff] [blame] | 2475 | /* Update high watermark before we lower total_vm */ |
| 2476 | update_hiwater_vm(mm); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2477 | do { |
Hugh Dickins | 2c0b381 | 2005-10-29 18:15:56 -0700 | [diff] [blame] | 2478 | long nrpages = vma_pages(vma); |
| 2479 | |
Linus Torvalds | 4f74d2c | 2012-05-06 13:54:06 -0700 | [diff] [blame] | 2480 | if (vma->vm_flags & VM_ACCOUNT) |
| 2481 | nr_accounted += nrpages; |
Konstantin Khlebnikov | 8463833 | 2016-01-14 15:22:07 -0800 | [diff] [blame] | 2482 | vm_stat_account(mm, vma->vm_flags, -nrpages); |
Hugh Dickins | a8fb561 | 2005-10-29 18:15:57 -0700 | [diff] [blame] | 2483 | vma = remove_vma(vma); |
Hugh Dickins | 146425a | 2005-04-19 13:29:18 -0700 | [diff] [blame] | 2484 | } while (vma); |
Linus Torvalds | 4f74d2c | 2012-05-06 13:54:06 -0700 | [diff] [blame] | 2485 | vm_unacct_memory(nr_accounted); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2486 | validate_mm(mm); |
| 2487 | } |
| 2488 | |
| 2489 | /* |
| 2490 | * Get rid of page table information in the indicated region. |
| 2491 | * |
Paolo 'Blaisorblade' Giarrusso | f10df68 | 2005-09-21 09:55:37 -0700 | [diff] [blame] | 2492 | * Called with the mm semaphore held. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2493 | */ |
| 2494 | static void unmap_region(struct mm_struct *mm, |
Hugh Dickins | e0da382 | 2005-04-19 13:29:15 -0700 | [diff] [blame] | 2495 | struct vm_area_struct *vma, struct vm_area_struct *prev, |
| 2496 | unsigned long start, unsigned long end) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2497 | { |
vishnu.ps | cc71aba | 2014-10-09 15:26:29 -0700 | [diff] [blame] | 2498 | struct vm_area_struct *next = prev ? prev->vm_next : mm->mmap; |
Peter Zijlstra | d16dfc5 | 2011-05-24 17:11:45 -0700 | [diff] [blame] | 2499 | struct mmu_gather tlb; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2500 | |
| 2501 | lru_add_drain(); |
Linus Torvalds | 2b04725 | 2013-08-15 11:42:25 -0700 | [diff] [blame] | 2502 | tlb_gather_mmu(&tlb, mm, start, end); |
Hugh Dickins | 365e9c87 | 2005-10-29 18:16:18 -0700 | [diff] [blame] | 2503 | update_hiwater_rss(mm); |
Linus Torvalds | 4f74d2c | 2012-05-06 13:54:06 -0700 | [diff] [blame] | 2504 | unmap_vmas(&tlb, vma, start, end); |
Peter Zijlstra | d16dfc5 | 2011-05-24 17:11:45 -0700 | [diff] [blame] | 2505 | free_pgtables(&tlb, vma, prev ? prev->vm_end : FIRST_USER_ADDRESS, |
Hugh Dickins | 6ee8630 | 2013-04-29 15:07:44 -0700 | [diff] [blame] | 2506 | next ? next->vm_start : USER_PGTABLES_CEILING); |
Peter Zijlstra | d16dfc5 | 2011-05-24 17:11:45 -0700 | [diff] [blame] | 2507 | tlb_finish_mmu(&tlb, start, end); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2508 | } |
| 2509 | |
| 2510 | /* |
| 2511 | * Create a list of vma's touched by the unmap, removing them from the mm's |
| 2512 | * vma list as we go.. |
| 2513 | */ |
| 2514 | static void |
| 2515 | detach_vmas_to_be_unmapped(struct mm_struct *mm, struct vm_area_struct *vma, |
| 2516 | struct vm_area_struct *prev, unsigned long end) |
| 2517 | { |
| 2518 | struct vm_area_struct **insertion_point; |
| 2519 | struct vm_area_struct *tail_vma = NULL; |
| 2520 | |
| 2521 | insertion_point = (prev ? &prev->vm_next : &mm->mmap); |
Linus Torvalds | 297c5ee | 2010-08-20 16:24:55 -0700 | [diff] [blame] | 2522 | vma->vm_prev = NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2523 | do { |
Michel Lespinasse | d373718 | 2012-12-11 16:01:38 -0800 | [diff] [blame] | 2524 | vma_rb_erase(vma, &mm->mm_rb); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2525 | mm->map_count--; |
| 2526 | tail_vma = vma; |
| 2527 | vma = vma->vm_next; |
| 2528 | } while (vma && vma->vm_start < end); |
| 2529 | *insertion_point = vma; |
Michel Lespinasse | d373718 | 2012-12-11 16:01:38 -0800 | [diff] [blame] | 2530 | if (vma) { |
Linus Torvalds | 297c5ee | 2010-08-20 16:24:55 -0700 | [diff] [blame] | 2531 | vma->vm_prev = prev; |
Michel Lespinasse | d373718 | 2012-12-11 16:01:38 -0800 | [diff] [blame] | 2532 | vma_gap_update(vma); |
| 2533 | } else |
Hugh Dickins | cfc0eb40 | 2017-06-19 04:03:24 -0700 | [diff] [blame] | 2534 | mm->highest_vm_end = prev ? vm_end_gap(prev) : 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2535 | tail_vma->vm_next = NULL; |
Davidlohr Bueso | 615d6e8 | 2014-04-07 15:37:25 -0700 | [diff] [blame] | 2536 | |
| 2537 | /* Kill the cache */ |
| 2538 | vmacache_invalidate(mm); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2539 | } |
| 2540 | |
| 2541 | /* |
KOSAKI Motohiro | 659ace5 | 2009-12-14 17:57:56 -0800 | [diff] [blame] | 2542 | * __split_vma() bypasses sysctl_max_map_count checking. We use this on the |
| 2543 | * munmap path where it doesn't make sense to fail. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2544 | */ |
vishnu.ps | cc71aba | 2014-10-09 15:26:29 -0700 | [diff] [blame] | 2545 | static int __split_vma(struct mm_struct *mm, struct vm_area_struct *vma, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2546 | unsigned long addr, int new_below) |
| 2547 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2548 | struct vm_area_struct *new; |
Chen Gang | e397589 | 2015-09-08 15:03:38 -0700 | [diff] [blame] | 2549 | int err; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2550 | |
Andi Kleen | a551643 | 2008-07-23 21:27:41 -0700 | [diff] [blame] | 2551 | if (is_vm_hugetlb_page(vma) && (addr & |
| 2552 | ~(huge_page_mask(hstate_vma(vma))))) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2553 | return -EINVAL; |
| 2554 | |
Christoph Lameter | e94b176 | 2006-12-06 20:33:17 -0800 | [diff] [blame] | 2555 | new = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2556 | if (!new) |
Chen Gang | e397589 | 2015-09-08 15:03:38 -0700 | [diff] [blame] | 2557 | return -ENOMEM; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2558 | |
| 2559 | /* most fields are the same, copy all, and then fixup */ |
| 2560 | *new = *vma; |
| 2561 | |
Rik van Riel | 5beb493 | 2010-03-05 13:42:07 -0800 | [diff] [blame] | 2562 | INIT_LIST_HEAD(&new->anon_vma_chain); |
| 2563 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2564 | if (new_below) |
| 2565 | new->vm_end = addr; |
| 2566 | else { |
| 2567 | new->vm_start = addr; |
| 2568 | new->vm_pgoff += ((addr - vma->vm_start) >> PAGE_SHIFT); |
| 2569 | } |
| 2570 | |
Oleg Nesterov | ef0855d | 2013-09-11 14:20:14 -0700 | [diff] [blame] | 2571 | err = vma_dup_policy(vma, new); |
| 2572 | if (err) |
Rik van Riel | 5beb493 | 2010-03-05 13:42:07 -0800 | [diff] [blame] | 2573 | goto out_free_vma; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2574 | |
Daniel Forrest | c4ea95d | 2014-12-02 15:59:42 -0800 | [diff] [blame] | 2575 | err = anon_vma_clone(new, vma); |
| 2576 | if (err) |
Rik van Riel | 5beb493 | 2010-03-05 13:42:07 -0800 | [diff] [blame] | 2577 | goto out_free_mpol; |
| 2578 | |
Konstantin Khlebnikov | e9714ac | 2012-10-08 16:28:54 -0700 | [diff] [blame] | 2579 | if (new->vm_file) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2580 | get_file(new->vm_file); |
| 2581 | |
| 2582 | if (new->vm_ops && new->vm_ops->open) |
| 2583 | new->vm_ops->open(new); |
| 2584 | |
| 2585 | if (new_below) |
Rik van Riel | 5beb493 | 2010-03-05 13:42:07 -0800 | [diff] [blame] | 2586 | err = vma_adjust(vma, addr, vma->vm_end, vma->vm_pgoff + |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2587 | ((addr - new->vm_start) >> PAGE_SHIFT), new); |
| 2588 | else |
Rik van Riel | 5beb493 | 2010-03-05 13:42:07 -0800 | [diff] [blame] | 2589 | err = vma_adjust(vma, vma->vm_start, addr, vma->vm_pgoff, new); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2590 | |
Rik van Riel | 5beb493 | 2010-03-05 13:42:07 -0800 | [diff] [blame] | 2591 | /* Success. */ |
| 2592 | if (!err) |
| 2593 | return 0; |
| 2594 | |
| 2595 | /* Clean everything up if vma_adjust failed. */ |
Rik van Riel | 5892753 | 2010-04-26 12:33:03 -0400 | [diff] [blame] | 2596 | if (new->vm_ops && new->vm_ops->close) |
| 2597 | new->vm_ops->close(new); |
Konstantin Khlebnikov | e9714ac | 2012-10-08 16:28:54 -0700 | [diff] [blame] | 2598 | if (new->vm_file) |
Rik van Riel | 5beb493 | 2010-03-05 13:42:07 -0800 | [diff] [blame] | 2599 | fput(new->vm_file); |
Andrea Arcangeli | 2aeadc3 | 2010-09-22 13:05:12 -0700 | [diff] [blame] | 2600 | unlink_anon_vmas(new); |
Rik van Riel | 5beb493 | 2010-03-05 13:42:07 -0800 | [diff] [blame] | 2601 | out_free_mpol: |
Oleg Nesterov | ef0855d | 2013-09-11 14:20:14 -0700 | [diff] [blame] | 2602 | mpol_put(vma_policy(new)); |
Rik van Riel | 5beb493 | 2010-03-05 13:42:07 -0800 | [diff] [blame] | 2603 | out_free_vma: |
| 2604 | kmem_cache_free(vm_area_cachep, new); |
Rik van Riel | 5beb493 | 2010-03-05 13:42:07 -0800 | [diff] [blame] | 2605 | return err; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2606 | } |
| 2607 | |
KOSAKI Motohiro | 659ace5 | 2009-12-14 17:57:56 -0800 | [diff] [blame] | 2608 | /* |
| 2609 | * Split a vma into two pieces at address 'addr', a new vma is allocated |
| 2610 | * either for the first part or the tail. |
| 2611 | */ |
| 2612 | int split_vma(struct mm_struct *mm, struct vm_area_struct *vma, |
| 2613 | unsigned long addr, int new_below) |
| 2614 | { |
| 2615 | if (mm->map_count >= sysctl_max_map_count) |
| 2616 | return -ENOMEM; |
| 2617 | |
| 2618 | return __split_vma(mm, vma, addr, new_below); |
| 2619 | } |
| 2620 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2621 | /* Munmap is split into 2 main parts -- this part which finds |
| 2622 | * what needs doing, and the areas themselves, which do the |
| 2623 | * work. This now handles partial unmappings. |
| 2624 | * Jeremy Fitzhardinge <jeremy@goop.org> |
| 2625 | */ |
| 2626 | int do_munmap(struct mm_struct *mm, unsigned long start, size_t len) |
| 2627 | { |
| 2628 | unsigned long end; |
Hugh Dickins | 146425a | 2005-04-19 13:29:18 -0700 | [diff] [blame] | 2629 | struct vm_area_struct *vma, *prev, *last; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2630 | |
Alexander Kuleshov | de1741a | 2015-11-05 18:46:54 -0800 | [diff] [blame] | 2631 | if ((offset_in_page(start)) || start > TASK_SIZE || len > TASK_SIZE-start) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2632 | return -EINVAL; |
| 2633 | |
vishnu.ps | cc71aba | 2014-10-09 15:26:29 -0700 | [diff] [blame] | 2634 | len = PAGE_ALIGN(len); |
| 2635 | if (len == 0) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2636 | return -EINVAL; |
| 2637 | |
| 2638 | /* Find the first overlapping VMA */ |
Linus Torvalds | 9be34c9 | 2011-06-16 00:35:09 -0700 | [diff] [blame] | 2639 | vma = find_vma(mm, start); |
Hugh Dickins | 146425a | 2005-04-19 13:29:18 -0700 | [diff] [blame] | 2640 | if (!vma) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2641 | return 0; |
Linus Torvalds | 9be34c9 | 2011-06-16 00:35:09 -0700 | [diff] [blame] | 2642 | prev = vma->vm_prev; |
Hugh Dickins | 146425a | 2005-04-19 13:29:18 -0700 | [diff] [blame] | 2643 | /* we have start < vma->vm_end */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2644 | |
| 2645 | /* if it doesn't overlap, we have nothing.. */ |
| 2646 | end = start + len; |
Hugh Dickins | 146425a | 2005-04-19 13:29:18 -0700 | [diff] [blame] | 2647 | if (vma->vm_start >= end) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2648 | return 0; |
| 2649 | |
| 2650 | /* |
| 2651 | * If we need to split any vma, do it now to save pain later. |
| 2652 | * |
| 2653 | * Note: mremap's move_vma VM_ACCOUNT handling assumes a partially |
| 2654 | * unmapped vm_area_struct will remain in use: so lower split_vma |
| 2655 | * places tmp vma above, and higher split_vma places tmp vma below. |
| 2656 | */ |
Hugh Dickins | 146425a | 2005-04-19 13:29:18 -0700 | [diff] [blame] | 2657 | if (start > vma->vm_start) { |
KOSAKI Motohiro | 659ace5 | 2009-12-14 17:57:56 -0800 | [diff] [blame] | 2658 | int error; |
| 2659 | |
| 2660 | /* |
| 2661 | * Make sure that map_count on return from munmap() will |
| 2662 | * not exceed its limit; but let map_count go just above |
| 2663 | * its limit temporarily, to help free resources as expected. |
| 2664 | */ |
| 2665 | if (end < vma->vm_end && mm->map_count >= sysctl_max_map_count) |
| 2666 | return -ENOMEM; |
| 2667 | |
| 2668 | error = __split_vma(mm, vma, start, 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2669 | if (error) |
| 2670 | return error; |
Hugh Dickins | 146425a | 2005-04-19 13:29:18 -0700 | [diff] [blame] | 2671 | prev = vma; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2672 | } |
| 2673 | |
| 2674 | /* Does it split the last one? */ |
| 2675 | last = find_vma(mm, end); |
| 2676 | if (last && end > last->vm_start) { |
KOSAKI Motohiro | 659ace5 | 2009-12-14 17:57:56 -0800 | [diff] [blame] | 2677 | int error = __split_vma(mm, last, end, 1); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2678 | if (error) |
| 2679 | return error; |
| 2680 | } |
vishnu.ps | cc71aba | 2014-10-09 15:26:29 -0700 | [diff] [blame] | 2681 | vma = prev ? prev->vm_next : mm->mmap; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2682 | |
| 2683 | /* |
Rik van Riel | ba470de | 2008-10-18 20:26:50 -0700 | [diff] [blame] | 2684 | * unlock any mlock()ed ranges before detaching vmas |
| 2685 | */ |
| 2686 | if (mm->locked_vm) { |
| 2687 | struct vm_area_struct *tmp = vma; |
| 2688 | while (tmp && tmp->vm_start < end) { |
| 2689 | if (tmp->vm_flags & VM_LOCKED) { |
| 2690 | mm->locked_vm -= vma_pages(tmp); |
| 2691 | munlock_vma_pages_all(tmp); |
| 2692 | } |
| 2693 | tmp = tmp->vm_next; |
| 2694 | } |
| 2695 | } |
| 2696 | |
| 2697 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2698 | * Remove the vma's, and unmap the actual pages |
| 2699 | */ |
Hugh Dickins | 146425a | 2005-04-19 13:29:18 -0700 | [diff] [blame] | 2700 | detach_vmas_to_be_unmapped(mm, vma, prev, end); |
| 2701 | unmap_region(mm, vma, prev, start, end); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2702 | |
Dave Hansen | 1de4fa1 | 2014-11-14 07:18:31 -0800 | [diff] [blame] | 2703 | arch_unmap(mm, vma, start, end); |
| 2704 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2705 | /* Fix up all other VM information */ |
Hugh Dickins | 2c0b381 | 2005-10-29 18:15:56 -0700 | [diff] [blame] | 2706 | remove_vma_list(mm, vma); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2707 | |
| 2708 | return 0; |
| 2709 | } |
Guenter Roeck | abfb486 | 2016-03-24 10:39:14 -0700 | [diff] [blame] | 2710 | EXPORT_SYMBOL(do_munmap); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2711 | |
Al Viro | bfce281 | 2012-04-20 21:57:04 -0400 | [diff] [blame] | 2712 | int vm_munmap(unsigned long start, size_t len) |
Linus Torvalds | a46ef99 | 2012-04-20 16:20:01 -0700 | [diff] [blame] | 2713 | { |
| 2714 | int ret; |
Al Viro | bfce281 | 2012-04-20 21:57:04 -0400 | [diff] [blame] | 2715 | struct mm_struct *mm = current->mm; |
Linus Torvalds | a46ef99 | 2012-04-20 16:20:01 -0700 | [diff] [blame] | 2716 | |
Michal Hocko | ae79878 | 2016-05-23 16:25:33 -0700 | [diff] [blame] | 2717 | if (down_write_killable(&mm->mmap_sem)) |
| 2718 | return -EINTR; |
| 2719 | |
Linus Torvalds | a46ef99 | 2012-04-20 16:20:01 -0700 | [diff] [blame] | 2720 | ret = do_munmap(mm, start, len); |
| 2721 | up_write(&mm->mmap_sem); |
| 2722 | return ret; |
| 2723 | } |
| 2724 | EXPORT_SYMBOL(vm_munmap); |
| 2725 | |
Heiko Carstens | 6a6160a | 2009-01-14 14:14:15 +0100 | [diff] [blame] | 2726 | SYSCALL_DEFINE2(munmap, unsigned long, addr, size_t, len) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2727 | { |
Michal Hocko | dc0ef0d | 2016-05-23 16:25:27 -0700 | [diff] [blame] | 2728 | int ret; |
| 2729 | struct mm_struct *mm = current->mm; |
| 2730 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2731 | profile_munmap(addr); |
Michal Hocko | dc0ef0d | 2016-05-23 16:25:27 -0700 | [diff] [blame] | 2732 | if (down_write_killable(&mm->mmap_sem)) |
| 2733 | return -EINTR; |
| 2734 | ret = do_munmap(mm, addr, len); |
| 2735 | up_write(&mm->mmap_sem); |
| 2736 | return ret; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2737 | } |
| 2738 | |
Kirill A. Shutemov | c8d78c1 | 2015-02-10 14:09:46 -0800 | [diff] [blame] | 2739 | |
| 2740 | /* |
| 2741 | * Emulation of deprecated remap_file_pages() syscall. |
| 2742 | */ |
| 2743 | SYSCALL_DEFINE5(remap_file_pages, unsigned long, start, unsigned long, size, |
| 2744 | unsigned long, prot, unsigned long, pgoff, unsigned long, flags) |
| 2745 | { |
| 2746 | |
| 2747 | struct mm_struct *mm = current->mm; |
| 2748 | struct vm_area_struct *vma; |
| 2749 | unsigned long populate = 0; |
| 2750 | unsigned long ret = -EINVAL; |
| 2751 | struct file *file; |
| 2752 | |
Joe Perches | 756a025f0 | 2016-03-17 14:19:47 -0700 | [diff] [blame] | 2753 | pr_warn_once("%s (%d) uses deprecated remap_file_pages() syscall. See Documentation/vm/remap_file_pages.txt.\n", |
| 2754 | current->comm, current->pid); |
Kirill A. Shutemov | c8d78c1 | 2015-02-10 14:09:46 -0800 | [diff] [blame] | 2755 | |
| 2756 | if (prot) |
| 2757 | return ret; |
| 2758 | start = start & PAGE_MASK; |
| 2759 | size = size & PAGE_MASK; |
| 2760 | |
| 2761 | if (start + size <= start) |
| 2762 | return ret; |
| 2763 | |
| 2764 | /* Does pgoff wrap? */ |
| 2765 | if (pgoff + (size >> PAGE_SHIFT) < pgoff) |
| 2766 | return ret; |
| 2767 | |
Michal Hocko | dc0ef0d | 2016-05-23 16:25:27 -0700 | [diff] [blame] | 2768 | if (down_write_killable(&mm->mmap_sem)) |
| 2769 | return -EINTR; |
| 2770 | |
Kirill A. Shutemov | c8d78c1 | 2015-02-10 14:09:46 -0800 | [diff] [blame] | 2771 | vma = find_vma(mm, start); |
| 2772 | |
| 2773 | if (!vma || !(vma->vm_flags & VM_SHARED)) |
| 2774 | goto out; |
| 2775 | |
Kirill A. Shutemov | 48f7df3 | 2016-02-17 13:11:15 -0800 | [diff] [blame] | 2776 | if (start < vma->vm_start) |
Kirill A. Shutemov | c8d78c1 | 2015-02-10 14:09:46 -0800 | [diff] [blame] | 2777 | goto out; |
| 2778 | |
Kirill A. Shutemov | 48f7df3 | 2016-02-17 13:11:15 -0800 | [diff] [blame] | 2779 | if (start + size > vma->vm_end) { |
| 2780 | struct vm_area_struct *next; |
| 2781 | |
| 2782 | for (next = vma->vm_next; next; next = next->vm_next) { |
| 2783 | /* hole between vmas ? */ |
| 2784 | if (next->vm_start != next->vm_prev->vm_end) |
| 2785 | goto out; |
| 2786 | |
| 2787 | if (next->vm_file != vma->vm_file) |
| 2788 | goto out; |
| 2789 | |
| 2790 | if (next->vm_flags != vma->vm_flags) |
| 2791 | goto out; |
| 2792 | |
| 2793 | if (start + size <= next->vm_end) |
| 2794 | break; |
| 2795 | } |
| 2796 | |
| 2797 | if (!next) |
| 2798 | goto out; |
Kirill A. Shutemov | c8d78c1 | 2015-02-10 14:09:46 -0800 | [diff] [blame] | 2799 | } |
| 2800 | |
| 2801 | prot |= vma->vm_flags & VM_READ ? PROT_READ : 0; |
| 2802 | prot |= vma->vm_flags & VM_WRITE ? PROT_WRITE : 0; |
| 2803 | prot |= vma->vm_flags & VM_EXEC ? PROT_EXEC : 0; |
| 2804 | |
| 2805 | flags &= MAP_NONBLOCK; |
| 2806 | flags |= MAP_SHARED | MAP_FIXED | MAP_POPULATE; |
| 2807 | if (vma->vm_flags & VM_LOCKED) { |
Kirill A. Shutemov | 48f7df3 | 2016-02-17 13:11:15 -0800 | [diff] [blame] | 2808 | struct vm_area_struct *tmp; |
Kirill A. Shutemov | c8d78c1 | 2015-02-10 14:09:46 -0800 | [diff] [blame] | 2809 | flags |= MAP_LOCKED; |
Kirill A. Shutemov | 48f7df3 | 2016-02-17 13:11:15 -0800 | [diff] [blame] | 2810 | |
Kirill A. Shutemov | c8d78c1 | 2015-02-10 14:09:46 -0800 | [diff] [blame] | 2811 | /* drop PG_Mlocked flag for over-mapped range */ |
Kirill A. Shutemov | 48f7df3 | 2016-02-17 13:11:15 -0800 | [diff] [blame] | 2812 | for (tmp = vma; tmp->vm_start >= start + size; |
| 2813 | tmp = tmp->vm_next) { |
Kirill A. Shutemov | 9a73f61 | 2016-07-26 15:25:53 -0700 | [diff] [blame] | 2814 | /* |
| 2815 | * Split pmd and munlock page on the border |
| 2816 | * of the range. |
| 2817 | */ |
| 2818 | vma_adjust_trans_huge(tmp, start, start + size, 0); |
| 2819 | |
Kirill A. Shutemov | 48f7df3 | 2016-02-17 13:11:15 -0800 | [diff] [blame] | 2820 | munlock_vma_pages_range(tmp, |
| 2821 | max(tmp->vm_start, start), |
| 2822 | min(tmp->vm_end, start + size)); |
| 2823 | } |
Kirill A. Shutemov | c8d78c1 | 2015-02-10 14:09:46 -0800 | [diff] [blame] | 2824 | } |
| 2825 | |
| 2826 | file = get_file(vma->vm_file); |
| 2827 | ret = do_mmap_pgoff(vma->vm_file, start, size, |
| 2828 | prot, flags, pgoff, &populate); |
| 2829 | fput(file); |
| 2830 | out: |
| 2831 | up_write(&mm->mmap_sem); |
| 2832 | if (populate) |
| 2833 | mm_populate(ret, populate); |
| 2834 | if (!IS_ERR_VALUE(ret)) |
| 2835 | ret = 0; |
| 2836 | return ret; |
| 2837 | } |
| 2838 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2839 | static inline void verify_mm_writelocked(struct mm_struct *mm) |
| 2840 | { |
Paul E. McKenney | a241ec6 | 2005-10-30 15:03:12 -0800 | [diff] [blame] | 2841 | #ifdef CONFIG_DEBUG_VM |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2842 | if (unlikely(down_read_trylock(&mm->mmap_sem))) { |
| 2843 | WARN_ON(1); |
| 2844 | up_read(&mm->mmap_sem); |
| 2845 | } |
| 2846 | #endif |
| 2847 | } |
| 2848 | |
| 2849 | /* |
| 2850 | * this is really a simplified "do_mmap". it only handles |
| 2851 | * anonymous maps. eventually we may be able to do some |
| 2852 | * brk-specific accounting here. |
| 2853 | */ |
Kees Cook | ba093a6 | 2016-08-02 14:04:54 -0700 | [diff] [blame] | 2854 | static int do_brk(unsigned long addr, unsigned long request) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2855 | { |
vishnu.ps | cc71aba | 2014-10-09 15:26:29 -0700 | [diff] [blame] | 2856 | struct mm_struct *mm = current->mm; |
| 2857 | struct vm_area_struct *vma, *prev; |
Kees Cook | ba093a6 | 2016-08-02 14:04:54 -0700 | [diff] [blame] | 2858 | unsigned long flags, len; |
vishnu.ps | cc71aba | 2014-10-09 15:26:29 -0700 | [diff] [blame] | 2859 | struct rb_node **rb_link, *rb_parent; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2860 | pgoff_t pgoff = addr >> PAGE_SHIFT; |
Kirill Korotaev | 3a45975 | 2006-09-07 14:17:04 +0400 | [diff] [blame] | 2861 | int error; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2862 | |
Kees Cook | ba093a6 | 2016-08-02 14:04:54 -0700 | [diff] [blame] | 2863 | len = PAGE_ALIGN(request); |
| 2864 | if (len < request) |
| 2865 | return -ENOMEM; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2866 | if (!len) |
Linus Torvalds | 5d22fc2 | 2016-05-27 15:57:31 -0700 | [diff] [blame] | 2867 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2868 | |
Kirill Korotaev | 3a45975 | 2006-09-07 14:17:04 +0400 | [diff] [blame] | 2869 | flags = VM_DATA_DEFAULT_FLAGS | VM_ACCOUNT | mm->def_flags; |
| 2870 | |
Al Viro | 2c6a101 | 2009-12-03 19:40:46 -0500 | [diff] [blame] | 2871 | error = get_unmapped_area(NULL, addr, len, 0, MAP_FIXED); |
Alexander Kuleshov | de1741a | 2015-11-05 18:46:54 -0800 | [diff] [blame] | 2872 | if (offset_in_page(error)) |
Kirill Korotaev | 3a45975 | 2006-09-07 14:17:04 +0400 | [diff] [blame] | 2873 | return error; |
| 2874 | |
Davidlohr Bueso | 363ee17 | 2014-01-21 15:49:15 -0800 | [diff] [blame] | 2875 | error = mlock_future_check(mm, mm->def_flags, len); |
| 2876 | if (error) |
| 2877 | return error; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2878 | |
| 2879 | /* |
| 2880 | * mm->mmap_sem is required to protect against another thread |
| 2881 | * changing the mappings in case we sleep. |
| 2882 | */ |
| 2883 | verify_mm_writelocked(mm); |
| 2884 | |
| 2885 | /* |
| 2886 | * Clear old maps. this also does some error checking for us |
| 2887 | */ |
Rasmus Villemoes | 9fcd145 | 2015-04-15 16:14:32 -0700 | [diff] [blame] | 2888 | while (find_vma_links(mm, addr, addr + len, &prev, &rb_link, |
| 2889 | &rb_parent)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2890 | if (do_munmap(mm, addr, len)) |
| 2891 | return -ENOMEM; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2892 | } |
| 2893 | |
| 2894 | /* Check against address space limits *after* clearing old maps... */ |
Konstantin Khlebnikov | 8463833 | 2016-01-14 15:22:07 -0800 | [diff] [blame] | 2895 | if (!may_expand_vm(mm, flags, len >> PAGE_SHIFT)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2896 | return -ENOMEM; |
| 2897 | |
| 2898 | if (mm->map_count > sysctl_max_map_count) |
| 2899 | return -ENOMEM; |
| 2900 | |
Al Viro | 191c542 | 2012-02-13 03:58:52 +0000 | [diff] [blame] | 2901 | if (security_vm_enough_memory_mm(mm, len >> PAGE_SHIFT)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2902 | return -ENOMEM; |
| 2903 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2904 | /* Can we just expand an old private anonymous mapping? */ |
Rik van Riel | ba470de | 2008-10-18 20:26:50 -0700 | [diff] [blame] | 2905 | vma = vma_merge(mm, prev, addr, addr + len, flags, |
Colin Cross | 3e4578f | 2015-10-27 16:42:08 -0700 | [diff] [blame] | 2906 | NULL, NULL, pgoff, NULL, NULL_VM_UFFD_CTX, NULL); |
Rik van Riel | ba470de | 2008-10-18 20:26:50 -0700 | [diff] [blame] | 2907 | if (vma) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2908 | goto out; |
| 2909 | |
| 2910 | /* |
| 2911 | * create a vma struct for an anonymous mapping |
| 2912 | */ |
Pekka Enberg | c5e3b83 | 2006-03-25 03:06:43 -0800 | [diff] [blame] | 2913 | vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2914 | if (!vma) { |
| 2915 | vm_unacct_memory(len >> PAGE_SHIFT); |
| 2916 | return -ENOMEM; |
| 2917 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2918 | |
Rik van Riel | 5beb493 | 2010-03-05 13:42:07 -0800 | [diff] [blame] | 2919 | INIT_LIST_HEAD(&vma->anon_vma_chain); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2920 | vma->vm_mm = mm; |
| 2921 | vma->vm_start = addr; |
| 2922 | vma->vm_end = addr + len; |
| 2923 | vma->vm_pgoff = pgoff; |
| 2924 | vma->vm_flags = flags; |
Coly Li | 3ed75eb | 2007-10-18 23:39:15 -0700 | [diff] [blame] | 2925 | vma->vm_page_prot = vm_get_page_prot(flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2926 | vma_link(mm, vma, prev, rb_link, rb_parent); |
| 2927 | out: |
Eric B Munson | 3af9e85 | 2010-05-18 15:30:49 +0100 | [diff] [blame] | 2928 | perf_event_mmap(vma); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2929 | mm->total_vm += len >> PAGE_SHIFT; |
Konstantin Khlebnikov | 8463833 | 2016-01-14 15:22:07 -0800 | [diff] [blame] | 2930 | mm->data_vm += len >> PAGE_SHIFT; |
Michel Lespinasse | 128557f | 2013-02-22 16:32:40 -0800 | [diff] [blame] | 2931 | if (flags & VM_LOCKED) |
| 2932 | mm->locked_vm += (len >> PAGE_SHIFT); |
Cyrill Gorcunov | d9104d1 | 2013-09-11 14:22:24 -0700 | [diff] [blame] | 2933 | vma->vm_flags |= VM_SOFTDIRTY; |
Linus Torvalds | 5d22fc2 | 2016-05-27 15:57:31 -0700 | [diff] [blame] | 2934 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2935 | } |
| 2936 | |
Linus Torvalds | 5d22fc2 | 2016-05-27 15:57:31 -0700 | [diff] [blame] | 2937 | int vm_brk(unsigned long addr, unsigned long len) |
Linus Torvalds | e4eb1ff | 2012-04-20 15:35:40 -0700 | [diff] [blame] | 2938 | { |
| 2939 | struct mm_struct *mm = current->mm; |
Linus Torvalds | 5d22fc2 | 2016-05-27 15:57:31 -0700 | [diff] [blame] | 2940 | int ret; |
Michel Lespinasse | 128557f | 2013-02-22 16:32:40 -0800 | [diff] [blame] | 2941 | bool populate; |
Linus Torvalds | e4eb1ff | 2012-04-20 15:35:40 -0700 | [diff] [blame] | 2942 | |
Michal Hocko | 2d6c928 | 2016-05-23 16:25:42 -0700 | [diff] [blame] | 2943 | if (down_write_killable(&mm->mmap_sem)) |
| 2944 | return -EINTR; |
| 2945 | |
Linus Torvalds | e4eb1ff | 2012-04-20 15:35:40 -0700 | [diff] [blame] | 2946 | ret = do_brk(addr, len); |
Michel Lespinasse | 128557f | 2013-02-22 16:32:40 -0800 | [diff] [blame] | 2947 | populate = ((mm->def_flags & VM_LOCKED) != 0); |
Linus Torvalds | e4eb1ff | 2012-04-20 15:35:40 -0700 | [diff] [blame] | 2948 | up_write(&mm->mmap_sem); |
Linus Torvalds | 5d22fc2 | 2016-05-27 15:57:31 -0700 | [diff] [blame] | 2949 | if (populate && !ret) |
Michel Lespinasse | 128557f | 2013-02-22 16:32:40 -0800 | [diff] [blame] | 2950 | mm_populate(addr, len); |
Linus Torvalds | e4eb1ff | 2012-04-20 15:35:40 -0700 | [diff] [blame] | 2951 | return ret; |
| 2952 | } |
| 2953 | EXPORT_SYMBOL(vm_brk); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2954 | |
| 2955 | /* Release all mmaps. */ |
| 2956 | void exit_mmap(struct mm_struct *mm) |
| 2957 | { |
Peter Zijlstra | d16dfc5 | 2011-05-24 17:11:45 -0700 | [diff] [blame] | 2958 | struct mmu_gather tlb; |
Rik van Riel | ba470de | 2008-10-18 20:26:50 -0700 | [diff] [blame] | 2959 | struct vm_area_struct *vma; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2960 | unsigned long nr_accounted = 0; |
| 2961 | |
Jeremy Fitzhardinge | d6dd61c | 2007-05-02 19:27:14 +0200 | [diff] [blame] | 2962 | /* mm's last user has gone, and its about to be pulled down */ |
Andrea Arcangeli | cddb8a5 | 2008-07-28 15:46:29 -0700 | [diff] [blame] | 2963 | mmu_notifier_release(mm); |
Jeremy Fitzhardinge | d6dd61c | 2007-05-02 19:27:14 +0200 | [diff] [blame] | 2964 | |
Rik van Riel | ba470de | 2008-10-18 20:26:50 -0700 | [diff] [blame] | 2965 | if (mm->locked_vm) { |
| 2966 | vma = mm->mmap; |
| 2967 | while (vma) { |
| 2968 | if (vma->vm_flags & VM_LOCKED) |
| 2969 | munlock_vma_pages_all(vma); |
| 2970 | vma = vma->vm_next; |
| 2971 | } |
| 2972 | } |
Jeremy Fitzhardinge | 9480c53 | 2009-02-11 13:04:41 -0800 | [diff] [blame] | 2973 | |
| 2974 | arch_exit_mmap(mm); |
| 2975 | |
Rik van Riel | ba470de | 2008-10-18 20:26:50 -0700 | [diff] [blame] | 2976 | vma = mm->mmap; |
Jeremy Fitzhardinge | 9480c53 | 2009-02-11 13:04:41 -0800 | [diff] [blame] | 2977 | if (!vma) /* Can happen if dup_mmap() received an OOM */ |
| 2978 | return; |
| 2979 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2980 | lru_add_drain(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2981 | flush_cache_mm(mm); |
Linus Torvalds | 2b04725 | 2013-08-15 11:42:25 -0700 | [diff] [blame] | 2982 | tlb_gather_mmu(&tlb, mm, 0, -1); |
Oleg Nesterov | 901608d | 2009-01-06 14:40:29 -0800 | [diff] [blame] | 2983 | /* update_hiwater_rss(mm) here? but nobody should be looking */ |
Hugh Dickins | e0da382 | 2005-04-19 13:29:15 -0700 | [diff] [blame] | 2984 | /* Use -1 here to ensure all VMAs in the mm are unmapped */ |
Linus Torvalds | 4f74d2c | 2012-05-06 13:54:06 -0700 | [diff] [blame] | 2985 | unmap_vmas(&tlb, vma, 0, -1); |
Hugh Dickins | 9ba6929 | 2009-09-21 17:02:20 -0700 | [diff] [blame] | 2986 | |
Andrea Arcangeli | 07b039e9 | 2017-09-06 16:25:00 -0700 | [diff] [blame] | 2987 | set_bit(MMF_OOM_SKIP, &mm->flags); |
| 2988 | if (unlikely(tsk_is_oom_victim(current))) { |
| 2989 | /* |
| 2990 | * Wait for oom_reap_task() to stop working on this |
| 2991 | * mm. Because MMF_OOM_SKIP is already set before |
| 2992 | * calling down_read(), oom_reap_task() will not run |
| 2993 | * on this "mm" post up_write(). |
| 2994 | * |
| 2995 | * tsk_is_oom_victim() cannot be set from under us |
| 2996 | * either because current->mm is already set to NULL |
| 2997 | * under task_lock before calling mmput and oom_mm is |
| 2998 | * set not NULL by the OOM killer only if current->mm |
| 2999 | * is found not NULL while holding the task_lock. |
| 3000 | */ |
| 3001 | down_write(&mm->mmap_sem); |
| 3002 | up_write(&mm->mmap_sem); |
| 3003 | } |
Hugh Dickins | 6ee8630 | 2013-04-29 15:07:44 -0700 | [diff] [blame] | 3004 | free_pgtables(&tlb, vma, FIRST_USER_ADDRESS, USER_PGTABLES_CEILING); |
Al Viro | 853f5e2 | 2012-03-05 14:03:47 -0500 | [diff] [blame] | 3005 | tlb_finish_mmu(&tlb, 0, -1); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3006 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3007 | /* |
Hugh Dickins | 8f4f8c1 | 2005-10-29 18:16:29 -0700 | [diff] [blame] | 3008 | * Walk the list again, actually closing and freeing it, |
| 3009 | * with preemption enabled, without holding any MM locks. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3010 | */ |
Linus Torvalds | 4f74d2c | 2012-05-06 13:54:06 -0700 | [diff] [blame] | 3011 | while (vma) { |
| 3012 | if (vma->vm_flags & VM_ACCOUNT) |
| 3013 | nr_accounted += vma_pages(vma); |
Hugh Dickins | a8fb561 | 2005-10-29 18:15:57 -0700 | [diff] [blame] | 3014 | vma = remove_vma(vma); |
Linus Torvalds | 4f74d2c | 2012-05-06 13:54:06 -0700 | [diff] [blame] | 3015 | } |
| 3016 | vm_unacct_memory(nr_accounted); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3017 | } |
| 3018 | |
| 3019 | /* Insert vm structure into process list sorted by address |
| 3020 | * and into the inode's i_mmap tree. If vm_file is non-NULL |
Davidlohr Bueso | c8c06ef | 2014-12-12 16:54:24 -0800 | [diff] [blame] | 3021 | * then i_mmap_rwsem is taken here. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3022 | */ |
Hugh Dickins | 6597d78 | 2012-10-08 16:29:07 -0700 | [diff] [blame] | 3023 | int insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3024 | { |
Hugh Dickins | 6597d78 | 2012-10-08 16:29:07 -0700 | [diff] [blame] | 3025 | struct vm_area_struct *prev; |
| 3026 | struct rb_node **rb_link, *rb_parent; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3027 | |
Chen Gang | c9d13f5 | 2015-09-08 15:04:08 -0700 | [diff] [blame] | 3028 | if (find_vma_links(mm, vma->vm_start, vma->vm_end, |
| 3029 | &prev, &rb_link, &rb_parent)) |
| 3030 | return -ENOMEM; |
| 3031 | if ((vma->vm_flags & VM_ACCOUNT) && |
| 3032 | security_vm_enough_memory_mm(mm, vma_pages(vma))) |
| 3033 | return -ENOMEM; |
| 3034 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3035 | /* |
| 3036 | * The vm_pgoff of a purely anonymous vma should be irrelevant |
| 3037 | * until its first write fault, when page's anon_vma and index |
| 3038 | * are set. But now set the vm_pgoff it will almost certainly |
| 3039 | * end up with (unless mremap moves it elsewhere before that |
| 3040 | * first wfault), so /proc/pid/maps tells a consistent story. |
| 3041 | * |
| 3042 | * By setting it to reflect the virtual start address of the |
| 3043 | * vma, merges and splits can happen in a seamless way, just |
| 3044 | * using the existing file pgoff checks and manipulations. |
| 3045 | * Similarly in do_mmap_pgoff and in do_brk. |
| 3046 | */ |
Oleg Nesterov | 8a9cc3b | 2015-09-08 14:58:31 -0700 | [diff] [blame] | 3047 | if (vma_is_anonymous(vma)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3048 | BUG_ON(vma->anon_vma); |
| 3049 | vma->vm_pgoff = vma->vm_start >> PAGE_SHIFT; |
| 3050 | } |
Srikar Dronamraju | 2b14449 | 2012-02-09 14:56:42 +0530 | [diff] [blame] | 3051 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3052 | vma_link(mm, vma, prev, rb_link, rb_parent); |
| 3053 | return 0; |
| 3054 | } |
| 3055 | |
| 3056 | /* |
| 3057 | * Copy the vma structure to a new location in the same mm, |
| 3058 | * prior to moving page table entries, to effect an mremap move. |
| 3059 | */ |
| 3060 | struct vm_area_struct *copy_vma(struct vm_area_struct **vmap, |
Michel Lespinasse | 38a7601 | 2012-10-08 16:31:50 -0700 | [diff] [blame] | 3061 | unsigned long addr, unsigned long len, pgoff_t pgoff, |
| 3062 | bool *need_rmap_locks) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3063 | { |
| 3064 | struct vm_area_struct *vma = *vmap; |
| 3065 | unsigned long vma_start = vma->vm_start; |
| 3066 | struct mm_struct *mm = vma->vm_mm; |
| 3067 | struct vm_area_struct *new_vma, *prev; |
| 3068 | struct rb_node **rb_link, *rb_parent; |
Andrea Arcangeli | 948f017 | 2012-01-10 15:08:05 -0800 | [diff] [blame] | 3069 | bool faulted_in_anon_vma = true; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3070 | |
| 3071 | /* |
| 3072 | * If anonymous vma has not yet been faulted, update new pgoff |
| 3073 | * to match new location, to increase its chance of merging. |
| 3074 | */ |
Oleg Nesterov | ce75799 | 2015-09-08 14:58:34 -0700 | [diff] [blame] | 3075 | if (unlikely(vma_is_anonymous(vma) && !vma->anon_vma)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3076 | pgoff = addr >> PAGE_SHIFT; |
Andrea Arcangeli | 948f017 | 2012-01-10 15:08:05 -0800 | [diff] [blame] | 3077 | faulted_in_anon_vma = false; |
| 3078 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3079 | |
Hugh Dickins | 6597d78 | 2012-10-08 16:29:07 -0700 | [diff] [blame] | 3080 | if (find_vma_links(mm, addr, addr + len, &prev, &rb_link, &rb_parent)) |
| 3081 | return NULL; /* should never get here */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3082 | new_vma = vma_merge(mm, prev, addr, addr + len, vma->vm_flags, |
Andrea Arcangeli | 19a809a | 2015-09-04 15:46:24 -0700 | [diff] [blame] | 3083 | vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma), |
Colin Cross | 3e4578f | 2015-10-27 16:42:08 -0700 | [diff] [blame] | 3084 | vma->vm_userfaultfd_ctx, vma_get_anon_name(vma)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3085 | if (new_vma) { |
| 3086 | /* |
| 3087 | * Source vma may have been merged into new_vma |
| 3088 | */ |
Andrea Arcangeli | 948f017 | 2012-01-10 15:08:05 -0800 | [diff] [blame] | 3089 | if (unlikely(vma_start >= new_vma->vm_start && |
| 3090 | vma_start < new_vma->vm_end)) { |
| 3091 | /* |
| 3092 | * The only way we can get a vma_merge with |
| 3093 | * self during an mremap is if the vma hasn't |
| 3094 | * been faulted in yet and we were allowed to |
| 3095 | * reset the dst vma->vm_pgoff to the |
| 3096 | * destination address of the mremap to allow |
| 3097 | * the merge to happen. mremap must change the |
| 3098 | * vm_pgoff linearity between src and dst vmas |
| 3099 | * (in turn preventing a vma_merge) to be |
| 3100 | * safe. It is only safe to keep the vm_pgoff |
| 3101 | * linear if there are no pages mapped yet. |
| 3102 | */ |
Sasha Levin | 81d1b09 | 2014-10-09 15:28:10 -0700 | [diff] [blame] | 3103 | VM_BUG_ON_VMA(faulted_in_anon_vma, new_vma); |
Michel Lespinasse | 38a7601 | 2012-10-08 16:31:50 -0700 | [diff] [blame] | 3104 | *vmap = vma = new_vma; |
Michel Lespinasse | 108d664 | 2012-10-08 16:31:36 -0700 | [diff] [blame] | 3105 | } |
Michel Lespinasse | 38a7601 | 2012-10-08 16:31:50 -0700 | [diff] [blame] | 3106 | *need_rmap_locks = (new_vma->vm_pgoff <= vma->vm_pgoff); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3107 | } else { |
Christoph Lameter | e94b176 | 2006-12-06 20:33:17 -0800 | [diff] [blame] | 3108 | new_vma = kmem_cache_alloc(vm_area_cachep, GFP_KERNEL); |
Chen Gang | e397589 | 2015-09-08 15:03:38 -0700 | [diff] [blame] | 3109 | if (!new_vma) |
| 3110 | goto out; |
| 3111 | *new_vma = *vma; |
| 3112 | new_vma->vm_start = addr; |
| 3113 | new_vma->vm_end = addr + len; |
| 3114 | new_vma->vm_pgoff = pgoff; |
| 3115 | if (vma_dup_policy(vma, new_vma)) |
| 3116 | goto out_free_vma; |
| 3117 | INIT_LIST_HEAD(&new_vma->anon_vma_chain); |
| 3118 | if (anon_vma_clone(new_vma, vma)) |
| 3119 | goto out_free_mempol; |
| 3120 | if (new_vma->vm_file) |
| 3121 | get_file(new_vma->vm_file); |
| 3122 | if (new_vma->vm_ops && new_vma->vm_ops->open) |
| 3123 | new_vma->vm_ops->open(new_vma); |
| 3124 | vma_link(mm, new_vma, prev, rb_link, rb_parent); |
| 3125 | *need_rmap_locks = false; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3126 | } |
| 3127 | return new_vma; |
Rik van Riel | 5beb493 | 2010-03-05 13:42:07 -0800 | [diff] [blame] | 3128 | |
Chen Gang | e397589 | 2015-09-08 15:03:38 -0700 | [diff] [blame] | 3129 | out_free_mempol: |
Oleg Nesterov | ef0855d | 2013-09-11 14:20:14 -0700 | [diff] [blame] | 3130 | mpol_put(vma_policy(new_vma)); |
Chen Gang | e397589 | 2015-09-08 15:03:38 -0700 | [diff] [blame] | 3131 | out_free_vma: |
Rik van Riel | 5beb493 | 2010-03-05 13:42:07 -0800 | [diff] [blame] | 3132 | kmem_cache_free(vm_area_cachep, new_vma); |
Chen Gang | e397589 | 2015-09-08 15:03:38 -0700 | [diff] [blame] | 3133 | out: |
Rik van Riel | 5beb493 | 2010-03-05 13:42:07 -0800 | [diff] [blame] | 3134 | return NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3135 | } |
akpm@osdl.org | 119f657 | 2005-05-01 08:58:35 -0700 | [diff] [blame] | 3136 | |
| 3137 | /* |
| 3138 | * Return true if the calling process may expand its vm space by the passed |
| 3139 | * number of pages |
| 3140 | */ |
Konstantin Khlebnikov | 8463833 | 2016-01-14 15:22:07 -0800 | [diff] [blame] | 3141 | bool may_expand_vm(struct mm_struct *mm, vm_flags_t flags, unsigned long npages) |
akpm@osdl.org | 119f657 | 2005-05-01 08:58:35 -0700 | [diff] [blame] | 3142 | { |
Konstantin Khlebnikov | 8463833 | 2016-01-14 15:22:07 -0800 | [diff] [blame] | 3143 | if (mm->total_vm + npages > rlimit(RLIMIT_AS) >> PAGE_SHIFT) |
| 3144 | return false; |
akpm@osdl.org | 119f657 | 2005-05-01 08:58:35 -0700 | [diff] [blame] | 3145 | |
Konstantin Khlebnikov | d977d56 | 2016-02-02 16:57:43 -0800 | [diff] [blame] | 3146 | if (is_data_mapping(flags) && |
| 3147 | mm->data_vm + npages > rlimit(RLIMIT_DATA) >> PAGE_SHIFT) { |
Konstantin Khlebnikov | f4fcd55 | 2016-05-20 16:57:45 -0700 | [diff] [blame] | 3148 | /* Workaround for Valgrind */ |
| 3149 | if (rlimit(RLIMIT_DATA) == 0 && |
| 3150 | mm->data_vm + npages <= rlimit_max(RLIMIT_DATA) >> PAGE_SHIFT) |
| 3151 | return true; |
| 3152 | if (!ignore_rlimit_data) { |
| 3153 | pr_warn_once("%s (%d): VmData %lu exceed data ulimit %lu. Update limits or use boot option ignore_rlimit_data.\n", |
Konstantin Khlebnikov | d977d56 | 2016-02-02 16:57:43 -0800 | [diff] [blame] | 3154 | current->comm, current->pid, |
| 3155 | (mm->data_vm + npages) << PAGE_SHIFT, |
| 3156 | rlimit(RLIMIT_DATA)); |
Konstantin Khlebnikov | d977d56 | 2016-02-02 16:57:43 -0800 | [diff] [blame] | 3157 | return false; |
Konstantin Khlebnikov | f4fcd55 | 2016-05-20 16:57:45 -0700 | [diff] [blame] | 3158 | } |
Konstantin Khlebnikov | d977d56 | 2016-02-02 16:57:43 -0800 | [diff] [blame] | 3159 | } |
akpm@osdl.org | 119f657 | 2005-05-01 08:58:35 -0700 | [diff] [blame] | 3160 | |
Konstantin Khlebnikov | 8463833 | 2016-01-14 15:22:07 -0800 | [diff] [blame] | 3161 | return true; |
| 3162 | } |
| 3163 | |
| 3164 | void vm_stat_account(struct mm_struct *mm, vm_flags_t flags, long npages) |
| 3165 | { |
| 3166 | mm->total_vm += npages; |
| 3167 | |
Konstantin Khlebnikov | d977d56 | 2016-02-02 16:57:43 -0800 | [diff] [blame] | 3168 | if (is_exec_mapping(flags)) |
Konstantin Khlebnikov | 8463833 | 2016-01-14 15:22:07 -0800 | [diff] [blame] | 3169 | mm->exec_vm += npages; |
Konstantin Khlebnikov | d977d56 | 2016-02-02 16:57:43 -0800 | [diff] [blame] | 3170 | else if (is_stack_mapping(flags)) |
Konstantin Khlebnikov | 8463833 | 2016-01-14 15:22:07 -0800 | [diff] [blame] | 3171 | mm->stack_vm += npages; |
Konstantin Khlebnikov | d977d56 | 2016-02-02 16:57:43 -0800 | [diff] [blame] | 3172 | else if (is_data_mapping(flags)) |
Konstantin Khlebnikov | 8463833 | 2016-01-14 15:22:07 -0800 | [diff] [blame] | 3173 | mm->data_vm += npages; |
akpm@osdl.org | 119f657 | 2005-05-01 08:58:35 -0700 | [diff] [blame] | 3174 | } |
Roland McGrath | fa5dc22 | 2007-02-08 14:20:41 -0800 | [diff] [blame] | 3175 | |
Andy Lutomirski | a62c34b | 2014-05-19 15:58:33 -0700 | [diff] [blame] | 3176 | static int special_mapping_fault(struct vm_area_struct *vma, |
| 3177 | struct vm_fault *vmf); |
| 3178 | |
| 3179 | /* |
| 3180 | * Having a close hook prevents vma merging regardless of flags. |
| 3181 | */ |
| 3182 | static void special_mapping_close(struct vm_area_struct *vma) |
| 3183 | { |
| 3184 | } |
| 3185 | |
| 3186 | static const char *special_mapping_name(struct vm_area_struct *vma) |
| 3187 | { |
| 3188 | return ((struct vm_special_mapping *)vma->vm_private_data)->name; |
| 3189 | } |
| 3190 | |
Dmitry Safonov | b059a45 | 2016-06-28 14:35:38 +0300 | [diff] [blame] | 3191 | static int special_mapping_mremap(struct vm_area_struct *new_vma) |
| 3192 | { |
| 3193 | struct vm_special_mapping *sm = new_vma->vm_private_data; |
| 3194 | |
| 3195 | if (sm->mremap) |
| 3196 | return sm->mremap(sm, new_vma); |
| 3197 | return 0; |
| 3198 | } |
| 3199 | |
Andy Lutomirski | a62c34b | 2014-05-19 15:58:33 -0700 | [diff] [blame] | 3200 | static const struct vm_operations_struct special_mapping_vmops = { |
| 3201 | .close = special_mapping_close, |
| 3202 | .fault = special_mapping_fault, |
Dmitry Safonov | b059a45 | 2016-06-28 14:35:38 +0300 | [diff] [blame] | 3203 | .mremap = special_mapping_mremap, |
Andy Lutomirski | a62c34b | 2014-05-19 15:58:33 -0700 | [diff] [blame] | 3204 | .name = special_mapping_name, |
| 3205 | }; |
| 3206 | |
| 3207 | static const struct vm_operations_struct legacy_special_mapping_vmops = { |
| 3208 | .close = special_mapping_close, |
| 3209 | .fault = special_mapping_fault, |
| 3210 | }; |
Roland McGrath | fa5dc22 | 2007-02-08 14:20:41 -0800 | [diff] [blame] | 3211 | |
Nick Piggin | b1d0e4f | 2008-02-09 01:15:19 +0100 | [diff] [blame] | 3212 | static int special_mapping_fault(struct vm_area_struct *vma, |
| 3213 | struct vm_fault *vmf) |
Roland McGrath | fa5dc22 | 2007-02-08 14:20:41 -0800 | [diff] [blame] | 3214 | { |
Nick Piggin | b1d0e4f | 2008-02-09 01:15:19 +0100 | [diff] [blame] | 3215 | pgoff_t pgoff; |
Roland McGrath | fa5dc22 | 2007-02-08 14:20:41 -0800 | [diff] [blame] | 3216 | struct page **pages; |
| 3217 | |
Andy Lutomirski | f872f54 | 2015-12-29 20:12:19 -0800 | [diff] [blame] | 3218 | if (vma->vm_ops == &legacy_special_mapping_vmops) { |
Andy Lutomirski | a62c34b | 2014-05-19 15:58:33 -0700 | [diff] [blame] | 3219 | pages = vma->vm_private_data; |
Andy Lutomirski | f872f54 | 2015-12-29 20:12:19 -0800 | [diff] [blame] | 3220 | } else { |
| 3221 | struct vm_special_mapping *sm = vma->vm_private_data; |
| 3222 | |
| 3223 | if (sm->fault) |
| 3224 | return sm->fault(sm, vma, vmf); |
| 3225 | |
| 3226 | pages = sm->pages; |
| 3227 | } |
Andy Lutomirski | a62c34b | 2014-05-19 15:58:33 -0700 | [diff] [blame] | 3228 | |
Oleg Nesterov | 8a9cc3b | 2015-09-08 14:58:31 -0700 | [diff] [blame] | 3229 | for (pgoff = vmf->pgoff; pgoff && *pages; ++pages) |
Nick Piggin | b1d0e4f | 2008-02-09 01:15:19 +0100 | [diff] [blame] | 3230 | pgoff--; |
Roland McGrath | fa5dc22 | 2007-02-08 14:20:41 -0800 | [diff] [blame] | 3231 | |
| 3232 | if (*pages) { |
| 3233 | struct page *page = *pages; |
| 3234 | get_page(page); |
Nick Piggin | b1d0e4f | 2008-02-09 01:15:19 +0100 | [diff] [blame] | 3235 | vmf->page = page; |
| 3236 | return 0; |
Roland McGrath | fa5dc22 | 2007-02-08 14:20:41 -0800 | [diff] [blame] | 3237 | } |
| 3238 | |
Nick Piggin | b1d0e4f | 2008-02-09 01:15:19 +0100 | [diff] [blame] | 3239 | return VM_FAULT_SIGBUS; |
Roland McGrath | fa5dc22 | 2007-02-08 14:20:41 -0800 | [diff] [blame] | 3240 | } |
| 3241 | |
Andy Lutomirski | a62c34b | 2014-05-19 15:58:33 -0700 | [diff] [blame] | 3242 | static struct vm_area_struct *__install_special_mapping( |
| 3243 | struct mm_struct *mm, |
| 3244 | unsigned long addr, unsigned long len, |
Chen Gang | 27f28b9 | 2015-11-05 18:48:41 -0800 | [diff] [blame] | 3245 | unsigned long vm_flags, void *priv, |
| 3246 | const struct vm_operations_struct *ops) |
Roland McGrath | fa5dc22 | 2007-02-08 14:20:41 -0800 | [diff] [blame] | 3247 | { |
Tavis Ormandy | 462e635e | 2010-12-09 15:29:42 +0100 | [diff] [blame] | 3248 | int ret; |
Roland McGrath | fa5dc22 | 2007-02-08 14:20:41 -0800 | [diff] [blame] | 3249 | struct vm_area_struct *vma; |
| 3250 | |
| 3251 | vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL); |
| 3252 | if (unlikely(vma == NULL)) |
Stefani Seibold | 3935ed6 | 2014-03-17 23:22:02 +0100 | [diff] [blame] | 3253 | return ERR_PTR(-ENOMEM); |
Roland McGrath | fa5dc22 | 2007-02-08 14:20:41 -0800 | [diff] [blame] | 3254 | |
Rik van Riel | 5beb493 | 2010-03-05 13:42:07 -0800 | [diff] [blame] | 3255 | INIT_LIST_HEAD(&vma->anon_vma_chain); |
Roland McGrath | fa5dc22 | 2007-02-08 14:20:41 -0800 | [diff] [blame] | 3256 | vma->vm_mm = mm; |
| 3257 | vma->vm_start = addr; |
| 3258 | vma->vm_end = addr + len; |
| 3259 | |
Cyrill Gorcunov | d9104d1 | 2013-09-11 14:22:24 -0700 | [diff] [blame] | 3260 | vma->vm_flags = vm_flags | mm->def_flags | VM_DONTEXPAND | VM_SOFTDIRTY; |
Coly Li | 3ed75eb | 2007-10-18 23:39:15 -0700 | [diff] [blame] | 3261 | vma->vm_page_prot = vm_get_page_prot(vma->vm_flags); |
Roland McGrath | fa5dc22 | 2007-02-08 14:20:41 -0800 | [diff] [blame] | 3262 | |
Andy Lutomirski | a62c34b | 2014-05-19 15:58:33 -0700 | [diff] [blame] | 3263 | vma->vm_ops = ops; |
| 3264 | vma->vm_private_data = priv; |
Roland McGrath | fa5dc22 | 2007-02-08 14:20:41 -0800 | [diff] [blame] | 3265 | |
Tavis Ormandy | 462e635e | 2010-12-09 15:29:42 +0100 | [diff] [blame] | 3266 | ret = insert_vm_struct(mm, vma); |
| 3267 | if (ret) |
| 3268 | goto out; |
Roland McGrath | fa5dc22 | 2007-02-08 14:20:41 -0800 | [diff] [blame] | 3269 | |
Konstantin Khlebnikov | 8463833 | 2016-01-14 15:22:07 -0800 | [diff] [blame] | 3270 | vm_stat_account(mm, vma->vm_flags, len >> PAGE_SHIFT); |
Roland McGrath | fa5dc22 | 2007-02-08 14:20:41 -0800 | [diff] [blame] | 3271 | |
Ingo Molnar | cdd6c48 | 2009-09-21 12:02:48 +0200 | [diff] [blame] | 3272 | perf_event_mmap(vma); |
Peter Zijlstra | 089dd79 | 2009-06-05 14:04:55 +0200 | [diff] [blame] | 3273 | |
Stefani Seibold | 3935ed6 | 2014-03-17 23:22:02 +0100 | [diff] [blame] | 3274 | return vma; |
Tavis Ormandy | 462e635e | 2010-12-09 15:29:42 +0100 | [diff] [blame] | 3275 | |
| 3276 | out: |
| 3277 | kmem_cache_free(vm_area_cachep, vma); |
Stefani Seibold | 3935ed6 | 2014-03-17 23:22:02 +0100 | [diff] [blame] | 3278 | return ERR_PTR(ret); |
| 3279 | } |
| 3280 | |
Dmitry Safonov | 2eefd87 | 2016-09-05 16:33:05 +0300 | [diff] [blame] | 3281 | bool vma_is_special_mapping(const struct vm_area_struct *vma, |
| 3282 | const struct vm_special_mapping *sm) |
| 3283 | { |
| 3284 | return vma->vm_private_data == sm && |
| 3285 | (vma->vm_ops == &special_mapping_vmops || |
| 3286 | vma->vm_ops == &legacy_special_mapping_vmops); |
| 3287 | } |
| 3288 | |
Andy Lutomirski | a62c34b | 2014-05-19 15:58:33 -0700 | [diff] [blame] | 3289 | /* |
| 3290 | * Called with mm->mmap_sem held for writing. |
| 3291 | * Insert a new vma covering the given region, with the given flags. |
| 3292 | * Its pages are supplied by the given array of struct page *. |
| 3293 | * The array can be shorter than len >> PAGE_SHIFT if it's null-terminated. |
| 3294 | * The region past the last page supplied will always produce SIGBUS. |
| 3295 | * The array pointer and the pages it points to are assumed to stay alive |
| 3296 | * for as long as this mapping might exist. |
| 3297 | */ |
| 3298 | struct vm_area_struct *_install_special_mapping( |
| 3299 | struct mm_struct *mm, |
| 3300 | unsigned long addr, unsigned long len, |
| 3301 | unsigned long vm_flags, const struct vm_special_mapping *spec) |
| 3302 | { |
Chen Gang | 27f28b9 | 2015-11-05 18:48:41 -0800 | [diff] [blame] | 3303 | return __install_special_mapping(mm, addr, len, vm_flags, (void *)spec, |
| 3304 | &special_mapping_vmops); |
Andy Lutomirski | a62c34b | 2014-05-19 15:58:33 -0700 | [diff] [blame] | 3305 | } |
| 3306 | |
Stefani Seibold | 3935ed6 | 2014-03-17 23:22:02 +0100 | [diff] [blame] | 3307 | int install_special_mapping(struct mm_struct *mm, |
| 3308 | unsigned long addr, unsigned long len, |
| 3309 | unsigned long vm_flags, struct page **pages) |
| 3310 | { |
Andy Lutomirski | a62c34b | 2014-05-19 15:58:33 -0700 | [diff] [blame] | 3311 | struct vm_area_struct *vma = __install_special_mapping( |
Chen Gang | 27f28b9 | 2015-11-05 18:48:41 -0800 | [diff] [blame] | 3312 | mm, addr, len, vm_flags, (void *)pages, |
| 3313 | &legacy_special_mapping_vmops); |
Stefani Seibold | 3935ed6 | 2014-03-17 23:22:02 +0100 | [diff] [blame] | 3314 | |
Duan Jiong | 14bd5b4 | 2014-06-04 16:07:05 -0700 | [diff] [blame] | 3315 | return PTR_ERR_OR_ZERO(vma); |
Roland McGrath | fa5dc22 | 2007-02-08 14:20:41 -0800 | [diff] [blame] | 3316 | } |
Andrea Arcangeli | 7906d00 | 2008-07-28 15:46:26 -0700 | [diff] [blame] | 3317 | |
| 3318 | static DEFINE_MUTEX(mm_all_locks_mutex); |
| 3319 | |
Peter Zijlstra | 454ed84 | 2008-08-11 09:30:25 +0200 | [diff] [blame] | 3320 | static void vm_lock_anon_vma(struct mm_struct *mm, struct anon_vma *anon_vma) |
Andrea Arcangeli | 7906d00 | 2008-07-28 15:46:26 -0700 | [diff] [blame] | 3321 | { |
Michel Lespinasse | bf181b9 | 2012-10-08 16:31:39 -0700 | [diff] [blame] | 3322 | if (!test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_node)) { |
Andrea Arcangeli | 7906d00 | 2008-07-28 15:46:26 -0700 | [diff] [blame] | 3323 | /* |
| 3324 | * The LSB of head.next can't change from under us |
| 3325 | * because we hold the mm_all_locks_mutex. |
| 3326 | */ |
Jiri Kosina | 572043c | 2013-01-11 14:31:59 -0800 | [diff] [blame] | 3327 | down_write_nest_lock(&anon_vma->root->rwsem, &mm->mmap_sem); |
Andrea Arcangeli | 7906d00 | 2008-07-28 15:46:26 -0700 | [diff] [blame] | 3328 | /* |
| 3329 | * We can safely modify head.next after taking the |
Ingo Molnar | 5a50508 | 2012-12-02 19:56:46 +0000 | [diff] [blame] | 3330 | * anon_vma->root->rwsem. If some other vma in this mm shares |
Andrea Arcangeli | 7906d00 | 2008-07-28 15:46:26 -0700 | [diff] [blame] | 3331 | * the same anon_vma we won't take it again. |
| 3332 | * |
| 3333 | * No need of atomic instructions here, head.next |
| 3334 | * can't change from under us thanks to the |
Ingo Molnar | 5a50508 | 2012-12-02 19:56:46 +0000 | [diff] [blame] | 3335 | * anon_vma->root->rwsem. |
Andrea Arcangeli | 7906d00 | 2008-07-28 15:46:26 -0700 | [diff] [blame] | 3336 | */ |
| 3337 | if (__test_and_set_bit(0, (unsigned long *) |
Michel Lespinasse | bf181b9 | 2012-10-08 16:31:39 -0700 | [diff] [blame] | 3338 | &anon_vma->root->rb_root.rb_node)) |
Andrea Arcangeli | 7906d00 | 2008-07-28 15:46:26 -0700 | [diff] [blame] | 3339 | BUG(); |
| 3340 | } |
| 3341 | } |
| 3342 | |
Peter Zijlstra | 454ed84 | 2008-08-11 09:30:25 +0200 | [diff] [blame] | 3343 | static void vm_lock_mapping(struct mm_struct *mm, struct address_space *mapping) |
Andrea Arcangeli | 7906d00 | 2008-07-28 15:46:26 -0700 | [diff] [blame] | 3344 | { |
| 3345 | if (!test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) { |
| 3346 | /* |
| 3347 | * AS_MM_ALL_LOCKS can't change from under us because |
| 3348 | * we hold the mm_all_locks_mutex. |
| 3349 | * |
| 3350 | * Operations on ->flags have to be atomic because |
| 3351 | * even if AS_MM_ALL_LOCKS is stable thanks to the |
| 3352 | * mm_all_locks_mutex, there may be other cpus |
| 3353 | * changing other bitflags in parallel to us. |
| 3354 | */ |
| 3355 | if (test_and_set_bit(AS_MM_ALL_LOCKS, &mapping->flags)) |
| 3356 | BUG(); |
Davidlohr Bueso | c8c06ef | 2014-12-12 16:54:24 -0800 | [diff] [blame] | 3357 | down_write_nest_lock(&mapping->i_mmap_rwsem, &mm->mmap_sem); |
Andrea Arcangeli | 7906d00 | 2008-07-28 15:46:26 -0700 | [diff] [blame] | 3358 | } |
| 3359 | } |
| 3360 | |
| 3361 | /* |
| 3362 | * This operation locks against the VM for all pte/vma/mm related |
| 3363 | * operations that could ever happen on a certain mm. This includes |
| 3364 | * vmtruncate, try_to_unmap, and all page faults. |
| 3365 | * |
| 3366 | * The caller must take the mmap_sem in write mode before calling |
| 3367 | * mm_take_all_locks(). The caller isn't allowed to release the |
| 3368 | * mmap_sem until mm_drop_all_locks() returns. |
| 3369 | * |
| 3370 | * mmap_sem in write mode is required in order to block all operations |
| 3371 | * that could modify pagetables and free pages without need of |
Kirill A. Shutemov | 27ba064 | 2015-02-10 14:09:59 -0800 | [diff] [blame] | 3372 | * altering the vma layout. It's also needed in write mode to avoid new |
Andrea Arcangeli | 7906d00 | 2008-07-28 15:46:26 -0700 | [diff] [blame] | 3373 | * anon_vmas to be associated with existing vmas. |
| 3374 | * |
| 3375 | * A single task can't take more than one mm_take_all_locks() in a row |
| 3376 | * or it would deadlock. |
| 3377 | * |
Michel Lespinasse | bf181b9 | 2012-10-08 16:31:39 -0700 | [diff] [blame] | 3378 | * The LSB in anon_vma->rb_root.rb_node and the AS_MM_ALL_LOCKS bitflag in |
Andrea Arcangeli | 7906d00 | 2008-07-28 15:46:26 -0700 | [diff] [blame] | 3379 | * mapping->flags avoid to take the same lock twice, if more than one |
| 3380 | * vma in this mm is backed by the same anon_vma or address_space. |
| 3381 | * |
Kirill A. Shutemov | 88f306b | 2016-01-15 16:57:31 -0800 | [diff] [blame] | 3382 | * We take locks in following order, accordingly to comment at beginning |
| 3383 | * of mm/rmap.c: |
| 3384 | * - all hugetlbfs_i_mmap_rwsem_key locks (aka mapping->i_mmap_rwsem for |
| 3385 | * hugetlb mapping); |
| 3386 | * - all i_mmap_rwsem locks; |
| 3387 | * - all anon_vma->rwseml |
| 3388 | * |
| 3389 | * We can take all locks within these types randomly because the VM code |
| 3390 | * doesn't nest them and we protected from parallel mm_take_all_locks() by |
| 3391 | * mm_all_locks_mutex. |
Andrea Arcangeli | 7906d00 | 2008-07-28 15:46:26 -0700 | [diff] [blame] | 3392 | * |
| 3393 | * mm_take_all_locks() and mm_drop_all_locks are expensive operations |
| 3394 | * that may have to take thousand of locks. |
| 3395 | * |
| 3396 | * mm_take_all_locks() can fail if it's interrupted by signals. |
| 3397 | */ |
| 3398 | int mm_take_all_locks(struct mm_struct *mm) |
| 3399 | { |
| 3400 | struct vm_area_struct *vma; |
Rik van Riel | 5beb493 | 2010-03-05 13:42:07 -0800 | [diff] [blame] | 3401 | struct anon_vma_chain *avc; |
Andrea Arcangeli | 7906d00 | 2008-07-28 15:46:26 -0700 | [diff] [blame] | 3402 | |
| 3403 | BUG_ON(down_read_trylock(&mm->mmap_sem)); |
| 3404 | |
| 3405 | mutex_lock(&mm_all_locks_mutex); |
| 3406 | |
| 3407 | for (vma = mm->mmap; vma; vma = vma->vm_next) { |
| 3408 | if (signal_pending(current)) |
| 3409 | goto out_unlock; |
Kirill A. Shutemov | 88f306b | 2016-01-15 16:57:31 -0800 | [diff] [blame] | 3410 | if (vma->vm_file && vma->vm_file->f_mapping && |
| 3411 | is_vm_hugetlb_page(vma)) |
| 3412 | vm_lock_mapping(mm, vma->vm_file->f_mapping); |
| 3413 | } |
| 3414 | |
| 3415 | for (vma = mm->mmap; vma; vma = vma->vm_next) { |
| 3416 | if (signal_pending(current)) |
| 3417 | goto out_unlock; |
| 3418 | if (vma->vm_file && vma->vm_file->f_mapping && |
| 3419 | !is_vm_hugetlb_page(vma)) |
Peter Zijlstra | 454ed84 | 2008-08-11 09:30:25 +0200 | [diff] [blame] | 3420 | vm_lock_mapping(mm, vma->vm_file->f_mapping); |
Andrea Arcangeli | 7906d00 | 2008-07-28 15:46:26 -0700 | [diff] [blame] | 3421 | } |
Peter Zijlstra | 7cd5a02 | 2008-08-11 09:30:25 +0200 | [diff] [blame] | 3422 | |
| 3423 | for (vma = mm->mmap; vma; vma = vma->vm_next) { |
| 3424 | if (signal_pending(current)) |
| 3425 | goto out_unlock; |
| 3426 | if (vma->anon_vma) |
Rik van Riel | 5beb493 | 2010-03-05 13:42:07 -0800 | [diff] [blame] | 3427 | list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) |
| 3428 | vm_lock_anon_vma(mm, avc->anon_vma); |
Peter Zijlstra | 7cd5a02 | 2008-08-11 09:30:25 +0200 | [diff] [blame] | 3429 | } |
| 3430 | |
Kautuk Consul | 584cff5 | 2011-10-31 17:08:59 -0700 | [diff] [blame] | 3431 | return 0; |
Andrea Arcangeli | 7906d00 | 2008-07-28 15:46:26 -0700 | [diff] [blame] | 3432 | |
| 3433 | out_unlock: |
Kautuk Consul | 584cff5 | 2011-10-31 17:08:59 -0700 | [diff] [blame] | 3434 | mm_drop_all_locks(mm); |
| 3435 | return -EINTR; |
Andrea Arcangeli | 7906d00 | 2008-07-28 15:46:26 -0700 | [diff] [blame] | 3436 | } |
| 3437 | |
| 3438 | static void vm_unlock_anon_vma(struct anon_vma *anon_vma) |
| 3439 | { |
Michel Lespinasse | bf181b9 | 2012-10-08 16:31:39 -0700 | [diff] [blame] | 3440 | if (test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_node)) { |
Andrea Arcangeli | 7906d00 | 2008-07-28 15:46:26 -0700 | [diff] [blame] | 3441 | /* |
| 3442 | * The LSB of head.next can't change to 0 from under |
| 3443 | * us because we hold the mm_all_locks_mutex. |
| 3444 | * |
| 3445 | * We must however clear the bitflag before unlocking |
Michel Lespinasse | bf181b9 | 2012-10-08 16:31:39 -0700 | [diff] [blame] | 3446 | * the vma so the users using the anon_vma->rb_root will |
Andrea Arcangeli | 7906d00 | 2008-07-28 15:46:26 -0700 | [diff] [blame] | 3447 | * never see our bitflag. |
| 3448 | * |
| 3449 | * No need of atomic instructions here, head.next |
| 3450 | * can't change from under us until we release the |
Ingo Molnar | 5a50508 | 2012-12-02 19:56:46 +0000 | [diff] [blame] | 3451 | * anon_vma->root->rwsem. |
Andrea Arcangeli | 7906d00 | 2008-07-28 15:46:26 -0700 | [diff] [blame] | 3452 | */ |
| 3453 | if (!__test_and_clear_bit(0, (unsigned long *) |
Michel Lespinasse | bf181b9 | 2012-10-08 16:31:39 -0700 | [diff] [blame] | 3454 | &anon_vma->root->rb_root.rb_node)) |
Andrea Arcangeli | 7906d00 | 2008-07-28 15:46:26 -0700 | [diff] [blame] | 3455 | BUG(); |
Konstantin Khlebnikov | 08b5270 | 2013-02-22 16:34:40 -0800 | [diff] [blame] | 3456 | anon_vma_unlock_write(anon_vma); |
Andrea Arcangeli | 7906d00 | 2008-07-28 15:46:26 -0700 | [diff] [blame] | 3457 | } |
| 3458 | } |
| 3459 | |
| 3460 | static void vm_unlock_mapping(struct address_space *mapping) |
| 3461 | { |
| 3462 | if (test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) { |
| 3463 | /* |
| 3464 | * AS_MM_ALL_LOCKS can't change to 0 from under us |
| 3465 | * because we hold the mm_all_locks_mutex. |
| 3466 | */ |
Davidlohr Bueso | 83cde9e | 2014-12-12 16:54:21 -0800 | [diff] [blame] | 3467 | i_mmap_unlock_write(mapping); |
Andrea Arcangeli | 7906d00 | 2008-07-28 15:46:26 -0700 | [diff] [blame] | 3468 | if (!test_and_clear_bit(AS_MM_ALL_LOCKS, |
| 3469 | &mapping->flags)) |
| 3470 | BUG(); |
| 3471 | } |
| 3472 | } |
| 3473 | |
| 3474 | /* |
| 3475 | * The mmap_sem cannot be released by the caller until |
| 3476 | * mm_drop_all_locks() returns. |
| 3477 | */ |
| 3478 | void mm_drop_all_locks(struct mm_struct *mm) |
| 3479 | { |
| 3480 | struct vm_area_struct *vma; |
Rik van Riel | 5beb493 | 2010-03-05 13:42:07 -0800 | [diff] [blame] | 3481 | struct anon_vma_chain *avc; |
Andrea Arcangeli | 7906d00 | 2008-07-28 15:46:26 -0700 | [diff] [blame] | 3482 | |
| 3483 | BUG_ON(down_read_trylock(&mm->mmap_sem)); |
| 3484 | BUG_ON(!mutex_is_locked(&mm_all_locks_mutex)); |
| 3485 | |
| 3486 | for (vma = mm->mmap; vma; vma = vma->vm_next) { |
| 3487 | if (vma->anon_vma) |
Rik van Riel | 5beb493 | 2010-03-05 13:42:07 -0800 | [diff] [blame] | 3488 | list_for_each_entry(avc, &vma->anon_vma_chain, same_vma) |
| 3489 | vm_unlock_anon_vma(avc->anon_vma); |
Andrea Arcangeli | 7906d00 | 2008-07-28 15:46:26 -0700 | [diff] [blame] | 3490 | if (vma->vm_file && vma->vm_file->f_mapping) |
| 3491 | vm_unlock_mapping(vma->vm_file->f_mapping); |
| 3492 | } |
| 3493 | |
| 3494 | mutex_unlock(&mm_all_locks_mutex); |
| 3495 | } |
David Howells | 8feae13 | 2009-01-08 12:04:47 +0000 | [diff] [blame] | 3496 | |
| 3497 | /* |
| 3498 | * initialise the VMA slab |
| 3499 | */ |
| 3500 | void __init mmap_init(void) |
| 3501 | { |
KOSAKI Motohiro | 00a62ce | 2009-04-30 15:08:51 -0700 | [diff] [blame] | 3502 | int ret; |
| 3503 | |
Tejun Heo | 908c7f1 | 2014-09-08 09:51:29 +0900 | [diff] [blame] | 3504 | ret = percpu_counter_init(&vm_committed_as, 0, GFP_KERNEL); |
KOSAKI Motohiro | 00a62ce | 2009-04-30 15:08:51 -0700 | [diff] [blame] | 3505 | VM_BUG_ON(ret); |
David Howells | 8feae13 | 2009-01-08 12:04:47 +0000 | [diff] [blame] | 3506 | } |
Andrew Shewmaker | c9b1d09 | 2013-04-29 15:08:10 -0700 | [diff] [blame] | 3507 | |
| 3508 | /* |
| 3509 | * Initialise sysctl_user_reserve_kbytes. |
| 3510 | * |
| 3511 | * This is intended to prevent a user from starting a single memory hogging |
| 3512 | * process, such that they cannot recover (kill the hog) in OVERCOMMIT_NEVER |
| 3513 | * mode. |
| 3514 | * |
| 3515 | * The default value is min(3% of free memory, 128MB) |
| 3516 | * 128MB is enough to recover with sshd/login, bash, and top/kill. |
| 3517 | */ |
Andrew Shewmaker | 1640879 | 2013-04-29 15:08:12 -0700 | [diff] [blame] | 3518 | static int init_user_reserve(void) |
Andrew Shewmaker | c9b1d09 | 2013-04-29 15:08:10 -0700 | [diff] [blame] | 3519 | { |
| 3520 | unsigned long free_kbytes; |
| 3521 | |
| 3522 | free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10); |
| 3523 | |
| 3524 | sysctl_user_reserve_kbytes = min(free_kbytes / 32, 1UL << 17); |
| 3525 | return 0; |
| 3526 | } |
Paul Gortmaker | a64fb3c | 2014-01-23 15:53:30 -0800 | [diff] [blame] | 3527 | subsys_initcall(init_user_reserve); |
Andrew Shewmaker | 4eeab4f | 2013-04-29 15:08:11 -0700 | [diff] [blame] | 3528 | |
| 3529 | /* |
| 3530 | * Initialise sysctl_admin_reserve_kbytes. |
| 3531 | * |
| 3532 | * The purpose of sysctl_admin_reserve_kbytes is to allow the sys admin |
| 3533 | * to log in and kill a memory hogging process. |
| 3534 | * |
| 3535 | * Systems with more than 256MB will reserve 8MB, enough to recover |
| 3536 | * with sshd, bash, and top in OVERCOMMIT_GUESS. Smaller systems will |
| 3537 | * only reserve 3% of free pages by default. |
| 3538 | */ |
Andrew Shewmaker | 1640879 | 2013-04-29 15:08:12 -0700 | [diff] [blame] | 3539 | static int init_admin_reserve(void) |
Andrew Shewmaker | 4eeab4f | 2013-04-29 15:08:11 -0700 | [diff] [blame] | 3540 | { |
| 3541 | unsigned long free_kbytes; |
| 3542 | |
| 3543 | free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10); |
| 3544 | |
| 3545 | sysctl_admin_reserve_kbytes = min(free_kbytes / 32, 1UL << 13); |
| 3546 | return 0; |
| 3547 | } |
Paul Gortmaker | a64fb3c | 2014-01-23 15:53:30 -0800 | [diff] [blame] | 3548 | subsys_initcall(init_admin_reserve); |
Andrew Shewmaker | 1640879 | 2013-04-29 15:08:12 -0700 | [diff] [blame] | 3549 | |
| 3550 | /* |
| 3551 | * Reinititalise user and admin reserves if memory is added or removed. |
| 3552 | * |
| 3553 | * The default user reserve max is 128MB, and the default max for the |
| 3554 | * admin reserve is 8MB. These are usually, but not always, enough to |
| 3555 | * enable recovery from a memory hogging process using login/sshd, a shell, |
| 3556 | * and tools like top. It may make sense to increase or even disable the |
| 3557 | * reserve depending on the existence of swap or variations in the recovery |
| 3558 | * tools. So, the admin may have changed them. |
| 3559 | * |
| 3560 | * If memory is added and the reserves have been eliminated or increased above |
| 3561 | * the default max, then we'll trust the admin. |
| 3562 | * |
| 3563 | * If memory is removed and there isn't enough free memory, then we |
| 3564 | * need to reset the reserves. |
| 3565 | * |
| 3566 | * Otherwise keep the reserve set by the admin. |
| 3567 | */ |
| 3568 | static int reserve_mem_notifier(struct notifier_block *nb, |
| 3569 | unsigned long action, void *data) |
| 3570 | { |
| 3571 | unsigned long tmp, free_kbytes; |
| 3572 | |
| 3573 | switch (action) { |
| 3574 | case MEM_ONLINE: |
| 3575 | /* Default max is 128MB. Leave alone if modified by operator. */ |
| 3576 | tmp = sysctl_user_reserve_kbytes; |
| 3577 | if (0 < tmp && tmp < (1UL << 17)) |
| 3578 | init_user_reserve(); |
| 3579 | |
| 3580 | /* Default max is 8MB. Leave alone if modified by operator. */ |
| 3581 | tmp = sysctl_admin_reserve_kbytes; |
| 3582 | if (0 < tmp && tmp < (1UL << 13)) |
| 3583 | init_admin_reserve(); |
| 3584 | |
| 3585 | break; |
| 3586 | case MEM_OFFLINE: |
| 3587 | free_kbytes = global_page_state(NR_FREE_PAGES) << (PAGE_SHIFT - 10); |
| 3588 | |
| 3589 | if (sysctl_user_reserve_kbytes > free_kbytes) { |
| 3590 | init_user_reserve(); |
| 3591 | pr_info("vm.user_reserve_kbytes reset to %lu\n", |
| 3592 | sysctl_user_reserve_kbytes); |
| 3593 | } |
| 3594 | |
| 3595 | if (sysctl_admin_reserve_kbytes > free_kbytes) { |
| 3596 | init_admin_reserve(); |
| 3597 | pr_info("vm.admin_reserve_kbytes reset to %lu\n", |
| 3598 | sysctl_admin_reserve_kbytes); |
| 3599 | } |
| 3600 | break; |
| 3601 | default: |
| 3602 | break; |
| 3603 | } |
| 3604 | return NOTIFY_OK; |
| 3605 | } |
| 3606 | |
| 3607 | static struct notifier_block reserve_mem_nb = { |
| 3608 | .notifier_call = reserve_mem_notifier, |
| 3609 | }; |
| 3610 | |
| 3611 | static int __meminit init_reserve_notifier(void) |
| 3612 | { |
| 3613 | if (register_hotmemory_notifier(&reserve_mem_nb)) |
Mitchel Humpherys | b1de0d1 | 2014-06-06 14:38:30 -0700 | [diff] [blame] | 3614 | pr_err("Failed registering memory add/remove notifier for admin reserve\n"); |
Andrew Shewmaker | 1640879 | 2013-04-29 15:08:12 -0700 | [diff] [blame] | 3615 | |
| 3616 | return 0; |
| 3617 | } |
Paul Gortmaker | a64fb3c | 2014-01-23 15:53:30 -0800 | [diff] [blame] | 3618 | subsys_initcall(init_reserve_notifier); |